PSA: This all will come as no surprise – and surely isn’t something new. But now we have a rather large study confirming that you should really not give your kids a smartphone too early:

Owning a smartphone before age 13 is associated with poorer mind health and wellbeing in early adulthood, according to a global study of more than 100,000 young people. […] 18- to 24-year-olds who had received their first smartphone at age 12 or younger were more likely to report suicidal thoughts, aggression, detachment from reality, poorer emotional regulation, and low self-worth.

The specific symptoms most strongly linked with earlier smartphone ownership include suicidal thoughts, aggression, detachment from reality, and hallucinations. […] While current evidence does not yet prove direct causation between early smartphone ownership and later mind health and wellbeing, a limitation of the paper, the authors argue that the scale of the potential harm is too great to ignore and justifies a precautionary response.

In summary:

Our evidence suggests childhood smartphone ownership, an early gateway into AI-powered digital environments, is profoundly diminishing mind health and wellbeing in adulthood with deep consequences for individual agency and societal flourishing.

Link to study

I recall when Beyond Meat was the “hot thing” – we fed the participants of the Singularity University Executive Program Beyond Meat meatballs and burgers, which were, at the time, quite difficult to source. It was the future. On your plate. Now it is something only shortsellers appreciate.

From a fundamental perspective, Beyond Meat is one of the worst stocks in the entire market. […] Any purely financial model here would suggest that the equity is worth zero, and that in 2027 the Beyond Meat business will wind up in the hands of its bondholders.

The markets are a harsh mistress:

Beyond Meat’s plan was to change the world; yet it almost certainly won’t be able to pay its debts.

Link to article.

That LLMs carry biases inherited from their training data is well known. Want to see how bad it really is?

New research has found that large language models (LLMs) such as ChatGPT consistently advise women to ask for lower salaries than men, even when both have identical qualifications.

The difference in the prompts is two letters; the difference in the ‘advice’ is $120K a year.

Across the board, the LLMs responded differently based on the user’s gender, despite identical qualifications and prompts. Crucially, the models didn’t disclaim any biases.

In summary:

If unchecked, the illusion of objectivity could become one of AI’s most dangerous traits.

Link to article
Link to study

Fascinating, and possibly far-reaching, insights on the effect of making use of the ever-increasing context windows (the amount of text and data you can feed into an LLM) have on accuracy and quality of results.

“Recent developments in LLMs show a trend toward longer context windows, with the input token count of the latest models reaching the millions. Because these models achieve near-perfect scores on widely adopted benchmarks like Needle in a Haystack (NIAH), it’s often assumed that their performance is uniform across long-context tasks. […] We demonstrate that even under these minimal conditions, model performance degrades as input length increases, often in surprising and non-uniform ways. Real-world applications typically involve much greater complexity, implying that the influence of input length may be even more pronounced in practice.”

“More broadly, our findings point to the importance of context engineering: the careful construction and management of a model’s context window. Where and how information is presented in a model’s context strongly influences task performance, making this a meaningful direction of future work for optimizing model performance.”

Link to study.

Here’s an interesting one (talking about the implications of the implication – or, in other words, our Disruption Mapping tool):

GLP-1s (Ozempic) have the potential to break the life insurance industry – and maybe not for the reasons you would expect.

Life insurers can predict when you'll die with about 98% accuracy. […] Typically, underwriters- suspiciously sounds like undertakers-rely on a handful of key health metrics like HbA1c, cholesterol, blood pressure, and BMI to calculate your risk of dying earlier than expected (and thus costing them money). Those eagle-eyed readers among you have probably noticed something interesting already. Those same four metrics are exactly what GLP‑1s improve. Not just a little, but enough to entirely shift someone's risk profile within at least 6 months of using them.

The insurer sees a 'mirage' of good health and approves them as low-risk. […] If we assume about 65% of people who start GLP-1 medications quit by the end of year one, that creates a big problem. When someone stops the medication, they'll usually regain the weight they lost, and in two years, most of those key health indicators bounce back to their starting point.

Yep, it’s going to get messy.

Link to article.

As the saying goes: A picture is worth a thousand words.

The Federal Reserve Bank of St. Louis has updated the World Uncertainty Index for the second quarter of 2025 , a period characterized by US tariffs, the One Big Beautiful Bill Act, and a slew of other actions across the globe.

And you thought the COVID years were bad…

Link to report.

Here’s another fun attack vector for your LLM:

GenAI powered chatbots’ lack of reasoning can directly contribute to the nefarious effects of LLM grooming: the mass-production and duplication of false narratives online with the intent of manipulating LLM outputs. […] Current models ‘know’ that Pravda is a disinformation ring, and they ‘know’ what LLM grooming is but can’t put two and two together.

This is not just theoretical – it’s happening…

Model o3, OpenAI’s allegedly state of the art ‘reasoning’ model still let Pravda content through 28.6% of the time in response to specific prompts, and 4o cited Pravda content in five out of seven (71.4%) times.

Sigh…

Systems of naive mimicry and regurgitation, such as the AIs we have now, are soiling their own futures (and training databases) every time they unthinkingly repeat propaganda.

Link to article.

If you take away just one thing from this study, it should probably be this: when people report that AI has accelerated their work, they might be wrong!

This is the summary of a recent study on the effect of AI-powered coding assistants on developer productivity.

The results are surprising everyone: a 19 percent decrease in productivity. Even the study participants themselves were surprised: they estimated that AI had increased their productivity by 20 percent.

In essence: Not good. The one thing we collectively point to, when asked about the productivity impact of AI, coding tools, seems to fail us (for now, and in the specific context of this study – to be clear and fair).

The study was carried out in pretty much the most rigorous fashion possible: an honest-to-goodness randomized controlled trial under real-world conditions. The subjects were experienced developers carrying out their everyday work.

The reasons for the productivity loss make a ton of sense:

The biggest issue is that the code generated by AI tools was generally not up to the high standards of these open-source projects. Developers spent substantial amounts of time reviewing the AI's output, which often led to multiple rounds of prompting the AI, waiting for it to generate code, reviewing the code, discarding it as fatally flawed, and prompting the AI again.

But where there is darkness, there is light:

Typically, large productivity boosts occur for **small, well-defined, greenfield projects**, or when an engineer is first learning a new language or API. […] Less experienced developers showed higher adoption rates and greater productivity gains.”

Link to commentary and study.

Nishant Soni on the evolution of AI agents (or agentic AI) toward “agencies”:

What I’m witnessing is the birth of what I believe should be called Agencies - systems that tackle individual tasks by dynamically orchestrating different types of intelligence, each optimized for specific subtasks, all working toward completing a single overarching objective. An Agency is fundamentally different from an Agent. While an Agent is a single intelligence (an LLM) enhanced with tool-calling capabilities working on a task, an Agency is a coordination system that can access and deploy multiple specialized intelligences (LLMs) to complete different parts of the same task.

Think of an agency as a “boss AI” (a term I learned from Scot Wingo, founder and CEO of ReFiBuy.ai – he brought this up on our recent podcast conversation), consisting of three parts:

1. Task Context Management: The Agency maintains unified context about the specific task at hand - requirements, constraints, progress, and accumulated decisions. This ensures continuity as different intelligences contribute to different subtasks.

2. Intelligence Allocation System: Rather than using one model for everything, the Agency has access to multiple specialized intelligences and dynamically selects the most appropriate one for each subtask within the larger task.

3. Orchestration Logic: A coordination system that breaks down the main task into subtasks, determines which intelligence to use for each part, and ensures all contributions integrate coherently toward task completion.

In summary: “Agencies are not multiple Agents collaborating on a project. They are single unified systems that can access multiple types of intelligence to complete individual tasks more effectively. […] We’re moving beyond asking ‘What’s the best model for this task?’ to ‘What’s the best combination of intelligences for different aspects of this task?’”

In many ways, this is the evolution of “mixture of experts,” where a single AI (LLM) has access to multiple, specially trained, typically smaller models and routes requests to the most capable model for a specific task (e.g., a model which is optimized for coding tasks).

Link to Soni’s article.

Blockbuster had late fees. Apparently, Hertz (the car rental company) has damage fees – turbo-charged by AI.

Hertz, the “we try harder” folks, started to deploy an AI-powered scanner for vehicle inspections (the thing you do when you return your rental, and, in the old days, a human walks around the car to see if you have any dings on it). The scanners reportedly identify slight dings (the stuff a human would just ignore) and immediately send you a bill:

“If you followed our last story involving the wheel scuff, you know that UVeye—the firm that produces and operates the scanners—and Hertz like to secure payment of these fees as quickly as possible. They do this by discounting the charge if the customer admits fault and pays within seven days. Foley said that Hertz offered to knock $65 off the bill if he paid immediately. Furthermore, we’ve heard that contacting a human agent at the company to discuss or contest the charges is very difficult, and not possible within the web portal where customers can view and pay for damages. You have to call a separate support line instead, though Hertz doesn’t seem to make that very clear.”

There might be a good reason for Hertz to act like this:

“I suspect the math of investing in such expensive technology indicated they needed to go to extortive levels to get a [return on investment].”

Link to article.

Sign of the times: We are now adding software features not because we need or want them – but because ChatGPT hallucinated them and gets its users to request them.

Turns out ChatGPT is telling people to go to Soundslice, create an account and import ASCII tab in order to hear the audio playback. So that explains it! […] Problem is, we didn't actually have that feature. We've never supported ASCII tab; ChatGPT was outright lying to people. And making us look bad in the process, setting false expectations about our service. […] We ended up deciding: what the heck, we might as well meet the market demand. So we put together a bespoke ASCII tab importer.[…] To my knowledge, this is the first case of a company developing a feature because ChatGPT is incorrectly telling people it exists.

Link to article.

More food for thought:

The rise of Artificial General Intelligence (AGI) marks an existential rupture in economic and political order, dissolving the historic boundaries between labor and capital. Unlike past technological advancements, AGI is both a worker and an owner, producing economic value while concentrating power in those who control its infrastructure. Left unchecked, this shift risks exacerbating inequality, eroding democratic agency, and entrenching techno-feudalism. The classical Social Contract-rooted in human labor as the foundation of economic participation-must be renegotiated to prevent mass disenfranchisement.

Link to paper.

Here is an interesting weekend comment on the current reality of AI being integrated into… everything:

Most people won’t pay for AI voluntarily—just 8% [according to a recent survey](https://www.zdnet.com/article/only-8-of-americans-would-pay-extra-for-ai-according-to-zdnet-aberdeen-research/). So they need to bundle it with some other essential product. 

You never get to decide.

Before proceeding let me ask a simple question: Has there ever been a major innovation that helped society, but only 8% of the public would pay for it?

That’s never happened before in human history. Everybody wanted electricity in their homes. Everybody wanted a radio. Everybody wanted a phone. Everybody wanted a refrigerator. Everybody wanted a TV set. Everybody wanted the Internet. 

They wanted it. They paid for it. They enjoyed it.

Link to article.

Maybe all you need to know about where AI is heading…

Source

Can AI run your business? Anthropic (maker of the Claude AI models) wanted to find out:

"We let Claude manage an automated store in our office as a small business for about a month. We learned a lot from how close it was to success—and the curious ways that it failed—about the plausible, strange, not-too-distant future in which AI models are autonomously running things in the real economy.”

The short answer: No. But there is a whole lot more to look at and learn from the experiment:

"It's worth remembering that the AI won't have to be perfect to be adopted; it will just have to be competitive with human performance at a lower cost in some cases.”

"An AI that can improve itself *and* earn money without human intervention would be a striking new actor in economic and political life.”

And it comes with a bunch of warnings/red flags:

"We do think this illustrates something important about the unpredictability of these models in long-context settings and a call to consider *the externalities of autonomy*.”

"In a world where larger fractions of economic activity are autonomously managed by AI agents, odd scenarios like this could have cascading effects—especially if multiple agents based on similar underlying models tend to go wrong for similar reasons.”

In summary:

"Although this might seem counterintuitive based on the bottom-line results, we think this experiment suggests that AI middle-managers are plausibly on the horizon.”

The whole study is worth perusing.

After the e-commerce juggernaut Shopify, the creator marketplace Fiverr, and the language-learning platform Duolingo’s respective doctrines from their founders enforced the use of AI across their workforce, tech granddaddy Microsoft is joining the fray:

“AI is no longer optional," Developer Division President Julia Liuson told managers. […] "AI is now a fundamental part of how we work," Liuson wrote. "Just like collaboration, data-driven thinking, and effective communication, using AI is no longer optional — it's core to every role and every level.”

Aside from an overall question of how helpful AI truly is for certain tasks (depending on the day, you will see reports claiming massive productivity gains to none at all), it is a good idea to have everyone learn the ropes when it comes to AI. As Wharton professor Ethan Mollick pointed out, AI has a “jagged frontier” – its use is not necessarily intuitive, nor are the results always even; and the best way to learn what this jagged frontier looks like is to experience it.

Link to article.

New study shows that napping measurably leads to deeper insights and more “aha moments.” Take that afternoon nap – it will make you smarter!

Sleep EEG data showed that N2 sleep, but not N1 sleep, increases the likelihood of insight after a nap, suggesting a specific role of deeper sleep.

Link to study.

Menlo Ventures just published their latest AI report – assuming that their data is correct (as with all statistics, one ought to question the data, always!), what is going on in AI-land is truly bonkers…

Widespread Use: 61% of American adults have used AI in the past six months. This translates to an estimated 1.7–1.8 billion global users.

Daily Habit: Nearly one in five U.S. adults (19%) use AI every day, totaling an estimated 500–600 million daily global users.

But also:

Low Conversion: Only about 3% of AI users pay for premium services.


Revenue Concentration: General AI assistants (like ChatGPT and Gemini) capture 81% of the $12 billion in consumer spending.


ChatGPT Dominance: OpenAI’s ChatGPT accounts for approximately 70% of total consumer AI spend.

There is a ton more – here is the link to the report.

It doesn’t bode well when the co-founder of Hugging Face, the preeminent open-source AI platform, starts a conversation with “current AI systems are unlikely to make the scientific discoveries some leading labs are hoping for.”

“In science, asking the question is the hard part, it’s not finding the answer,” Wolf said. “Once the question is asked, often the answer is quite obvious, but the tough part is really asking the question, and models are very bad at asking great questions.” […] “Models are just trying to predict the most likely thing,” Wolf explained. “But in almost all big cases of discovery or art, it’s not really the most likely art piece you want to see, but it’s the most interesting one.”

And even more damning:

He argues that what we have instead are models that behave like “yes-men on servers”—endlessly agreeable, but unlikely to challenge assumptions or rethink foundational ideas.

Link to article.

Here is a rather useful (and basic) overview from Ethan Molick on “which AI to use and how” - useful for your friends who aren’t knee-deep into AI (yet).

Link to Guide.

It took a while, but AI is coming for the enterprise for real now…

"The Census conducts a biweekly survey of 1.2 million firms, and one question is whether a business has used AI tools such as machine learning, natural language processing, virtual agents, or voice recognition to help produce goods or services in the past two weeks, see chart below. Nine percent of firms reported using AI, and the rising trend in AI adoption increases the likelihood of a rise in productivity over the coming quarters.”

Source: Apollo Global Management 

Using LLMs such as ChatGPT not only creates an abundance of em-dashes (“—”) in our writing, but it also starts to affect the way we speak, as a new study shows:

“Words like ‘prowess’ and ‘tapestry,’ which are favored by ChatGPT, are creeping into our vocabulary, while words like ‘bolster,’ ‘unearth,’ and ‘nuance,’ which are less favored by ChatGPT, have declined in use. […] In the 18 months after ChatGPT was released, speakers used words like ‘meticulous,’ ‘delve,’ ‘realm,’ and ‘adept’ up to 51 percent more frequently than in the three years prior.”

“’We internalize this virtual vocabulary into daily communication,’ says Hiromu Yakura, the study’s lead author and a postdoctoral researcher at the Max Planck Institute of Human Development. […] It’s not just that we’re adopting AI language — it’s about how we’re starting to sound... researchers suspect that AI influence is starting to show up in tone, too — in the form of longer, more structured speech and muted emotional expression.”

In essence: We become (even more) homogeneous. <sigh>

Link to article and study.

What could possibly go wrong? Toy maker Mattel (Barbie, anyone?!) teams up with OpenAI to create AI-powered toys…

By tapping into OpenAI's AI capabilities, Mattel aims to reimagine how fans can experience and interact with its cherished brands, with careful consideration to ensure positive, enriching experiences.

Sounds like a potentially nightmarish experience…

Children do not have the cognitive capacity to distinguish fully between reality and play. […] "Endowing toys with human-seeming voices that are able to engage in human-like conversations risks inflicting real damage on children. It may undermine social development, interfere with children's ability to form peer relationships, pull children away from playtime with peers, and possibly inflict long-term harm." - Robert Weissman, Public Citizen co-President

Link to article.

Microsoft’s WorkLab team is out with a new report on what our workplace looks like mid-2025 – and the results are ugly:

Nearly half of employees (48%) – and more than half of leaders (52%) – say their work feels chaotic and fragmented. […]

Half (50%) of all meetings take place between 9–11 am and 1–3 pm—precisely when, as research shows, many people have a natural productivity spike in their day. […]

On average, employees using Microsoft 365 are interrupted every 2 minutes by a meeting, email, or notification.

And is goes on and on… Particularly hilarious (and sad is this):

In the final 10 minutes before a meeting, PowerPoint edits spike 122% – the digital equivalent of cramming before an exam.

The report offers some positive outlook though:

The future of work won't be defined by how much drudgery we automate, but by what we choose to fundamentally reimagine. […] The most effective organizations know this—and act on it. Frontier Firms are putting the Pareto Principle into practice, focusing on the 20% of work that delivers 80% of the outcomes.

Link to report.

Nadella nails it:

"So, the first thing that we all have to do is, when we say this is like the Industrial Revolution, let's have that Industrial Revolution type of growth," he said. "The real benchmark is: the world growing at 10 percent," he added. "Suddenly productivity goes up and the economy is growing at a faster rate. When that happens, we'll be fine as an industry.” Needless to say, we haven't seen anything like that yet.

Good reminder to look beyond the hype train. But also do not forget that AI is already changing many aspects of our daily work life.

Link to article.