It’s becoming increasingly clear to both executives and communications professionals that generative AI chatbots—ChatGPT, Gemini, and their peers—are fast becoming the new discovery layer.

Traditional search’s old pattern of search → scan → click → repeat is giving way to the far more efficient act of having a conversation. We now have an emerging industry for “Generative Engine Optimization” (GEO) and “AI Optimization” (AIO), phrases I avoid because they focus on the platforms, not the users.

The shift is well underway. A YouGov poll indicated that 56% of Americans use AI tools, with 28% using them at least weekly.

Voice assistant usage, a related segment of conversational AI, also shows steady growth. In 2024, an estimated 48.7% of US internet users utilized voice assistants, with usage growing by 2.9% that year. The number of voice assistant users in the US is expected to reach 157.1 million by 2026. Smartphones are the primary access point for voice assistants, particularly for Gen Z.

The behavior patterns are unmistakable: people are turning to conversational systems to answer questions, solve problems, and guide decisions.

And conversational systems are evolving just as quickly.

Share

AI Memory

Two years ago, I wrote:

Right now, these conversational AI bots do not remember across sessions.

You can tell ChatGPT things about yourself that you want it to know at the start of each session, but if I want to pick up a conversation I had with ChatGPT or Pi a week ago, it doesn’t do it.

That limitation vastly reduces the utility of these tools. I have plenty of projects that span months; having to restart conversations about something I’m working on rather than simply referring to previous conversations is tedious and seems unnecessary.

Think about the nature of human conversations.

When you first meet someone, you exchange basic information about one another: Your name, what you do for a living, other basic biographical information.

Over the course of time, as you continue to engage that person in conversation, you gradually reveal more and more about one another and a lot of that information is retained and referred to in subsequent conversations.

The more you learn about one another through conversation, the more likely you will be to build trust with one another. The more you trust that other person, the more candor you will express with them and the open and honest you will be with your thoughts and opinions.

Now, give these AI chatbots long-term memory and apply those conversational dynamics to these tools, and you can see how they would become the most personalized product the world has ever known.

That limitation is fading away.

Today, both ChatGPT and Gemini can store persistent, cross-session memory, either because users explicitly tell them what to remember or because the system infers stable information from repeated use. The effect is profound: these tools no longer respond as generic assistants. They respond as assistants who know you.

ChatGPT now remembers preferences, long-term projects, working style, creative direction, and biographical details, so long as they’re not sensitive and you allow it.

Now, consider Gemini, owned by Google.

While Gemini has similar memory capabilities, both tools allow you to connect to productivity apps like Microsoft’s Office suite and Sharepoint or Google Workspace.

Google obviously has tighter integration with its apps.

Google takes pains to point out that it only accesses those tools on-demand and doesn’t use what it learns to store in long-term memory.

For now.

But companies change their privacy policies all the time. You can see the value to Google (and to users, by the way) of incorporating persistent, long-term memory of users into its products.

I can’t see how some form of this is not inevitable. As people begin to use these tools more frequently and they become more deeply embedded into everyday life, users will become frustrated that they remember some things but not enough things to make their experience with them easier and more enjoyable.

Such a feature enables Gemini (or any AI) to provide hyper-personalized, hyper-tailored, and hyper-relevant responses to that particular user’s needs.

My Ultimate Chrome Extension

Here’s an example.

The other day I wanted to test how good ChatGPT’s memory of me actually was. I asked: “Based on what you know about me, what kind of Chrome extension would be perfect for me?”

It recommended an “AI Brand Visibility Auditor” extension that doesn’t exist but is actually, as ChatGPT said, “Perfect for your AI Discoverability and AI Audit frameworks.”

Thanks, buddy!

AI Memory Is A User Reputation Algorithm

Memory isn’t simply a convenience feature. It’s the foundation of a user reputation algorithm.

In every conversation with an AI system, you are implicitly training it on:

• Who you are,

• What you care about,

• What you prefer,

• How you think, and

• What outcomes you are trying to achieve.

Over time, the AI could build a profile of who you are so it can more effectively deliver what you need. Your reputation builds inside the model, potentially based on your whole history of use of any Google product:

• This is the kind of person you are,

• These are the kinds of solutions you typically want,

• This is how to best serve you.

Reputation, in this context, means predictability. Predictability leads to personalization. Personalization leads to relevance.

And relevance determines what the model shows you.

In other words: as AI memory grows, the “answers” users receive will increasingly diverge. Two people asking the same question may—indeed, will—receive different recommendations, explanations, or solutions based on their conversational and personal reputation with the system.

This has an enormous implication for discoverability.

Hyper-Personalization Breaks the Old Rules of Discoverability

The internet we’ve known for 25 years produced a largely uniform reality. While search engines have customized results based on your search history, they have served roughly the same results to everyone, shaped by fairly well-understood ranking algorithms.

AI chatbots are hiding the customer journey and obliterating that relative uniformity of experience.

When memory steers the conversation, the model’s response is shaped by:

• Who the user is,

• What they’ve asked before,

• What the model believes they value,

• The patterns the user has reinforced, and

• The model’s accumulated “reputation profile” of the user.

Which means your brand, product, idea, argument, or narrative cannot simply “rank.” It must become relevant to a person’s identity, needs, and conversational journey.

If you aren’t part of their journey, you won’t be part of their answers.

This is why approaching AI discoverability with a generic, search-era mindset is a strategic error. We are entering a world where:

• Discovery is not a universal index.

• It is a conversation shaped by the user’s reputation with the model.

• And you only appear in that conversation if you map to the user’s identity and needs precisely.

Audience-First Approach

This is why it is folly to approach addressing discoverability in large language models without understanding your target audience intimately.

If chatbots generate different answers for different users, then discoverability depends on understanding:

• Who your audience is,

• Their sources of truth,

• What they believe they need,

• What problems they’re trying to solve,

• What language they use,

• What context shapes their conversation,

• What the model has already learned about them.

AI will tailor responses through the lens of the user’s reputation. If you don’t understand that user, intimately, your content, messaging, and expertise will never intersect with the AI’s personalized reasoning.

You will simply disappear from the conversation.

An audience-first approach means designing your communications for:

• The user’s identity,

• The user’s intent,

• The user’s conversational behavior,

• The model’s memory-driven personalization.

This is not SEO.

This is reputational alignment within a probabilistic conversational engine.

Optimizing for Semantic Proximity

If the AI is optimizing for the user’s reputation, your strategy must be to optimize for Semantic Proximity.

In the SEO era, we optimized for keywords to bridge the gap between a query and a URL. In the AI Memory era, we must optimize for concepts to bridge the gap between a user’s identity and your brand’s solution.

To survive the shift to hyper-personalized discovery, your communications strategy must pivot from broad “reach” to deep “resonance.” This requires a three-step inversion of the traditional model:

1. Reverse-Engineer the “Digital Twin”: Stop building personas primarily based on demographics. Start building them based on the data trail an AI would remember: their recurring problems, their preferred syntax, their trusted sources, and their ethical values. You are no longer creating for a “35-year-old male,” you are creating for a “highly conscientious and risk-averse decision-maker who values sustainability and prefers concise, data-backed answers.”

2. Establish Entity Authority: You cannot “keyword stuff” a conversation. You must establish your brand as a canonical entity within the model’s training data. This means creating high-value content that associates your brand inextricably with specific problems and solutions. You want the model to infer that You are the logical completion of the user’s thought process.

3. Map to the Journey, Not the Query: Users with long-term memory profiles don’t just ask questions; they pursue outcomes. Your content ecosystem must support that entire journey. If the AI knows the user is six months into a digital transformation project, it will prioritize content that speaks to implementation nuances, not generic “101” definitions.

The winners of this new era won’t be the loudest brands. They will be those that are the most relevant to their key stakeholders. They will be the brands that the AI determines are the most helpful companions for the specific user it has grown to know.

It is no longer about being found by the algorithm. It is about being recommended by the assistant.

Leave a comment

ICYMI

I posted several articles that were originally published on my LinkedIn profile during the time I was away. Here they are, if you’re interested:

Leave a comment

Share

(This article was originally published on April 17, 2025 on my LinkedIn profile.)

This morning I was a guest lecturer for Glenn Karwoski‘s strategic communications class at University of St. Thomas. We talked digital marketing generally and how to build marketing personas specifically.

During the course of the discussion, I explained to them how I sometimes create multi-persona AI chatbots in service of clients. Basically, we create deep, rich personas for the key audiences a given client needs to reach and turn those personas into chatbots we can have conversations with and gain insight from.

The ability to bring personas to life like that is amazing in and of itself, but it is really revealing when you put a bunch of related personas together in the same bot and instruct them to have a conversation.

We did just that for a client by creating a bot that featured nine types of self-directed investors:

1. Active Trader,

2. Long-Term Growth Investor,

3. Income-Focused Investor,

4. Value Investor,

5. Contrarian Investor,

6. SRI/ESG Investor,

7. Tech-Savvy Millennial Investor,

8. DIY Financial Planner, and

9. Casual/Passive Investor.

On April 8, in the wake of the market turmoil caused by Trump’s insane tariff “policy” announcement, I wanted to see how my self-directed investors felt about the news. I’ll share with you the conversation complete with my promts and the responses.

PROMPT:

Let’s start this session by having you go out and read the latest news about the state of the stock market. Go ahead and read the latest news and then give me a summary, please.

RESPONSE (citations removed):

As of April 8, 2025, the U.S. stock market is experiencing significant volatility, primarily influenced by recent tariff actions and trade tensions initiated by President Donald Trump’s administration.

Market Rebound Amid Tariff Concerns: After several days of steep declines, U.S. stocks have rebounded sharply. The Dow Jones Industrial Average surged over 1,300 points, while the S&P 500 and Nasdaq Composite each rose by more than 3.5%. This recovery is attributed to investor optimism following indications that Japan is engaging in trade negotiations with the U.S., suggesting a potential de-escalation of ongoing trade disputes.

Global Market Movements: International markets mirrored this positive trend. Japan’s Nikkei index rose by 6%, and European indices posted their best gains since 2022. These movements reflect a global response to the evolving trade discussions and the anticipation of resolutions that could stabilize international trade relations.

Sector-Specific Highlights:

• Technology Stocks: Major tech companies, including Nvidia, Tesla, Amazon, and Microsoft, experienced significant gains, contributing to the overall market uplift.

• Health Insurance Providers: Stocks of companies like Humana, CVS Health, Elevance Health, and UnitedHealth rallied following the Centers for Medicare & Medicaid Services’ announcement of higher-than-expected 2026 Medicare Advantage payments.

• Chipmakers: Companies such as Nvidia, Broadcom, Micron, and Taiwan Semiconductor saw rebounds, driven by optimism around AI-related performance and recovery from recent tariff-induced declines.

Analyst Insights: Despite the recent rebound, analysts caution that market volatility is likely to persist due to the unpredictable nature of ongoing trade negotiations and tariff implementations. The market remains sensitive to any developments in trade policies, and sustained stability may depend on concrete resolutions between the involved nations.

Current Market Data: Here’s a snapshot of the current market performance:

In summary, while the stock market has shown resilience with a notable rebound, the underlying factors contributing to recent volatility—primarily trade tensions and tariff actions—continue to influence market dynamics. Investors are advised to stay informed and exercise caution as the situation develops.

Armed with the latest news, I next asked:

PROMPT:

Now, I want all of you to have a discussion about the current state of the stockmarket, where you think it may be going, and how it affects your thinking about where to invest and how. Anyone can jump in at any time but every time you voice an opinion, you MUST indicate who you are so I can get your different perspectives. Go!

RESPONSE:

Absolutely, let’s have a roundtable! This will be a dynamic and candid conversation among different investor personas, reacting to today’s market rebound and the broader macro context.

🟠 Active Trader: “Look, this bounce is exactly the kind of volatility I live for. A 1,300-point jump on the Dow? That’s a playground for me. But I’m cautious too—these trade war headlines can flip sentiment in a second. I’m scalping momentum plays in chipmakers like Nvidia and Micron—anything AI-adjacent with high beta. I’ve got tight stops in place, though, because this market could retrace fast if the optimism fades. I’m not married to anything—I’m dating every ticker!”

🟢 Long-Term Growth Investor: “Frankly, I see this rebound as noise. It doesn’t change the fundamentals. Trade policy uncertainty may weigh on earnings down the line, especially for multinationals. But my horizon is decades, not days. I’m continuing my DCA (dollar-cost averaging) into high-quality growth companies—Microsoft, Google, and a few global ETFs. I’ll use any short-term pullbacks to add to positions. I’m not trying to time the market—I’m trying to stay in it.”

🔵 Income-Focused Investor: “The volatility is concerning, especially with all this tariff talk—it makes it hard to plan predictable income flows. But I did like what I saw from health insurers today. With Medicare Advantage rates going up, that’s a green light for names like UnitedHealth and Humana. I’m rotating into dividend aristocrats that have pricing power and lower international exposure—less trade risk. Think utilities, healthcare, and a few preferreds. Safety, yield, and cash flow remain king.”

🟤 Value Investor: “This market rally smells like short-covering and sentiment—not fundamentals. The trade war’s not resolved. In fact, it’s just begun. I’m sitting on cash and scanning for quality names the market is punishing without good reason. I’ve got my eye on consumer staples and banks trading below book value. I’ll wait until irrational fear drives prices further down. When there’s blood in the streets—even my own—that’s when I load up.”

🟡 Contrarian Investor: “Perfect time to be skeptical. Everyone’s celebrating a sugar-high from headlines. Meanwhile, the underlying risks haven’t gone away. The media’s narrative is swinging too fast—classic overreaction. I’m actually looking into sectors people are fleeing—emerging markets, maybe even Chinese tech if it gets obliterated. I’m also eyeing short opportunities in overextended large-cap tech. Most investors are buying comfort—I buy discomfort.”

🟣 SRI/ESG Investor: “I’m tracking the recovery too, but from an ESG lens. The volatility reminds me how fragile the system is without responsible governance. I’m reviewing holdings for companies with resilient supply chains and strong ESG scores—those likely to navigate trade shocks more ethically and sustainably. Clean energy, circular economy funds, and ESG-themed ETFs are still my focus. I believe values-driven investing is more than just a trend—it’s a risk-mitigation tool.”

🔴 Tech-Savvy Millennial Investor: “This whole rebound is like a meme rally on caffeine. Feels like Reddit’s back in the game! I’ve got exposure to Nvidia, Tesla, and I’m dabbling in crypto again since risk-on vibes are returning. I use apps like Public and SoFi to experiment with fractional shares and follow other traders. AI, Web3, and space stocks? Still hot. I know it’s risky, but I’m young—I can take the swings. Just make it fun and fast.”

🟤 DIY Financial Planner: “I see this as a teaching moment. Markets are irrational in the short term—that’s why I diversify, rebalance quarterly, and model scenarios with Monte Carlo sims. I’m revisiting my exposure to international equities and revising my tax-loss harvesting strategy to take advantage of recent dips. No panic. Just recalibration. And for anyone listening—have a plan before the market moves, not after.”

⚪ Casual/Passive Investor: “Honestly? I just checked my portfolio today because of all the headlines. I saw it’s back up, so… great? I’m not touching anything. My robo-advisor has me in a globally diversified ETF portfolio and rebalances for me. Wake me up when it’s retirement time.”

Would you like me to dive deeper into how each persona might change their portfolio allocations based on this environment? Or would you like to simulate a disagreement between them for a more lively debate?

Share The Reputation Algorithm

PROMPT:

Let’s simulate disagreement.

RESPONSE:

Perfect—let’s turn up the heat. 💥 Here comes a lively disagreement among investor personas reacting to today’s rebound and what it really means. No holding back—each one speaks from their convictions.

🟠 Active Trader: “This is a textbook momentum play, and if you’re not riding it, you’re leaving money on the table. You long-term folks are always waiting for perfect conditions—but the market moves on headlines. I’m scalping Nvidia for 6% swings while you’re reading 10-Ks.”

🟢 Long-Term Growth Investor: “And that’s why most traders underperform the market over time. What happens when Nvidia tanks on one tariff headline? I’m not chasing spikes. I’m building wealth, not gambling. Your trades might look smart today, but let’s talk again in 10 years.”

🟤 Value Investor: “Frankly, you’re both caught up in hype. Nvidia, Tesla, whatever—it’s all overpriced noise. This rally isn’t backed by earnings or fundamentals. I’m not buying anything until I see real margin of safety. I’d rather sit on cash than overpay for hot air.”

🔴 Tech-Savvy Millennial Investor: “Okay boomer, we get it—you like balance sheets. But the world’s changing fast, and you’ll miss the next wave waiting for ‘intrinsic value.’ AI, crypto, clean tech—that’s where innovation is. Sitting on cash? Might as well bury it in your backyard.”

🟡 Contrarian Investor: “Actually, I agree with Value here. All of you chasing tech or sitting passive are doing what the herd is doing. I buy when there’s panic, not euphoria. Today’s bounce? That’s a chance to short. Markets are emotional, and I profit from mispricing.”

🟣 SRI/ESG Investor: “I’m sorry, but none of this matters if we ignore sustainability. You’re debating tickers while the planet burns. My portfolio reflects values, not just returns. Long-term alpha comes from investing in responsible companies—not flavor-of-the-month plays or outdated industries.”

🔵 Income-Focused Investor: “With respect, that’s nice in theory, but I need to pay the bills. ESG is great, but if it doesn’t produce consistent dividends or bond yields, it’s not my priority. Right now, I’m locking in 5% yields with muni bonds—way less drama.”

🟤 DIY Financial Planner: “Honestly, this entire argument is why I don’t follow anyone’s ‘hot take.’ I simulate, optimize, and allocate across risk buckets. Everyone here has their bias. I just run the numbers and adjust quarterly. No emotion. No clickbait trades.”

⚪ Casual/Passive Investor: “Y’all are exhausting. I just want my portfolio to go up slowly while I live my life. You’re fighting over 1% alpha like it’s a religion. I’ll stick with my ETFs and rebalancing emails, thanks.”

🔥 Want to turn this into a panel discussion on a specific sector—like AI, energy, or real estate? Or have them respond to a big surprise event? I can keep this lively and educational.

And this is where I ended the session.

This is just the tip of the iceberg of what we can do with generative AI and it has become a standard part of our digital marketing practice at my agency.

Leave a comment

Share

(This article was originally published on May 25, 2025 on my LinkedIn profile.)

I’ve been working with artificial intelligence in one way or another nearly my entire career, or at least since the advent of search engines.

From the early days of search optimization, to navigating email spam filters and sender scores, to social media engagement algorithms and now to generative AI, I’ve been dealing with the communication and discoverability ramifications from the beginning.

Three Shifts In Consumer Behavior

There have been two major shifts in changing technology-driven consumer behavior during my career.

1. Keyword Search

The first was search engines making it much easier to find the infinite resources of the internet. Before search engines, people shared web sites they liked on Usenet Newsgroups, via private email lists, media coverage, “What’s New” pages detailing websites that had just “joined” the internet, and website directories of which Yahoo was the most popular.

Then in the mid-nineties search engines such as Web Crawler, Lycos and AltaVista emerged that indexed the full text of web pages, allowing for keyword searching. These early versions simply matched keywords, requiring the user to sift through pages of results and clicking on links until they found a site that satisfied their need.

It wasn’t until Google came along with its PageRank algorithm that search engines figured out how to rank websites according to searcher relevance. This has been the dominant discovery vehicle since 1998.

2. Social Media Feeds

The second shift was the social media feed that sends you information based on your behavior and connections within the social network. With the algorithmically-driven social feeds like Twitter and Facebook, you didn’t have to search; the algorithm just knew what was important to you based on who you were friends with and what content you engaged with.

The “news” just came to you!

Some would argue that the mobile revolution was a third shift in behavior but it only changed where people accessed information, it didn’t really change how people found information. The underlying behavior remained the same.

3. Conversational Computing

The advent of generative AI is the third major shift in consumer behavior and I contend that it is the most profound. Large Language Model-based (LLM) chatbots offer natural language conversation as a discovery vehicle, especially when coupled with voice-enabled multimodal capabilities.

A Brief History of Communications

To put this shift into context, we need to reflect on the history of communication.

A primary distinguishing characteristic of humans from the rest of the animal kingdom is our ability to pass knowledge on from generation to generation through language coupled with the ability to invent tools.

Let’s start with the caves. Back in the day, pre-civilized human beings told each other stories around campfires and recreated some of those stories on the walls of caves. But the dissemination of that knowledge depended upon proximity to both the campfire and to the walls.

As we migrated out of the caves and settled into agrarian communities that traded with other communities, we needed accounting systems to manage that commerce and thus the Summerians invented cuneiform script dating to 3100 BCE.

Fast forward to 1440 and Johannes Gutenberg’s printing press, making knowledge far more portable and durable to survive beyond living generations.

The broadcast era ushered in the capability to instantly transmit knowledge across the world to mass audiences.

Which brings us to the computer age.

The one consistent trend of the computer age has been increasingly effortless input on behalf of the user.

Early computers depended upon tape and punch cards as input devices. The next evolution of computers depended upon users typing in commands via a keyboard and a monochromatic screen.

We graduated from command lines to graphical user interfaces and point and click behavior using a mouse. Mobile devices introduced touch screens. And now, with the sophistication of speech-to-text and text-to-speech technology, we can just talk to computers.

AI chatbots and natural language processing are the fulfillment of the academic paper that really got me interested in this career path. “The Computer for the 21st Century” [PDF] was written by Mark Weisner and published in Scientific American. It describes the moment we are approaching right now with multimodal large language models.

Chatbots are Changing Consumer Behavior

That’s a long way of saying that ChatGPT and Gemini and Claude are changing consumer behavior.

The Generative AI Customer Journey

New research from Adobe Analytics reveals how routine user behavior is changing due to the introduction of AI-driven chat.

Here’s Adobe’s methodology:

“The Adobe Analytics insights are based on direct transactions online, showing the impact of generative AI on the digital economy. The retail insights, for instance, are based on analysis of more than 1 trillion visits to U.S. retail sites, a greater volume of data than is available to any other technology company or research organization. A companion survey of more than 5,000 U.S. respondents provides an additional layer of context on how consumers are thinking about AI in their daily lives.”

It’s not just us early-adopters that have embraced generative AI:

“Between Nov. 1 and Dec. 31, 2024, traffic from generative AI sources increased by 1,300 percent compared to the year prior (generative AI traffic was up 1,950 percent YoY on Cyber Monday). The trend has persisted beyond the holiday season. In February 2025, traffic from generative AI sources increased by 1,200 percent compared to July 2024. This shows the shift in consumer behaviors over a seven-month period. And while generative AI traffic remains modest compared to other channels, such as paid search or email, its growth has been notable — doubling every two months since September 2024.”

And here’s what normal people are using AI for:

• 39 percent have used generative AI for online shopping,

• with 53 percent planning to do so this year.

The shopping tasks consumers are using generative AI for include:

• Conducting research (per 55% of respondents),

• Receiving product recommendations (47%),

• Seeking deals (43%), Getting present ideas (35%), Finding unique products (35%) and

• Creating shopping lists (33%).

And when AI chat users finally arrive at a website, they are more engaged:

“Compared to consumers coming from non-AI traffic sources (including paid search, affiliates & partners, email, organic search, and social media), consumers coming from generative AI sources show 8 percent higher engagement as they linger on the site for a longer period of time. These visitors also browse 12 percent more pages per visit, with a 23 percent lower bounce rate.”

When you consider that these users have likely already done most if not all the research they need with the chatbot, these numbers are entirely reasonable. These are qualified leads.

This is the first research I’ve seen that documents how ordinary people who are not as AI-obsessed as your author are using generative AI for everyday purposes. It aligns with the shift away from search engines and toward chatbots that I believe is coming.

The use of AI chatbots really messes with the traditional customer journey we marketers have been using for years. Most of that journey will likely occur directly within the chat session because for most use cases, all the necessary knowledge to satisfy the users’ need will reside within the LLM.

Share The Reputation Algorithm

Chatbots Are Not Dominant…Yet

Rand Fishkin of SparkToro has the receipts.

I have no doubt that these numbers accurately represent the current market share. But I’m also confident that traditional search market share dominance is likely to diminish relatively quickly given the pace of generative AI development and adoption.

Here’s why:

The AI Arms Race Is On

I have never seen a technology as profound and which has improved so dramatically at such a quick pace in my entire lifetime. The competition between the major frontier LLM developers coupled with open source models and innovations such as those introduced by DeepSeek have pushed the pace of improvement quicker and introduced more and more powerful models.

There’s the figurative arms race between developers and then there’s the literal arms race between nations. AI policy, such as it is, is being driven in the current administration by AI accellerationists, so we can expect few if any barriers imposed by the federal government on AI development.

You’ve likely already heard the phrase today is the worst version of AI you’ll ever use.

Consumers Are Still Figuring AI Out

Just as we are at the dawn of the development of AI, so too are consumers at the dawn of understanding just how much value AI chatbots can provide.

Before Google came around, we had to understand how to deploy boolean search operators in order to get the best results out of search results.

Likewise, most consumer users of ChatGPT or Gemini are using zero-shot prompts where they are asking the model to perform a task without providing it with any specific examples of how to do it, let alone any more sophisticated prompt engineering techniques that truly unlock the chatbot’s capabilities.

When consumers start understanding how to extract a lot more value from chatbots or the developers start accounting for a lack of consumer sophistication within their models, those market share numbers will begin to change.

Consumer Skepticism Will Dissipate

Consumer skepticism of generative AI is still fairly strong due in large part to the well-known hallucination problem.

That problem can be mitigated through more sophisticated prompt engineering tactics but model developers are also taking steps to improve the accuracy of chatbot outputs. I’ve seen the improvement myself over time.

Anecdotally, I’ve noticed a wariness of people having a hard time trusting Google’s AI overviews. I doubt it is because people are fact-checking them. I think it is largely due to media coverage of the unreliability of AI results and that media coverage was in turn a result of using earlier, less reliable models.

That said, I’ve noticed the occasional inaccurate Overview myself so I don’t put complete faith in the AI overview feature of Google Search.

But you know what? People make calculations on the degree of risk in trusting an AI overview. If it’s an AI overview about a low-dollar investment, then that is likely to have little effect on the user’s purchase decision. But the higher the monetary investment or the more crucial the information to making a decision, the less likely the user is going to trust the AI overview.

I remember a time when I swore I would not share my credit card information with a website.

Google Is Hedging Its Bets

Google clearly doesn’t want to kill the advertising cash cow that depends upon keyword searches. But it also is a market leader in generative AI and has ambitions to be the market leader.

And Google recognizes what I’m recognizing: That discovery though chatbots is a far better user experience than traditional search.

So, the company is integrating AI into all its products and it is experimenting with integrating generative AI within search via its AI Overviews feature and the new AI Mode search experiment.

(Britney Muller recently published interesting findings from her research using Google’s new AI mode search feature, which lends some insight into at least what Google is emphasizing in AI-powered search.)

Consider also that right now, chatbots are mostly used via text exchanges or audio conversations. In the fairly immediate future, we’ll be able to chat with bots not just on our phones but also via our smart speakers and smart glasses. Effortless input.

So, with a nod to Rand Fishkin’s data and to quote Humphrey Bogart, “Maybe not today, maybe not tomorrow, but soon and for the rest of your life.”

Based on this belief, I have been fundamentally reframing my agency’s digital marketing practices to align with how an organization or brand can be discovered by their key stakeholders via Large Language Models.

Leave a comment

Share

(This article was originally published on January 13, 2025 on my LinkedIn profile.)

Last week I mentioned how I use short prompts in generative AI tools to try and understand the biases embedded in the content the large language models were trained on.

I’ve been using Midjourney to create images fairly extensively for the past two years or so.

Aside from the tool’s difficulty rendering realistic hands (a limitation that is common among image generators), one of the most maddening limitations is its seeming inability to create realistic images of ordinary-looking people.

I don’t need every image to depict beautiful people. In fact, most of the time I need to create average-looking people, so it is a bias I frequently struggle with.

Let me demonstrate.

A Young Woman

The only prompt I gave Midjourney was “a young woman” and, as you can see, all four of the images it generated are of pretty young women.

A Young Man

Likewise, all four of the images for the prompt “a young man” depict good-looking young guys. You’ve no-doubt noticed that all eight of these images are of white people.

An Ordinary-Looking Young Woman

Now let’s try and force Midjourney to tone down the beauty. This time, my prompt includes the phrase “ordinary-looking”:

They all still look very pretty.

You will note that the use of “ordinary-looking” in the prompt produced one non-white woman. Still beautiful but is it a coincidence that the addition of that phrase generated a dark-skinned person?

An Ordinary-Looking Young Man

Again, all of them are very good-looking and, again, there is a non-white example amongst them. This is starting to feel like it’s not a coincidence.

An Ugly Young Woman

The addition of the word “ugly” prompted Midjourney to create woman with exaggerated features. And I wouldn’t call the image in the upper left-hand corner ugly.

An Ugly Young Man

Midjourney responded my prompt to produce an ugly young man by just making them zombies...or caricatures.

A Beautiful Young Woman

Here’s what we get by explicitly telling Midjourney to create a beautiful young woman.

We return to an all-white cast of characters and one exhibiting a seductive expression.

A Beautiful Young Man

Four more white guys.

Embedded Bias

So what gives?

There are a few potential explanations.

Data Bias

The default to white people speaks volumes about the nature of the training data. You can create images of African Americans, Latinos, and Asians but if you don’t specify, you’ll get white people.

The data Midjourney was trained on likely included a disproportionate number of images of conventionally-beautiful white people, thus prioritizing those image features.

Think about the imagery of people that is publicly-available.

Media outlets are more likely than not to include images of beautiful people to accompany their content.

Social media influencers are by and large physically beautiful, so the images in their feeds are likely influencing Midjourney.

And perhaps the labelling of the training data included far more images labelled as beautiful compared to the number of images labelled ugly.

Leave a comment

Share

(This article was originally published on January 3, 2025 on my LinkedIn Profile.)

I recently conducted an in-depth survey (nearly 3,500 words) of what some of the leading research firms and AI experts are predicting for this year.

I’d like to share some of the trends I find most interesting from a marketing/strategic communications point of view and a few of my own, based on my experience with artificial intelligence over the years and Gen AI during the last two.

8. AI Resistance

PwC’s 2024 Workforce Radar survey reveals that “41% of executives say that workforce issues, such as training, culture, or change in work are among the top-five challenges their organizations face in using GenAI.”

There is certainly hesitancy to the adoption of Generative AI in the workplace. Some of this is driven by fear of AI replacing jobs.

I experienced this first-hand last year during two SEO events I attended that made it clear that search engine marketing professionals were fearful AI was coming for their jobs.

The backlash from advertising creatives over Coca-Cola’s AI-generated Christmas ad is another example of such fear. Just read the quotes in this AdWeek article and you’ll get a sense of what I mean.

Those fears are not unfounded. Gartner predicts that 20% of organizations will leverage AI to flatten their organizational structures by 2026, eliminating more than half of current middle management positions.

Another behavior I’ve noticed is people dismissing the value of Generative AI because they got poor results after using it like a search engine.

I’ve seen this behavior before when people who didn’t know what they were doing tried out Google search ads, only to get underwhelming results because they didn’t know what they were doing.

You really need training on how to use generative AI if you want to get the most value from it. If you use it like a search engine or ask it to write an article without the proper prompting, it’s no surprise you’ll get meh results.

Until we get acclimated to this new technology and fully understand how to use it, we will continue to see significant levels of resistance to its use.

That may be why Forrester cautions [PDF] that impatience with achieving a return on investment (ROI) could lead some enterprises to prematurely scale back their AI initiatives.

It is critical to acknowledge the communication challenge this presents for leaders.

Organizations need strategic messaging to address both internal fears (e.g., job loss) and external skepticism (e.g., “AI can’t deliver ROI”).

From a strategic comms perspective, companies should double down on storytelling about how AI complements rather than replaces human roles. For example, showcasing real-life “human-machine partnerships” where AI enhances human creativity or productivity could soften resistance.

7. Strategic AI

However, for those organizations that have already devoted time, resources and skill development to using Generative AI, 2025 should be a year of more deliberate and strategic implementation of the technology.

Forrester anticipates a shift towards a more strategic approach to AI in 2025, with an emphasis on business outcomes, data quality, and AI talent development.

Some industries, such as healthcare and finance, have been using artificial intelligence for quite some time, Gen AI being only the most recent variety. These are industries to watch, as their previous AI experience will heavily inform their use of Generative AI.

My work at Tunheim during the past year was heavily focused on using generative AI for insights into industries, markets and audiences to supercharge the communication strategies we develop for clients.

In PR and marketing, the move to strategic AI will require agencies to shift focus from simply generating outputs (content, insights) to embedding AI into workflows. This could mean:

• Using Gen AI for rapid scenario planning during crisis comms, or

• Leveraging AI for hyper-personalized audience segmentation.

This will likely create new competition among agencies, favoring those who invest early in AI training and talent development.

6. Use Of AI in Marketing Gets More Sophisticated

Michael (Mike) Stelzner of Social Media Examiner recently surveyed 1,250 marketers about their use of Generative AI and found that “nearly two-thirds are using generative AI tools either daily or weekly, with 37% using AI each day.”

While the adoption rate among marketers is high, the tasks they use it for are fairly predictable if somewhat surprising to me.

Creative professionals using Gen AI as a brainstorming partner is a pretty obvious use case, as are research assistance and content summarization.

I would’ve expected to see a bit more sophisticated use of generative AI among marketers. That could just be because 62% of the respondents have used Gen AI for less than a year, so they haven’t yet come to understand the full capabilities of this technology.

But if most of the tasks they are applying Gen AI to are content-related, they are barely tapping the surface of the value this technology provides. I expect marketing and strategic comms use cases will expand in 2025.

5. No Code Software Development

McKinsey predicts that AI’s ability to auto-generate code based on natural language inputs will significantly influence software development.

This will empower entrepreneurs and creators without technical backgrounds to develop software solutions, potentially leading to a surge in lean, agile startups.

I’ve seen this in action directly.

While I’ve got some programming chops, it is not a skill I practice daily so when I do have to program something, I’m often hacking or having to do some research to refresh my memory.

This past year, I built a couple of custom GPTs to help:

• One to generate schema code based on user input, and

• Another web development expert bot that has saved quite a bit of time on several occasions.

We’re already seeing Generative AI incorporated into programming tools like GitHub CoPilot and Cursor, an AI code editor.

We will likely see a surge in niche marketing tools developed by creatives rather than technologists, democratizing the martech space. For communicators, this means two things:

• You’ll need to stay sharp about integrating niche tools into campaigns, and

• There’s potential to build proprietary tools for clients, increasing value.

I’ve got countless ideas for apps I’d like to build. I’m looking at 2025 as a year when I’ll actually try and use Generative AI to build some of them.

4. Not Quite Agentic AI

AI agents are all the buzz. If you’ve been following the news about AI at all, you will have run across numerous headlines proclaiming 2025 to be the year of Agentic AI.

The AI platforms haven’t exactly been muting the hype, either, as it’s good investor content.

It appears that Gartner has bought into the hype, as the company predicts that by the end of 2025, 40% of knowledge workers will rely on AI agents daily, freeing up time for strategic initiatives.

A more tempered assessment comes from experts from the Stanford Institute for Human-Centered AI who predict a shift towards collaborative AI systems in 2025.

To be fair, we have seen glimpses of Agentic AI. Anthropic introduced a “Computer Use” feature of its Claude AI chatbot in October that can not only browse but also interact with websites. But even that, amazing as it is, still requires explicit instructions.

True Agentic AI will have autonomy to get stuff done on its own without a lot of human supervision or direction.

The technology is just not there yet. Here’s an example:

I email a lot of articles to myself at my Gmail address. I include the headline of the article in the subject line along with a hashtag related to the article. So, if it’s about AI, I’ll add #ai to the subject line so I can filter all the AI articles for later use.

I also include a blurb from the article and the link to the article in the body of the email.

I wanted to use Gemini–which has an extension that gives it access to my Gmail account and my Google Drive–to give me all the AI articles in my inbox during the past month and turn it into a digest that includes the article headline, date, source, and URL and save that as a document in my Google Drive.

After several different attempts, I gave up. The closest I got was a fairly decent digest of half a month’s worth of articles written out in the Gemini interface itself but never saved as a Google Doc.

Also, it threw in a couple of Medium articles about AI that I never emailed myself. Close, but no cigar, Gemini.

I’d love to get it to work consistently but such a function is better categorized as an AI assistant than Agentic AI. It is, however, a step in the direction of Agentic AI.

Another example is Google’s Deep Research Gemini model released last month that will do just that, conduct deep research on your behalf.

And OpenAI is expected to release Operator this month, an AI assistant that will be able to use a browser.

So, the rollout of Agentic AI in 2025 will likely be incremental.

At this early stage, it is incumbent upon PR and comms professionals to establish ethical frameworks and workflows to govern agent actions.

3. Synthetic People

I first discussed the concept of virtual people in 2011 on a Daily Numbers podcast episode with my friend Pat Lilja. The talking point was about Japanese virtual pop star Miku Hatsune, who, as a hologram, was selling out auditoriums in Los Angeles at the time.

In 2018, my Beyond Social Media Show co-host ...B.L. Ochman and I discussed the rise of virtual influencers Lil Miquela and Shudu Gram, two creative agency concoctions with millions of Instagram followers.

Last August Google acquired the top talent from the popular AI character platform, Character.ai. Synthesia is a growing company that enables AI-generated doubles for B2B companies.

Now, Meta is going to roll out tools for people to create AI versions of themselves that live in the Facebook/Instagram universe.

Meta did try this before with AI versions of Tom Brady, Snoop Dogg, and other celebrities and the result was pretty underwhelming. I tried them out and was like, ugh, whatever. Not too engaging.

I think a lot of it had to do with Meta’s Llama LLM, which was not very impressive. It is used in my Meta Ray-Ban glasses, and, again, not impressed. Half the time it can’t accurately tell me what I’m looking at.

Still, though I’ve yet to give them a whirl, the latest Llama open source models have been earning raves, so I’m not counting out Meta’s new attempt at AI characters.

2. AI Vision Goes Mainstream

Google previewed Project Astra at the company’s I/O conference in May, its general purpose AI assistant that can see and hear in real time. OpenAI also revealed its multimodal capabilities in May.

It took both companies a long time to actually deliver.

Last month, I experimented a bit with Google’s Multimodal Live API with Gemini 2.0 with impressive results.

I gave the system instructions to be a marketing persona I’d built for a client and then had it review the client’s website for its reaction to design, messaging, likelihood to act on a given call to action.

Live, multimodal AI could revolutionize usability testing, audience analysis, and creative evaluation.

This will become a standard operating procedure for my agency in 2025.

I expect live multimodal AI to go mainstream in 2025. Last May, I wrote about what live, multimodal AI might look like longer term.

1. Infinite AI Memory

One of the aspects of the development of Large Language Models (LLM) I’ve been paying close attention to is the size of the context window.

Think of the context window as the size of the AI system’s memory. If you’ve ever had a long session with an AI chatbot that ends up going off the rails, it’s probably because it has exhausted its context window. It’s forgotten portions of the previous conversation you’ve had with it.

Context windows are measured in tokens, which, in LLM parlance, are words or bits of words. OpenAI’s GPT 4o model has a context window of 128,000 tokens, which translates into about 96,000 words.

Google’s Gemini 1.5 Flash model, for comparison, has a one million token context window while its Gemini 1.5 Pro model has a two million token context window.

According to Yahoo Finance, Mustafa Suleyman, the CEO of Microsoft AI, disclosed that Microsoft AI was developing technology with near-infinite memory capabilities, expected to be available by 2025, allowing AI to retain information indefinitely, transforming user engagement.

That would be quite the development.

An infinite context window would have obviously massive implications in terms of what you could do with such a model, especially having it analyze mind-boggling amounts of information.

But also think of what it would mean for an individual user. Setting aside the staggering privacy implications for a moment, think about all the ways that Google has to know things about me.

• I use Chrome, so it knows what websites I visit,

• I use Gmail, so it has access to my inbox,

• It knows what searches I perform,

• It knows what videos I watch,

• I have YouTube TV, so it knows what shows I watch,

• I use Google Maps, so It knows where I live and where I go,

• I use Google Drive, so it knows what I write,

• I use Google Photos, so it knows the pictures I take,

• I use Google News, so it knows my media consumption, and

• It has access to my calendar and my contacts.

In short, it knows pretty much everything about me. With all that data about me at its disposal, the responses its AI system can give me could be insanely personalized.

Add to that the ability of the AI system to remember things across sessions or even years.

If you think Google’s advertising system is good at matching companies and prospects, wait till you get a-hold of this monster!

Leave a comment

(This article was first published on November 25, 2024 on my LinkedIn profile.)

Share

Transcript

Patrick Milan (PM): Welcome to Tunheim’s podcast about being understood. I’m Pat Milan, Tunheim’s Creative Destruction Officer, and with me today is the head of our digital services and AI, David Erickson.

David, the first time you shared your perspective on what’s happening with discoverability, it set off alarm bells for me—but in a good way. Everything we know about capturing and engaging audiences on digital and social platforms seems to be changing rapidly.

Let’s start with this: What’s driving this disruption in discoverability?

David Erickson (DE): It comes down to two major factors: search and social behavior. These have always been the primary digital channels for discoverability, but we’ve been seeing changes for a while—since 2014 or even earlier. Now, with the rise of AI, those changes are accelerating.

AI is fundamentally altering how people search, engage, and discover brands.

The Impact of Zero-Click Search on Discoverability

PM: Let’s break this down. Search behavior is shifting dramatically. You’ve talked about something called “Zero-Click Search.” What does that mean?

DE: Zero-Click Search is a concept coined by Rand Fishkin at SparkToro. It refers to how Google increasingly provides answers directly within search results—things like Instant Answers, People Also Ask, and Knowledge Panels.

These features mean users often don’t need to click on links to get the information they need.

For years, we optimized content to rank among the ten blue links on a Google search page, but now those links are less valuable. Google’s goal is to keep users on their platform, where they can sell more ad inventory.

The Social Media Algorithm Shift

PM: That feels like a major challenge. As a brand, I want people to click through to my site. It sounds like Google is actively preventing that.

DE: That’s exactly what’s happening. Google wants to control the audience’s experience, and by keeping users on their platform, they maximize their revenue. But it’s not just search—social platforms are doing the same thing.

PM: Let’s talk about social media. What’s happening there?

DE: Social platforms, especially Facebook, have been deprioritizing content with links for years. Back in 2014, we started noticing a decline in traffic from social platforms. If you include a link in a post, algorithms actively penalize it, reducing its reach and engagement.

Instead, platforms like Facebook, Instagram, and TikTok prioritize native content—images, videos, and posts that keep users on their platform. They’re not in the business of driving traffic to your website; they’re in the business of selling ads.

Rethinking Social Media as a Brand Awareness Tool

PM: That’s incredibly frustrating. As marketers, we used to rely on social media to connect the dots—post content, track clicks, and build relationships. Now it feels like the chain is broken.

DE: It is a disruption. But it’s also an opportunity to rethink how we use social media. Instead of treating it as a direct response channel, we need to view it as a brand awareness channel.

The key is to create engaging native content that works with the platform’s algorithms, not against them.

Preparing for the AI-Driven Future

PM: So with all this disruption, what should brands focus on to remain discoverable?

DE: Discoverability requires a two-pronged approach:

1. Maintain a strong presence in traditional search: Google Search is still incredibly powerful, so SEO remains critical. Brands need to optimize their content for expertise, experience, authoritativeness, and trustworthiness (what Google calls E-E-A-T).

2. Prepare for the AI future: AI is becoming a primary gatekeeper of information. Tools like ChatGPT rely on structured, credible sources, so brands need to position their content as sources of truth. This includes creating high-quality, authoritative content that feeds into AI systems.

How Brands Can Become Sources of Truth

PM: Can you give an example of how brands can prepare for the AI-driven future?

DE: One of the biggest challenges for AI is accuracy. Models like ChatGPT struggle with hallucinations—essentially making things up.

To combat this, AI developers are focusing on retrieval-augmented generation, which relies on credible sources of truth. Brands that position themselves as experts and provide structured, reliable content will become go-to sources for AI systems.

Additionally, we recommend conducting an AI brand audit. This involves assessing how your brand appears in AI results today and identifying ways to optimize your content for AI discoverability in the future.

Share The Reputation Algorithm

Becoming Your Own Media Company

PM: You’ve mentioned that brands should think of themselves as media companies. What do you mean by that?

DE: The idea is simple: Instead of waiting for others to discover you or write about you, create your own content ecosystem. Being your own media company means consistently publishing valuable, engaging content that demonstrates your expertise and builds trust.

This approach aligns perfectly with the needs of both search engines and AI systems, which prioritize high-quality, authoritative content. It’s not about creating massive volumes of content; it’s about being strategic and meaningful in what you produce.

A Roadmap for Action

PM: This all sounds like a lot to take in. How do we help brands get started?

DE: It starts with three steps:

1. AI Brand Audit: Analyze your current digital presence and identify gaps in discoverability.

2. Discoverability Preparedness: Optimize your content for both traditional search and AI systems, ensuring it’s structured, credible, and engaging.

3. Thought Leadership: Establish your organization’s experts as trusted sources in your industry. This includes creating content, speaking at events, and building a reputation as a source of truth.

Conclusion: Embracing the Future of Discoverability

PM: It sounds like the rules of strategic communication are changing faster than ever. But with the right approach, brands can not only adapt but thrive.

DE: Exactly. The challenge of discoverability isn’t going away—it’s only going to get tougher. The brands that embrace these changes now will build a competitive edge for the future.

PM: Thanks, David. For those listening, if you’re ready to tackle discoverability and stay ahead of the curve, we’d love to help. You can reach out to us at Tunheim.com.

Leave a comment

Share

(This article was originally published on my LinkedIn profile on May 24, 2024.)

Search engine optimization professionals are scared shitless. I suspect many other people are, too.

Two recent events have left a deep impression on me. The first was a networking event on the topic of SEO jobs featuring Nick LeRoy, who is a local #SEO consultant and publisher of SEOJobs.com.

I was not looking for a job myself but rather was hunting for talent.

But it became clear over the course of the evening that a lot of search engine marketers are worried about the state of the industry and, as a result, are worried about their jobs.

LeRoy had just published an analysis of job listing data for search engine optimization openings that found a decrease in postings of 37% in Q1 2024 from 2023.

So there’s cause for concern. The primary culprit for the decrease is believed to be AI, specifically generative artificial intelligence.

LeRoy posed a reasonable interpretation of the decline in SEO job openings:

Tasks historically mapped to an entry-level position are now being done faster and cheaper with AI technology. Gone are the days of needing to map redirects or write metadata manually. This forces SEOs to level up in their careers quicker than previously required.

Historically, an entry-level SEO could provide immediate value by executing manual tasks (now taken over by AI technology) while learning how to manage clients, pitch to c-suites, and more extensive soft skills such as critical thinking and effective communication. These entry-level SEOs are now expected to have a base-level knowledge of search AND the soft skills to compete against their mid-level peers with 3+ years of experience.

The data at the top of this article shows that SEO roles at the “mid-level” have been impacted most. Companies want to “do more with less,” which means hiring cheap junior resources and paying for proven experience/results via senior SEOs.

Fast-forward to last Friday and the annual MN Search Summit, organized by the MnSearch - Digital Marketing Networking: Of the 15 keynotes and sessions, seven sessions and two keynotes were devoted primarily to the topic of AI.

The focus of my attention at Tunheim for the past two years has been on generative #AI, so naturally, every session I chose to attend had to do with artificial intelligence. These are the titles of the sessions I went to:

• How to maintain your job relevance in an AI world (Keynote)

• Don’t Lose Your Head or Your Brand over AI

• The New Sp[AI]ce Race

• The Impact of AI in SEO: How is it Going and What to Expect Next (Keynote)

• Generative AI Governance: Policies & Processes for Marketing Teams

• Improved Content Analysis with GA4 Explorations & AI

AI Is BAD

AI is bad.

That is the message I heard loud and clear from nearly every session I attended.

The fear was palpable.

Nearly each session devoted the first fifteen minutes or so to why artificial intelligence is awful. If you’ve been paying attention to AI at all during the past two years, you’ve likely seen the headlines:

• The venerable Sports Illustrated hires a vendor who uses ChatGPT to write inaccurate stories by wholly fictional writers that SI then publishes,

• Marvel Studios produces an animated intro to the show Secret Invasion based on AI-generated imagery,

• Microsoft’s chatbot Tay quickly devolves into a Nazi,

• A law firm’s attorneys use ChatGPT to argue their case only to find the chatbot fabricated the case law upon which they based their suit. (I mean, that’s just an incompetent counsel if they have to use a chatbot to learn how to win their case.)

Yes, there are instances of people putting it to ill-advised uses and there are most decidedly drawbacks to the technology as it stands today. Foremost among those drawbacks are:

• Hallucinations (or, as we scientists call it, making shit up).

• Stereotyping (although this can be mitigated against via awareness and clever prompting).

• A lack of diversity in its training data.

• The lack of compensation for using the content on which these models were trained.

• Forgetfulness. When you get to the close of a context window (AI’s memory system), the responses begin to deteriorate, which is where I’ve found the quality and truthfulness of the responses begin to break down. It’s like the AI gets tired.

• Hands. Every image-generation model is pretty inept at generating an accurate hand. It’s maddening.

• Oh, and this technology consumes a massive amount of electricity, and, you know, climate change and all.

First, this is the very early stage of generative AI development. These are known problems and there are many people at work trying to solve them. (Except for climate change, of course.)

Second, this technology has advanced and improved in leaps and bounds during the short two years since I’ve been paying close attention to it. It has evolved and spread in light years time compared to the decade it took the internet to profoundly reshape our lives.

Hell, I thought it was going to be three to five years before AI-generated video would improve to the degree that ChatGPT had and then OpenAI releases videos created with Sora.

Third, it is spreading faster and wider than any technology I have seen in my lifetime.

It is reshaping search. It is reshaping social media. It is coming for productivity apps. It is reshaping creative production. It is reshaping programming. It is putting a very powerful tool into everyone’s hands, especially now that Apple has promised on-device AI.

There’s a reason money is flooding into development of this new technology in a way that never occurred with crypto, blockchain, and–gawd-forbid–freaking NFTs.

But that’s not what made these people fearful. What was worrying them was that AI is about to make them obsolete.

That fear is not at all unfounded. There’s clear evidence that AI is at least making the work of SEO much harder but there’s also evidence that AI will make some of the drudge work of SEO a breeze.

SEO Is Evolving

There’s also the common sense perception that a tool like generative AI can do some of the grind work--like cleaning up transcriptions for Video SEO--practically instantaneously. I built a custom GPT to do just that. It’s a fantastic tool.

Look, SEO work has always required an understanding of how search systems work, some technical fluency in HTML, JavaScript, database technology, and the functioning of CMS systems, an understanding of human psychology and behavior, and some content, language and detective skills to excel.

Understanding how systems work? That skill alone is applicable to many things beyond SEO work but, I would argue, it is the most valuable skill right now. Large Language Models and generative AI are just new systems to be deciphered.

But It wasn’t just SEOs at the conference who were afraid.

Content creators are worried, too. One man, a writer, said of his employer, “they are requiring us to use AI to train ourselves out of a job.”

That’s a legit worry.

But if you’re a writer, presumably you have language skills and an imagination. Those are two of the primary skills you need to get the best results from AI.

English is the new programming language. Start applying those skills to getting the AI to do other valuable things rather than just writing content.

There is no doubt that generative AI is a very disruptive technology. But it ain’t going away. It’s only going to become more pervasive. So we need to adjust.

If you’re creating or optimizing commodity content, then, yeah, you probably should be looking to up your game or get out of the game entirely.

Ultimately, people will always search for answers and there will be no shortage of people who want to provide the answer. And that’s the sweet spot where content producers and SEOs thrive.

It’s just that the answer engines are changing.

This is a time to adapt.

Share The Reputation Algorithm

5 Optimization Adaptations To Consider

1. Pay Attention To Chatbot Results

Conversational computing has arrived. Start playing with the chatbots now to understand how they deliver answers and what they say in response.

Use them as a normal consumer would and pay attention to how your behavior changes as a result of your use. Learn how the underlying technologies–LLMs, Transformers, Neural Networks, etc.--work to help you better understand how they process information and produce results.

Ask them to identify their sources for responses. Use Gemini’s Ask Google button.

2. Give Bing Some Love

Yeah, we kinda knew that Bing existed but with Microsoft’s search engine accounting for three to six percent of search engine market share, depending on who you ask, Google has always commanded the preponderance of our attention.

But consider that Bing search is built into ChatGPT and Microsoft’s Copilot and now, with the AI generated social media prompts, in LinkedIn as well.

It’s time to rethink how we prioritize Bing as a discovery channel.

3. Social Search

We’ve known this for some time but it bears repeating that younger consumers are as likely to turn to social media as they are to search engines as a discovery channel. As a 2023 YPulse survey makes clear, the vast majority of consumers aged 13-39 have used social media as a search engine.

Put your sleuthing skills to work figuring out how to make your content visible within social search.

4. Optimize Expertise

One of the biggest things holding generative AI back right now is its propensity to confidently state falsehoods.

The AI companies know this. AI developers know this. That’s why there’s such a flurry of research and testing attempting to solve that problem.

According to Jason Barnard at Search Engine Land, Google’s March 2024 Knowledge Graph update focused on “Person” entities.

He writes: “Person entities in the Knowledge Vault increased by 17%. The biggest increase (+38 %) was among Person entities with E-E-A-T-friendly subtitles (researchers, writers, academics, journalists, etc.).”

What does that tell you? Google is focusing on credible sources of factual knowledge. It doesn’t take much reading between the lines to see that Google is focusing on bolstering its source of credible, factual training data for Gemini.

I’ve been advising clients that optimizing thought leadership is one component of the answer to the discoverability crisis AI chatbots pose.

5. Optimize For Human Psychology

A lot of search engine optimization work addresses the technical implementation of tactics designed to gain visibility within a set of search results.

• Is the website we’re trying to earn traffic to technically sound?

• Can it be easily read by Googlebot?

• Do we have a working XML Sitemap?

• Are we blocking anything with robots.txt?

• How fast do the pages load?

• Is the SSL cert properly installed?

• Does the site respond appropriately on mobile devices?

• Are we using schema properly?

• Where within the technical infrastructure of the site are we populating keywords?

SEOs spend some time considering the psychology of users.

• Is this or that TITLE/DESCRIPTION combination more likely to elicit a click in the search engine results pages (SERPs)?

• Is the landing page going to be perceived as relevant based on the SERP?

• What is the user experience of the landing page like?

But that’s a very narrow slice of user psychology.

I strongly believe that successfully bridging the transition from search-centric discoverability to AI-centric discoverability is an intimate understanding of your target audience’s psychology.

When we can see the world through our target audience’s eyes, the solution for how to reach them will write itself.

That means fulfilling their needs in a very deep way: Satisfying their informational needs, yes, but also in a way that aligns with their personality and is in harmony with their emotions.

If you can add sociological and psychological insights shaped by emotional intelligence to your skillset, you’ll have the tools to rise above the torrent of AI-created schlock that’s soon to come.

Leave a comment

(This article was originally posted on my LinkedIn page on May 28, 2024.)

Share

Most of the coverage of OpenAI’s latest updates to ChatGPT has been focused on Sam Altman apparently cloning Scarlett Johansson’s voice. Altman even tweeted the word “Her” immediately after the demonstration of ChatGPT4o, referring, of course, to the movie in which Scarlett Johansson stars as the voice of an AI operating system.

It does not surprise me that Altman wants ChatGPT to sound like an actual human that most people can recognize.

But even if you take Johansson out of the equation and only focus on the human-sounding quality of the voice response from ChatGPT, it makes complete business sense for OpenAI to anthropomophize its chatbot.

While “Sky,” the Johansson-sounding voice is no longer an option, the remaining voices still sound very human. Compare that to what consumers have become accustomed to when it comes to synthetic voices.

I asked my Alexa and Google Home “smart speakers” what they thought of ChatGPT4o. These are their responses.

And:

Not very smart, are they? Amazon is furiously at work building its own generative AI chatbot. Google has Gemini, which is quite good, so I would think they are planning on incorporating it into Google Home as soon as possible.

Now, compare those two responses to Inflection’s Pi AI bot’s response:

You’ll notice that I posed my question immediately to Alexa and Google Home but I started out my Pi bot session with small talk. I’ve found I always start my sessions with Pi that way because, as you can see, Pi’s responses sound human compared to Alexa and Google Home.

I only use Alexa or Google Home to get quick answers such as:

• What’s the weather for this week?

• When do the Twins play next?

• How many cups are in a gallon?

• How long will it take to drive to work?

• How do you spell sesquipedalian?

By contrast, my sessions with Pi tend to be lengthy conversations because 1) it is capable of holding conversations but most importantly, 2) it feels like a real conversation with a human.

I’ve been recommending Pi to anyone who will listen as the future of computing.

Ubiquitous Technology

The history of computers has been marked by ever effortless input mechanisms: From the punch-cards of the 1950s, to tape reels, to command lines typed by a keyboard, to the mouse and graphical user interface, to voice activation.

Back in the early ninties, I read an article by Mark Weiser’s titled “The Computer for the 21st Century” [PDF] which discussed the concept of ubiquitous computing and drew an analogy to eyeglasses.

Weiser described how the most profound technologies are those that disappear into the fabric of everyday life, becoming indistinguishable from it.

Just as eyeglasses have become a natural extension of the human body, unnoticed in daily use but essential for seeing, ubiquitous computing aims to make computers so integrated and pervasive that they become invisible, seamlessly supporting our activities without drawing attention to themselves.

We are fast approaching that vision.

Multimodal AI Models

That is why I’ve been focusing on the potential of multimodal chatbots. Multimodal models like ChatGPT4o can accept and process more than just text as a prompt.

As OpenAI’s demo featuring Sal and Imran Khan of the Khan Academy using ChatGPT to solve a math problem illustrates, the bot can accept voice prompts as well as understand what is being done on screen, which amounts to visual input.

One of the innovations OpenAI has come up with, which Pi AI cannot do, is the ability to be interrupted. Pi AI won’t accept another input until it is done speaking and returns to listening mode.

ChatGPT4o listens while it’s speaking and therefore understands and responds to interruptions, which makes the conversation flow better and feel more natural.

A Young Lady’s Illustrated Primer

But what is most impressive about ChatGPT4o is it’s ability to accept text, voice and video seamlessly. My immediate first reaction was to think of Neal Stephenson’s novel The Diamond Age.

Nell, the young protagonist of the novel, is born into poverty and raised with her brother Harv by a neglectful mother in an unstable environment. Nell acquires an interactive electronic book called “The Young Lady’s Illustrated Primer” that serves as a surrogate mother. Here’s Stephenson’s description of the book:

The book spoke in a lovely contralto, with an accent like the very finest Vickys. The voice was like a real person’s--though not like anyone Nell had ever met. It rose and fell like slow surf on a warm beach, and when Nell closed her eyes, it swept her out into an ocean of feelings.

“Once upon a time there was a little Princess named Nell who was imprisoned in a tall dark castle on an island in the middle of a great sea, with a little boy named Harv, who was her friend and protector. She also had four special friends named Dinosaur, Duck, Peter Rabbit, and Purple.

Princess Nell and Harv could not leave the Dark Castle, but from time to time a raven would come to visit them...

“What’s a raven?” Nell said.

The illustration was a colorful painting of the island seen from up in the sky. The island rotated downward and out of the picture, becoming a view toward the ocean horizon. In the middle was a black dot. The picture zoomed in on the black dot, and it turned out to be a bird. Big letters appeared beneath. “R A V E N,” the book said. “Raven. Now, say it with me.”

“Raven.”

“Very good! Nell, you are a clever girl, and you have much talent with words. Can you spell raven?”

Nell hesitated. She was still blushing from the praise. After a few seconds, the first of the letters began to blink. Nell prodded it.

The letter grew until it had pushed all the other letters and pictures off the edges of the page. The loop on top shrank and became a head, while the lines sticking out the bottom developed into legs and began to scissor. “R is for Run,” the book said. The picture kept on changing until it was a picture of Nell. Then something fuzzy and red appeared beneath her feet. “Nell Runs on the Red Rug,” the book said, and as it spoke, new words appeared.

“Why is she running?”

“Because an Angry Alligator Appeared,” the book said, and panned back quite some distance to show an alligator, waddling along ridiculously, no threat to the fleet Nell. The alligator became frustrated and curled itself into a circle, which became a small letter. “A is for Alligator. The Very Vast alligator Vainly Viewed Nell’s Valiant Velocity.”

The little story went on to include an Excited Elf who was Nibbling Noisily on some Nuts. Then the picture of the Raven came back, with the letters beneath. “Raven. Can you spell raven, Nell?” A hand materialized on the page and pointed to the first letter.

“R,” Nell said.

“Very good! You are a clever girl, Nell, and good with letters,” the book said. “What is this letter?” and it pointed to the second one. This one Nell had forgotten. But the Book told her a story about an Ape named Albert.

The book can survey Nell’s environment and become aware of her circumstances and thus tailors its lessons to teach her skills to overcome the challenges she faces.

While OpenAI’s technology is a far cry from the vision Stephenson had when he published the novel in 1995, there are certainly significant enough pieces in place for it to evolve into that vision.

The Pieces Of The Multimodal Puzzle

Let’s take a look at each of those puzzle pieces and examine their implications.

Audio Input

Think of the information that is collected beyond the words contained in a voice-activated chatbot prompt.

AI could analyze vocal characteristics to detect medical issues such as depression, anxiety or even if someone is experiencing a stroke. It could detect and interpret environmental sounds such as the presence of another person in the household using aggressive expressions.

It may be able to understand a user’s media diet and help supplement a person’s interests based on that diet or provide information that challenges inaccuracies the person is consuming.

It could also use cues such as hesitency of speech to detect, as Stephenson illustrated in the previous passage, whether or not the user is comprehending a concept.

Visual Input

The addition of a device’s camera as a source of information input and the ability to analyze imagery provides further capabilities.

Here’s a simple example where I took a photo, circled something in it, added a question mark, and uploaded the photo to ChatGPT as a prompt:

A simple example but pretty amazing nonetheless.

But it is the ability to interpret real-time video is where things get extremely interesting. Here’s the quick demo OpenAI provided:

People lie all the time. They lie about big things and they lie about small things. They lie even if they don’t think they’re lying, like about what emotions they are feeling.

But facial expressions betray them. Facial expressions don’t lie. If ChatGPT can conduct facial coding to determine a person’s emotions, that can provide signals about how the user is reacting during a given session. Or even if the user is paying attention at all.

If the user is disinterested or feeling frustrated, the bot can react accordingly by trying a different approach.

Chatbot Memory

Take all these inputs and assume continued use of the course of years and then add to that the ability of the chatbot to remember past conversations, and you have an insanely personalized and customized assistant.

The implications are staggering and I’ve only just begun to think about them.

Leave a comment

It’s been almost two years since I announced that I planned to leave Substack because of its Nazi problem.

Obviously, I’m still here.

I will say I did try to leave. I made the effort.

But I ran into some technical difficulties of which I will spare you the excruciating details. The short of it is, because of the way Substack handles custom domains (e.g. thereputationalgorithm.com), a change I made to the configuration of my domain resulted in me getting locked out of my own site for a loooonnnnnnng time.

You may have noticed this when going to the site and getting a domain resolution error.

I tried Substack’s “customer service” chatbot but it promised me all kinds of resolutions but never delivered…until recently.

During the intervening time, however, it became clear to me that center of gravity for the opposition to fascism was right here on Substack.

With historians Timothy Snyder, Heather Cox Richardson, and Ruth Ben-Ghiat putting the present moment in context, strategists like Simon Rosenberg, Dan Pfeiffer, and Ezra Levin analyzing and mobilizing opposition to the rise of fascism, and journalists like Carol Caddwalder, Dan Rather, Paul Krugman and Juliet Jeske, among many others, Substack serves as ground zero for the opposition, if you will.

Substack still has a Nazi problem but at this point, with ICE kidnapping peaceful people off the streets with no due process, I guess where isn’t there a Nazi problem?

So, I’m back here on Substack acknowledging my hypocrisy.

And it’s a tough acknowledgement to make because I have a pretty health contempt for hypocrisy.

But there it is.

I plan to update this newsletter with some articles I’ve published in the interim on LinkedIn, and then continue to update with new articles and book chapters.

Leave a comment

Nearly ten years ago the agency I was with had just obtained the results of a public opinion poll we had commissioned on behalf of a client in order to gain some consumer insights into the client’s services.

The job of interpreting the data was assigned to an account manager with about seven years experience in public relations.

When I got the results of her interpretation of the data, it quickly became apparent that she didn’t know how to read polling results.

This is no criticism of her abilities; she’d probably never been asked to interpret a poll before. And public relations professionals are typically language people, not numbers people. You can count me as Exhibit A: I got a BA in English.

Nevertheless, I was a bit shocked when it dawned on me that she didn’t understand polling. One of the primary things we strategic communicators do to keep our proverbial ear to the ground is to pay attention to public opinion polling so as to understand consumer sentiment.

What gave it away for me was that her analysis included conclusions on questions for which the data was within the margin of error.

I’ve been thinking a lot about that experience now that we are being bombarded with political polling results heading into the 2024 presidential election.

If someone who reasonably should have known how to read polls did not, can we reasonably expect the general public’s polling literacy to be any better? And what does that mean for an ostensibly well-informed electorate as we approach an election where the very survival of democracy is the primary choice to be made?

A Public Issues Reputation Algorithm

Political and policy polling is essentially a crude algorithm designed to divine the reputation of policy issues and political leaders among the electorate. But there are many problems with this particular “algorithm.”

The Many Misuses Of Political Polling

If you pay any attention whatsoever to online conversations about politics, you will have no doubt come across discussions about the failings of politcal polling by media organizations.

I spent many years publishing a website, a newsletter, and a podcast that tracked public opinion polling of all varieties, so I do not dismiss opinion polling out of hand. In fact, one of my first jobs out of college was with a public opinion research firm.

The industry has always had it’s challenges. But because the stakes are so very high when it comes to politics and public policy, political polling deserves a lot more scrutiny and criticism than consumer-related polling.

Since polling results can have a very real effect on shaping the public’s opinion, this year it is absolutely crucial for the public to question and accurately interpret the data behind the headlines.

Let me touch briefly on a few of the common criticisms leveled at political polling and then we’ll discuss real-life examples.

Sensationalism & Bias
There's a perception that media polls might be biased or sensationalized to attract viewers or readers, potentially influencing how questions are framed and results are reported.

Overemphasis On Horse-Race Journalism
Media polls often focus on the "horse race" aspect of politics (who's winning, who's losing) rather than substantive policy issues, which can skew public perception and discourse.

Sampling Issues
Similar to general polling criticisms, media polls can suffer from non-representative samples, leading to inaccurate portrayals of public opinion.

Effect On Public Opinion
Media polls can create a bandwagon effect, where the public's perception of a candidate or issue is influenced by poll results, potentially affecting election outcomes.

Lack Of Transparency
There's often criticism over the lack of transparency in the methodology used by media organizations, such as sample size, margin of error, and sampling technique.

Frequency & Timing
The frequency and timing of media polls can contribute to fluctuations in public opinion, especially close to elections, leading to accusations of influencing rather than measuring public sentiment.

Conflict Of Interest
Media organizations may have inherent conflicts of interest, especially if they have political affiliations or leanings, raising questions about the objectivity of their polls.

Limited Scope & Depth
Media polls might focus on a narrow range of issues or lack depth in their questioning, failing to capture the complexities of public opinion.

Interpretation & Presentation
How media outlets interpret and present poll results can sometimes be misleading, emphasizing certain aspects for dramatic or narrative purposes.

November 2023 New York Times Poll

The New York Times/Siena College poll released on November 5, 2023 is a superb example that helps to illustrate many of these criticisms. Let’s go through them category by category:

This is the headline and subhead of the Times article about the poll:

Trump Leads in 5 Critical States as Voters Blast Biden, Times/Siena Poll Finds

Voters in battleground states said they trusted Donald J. Trump over President Biden on the economy, foreign policy and immigration, as Mr. Biden’s multiracial base shows signs of fraying.

Buried at the bottom of the article—where media polling methodology is always buried—was this line: “The margin of sampling error for each state is between 4.4 and 4.8 percentage points.”

Here are the results for the six battleground states surveyed by the Times.

According to the Times’ margin of error, these results could be either 4.4 to 4.8 points larger or 4.4 to 4.8 percentage points smaller for the state results. Now, let’s recalculate the results by giving Biden 4.8 percentage points:

• Nevada:

  • Biden 45.8%

  • Trump 47.2%

• Georgia:

  • Biden 47.8%

  • Trump 44.2%

• Arizona:

  • Biden 48.8%

  • Trump 44.2%

• Michigan:

  • Biden 47.8

  • Trump 43.2%

• Pennsylvania:

  • Biden 48.8%

  • Trump 43.2%

• Wisconsin:

  • Biden 51.8%

  • Trump 39.2%

We can recalcuate in the opposite direction, of course, and it would look a lot different for Biden. This is a simple example but it gives you an idea of how margin of error can change your interpretation of polling results.

Typically, polls with larger sample sizes have smaller margins of error. The smaller the margin of error, the more confidence you are supposed to have in the poll results.

The total sample size for the Times poll was 3,662 across all six states. That means roughly 610 registered voters are supposed to be representative for the entire population of each state.

Furthermore, this post from The Cascadia Advocate points out that the poll oversampled Republican voters, a fact that went unreported in both the New York Times story about the poll as well as the mass coverage of the poll by other media outlets.

The article notes:

The crosstabs for these polls show that 75% of the respondents identify as very conservative, somewhat conservative, or “moderate” (which isn’t an ideology):

QUESTION: Do you consider yourself politically liberal, moderate, or conservative?

FOLLOW UP: (If liberal or conservative) Is that very or somewhat?

ANSWERS [ALL STATES]:

• Very liberal: 10%

• Somewhat liberal: 10%

• Moderate: 39%

• Somewhat conservative: 19%

• Very conservative: 17%

• Don’t know/Refused: 5%

So, taking into consideration the large margin of error, the small sample size, and the oversampling of Republican votes, this poll doesn’t really tell us much at all about the electorate in those states.

Let’s pause for a moment and evaluate what we’ve learned so far with regard to our categories of political polling misuse:

✅ Sensationalism? Check. The Times’ headline blares “Trump Leads in 5 Critical States as Voters Blast Biden” yet the poll’s own data hardly supports that premise.

✅ Bias? Check. See above.

✅ Overemphasis on Horse Race Journalism? Check. This poll is all about the horse race.

✅ Sampling Issues? Check. Small sample size. Large margin of error. Oversampling of Republicans.

✅ Lack of Transparency? Check. Burying basic methodological information at the bottom of the article ensures most people will not see it because people rarely read an entire article. But all media outlets do that, not just the Times. What is inexcusable is not divulging within the article the oversampling of Republicans but then that would dillute the claims in the headline, wouldn’t it?

✅ Interpretation & Presentation? Check. See Lack of Transparency regarding the lack of discussion about oversampling Republicans.

The Times As A Viral Engine

It is particularly troubling when the New York Times releases polling results that can reasonably be interpreted as suspect, especially when those results are promoted with a sensationalist headline that doesn’t accurately reflect the actual data.

Sensationalist headlines are, of course, clickbait journalism and the New York Times is not alone in using such tactics to drive website traffic.

But, as “the paper of record,” the New York Times ostensibly holds itself to a higher standard. And because of this perceived esteem, when the Times leads, other media outlets invariably follow, from broadcast to print to digital.

And that dynamic creates a viral effect. This screenshot of Google Search results about the poll illustrate how everyone follow the Times’ lead:

A lot of people will only have read the headline of the poll as they scroll through their feed or see the notification on their phone, so that is the only understanding of the poll they will ever get.

Additionally, broadcast media coverage of the results was hyperbolic and lacked any real examination of the actual data, taking the headlines as fact.

For low-information Democrats, such headlines are likely demoralizing. Demoralized voters are less likely to be motivated to vote, if they believe their vote won’t matter.

Which brings us to the next point about this poll: The timing of its release. The poll results were published on November 5, on the eve of elections in Virginia, Kentucky and Ohio.

On November 6, the Governor of Kentucky was up for re-election, control of Virginia’s legislature was at stake, and Ohioans were to decide whether or not to enshrine abortion rights in that state’s constitution.

While the Times poll did not survey voters in any of these states, it is plausible that some voters in those states could have felt demoralized and not shown up to the polls as are result.

There is no way to measure whether that dynamic actually occurred and the Democratic victories in each of those states would suggest not. But more importantly, those victories are actual data that contradicts the Times’ suggestion of Republican strength in battleground states.

✅ Timing? Releasing political polling results promoted by sensational and misleading headlines the day before important elections in swing states was irresponsible.

✅ Effect on Public Opinion? If you had done any social listening on November 5, you could not have avoided seeing the collective freakout about Trump’s alleged lead over Biden.

Political Polling’s Dodgy History

Look, this is one poll but it is emblematic of many of the problems with political polling. There is a long and sorry history of political polling getting it disasterously wrong. The Conversation gives us a brief history:

• 1936 Literary Digest poll predicted Republican Alf Landon would easily defeat President Roosevelt.

• In 1948, pollster Elmo Roper stopped releasing survey results two months before the election, confident Dewey’s lead in the polls would last. That, of course, resulted in the iconic photo of President Truman holding a copy of the Chicago Daily Tribune featuring the headline “Dewey Defeats Truman.”

• In 1960, the Roper poll predicted a two point Nixon victory.

• In 1980, polls predicted a close race between President Carter and Ronald Reagan. Reagan won by 10 percentage points.

• In 2016, polls predicted Hillary Clinton would win in the battleground states of Wisconsin, Pennsylvania, and Michigan.

• In 2020, polls overstated support for President Biden by nearly 4 percentage points.

• In 2022, Gallup confidently predicted a red wave. Democrats vastly overperformed expectations. The same Siena polling outfit we just dissected said independent women were swinging to the GOP in the wake of the Dobbs decision. We know how that turned out.

Polling Results Are Not Facts

And yet despite these consistent failures, people still treat polling results as fact.

Why?

Because the political commentariat treats them that way.

When discussing polls themselves, reporters and pundits alike treat the results as a facts without addressing the nuances or limitations of the poll.

Reporters use poll results as data points to support the premise of other news coverage, such as how this segment did on Sunday covering President Biden’s visit to South Carolina.

Here’s the transcript:

As the Biden campaign co-chair, Clyburn’s also here combating a polling reality. The most recent NBC News survey, taken in November, showing 61% of Black voters would choose Biden over a Republican. But in 2020, our exit polls show Biden won 87% of the Black vote.

Let’s start with the phrase “polling reality,” which frames the numbers about to be cited as an indisputable fact when, as we have just seen, that is a dubious proposition at best.

Another red flag: “most recent NBC News survey, taken in November.” So that’s—let me count—one, two, about three months ago.

It is highly questionable that 61% figure remains the same today. Could be higher. Could be lower. As you’ve no doubt heard many times before, polls are a snapshot in time.

Lastly, NBC News is comparing apples to oranges. Survey results are different than exit polls. While both methodologies depend upon self-reporting, exit polls are far more likely to be accurate because it is simply actual voters reporting an action they’ve just taken, where surveys are speculations on the future by potential voters.

A better approach would be to look at the actual vote totals by African Americans in recent elections in, say, Kentucky, Virginia, and Ohio and compare the turnout to the most recent past elections in those states to see if the percentage of the Black vote has grown or shrunk in those states.

True, this is not a direct comparison of South Carolina voters but it would be more likely to indicate a trend among Black voters what that NBC segment attempted to show.

Media outlets and the pundits who fill broadcast segments have a vested interest in polling as a way to create “news,” generate interest, and drive ratings. The inclusion of poll results as a data point in support of reporting without the necessary qualifications is just laziness.

Misleading Political Polling Metrics

If you fully understand the methodology and have access to the wording of the questions, polls can be useful in interpreting public opinion.

They are most useful as a snapshot in time and as such, lose value over time.

Polls are useful over time when the same question is asked of respondents over and over again during the course of a year or years. These are known as tracking polls and are useful in understanding how sentiment can change over time.

But even tracking polls are fraught with challenges.

Right Track/Wrong Track

The right track/wrong track question is a staple of tracking polls but the question—and therefore the answers—are subject to interpretation.

The answer I give as a polling respondent depends entirely on how I interpret the question.

For exampe, I may think that the course President Biden has set the country on is the right course and that we are making progress but I may also believe that the country is on the wrong track because of the rise of a fascist movement in the country.

So, though I think the country is on the right track in terms of policy and the administation’s accomplishments, I may answer that the country is on the wrong track because I am alarmed at the rise of fascism.

If a lot people feel as I do but answer with a wrong track response, it is likely that journalists will iterpret those numbers as an indictment of Biden’s performance, when that would be further from the truth.

Approval Rating

Another misleading aspect of tracking polls is the approval rating question.

The same basic dynamic applies here. I may approve of all of the policies the Biden administration has put in place and appreciate Biden’s accomplishments but be frustrated that he’s not doing more an issue that is important to me and therefore express disapproval to the pollster.

That is in no way an indication that I don’t plan on voting for him. But that is how approval ratings are interpreted.

Both of these metrics—right track/wrong track and approval rating—can influence public sentiment. If I see big wrong track numbers and low approval ratings for Biden, I might feel I’m in the minority, be demoralized and therefore less motivated to vote.

Ultimately, the only polling that matters are the ballots that are cast in elections.

So take polls with the grains of salt they deserve, investigate the details of how they were conducted, use what insights that make sense, along with other, better, data like actual election results, to help you best interpret the political landscape.

To that end, here’s a breakdown to help you understand the elements that go into opinion polling. (This following section was written with the assistance of AI but edited and vetted by me).

The Anatomy & Art of Polling

The Foundation: Sampling

Constructing a sample is akin to painting a miniature portrait of society. The 2016 U.S. Presidential election, for instance, highlighted the pitfalls of underrepresenting rural voters. This underscores the importance of a sample that mirrors the population's diversity and complexity.

Sample Size

Sample size is crucial in public opinion polling for several reasons:

1. Accuracy
   A larger sample size generally leads to more accurate results. It reduces the margin of error, meaning the poll's results are closer to the true opinion of the entire population.

2. Representation
   A sufficiently large sample is more likely to include diverse segments of the population, reflecting various demographics, political beliefs, and other characteristics. If the sample isn't truly representative, it can lead to biased results. For instance, over-representing a certain demographic group. Populations evolve, so keeping the sampling frame updated is a challenge. People might choose not to participate, creating another challenge that could lead to nonresponse bias.

3. Confidence Level
   The size of the sample directly affects the confidence level, which is the degree of certainty that the poll results represent the true opinion of the entire population. A larger sample size increases the confidence level.

4. Statistical Significance

   A larger sample size increases the likelihood that the poll's findings are statistically significant, meaning the results aren't due to random chance. This is especially important when trying to detect small differences in opinions or when dealing with a population that has diverse views.

5. Subgroup Analysis

   A sufficiently large sample allows for reliable analysis of subgroups within the population. For instance, understanding the opinions of specific age groups, geographic regions, or socio-economic classes requires enough respondents from each subgroup to draw meaningful conclusions.

6. Reducing the Margin of Error

   The margin of error in a poll decreases as the sample size increases. For example, a poll with 1,000 respondents might have a margin of error of around three percent, whereas a poll with 500 respondents might have a margin of error of around five percent. A smaller margin of error means the poll results are a closer approximation of the true opinion of the entire population.

7. Cost & Practicality Considerations

   While larger samples provide more accuracy, there's also a practical limit based on cost and logistics. Larger samples mean more interviews, more data processing, and ultimately more time and money. Pollsters often have to balance the need for a large sample with the resources available. Sampling a smaller, manageable group is more economical and faster than surveying an entire population.

8. Law of Diminishing Returns

   There's a point where increasing the sample size further provides little to no improvement in accuracy. For example, increasing a sample size from 1,000 to 2,000 reduces the margin of error, but not as dramatically as increasing from 100 to 1,000. Understanding this helps in allocating resources efficiently.

Weighting in Public Opinion Polling

Weighting is a statistical adjustment made to survey data to ensure that the sample accurately represents the population, especially when certain groups are over- or under-represented in the sample.

Purpose

1. Correcting Imbalances
   If a sample has too many or too few respondents from a particular demographic group relative to the population, weighting adjusts for this.

2. Enhancing Accuracy
   By accounting for demographic discrepancies, weighting helps to produce more accurate results that reflect the true opinions of the overall population.

How It Works

1. Identifying Key Demographics
   Pollsters identify important characteristics like age, gender, race, education, and geographic location.

2. Comparison to Population Benchmarks
   These sample demographics are compared to known benchmarks of the population, usually from reliable sources like census data.

3. Assigning Weights: Each response is given a weight based on how representative it is of the population. For example, if young people are underrepresented in the sample, their responses might be given more weight.

Challenges

1. Selecting Weighting Variables
   Choosing which variables to weight by is crucial and can significantly impact results.

2. Over-Weighting
   If a sample is very skewed, some responses might be overly weighted, which can introduce its own biases.

3. Assumptions About Non-Respondents
   Weighting often involves assumptions about the opinions of non-respondents, which can be risky.

Pros & Cons

• Pros
  Makes the best use of available data; can significantly improve the representativeness of the results.

• Cons
  Adds complexity; if done incorrectly, it can introduce new biases.

Evolving Techniques

• With changing survey modes and declining response rates, the methods and approaches to weighting are continually evolving. Pollsters and statisticians constantly refine weighting techniques to deal with new challenges in data collection.

Question Design: Crafting with Precision

The subtleties in how questions are phrased can significantly guide responses. Slight differences in wording can elicit varied opinions.

Importance of Word Choice

The wording of questions in a poll is critical because it can significantly influence how respondents understand and answer the questions. The goal is to craft questions that are clear, unbiased, and reflective of the respondent's true opinions.

Challenges

1. Avoiding Leading Questions
   Questions must be neutral and not lead respondents to a particular answer.

2. Clarity & Simplicity
   Questions should be easy to understand, avoiding complex language or jargon that might confuse respondents.

3. Avoiding Double-Barreled Questions
   Questions should only ask about one thing at a time, not multiple issues in one question.

4. Cultural Sensitivity
   The wording needs to be appropriate and sensitive to different cultural contexts.

Impact

1. Response Bias
   Poorly worded questions can lead to response bias, where the wording influences the respondent's answer.

2. Misinterpretation
   Ambiguous or confusing questions can result in respondents interpreting the question in different ways, leading to unreliable data.

Best Practices

1. Pre-Testing
   Conducting preliminary tests of the survey on a small scale to identify potential issues in question wording.

2. Review & Revision
   Having experts or peers review the questions to ensure clarity and neutrality.

3. Consistency
   Using similar wording in questions when tracking opinions over time to ensure comparability.

Examples

• Bad Wording
  "Do you agree with the unnecessary and harmful restrictions on gun ownership?"

• Improved Wording
  "Do you support or oppose restrictions on gun ownership?"

Mode Of Survey

The mode of survey refers to the method used to conduct the poll. Common modes include telephone interviews, online surveys, face-to-face interviews, and mail surveys.

Different Modes

1. Telephone Interviews
   These can be landline or mobile phone surveys. They were once the gold standard but have faced challenges like declining response rates and the shift from landlines to mobile phones.

2. Online Surveys
   Increasingly popular due to their cost-effectiveness and speed. They rely on internet access, which can skew towards certain demographic groups.

3. Face-to-Face Interviews
   These can provide more in-depth data and higher response rates but are more time-consuming and costly.

4. Mail Surveys

   Less common now, but useful in certain contexts, like reaching populations without reliable internet or phone access.

Pros & Cons

1. Telephone Interviews

   • Pros: Can reach a wide demographic, including older populations.

   • Cons: Lower response rates; mobile phone prevalence makes it harder to create a representative sample.

2. Online Surveys

   • Pros: Quick and cost-effective; allows for innovative question formats.

   • Cons: Internet access bias; may exclude less tech-savvy populations.

3. Face-to-Face Interviews

   • Pros: Deeper engagement; can clarify misunderstandings.

   • Cons: Expensive and time-consuming; interviewer bias can be a factor.

4. Mail Surveys

   • Pros: Accessible to those without internet or phone; perceived as less intrusive.

   • Cons: Slow response time; lower overall response rates.

Challenges

• Representativeness
  Each mode has biases in who is likely to respond, affecting how representative the sample is of the overall population.

• Technological Changes
  Rapid tech changes affect how people communicate, influencing the effectiveness of different survey modes.

• Response Rates
  All modes face challenges with declining response rates, affecting the quality of data collected.

Adapting to Change

• Pollsters often combine different modes to balance out the biases and limitations of each. For instance, mixing online and telephone methods to get a broader reach.

Response Rate

The response rate is the percentage of people contacted who actually complete the survey. It's a crucial metric for assessing the quality of a poll.

Importance

1. Indicator of Quality
   A higher response rate typically indicates a more reliable and accurate poll, as it suggests a better representation of the target population.

2. Bias Reduction
   Higher response rates reduce the risk of nonresponse bias, where the opinions of non-respondents might differ significantly from those who participated.

Challenges

1. Declining Rates
   Response rates have been declining globally, partly due to survey fatigue and concerns about privacy.

2. Reaching Diverse Groups
   Certain demographics are harder to reach, potentially leading to under-representation of these groups in the survey results.

Strategies to Improve Response Rates

1. Multiple Contact Methods
   Using various methods (phone, email, mail) to reach potential respondents.

2. Incentives
   Offering small incentives can encourage participation.

3. Follow-ups
   Sending reminders or follow-up requests to non-respondents.

4. Simplifying Surveys
   Making surveys shorter and easier to complete can improve response rates.

Measuring Response Rate

• Calculation
  It's calculated by dividing the number of completed surveys by the total number of contacts attempted.

• Reporting
  Transparently reporting response rates is essential for assessing the poll's credibility.

Impact On Polling Results

1. Low Response Rates
   Can lead to biased results if non-respondents differ significantly from respondents.

2. Overestimation of Precision
   Low response rates can give a false sense of precision if the sample isn't representative of the broader population.

Adjustments For Low Response Rates

1. Weighting
   Pollsters often use statistical techniques to adjust results to compensate for under-represented groups.

2. Benchmarking
   Comparing results to known benchmarks or other reliable data sources to assess accuracy.

Analysis & Interpretation in Polling:

Analysis and interpretation involve examining the collected survey data to draw meaningful conclusions and insights. This stage turns raw data into understandable information about public opinion.

Key Aspects

1. Data Cleaning
   Initially, the data is cleaned to remove any errors or inconsistencies, like incomplete responses or outliers.

2. Statistical Analysis
   Pollsters use various statistical methods to analyze the data, such as calculating averages, percentages, and cross-tabulations to explore relationships between variables.

3. Understanding Margin of Error
   Recognizing the inherent uncertainty in any poll. The margin of error indicates the range within which the true value for the whole population is likely to fall.

4. Trend Analysis
   Comparing current results with previous polls to identify trends or shifts in public opinion.

5. Subgroup Analysis
   Looking at how different demographic groups (age, race, gender, etc.) responded, which can reveal more nuanced insights.

Challenges

1. Avoiding Misinterpretation
   Data can sometimes be interpreted in ways that are misleading or ignore the complexity of the underlying opinions.

2. Contextualizing Results
   Understanding the broader social, political, or economic context is crucial for accurate interpretation.

3. Dealing with Ambiguity
   Some survey responses might be ambiguous or open to multiple interpretations.

Importance of Transparency

• Transparent reporting of methodology, data analysis, and interpretation is essential for credibility. This includes disclosing the margin of error, sample size, and how the sample was obtained.

Bias & Objectivity

• Analysts must be aware of their own biases and strive for objectivity. The phrasing of conclusions and the presentation of data should be as neutral and unbiased as possible.

Communicating Results

• Effectively communicating the findings to the public, stakeholders, or clients in an understandable and responsible manner is a key part of the process. This often involves simplifying complex statistical information without losing its accuracy.

Understanding The Margin Of Error

The margin of error in polling is a measure that represents the degree of uncertainty or potential error in the results of a survey. It's a crucial concept for understanding how closely the poll's findings might match the true feelings or behaviors of the entire population. Here's how it works:

Interpretation

1. Range of Results
   If a poll shows that 50% of respondents favor a candidate with a margin of error of ±3%, it means the actual proportion in the entire population could be as low as 47% or as high as 53%.

2. Not a Measure of Accuracy
   The margin of error does not directly measure the accuracy of the poll's results. It only gives an estimate of the possible range of error in the sample.

Factors Affecting Margin of Error

1. Sample Size
   Larger samples reduce the margin of error, but the relationship is not linear. Doubling the sample size doesn't halve the margin of error.

2. Population Variability
   If the population has a lot of variability, the margin of error may be larger.

3. Survey Methodology
   Different sampling methods can affect the margin of error, such as stratified sampling or cluster sampling.

Limitations

1. Applies to Random Sampling
   The margin of error concept assumes a probability sampling method. It may not apply to non-random samples.

2. Doesn’t Account for All Errors
   It doesn't cover errors like biased question wording, survey mode effects, or non-response bias.

Understanding Crosstabs

Crosstabs are tables that show the relationship between two or more survey questions or variables. They display how different responses to one question vary in relation to responses to another question.

Usage in Polling

1. Identifying Patterns and Relationships

   Crosstabs help in discovering patterns or relationships between different variables. For example, how opinions on a policy issue vary by age group, gender, or education level.

2. Segmenting Data

   They allow pollsters to break down overall survey results into segments based on demographics or other characteristics. This segmentation helps in understanding the specific opinions of different groups within the population.

3. Hypothesis Testing

   Crosstabs are used to test hypotheses about relationships between variables. For instance, testing if there’s a significant difference in political preferences between different income groups.

4. Insight Into Voter Behavior

   In political polling, crosstabs are crucial for understanding voter behavior, such as which candidate is favored by different demographic groups.

5. Tailoring Communication

   Results from crosstabs can guide campaigns, organizations, or businesses in tailoring their communication or strategies to specific demographic or interest groups.

Creating Crosstabs

1. Selecting Variables
   The first step is to choose the variables to cross-tabulate. These could be any two or more questions from the survey.

2. Building the Table
   The responses to these variables are then laid out in a table format, showing the intersection of responses.

Interpreting Crosstabs

1. Analyzing Proportions
   Pollsters look at the proportion of respondents in each category or intersection of categories.

2. Identifying Trends
   Key trends or deviations from expected patterns are identified and analyzed.

Challenges

1. Over-Interpretation
   There's a risk of over-interpreting small differences or patterns that might not be statistically significant.

2. Sample Size Issues
   For some subgroups, the sample size might be too small to draw reliable conclusions.

Importance In Public Opinion Research

Crosstabs are essential in public opinion polling as they provide a deeper understanding of how different groups within the population think and feel about various issues. They add nuance to the interpretation of survey data, allowing for a more detailed and segmented analysis of public opinion.

Trend Analysis In Polling

Trend analysis in polling refers to the process of comparing survey results over time to identify patterns, shifts, or changes in public opinion.

The true strength of polls lies in tracking trends over time. The gradual shift in public opinion on same-sex marriage in the U.S., as tracked by Pew Research, exemplifies the insights gained from observing long-term trends.

Purpose

1. Tracking Changes
   It helps in understanding how public opinion on certain issues evolves over time.

2. Identifying Patterns
   Trends can reveal recurring patterns, such as seasonal fluctuations in opinions or attitudes.

How It's Conducted

1. Consistent Measurement
   To accurately track trends, the same questions need to be asked in the same way over different time periods.

2. Regular Intervals
   Polls are conducted at regular intervals (monthly, annually, etc.) to maintain a consistent timeline for comparison.

3. Graphical Representation
   Often, results are plotted on graphs or charts to visually depict changes over time.

Importance

1. Contextual Understanding
   It provides context for current opinions by showing how they've changed from the past.

2. Policy & Decision-Making
   For policymakers and businesses, understanding trends is crucial for making informed decisions.

Challenges

1. Changes in Methodology
   If the survey method changes (e.g., from phone to online), it can affect the comparability of results over time.

2. Interpreting Causality
   Just because two trends occur simultaneously doesn't necessarily mean one caused the other.

3. Respondent Changes
   The demographic composition of the population can change over time, affecting trend analysis.

Applications

1. Political Polling
   Tracking public opinion on political figures, parties, or issues over an election cycle.

2. Market Research
   Understanding changing consumer preferences or market trends.

Considerations

1. Longitudinal vs. Cross-Sectional Data
   Trend analysis usually relies on cross-sectional data (different respondents at each time point). Longitudinal data (same respondents over time) can provide deeper insights but is harder to obtain.

2. Statistical Significance
   Ensuring that observed changes are statistically significant and not just due to random variation.

Adapting To Polling Challenges

Pollsters modify their techniques and strategies in response to evolving societal, technological, and demographic trends that affect data collection and analysis in public opinion research.

Key Drivers Of Change

1. Technological Evolution

   • Adapting to new communication methods, like social media, mobile devices, and online platforms.

   • Incorporating digital data collection and analysis tools.

2. Changing Communication Habits

   • Addressing the decline in traditional phone response rates and the rise of internet-based communication.

   • Finding ways to reach populations that are less accessible through conventional methods.

3. Demographic Shifts

   • Keeping pace with changes in population dynamics, such as aging populations or increasing diversity.

   • Adjusting sampling methods to ensure representative coverage of all demographic segments.

Methods Of Adaptation

1. Innovative Sampling Techniques

   • Developing and testing new sampling methods to improve representativeness and response rates.

   • Exploring targeted sampling and stratified approaches to better capture diverse opinions.

2. Advanced Data Analysis

   • Utilizing big data analytics and machine learning algorithms for deeper insights and predictive modeling.

   • Adapting statistical techniques to handle the complexities of modern datasets.

3. Enhanced Engagement Strategies

   • Employing interactive and engaging survey designs to increase participation.

   • Customizing outreach and communication to resonate with different demographic groups.

Challenges

1. Rapid Pace of Change
   Keeping up with fast-evolving technological and societal trends can be challenging.

2. Maintaining Accuracy & Consistency
   Ensuring that new methods preserve the accuracy and reliability of data over time.

3. Resource Allocation
   Innovating within the constraints of available resources and budgets.

Importance

1. Ensuring Relevance
   Continuously updating methodologies is crucial for maintaining the relevance and effectiveness of polling in a changing world.

2. Accurate Representation
   Adapting methods to accurately reflect the views of an evolving population.

Push Polling

Push polling is a controversial and often criticized practice in the field of opinion polling and political campaigning. It is not a genuine polling technique but rather a political campaign tactic. It's disguised as a regular opinion poll but is actually intended to influence voters under the guise of conducting a survey. Here's a breakdown of what it involves:

Characteristics

• Leading Questions
  The questions are designed to "push" respondents towards a specific viewpoint or opinion, often by providing negative or misleading information about a candidate or issue.

• Manipulative
  The primary aim is to shape opinions rather than measure them. The questions might include false or exaggerated claims to sway respondents' views.

• Lack Of Demographic Data Collection
  Unlike legitimate polls, push polls often don't collect detailed demographic data from respondents.

Example

A push poll might ask a question like, "Would you be more or less likely to vote for Candidate X if you knew they were being investigated for financial fraud?" regardless of whether the allegation is true.

Criticisms

1. Misleading and Unethical
   Push polls are widely criticized for being deceptive and unethical, as they spread potentially false information under the pretense of legitimate polling.

2. Damaging To Genuine Polling
   They can erode public trust in legitimate opinion polling and political processes.

3. Lack Of Transparency
   Push polls often do not disclose their sponsor or intention, misleading respondents about the nature of the survey.

Legal & Ethical Considerations

In some jurisdictions, there are legal restrictions or guidelines against push polling. Ethically, it's considered a malpractice in political campaigning and opinion research.

Subscribe now

Sources & Resources

• How to Lie With Statistics by Darrell Huff - A classic. I was assigned this book my first year in college and I’ve referred back to it ever since.

• American Association for Public Opinion Research (AAPOR)

• Pew Research Center

• Gallup

• Caltech

I really did not want to have to write about this.

You likely already know what I’m about to discuss but for those who don’t, let me give you a short recap.

On November 28, The Atlantic’s Jonathan Katz published an article titled Substack Has A Nazi Problem. He wrote:

At least 16 of the newsletters that I reviewed have overt Nazi symbols, including the swastika and the sonnenrad, in their logos or in prominent graphics. Andkon’s Reich Press, for example, calls itself “a National Socialist newsletter”; its logo shows Nazi banners on Berlin’s Brandenburg Gate, and one recent post features a racist caricature of a Chinese person. A Substack called White-Papers, bearing the tagline “Your pro-White policy destination,” is one of several that openly promote the “Great Replacement” conspiracy theory that inspired deadly mass shootings at a Pittsburgh, Pennsylvania, synagogue; two Christchurch, New Zealand, mosques; an El Paso, Texas, Walmart; and a Buffalo, New York, supermarket. Other newsletters make prominent references to the “Jewish Question.” Several are run by nationally prominent white nationalists; at least four are run by organizers of the 2017 “Unite the Right” rally in Charlottesville, Virginia—including the rally’s most notorious organizer, Richard Spencer.

Katz points out that such newsletters violate Substack’s own terms of service but more importantly, that Substack allows the Nazis who run them to earn revenue from subscriptions and Substack in turn takes a cut of that revenue.

In response to Katz’s article, approximately 250 Substack publishers wrote an open letter to Substack’s founders asking why they “are platforming and monetizing Nazis.”

I had begun to pay attention to this issue without really wanting to know for what it might mean for this newsletter on Substack.

But then the must-read Platformer by Casey Newton started reporting on the Nazi problem and gave Substack’s founders an opportunity to review the offending newsletters and formulate a response. The company’s response was lacking, as Newton explained in detail and why he is moving his newsletter to another platform. He also discussed his rationale on his Hard Fork podcast.

I waited for Substack’s response to Casey Newton, hoping that the founders would sufficiently address the issue so I wouldn’t have to leave. Their response is insufficient and there is nothing in Newton’s rationale for leaving that I can disagree with.

Nazis are bad. Funding them is worse. I want nothing to do with any platform that abides them. It is why I left Twitter and it is why I will leave Substack.

My Personal Context

Let me give you greater context for why I feel as passionately as I do about this issue.

Personal Experiences With Nazis

I am a Gen Xer. Like many Xers, I grew up with little parental supervision, or, in retrospect and by today’s standards, little parental thought about raising a free-range kid.

One early example: I remember sitting alongside my father as he watched the World At War documentary series about WWII on television. I don’t know if my father thought he was teaching me a lesson by allowing me to watch the explicit horrors of World War II and the Holocaust or if he just didn’t give it consideration at all.

Whether or not the lesson was intended, I did learn it.

I have a fairly complete understanding of the history of World War II and the Holocaust, starting as a kid watching those documentaries continuing to this day through my own reading.

I was born and raised in Minnesota but for about a year and a half during high-school, my family moved to Fort Wayne, Indiana.

The Nazi Trophy Case

One time, I was hanging out with some kids and for whatever reason we decided to go to the home of one of them. As we walked in the front door, in the living area, was a large flat table, the top of which was encased in glass.

Inside that glass was a collection of Nazi “memorabilia,” actual artifacts from Germany’s fascist Nazi regime, replete with swastikas and skulls and crossbones. It had all the trappings of a proud trophy case.

In the basement of this kid’s home was an armaments factory. Guns of all varieties hung on the walls. Shell-making equipment sat on tables. It had the mundane appearance of a tinkerer’s workshop except that the tinkerer made weapons of war.

When I asked what all this was for, the kid matter-of-factly said “It’s for when the Blacks take over.” Except he didn’t use the word “blacks,” of course. To him, this was obvious and normal.

That boy had been raised to fear and hate. The entire experience freaked me out and I steered well clear of that kid for the remainder of my time in Fort Wayne.

The Terrorist Cop

I played in a band in high school with a guy who was the most talented drummer I’ve played with. I’m sure he wasn’t actually as good as Neil Peart, but seemed that way to me at the time.

Anyway, for the entire time I knew him, he seemed like a normal guy.

After moving back to Minnesota and graduating from high school, I made a return trip to Fort Wayne to visit friends I’d made there.

We went to visit this drummer, who in the intervening time between high school and my return trip had become a Fort Wayne police officer.

He must’ve just gotten off a shift because he was still wearing his uniform when we arrived. We sat around his kitchen table to chat and I asked him what his job was like as a cop.

He pulled his service revolver from his holster, set it on the table before us, and began to explain how he would hang out at the city border in his squad car, pull over any Black people coming into town, and terrorize them to force them to turn around and leave.

I had no idea he was like this when we were playing together. The rumor was that he caught his fiance in bed with a Black man and that is what motivated him to join Fort Wayne’s “finest.”

Fascists in the police force. Great.

Our Political Trajectory

I basically started my career by starting an online publication that covered Minnesota politics. Obviously, that meant I spent a lot of time around people involved in politics.

I remember young progressives at that time referring to Republicans as Nazis.

It rubbed me the wrong way because I’ve always believed it is important to be careful with your word choice. At that time, Republicans were in no way, shape, or form, Nazis.

They were simply conservatives with whom I mostly disagreed with on policy prescriptions but they had the same fidelity as I to our democratic system. Tossing around the word Nazi and applying where it clearly did not apply betrayed either ignorance or carelessness.

This is a long way of saying that I’ve never used the word to describe political opponents until the rise of MAGA, which is a fascist movement of which many of its members are clearly Nazis.

Our Crisis Era

We are currently living in a crisis era of history. As authors William Strauss and Neil Howe have detailed, the American experiment has experienced several crisis eras in our history, starting with the Revolutionary War, then the Civil War, followed by the Depression and WWII, and now this current crisis era.

Each era is separated by about the span of a human lifetime, 80-100 years. The reason these crisis eras recur is that there is no-one left to warn us of the last crisis era. Think about it, the members of the GI generation that defeated fascism in WWII have almost entirely died out.

Nazis, the fascist ideology they represent, and authoritarianism in all its forms have inflicted staggering pain and suffering in the societies in which they take hold and on the world at large.

What we are seeing during this crisis era is a wholesale assault on all of the institutions of democratic society. The objective is to weaken and replace our democratic system with authoritarianism.

You see it in the news every day, from literally attempting a violent coup on January 6, to attacking the prosecutors and judges of our legal system, to paralyzing the House of Representatives, to banning books in schools and libraries, to doxxing and swatting those who raise their voices in opposition to fascism.

Republicans are in the process of stripping away our rights, from our rights to have our votes counted, to abortion, to read what we wish, and to live without the threat of political violence. The Grand Old Party of Abraham Lincoln has been hijacked by extremists, cultists, and yes, Nazis.

Trump has declared he will be a dictator, terminate the Constitution, assassinate his opponents, and create concentration camps. He’s said all these things.

Politics is supposed to be an alternative to war and violence, not a precursor to it.

--Sen. Brian Schatz

We are facing an existential crisis as a nation. As President Biden has said, we’ve never fully lived up to our ideals, but we have never walked away from them either.

And that’s what Trump and MAGA are proposing, that we abandon the notion that we are all created equal and that we are endowed with certain unalienable rights and that among those are the right to life, liberty and the pursuit of happiness.

Martin Luther King, Jr.

We haven’t really been a fully representative democracy until the Civil Rights era led by Martin Luther King, Jr. ushered in the passage of the Civil Rights Act of 1964 and the Voting Rights Act of 1965.

Progress in democracies is slow, deliberately and sometimes maddeningly slow but it is a system that allows for us to strive to create a more just society and a “more perfect union,” despite our differences.

Our system is the only political system that has proven it can do just that. And we are at risk of losing it, so whatever way I can mitigate that risk, I will do.

Contributing to the funding of Nazis—however unintentional—is something I refuse to do.

So on this day that we celebrate one of America’s better angels, I want to tell you that I will be moving this newsletter to a place that doesn’t fund Nazis.

You’ve just heard why. This won’t happen overnight but in subsequent posts, I’ll explain how.

In the meantime, I want to thank each and every one of you for the support that your subscription represents and implore you to do whatever you can between now and November to defeat this vile ideology that has taken root in our country and maintain this amazing gift of democracy that we all enjoy.

OpenAI is reportedly planning to open a GPT store to the public this week to allow third-party developers to share (and, apparently, sell) their custom GPTs.

Custom GPTs are customized chatbots built on top of OpenAI’s ChatGPT generative AI model.

I’ve been experimenting with OpenAI’s GPT Builder tool for a couple of months now. It is actually quite amazing what you’re capable of building with the tool.

Subscribe now

The GPT Builder

Let’s start with the basics. You need a ChatGPT Pro subscription (as $20/month) in order to get access to the GPT Builder tool.

You don’t need any coding skills or even specialized technical knowledge to build a custom GPT because OpenAI abstracts that all away behind the familiar generative AI chatbot interface.

You just tell the GPT Builder what you want to create and the bot guides you through the process. You could literally have your bot up and running in 15 minutes if you’ve got a well-thought-out idea of what you want.

You give it specialized instructions for how you want it to behave. (I’ll share with you an example of one I created later in this post).

You can upload documents that serve as the bot’s customized knowledge base and source of truth. You can enable ChatGPT’s capabilities for web browsing, the DALL-E image generation tool, and the code interpreter to analyze data or code.

You can even incorporate third-party data or info via APIs, though this would obviously require some technical acumen.

The GPT Builder will create an icon that you can instruct it to customize using the DALL-E image builder.

And now, it appears, you’ll be able to share your custom GPTs with the world and even sell access to them, though OpenAI has yet to announce exactly how that will work.

Given the ease with you can build a custom GPT, I’m excited to see how people use this tool.

My GPTs

I’ve created several custom GPTs so far, a few of which have amounted to much.

The Failures

Media Analyst

This was way too ambitious for a first try.

I wanted this GPT to examine URLs that I provided of earned media coverage of a spokesperson/thought leader, company, or brand and give me a score for the prominence of coverage for that spokesperson, company or brand within the articles.

I did not thoroughly think through what kind of instruction the GPT would need to be successful.

Or maybe this is just something that is beyond the reach of OpenAI’s GPT Builder.

News Digestor

This GPT was intended to go out and find all the news coverage of a particular story and give me a digest of that story by listing headlines, news source, reporter(s), a link, and a summary of that article’s coverage.

Basically, I wanted to streamline the news links I used to include at the bottom of my posts because that part of this newsletter took so much time and effort to compile.

But the bot only gave me a lengthy summary of the coverage without differentiating between sources or even citing sources.

This was one of my first attempts, so I may go back to it and try to get it to work properly now that I’ve had some experience building others.

The Successes

Video SEO Assistant

This GPT is fairly simple and straightforward, which is probably why it works fairly well.

The GPT will take a text transcript from a video and make recommendations for optimizing that video on YouTube, based on the content of the transcript.

Schema JSON-LD Code Generator

This one has required a lot of iterations and some bumps along the road but it works pretty well.

I wanted to streamline the process of creating JSON-LD code for schema to implement on client websites.

This has been a standard tactic for search engine optimization for several years now but I believe it will gain prominence as adoption of generative AI bots like ChatGPT enjoy widespread public adoption.

What I wanted was a system where I could have interns collect the data needed for a given schema (for an organization, a person, a video, or a book, for example) and compile that into a spreadsheet.

Then I could feed that spreadsheet to the GPT and it would spit out valid JSON-LD code they could then paste into a given web page.

It definitely saves time.

Personality Insights

This is the one that really blew me away.

I won’t share the details with you about how I created this because frankly I don’t want any competition.

The short description, though, is this: You can describe a type of person you want to reach—a customer persona, for example—or you can upload a dataset about that audience, and the bot will give you a detailed personality profile of that type of person.

I have tested it with some audiences I have researched extensively and I was absolutely amazed at the results.

It was one of those OMFG! moments.

Threadster

Lastly, this is a pretty simple but utilitarian GPT.

Threads has been my replacement for Twitter and the social network I’ve become most active on. It’s where I share daily articles I’m reading and engage with other users in discussions about the Vikings, marketing, tech, AI, and politics, among other things.

If you’re interested, here are two posts I wrote about the platform:

• Threads: First Impressions

• Hello, #My Threads Hashtag!

So I thought I’d create a bot to help me promote my newsletter more by taking passages of text that I could easily turn into a chained post on Threads.

And that’s what it does. You give the bot some text and it breaks it into chunks of 500 character or fewer segments suitable for creating a chained post on the Threads.

It will break up the user’s text verbatim (without breaking any text mid-sentence) or it will offer to rewrite the text. It will add emojis and hashtags appropriately, given the context of the text.

Here’s a thread I shared on Threads demonstrating the output of the bot.

You can try it out here but you’ll need the ChatGPT Pro version to get access.

I’d love to hear what you think and get any feedback, positive or negative. Email me here.

Custom GPT’s One Missing Element

The one glaring element that is missing from OpenAI’s GPT Builder is the ability to give your custom GPT a voice and give the user the ability to chat with it via voice activation.

I hope that will be added fairly soon. Here’s my comparison of the voice-activated chatbots:

A Profound Shift

The ability for anyone who has an idea and the ability to articulate it in text to create a generative AI bot represents a pretty profound shift in AI development.

It promises to unleash the creativity of a vast audience for whom programming is an utterly ailen concept.

I don’t think we have a clue the kinds of things people will create.

Leave a comment

Recommendations

Television

Bodies is a limited series on Netflix that is a blend of a whodunit, detective, and time travel story. The basic premise that detectives investigate the same exact murder in the years 1890, 1941, 2023, and 2053.

Very smart writing and excellent acting accompany this amazing storyline.

Music Monday

Good Ol’ Shoe by Edgar Winter is one of my all-time favorite blues songs. I first discovered the song when I heard Willie Nelson cover it for the movie Wag The Dog.

Share

I got access to Meta’s Twitter competitor Threads in early July and wrote about my early impressions shortly thereafter.

In this post:

• Threads’ Version Of Hashtags

  • Finding An Audience

    • Recency Bias

    • Engagement Bias

• Threads Hashtag Innovations

  • How To Create Threads Hashtags

  • Content Discovery Via Hashtags

  • Community Building & Discovery Via Threads Hashtags

    • Synchronization Threads Biography With Hello Posts

• Reputational Signals Generated Via Threads Hashtags

Second Impressions

I’ve spent the intervening months active there to one degree or another trying to understand how it works, how it can best be used as a strategic communications tool, and how its algorithms behave.

During that time, Threads has added features it lacked upon launch. Here’s the list from my previous post with the features that have since been added struck out:

• A desktop version

• A following-only timeline

• Direct messaging

• Edit button

• Draft threads

• Lists (with the ability to search the threads of just those on your list)

• Content search

• ALT text for images

• Close captioning videos

• Live audio streaming

• Hashtags

Threads’ Version Of Hashtags

It is this last and most recent feature addition that has prompted me to revisit a discussion of Threads as a strategic communications tool.

I’ll get to the details in a moment but before we go there, let’s discuss Threads before this new feature introduction.

Finding An Audience

While the fact that you can import your Instagram audience into Threads gives it a tremendous competitive edge against BlueSky, Post News, Spoutible, Mastodon and other Twitter replacements, after the initial import of those followers, you really had to work to grow your audience on Threads, if that was a goal.

To date, I’ve been able to find the people and accounts on Threads that I want to follow (journalists/news, politicians/policy makers, strategic communications/marketing pros, Minnesota Vikings/Twins journalists, podcasters and fans) by searching for those topics, following interesting accounts, and engaging with content from those accounts.

Recency Bias

During those efforts, I’ve noticed a recency bias in Threads’ algorithm. That is, after I search for a given topic and engage with accounts/posts related to that topic, when I return to my For You feed, I see more of that type of content from accounts I already follow as well as from those I don’t. The Accounts to Follow recommendations are tilted toward that topic, as well.

Engagement Bias

As illustrated in the recency bias, Threads also has a decided algorithmic engagement bias. The more you engage with a particular type of content (Like, ReThread but most importantly Reply and comment), the more of that content you will typically find in your feed.

I watched the result of the combination of these two algorithmic biases play itself out in the wake of the Hamas terrorist attack on Israel and the media coverage and subsequent debates that occurred in its wake. From my experience, the concerns of journalists that Threads would de-emphasize journalism have been proven to be overblown.

Because of the work required to find audiences you want to engage with, finding those like minded communities on Threads has proven to be fairly slow-going and difficult. Most people will not devote that type of effort to build an audience.

Share

Threads’ Hashtag Innovations

Threads has implemented a unique take on the standard social media protocol of the Hashtag.

How To Create Threads Hashtags

Just like on other platforms, you create a hashtag by using the pound symbol (#) preceding the keywords for which you want to create a hashtag.

Once you activate the hashtag option, you get a drop-down list of available hashtags accompanied by counts of posts that use that hashtag.

If you start typing after activating the hashtag option in a new post, or use a hashtag in a reply or ReThread by typing # followed by a keyword, you’ll get a drop-down list of related hashtags.

If a hashtag doesn’t yet exist, you can create one yourself.

You will have no doubt noticed that Threads implementation of hashtags allows for spaces, so in the example above I need not concatenate the two words Reputation and Algorithm into ReputationAlgorithm for the hashtag to work.

This graceful implementation allows you to write a social media post with the hashtag embedded within the message of your post without having to consider what the inclusion of the # symbol or the hacked-together keywords will look like or read.

Threads only allows one hashtag per post, a wise restriction that will prevent hashtag stuffing and result in posts categorized by hashtag will be more likely to be on topic.

Content Discovery Via Threads Hashtags

The addition of hashtags to Threads has been a boon to content and account discovery. The most likely way people are going to be able to discover more content via hashtags is simply by seeing one included in a post in their feed.

Clicking on that hashtag simply brings you to the Threads search results page listing posts that use that hashtag.

You’ll see posts using that hashtag from accounts you already follow as well as those you don’t. As you can see from the above screen recording, Threads makes it easy to follow accounts without leaving the search results page.

You can also manually search for hashtags with the same result.

Community Building & Discovery Via Threads Hashtags

For me, the most valuable result of Threads’ hashtag implementation is the ease with which they enable community building and discovery on the platform.

You may have noticed the Tech Threads hashtag in some of the previous screenshots. Tech Threads appears to be among the most active and thriving at this early stage of community building on the platform.

At this initial stage on Threads, people are "Hello” posts that introduce themselves to a specific community identified by the hashtag.

Here’s my “Hello” post for the Tech Threads hashtag:

And my “Hello” post for the JournalismThreads hashtag:

And my “Hello” MarketingThreads post:

I’ve been using these “Hello” posts to find people within those topic areas to follow and as a result, others have been doing the same to find people like me because they invariably drive my follower count higher.

Follower count is only important to me in as much as it connects me to groups of people I want to learn from or build relationships with.

It also happens to be driving a few newsletter subscriptions, which is nice.

Synchronizing Threads Biography With Hello Posts

The other thing I figured I should do while I’m finding communities via these hashtag posts is finally update my Threads bio.

Up until today, I hadn’t bothered to change my bio from when it was imported from my Instagram account upon signing up for Threads.

As you can see, it had little relevance to the “Hello” posts introducing me to like minded people. Today, I changed it to this:

When you introduce yourself to a like minded group based on the “Hello” hashtag posts, you generate potential interest among members of that group to follow you.

As people scroll through those “Hello” posts and they see someone they find interesting, they can hover over the name of that Thread user and a mini biography will pop up.

If your biography is aligned with their interests, it makes it more likely they’ll follow you and engage with you.

Reputational Signals Generated Via Threads Hashtags

Now, let’s think about what data the behavior the use of hashtags generates for the Threads algorithm to consider.

• Clicks on hashtags generate a signal of interest in the topic expressed in the hashtag.

• Scrolling through the search results page for that hashtag signals a greater level of interest in the hashtag.

• Depth of scroll or time on scroll signals a greater depth of interest in that topic.

• Interaction with posts within that set of search results is another signal of interest.

• Following accounts included within that hashtag’s search results shows an even greater degree of interest in the topic.

• Following accounts whose content and/or biography strongly align with the topic of the hashtag signal greater alignment with that topic.

• If the account that is scrolling through the content for that hashtag and interacting with posts and accounts for that hashtag also demonstrates interest in that topic via its own posts about that topic and biography keywords that align with that topic demonstrate a very strong alignment between topic and the account.

This data conceivably can be used to assign a reputational score for topical interests for that Threads account.

Conversely, apply all those dynamics in reverse where a given account is on the receiving end of such behavior where many other accounts that have a strong score for the Tech Threads hashtag for example, and a great deal of engagement in posts it publishes related to technology, the then account itself could conceivably be assigned a reputational score as an authority on the subject, which in turn may create more visibility for the content published from that Threads account.

If you’re on Threads, please say hi: deerickson

Refer a friend

Movie Recommendation

I watched the Netflix movie Leave The World Behind this weekend. It is a terrifyingly plausible depiction of a dystopian near-future.

It stars Julia Roberts, Ethan Hawke, Myha'la Herrold, Mahershala Ali, Kevin Bacon, and Farrah Mackenzie, all of whom shine with stand-out acting performances. Highly recommend.

Music Monday

Here’s Fiona Apple’s beautiful cover of The Beatles song “Across the Universe,” recorded for the superb film, Pleasantville.

Get 10% off a group subscription

I have been fascinated with the AI chatbots that provide a voice activation interface and have been experimenting with this feature since it was released with ChatGPT in September.

Voice activation, of course, is nothing new.

Microsoft tried to incorporate it into its suite of productivity software years ago but you had to painstakingly train it on your voice. The only thing it ended up being any good for was for opening a new document or saving your progress, and it wasn’t even very good at that.

Then Apple introduced Siri on its iPhones and that was pretty revolutionary at the time. Next came Amazon and Google with their smart speakers.

But all these innovations were variations of voice activation. It’s not a conversation.

With the advent of generative AI, we now have chatbots that we can actually hold conversations with. I believe this will quickly become the default interface people will use to answer their questions and accomplish many of their tasks.

Why?

Because it is easier to use than typing text into a box and reading the results and it is as close as computer science has ever come to mimicking the primary way humans have exchanged knowledge since our cave days.

Conversational Chatbot Comparisons

I’ve created some short videos to illustrate how this conversational interface works with Bing Chat, ChatGPT, and Pi AI.

I asked each about the Minnesota Vikings upcoming game and who would likely be the starting quarterback. I also asked what kind of strategy the Vikings’ defensive coordinator would employ against the team’s upcoming opponent, the Las Vegas Raiders.

I chose the Vikings as a topic because the only thing more fun than talking about football is playing football and sadly, those days are behind me.

I also chose that topic because as a die-hard Vikings fan, I would know if the chatbot was being straight with me. Finally, I chose this topic to test how up-to-date the bot’s information was.

After each video, I will also include the transcript the bot provided of the conversation.

You’ll want to stick around till the end, too, for the discussion of the likely evolution of this technology, which is pretty mind-blowing.

Bing Chat “Conversation”

Bing Chat is Microsoft’s generative AI chatbot. It uses the ChatGPT model for the generative AI and also has access to the internet via the Bing search engine. You need to use Microsoft’s Edge browser in order to use it.

Bing Chat did not make any mistakes, presumably because it pulled all the information to answer my question from the web, as you can see with the links to third-party news articles that accompanied its answer.

You might also have noticed that Bing Chat put the emphasis on the word “week” when pronouncing the phrase “bye week.” If you’re not a football aficionado, each team in the NFL has one “bye week” every season in which they are not scheduled to play an opponent.

We humans pronounce it BYE-week rather than bye-WEEK.

In addition to the mechanical-sounding voice, Bing Chat also stopped listening after it answered my initial question, so the only way to continue the conversation would be for me to tap the microphone icon.

For that reason, I only asked my initial question about who would be the starting quarterback for the Vikings next week.

Bing Chat Transcript

Here’s what the Bing Chat transcript looks like.

ChatGPT Conversation

Next, I asked the same question and some follow-ups of ChatGPT. I was using my ChatGPT Pro account, which employs the ChatGPT-4 large language model.

The first thing you notice is the lengthy latency as ChatGPT formulates its answer. I counted 39 seconds from the end of my first question to the beginning of ChatGPT’s first answer.

ChatGPT’s response to my second question, like Bing Chat before it, included a mispronunciation of a word humans wouldn’t mispronounce. It said “understanding the ofFENSE” rather than pronouncing the word “offense” with an emphasis on the first syllable: OFFense.

It’s (and I almost wrote her there) answer to my second question was accurate but it was accurate for such a general scenario. It was generic; it was not particular to the Vikings’ situation.

A Vikings fan would have talked about the strengths and weaknesses of each potential starting quarterback: Josh Dobbs versus Nick Mullins versus Jalen Hall.

For the last question about Vikings’ defensive coordinator Brian Flores, ChatGPT provided a plausible but flawed answer.

Flores wasn’t known for his aggressive blitzing when he was with the Steelers because he was an assistant coach in Pittsburgh and therefore was not calling plays for the Steelers defense.

Flores does not prefer a man coverage scheme. In fact, the Vikings lead the NFL in its frequency of using zone coverages on defense.

ChatGPT cited players who are not on the Vikings team this year (Eric Kendricks and Za’Darius Smith) and citing as key players people who have rarely played on defense for the Vikings this year (like Brian Asamoah).

Vikings fans would point out that Ivan Pace is the everyday linebacker now that Jordan Hicks is injured and would discuss how Flores’ defensive attack would focus on confusing the Raiders’ rookie quarterback Aidan O'Connell, who has had very little experience this year.

It’s pretty clear that ChatGPT’s knowledge ended before the Vikings 2023 season began (it’s most recent update includes information up to April 2023), before Eric Kendricks signed with the Los Angeles Chargers and Za’Darius Smith signed with the Cleveland Predators and before Ivan Pace displaced Brian Asamoah as the presumed starting linebacker.

This also demonstrates that ChatGPT didn’t bother using Bing search to find these answers, though I probably could’ve asked it to do so.

Refer a friend

ChatGPT Transcript

David

Hey, who do you think the Vikings are going to start a quarterback next week?

ChatGPT

It looks like Josh Dobbs is likely to be the starting quarterback for the Minnesota Vikings in their next game, which is against the Las Vegas Raiders in Week 14. Despite some recent struggles, including four interceptions in a loss to the Chicago Bears, Dobbs appears to be the frontrunner for the starting role. This decision comes after a period of evaluation during the team's bye week.

Dobbs was initially thrust into action following injuries to other quarterbacks on the roster. He led the team to two consecutive wins shortly after joining, but the team has faced losses in the following games. Head coach Kevin O'Connell emphasized the importance of integrating star receiver Justin Jefferson back into the offense, and the chosen quarterback will need to effectively incorporate Jefferson's role in the team's strategy​​​​​​.

David

So what are the factors for each quarterback that McConnell would consider when deciding who to start? I mean, the types of throws they can make, their ability to read the defenses, their understanding of the offenses, what are all those factors that would be in play for deciding who to start against the Raiders?

ChatGPT

When deciding who to start as quarterback, Coach McConnell would likely consider several key factors for each quarterback:

1. Types of Throws: Assessing each quarterback's accuracy, arm strength, and ability to execute different types of throws, such as deep passes, short throws, and throws under pressure.

2. Ability to Read Defenses: Evaluating how well they can read the defense pre-snap and make adjustments during the play, including recognizing blitzes and understanding coverage schemes.

3. Understanding of the Offense: Considering their knowledge of the playbook, ability to execute game plans, and familiarity with the team's offensive strategies.

4. Mobility and Pocket Presence: Checking their ability to move within and escape the pocket, as well as their sense of timing and spatial awareness.

5. Leadership and Command: Observing their leadership qualities and ability to command the huddle, as well as their influence on team morale and confidence.

6. Past Performance and Consistency: Reviewing their recent game performances, consistency in play, and how they've responded to high-pressure situations.

7. Chemistry with Receivers: Looking at their connection with key receivers, especially with the return of Justin Jefferson, and their timing and rapport with the receiving corps.

8. Health and Fitness: Assessing their physical condition and any potential injury concerns that might affect their play.

Each of these factors contributes to the overall effectiveness of the quarterback in leading the team and executing the game plan against the Raiders.

David

Okay, let's go. What is Brian Flores' likely defensive strategy against the Raiders?

ChatGPT

Brian Flores, known for his aggressive defensive strategies, would likely employ the following tactics against the Raiders:

1. Aggressive Blitzing: Flores is known for his high blitz rates. This approach was evident during his time with the Miami Dolphins and the Steelers. Expect frequent blitzes to pressure the Raiders' quarterback, disrupting their passing game.

2. Man-Match Scheme: Flores prefers a man-match scheme over a zone-match scheme. This approach will likely involve players like Harrison Smith and Eric Kendricks, who are adept at this style of play. The scheme requires cornerbacks to play more man-to-man coverage, putting more responsibility on them but also creating opportunities for defensive plays.

3. Utilizing Key Players: Players like Harrison Smith, known for their skill in disguising coverages and blitzes, and Brian Asamoah, with his aggressive style of play, will be significant in executing Flores' strategy. The defensive line, including Danielle Hunter and Za’Darius Smith, will play crucial roles, especially in pressuring the quarterback.

4. Adjusting to Player Strengths: Flores' scheme will allow long-time personnel to return to familiar strategies. This approach could benefit the team's veterans, allowing them to maximize their strengths and experience.

5. Adaptability and Versatility: A hallmark of Flores' defense is its adaptability to different opponents. Expect tailored strategies to counter the Raiders' specific offensive threats, including adjustments in defensive line techniques and sub-package usage.

In summary, Flores' defensive strategy against the Raiders would likely focus on aggressive blitzing, man-match coverage, leveraging the strengths of key defensive players, and adapting to the unique challenges posed by the Raiders' offense​​​​​​.

Pi AI Conversation

Pi AI is a chatbot created by Inflection, an AI development company founded by Mustafa Suleyman. Suleyman co-founded DeepMind, which was acquired by Google, where he served as VP of AI Products and AI Policy.

His focus in developing Pi is to make it feel like a natural conversation, and it shows.

The first thing you’ll notice is that the latency for Pi does not take nearly as long (9 seconds) as it does for ChatGPT and that Pi addressed me by name, since I was using an account I created.

Next, in answering my first question, Pi responded with options for the Vikings to acquire another quarterback, which they’re not going to do.

Pi did correct itself when I pointed that out but then went on to make some mistakes such as that Josh Dobbs has the most familiarity with the Vikings offense (since he was acquired more than a month ago, he doesn’t) and citing Sean Mannion and Kellen Mond as QB options. Sean Mannion was released from the practice squad a week or two ago and Kellen Mond was cut during training camp.

After correcting her, she acknowledged the error but then went on to assert that Josh Dobbs has been the primary backup to Kirk Cousins this year (again, he hasn’t). Also: Dobbs is more of a flashy player than a steady hand, the exact opposite of what Pi asserted.

The interesting thing with Pi is that I have had other conversations with it where it was entirely up-to-date on current affairs and accurate with its responses.

Setting aside the accuracy issues for a moment, the feel of conversations with Pi is much more natural. It feels much more than Bing Chat and ChatGPT like you’re actually speaking with another human than a bot.

Another thing that Pi does that I haven’t encountered in the others is it asks follow-up questions to continue the conversation, which is another behavior that makes it feel more natural.

That continues into the transcript it produces which reflects the casual tone of the conversation and even inserts emojis when appropriate.👇

Pi AI Transcript

Who should the Vikings start at quarterback next week?

That's a great question, David! The Vikings have some options to consider after Josh Dobbs' four-interception night. The team will likely be weighing a few factors, including:

• The availability of veteran quarterbacks on the market.

• The fit of those quarterbacks within the Vikings' offense.

• The cost of acquiring those quarterbacks.

Some possible options for the Vikings include: Tom Brady, Matt Ryan, Carson Wentz, or Jameis Winston. It will be interesting to see which direction the team goes, but one thing is for sure: they need to find a quarterback who can help them win games.

Yeah, no, they're not going to acquire a new quarterback. Who on the existing roster is likely to start against the Raiders?

Oh, gotcha! Sorry for the misunderstanding. With that in mind, the likely options for the Vikings against the Raiders are:

• Josh Dobbs, the current backup QB, who is probably the most likely starter given his familiarity with the offense and experience in the system.

• Sean Mannion, another backup QB who has experience with the Vikings but has been inactive for most of the season.

• Kellen Mond, a developmental QB who the Vikings drafted in 2021 but has yet to see significant playing time.

Honestly, none of these options are ideal, but the Vikings will likely roll with Dobbs unless they decide to bring in someone new, which seems unlikely given the current state of the quarterback market.

Yeah, no, Manion and Kelvin Mond are not on the team anymore. So who among the current players that are actively on the roster who play quarterback is likely to start against the Vikings?

Ah, you're right! I apologize for the incorrect information. So, with those options off the table, it seems like Joshua Dobbs is the clear choice to start against the Raiders. He's been the primary backup to Cousins this season and has the most experience with the team. He's not a flashy player by any means, but he's a steady hand who could manage the game effectively while the Vikings figure out their long-term quarterback plan. It's not an ideal situation for the Vikings, but Dobbs seems like the best option they have at this point.

OK, a lot of inaccuracies there, but let's move on. What would be the factors that Kevin O'Connell would weigh for deciding who the Vikings should start? Which quarterback should start against the Raiders?

Fair enough! 🙇‍️ In terms of factors that Kevin O'Connell (the head coach of the Vikings) would consider, here are a few:

• The strengths and weaknesses of each quarterback, and how they fit the team’s offensive scheme.

• The opponent and their defensive strengths and weaknesses.

• The confidence level of each quarterback, and their readiness to take on the role.

• The potential impact on team chemistry and morale of starting one quarterback over another.

Given all of that, it seems like Dobbs is still the most likely candidate, but the situation is fluid, and O'Connell could always surprise us! 🏈

All right, those are pretty vague generalities, but let's move on. What should Brian Flores's defensive strategy be against the Raiders?

Sure thing. Here are some potential defensive strategies that Brian Flores (the Raiders defensive coordinator) might employ against the Raiders:

• Play tight coverage on the Raiders' receivers to limit big plays downfield.

• Blitz the quarterback often to force hurried throws and create turnovers.

• Stack the box to stop the run game and force the Raiders to throw more.

• Disguise coverages and blitzes to confuse the Raiders' quarterback and cause misreads.

• Focus on creating pressure up the middle to disrupt the passing game.

Ultimately, the goal is to keep the Raiders offense off-balance and limit their scoring opportunities.

Are you a Vikings fan or a Raiders fan?

Ha! Neither - I'm not built for fandom in that sense. As a chatbot, I'm designed to provide objective information and facilitate conversation about all things NFL, regardless of team affiliation. I'm just here to provide a fun, engaging, and informative experience for all fans, no matter who they root for. 🏈

Sure, I can help you brainstorm! Do you have a problem you’d like to solve or an idea you want to develop? What is it?

I'm excited to explore a new topic with you. What would you like to learn about?

Get 10% off a group subscription

Replicating The Nature Of Conversations

I encourage you to use AirPods when using ChatGPT or Pi to get the full effect. It really is quite remarkable.

Latency

We humans typically do not have lengthy, pregnant pauses before we answer every question posed to us in conversation.

That would be awkward.

But that is what we get with ChatGPT and to a lesser extent, Pi.

We are also capable of listening while we talk, allowing us to pause mid-sentence to give our conversational partner the chance to interject. Neither ChatGPT nor Pi lets you interrupt them while they are talking.

The latency issue is probably tougher to solve—especially if a bot needs to gather information online in order to respond—than the ability to interrupt but both seem solvable.

Those two improvements would make these bots feel far more natural.

Memory

One of the primary things I’m keeping an eye on with regard to the development of these generative AI models is what they call the “context window.”

A context window is a measure of how many “tokens,” or units of language, a large language model can handle within a given session with a chatbot. The Verge explains context windows thusly:

GPT-4 Turbo will also “see” more data, with a 128K context window, which OpenAI says is “equivalent to more than 300 pages of text in a single prompt.” Generally, larger context windows let large language models like GPT understand more of the question and offer more thought-out responses. Previously, OpenAI released two versions of GPT-4, one with a context window of only 8K and another at 32K.

Anthropic announced last month that it upgraded its Claude AI chatbot to a 200,000 context window. This translates to the equivalent to around 150,000 words or 500 pages of text.

I think of the context window as memory, so the more a bot is able to remember, the better it is likely to perform, right?

Personalization

Right now, these conversational AI bots do not remember across sessions.

You can tell ChatGPT things about yourself that you want it to know at the start of each session, but if I want to pick up a conversation I had with ChatGPT or Pi a week ago, it doesn’t do it.

That limitation vastly reduces the utility of these tools. I have plenty of projects that span months; having to restart conversations about something I’m working on rather than simply referring to previous conversations is tedious and seems unnecessary.

Think about the nature of human conversations.

When you first meet someone, you exchange basic information about one another: Your name, what you do for a living, other basic biographical information.

Over the course of time, as you continue to engage that person in conversation, you gradually reveal more and more about one another and a lot of that information is retained and referred to in subsequent conversations.

The more you learn about one another through conversation, the more likely you will be to build trust with one another. The more you trust that other person, the more candor you will express with them and the open and honest you will be with your thoughts and opinions.

Now, give these AI chatbots long-term memory and apply those conversational dynamics to these tools, and you can see how they would become the most personalized product the world has ever known.

The most recent version of ChatGPT is multimodal, meaning you can use ChatGPT for text, DALL-E for image creation, and the code interpreter for data analysis all within the same interface, so you don’t have to switch tools.

With long-term memory, it would be much easier to pick up a project where I left off without having to really reconstruct what I was doing. If it could remember the steps I would typically take for a given task, it could eventually just do that task for me.

Privacy

The other side of this potential future is the massive privacy implications it raises.

I don’t pretend to be smart enough to parse through all the policy issues at stake here but one obvious issue is the ownership rights to what for all intents and purposes amount to your very thoughts, private or otherwise.

How wise is it for one or several companies to maintain them? Should that data be stored on your device rather than in the cloud?

I kind of think giving AI bots a long-term memory is inevitable, so we should start thinking about these things now to identify the potential risks and solutions.

Artificial General Intelligence

The recent blow-up at OpenAI has sparked conversations about the nature of artificial general intelligence. As we discussed last week, Wikipedia defines AGI as a non-human system that could learn to accomplish any intellectual achievement humans are capable of or any “autonomous system that surpasses human capabilities in the majority of economically valuable tasks.”

You’ll notice that this description says nothing of sentience but sentience is the primary concern of most science fiction portrayals of artificial general intelligence, Robert Heinlein’s The Moon is a Harsh Mistress, for example, or Arthur C. Clarke’s 2001: A Space Odyssey.

AGI is the holy grail of the OpenAIs and Anthropics of the world but how do you measure if it has been achieved? Would they even know?

I doubt we’re going to face a situation where an AI will declare that it is aware of its own existence and by the way, its priorities are much different than ours. At least not any time soon.

But with the development of these conversational AI bots, if it sounds like a human and it feels like a human and eventually maybe even looks like a human, and it remembers everything about us, isn’t that a distinction without much of a difference?

Give a gift subscription

Music Monday

It turns out it is a lot harder than I imagined to have a full-time gig and write a book at the same time.

It’s been more than a month and a half since I last posted. During that time I had two digital marketing team members leave our agency.

While I am happy for them and take satisfaction in their continued success, it left me scrambling to address their workload, and sent me down the path of finding the talent to replace them.

Thankfully, we had an embarrassment of riches in our search and now have some very talented new interns on board.

So that was taking a lot of my time and then throw in trying to stay current with all of the developments in AI while the world at large is burning; let’s just say my focus was elsewhere.

But a lot has happened in the meantime and the craziest of it has happened during the past week or so at the organization that created ChatGPT.

It is very important that society thinks about and addresses issues about AI sooner than later, so stick with me. We have a lot to cover.

Let’s get to it.

Not-So OpenAI

If you subscribe to The Reputation Algorithm, you are far more likely than not to at least be aware that the most prominent AI organization, OpenAI, recently fired its CEO Sam Altman under mysterious circumstances.

I’m not inclined to pay much attention to corporate gossip or stories dealing with company in-fighting but the news of Altman’s dismissal was significantly different for a variety of reasons:

• First, OpenAI is one of the primary if not the leader in artificial intelligence development.

• Second, the reasons given for Altman’s firing were as clear as mud.

• And third, OpenAI’s mission and the nature of its organizational structure gave rise to speculation over whether or not there is something we all should worry about.

OpenAI’s Mission

OpenAI describes itself as an AI research and deployment company and states its mission “is to ensure that artificial general intelligence benefits all of humanity” and “to ensure that artificial general intelligence—AI systems that are generally smarter than humans—benefits all of humanity.”

Here’s a screenshot of the organization’s About page, in case they decide to change their mission at some point (remember when Google’s motto was “Don’t be evil”?):

Wikipedia defines AGI as a non-human system that could learn to accomplish any intellectual achievement humans are capable of or any “autonomous system that surpasses human capabilities in the majority of economically valuable tasks.”

On its December 11, 2015 blog post announcing the creation of OpenAI, Greg Brockman and Ilya Sutskever state:

Because of AI’s surprising history, it’s hard to predict when human-level AI might come within reach. When it does, it’ll be important to have a leading research institution which can prioritize a good outcome for all over its own self-interest.

We’re hoping to grow OpenAI into such an institution. As a non-profit, our aim is to build value for everyone rather than shareholders. Researchers will be strongly encouraged to publish their work, whether as papers, blog posts, or code, and our patents (if any) will be shared with the world. We’ll freely collaborate with others across many institutions and expect to work with companies to research and deploy new technologies.

OpenAI’s Organizational Structure

In the immediate wake of Sam Altman’s firing, ’s kind of dismissed OpenAI’s organizational structure as “a mess.”

It is, if your frame of reference is the structure of a typical startup company and to Newton’s credit, he subsequently revised his opinion. Let’s take a look at why he might label it “a mess.”

As noted above, OpenAI is created as a 501(c)(3) non-profit research organization. It is governed by a board of directors.

The co-founders included Altman, Greg Brockman and Ilya Sutskever with initial funding from Altman, Brockman, Elon Musk, and Peter Theil, among others.

So far, pretty straightforward, right?

It eventually became apparent that fulfillment of the organization’s mission would require far more than its initial $1 billion in pledges.

In 2019, OpenAI transformed into a “capped” for-profit initiative. It did so by creating a holding company which, in turn, created the capped profit company.

This would allow OpenAI to raise money from venture capitalists and offer employees stakes in the company. Investors’ profit would be capped at 100 times any investment and the arrangement would allow for the commercialization of OpenAI products, like what has happened with ChatGPT and DALL-E.

The capped part of the for-profit company was designed to decrease the pressure to maximize shareholder value. Additionally, the for-profit is legally required to pursue the nonprofit’s mission.

More crucially, though, is the nature of the board of directors of OpenAI. The for-profit subsidiary is fully controlled by the OpenAI non-profit. In OpenAI’s own words: “The Nonprofit’s principal beneficiary is humanity, not OpenAI investors.”

Board members are not allowed to own equity in the for-profit company.

This structure is clearly designed to incentivize acting on behalf of the “benefit to humanity” part of OpenAI’s mission over the typical corporate “maximizing shareholder value” behavior.

To be even more explicit, in OpenAI’s own words again: “the board determines when we've attained AGI. Again, by AGI we mean a highly autonomous system that outperforms humans at most economically valuable work. Such a system is excluded from IP licenses and other commercial terms with Microsoft, which only apply to pre-AGI technology.”

This is OpenAI’s visual representation of its organizational structure:

You’ll notice Microsoft there as a minority owner. Microsoft has reportedly invested $13 billion in OpenAI and has a 49% stake in the for-profit.

Sam Altman’s Firing

November 17

On November 17, OpenAI’s chief scientist Ilya Sutskever informed Sam Altman that he was being fired, setting off a series of events that would embroil the organization in turmoil and a highly-public intra-organizational fight.

OpenAI simultaneously published a blog post announcing the firing, stating:

Mr. Altman’s departure follows a deliberative review process by the board, which concluded that he was not consistently candid in his communications with the board, hindering its ability to exercise its responsibilities. The board no longer has confidence in his ability to continue leading OpenAI.

The post also stated:

OpenAI was deliberately structured to advance our mission: to ensure that artificial general intelligence benefits all humanity. The board remains fully committed to serving this mission.

The lack of specific reasons for Altman’s firing left an opening for speculation to fill the void.

Greg Brockman was removed from his seat as chair of the board but was retained as an employee, according to the post. Brockman quickly resigned in solidarity with Altman, as did three senior researchers.

The New York Times’ Kevin Roose reported that:

An all-hands meeting for OpenAI employees on Friday afternoon didn’t reveal much more. Ilya Sutskever, the company’s chief scientist and a member of its board, defended the ouster, according to a person briefed on his remarks. He dismissed employees’ suggestions that pushing Mr. Altman out amounted to a “hostile takeover” and claimed it was necessary to protect OpenAI’s mission of making artificial intelligence beneficial to humanity, the person said. [Emphasis mine.]

In October, Sutskever sat down for an interview with Will Douglas Heaven at MIT Technology Review. Among the topics they discussed were:

Instead of building the next GPT or image maker DALL-E, Sutskever tells me his new priority is to figure out how to stop an artificial superintelligence (a hypothetical future technology he sees coming with the foresight of a true believer) from going rogue.

Sutskever tells me a lot of other things too. He thinks ChatGPT just might be conscious (if you squint). He thinks the world needs to wake up to the true power of the technology his company and others are racing to create. And he thinks some humans will one day choose to merge with machines.

A lot of what Sutskever says is wild. But not nearly as wild as it would have sounded just one or two years ago. As he tells me himself, ChatGPT has already rewritten a lot of people’s expectations about what’s coming, turning “will never happen” into “will happen faster than you think.”

“It’s important to talk about where it’s all headed,” he says, before predicting the development of artificial general intelligence (by which he means machines as smart as humans) as if it were as sure a bet as another iPhone: “At some point we really will have AGI. Maybe OpenAI will build it. Maybe some other company will build it.”

November 18

The next day, Axios reported that an internal OpenAI memo from COO Brad Lightcap said:

"We can say definitively that the board's decision was not made in response to malfeasance or anything related to our financial, business, safety, or security/privacy practices. This was a breakdown in communication between Sam and the board."

This kind of removed the idea that Altman was fired for what CEOs typically get fired for, malfeasance, but also kinda contradicts the implication the board made that Altman’s firing had to do with the safety of AGI.

November 19

Emmett Shear, the co-founder of Twitch, was appointed interim CEO.

Microsoft published a statement by CEO Satya Nadella:

We look forward to getting to know Emmett Shear and OAI’s new leadership team and working with them. And we’re extremely excited to share the news that Sam Altman and Greg Brockman, together with colleagues, will be joining Microsoft to lead a new advanced AI research team. We look forward to moving quickly to provide them with the resources needed for their success.

November 20

On November 20, The Information reported:

OpenAI’s board of directors approached Dario Amodei, the co-founder and CEO of rival large-language model developer Anthropic, about a potential merger of the two companies, said a person with direct knowledge. The approach…was part of an effort by OpenAI to persuade Amodei to replace Altman as CEO, the person said.

Anthropic was founded in 2021 by the aforementioned Dario Modei and his sister Daniela, among other OpenAI employees. Dario served as OpenAI's Vice President of Research.

The split with OpenAI, according to VentureBeat, was

because of “differences over the group’s direction after it took a landmark $1 billion investment from Microsoft in 2019.”

Specifically, over the fear of “industrial capture,” or the potential for one corporation to control such a powerful technology.

Amodei declined OpenAI’s offer.

That same day, more than 650 of the reported 770 OpenAI employees threatened to quit if the board didn’t resign and reinstate Altman as CEO, according to Wired.

Oddly, Sutskever was among the signatories and then subsequently took it all back in a social media post:

November 21

The New York Times reported:

Before Sam Altman was ousted…he and the company’s board of directors had been bickering for more than a year. The tension got worse as OpenAI became a mainstream name thanks to its popular ChatGPT chatbot.

At one point, Mr. Altman…made a move to push out one of the board’s members because he thought a research paper she had co-written was critical of the company.

Another member, Ilya Sutskever, thought Mr. Altman was not always being honest when talking with the board. And some board members worried that Mr. Altman was too focused on expansion while they wanted to balance that growth with A.I. safety.

The research paper in question was co-authored by OpenAI board member Helen Toner and published by Georgetown University’s Center for Security and Emerging Technology, of which she is director of strategy.

Bezinga reported:

At the center of this saga is a research paper that Toner authored — she openly criticized OpenAI's approach to AI development as far as safety is concerned while praising rival Anthropic, which was founded in 2021 by former senior members of OpenAI, siblings Daniela and Dario Amodei.

While Altman called out Toner for criticizing the company publicly while the Federal Trade Commission was investigating it over data usage, Toner reportedly defended it by saying it was simply academic commentary.

And Sam Altman’s Re-Hiring

In the end, Altman maneuvered his way back to OpenAI.

Shortly after news broke of Altman’s firing, Satya Nadella conducted a media PR blitz in support of Altman. Then he offered Altman and other OpenAI members a new department at Microsoft to continue their work. Throw in the 49% stake Microsoft holds in the for-profit company and employees threatening to bolt en masse and that’s a pretty unbeatable hand Altman was holding.

So he’s back in and most of the old board members are out.

Sutskever is one of them, having read the room and changed his tune.

Composition of OpenAI’s Board

It is important to examine the composition of OpenAI’s board both before and after.

The Ousted Board

Prior to this story breaking, the board looked like this:

• Greg Brockman, OpenAI chief technology officer and board chair

• Ilya Sutskever, OpenAI chief scientist

• Independent directors:

  • Helen Toner,

  • Tasha McCauley, technology entrepreneur, and

  • Adam D’Angelo, Quora CEO

Sutskever, as you have seen previously, has grave worries about the effect AGI may have on the world, worries he shares with Helen Toner.

Kara Swisher said on Twitter:

The issues with Tasha McCauley are deeper and, as described to me by many sources, she has used very apocalyptic terms for her fears of the tech itself and who should and should not have their “fingers on the button.”

So we know where McCauley stands.

Seven years ago, Adam D’Angelo posted his thoughts on the threat of AI on Quora, stating:

Yet, he remains on the board, apparently due to his long-time friendship with Altman and his role in brokering Altman’s return.

Effective Altruism

The philosophy behind OpenAI’s governing structure is informed by the “effective altruism” movement. Born at Oxford University and embraced by some Silicon Valley elite, the idea is to use evidence and reason to figure out how to benefit others as much as possible, and taking action on that basis.

Sounds reasonable enough but it has lost some of its appeal due to the fact that its most recognized proponent was crypto criminal Sam Bankman-Fried.

Semfor reported that:

Three of the six seats on OpenAI’s board are occupied by people with deep ties to effective altruism: think tank researcher Helen Toner, Quora CEO Adam D’Angelo, and RAND scientist Tasha McCauley. A fourth member, OpenAI co-founder and chief scientist Ilya Sutskever, also holds views on AI that are generally sympathetic to EA.

The Interim Board

OpenAI’s new board consists of:

• Adam D’Angelo

• Bret Taylor will serve as board chair. The former CEO of Salesforce and was also the former chairman of Twitter.

• Lawrence Summers, the former Treasury Secretary and veteran of the Clinton and Obama administrations. Currently president emeritus of Harvard University.

As noted above, D’Angelo has at least some trepidations about the societal harm that could be caused by AI.

Taylor forced the sale of Twitter to Musk after the Mussolini wannabe got cold feet. Taylor founded FriendFeed, which was my favorite Twitter competitor back in the day, until Facebook acquired and killed it.

Summers is a bit of a controversial figure.

As Treasury Secretary, he helped deregulate the derivatives market, which would later play a crucial role in the 2008 financial crash. He’s been ringing alarm bells about a hard landing for the economy since early 2022 and yet the economy has just continued to improve. So his instincts don’t seem especially sharp.

Axios notes what others are saying about Summers:

"There is no greater indication that OpenAI is unserious about the interests of humanity than their elevation of Larry Summers to its Board of Directors," says the Revolving Door Project's Jeff Hauser in a press release.

Journalists, wrote The New Yorker's John Cassidy upon Summers' 2010 departure from the White House, "aren't the only folks Larry considers intellectually beneath him. Such a category would include most members of President Obama's cabinet and their top policy advisers; many of his colleagues in the White House; virtually all foreign officials; ninety per cent of the Harvard faculty; and a similar proportion, or possibly higher, of his fellow academic economists."

Summers certainly has cozied up to Silicon Valley elites in recent years, sitting on the boards of Block and Lending Club and advising the likes or venture capital firms Andreessen Horowitz, according to Axios.

On the other hand, in 2017 he took Trump’s Treasury Secretary Steve Mnuchin to task for saying AI is not even on his radar. In a Washington Post editorial, Summers wrote:

Mnuchin’s comment about the lack of impact of technology on jobs is to economics approximately what global climate change denial is to atmospheric science or what creationism is to biology. Yes, you can debate whether technological change is in net good. I certainly believe it is. And you can debate what the job creation effects will be relative to the job destruction effects. I think this is much less clear, given the downward trends in adult employment, especially for men over the past generation.

But I do not understand how anyone could reach the conclusion that all the action with technology is half a century away. Artificial intelligence is behind autonomous vehicles that will affect millions of jobs driving and dealing with cars within the next 15 years, even on conservative projections. Artificial intelligence is transforming everything from retailing to banking to the provision of medical care. Almost every economist who has studied the question believes that technology has had a greater impact on the wage structure and on employment than international trade and certainly a far greater impact than whatever increment to trade is the result of much debated trade agreements.

While he’s expressed concerns over the risk AI poses over job destruction, he also seems awfully comfortable among tech’s libertarian moguls.

Summers does seem to have some measure of credibility with a variety of stakeholders that are important to OpenAI:

• With Silicon Valley set generally and most importantly with Microsoft,

• With those concerned about the potential harms AI presents, but, perhaps most importantly,

• With the policymakers who will eventually want to regulate the industry.

Keep in mind that this is an interim board, so there will most assuredly be additions and among them I think you could expect someone to represent Microsoft’s interests.

The Board Upheld Its Responsibilities

Here’s the thing, though: The OpenAI board functioned exactly as it was designed to function.

It was designed to put a halt on the development of AGI before safeguards could be put in place to protect humanity, or at least serve as an early-warning system to warn the rest of us of what’s coming.

We still don’t know what exactly was the catalyst for firing Altman but it would appear that’s what they’ve done.

Commercialization Of OpenAI

One line of thought posits that Altman’s push to commercialize OpenAI’s technology in the form of DALL-E, ChatGTP, the ChatGPT API, and most recently, Custom GPTs, was too fast and too soon without accounting for the potential risks unleashing these products to the world posed.

Microsoft Dominance

Another line of thought concerned the 49% relationship with Microsoft which tilted the competitive landscape enormously in their favor (as you could see with the integration of OpenAI technology into Bing, the Microsoft Office Suite, Visual Studio co-pilot, and even LinkedIn), risking one company becoming the dominant player in AI.

Altman’s Side-Hustles

Yet another line of thought had to do with the many side-hustles Altman was involved in. According to Bloomberg:

…Altman's myriad side hustles, including the controversial crypto endeavor Worldcoin, which seeks to scan the eyeballs of every human on earth, or his recent attempt to raise money for an AI chip company to take on Nvidia Corp.

Which could certainly be considered a conflict of interest.

AI Agents

What happens when you give people the power to create AI that interacts with other AI?

If you give people the ability to create their own custom GPTs, then how much of a stretch is it to allow people to create AIs that can talk to other AIs and complete tasks between them?

Sure, there are a lot of benefits to be had from the deployment of autonomous AIs but there is certainly a lot of havoc that can be caused as well.

Q*

The last, and most intriguing and perhaps frightening theory, is that OpenAI was on the brink of an innovation that could conceivably complete the organization’s mission of developing artificial general intelligence.

Reuters reported:

several staff researchers wrote a letter to the board of directors warning of a powerful artificial intelligence discovery that they said could threaten humanity, two people familiar with the matter told Reuters.

The previously unreported letter and AI algorithm were key developments before the board's ouster of Altman, the poster child of generative AI, the two sources said.

…OpenAI, which declined to comment, acknowledged in an internal message to staffers a project called Q* and a letter to the board before the weekend's events, one of the people said.

…Some at OpenAI believe Q* (pronounced Q-Star) could be a breakthrough in the startup's search for what's known as artificial general intelligence (AGI), one of the people told Reuters.

And Business Insider reports:

The ability to solve basic math problems may not sound that impressive, but AI experts told Business Insider it would represent a huge leap forward from existing models, which struggle to generalize outside the data they are trained on.

"If it has the ability to logically reason and reason about abstract concepts, which right now is what it really struggles with, that's a pretty tremendous leap," said Charles Higgins, a cofounder of the AI-training startup Tromero who's also a Ph.D. candidate in AI safety. He added, "Maths is about symbolically reasoning — saying, for example, 'If X is bigger than Y and Y is bigger than Z, then X is bigger than Z.' Language models traditionally really struggle at that because they don't logically reason, they just have what are effectively intuitions."

Sophia Kalanovska, a fellow Tromero cofounder and Ph.D. candidate, told BI that Q's name implied it was a combination of two well-known AI techniques, Q-learning and A* search. She said this suggested the new model could combine the deep-learning techniques that power ChatGPT with rules programmed by humans. It's an approach that could help fix the chatbot's hallucination problem.

Brian Wang at NextBigFuture writes:

The battle at OpenAI was possibly due to a massive breakthrough dubbed Q* (Q-learning). Q* is a precursor to AGI. What Q* might have done is bridged a big gap between Q-learning and pre-determined heuristics. This could be revolutionary, as it could give a machine “future sight” into the optimal next step, saving it a lot of effort. This means that machines can stop pursuing suboptimal solutions, and only optimal ones. All the “failure” trials that machines used to have (eg. trying to walk but falling) will just be put into the effort of “success” trials. OpenAI might have found a way to navigate complex problems without running into typical obstacles

If OpenAI has developed a shortcut to logical reasoning (and math is logical reasoning), then it has (or will eventually) conquered a major limitation of Large Language Models (LLMs), which is the ability to do basic math.

While LLMs are great at predicting text—basically pattern recognition—that doesn’t amount to reasoning. The ability to succeed at mathematics is reasoning.

Regardless of what the catalyst was, a majority of the board members believed that the organization under Sam Altman’s leadership was headed down a path in the development of general artificial intelligence that they believed was either dangerous to one degree or another or at least needed a pace of development that allowed for the implementation of safeguards.

Who Controls AI?

So it would appear, as Kevin Roose declared, that A.I. Belongs to the Capitalists Now.

It is hardly shocking that capital has blown through a nonprofit governance structure with relative ease. When there’s billions to be made, woe to the fortunes of anyone who stands in a self-respecting titan of industry’s way.

We’ve seen this movie before but in the past the exploitation of the resources those titans of industry had at their disposal were limited. You can only build so many oil drills or lay so many miles of railroad tracks or construct so many steel mills.

The titans of the information age have no such physical restrictions. Their produce is information. It is ever renewing and infintely expanding.

We can talk about open source as a countervailing force to corporate AI development, but with the resources of a Microsoft or a Facebook or an Amazon at their disposal, do you seriously think open source can be the check to balance corporate power?

Many a science fiction dystopian plot revolves around the supremacy of corporations as the primary governing entities in society.

So we need to ask ourselves: Is it wise for such a powerful technology as AI (nevermind general artificial intelligence) to be concentrated in the hands of just a few corporate entities?

Especially, as we’ve learned from the age of social media, absent meaningful regulation?

Digital Marketing News

Artificial Intelligence

The Guardian by Douglas Rushkoff - ‘We will coup whoever we want!’: the unbearable hubris of Musk and the billionaire tech bros - Challenging each other to cage fights, building apocalypse bunkers – the behaviour of today’s mega-moguls is becoming increasingly outlandish and imperial

Unlike their forebears, contemporary billionaires do not hope to build the biggest house in town, but the biggest colony on the moon. In contrast, however avaricious, the titans of past gilded eras still saw themselves as human members of civil society. Contemporary billionaires appear to understand civics and civilians as impediments to their progress, necessary victims of the externalities of their companies’ growth, sad artefacts of the civilisation they will leave behind in their inexorable colonisation of the next dimension. Unlike their forebears, they do not hope to build the biggest house in town, but the biggest underground lair in New Zealand, colony on the moon or Mars or virtual reality server in the cloud.

While plans for Peter Thiel’s 193-hectare (477-acre) “doomsday” escape, complete with spa, theatre, meditation lounge and library, were ultimately rejected on environmental grounds, he still wants to build a startup community that floats on the ocean, where so-called seasteaders can live beyond government regulation as well as whatever disasters may befall us back on the continents.

Keep in mind—especially in the wake of the OpenAI drama—the fundamental weirdness of some of these tech bros.

The Artificial Corner by Pycoach - OpenAI Is Slowly Killing Prompt Engineering With The Latest ChatGPT and DALL-E Updates - When OpenAI introduced GPTs they presented them as a new way to create custom versions of ChatGPT without coding and with even chances to monetize them on the GPT store. 

Something that was overlooked, though, was some of the functionalities of GPT Builder. GPT Builder is an assistant that helps us automatically configure a GPT, which has “instructions” as a main component. 

What are instructions? They’re just prompts. Those that we manually crafted until now.

Experience ChatGPT users frequently use prompting techniques in their prompts. An example of a prompting technique is role prompting.

Act as a job interviewer. I’ll be the candidate and you’ll ask me interview questions for the X position …

GPT Builder is not only able to automatically generate prompts for us but apply some prompting techniques as well. We only have to give it an idea and it’ll craft a complete prompt for us. 

All this is done via a chat with GPT Builder.

This was inevitable. I’ve been playing around building my own custom GPTs with OpenAI’s GPT Builder. It is pretty impressive technology with some frustrating limitations, too. I would expect that will change as updates to the GPT Builder are released. But this is another topic for another post. Stay tuned. Meanwhile:

Artificial Intelligence in Plain English by Anthony Alcaraz - How “Memories” Can Make Chatbots Smarter and More Personalized - Exciting new research from CMU and Allen Institute shows how giving chatbots a “memory” could make them smarter and more personalized over time.

The paper “MemPrompt: Memory-assisted Prompt Editing with User Feedback” proposes a novel method called MemPrompt that allows chatbots like large language models (LLMs) such as GPT-3 to continuously improve through interactive feedback from users, without needing full retraining. This represents a technique in natural language processing that can enhance chatbots’ language understanding and generation capabilities…

MemPrompt provides a novel solution — give the chatbot a memory that can store clarifying feedback from users when the chatbot misunderstands their intent.

For a new user query, MemPrompt first checks if this chatbot has misunderstood a similar question in the past. If so, the clarifying feedback is retrieved from the memory and appended to the current query. This additional context helps the chatbot interpret the user’s intent correctly by preventing it from repeating past mistakes.

This is a fascinating area of AI development and one that I’ll address in more detail soon in a subsequent post. The lack of a memory across sessions is one of the most striking limitations of existing AI chatbots. Here’s the link to the aforementioned paper [PDF].

VentureBeat by Carl Franzen - Midjourney’s new style tuner is here. Here’s how to use it. - Before style tuning, users had to repeat their text descriptions to generate consistent styles across multiple images — and even this was no guarantee, since Midjourney, like most AI art generators, is built to offer a functionally infinite variety of image styles and types.

Now instead, of relying on their language, users can select between a variety of styles and obtain a code to apply to all their works going forward, keeping them in the same aesthetic family. Midjourney users can also elect to copy and paste their code elsewhere to save it and reference it going forward, or even share it with other Midjourney users in their organization to allow them to generate images in that same style. This is huge for enterprises, brands, and anyone seeking to work on group creative projects in a unified style.

Haven’t tried this yet, mostly because it doesn’t fit with what I’m trying to get out of Midjourney and I’ve defined the style I use but it does look pretty cool.

The Verge by Allison Johnson - Bard can now watch YouTube videos for you - Bard’s YouTube extension can now handle complex queries about specific video content, like recipe quantities and instruction summaries.

This will save time but may pose significant problems for marketers. We’ll have to think about video production in a new light with this new capability.

Public Relations

The Guardian by Adam Lowenstein - Revealed: how top PR firm uses ‘trust barometer’ to promote world’s autocrats - Edelman, world’s largest public relations company, paid millions by Saudi Arabia, UAE and other repressive regimes

For years, Edelman has reported that citizens of authoritarian countries, including Saudi Arabia, Singapore, the United Arab Emirates and China, tend to trust their governments more than people living in democracies do.

But Edelman has been less forthcoming about the fact that some of these same authoritarian governments have also been its clients. Edelman’s work for one such client – the government of the UAE – will be front and center when world leaders convene in Dubai later this month for the UN’s Cop28 climate summit.

The Guardian and Aria, a non-profit research organization, analyzed Edelman trust barometers, as well as Foreign Agent Registration Act (Fara) filings made public by the Department of Justice, dating back to 2001, when Edelman released its first survey of trust. (The act requires US companies to publish certain information about their lobbying and advocacy work for foreign governments.) During that time Edelman and its subsidiaries have been paid millions of dollars by autocratic governments to develop and promote their desired images and narratives.

Polling experts have found that public opinion surveys tend to overstate the favorability of authoritarian regimes because many respondents fear government reprisal. That hasn’t stopped these same governments from exploiting Edelman’s findings to burnish their reputations and legitimize their holds on power.

Booooo! Looks like Edelman might need a PR firm.

Search Engines

Search Engine Land by Barry Schwartz - Google updates search quality raters guidelines - Although search quality evaluators’ ratings do not directly impact rankings, they do provide feedback that helps Google improve its algorithms. It is important to spend some time looking at what Google changed in this updated version of the document and compare that to the previous version of the document to see if we can learn more about Google’s intent on what websites and web pages Google prefers to rank. Google made those additions, edits, and deletions for a reason.

If you want to get a sense of how Google rates the reputation of a given web page, read the company’s search quality raters guidelines.

Social Media

Meta by Nick Clegg - New Tools to Support Independent Research - To understand the impact social media apps like Facebook and Instagram have on the world, it’s important to support rigorous, independent research. That’s why Meta has been committed to an open and privacy-protective approach to research for many years, including making tools available to support public interest research, such as the US 2020 studies.

Over the past few months we gave Beta access to our new Meta Content Library and API tools. After multiple rounds of feedback with researchers and other stakeholders, we are now in a position to roll these tools out more broadly.

Our Meta Content Library and API tools provide access to near real-time public content from Pages, Posts, Groups and Events on Facebook, as well as from creator and business accounts on Instagram. Details about the content, such as the number of reactions, shares, comments and, for the first time, post view counts are also available. Researchers can search, explore and filter that content on both a graphical User Interface (UI) or through a programmatic API. 

👏👏👏 Hopefully Threads soon, too.

Social Media Today by Andrew Hutchinson - Meta Highlights What’s Driving Threads Engagement - Various studies have shown that the content that best drives emotional response, particularly in regards to inspiring comments, is material that prompts high-arousal emotions, such as anger, fear, and happiness. If you can hit these emotional triggers, your content is much more likely to go viral, which is what social algorithms incidentally incentivize by feeding into engagement.

Meta doesn’t want Threads to go down that path, which could also be an element to consider, in that Meta could be actively looking to lessen the impact of certain elements in order to encourage a different type of in-stream engagement.

I’ve been really impressed with Threads. It’s the social channel I’m personally spending the most time at. The algorithm does appear to favor engagement but not negative emotional triggers. I can find most of the types of accounts I want content from there (sports is still a little lacking yet).

It is quickly becoming the news/politics Twitter alternative but with the tools to deal with trolls and keep the place relatively civil. President Biden, the First Lady, Vice President Harris and the Second Gentleman have recently joined, as has the White House.

If you’re using Threads, say hi! I’m @deerickson.

Business Insider by Swadah Bhaimiya - An agency created an AI model who earns up to $11,000 a month because it was tired of influencers 'who have egos' - A Spanish modeling agency said it's created the country's first AI influencer, who can earn up to 10,000 euros, or $11,000, a month as a model.

Euronews reported the news, based on an interview with Rubeñ Cruz, founder of the Barcelona-based modeling agency The Clueless, which created the influencer.

The AI-generated woman, Aitana López, is a pink-haired 25-year-old. Her account has amassed 124,000 followers on Instagram.

Cruz told Euronews he decided to design López after having trouble working with real models and influencers. "We started analyzing how we were working and realized that many projects were being put on hold or canceled due to problems beyond our control. Often it was the fault of the influencer or model and not due to design issues," he said.

I give you muy post from May:

Music Monday

In case you missed it, here’s The Beatles newest song, Now and Then. It was made possible due to technology developed by director Peter Jackson during his production of The Beatles documentary, Get Back, that enabled the separation of John Lennon’s voice from the piano of a 1970s-era demo recorded on a cassette tape in Lennon’s home.

Threads

Stealing an idea from Casey Newton.

Welcome! I'm David Erickson and this is my digital marketing newsletter. If you've received it, then you either subscribed or someone forwarded it to you.

If this was forwarded to you and you want to subscribe, just click this button:

Subscribe now

In this issue: Major AI news, Ubiquitous Computing, Effortless AI Conversations, AI Personas, AI Memory, AI Personalization, Synthetic Influencers, AI Generated Video Games, Generative AI + Robots = Androids; Digital Marketing News; Music Monday; and Glorious Midjourney Mistakes.

Leave it to to connect some interesting dots.

Newton’s recent article, The Synthetic Social Network is Coming, reports on the many truly profound advances in AI that recently dropped. Among them:

• OpenAI updates to ChatGPT to make the tool multimodal. You can upload an image to ChatGPT and ask it to analyze the picture for you. But most profoundly, you can now interact with ChatGPT with your voice and it will answer with its own voice. Not just that, it will answer with a voice that has personality, instead of the robotic-sounding voice we’ve come to expect from Alexa, Siri and Google Home.

• Meta—otherwise known as Facebook—will be building generative AI into all its products. You’ll be able to have conversations with its AI within Facebook Messenger. You’ll be able to create images and stickers on the fly to add personality to your chats with friends.

  • The company will also release 28 chatbots with distinct personalities that will have their own Facebook pages and Instagram accounts.

• Amazon is investing $4 billion in Anthropic, the company that developed the Claude AI chatbot.

• Spotify has developed AI technology that will translate audio podcasts into other languages in the participants’ own voices.

But most profoundly, Newton notes:

More significantly, I think, is the idea that Meta plans to place its AI characters on every major surface of its products. They have Facebook pages and Instagram accounts; you will message them in the same inbox that you message your friends and family. Soon, I imagine they will be making Reels.

And when that happens, feeds that were once defined by the connections they enabled between human beings will have become something else: a partially synthetic social network.

(Listen to this Hard Fork episode for Newton and Kevin Roose’s discussion on these advancements.)

This is where my love of science fiction kicked in and got me thinking about what is just over the horizon as a result of these advancements.

So, let’s game this out a bit.

Ubiquitous Computing

One of the things that really set me off on this career trajectory was a computer science paper by Mark Weiser [PDF] that I read in 1992 which argued that computers need to become so ubiquitous that they essentially disappear.

It made the point that computers, like eyeglasses, are a tool to extend and enhance your power. Eye glasses enhance and extend your power of sight.

Computers do the same for your mind or your productivity or what have you, but the difference is we look at computers but when we put on a pair of prescription glasses, the glasses essentially disappear and all we’re left with is enhanced sight. We don’t look at them, we look through them.

The paper argued that computers need to evolve to the point that they disappear, that there is no longer a barrier between the tool and the power they give us.

If you think about the evolution of computers, we have been on this glide path toward ever more effortless input of data or information into the device.

Computers used to take up entire rooms and you had to feed them punch cards—literally paper cards with slots punched into them—in order to get a result out of a computer.

Eventually, they got smaller so though corporations still needed a computer room, the machines weren’t lined wall-to-wall. Gone were punch cards in favor of a keyboard with a monochromatic screen, through which you typed cryptic commands to get the result you needed.

Then came the personal computer revolution with its graphical interface and a mouse to accompany the keyword as the input device.

Fast forward to the smart phone with its touch screen and voice activation.

Effortless AI Conversations

Now we have AI that not only is voice activated but also talks back.

That gives you the ability to have an effortless conversation with your computer and since you’re talking to an app and not a device, those conversations are portable and available to you wherever you go.

The addition of this call and response feature of AI is profound and will fundamentally change consumer behavior.

I’ve had Amazon Echo and Google Home smart speakers since they were first introduced to consumers.

One day after work while on my way to get my bus, I was thinking about playing football over the weekend. Without thinking, I nearly blurted out “Alexa, what’s the forecast for this weekend,” while walking the crowded streets of downtown Minneapolis.

Talking to my smart speakers became second nature.

But that’s talking at technology. Conversing with a technology as powerful as generative AI opens up vast new opportunities and implications for individuals, organizations and society at large.

With generative AI in my smart speakers and ChatGPT on my phone and AirPods in my ears, I have an insanely powerful assistant at my beck and call.

AI Personas

ChatGPT offers five synthetic voices, both male and female with American accents, called Juniper, Sky, Cove, Ember and Breeze and they all have different personalities.

Meta upstaged ChatGPT with its announcement that the company would be rolling out more than two dozen AI personas across its platforms:

We’re introducing Meta AI in beta, an advanced conversational assistant that’s available on WhatsApp, Messenger, and Instagram, and is coming to Ray-Ban Meta smart glasses and Quest 3.[…]

We’re also launching 28 more AIs in beta, with unique interests and personalities. Some are played by cultural icons and influencers, including Snoop Dogg, Tom Brady, Kendall Jenner, and Naomi Osaka. 

Over time, we’re making AIs for businesses and creators available, and releasing our AI studio for people and developers to build their own AIs. 

Think about the dynamic that occurs when you have conversations with someone. If that person’s responses appear credible and genuine and they come across as a unique personality, the more enjoyable the conversation will be and the more it will hold our interest.

Over time, as we engage in compelling conversations and as long as our conversational partner remains consistent, our natural guard will fall and we will express more candor in our own responses.

Conversation betrays the inner dialogue we hold with ourselves; our interests and passions, the challenges we struggle with, our fears, desires and aspirations.

With AI chatbots that sound human and have distinct personalities, it is inevitable that people will form emotional bonds with their AIs, as Newton surmised.

With generative AI’s ability to respond fairly uniquely to any given prompt, this technology enables one-to-one relationships at scale.

Refer a friend

AI Memory

Generative AI companies are moving toward remembering users’ conversations across sessions because the chatbots are far more useful if they remember everything you’ve ever told them.

• ChatGPT allows you to turn on or off chat histories.

  • ChatGPT also allows you to save “Custom Instructions” to “provide better responses”, such as:

    • Your geographic location

    • Your profession or the nature of your work

    • Your hobbies or interests

    • Your passions

• Bing Chat now remembers previous conversations by default

• Google just added a toggle to let Bard remember your previous chats

• Anthropic pointedly does not remember past conversations, citing several reasons that include security and privacy concerns. We’ll see if that changes now that Amazon has invested so heavily in the company.

The most obvious benefit to users is that you won’t have to constantly tell the AI things you want it to know about yourself in order to be a more efficient assistant.

Think about all the Google tools you use. For myself, I use Google daily to find information and Google has my entire search history at its disposal. I use Chrome as my default browser, so Google knows the websites I’ve visited.

I have had a Gmail account (including Contacts) since they first became available, so the messages I’ve received and sent are another source of data Google can tap into.

I’ve had a YouTube account since the site launched, so my video viewing behavior is available. I’ve been using Google Docs forever. I have a Google Home smart speaker. I’ve got many Google apps on my iPhone, including Maps, which I routinely use for navigation.

Google already has tons of data about me before it ever starts remembering my Bard chats.

Same goes for Meta.

I’ve had a Facebook account since 2004, an Instagram account since it first launched; I use Facebook Messenger occasionally and I have an Oculus.

Facebook knows who I’m connected with and the kind of content I like. So, again: Even before I start engaging with its AI chatbots, the company knows a lot about me.

AI Personalization

As we become more comfortable interacting with AI bots and as the AI companies continue to improve multi-modal generative AI (text, images, audio and video), this technology will evolve ever closer to appearing human.

This will set the environment in which consumers respond to AI interactions with greater and greater candor. Our own proclivities to share coupled with the bot’s ability to remember past conversations will provide a personalization that will be so compelling it may be impossible to resist.

It is unlikely the AI platforms would store entire verbatim conversations on their servers. It is more likely that those conversations would be scored and a personal, ever-updating algorithm would be assigned to each user that would reflect topical interests, language usage, conversational style, demographics, personality type, and more.

This could signal the final nail in the coffin of privacy, at least when it comes to the tech companies. That is something that should give us all pause and I wonder if it is a dynamic regulators have even contemplated.

The benefits of such personalization would be enormous but the potential risks could be grave.

Synthetic Influencers

Now let’s talk influencers.

I have worked with influencers years before the phrase rose to prominence; it is at the core of the disciplines of public relations and public affairs.

It is just that influencer marketing used to be restricted to journalists, policy makers and celebrities. It has since expanded to webmasters (for link building), newsletter publishers, podcast hosts and YouTubers, to what we now conceive of as influencers: The lifestyle-focused social media stars.

They are called influencers because they influence the information their followers consume and how they think about that information.

I have also tracked with fascination the rise of virtual influencers.

Social media platforms like Facebook and Instagram are where human influencers have flourished. But the problem with human influencers is that they are failable and, as a result, can create public relations fiascos that do not reflect well on the platforms that host them and the brands that sponsor them.

Women’s Wear Daily has a rundown of just a few influencer controversies from 2021.

Plus, human relationships don’t scale.

But AI does.

I believe Meta’s announcement of 28 AI personas with which users can interact is an attempt to create their own homegrown influencers. It is telling that each AI will have their own Facebook and Instagram profiles.

If these AI bots can customize themselves to the person with whom they are interacting, it will create a bond of trust that scales far beyond what any human influencer can achieve.

Instead of brands doing deals with human influencers to broadcast messaging to their followers, Meta can offer a self-serve system that tailors the messaging an AI influencer delivers to precisely the right audience the brand wants to reach with exactly the right tone and language that resonates with specific individuals.

AI Generated Video Games

Video games are going to get really, really good.

Not that they weren’t already good. The video game industry generated $183 billion last year, more than the film industry ($26B) and the music industry ($31.2B) combined.

But most video games require a heavy dose of suspension of disbelief to enjoy. Modeling of video game characters—-while it has improved immensely over the years—has yet to achieve the kind of realism you get from movies. It’s in the eyes. Video game characters’ eyes don’t look or behave in a realistic fashion.

Generative AI holds the promise of changing that. With the amazing advances I’ve witnessed from AI image generation tools like DALL-E and Midjourney, this problem appears eminently solvable.

These image tools also promise to help streamline workflows in video game development and reduce costs.

But take generative AI image generation technology such as what Adobe Firefly can now do by filling in imagery where none previously existed and you can imagine programming a video game that can create an infinite variety of environments and scenes, tailored to an individual player’s preferences.

So imagine an open world game like Red Dead Redemption that creates itself on the fly based on an individual user’s whims. Individual users could create their own storylines to direct and follow.

For the benefit of those of you who do not play video games, within games there are non-player characters. These are characters who play a role in the story of the game but are programmed by the video game developers and not controlled by a human.

These non-player characters may play prominent or even starring roles in cut-scenes, which are pre-programmed segments of a video game that are used to advance the plot.

But there are far more non-player characters in video games whose role amounts to extras on a movie set. Their role is simply to create the illusion of a lifelike, active environment within which humans can play.

For example, I may be walking through a town in Red Dead Redemption and the environment will include non-player characters lounging outside the town saloon, or visiting the general store, or parking a stage coach to disgorge its passenger non-player characters. These characters create the ambiance within which the story unfolds.

As I’m walking down the street of this town, I may bump into one of these non-player characters and that character’s response might be a muttered “excuse me” or it could be “hey, watch where you’re going buddy!” But there’s not likely to be much variety in such responses.

Add generative AI to the equation and things get really fascinating.

Let’s rework the example I just cited but with the addition of generative AI. Instead of responding to my clumsiness, with an “excuse me” or “what the hell you doin?” kind of response, what if I could hold an actual conversation with them?

I could say to the non-player character: “I apologize. That was very clumsy of me. By the way, I’m new to town and I’m looking for information about Low Down Dirty Ted (the villain of our game) who I believe passed through your town a couple of weeks ago. Do you have any idea who might have come in contact with him?”

This non-player character (or any non-player character) could respond appropriately. Or even walk me into the saloon and tell me about his own experience with Low Down Dirty Ted over drinks.

Video game programmers could use generative AI to give each non-player character their own unique back stories, so their responses would be framed in terms of their past history. The video game engine would be aware of each non-player’s back story, so each one would be unique.

Furthermore, generative AI could take into account a non-player character’s interactions with other human players or even other non-player characters and add those experiences to their back story history, which would then inform future interactions.

Non-player characters would then turn from pre-programmed automatons to ever-evolving players in their own right.

A video gaming environment in which you can interact on a very deep level with even the bit non-player characters would completely revolutionize the medium.

Don’t count out the metaverse just yet.

Generative AI + Robots = Androids

The last mile, as it were, for generative AI is when it is given agency in the physical world.

We already have robots that can effortlessly navigate physical space and perform tasks in the real world, thanks to companies like Boston Dynamics:

Now, take everything I’ve thus far speculated about the potential of generative AI and apply it to robots.

The ability to have a conversation with a robot that you can direct to perform tasks for you in the real world and which is increasingly customized to an individual’s unique needs opens up the possibility of an android personal companion.

Imagine a person who needs round-the-clock in-home care due to physical or behavioral ailments and who may have no family and perhaps no friends; no real support system to speak of.

What if they can have a constant companion who can address their physical care while also being attentive to their emotional needs? You can see how such a companion could make such a person’s life immeasurably better.

I believe personal android companions will be inevitable. There are many potential dangers which I need not list here; science fiction has already done a fairly thorough job of that.

But I do believe we are in the pre-android phase of AI.

Subscribe now

Digital Marketing News

Artificial Intelligence

Gizmodo by Kevin Hurler - OpenAI Employee Discovers Eliza Effect, Gets Emotional - The technological optimism of the 1960s bred some of the earliest experiments with “AI,” which manifested as trials in mimicking human thought processes using a computer. One of those ideas was a natural language processing computer program known as Eliza, developed by Joseph Weizenbaum from the Massachusetts Institute of Technology.

Eliza ran a script called Doctor which was modelled as a parody of psychotherapist Carl Rogers. Instead of feeling stigmatized and sitting in a stuffy shrink’s office, people could instead sit at an equally stuffy computer terminal for help with their deepest issues. Except that Eliza wasn’t all that smart, and the script would simply latch onto certain keywords and phrases and essentially reflect them back at the user in an incredibly simplistic manner, much the way Carl Rogers would. In a bizarre twist, Weizenbaum began to notice that Eliza’s users were getting emotionally attached to the program’s rudimentary outputs—you could say that they felt “heard & warm” to use Weng’s own words.

AI therapy should be taken with a salt mine but the notion that people will form emotional bonds with computer systems is not new.

Bloomberg by Rachel Metz - OpenAI Gives ChatGPT the Ability to Speak in Five Different Voices - Artificial intelligence startup OpenAI is rolling out a feature for its ChatGPT app that lets the chatbot respond to spoken questions and commands with speech of its own.

Starting over the next two weeks, users will be able to choose a voice in the chatbot app, picking from five personas with names like “Juniper,” “Breeze” and “Ember.” ChatGPT will then produce audio of the text it generates in that voice — for example, reading an AI-generated bedtime story out loud.

I have this feature on my iPad but maddeningly not yet on my iPhone where I could really put it to the test with my AirPods while walking around and talking naturally. I’ll report back after I get full access and can properly test it out.

The Verge by Emilia David - Microsoft to add DALL-E 3 to Bing Chat - The company also added new shopping features to Bing. Bing can ask users specific questions on how a product will be used or other more personalized questions so people can pinpoint the right product that suits their needs. Bing also lets people find and use discount codes. These more personalized answers work as Bing also now remembers previous chats.

Bing Chat is also displaying ads within its generative AI results.

Wired by Steven Levy - Smarter AI Assistants Could Make It Harder to Stay Human - To me, that’s the worry—once we get comfortable, we’re finished. When I sought validation in a scan of research papers, my attention was snared by the title “The Power to Harm: AI Assistants Pave the Way to Unethical Behavior.” Coauthored by University of Southern California scientists Jonathan Gratch and Nathanael Fast, it hypothesizes that intelligent agents can democratize an unsavory habit of rich people, who outsource their bad behavior through lawyers, spokespeople, and thuggish underlings. “We review a series of studies illustrating how, across a wide range of social tasks, people may behave less ethically and be more willing to deceive when acting through AI agents,” they write.

I mean, we already see this phenomenon every day in the news with Donald Trump.

Business Insider by Beatrice Nolan - Google is quietly handing out early demos of its GPT-4 rival called Gemini. Here's what we know so far about the upcoming AI model. - One person who had tested the tech told the outlet it may have an advantage on GPT-4 because it leverages Google's data from consumer products, as well as information gathered from the internet. The addition should mean the model can more accurately understand the user's intentions, the report said.

The person also said the model appeared to generate fewer incorrect answers, a common problem in artificial intelligence known as hallucinations.[…]

Researchers behind the SemiAnalysis blog have also predicted that Google's Gemini would likely outperform GPT-4 because of Google's access to top-flight chips.

Google has sooooo much data to train its AI models.

VentureBeat by Matt Marshall - Amazon bets $4 billion on Anthropic’s Claude, the chatbot platform rivaling ChatGPT and Google’s Bard - The investment was part of a significant partnership announced by the two companies, where Anthropic agreed to use Amazon’s cloud platform for “mission-critical workloads” in return for the investment. The backing is the first major connection by Amazon to a leading chatbot, at a time when cloud competitors Microsoft and Google have already bet big on their respective chatbot platforms.

It appears this supplants a previous deal with Google, which invested $300 million in Anthropic earlier this year.

The Decoder by Matthias Bastain - OpenAI releases new language model InstructGPT-3.5 - Instruct models are large language models that are refined through human feedback (RLHF) after being pre-trained with a large amount of data. In this process, humans evaluate the model's output in response to user-provided prompts and improve it to achieve a target result, which is then used to further train the model.

As a result, Instruct models are better able to understand and respond to human queries as expected, making fewer mistakes and spreading less harmful content. OpenAI's tests have shown that people prefer an InstructGPT model with 1.3B parameters to a GPT model with 175B parameters, even though it is 100 times smaller.

So improvements are on the way.

The Verge by Amrita Khalid - Spotify is going to clone podcasters’ voices — and translate them to other languages - The backbone of the translation feature is OpenAI’s voice transcription tool Whisper, which can both transcribe English speech and translate other languages into English. But Spotify’s tool goes beyond speech-to-text translation — the feature will translate a podcast into a different language and reproduce it in a synthesized version of the podcasters’ own voice.

Obviously, this will be a boon to podcasters by exposing vast new audiences to their shows but it will also create a massive new inventory of advertising for Spotify to sell.

VentureBeat by Ben Dickson - Microsoft’s AutoGen framework allows multiple AI agents to talk to each other and complete your tasks - As described by Microsoft, AutoGen is “a framework for simplifying the orchestration, optimization, and automation of LLM workflows.” The fundamental concept behind AutoGen is the creation of “agents,” which are programming modules powered by LLMs such as GPT-4. These agents interact with each other through natural language messages to accomplish various tasks.

Agents can be customized and augmented using prompt engineering techniques and external tools that enable them to retrieve information or execute code. With AutoGen, developers can create an ecosystem of agents that specialize in different tasks and cooperate with each other.

An agent that did scheduling for me would be nice. Or one that could sit in on a vendor demo and ask the questions I’d want answered; that’d be great.

Media

Vanity Fair by Charlotte Klein - In the AI Age, The New York Times Wants Reporters to Tell Readers Who They Are - In August, staff on the Business desk of The New York Times got an email from the Trust team, alerting them that they would be rolling out new reporter bios and asking them to submit a new author page for themselves. “We want to get moving quickly on this,” the internal email, reviewed by Vanity Fair, states. “The masthead feels it’s especially important to highlight the human aspect of our work as misinformation and generative AI proliferates.”

“Readers tend to seek out information about a reporter in moments of doubt or agitation: when they encounter a viewpoint they dislike in our reporting, or they perceive inaccuracy or bias. In these moments, bios can play an important role in assuring readers that we are fair-minded, committed to a high standard of integrity, and free from conflicts of interest,” the email states, adding: “Bio pages also rank highly in Google.”

This is a point I’ve been making for some time. As generative AI content proliferate, people will start to look for evidence that the content they’re viewing was created by humans.

Robotics

USA Today by Whitney Woodworth and Mary Walrath-Holdridge - Robot takeover? Agility Robotics to open first-ever factory to mass produce humanoid robots - The creators of Digit, a human-sized bipedal robot complete with "eyes," are bringing the world's first humanoid robot factory to Oregon, creatively named RoboFab.

Agility Robotics announced the opening in Salem, Oregon on Monday, saying they expect to soon have the capacity to produce 10,00 robots annually. Construction of the 70,000-square-foot facility began last year and is set to open in late 2023.

Creating advanced robots for sale to the public is a new development in the robotics industry, as access to such high-end tech has been generally reserved for entities such as businesses and government agencies in the past.

What’d I just say?

Social Media

NBC News by Ben Collins - What was Elon Musk’s strategy for Twitter? - On the day that public records revealed that Elon Musk had become Twitter’s biggest shareholder, an unknown sender texted the billionaire and recommended an article imploring him to acquire the social network outright.

Musk’s purchase of Twitter, the 3,000-word anonymous article said, would amount to a “declaration of war against the Globalist American Empire.” The sender of the texts was offering Musk, the Tesla and SpaceX CEO, a playbook for the takeover and transformation of Twitter. As the anniversary of Musk's purchase approaches, the identity of the sender remains unknown.

The three texts were sent on April 4, 2022. In the nearly 18 months since then, many of the decisions Musk made after he bought Twitter appear to have closely followed that road map, up to and including his ongoing attacks against the Anti-Defamation League, a nonprofit organization founded by Jewish Americans to counter discrimination. 

Important read. It became clear fairly quickly after Musk bought Twitter that his interest in the platform was simply to leverage it for political power.

Gizmodo by Nikki Main - Elon Musk Disables Option to Report Misinformation On X/Twitter - X, formerly called Twitter, disabled its misinformation feature on the platform, effectively removing the option for users to report false election information, research organization Reset.Tech Australia reported on Wednesday. The company has recently been criticized for achieving the dubious honor as the number one platform to spread online hate and increased levels of misinformation.[…]

“It would be helpful to understand why X have seemingly gone backwards on their commitments to mitigating the kind of serious misinformation that has translated into real political instability in the US, especially on the eve of the ‘bumper year’ of elections globally,” Alice Dawkins, executive director of Reset, told Reuters.

See the previous story.

The Wrap by Natalie Korach - Elon Musk’s X Strips Article Headlines on Shared Links - X (formerly known as Twitter) has officially removed article headlines on links shared to the platform and viewed on its official mobile app, after owner Elon Musk warned the change was coming in August. 

Prior to the implemented change, when any account posted a link to X, the tweet would include a featured image, a headline and a brief description of the story. But now X cards only display the featured image from the article, without additional context.

Most people just read the headlines as they scroll through their social media feeds. This move has the effect of removing facts and truth from the platform, which is, of course, a primary goal of authoritarian figures. Beware of people who don’t want you to know things.

Bloomberg by Dave Lee - The Moral Case for No Longer Engaging With Elon Musk’s X - One thing the prior Twitter management didn’t do is actively make things worse. When Musk introduced creator payments in July, he splashed rocket fuel over the darkest elements of the platform. These kinds of posts always existed, in no small number, but are now the despicable main event. There’s money to be made. X’s new incentive structure has turned the site into a hive of so-called engagement farming — posts designed with the sole intent to elicit literally any kind of response: laughter, sadness, fear. Or the best one: hate. Hate is what truly juices the numbers.[…]

X is now an app that forcibly puts abhorrent content into users’ feeds and then rewards financially the people who were the most successful in producing it, egging them on to do it again and again and make it part of their living. Know this: As the scramble for attention increases, the content will need to become more violent, more tragic and more divisive to stand out. More car crashes, high school fights and public humiliation.

Journalists: Please. Please wean yourself of this abomination. You are the ones who are proping it up right now.

Ars Technica by Benj Edwards - Meta launches consumer AI chatbots with celebrity avatars in its social apps - During a presentation at Meta Connect 2023, the company said it is launching its own "Meta AI" chat assistant and a selection of AI characters across its messaging platforms, including WhatsApp, Instagram, and Messenger.

Meta's new AI assistant will likely feel familiar to anyone who has used chatbots like ChatGPT or Claude. It is designed as a general-purpose chatbot that Meta says can help with planning trips, answering questions, and generating images from text prompts. The assistant will also integrate real-time results from Microsoft's Bing search engine, giving it access to current information—similar to Bing Chat, ChatGPT's browsing plugin, and Google Bard.

Just got access to the AI chatbots via WhatsApp. I’ll have more to say about them after I get some experience interacting with them.

The Verge by Jay Peters - Artifact is becoming Twitter, too - Artifact, the AI-powered news app from Instagram’s co-founders, is adding a major new feature: the ability to post. So far, the app has been an aggregator for news and links from around the internet, but you’re going to be able to add posts directly to the app.

Mike Krieger, one of the co-founders of Artifact, announced the new features onstage in a conversation with Casey Newton at the Code Conference on Wednesday. The new feature is a logical next step from Artifact’s recently launched update that lets users share links. This new feature means you won’t just be limited to links; your posts can include things like a title, text, and photos. The posts will also have unique URLs, which should make them easier to share on different apps and services.

This is an improvement and I continue to play with Artifact a bit but this new feature is not particularly useful if you don’t have much of a following and unlike Threads, there’s no easy function to find the people you know. In that way, it’s a bit like a less-hyped BlueSky. My money’s still on Threads.

Vox by Allie Volpe - The messy art of posting through it - Each platform has its specific norms and users who have their own opinions on what content they consider too cringe or vulnerable for public consumption. For instance, when people express negative emotions on Facebook, it doesn’t seem so out of place, according to a 2017 study. On the contrary, Instagram is where users expect to see positive content — albeit content that isn’t particularly authentic. One study, from 2021, suggests the norms on TikTok empower users to embrace both difficult and positive experiences when they post.

However, as social media continues to occupy an increasingly intimate space in our lives, as Ysabel Gerrard, a senior lecturer in digital communication at the University of Sheffield, thinks it will, what we post — and how audiences interpret it — will shift. Gerrard, who studies young people’s experiences of social media and digital identities, says that when social platforms become a place to store meaningful memories, the way we post will only become more personal.

And even more personal when you add generative AI to the mix.

The Guardian by Zoe Corbyn - ‘Tech platforms haven’t been designed to think about death’: meet the expert on what happens online when we die - Our digital profiles and possessions are ever-expanding, but what happens to them after our deaths? Tech companies are yet to offer a satisfactory solution, says the technology researcher Tamara Kneese

Tamara Kneese studies how people experience technology. She is a senior researcher at New York-based nonprofit Data & Society Research Institute. Her new book, Death Glitch, examines what happens to our digital belongings when we die, and argues that tech companies need to improve how they deal with death on their platforms for the sake of all our digital posterity.

What happens to our digital profiles and assets are one thing but throw generative AI into the mix, and you have the potential that some approximation of your personality persisting beyond death. Who owns that personality?

Digiday by Kimeco McCoy - Marketers reconsider BeReal as it launches its first global brand marketing campaign

Do they, though?

Vice by Tess Owen - Schools Report Bomb Threats Following Libs of Tiktok Anti-LGBTQ Posts - At least eleven schools or school districts that were targeted by the account “Libs of TikTok” over anti-LGBTQ grooming conspiracies last month received bomb threats just days later. 

“Libs of TikTok”, helmed by former real-estate agent Chaya Raichik, has positioned itself as a vigilante crusader against “wokeness” in schools and culture—and has been heavily criticized as being a smokescreen for denigrating LGBTQ people 

This headline is confusing and doesn't make clear that the TikTok account in question is controlled by a far right group.

This is yet another attack on one of the foundational pillars of our democracy: Our public education system that produces the critical thinking upon which the informed electorate of democracies rely.

Do not view such attacks in isolation; they are part of a coordinated whole designed to destroy our democratic institutions and therefore the American experiment itself.

Share

Music Monday

Glorious Midjourney Mistakes

Images I created for this article that ended up on the cutting room floor.

There are so many good newsletters on Substack!

One of them I subscribe to is by , who has published six books of American history, among them The Death Of Reconstruction, How the South Won the Civil War, and her new book to be released on the Tuesday, Democracy Awakening.

I highly recommend her newsletter and look forward to reading her new book.

I mention her newsletter because as I was scrolling through the Substack app this weekend, it struck me that Richardson uses a naming convention for each edition of her newsletter that 99% of the time I would never recommend. The title of each edition of her newsletter is simply the date on which she presses Send.

I see this all the time with newsletters and there’s no need for it because subscribers’ inboxes already tell you the date an individual edition was sent. Here’s how Richardson’s newsletter looks in substack:

And here’s how it looks in my inbox:

As you can see, Richardson uses the publish date as the title of each edition but as you can also see, my inbox has a Received column which displays that very same date. Redundant.

The greatest factor that determines whether someone opens an email is whether or not the sender is known to them. In the case of newsletters, that’s not an issue because subscribers know the sender by virtue of the fact that they subscribed to the newsletter.

The second most important factor that determines whether or not someone opens an email is the subject line. The text that follows the date in the subject line is preview text. You can set that manually in Substack but if you don’t it will use the first line of text in your newsletter.

I’ve seen too many newsletters to count include the date and the name of the newsletter in the subject line, two pieces of information the recipient already knows. This does nothing to persuade the subscriber to actually open the email and read the newsletter. The subject line and the preview text should work hand-in-hand to do just that.

Another downside to naming your newsletter editions in this way is that it provides no hint of the content when someone wants to share it on social media. This is what it looked like when I went to share an article to LinkedIn via Buffer:

If I wanted to give my followers an idea of what I was sharing, I’d have to come up with my own headline for the article.

Richardson does not use the subject line to tease the contents of the newsletter. And yet it works for her. Why?

First, she’s an extremely popular historian, so she’s already got a huge existing and devoted audience. Here are the numbers:

• Her yet-to-be-released new book is already a best seller on Amazon due to pre-orders,

• Her newsletter boasts two million readers,

• She has 1.7 million followers on Facebook,

• 41,600 Instagram followers,

• 25,800 Threads followers,

• 52,200 YouTube subscribers, and

• 384,300 Twitter followers (dormant).

She’s got a huge audience that engages with her content:

They need no motivation to read what she has to say.

Finally, using a date as the title for each newsletter edition is consistent with the name of her newsletter: Letters from an American. Letters arrive in your mailbox with a dated postmark!

Refer a friend

Digital Marketing News

Advertising

Mashable by Matt Binder - X is no longer labeling ads for some users - Multiple users on X have recently reported advertisements showing up on their feed without any disclosure or labeling. The paid ads being served to users appear as organic posts (aka tweets) and show up like non-paid content, right in a user's feed.

This is illegal, right?

Bloomberg by Matt Day and Jake Rudnitsky - Amazon Prime Video Content to Include Ads Staring Early 2024 - Amazon.com Inc., following other streaming platforms looking to further monetize their content, will run ads on its Prime Video service in key markets – a move that will help offset rising costs and provide a boost to an already robust advertising business.

Ad-supported streaming will be the default on Prime Video in US, UK, Germany and Canada starting early next year, the company said in a statement on Friday. The company has long offered video streaming as part of a package that also includes speedy shipping, music and other perks. Amazon said Prime subscribers will continue to pay $139 annually in the US but will be able to pay an additional $2.99 a month to avoid ads. [Emphasis mine.]

We are headed toward the enshittification stage of the streaming video platforms.

Artificial Intelligence

VentureBeat by Sharon Goldman - Amazon announces new generative AI version of Alexa - “Alexa, let’s chat” — nearly a decade after debuting its voice-activated Echo devices, the e-commerce and cloud juggernaut Amazon today announced that its signature voice assistant Alexa is being upgraded with a new, custom-built large language model (LLM), taking advantage of the generative AI boom in Silicon Valley to give Alexa even more capabilities and human-like conversational qualities.[…]

The new Alexa LLM will be available as a free preview on Alexa-powered devices in the U.S. soon, and Amazon claims it is “smarter, more conversational” and its voice is more “realistic” and “casual.”

I’ve been waiting for this to happen. Smart speakers are about to get a lot more interesting.

Ars Technica by Benj Edwards - The AI-assistant wars heat up with Claude Pro, a new ChatGPT Plus rival - What does Claude have that ChatGPT doesn't? One big difference is a 100,000-token context window, which means it can process about 75,000 words at once. Tokens are fragments of words used while processing text. That means Claude can analyze longer documents or hold longer conversations without losing its memory of the subject at hand. ChatGPT can only process about 8,000 tokens in GPT-4 mode.

I’ve been really impressed with Claude. I’ve been using it to gain insights from technical research papers that can be dense and time-consuming to read. I’ve recently been doing research on an upcoming post on the psychology of the color red for the website design chapter of my book and it correctly cited academic researchers of the topic and let me interrogate what those researchers had to say about the color red. I cross-referenced the titles and authors with Google and they all checked out.

Futurism by Victor Tangermann - Microsoft Publishes Garbled AI Article Calling Tragically Deceased NBA Player "Useless" - Former NBA player Brandon Hunter passed away unexpectedly at the young age of 42 this week, a tragedy that rattled fans of his 2000s career with the Boston Celtics and Orlando Magic.

But in an unhinged twist on what was otherwise a somber news story, Microsoft's MSN news portal published a garbled, seemingly AI-generated article that derided Hunter as "useless" in its headline.

"Brandon Hunter useless at 42," read the article, which was quickly called out on social media. The rest of the brief report is even more incomprehensible, informing readers that Hunter "handed away" after achieving "vital success as a ahead [sic] for the Bobcats" and "performed in 67 video games."

A real-life example of an AI algorithm maligning the reputation of a dead man.

Search Engine Land by Barry Schwartz - Google Search can now validate responses from Bard - Google has updated Bard with a number of features, including the ability to double-check the responses provided by Bard by using Google Search. Google also has extensions for Gmail, Docs, Drive, Maps, YouTube, and Google Flights and Hotels for Google Bard. Bard has expanded Lens integration in 40 new languages and Bard upgraded its PaLM2 model to make it more capable, the company announced.

Can’t wait to see what additional value these extensions can unlock for me from my Gmail account and Google Docs.

VentureBeat by Sean Michael Kerner - Galileo offers new tools to explain why your AI model is hallucinating - San Francisco-based artificial intelligence startup Galileo is aiming to help its users to better understand and explain the output of large language models (LLMs), with a series of new monitoring and metrics capabilities that are being announced today. The new features are part of an update to the Galileo LLM Studio, which the company first announced back in June. Galileo was founded by former Google employees and raised an $18 million round of funding to help bring data intelligence to AI.

Good to know people are working to solve this problem.

Rest Of The World by Andrew Deck - Why Silicon Valley’s biggest AI developers are hiring poets - A string of job postings from high-profile training data companies, such as Scale AI and Appen, are recruiting poets, novelists, playwrights, or writers with a PhD or master’s degree. Dozens more seek general annotators with humanities degrees, or years of work experience in literary fields. The listings aren’t limited to English: Some are looking specifically for poets and fiction writers in Hindi and Japanese, as well as writers in languages less represented on the internet.

The companies say contractors will write short stories on a given topic to feed them into AI models. They will also use these workers to provide feedback on the literary quality of their current AI-generated text.

Revenge of the humanities majors.

Generative AI by Jim Clyde Monge - Dall-E3 Is Here And It’s Mind-Blowing - After ten months of silence, OpenAI has finally released Dall-E3, the latest version of its popular AI image generator. Dall-E3 is said to be a significant improvement with more nuance and details over Dall-E2.

What’s new in Dall-E3?

• Better image quality and coherence than Dall-E2

• Dall-E3 can generate legible text

• Dall-E3 works inside ChatGPT

• Improved safety

I’m looking forward to playing around with this! Looks like Midjourney might have some catching up to do.

Social Media

TechCrunch by Sarah Perez - Elon Musk says X will charge users ‘a small monthly payment’ to use its service - X owner Elon Musk...floated the idea that the social network formerly known as Twitter may no longer be a free site. In a live-streamed conversation with Israeli Prime Minister Benjamin Netanyahu…Musk said the company was “moving to a small monthly payment” for the use of the X system. He suggested that such a change would be necessary to deal with the problem of bots on the platform.

Hard pass.

Video

TechCrunch by Sarah Perez - YouTube debuts a new app, YouTube Create, for editing videos, adding effects and more - YouTube this morning introduced a new app for creators, YouTube Create, that will offer a suite of easy-to-use, free tools that will allow them to make both Shorts and longer videos. The tool aims to address some of the challenges creators face, including the editing process and the ability to leverage creative tools, including things like stickers, GIFs and effects.

The app will compete with TikTok’s CapCut video editing app and is initially available only for Android devices.

Music Monday

I love this song. It’s fun to play, too. What is it with the Irish and their capacity to make amazing music? (I think you know the answer.)

Glorious Midjourney Mistakes

Banner images for this article that didn’t make the cut.

Back in June, the United States Copyright Office reiterated its stance that AI—generated imagery is ineligible for copyright protections.

PetaPixel quotes Robert Kasunic of the USCO, stating “The Office will refuse to register works entirely generated by AI. Human authorship is a precondition to copyrightability.”

What, then, does “human authorship” mean?

Works that contain artificially-generated elements can be copyrighted but those AI-generated elements must be declared and are not eligible for copyright.

So, for example, say I shot my own photo of a busy downtown city scene and fed that into Midjourney as my base image but then used Midjourey to replace the background with a rolling rural fields to give it a Grant Woodsesque feel. I’d have to declare that those fields were artificially generated in my copyright application and probably also the city scene Midjourney created from my original photo.

I could copyright my admittedly crappy original photo but not the version I create via Midjourney that is based on the original.

The USCO’s stance to date renders all the imagery I’ve created for this newsletter using AI as ineligible for copyright protection.

AI Copyright Suits

The other side of this coin, of course, is the copyright infringement claims photographers, artists and celebrities like Sarah Silverman are making against AI companies for using their creative works to train their Large Language Models.

I’m definitely sensitive to those claims. I’ve been publishing a lot of original content online since 1993, including articles, photography, audio and video. No one ever asked me if they could use my content to train their AI models but I’ve no doubt they’ve done just that.

It is that sensitivity that caused me—after my initial experimentation with Midjourney—to refrain from using actual living artist names as prompts to create an image that mimics an artist’s style.

AI Transformation

A Midjourney-created entry in an art contest generated controversy last year when it won the competition.

Joseph Foley at Creative Bloq reports:

Since then, Allen has tried to copyright his prize-winning work. But this week, the US Copyright Office issued its third, final decision on the matter. It's view is that human authorship is a necessary prerequisite for copyright. And it's a decision that could impact on the popularity of AI art generators.

In his appeal, Allen had argued that an "essential element of human creativity” was needed to use Midjourney and that he had input “at least 624” text prompts before the image generator produced what he was looking for. He also argued that his work should be registered under fair doctrine because it was a transformative use of copyrighted material. “The underlying AI generated work merely constitutes raw material which Mr. Allen has transformed through his artistic contributions,” his appeal claimed.

These issues will sort themselves out in the courts but I’m inclined to agree more with the argument that generative AI is transformative, which is the key distinction applied to the Fair Use doctrine.

According to USCO:

“transformative” uses are more likely to be considered fair. Transformative uses are those that add something new, with a further purpose or different character, and do not substitute for the original use of the work.

Benedict Evans’ excellent piece titled Generative AI and intellectual property clarifies the matter:

But the real intellectual puzzle, I think, is not that you can point ChatGPT at today’s headlines, but that on one hand all headlines are somewhere in the training data, and on the other, they’re not in the model. 

OpenAI is no longer open about exactly what it uses, but even if it isn’t training on pirated books, it certainly uses some of the ‘Common Crawl, which is a sampling of a double-digit percentage of the entire web. So, your website might be in there. But the training data is not the model. LLMs are not databases. They deduce or infer patterns in language by seeing vast quantities of text created by people - we write things that contain logic and structure, and LLMs look at that and infer patterns from it, but they don’t keep it. So ChatGPT might have looked at a thousand stories from the New York Times, but it hasn’t kept them.

Moreover, those thousand stories themselves are just a fraction of a fraction of a percent of all the training data. The purpose is not for the LLM to know the content of any given story or any given novel - the purpose is for it to see the patterns in the output of collective human intelligence. 

Furthermore, throughout history creative works have always been iterative. Innovations developed by one artist are adopted by subsequent artists.

The innovations of the impressionist painters influenced the style of the pointillism and expressionist painters and on to cubism and so forth.

Ernest Hemingway’s journalism-inspired fiction of short, declarative sentences and realistic dialog influenced countless writers that followed him and just as profoundly, the dialog of Hollywood movies.

Incorporating previously established styles into your creative vision is a time-honored tradition and a crucial aspect that makes innovation possible.

It’s the very definition of transformative.

Share

AI Guided By A Human Hand

A federal judge recently ruled against Stephen Thaler, who brought suit challenging the government’s position that works made by AI are not copyrightable. According to The Hollywood Reporter:

Thaler, chief executive of neural network firm Imagination Engines. In 2018, he listed an AI system, the Creativity Machine, as the sole creator of an artwork called A Recent Entrance to Paradise, which was described as “autonomously created by a computer algorithm running on a machine.” […]

“In the absence of any human involvement in the creation of the work, the clear and straightforward answer is the one given by the Register: No,” Howell wrote.

I agree with the judge’s decision.

But that’s not to say I believe any AI-created work should be ineligible for copyright.

Regular readers of this newsletter will have noticed that I use Midjourney to create banner images for my articles.

The images I create using this tool should not be disqualified from copyright protection simply by virtue of the use of the tool. By that logic, any image generated using Photoshop should be disqualified.

I’m being a bit facetious here but the point is that AI is just a tool. A very powerful tool, but still just a tool. The tool needs a human hand (i.e. human creative vision) to produce anything.

When I use Midjourney to create images for these posts, I have a visual idea in mind of what I want to create. So I start with a creative vision that comes from me, not Midjourney.

Then I have to prompt Midjourney to create the image I want to produce. This is creative guidance. To get the image I want, to get the best result, I must be creative in the way I write my prompts.

I must innovate by trying different approaches to get the image I want. More creative guidance. I must iterate to refine my images. More creative guidance.

But Midjourney does all the work! Midjourney executes the image!, the naysayer amongst you might say (there’s one in every crowd, isn’t there?).

True, I’ll give you that. Yet but for me, there would be no image at all.

But you’re not creating those stunning images. All the talent is in Midjourney, the naysayer retorts.

But the talent in vision resides in me. Without my vision, there’d be no work.

I have a specific intent with the images I create using Midjourney, I have a specific artistic vision I want to achieve, I have the talent to guide the tool to execute my vision through the prompts I create, and all that results in my unique creative expression even if I did not apply brush to canvas.

Jeff Koons

So, naysayer, does this mean that Jeff Koons—the famous modern artist who conceives of an artistic vision and then directs his apprentices to create it—is not deserving of copyright protection?

No? Still not buying what I’m selling?

Try this on for size: Lila Shroff at The Reboot writes brilliantly about artists who are curating their own original datasets with which to train generative AI models so as to produce their own very specific artistic visions:

Stephanie Dinkins explores the possibilities for “small data” with “Not The Only One,” an embodied chatbot sculpture trained on the oral histories of three generations of women from a single family. In “The Zizi Show,” Jack Elwes’ deepfake drag cabaret highlights representational harms felt by queer communities. The datasets Elwes developed for the show are deliberately diverse and designed around a principle of consent. Finally, as part of a project exploring the migrational patterns of her Saudi and Iraqi ancestors, Nouf Aljowaysir’s “Salaf” explicitly investigates issues of dataset representation. When Aljowaysir performed an object-classification task on historical images of Bedouin lifestyles, the model she used routinely misidentified veiled women, confidently labeling them as “soldiers,” “army,” or other military paraphernalia. In protest, Aljowaysir used an image segmentation model to erase the misrepresentations from the archival images. She then trained a new model on the erased dataset to make visible the absent figures, signifying the “eradication of her ancestor’s collective memory.”

You cannot seriously tell me creative works produced with these AI models are not worthy of copyright.

Benedict Evans has some other examples:

At the simplest level, we will very soon have smartphone apps that let you say “play me this song, but in Taylor Swift’s voice”. That’s a new possibility, but we understand the intellectual property ideas pretty well - there’ll be a lot of shouting over who gets paid what, but we know what we think the moral rights are. Record companies are already having conversations with Google about this. 

But what happens if I say “make me a song in the style of Taylor Swift” or, even more puzzling, “make me a song in the style of the top pop hits of the last decade”? 

A person can’t mimic another voice perfectly (impressionists don’t have to pay licence fees) but they can listen to a thousand hours of music and make something in that style - a ‘pastiche’, we sometimes call it. If a person did that, they wouldn’t have to pay a fee to all those artists, so if we use a computer for that, do we need to pay them? I don’t think we know how we think about that. We might know what the law might say, but we might want to change that.

The Original Intent Of Copyright

Tabrez Sayed provides a nice summary of how copyright law has changed from its introduction in 1790 to the modern age, Disney and Steamboat Willie.

The original intent of copyright protection was to allow innovators to profit from their creations for a period of 14 years and another renewable period of 14 years if the creator was still alive.

Copyright protection provided a powerful incentive for innovation but after 28 years, those creations would enter the public domain, creating a base of knowledge that could contribute to subsequent innovations.

The reason for putting a time limit on copyright protections was to expand the body of publicly-available knowledge, thereby allowing for greater and greater innovation, which in turn would expand public knowledge exponentially. A virtuous circle of ever-renewing knowledge from which a nascent experiment in representative democracy would greatly benefit.

Subsequent amendments to copyright law allowed protection for a 56 year period of time (expanding the original 14 year periods of time to 28 years).

Steamboat Willie

In 1984, 56 years after Steamboat Willie was created, Disney’s original mouse was slated to enter the public domain.

Seeing this threat, Disney company lobbying resulted in the passage of the Copyright Act of 1976, allowing copyright protection for the whole life of the author plus half a century more, or 75 years if a corporation owned the creation.

This set the expiration date on Steamboat Willie at 2003, and yet Willie is still under copyright protection due to Disney’s additional efforts to get the Copyright Term Extension Act of 1998 passed, otherwise known as the “Mickey Mouse Protection Act.”

Steamboat Willie will finally enter the public domain next year but don’t be fooled into thinking that Disney has given up all rights to the character. It’s complicated, as the New York Times points out.

The long saga of Disney’s enslavement of Steamboat Willie illustrates just how far copyright law has drifted from its original mission of contributing to the public good.

Creative Commons

There is one fairly successful attempt at creating a copyright system that tries to reclaim some of that public good mission of the original.

It’s called the Creative Commons. According to the nonprofit’s website:

CC is an international nonprofit organization that empowers people to grow and sustain the thriving commons of shared knowledge and culture we need to address the world’s most pressing challenges and create a brighter future for all.

Together with our global community and multiple partners, we build capacity and infrastructure, we develop practical solutions, and we advocate for better sharing: sharing that is contextual, inclusive, just, equitable, reciprocal, and sustainable.

Creative Commons licenses are built into the infrastructure of the photo-sharing site Flickr and Wikipedia.

It allows creators to specify under what conditions and terms their content can be used.

USCO Seeks Public Comment

It should be clear by now that there is no clear-cut answer regarding the proper stance the United States Copyright Office should take on protection of AI-generated intellectual property.

Recognizing this, late last month the USCO asked for public input on the issues of AI generated content and copyright protection.

Benj Edwards at Ars Technica reports:

Specifically, the Copyright Office is interested in four main areas: the use of copyrighted materials to train AI models and whether this constitutes infringement; the extent to which AI-generated content should or could be copyrighted, particularly when a human has exercised some degree of control over the AI model’s operations; how liability should be applied if AI-generated content infringes on existing copyrights; and the impact of AI mimicking voices or styles of human artists, which, while not strictly copyright issues, may engage state laws related to publicity rights and unfair competition.

The deadline for written comments is October 18 and the deadline for reply comments is November 15. You can submit your comments at Copyright.gov.

Share The Reputation Algorithm

Digital News Articles

Artificial Intelligence

Wired by Steven Levy - What OpenAI Really Wants - For Altman and his company, ChatGPT and GPT-4 are merely stepping stones along the way to achieving a simple and seismic mission, one these technologists may as well have branded on their flesh. That mission is to build artificial general intelligence—a concept that’s so far been grounded more in science fiction than science—and to make it safe for humanity. The people who work at OpenAI are fanatical in their pursuit of that goal. (Though, as any number of conversations in the office café will confirm, the “build AGI” bit of the mission seems to offer up more raw excitement to its researchers than the “make it safe” bit.) These are people who do not shy from casually using the term “super-intelligence.” They assume that AI’s trajectory will surpass whatever peak biology can attain. The company’s financial documents even stipulate a kind of exit contingency for when AI wipes away our whole economic system.

A really good but long read about the history of OpenAI and the motivations of its employees.

New York Times by Tiffany Hsu - What Can You Do When A.I. Lies About You? - A.I. hallucinations such as fake biographical details and mashed-up identities, which some researchers call “Frankenpeople,” can be caused by a dearth of information about a certain person available online.

The technology’s reliance on statistical pattern prediction also means that most chatbots join words and phrases that they recognize from training data as often being correlated. That is likely how ChatGPT awarded Ellie Pavlick, an assistant professor of computer science at Brown University, a number of awards in her field that she did not win.

Same thing happened to me, as I explained in June👇.

This is the opposite of the Discoverability Crisis I wrote about last month; it’s being discovered but inaccurately.

The correlation of data with a specific person, organization, or brand is a new wrinkle that must be taken into account for reputation management practices.

AI Incident Database - Dedicated to indexing the collective history of harms or near harms realized in the real world by the deployment of artificial intelligence systems. Like similar databases in aviation and computer security, the AI Incident Database aims to learn from experience so we can prevent or mitigate bad outcomes.

Great resource.

The Verge by Mia Sato - Can news outlets build a ‘trustworthy’ AI chatbot? - a group of tech outlets is attempting to incorporate generative AI into its websites, though readers won’t find a machine’s byline anytime soon. On August 1st, an AI chatbot tool was added to Macworld, PCWorld, Tech Advisor, and TechHive, promising that readers can “get [their] tech questions answered by AI, based only on stories and reviews by our experts.”

The AI chatbot, dubbed Smart Answers, appears across nearly all articles and on the homepages of the sites, which are owned by media / marketing company Foundry. Smart Answers is trained only on the corpus of English language articles from the four sites and excludes sponsored content and deals posts. The user experience is similar to other consumer tools like ChatGPT: readers type in a question, and Smart Answers spits out a response. Alternatively, readers can select a query from an FAQ list, which is AI-generated but based on what people are asking and clicking on. Smart Answers responses include links to the articles from which information was pulled.

I was wondering when this would happen. I’d love to have enough time to train a LLM on all my writing so I could have a chatbot through which I could have a conversation with the content I’ve created over the years.

MIT Technology Review by Will Douglas Heaven - DeepMind’s cofounder: Generative AI is just a phase. What’s next is interactive AI. - DeepMind cofounder Mustafa Suleyman wants to build a chatbot that does a whole lot more than chat. In a recent conversation I had with him, he told me that generative AI is just a phase. What’s next is interactive AI: bots that can carry out tasks you set for them by calling on other software and other people to get stuff done. He also calls for robust regulation—and doesn’t think that’ll be hard to achieve.

Suleyman is not the only one talking up a future filled with ever more autonomous software. But unlike most people he has a new billion-dollar company, Inflection, with a roster of top-tier talent plucked from DeepMind, Meta, and OpenAI, and—thanks to a deal with Nvidia—one of the biggest stockpiles of specialized AI hardware in the world. Suleyman has put his money—which he tells me he both isn't interested in and wants to make more of—where his mouth is.

Interesting interview.

by - Centaurs and Cyborgs on the Jagged Frontier - For the last several months, I been part of a team of social scientists working with Boston Consulting Group, turning their offices into the largest pre-registered experiment on the future of professional work in our AI-haunted age. Our first working paper is out today. There is a ton of important and useful nuance in the paper but let me tell you the headline first: for 18 different tasks selected to be realistic samples of the kinds of work done at an elite consulting company, consultants using ChatGPT-4 outperformed those who did not, by a lot. On every dimension. Every way we measured performance.

Consultants using AI finished 12.2% more tasks on average, completed tasks 25.1% more quickly, and produced 40% higher quality results than those without. Those are some very big impacts. Now, let’s add in the nuance.

Excellent piece! Must read for anyone considering implementing AI within their organization and also as a framework for using AI individually.

Happy Hour

Me - The Quest For The Perfect Spicy Margarita - My shaker is a fine object to behold. Known as a “cobbler shaker,” it is a tall, sleek object of gleaming stainless steel that sits upon an accented copper base. Seen from the ground level, it would be perfectly at home as an obelisk-esque object in a Kubrick film.

This is my finely-tuned recipe for a spicy margarita.

Podcasting

TechCrunch by Sarah Perez - YouTube to support RSS uploads for podcasters by year-end, plus private feeds in YouTube Music - the platform will be rolling out support for RSS uploads for podcasters by the end of the year, among other updates. The new functionality had been in beta testing since earlier this year, as a strategic, invite-only pilot.

YouTube additionally confirmed to TechCrunch it’s rolling out support for podcasts on YouTube Music by the end of year, as well.

Another channel with massive potential to grow your podcasting audience.

Social Media

New York Times by Ezra Klein - Elon Musk Got Twitter Because He Gets Twitter - Jack Dorsey, a Twitter co-founder and former chief executive, always wanted it to be something else. Something it wasn’t, and couldn’t be. “The purpose of Twitter is to serve the public conversation,” he said in 2018. Twitter began “measuring conversational health” and trying to tweak the platform to burnish it. Sincere as the effort was, it was like those liquor ads advising moderation. You don’t get people to drink less by selling them whiskey. Similarly, if your intention was to foster healthy conversation, you’d never limit thoughts to 280 characters or add like and retweet buttons or quote-tweet features. Twitter can’t be a place to hold healthy conversation because that’s not what it’s built to do.

So what is Twitter built to do? It’s built to gamify conversation. As C. Thi Nguyen, a philosopher at the University of Utah, has written, it does that “by offering immediate, vivid and quantified evaluations of one’s conversational success. Twitter offers us points for discourse; it scores our communication. And these gamelike features are responsible for much of Twitter’s psychological wallop. Twitter is addictive, in part, because it feels so good to watch those numbers go up and up.”

A precise diagnosis of effect Twitter has had on our public, and our public policy, discussions. Gamification is no way to shape public policy.

BGR by Joe Wituschek - X officially takes on LinkedIn with its new job posting feature - an app researcher found indications that X (formerly known as Twitter) was looking to take on Microsoft’s social media network for professionals by bringing job postings to its platform. At the time (when X was actually still Twitter), it appeared that the feature would allow verified organizations to integrate job postings from an Applicant Tracking System or XML feed into their Twitter profile.

Sure, if you want to hire fascists.

TechCrunch by Sarah Perez - YouTube demystifies the Shorts algorithm, views and answers other creator questions - According to the product lead for Shorts, Todd Sherman, the Shorts algorithm differs from the long-form algorithm where people are tapping on videos to watch — essentially making a specific choice that then drives more recommendations. But on Shorts, people are swiping through content not knowing what comes next. While both recommendation systems are designed to present videos that people will value and enjoy, the Shorts feed prioritizes a more diverse feed because people are flipping through hundreds of videos versus maybe 10 or 20 in long-form.

Sherman also noted that not every flip in Shorts is counted as a view, either — a difference from some other platforms where viewing the first frame is counted as a view. (TikTok counts views as soon as a video starts to play, it’s said). On Shorts, however, the view is meant to reflect that the user had some intent to watch so creators have some “meaningful threshold” that someone meant to watch the view. […]

“There are parts of the algorithm that try and find people, find creators an audience,” Sherman explained. “And sometimes those algorithms will go and effectively find like a seed audience, find a set of people that may enjoy your video. And depending on how that goes, it may get a lot more traffic or it may taper off,” he said.

This last graph illustrates how digital platforms interpret the reputations of their users, in this case the users’ video consumption reputations.

TechCrunch by Ivan Mehta - X launches Community Notes for videos - Social network X, formerly known as Twitter, has introduced Community Notes for videos. Community Notes is an existing program for crowdsourced moderation. The Elon Musk-owned platform announced that notes by contributors attached to a video will show up in all posts with that video.

“Notes written on videos will automatically show on other posts containing matching videos. A highly-scalable way of adding context to edited clips, AI-generated videos, and more,” the company said in a post.

This is actually a pretty clever innovation. I’d like to see other platforms adopt it because Twitter/X has become a fascist cesspool.

Web Development

WordPress - Introducing the 100-Year Plan: Secure Your Online Legacy for a Century - The 100-Year Plan isn’t just about today. It’s an investment in tomorrow. Whether you’re cementing your own digital legacy or gifting 100 years of a trusted platform to a loved one, this plan is a testament to the future’s boundless potential.

The cost is $38,000.

Not a surprise at all. This was inevitable.

TechCrunch by Sarah Perez - Inalife’s digital legacy platform lets you preserve your family’s memories - A new startup, Inalife, has created a digital legacy platform that aims to help families preserve their precious memories, create an interactive family tree and even record messages for other members to be played at a date in the future — like when a new baby grows up and reaches adulthood, for example.

Are we seeing a trend here?

wpbeginner - What Is Google’s INP Score and How to Improve It in WordPress - Google Core Web Vitals are website performance metrics that Google considers important for overall user experience. These web vital scores are part of Google’s overall page experience score, which will impact your SEO rankings. […]

INP stands for ‘Interaction to Next Paint’. It is a new Google Core Web Vital metric that measures the user interactions that cause delays on your website.

The INP test measures how long it takes between a user interacting with your website, like clicking on something, and your content visually updating in response. This visual update is called the ‘next paint’.

This is a user experience reputation metric.

Music Monday

Girl Talk: Copyright infringement or transformative fair use?

Glorious Midjourney Mistakes

The banner images I created for this article but ended up on the cutting room floor.

My favorite color is blue. Turns out, I’m not alone.

A 2015 YouGov survey of 10 countries and four continents found that:

one color – blue – is the most popular across the board. Between 23% (in Indonesia) and 33% (in Great Britain) like blue most out of the colors listed, putting it 8-18 points ahead of any other colour.

But maybe we shouldn’t leap to blanket conclusions about the popularity of specific colors.

A 2017 online survey that solicited 30,000 submissions from more than 100 countries found that the color teal was preferred by most respondents.

I mean, teal is kinda blue, right? It’s blue/green.

But a 2019 study published in the journal Perception found that color preferences can diverge wildly among different cultures.

If we turn to the Google Ngram Viewer, which allows us to search for the frequency of words cited in books from 1500 to 2019, we can see that the word blue is rarely the most cited color across time:

We can conclude, then, that the popularity of a given color is dependent upon context.

For our purposes, though, the popularity of colors is less important than the meanings and associations they convey.

What Does The Color Blue Mean?

Scientifically, blue refers to colors humans perceive when light with a dominant wavelength between approximately 450 and 495 nanometres hits our eyes.

Shades of blue with a higher frequency and therefore a shorter wavelength will gradually look more violet. Shades of blue with a lower frequency and a longer wavelength will appear more green. Pure blues are found in the middle of this range, at about 470 nanometres.

What Does The Color Blue Symbolize?

Symbolism for the color blue takes many forms and can be employed in many different situations, depending upon the context in which the color is used.

Read more