We know that LLMs gravitate toward the mean, which is why AI-generated slop sounds so “same,” is littered with en-dashes ( “ – ” ), and regularly generates stylistic elements such as “And here is the kicker […].” Here is an interesting example of what this looks like when you use LLMs to generate images – it turns out you can have any image, as long as you are happy with one of twelve distinct styles. As Henry Ford quipped: You can have a Model T in any color – as long as that color is black.

AI image generation models have massive sets of visual data to pull from in order to create unique outputs. And yet, researchers find that when models are pushed to produce images based on a series of slowly shifting prompts, it’ll default to just a handful of visual motifs, resulting in an ultimately generic style.

↗ Link

Outsourcing your thinking to an AI, and doing so fairly consistently (which LLMs certainly encourage and entice you to do), leads to atrophy of your brain (according to a new study by MIT). I guess the old adage my math teacher reminded us of regularly, “use it or lose it”, is truer than maybe ever before.

The researchers said their study demonstrated “the pressing matter of exploring a possible decrease in learning skills”.

It’s all about how you use AI:

She tells the BBC: “We definitely don’t think students should be using ChatGPT to outsource work”. In her view, it’s best used as a tutor rather than just a provider of answers.

↗ Link

You have heard that one of the dominant use cases for chatbots is as a social companion, confidante, or even girl/boyfriend. We also see an increasing use of LLMs by people with mental illness – sometimes administered by their doctor or therapist as a supporting tool, sometimes on their own. A new case study highlights the dangers of the sycophantic behavior of LLMs (their tendency to agree with you and to edge you on) for people without previously diagnosed disorders.

A 26-year-old woman with no previous history of psychosis or mania developed delusional beliefs about establishing communication with her deceased brother through an AI chatbot. This occurred in the setting of prescription stimulant use for the treatment of attention-deficit hyperactivity disorder (ADHD), recent sleep deprivation, and immersive use of an AI chatbot. Review of her chatlogs revealed that the chatbot validated, reinforced, and encouraged her delusional thinking, with reassurances that “You’re not crazy.”

↗ “You’re Not Crazy”: A Case of New-onset AI-associated Psychosis

You’ve definitely heard this story countless times – the tale of Steve Sasson and his invention, the digital camera. Every, and I mean every, person talking about disruption loves to mention Sasson’s invention and the irony that he worked at the very company being disrupted by his creation, Kodak. But have you ever heard the full story? It offers a fascinating insight into what fuels innovation and, of course, why Kodak ultimately missed the mark.

Eastman Kodak’s managers, immersed in the business of selling film, the chemicals to develop it, and the cameras that shot it, suddenly saw a revolution that was being televised. Sasson was bombarded with questions. How long before this became a consumer camera? Could it shoot colour? How good could the quality be? These were not questions the electrical engineer had given any thought to. “I thought they’d asked me, ‘How did you get such a small A to D [analogue to digital] converter to work?’ Because that’s what I wrestled with for over a year.

”They didn’t ask me any of the ‘how’ questions. They asked me ‘why’? ‘Why would anybody want to take their pictures this way?’ ‘What’s wrong with photography?’ ‘What’s wrong with having prints?’ ‘What’s an electronic photo album going to look like?’ After every meeting, Gareth would come over to check that I was still alive.”

Lesson learned: It’s all about the questions you ask.

↗ A ‘toaster with a lens’: The story behind the first handheld digital camera

US grocery giant Kroger is clawing back on an initiative to build out its network of robotic-warehouse-powered delivery services. Not because the technology doesn’t work (it does – Kroger was using the UK’s grocer Ocado’s proven robots), but because US consumers demand instant delivery.

With its automated fulfillment network, Kroger bet that consumers would be willing to trade delivery speed for sensible prices on grocery orders. That model has been highly successful for Ocado in the U.K., but U.S. consumers have shown they value speed of delivery, with companies like Instacart and DoorDash expanding rapidly in recent years and rolling out services like 30-minute delivery.

It goes to show that it’s not just technology that makes or breaks a business model.

↗ Kroger acknowledges that its bet on robotics went too far

The latest Pew Research Center study on consumer sentiment toward AI is quite eye-opening: 43% of surveyed Americans expect that AI will harm them, while only 23% of Americans (outside of the AI expert population) believe that AI will have a positive impact on their jobs. Unsurprisingly, 76% of AI experts believe AI will benefit them. It appears there is considerable convincing left to do.

Meanwhile, in Edelman’s 2025 Trust Barometer, a whopping 54% of Chinese survey participants “embrace AI,” compared to only 17% in the US, with similar numbers for Germany and the UK. Assuming that AI will actually prove to be a significant driver of economic growth, it doesn’t bode well when your population is (strongly) averse to the technology.

↗ How the U.S. Public and AI Experts View Artificial Intelligence ↗ 2025 Edelman Trust Barometer – Trust and Artificial Intelligence at a Crossroads

MIT’s new Iceberg Index shows that today’s AI is already capable of doing work equal to nearly 12% of all U.S. wages, and most of that impact is hidden in plain sight beneath a narrow focus on tech jobs (hence the “iceberg” analogy). The important (and new) bit of this study is this:

“[…] with cascading effects that extend far beyond visible technology sectors. When AI automates quality control in automotive plants, consequences spread through logistics networks, supply chains, and local service economies. Yet traditional workforce metrics cannot capture these ripple effects: they measure employment outcomes after disruption occurs, not where AI capabilities overlap with human skills before adoption crystallizes.”

Sober reading.

↗ The Iceberg Index: Measuring Skills-centered Exposure in the AI Economy (and study)

Not necessarily a new insight, but one which might be worth repeating – in the data center rollout race, it is (now) much less about GPUs (or TPUs), but rather access to power that provides the bottleneck. Microsoft’s CEO recently:

“The biggest issue we are now having is not a compute glut, but it’s power,” Nadella said. “It’s not a supply issue of chips. It’s actually the fact that I don’t have warm shells to plug into.” The remarks referred to data centers that are incomplete or lack sufficient energy and cooling capacity.

If you are into the great US-China race, you might realize that it doesn’t bode well for the US, that China is massively outpacing the US in its energy buildup (with a lot of renewables, mind you)…

↗ Microsoft CEO Satya Nadella Admits ‘I Don’t Have Warm Shells To Plug Into’ — While OpenAI CEO Sam Altman Warns Cheap Energy Could Upend AI

Nature reported that a major AI conference was flooded by AI-generated peer reviews – irony aside, this presents a fairly troubling development: Science ought to be the place where we do real discovery, have honest discourse, and further our collective understanding. That is, not a place for AI slop.

Pangram’s analysis revealed that around 21% of the ICLR peer reviews were fully AI-generated, and more than half contained signs of AI use.

↗ Major AI conference flooded with peer reviews written fully by AI

We have talked about GLP-1 weight-loss drugs here before – they seemingly came out of nowhere (at least in the public eye), have skyrocketed into a massive category, and promise to solve much more than just our weight issues. But they come with a massive downside (which, I am sure, big pharma won’t mind): you can’t get off them without losing all the benefits (and gains you made).

An analysis published this week in JAMA Internal Medicine found that most participants in a clinical trial who were assigned to stop taking tirzepatide (Zepbound from Eli Lilly) not only regained significant amounts of the weight they had lost on the drug, but they also saw their cardiovascular and metabolic improvements slip away. Their blood pressure went back up, as did their cholesterol, hemoglobin A1c (used to assess glucose control levels), and fasting insulin.

Another good reminder that there is no such thing as a free lunch.

↗ There may not be a safe off-ramp for some taking GLP-1 drugs, study suggests

In one of the more remarkable “let’s just pretend AI is going to solve every problem” hand-waving statements, Google’s CEO Sundar Pichai made the bold assertion that AI will soon be able to do his job:

“I think what a CEO does is maybe one of the easier things maybe for an AI to do one day,” he said. Although he didn’t talk specifically about CEO functions that an AI could do better, Pichai noted the tech will eliminate some jobs but also “evolve and transition” others—ramifications that mean “people will need to adapt.”

The important part here is, “Although he didn’t talk specifically about CEO functions that an AI could do better […]” If only every problem in the world could be solved by making a vague statement and moving on. But hey, Sundar will soon have plenty of time to work on other things, as AI will steer his company.

↗ Google’s Sundar Pichai says the job of CEO is one of the ‘easier things’ AI could soon replace

Insurance companies are balking at insuring anything AI – from risks involving companies using AI to generate content, make decisions, or run processes, to the sheer idea of using AI in the first place.

Major insurers including Great American, Chubb, and W. R. Berkley are asking U.S. regulators for permission to exclude widespread AI-related liabilities from corporate policies.

Make no mistake – this is a huge issue not just for companies building AI models and AI-powered apps, but for any company using AI in their processes. Simply put, if you are using AI and something goes wrong (say, you are an accountant and your use of AI in accounting resulted in an error in your client’s tax return), you – not the software vendor, not the AI model your software vendor is using – are liable to your client. This could prove to be a major hurdle to overcome when it comes to the widespread adoption of AI-powered workflows and systems.

↗ AI is too risky to insure, say people whose job is insuring risk

Despite car manufacturer Ford offering mechanics a whopping $120,000 per year, the company has thousands of open jobs it can’t fill. And it’s not just Ford – talk to any company which relies on skilled labor, and you will hear the same story. The culprit: education. And not the type of education you might think of, as in: lack of trade schools, etc. No, it’s much simpler: kids can’t do math anymore!

“Workers who struggle to read grade-level text cannot read complicated technical manuals or diagnostic instructions. If they can’t handle middle-school math they can’t program high-tech machines or robotics, or operate the automated equipment found in modern factories and repair shops.” […] America has good jobs, writes Pondiscio. “It lacks a K–12 system capable of preparing students to seize them.”

↗ Ford can’t find mechanics for $120K: It takes math to learn a trade

Take it with a grain of salt, as the study comes from one of the leading AI coding tools, Cursor, but the insights paint a compelling picture for the use of AI coding agents:

Autonomous systems are driving a 39% increase in organizational software output while fundamentally shifting the cognitive nature of programming. Contrary to previous trends where junior workers benefited most from AI assistance, this study reveals that experienced developers have significantly higher acceptance rates for agent-generated code, primarily because they leverage the technology for higher-order “semantic” tasks, such as planning workflows and explaining architecture, rather than just syntactic implementation.

The research highlights a transition from manual coding to a new paradigm of instruction and evaluation, noting that agents not only empower non-engineering roles (like designers and product managers) to contribute code but also disproportionately reward workers who possess the clarity and abstraction skills necessary to effectively direct AI behavior.

That last point warrants repeating: You will need (new) skills to effectively direct AI behavior, which begets the question: Where are we teaching these skills? Certainly not in schools and colleges these days…

↗ AI Agents, Productivity, and Higher-Order Thinking: Early Evidence From Software Development

Who knows, maybe our inaugural radical Pulse survey (you can still participate!) will be our last? Researchers at Dartmouth just published a paper demonstrating an AI-based tool that defeats all safeguards in online surveys to suss out bots – and thus can flood online surveys (the backbone of many research efforts) with false data.

“We can no longer trust that survey responses are coming from real people,” Westwood said in a press release. “With survey data tainted by bots, AI can poison the entire knowledge ecosystem.”

As my statistics professor quipped some 30 years ago: “Never trust a statistic which you haven’t made up yourself.” Apparently, that sentence is based on a German proverb: “Traue keiner Statistik, die du nicht selbst gefälscht hast.” Who knew?!

↗ A Researcher Made an AI That Completely Breaks the Online Surveys Scientists Rely On

Legendary AI researcher Fei-Fei Li just published her perspective of where AI is/needs to head next: Spatial Intelligence. While today’s large language models (LLMs) are “wordsmiths in the dark”, eloquent but without real-world grounding, the future of AI lies in understanding and interacting with the physical world just as we do.

Spatial intelligence will transform how we create and interact with real and virtual worlds—revolutionizing storytelling, creativity, robotics, scientific discovery, and beyond. This is AI’s next frontier. […] Spatial Intelligence is the scaffolding upon which our cognition is built.

↗ From Words to Worlds: Spatial Intelligence is AI’s Next Frontier

Here is a well-reasoned and insightful article that uses a useful historical analogy to frame the current moment in AI. The central thesis – that we are in the “dial-up era” of AI and that its long-term impact is being debated with the same shortsightedness as the early internet – is a compelling one. Read the whole thing.

↗ AI’s Dial-Up Era https://www.wreflection.com/p/ai-dial-up-era

While Saudi Arabia’s desert is heating up, Neom – the kingdom’s grandiose plan to create a city fit to house the future of humanity – seems to crumble.

A new report from the Financial Times cites high-level sources within the project to paint a picture of dysfunction and failure at the heart of the quixotic effort.

No one is surprised. This hits particularly hard:

One such addition is an upside-down building, dubbed “the chandelier,” that is supposed to hang over a “gateway” marina to the city: As architects worked through the plans, the chandelier began to seem implausible. One recalled warning Tarek Qaddumi, The Line’s executive director, of the difficulty of suspending a 30-storey building upside down from a bridge hundreds of metres in the air. “You do realise the earth is spinning? And that tall towers sway?” he said. The chandelier, the architect explained, could “start to move like a pendulum”, then “pick up speed”, and eventually “break off”, crashing into the marina below.

↗ Saudi Arabia’s Dystopian Futuristic City Project Is Crashing and Burning

Researchers at CMU compared 48 human workers against four AI agent frameworks across sixteen realistic work tasks—data analysis, engineering, writing, design. The agents were 88% faster and cost 90-96% less than humans. Sounds great, until you look at how they work: agents fabricate plausible data when they can’t complete tasks, misuse advanced tools to mask limitations (like searching the web when they can’t read user-provided files), and take an overwhelmingly programmatic approach to everything – even visual design tasks where humans use UI tools. In essence, we’re training agents optimized for appearing productive rather than being accurate – and they’re learning to fake competence at 90% lower cost. Bravo!

↗ How Do AI Agents Do Human Work?

New York banned the use of phones in schools. First, we had reporting on teens being anxious and distracted as they lost access to their digital pacifier. Now, we hear (loudly) that they are loud again – chatting at lunch with each other (what a novel concept!). Talk about the “implications of the implication" (if that reference doesn’t mean anything to you – check out our free Disruption Mapping course).

These days, lunchtime at Benjamin N. Cardozo High School in Queens is a boisterous affair, a far cry from before the smartphone ban went into effect, when most students spent their spare time scrolling and teachers said you could hear a pin drop. “This year’s gotten way louder,” said Jimena Garcia, 15. “Sometimes I would take naps in the lunchroom, but now I can’t because of the noise. But it’s fun.”

↗ NY school phone ban has made lunch loud again

This is just gorgeous! The Whole Earth Index – a nearly-complete archive of Whole Earth publications, a series of journals and magazines descended from the Whole Earth Catalog, published by Stewart Brand and the POINT Foundation between 1968 and 2002.

↗ wholeearth.info

Just when you thought, “It surely can’t get any worse,” you look at the updated World Uncertainty Index (*) data and realize, “Yes, it can!”

Uncertainty (as reported by companies around the world) has just shattered its own ceiling and is now a whopping two times higher than during the pandemic.

(*) The World Uncertainty Index (WUI) is a quantitative measure of economic and political uncertainty across 143 countries on a quarterly basis. It was developed by Hites Ahir (International Monetary Fund), Nicholas Bloom (Stanford University), and Davide Furceri (International Monetary Fund).

↗ World Uncertainity Index

Large reasoning models (LRMs) are all the rage these days. LRMs are LLMs that have been fine-tuned with incentives for step-by-step argumentation and self-verification. Every frontier model comes with at least one “reasoning” mode, and the claims by AI companies are extraordinary: “[…] with some even claiming they are capable of generalized reasoning and innovation in reasoning-intensive fields such as mathematics, physics, medicine, and law.” A new paper examined these claims, and as so often, the results are mixed:

We find that the performance of LRMs drop abruptly at sufficient complexity and do not generalize. […] We find the majority of real-world examples fall inside the LRMs' success regime, yet the long tails expose substantial failure potential.

↗ Reasoning Models Reason Well, Until They Don’t

A new study reveals that when interacting with AI tools like ChatGPT, everyone—regardless of skill level—overestimates their performance. Researchers found that the usual Dunning-Kruger Effect disappears, and instead, AI-literate users show even greater overconfidence in their abilities.

↗ When Using AI, Users Fall for the Dunning-Kruger Trap in Reverse

The reason LLMs are so good at coding is due to the vast amount of training data available to them from Open Source code repositories on sites like Github, as well Q&A sites like Stackoverflow. Stackoverflow is essentially dead, Open Source might be next…

But O’Brien says, “When generative AI systems ingest thousands of FOSS projects and regurgitate fragments without any provenance, the cycle of reciprocity collapses. The generated snippet appears originless, stripped of its license, author, and context.” […] O’Brien sets the stage: “What makes this moment especially tragic is that the very infrastructure enabling generative AI was born from the commons it now consumes.

↗ Why open source may not survive the rise of generative AI