Machines that learn. Below, everything that touches this field, including pieces that live mostly in neighboring territory.
On Its Head You Can Only Manage What You Can’t Measure Inside the box thinking Why Optimization Is Often Sub-Optimal Simplicity Is Overrated Right-Brain Analytics The New LBO: Left-Brain Outsourcing The Return…
Understanding Cognitive Biases Through a Sampling Limitations Framework: A Comprehensive Review and Theoretical Integration Abstract This paper presents a comprehensive theoretical framework that…
TL;DR: leaderboard rankings are unreliable This picture is worth a t̶h̶o̶u̶s̶a̶n̶d̶ billion words . Based on an preprint on the ArXiv server (The Leaderboard Illusion) by a team led by Shivalika Singh, Marzieh…
As a 30-year groupie of evolutionary computation, I am particularly sensitive to this trend: evolutionary AI is on the rise. The latest case in point, of course is @google deepmind’s AlphaEvolve, which uses…
AI agents have been around for quite some time ("multi-agent systems" are one classical embodiment), but of course in 2024 LLM Agents are all the rage. With the explosion of agentic applications and agents'…
That's the title of a short IEEE Spectrum piece ( and a case in point as a second follow-up to Hugging Face's Thomas Wolf's post on how mediocre current AI systems, particularly LLMs, are at asking…
When people talk about “Human-in-the-Loop” (HITL), what they are really talking about is “Human-in-the-Current-Loop”. I am the human in the current loop when I use Claude Code or Codex and it is a f***g…
“More people: 100x more people will gain the confidence to express their ideas visually and engage with creative tools again (most of whom abandoned drawing as children and never attempted creativity again).…
A great take on genAI hallucinations by Colin F. "Hallucinations", however we define the concept, are a core property (and perhaps the core value proposition) of LLMs and genAI systems. My view: Trying to root…
The announcement of Cursor's $60B acquisition by SpaceX is timely, for two reasons: 1⃣ It is bolstering the share price right now, as confirmed by the 10% bump today. As my colleague Raphaelle d'Ornano wrote,…
In a stimulating post (link in the comments), @bloomberg beta’s @roy bahat highlighted a fascinating 2017 (yes, 8 years ago!) study from this dense urban center that is @the university of Illinois…
This is an "old" post by John List (4 months in AI time is like 4 years). But it is still worth commenting upon: 1⃣ First, I agree with the fact that AI is making people with knowledge and critical thinking…
If you build a company that depends critically on another company's technology, your business is at the whim of that other company's strategy. It happened before with Google (e.g., changes in its ranking…
A recent article about "AI Deception" ("AI deception: A survey of examples, risks, and potential solutions": reviews some AI systems' "ability to deceive via techniques such as manipulation, sycophancy, and…
I am saying this with great embarrassment, I don't think I had fully internalized until today that big AI compute deals are announced... in MW or GW (GW being the new MW). It has probably been 6 months but it…
In his 60 minutes interview with Anderson Cooper two years ago, legendary music producer Rick Rubin gave viewers, as usual, incredible insights into his success and also about the future (of humans and AI).…
? I had fun reading Tyler Cowen’s “The Marginal Revolution. Rise and Decline and the Pending AI Revolution”, a book about marginalism, “the economic doctrine that we can best understand value by considering…
In this recent Scientific American article, a well known historian of science provides a rather misleading description of Occam's razor or, more precisely, offers "counterexamples" that are not. The author's…
I am a scientist with lots of ideas, all the time. But I make a lot of calculation errors, and I am a slow coder. Most of my ideas go untested. On the flip side, I can easily and quickly verify whether a…
I was hit hard by learning about Danny Kahneman's very deliberate decision to end his life by assisted suicide in Switzerland last year (Danny's former collaborator on the book "Thinking, Fast and Slow" and…
Just for the fun of it, I applied the methodology described in detail in Michael Mauboussin and Dan Callahan, CFA's latest installment in the delightful series of exposés we have come to expect from Mike and…
Before Insilico Medicine, Recursion, Atomwise and many others, CoalesiX, launched in 2006, used another form of AI, Evolutionary Computation (EC), to turbocharge the ability of medicinal chemists to explore a…
With all the healthy chatter about "verifiability" ignited by Andrej Karpathy, I want to point out that even if "verifying" (a program) can be hard, it is still easier than coming up with a good idea for a…
A short article by WIRED's Will Knight may be one of the most important pieces anyone interested in AI needs to read right now (sadly behind a paywall, so I'll try to distill its essence). Although it focuses…
Even if you happen to intuitively agree with the conclusions, this the kind of dangerous and uninformed statement from a bestselling author that spreads like wildfire and becomes impossible to contain. The…
An interesting paper by the research nonprofit METR (Model Evaluation & Threat Research) titled "Measuring AI Ability to Complete Long Tasks"(link in comments) has been generating some deserved buzz. But it…
At a time of tense relationships between the USA and Europe, I think I have found the reason . When you are a French-American citizen, you have to entertain two opposing ideas at all times : a leisurely lunch…
This 8-years old video by Josh Gaines (aka Josh Darnit) has re-emerged recently —for good reason! (Notice the super intentional emdash). The Exact Instructions Challenge is a great reminder that "English is…
In a short Nature commentary "Does AI already have human-level intelligence? The evidence is clear," ( authors Eddy Keming Chen, Mikhail Belkin, Leon Bergen, and David Danks argue that the long-standing goal…
's requests for startups are always a good place to peek into the zeitgeist (at least the Silicon Valley zeitgeist). A few comments (on things I know a little bit about): #3: the n-of-1 is great on paper…
, a preprint by a team from Carnegie Mellon University, University of Oxford, Massachusetts Institute of Technology and UCLA offers a well-designed "randomized controlled experiment (N = 354) on…
After yesterday's post on the narrow search effect (how prior beliefs influence how you search for information, leading [usually] to search results that tend to confirm and/or reinforce said beliefs) I decided…
A study by Eugine Leung and Oleg Urminsky showed up in my feed today through Richard Hahn by way of Stefano Puntoni (thank you all!!). "The narrow search effect and how broadening search promotes belief…
Some 27 years ago, I was so impressed by Pablo Funes' work (with his PhD advisor the always ideating Jordan Pollack) (P. Funes and J.B. Pollack, “Evolutionary Body Building: Adaptive Physical Designs for…
Ok, by evolution I mean evolutionary computing (EC), meaning algorithms inspired by natural evolution such as genetic algorithms, evolution strategies, genetic programming and more. Why? Everyone, their…
You might be missing the iceberg ahead. Follow up from the timely HBR article I commented on recently. A Search Engine Marketing (SEM) agency decided recently to invest heavily in AI, and more specifically…
I have heard or read this so many times it is infuriating: LLMs are not smart because they just do "next-token prediction". Whether or not LLMs are smart or can reason or have feelings (they don't ), they are…
That’s the title of blog post by @jesse johnson, Scaling Biotech. I believe, with characteristic humility, that the question is ill-posed, even though the framework the author proposes can be useful in…
Aparna Chennapragada, Cassie Kozyrkov, Thomas Wolf, Kerem Tomak, Stefan Weitz, Charlene Li, Raphaelle d'Ornano, Peter Lee, Alex Komoroske, Blaise Agüera y Arcas Productivity gains from using AI are captured by…
It sounds obvious. But a few weeks ago, in the heat of the SaaS-pocalypse moment, Jasmine W. and Deirdre Bosa vibe-coded what they described as a clone of project management tools from enterprise SaaS company…
What has been happening after the publication of Google Quantum AI's paper on Shor's algorithm for elliptic curve cryptography is just stunning. The story: the Google team and colleagues reached a 10x…
Michael Kratsios, Director of the Office of Science and Technology Policy in the Executive Office of the President, issued an interesting memo yesterday on the "Adversarial Distillation of American AI Models."…
Thanks to Elissa Bowes-Arbeitman I found a video of my awesome "Draw Together Strangers" experience at The Aspen Institute's health festival in June. DrawTogether Strangers is a clever, fun and incredibly…
Wow, I find myself agreeing with Stephen Wolfram much more than I expected! It is also possible, as I am almost certain he would argue, that I fully misunderstood his points. But I think he nails it on many…
This guy does not look open-minded. In fact, when this picture was taken, I (yes, it’s me, I haven’t changed a bit) had just started my first semester at École Polytechnique Polytechnique, a French science and…
Raffi Krikorian this is great! Gen AI can "generate" (an infinite amount of) novel things riffing on a theme, but I am sure that humans were involved in curating what was carved into stone. That's augmented…
In one version of the report ( the authors discuss the Iceberg Index, a measure of "where AI capabilities overlap with human skills before adoption crystallizes" in a sort of digital twin of the US labor…
In one version of the report ( the authors discuss the Iceberg Index, a measure of "where AI capabilities overlap with human skills before adoption crystallizes" in a sort of digital twin of the US labor…
At the start of 2025, DeepSeek AI took the Western AI ecosystem by storm. The team was able to meet or beat state-of-the-art benchmark targets with a smaller, clever model running on non-state-of-the-art GPUs…
I have always been slightly uncomfortable with the idea of "ambient AI", AI that listens and watches you and makes decisions on your behalf because "it knows you." The problem, for me, is that it can nudge…
I have had it with the pointless AGI talk (will it be 2027, 2035, or is "it" already here?), the slow singularity (can you feel it?) but also the recent flare up about whether LLMs are "thinking". Both the…
Of course this caught my attention: LLMs and Swarm Intelligence! According to this study by a team from Renmin University of China, LLMs differ significantly in decentralized swarm scenarios based on their…
The title of a recent preprint from an @amd team (Chaitanya Manem, Pratik Prabhanjan Brahma, Prakamya Mishra, Zicheng Liu and Emad Barsoum) got me excited for a moment: “SAND-Math: Using LLMs to Generate…
I am utterly disappointed in “nice guy” @brian chesky’s unwavering support of @sam altman. Chesky’s description of his role during the short firing of Altman from OpenAI is nothing short of sycophantic: poor…
...until a large number of vulnerabilities can be addressed. The “vibe workflow” mindset that has dominated the agentic space is creating a ton of opportunities for bad actors to exploit and a lot of failure…
That’s a section of the super clear and insightful article by @anjali Shrivastava, “A broken pricing paradigm”. It was published more than 2 years ago and yet it is more relevant than ever in the “agentic…
Fun preprint I just came across through the ICLR 2025 submissions site ( Even though it is still a preprint, there have been extensive reviews. First, I think the authors from Google, Google DeepMind, Google…
I have been thinking a lot about “surprising behavior,” including unintended consequences, from algorithms that rely one way or another on optimization under constraint. That includes a lot of AI models as…
Not ONE but TWO AI (Co-)Scientist articles popped up in Nature today under Accelerated Review. 1⃣ A multi-agent system for automating scientific discovery by the never-shy folks from FutureHouse/Edison…
About 15 years ago, with the Icosystem team, I wanted to apply AI (such as it was back then, at the beginning of the deep learning revolution, the prehistoric dawn of cyborg life it seems) to... IFTTT. I loved…
Many have seen this chart of the "number of questions asked on stack overflow" over time and concluded, rightly, that ChatGPT precipitated its demise. The story is more complicated, and some it is explained in…
Charts from a16z always attract attention. This one (link in comments) is interesting but comes with caveats. First, what do we have here? 1⃣ Each quarter's value is the percent change versus the same quarter…
A clever computational model leveraging differential exposures to daylight by US location and the CDC PLACES health data by county was used to determine the possible health outcomes of switching from the…
Sometimes a Quanta Magazine article makes me feel like I understand something very complex and it is only when I try to explain it to someone else, or to myself, that I realize I didn't understand much at all.…
This article from @Anil Ananthaswamy for @quanta magazine (my favorite science magazine!) delivers a concise description of my own experience of using AI for scientific discovery: Hallucinations are a feature,…
Thanks to an amazing (please read below) paper in Scientific Reports we now know it is likely that Rambo was having a stroke, which may explain his erratic behavior. In related news, said Scientific Reports…
The year is 1993, he is a 25-yr-old research engineer at what is now Orange Labs in Brittany, in the Neural Networks team headed by Daniel Collobert (Ronan Collobert’s dad), deeply hidden inside the Flat…
This graph is from an article in The Economist on 7/14/2025 (AI is killing the web. Can anything save it?). The first-order interpretation is that traditional search is imploding: people are increasingly…
This is very different from what I usually write about. The podcast Rwanda Rising ( is an ode to humanity and resilience, hope and even joy in the face of unspeakable evil. Agatha (Agathe) is a genocide…
In view of the highly anticipated SpaceX IPO, I wanted to understand how the rest of the US stock market might be impacted, specifically the top stocks comprising the indices (e.g., Nasdaq 100, S&P 500) that…
The title of this nice BGR article did not surprise me (first because the Turing Test, as the article rightfully reminds us, is not very scientific and has highly limiting quirks; second because LLMs have…
A super interesting article in PNAS, "AI–AI bias: Large language models favor communications generated by large language models" by Walter Laurito (FZI Research Center for Information Technology), Benjamin…
“I think there is a world market for maybe five computers.” — Thomas Watson, 1943, showing more restraint than npm package authors In lieu of my favorite cartoons, I will be sharing a sequence of spot-on,…
There is something happening in all kinds of business contexts, but most prevalent in the expensive content business, aka consulting: a lot of mediocre content (AI slop) is being generated by AI tools,…
Anna Marie Wagner and Christina Agapakis thank you for this! When I read Ted Chiang's article last September I thought it was really weak and disappointing as an argument for why AI can't make art (although I…
My mind was a little bit prepared this time. I love how Ethan Mollick shares his thoughts. And I love Bruce Sterling (the Sci Fi author who asked me a question about heuristocrats at etech 2006, a proud…
At a recent "Workshop on AI in the Cloud" at UC Berkeley organized by Industry-Academia Partnership (IAP, see Ion Stoica (UC Berkeley Professor and a co-founder of Databricks, Anyscale, LMArena) gave a…
I found a description of the following experiment at (an intriguing discussion of AlphaGo's famous "random" or "genius" move 37). These two pictures were presented in 1966 by A. Michael Noll, a Bell Labs…
LLM "hallucinations" persist largely because of how we evaluate models, not just how we train them Adam Kalai (OpenAI), Ofir Nachum (OpenAI), Santosh Vempala (Georgia Institute of Technology), Edwin Zhang…
Finally, a sound reminder that AI, in and of itself, is not creating a NEW, SUSTAINABLE advantage for companies that adopt it. The article, by Jay Barney and Martin Reeves in the upcoming issue of Harvard…
There is much I dislike about Ted Chiang's manifesto in The New Yorker two weeks ago ( But his narrow understanding of AI, and specifically gen AI and LLMs, is a topic for another time. A more positive take…
That's an observation from "Artificial intelligence tools expand scientists’ impact but contract science’s focus," an article by 郝千越, Fengli Xu, Yong LI and James Evans that has been circulating as a preprint…
I really like Erik Brynjolfsson's work and thinking. This figure from his 2022 article in the American Academy of Arts & Sciences Daedalus journal ("The Turing Trap: The Promise & Peril of Human-Like…
3 days ago Mike Krieger resigned from Figma's board "the same day that The Information reported that Anthropic’s next model, Opus 4.7, will include design tools that could compete with Figma’s primary…
On the topic of "extraordinary claims require extraordinary evidence", I always thought that the claims made from Stranford's Walter Mischel's famous "Marshmallow Test" required better evidence. For example,…
You will find a number of really interesting graphs in the Stanford Institute for Human-Centered Artificial Intelligence (HAI) 2026 AI Index Report. This one is both revealing and a bit misleading. Revealing…
Two posts on the same preprint (the preprint is behind a paywall as far as I can see, which is a first for me) about a large-scale study of students using AI in China. Long story short: 1⃣ According to Gabriel…
As we're witnessing generative exaggeration in humans on social media discussing LLM social agents (or artificial social crustaceans as David Ha would put it), a timely article by Walter Quattrociocchi and…
I really, really enjoy Jeremy Howard 's thoughts... usually. A few weeks ago, he came up with a thread on X (as older people, formerly known as Twitter ) ( that can be summarized by one the posts: "Absolutely…
? I periodically remind myself to re-read this 2012 by Cormac Herley at Microsoft, or at least the abstract. The answer to the question, in a nutshell, is that if someone responds, they are much more likely to…
In a fun and short speculative PNAS article titled “Could humans and AI become a new evolutionary individual?” (PNAS 122 (2025) No. 37 e2509122122), Paul Rainey (Max Planck Institute for Evolutionary Biology…
A 2001 picture of Salesforce founder Marc Benioff feels very relevant. The rather indiscriminate "saaspocalypse" and its modern prophets of software doom should remember that back in 2001, according to…
12 years ago, as an ignorant and self-confident (the two often come as a bundle) founding Dean of Computational Sciences, I discovered Allen Downey's blog and books and used them as inspiration to create a…
If you have never read, or don't know about, Shel Silverstein's The Homework Machine (from the book of poems A Light in the Attic, first published by Harper & Row Junior Books in 1981), here it is. 44 years…
I made my own leaderboard picture from the Arena agent leaderboard. A few things to note. 1⃣ The first, and to me one reason why Arena (with Ion Stoica and Anastasios Angelopoulos) deserves a lot of attention,…
Thank you for that Charles H. Martin, PhD. I am lucky to have collaborated with Jack Cowan at Bios Group in the 1990s, where he regaled us with how he had actually worked and became friends with McCulloch and…
Lots of insights to unpack in Gigi Levy-Weiss's latest post on the NFX blog. And perhaps a few provocative ideas that don't apply everywhere all the time. But it captures the zeitgeist and zeroes in on…
Lots of insights to unpack in Gigi Levy-Weiss's latest post on the NFX blog. And perhaps a few provocative ideas that don't apply everywhere all the time. But it captures the zeitgeist and zeroes in on…
I love love love Jim Fan's commentary on OpenAI's o3 latest FrontierMath benchmark results, in some ways a super-intelligence tour de force. I like the concept of "single-point [RL] super-intelligence" the…
I don't know Lauren Ducrey but she recently posted what I think is the most concise summary of the uses of genAI: "This is my biggest question when I'm using AI: Am I being lazy or am I being curious?" (Lauren…
I didn't pay much attention to this article ("Bayesian teaching enables probabilistic reasoning in large language models", Linlu Qiu, Fei Sha, Kelsey Allen, Yoon Kim, Tal Linzen & Sjoerd van Steenkiste, Nat…
My apologies for being all over the place today, that's the problem when living at the intersection of AI, bio and aging. But one piece of news yesterday caught my attention: GitHub announced the launch of the…
On 12/2/1998, I saw my name for the first time in print in the US [the very first press mention came in Le Monde on 12/29/1995 for a Science Magazine article] based on an interview with Sue Goetinck Ambrose of…
That's the title of a fantastic blog post by Anu A. (link in the comments). The snapshot from jack friks is from the post and I included it because I can relate soooooo much. Another nugget comes from a link…
One consequence of the current genAI hype is companies, large and small, rushing to apply it to use cases that are either inappropriate or for which the technology is not yet ready. Of course, the more…
We deserve better than a vague, self-serving, oversimplifying, unserious, overhyped “report” about the 95% failure rate of AI pilots in organizations (@kervin Werbach has a great post about this marketing…
I love to read about Terence Tao's incredibly insightful experience reports with gen AI, in particular with chatGPT and now the new o1-preview model ( His goal when using an AI is to support him in solving…
A new wave of AI generation tools for images and videos has triggered a lot of excitement and fear. The reaction goes something like this: "Wow! Look at how realistic this image is. Can you tell this image is…
It's a fascinating exploration of Anthropic's Claude 3.5 Haiku model. Lots of interesting findings based on the concept of "attribution graphs", which reveal computational graphs in LLMs, showing how different…
Can't argue with the title of this SEMAFOR article by Reed Albergotti. Yes, this week (or the past month) was Anthropic's DeepSeek moment on Wall Street. But this time it was felt broadly by the "software…
A Communications of the ACM article ( with that title by Victor Dibia, PhD of Microsoft Research summarizes how many people in the field (people whose job it is to keep up with AI research) feel, especially if…
I am late to the armchair quarterbacking on DeepSeek, as I wanted to let the dust settle a little before forming an opinion. I think this work by a team from HKU, University of California, Berkeley, Google…
" My favorite line from Arcadia, a 1993 play by Tom Stoppard, the incredibly smart and witty British playwright and writer who passed away this week at 88. Funnily enough, Arcadia is not mentioned in his BBC…
An intriguing article, "How Might Artificial Intelligence Influence Human Evolution?" by Robert C Brooks (UNSW in Sydney) is exploring the possible impact of AI on the evolution of our species (it is funny how…
The concept of "jagged frontier", introduced about two years ago (Fabrizio Dell'Acqua, Edward McFowland III, Ethan Mollick, Dr. H /Hila Lifshitz (Hán, X也), Katherine Kellogg, Saran Rajendran, Lisa Krayer, PhD,…
The concept of "jagged frontier", introduced about two years ago (Fabrizio Dell'Acqua, Edward McFowland III, Ethan Mollick, Dr. H /Hila Lifshitz (Hán, X也), Katherine Kellogg, Saran Rajendran, Lisa Krayer, PhD,…
Tim O'Reilly wrote a concept-rich piece that I found incredibly novel and refreshing. Lots of ideas in Tim's article, and the one that stuck is best captured here: "... in the long term, if people stop…
Bruce Schneier and Barath Raghavan nail it in this short and delightful article in IEEE Spectrum. Imagine you work at a drive-through restaurant. Someone drives up and says: “I’ll have a double cheeseburger,…
A beloved French TV personality, Bernard Pivot passed away two weeks ago at 89. His literary shows, including his flagships "Apostrophes" and "Bouillon de Culture", were cultural gems, which inspired James…
I find this article by Rachel Draelos, MD, PhD super insightful. Beyond the obvious implications for medicine, it also raises questions/issues that apply to a lot of different contexts. Of the four problems…
Bruce Schneier and Nathan Sanders penned an insightful opinion piece in IEEE Spectrum (link in comments), specifically in the context of security, but with broader applicability. Millions of users of genAI…
The first evening at DOC, we got Reid Hoffman, Mike Krieger, Anitha Kannan, Neil Parikh debating whether AI will replace doctors, with Jordan Shlain, MD as moderator. Two versions of Anthropic's Claude…
In my daily newsletter from MIT Technology Review today, there was a (paywalled) story about paleo robots as models of prehistoric animals: "In the last few years, paleontologists have developed a new trick…
Here is a clever experiment, and one with interesting implications for AI-human interactions. But should we be really surprised that what is essentially a sampler of (a large percentage of all English) written…
The ability to create agentic workflows that can call tools and LLMs (Large Language Models) is opening up a universe of possibilities (and dangers too, but that’s for another post). Among them is verification…
I am on a roll. Two French intellectuals/philosophers analyzed the world around them almost 200 years ago (Tocqueville: Democracy in America, 1835 French and 1838 English) and 75 years ago (Weil: On the…
Delusional. If you need "founder mode" at a $82bn (market cap) company to make its customer service work, something is wrong. I am obviously not talking about Twitter, sorry, X since it is worth a lot less…
An article in Nature today (behind paywall, Artificial intelligence and illusions of understanding in scientific research) raises some interesting issues and questions about the use of ai and genai in…
In a conversation with economist Tyler Cowen last month, Peter Thiel expresses his view that LLMs will make things "worse for the math people than the word people." He continues: "What people have told me is…
I recently had the privilege of hearing about an unpublished study that would save (academic, corporate or nonprofit) scientists in the field a lot of misery if they knew about it. But the study “failed” in…
In the first week of January 2026, I have finally caught up with Andrej Karpathy: I am overwhelmed. I have crossed a threshold that I don't know how to uncross: I am spending more than 5 hours daily trying to…
Concise depiction of the strategic differences between OpenAI and Anthropic by Robinhood's Sherwood News. The interactive timeline is also a reminder (should one need one) of the insane shipping pace of both…
" This poetic failure mode ( which should at least ensure that we all go out in style, is apparently the most likely with complex AI models. As models become more intelligent and tackle harder tasks, their…
There is the occasional interdisciplinary physics paper that I find insightful and this is one of them: Nonuniversality for crossword puzzle percolation, by Alexander Hartmann of the Carl von Ossietzky…
"But, in truth, the David who solved the K(π,1) conjecture is a social parasite of the much better mathematician, the David who crafted Definitions 2.4 and 9.3." A post by David Bessis (by way of Alexander…
Right off the heels of Anthropic’s announcement that it is “donating the Model Context Protocol (MCP) to the Agentic AI Foundation (AAIF), a directed fund under the Linux Foundation, co-founded by Anthropic,…
What an incredibly naive statement Andrew Ng , that we will see it coming when AI wants to deceive us. AI is already deceiving us in so many ways, from social algorithms (think TikTok) that lead to addiction…
Here is a nice opinion piece in PNAS by Ross Hammond and Shari Barkin that makes a strong case for combining ABM and RCTs to alleviate the intrinsic limitations of RCTs, particularly when interactions between…
That is an interesting tidbit about Codex from OpenAI's Codex engineering lead Thibault Sottiaux in Alistair Barr's Business Insider article "Silicon Valley is buzzing about this new idea: AI compute as…
Last week I wrote about buying raicursive.com (I did acquire the URL) in the wake of recursive's and ricursive's big raises at stratospheric valuations. 1,800 impressions, one reaction. No one laughed. What is…
An important Nature paper by a Google DeepMind team (first author Junhyuk Oh) became available online unedited to accelerate the dissemination of ideas: this paper shows that Reinforcement Learning (RL)…
An interesting strategy+business article ( has an intriguing set of estimates of genAI impact on operating profit margins in a variety of industries. There is so much to criticize. For example, this is really…
Cool Springer Nature article on "emergent misalignment" (Training large language models on narrow tasks can lead to broad misalignment, Jan Betley, Niels Warncke, Anna Sztyber, Daniel Tan, Xuchan Bao,Martín…
I have recently seen an upsurge of interest in... stigmergy! I was wondering why for a minute and then realized it is central to the "new" paradigm of agentic orchestration. If you don't know stigmergy, the…
I hadn't thought of it but AI compute as part of your compensation package kind of makes sense if you are working in AI (and let's be honest, who is not working in AI? Just kidding but I do live in the Bay…
I hadn't thought of it but AI compute as part of your compensation package kind of makes sense if you are working in AI (and let's be honest, who is not working in AI? Just kidding but I do live in the Bay…
Ethan Mollick doesn't need more publicity... but I think this is important: #genAI is empowering "the masses" to do things that were out of reach before, not just experts and nerds . What they produce may not…
? That's the provocative title of a fun BMJ Digital Health article by Jonathan H. Chen, Michael Pfeffer and Christopher Longhurst, MD, MS. Before pushing back in gasping horror, it is more an exercise in…
In last week's issue of Science Magazine, Melanie Mitchell reminded us that the narratives we weave depend crucially on the metaphors we use to describe a situation, an event, a concept, a tool, or AI. We are…
I find that "going to the gym with an exoskeleton" is of the best metaphors for (lack of) skill acquisition with AI: it might be good to have an exoskeleton at some point in the real world, but you need to…
In continuing with the idea (or my theory) that the ability to evaluate/critique/curate the output is the critical skill required to extract value out of genAI (more so than prompting), once again Ethan…
wrote a great post about the curse of knowledge and how not only can't you unsee/unhear/unlearn what you have seen/heard/learnt, but you also instantly assume that everyone now has the same knowledge and…
A set of recent experiments with large language models (LLM) such as chatGPT suggest that using super-long prompts that contain up to hundreds or even thousands of examples outperforms fine-tuning. Given that…
Perhaps my biggest source of quotables came (unexpectedly?) from You.com's brilliant founder and CEO Richard Socher during his humanx Q&A session. In order of appearance during the session: 1⃣ He said…
After seeing Hector Zenil's most recent preprint on arXiv ( not being aware of the history behind the animosity (see blog link below), I thought it was a hit piece on the team behind Assembly Theory. Even…
Sorry to be the Grinch here, but while this technology looks impressive, the example you chose for your demo video, Charles A. Kantor, is highly problematic. Ask any recruiter or anyone who has ever used…
Daniel Kahneman passed away at 90. He was not just a 2002 Nobel laureate in economics and the co-creator with the late Amos Tversky of some of the most insightful (and simple) psychology experiments that paved…
Without any particular target in mind (or too many to list ), I like to remind myself on occasion that outsized success is not usually a consequence of outsized skill and talent but of outsized risk taking and…
Fascinating preprint by some reasonably credible mathematicians (Terence Tao, Javier Gómez-Serrano) working with Google DeepMind's Bogdan Georgiev and Adam Zsolt Wagner: MATHEMATICAL EXPLORATION AND DISCOVERY…
I'll end this series of yummy quotables with Mistral AI's founder and CEO Arthur Mensch. Arthur is very serious (in public), so here are two serious takes from his fireside chat: 1⃣ "LLMs are not intrinsically…
A nuanced post well worth reading about chatbots and mental health from someone who knows a little bit about the space, Thomas Insel MD. 1⃣ The idea of mental health chatbots as first line/triage/gateway: "It…
Every time I think of ways to make decisions fairer, I realize very quickly that fairness is not a one-dimensional concept, in fact it is a high-dimensional, multi-faceted ball of complexity. Not only may…
Thank you for that post Conor Grennan. In a comment, Eric Fraser makes a great point -as someone who TRIED to use Coq, I can say that the learning curve is steep. On the other hand, the typical genAI chatbot…
A lot of reactions yesterday about the acqui-partnership between OpenAI and Jony Ive. But after I came across (thank you Carmen!) this short comment in Nature Communications ( by Andy Clark, one of the best…
Back to the 2026 Stanford Institute for Human-Centered Artificial Intelligence (HAI) AI Index Report. There is a plot of public sentiment toward AI: are you expecting AI to eliminate jobs or to create new…
Not only did a certain 2002 paper ( make the list, it turns out to be the most cited article, with more than 6,000 citations as of May 11th, 2024. There is no exciting breakthrough in that paper, no…
Not disagreeing with the conclusion ("stay invested") but the "missing the 10 best days" argument always makes me laugh. So I decided to calculate the annualized returns when missing the worst trading days. It…
Very cool article in Science Magazine (Luke D. Fannin (Dartmouth College), Chalachew Seyoum (University of Arizona), Vivek Venkataraman (University of Calgary), Justin Yeakel (yay Santa Fe Institute!!),…
Humans may evolve, in a biological sense, circuits and chemical networks to form a complete dependence on AI because it will be required for survival. For example, two individuals may not have the same ability…
As we say on LinkedIn, I am humbled and honored to have been voted the most Eric Bonabeau lookalike by hysterical hordes of fans in one of the most landslide way possible. The pluralistic quasi-majority of the…
Fascinating article in Science on "The levers of political persuasion with conversational artificial intelligence" by a multidisciplinary team (Kobi Hackenburg, Ben Tappin, Luke Hewitt, Ed Saunders, Sid Black,…
Fascinating article in Science on "The levers of political persuasion with conversational artificial intelligence" by a multidisciplinary team (Kobi Hackenburg, Ben Tappin, Luke Hewitt, Ed Saunders, Sid Black,…
I finally got to meet Hugging Face/Pollen Robotics' Reachy Mini "in person." We are all striving to be as charming. Stefan Weitz is trying hard as uncle Thomas is watching. Reachy is not just cute, it is a…
Dario Amodei's article "Machines of Loving Grace", with a title inspired by Richard Brautigan's 1967 poem, is a great read. Introducing himself as a "perceived pessimist" is a little bit of a strawman... but I…
The delightfully named Bullshit Benchmark (really the BS Detection Benchmark) by Peter Gostev is really worth looking at. Yet another benchmark you say? Another benchmark, yes, but one that addresses how LLM…
An interesting statement made by Erik Brynjolfsson during his AI, Science and Society Conference AIActionSummit at École Polytechnique/Institut Polytechnique de Paris. The "at least for now" part. Erik made…
What does it say about me that Anthropic's Claude uses the words "honest" and "honestly" in pretty much every sentence it utters? Like, "the honest take is that your method doesn't work", or "Let's report…
My family's terrible sense of (market) timing must be genetic... ... I think. I found this surprising stock certificate for the Bonabeau Enterprise (I had no idea there was such as thing) dated 15 January…
I will come back to the topic of food printing in the context of generative thinking -and, of course, generative AI. That was the reason behind our first (2005) foray into food printing and the premise of…
Thank you Kes Sampanthar for your read on this paper. All the assumptions made by Daron, all the things he neglects, suggest a linear approximation whereby he applies a very short term gradient (most gains…
A fascinating (and a little scary) study on variance in values across 76 countries: Not surprisingly, homosexuality, euthanasia, divorce top the diverged leaderboard. The leading converged values: more…
A Perspectives article from June 2024 is making the case, as its title indicates, for language as a tool for communication rather than thought. Regardless of where you fall on this longstanding philosophical…
Fine, I never was, but I found this intro from James Fallows in March 2011 that comes as close as it gets to the proof of the possibility that I could have been in another life. Jim had introduced me to the…
This is a very interesting development coming out of Thinking Machines Lab, Mira Murati's company. But what caught my attention when following the link is that Danny Hillis is cited right at the top of the…
A short Science Magazine review worth reading (sadly not Open Access) by Nils Opel and Michael Breakspear "Transforming mental health research and care through artificial intelligence." There have been…
There is this graph from a Reddit post making the rounds. Replace "laptop" with (affluent) consumer hardware; to be fair the post's title is "If trends hold, Mythos-class capability may be running on high-end…
So it looks like the way I appropriated this quote is: "Managers would rather live with a problem they can't solve than with a solution they don't understand", according to GovExec: But, unsurprisingly, I…
Ethan Mollick 's post on capability thresholds (distributed unevenly) greatly clarifies the concept of jagged frontier: thousands or millions of human "tasks" that seem equally difficult to a human can have…
An exciting article from (the always insightful) Mikhail Belkin and colleagues got its first release in Science Magazine a few days ago (paywall). A preprint version from May 2023 is available ( The main…
Ok, a little different from food printing. A Nature article today adds to the inflection point that AI is creating in science and engineering advances. One of the big (multiple) obstacles to achieve fusion…
! It reminds me of a time, long, long ago, when having a child named Alexa (and calling her name loudly) was a sure way to receive unwanted packages from Amazon: the granddaddy of prompt injections. Not to…
The CEO of Klarna, the famous F̶i̶r̶e̶ ̶N̶o̶w̶,̶ ̶R̶e̶h̶i̶r̶e̶ ̶L̶a̶t̶e̶r̶ Buy Now, Pay Later fintech company, said in an on-stage interview at SXSW London that "offering human customer service is always going…
Matt Watson I am reaching the same conclusion but with a twist. The ability to evaluate the output is what matters most: if I can look at the output code and determine whether or not it does what I want…
A recent (humorous) post by Eduardo Ordax brought back memories of my 2004 Harvard Business Review article "The Perils of the Imitation Age", where I quoted (then fashionable) former Fed Chair Alan Greenspan,…
Not that I am that excited (or end-of-the-world worried) about AGI, but I agree wholeheartedly with François Chollet that "LLMs have sucked the oxygen out of the room." OpenAI's chatgpt has precipitated the…
Just trying to understand here. If, as Dario Amodei alluded to yesterday, Anthropic had an 80-fold growth in revenue for the whole of 2026 (the company did see 80x growth in the first quarter on an annualized…
Reid Hoffman I particularly like your "blind spot" comment. In venture capital there is a (very human) tendency to look for things that have "worked in the past" while at the same time pretending to support…
Brian Christian's 2012 videotaped talk at the Santa Fe Institute contains many little treasures. One of then begins at 15:40 in the video: Claude Shannon's wife was a computer... Until the middle of the 20th…
With the upcoming megacap/thin-float IPOs of SpaceX, OpenAI and Anthropic, the impact of S&P Global's (rational) decision to not change the requirements for entry into its major indices will have a huge impact…
is a unique human being in the world of AI. A "recovering statistician", she wonders who decides that something should be measured, worth applying statistics to, that something matters. That applies to the…
Some 18 years ago, with Hod Lipson, we were dreaming up a 3d printer/food replicator that would one day be on everyone's kitchen countertop, just like a food processor. You would buy the ingredient cartridges,…
Bill Viola, the "Rembrandt of the video age", passed away on July 12th after a long struggle with Alzheimer's Disease. I encountered his work for the first time at the Whitney Museum of American Art in 1997.…
As I am (obviously) going through a late-mid-life crisis and also through my archives in a hopeless attempt to de-clutter, I find little nuggets from the past. Here a pic with the amazing NYT journalist Frank…
Thank you Compound and Dr. Shelby for maintaining this repository of hashtag#techbio foundation models. I don't know whether to be excited by a Cambrian explosion of LLMs for bio or deflated by the emerging…
Prompt injections in the physical world are not as far-fetched as you might think. When input data is code, sh*t will hit the fan. Worth reading the article by Luis Burbano, Diego Ortiz Barbosa, Qi (Laura)…
Ok, to me this is more surprising than AlphaGo's Move 37 against Lee Sedol. Impressive feat of AI and robotics by a Sony/SonyAI team. The setup is cumbersome for your playroom but that's a milestone.
Bonabeau circa 2014, attempting to explain Minerva University's Formal Analyses curriculum to our first prospective cohort. And with that, the famous "What IS Formal Analyses?" What a blast, 2013-2016, for…
Now you know why you h̶a̶t̶e̶ are frustrated with Opus 4.7. Anastasios Angelopoulos and team at Arena are nailing it with their approach. I will write a longer post on Arena, which I think is a company well…
You want to simulate the impact of the thin-float, fast-entry, upcoming mega IPOs? You want to know what might happen to SpaceX insiders? OpenAI insiders? Head to the webpage and read the full description of…
Don't forget to state the constraints or your chatbot (or Claude Code) will find shady shortcuts and loopholes. When you live in flatland, shortcuts abound in the third dimension. Sadly, the infinite richness…
Science paper ( with unsurprising but specific results. The authors measured the fraction of Python functions in more than 30 million GitHub commits from 160k software developers they could detect were…
? Excellent piece by Toby Nangle on Alphaville about the semi-forced buying of SpaceX stock by benchmark-constrained fund managers who NEED to buy "in the stock memes"... Thank goodness S&P Global made the…
Vemödalen The Fear That Originality Is No Longer Possible Day 2 in John Koenig's rabbit hole, The Dictionary of Obscure Sorrows, another concept that resonates in the age of AI everywhere. Now the PhD…
Me showing up at the app gym with my Claude Code Fable exoskeleton. I am not exercising any muscle but I do look (and feel) powerful. In fact I am powerful, just don't take my exoskeleton away. Please.
I saw a typo today about Pasteur's famous quote. It turns out the typo was in my head. Still, it is too delicious a nonexistent typo not to share it. It even makes sense.
Nice explainer: Beyond performance: how design choices shape chemical language models
I am not sure many people realize how Hugging Face is quietly becoming one of the most important companies in AI. Thomas Wolf, Clem Delangue , Julien Chaumond
I recently discovered John Koenig's The Dictionary of Obscure Sorrows. And I found this concept that resonates way too strongly with me.
Anyone shocked by the terrible example used on Clawdbot's GitHub page? Scammers on the streets of Marrakech? Spreading racist stereotypes about Morrocans?
The most concise description by Victor Coimbra of a Jevons-like inflation in AI costs while token prices are free falling.
First installment: you think you are in control but your [social media, AII chatbot, ...] is controlling you.
Not sure what to add.
gives the clearest definition of the mysterious and often elusive "context layer".
Precision matters. Or why English is NOT the new programming language.
Open-endedness represents perhaps the most profound challenge in computational science: how do systems transcend their initial constraints to generate genuinely novel, complex, and meaningful structures? This…
Eli Lilly's Chorus unit, established in 2002, emerged as one of the most successful innovations in pharmaceutical R&D productivity, achieving 3-10x productivity improvements over traditional development models…
Open-endedness is the capacity of a system to keep generating novelty, complexity, and diversity without ever settling into a fixed endpoint. Biological evolution is the reference case: four billion years of…
Open-endedness represents one of the most profound and elusive concepts in computational science, spanning domains from artificial life to creative systems. Rather than focusing on specific algorithms, this…
Gordon Pask's electrochemical devices represent one of the most remarkable achievements in cybernetics and early artificial life research. Between the 150s and 160s, Pask constructed genuinely self-organizing…
The core idea Some processes are easy to run in one direction and hard to run in the other. Mix yellow and blue paint and you get green in a second. Hand someone a can of green paint and ask which yellow and…
Recall and recognition are two ways of retrieving the same stored information, and they differ sharply in how much work they demand. Recall means producing a target from memory with little or no external cue.…
Gordon Pask's electrochemical devices from the 150s and 160s are entirely unique in AI history because they challenged conventional computational orthodoxy by demonstrating that complex intelligence,…
That's my favorite quote from Derek Lowe's Science opinion, "The End of Disease". And Derek was kind enough to collect links to all his AI-based drug discovery opinions from the past few years, a treasure…
I saw a post by @juergen schmidhuber recently ( that gave me pause. About convolutional neural networks and the neocognitron, a 1979-80 neural network architecture by the Japanese scientist Kunihiko Fukushima…
I was heartbroken when Harvard Business Review editors decided to title my 2003 article “Don’t Trust Your Gut” ( certainly a buzzier catchphrase than “How To Leverage Your Intuition With Analytics” or other…
I am sure many have seen this Y Combinator's Summer 2025 Request for Startups by Jared Friedman. When I founded Icosystem in 2000, I had this exact concept in mind: if we can do analytics and machine learning…
As Insilico Medicine has released fascinating R&D benchmarks ( I find it useful to get back to basics to understand the fundamentals. Borrowing from what I consider the gold standard in defining it (2010…
I was heartbroken when Harvard Business Review editors decided to title my 2003 article “Don’t Trust Your Gut” ( certainly a buzzier catchphrase than “How To Leverage Your Intuition With Analytics” or other…
I saw a post by Jürgen Schmidhuber recently (all links in comments) that gave me pause . About convolutional neural networks and the neocognitron, a 1979-80 neural network architecture by the Japanese…
In a recent substack post ( the always insightful Rishad Tobaccowala presents "10 thoughts about AI, Humans and Work in 10 minutes". They are all worth pondering (although I would not invoke Suleyman in # 3)…
Timely article by Xuekun WU, Matthew Wu, James Zou, Nicole Kleinstreuer, Joseph C. Wu: Over 90% of drug candidates that succeed in preclinical animal studies fail in human trials. One (but not the only) big…
Or, evolving on chemical space(s)! The appeal of the concept for me comes from the fact that the topology of a chemical space (the "construction operators") and its boundary conditions are so natural to map to…
Always stimulated by Derek Lowe's writings (in spite of an average 50% agreement rate), I tried (a condensed version of) his OpenAI Deep Research prompt on the toxic effects of thalidomide with Perplexity Deep…
What sounded like a (half-)joke by Joanna Maciejewska ( a little less than two years ago feels a bit more urgent today. Obviously, it resonated then with a lot of people, generating 3.3M views on X and…
An intriguing article ("Imbalance in gut microbial interactions as a marker of health and disease", Roberto Corral López, Juan Bonachela, Maria Gloria Dominguez Bello, Michael Manhart, Simon Levin, Martin…
Great post by Ash Jogalekar. Drug discovery is still a prime example of "competence without comprehension" ( we build things that work without always understanding how or why. When a mechanism of action (MOA)…
Now for something unexpected: Midjourney Medical. Yes, the company that generates images from natural language prompts and is the subject of multiple lawsuits for copyright infringement. But it was never your…
An interesting take from one of the most brilliant minds of microbiology, Itzhak Mizrahi. I love the paper he is highlighting, the results are impressive and exciting, it took a lot of "fundamental…
So there is a lot to unpack in Elliot's latest post. To be clear, I am not trying to unpack Elliot but his post . TL;DR: disruption is here, commoditization is coming and the nexus (or nexi, nexuses?) of value…
A useful preprint (Future of Work with AI Agents: Auditing Automation and Augmentation Potential across the U.S. Workforce; by Stanford University's Yijia Shao, Humishka Zope , Yucheng Jiang, Jiaxin Pei, David…
I love this introduction to scent design by Taylor Rayne at Asimov Press (Scent, in silico, and the role that machine learning and genAI is playing in the field's recent advances: Google DeepMind's Principal…
There is this concept of AI-driven cognitive atrophy going around. It may happen, there is ample evidence that our ability to perform some tasks degrades when we stop performing them. But the key here is to…
Kurt Vonnegut’s 1952 novel, Player Piano, feels terrifyingly prescient: a dystopian society where machines and computers have replaced most human workers, leaving only engineers and managers with meaningful…
That’s the title of a great Trends in Biotechnology article led by @the align foundation’s @erika debenedictis. I loved the preprint, I am thrilled to see it published. The Align Foundation is a @schmidt…
Super-interesting Nature Medicine article (A multimodal sleep foundation model for disease prediction, link in comments) from a team supervised by Emmanuel JM Mignot and James Zou at Stanford University…
Some folks are wondering why all of a sudden their LinkedIn feed is inundated with my posts and in particular a picture of a much younger-looking Bonabeau eating a 3D-printed appetizer in a weird T-shirt. Both…
Some folks are wondering why all of a sudden their LinkedIn feed is inundated with my posts and in particular a picture of a much younger-looking Bonabeau eating a 3D-printed appetizer in a weird T-shirt. Both…
The estate of Ansel Adams complained to Adobe on Threads about images generated by invoking an "Ansel Adams-style" landscape. Using genai to produce content that is "in the style" of an artist goes against…
I just received Blaise Aguera y Arcas's wonderful little treasure, 'What is Life?", a play on Erwin Schrödinger's opus of the same title, published as the first tome of the Antikythera series as part of…
I was struck by a statement I read today about a Techbio company that "uses large language models to create billions of druglike molecules". I have nothing against this company and I think they do very…
A recent fantastic Science Magazine article by a group at the Massachusetts Institute of Technology led by Richard Young and Henry Kilgore of the Whitehead Institute and Regina Barzilay of MIT Computer Science…
Intriguing article by a team led by Moritz Gerstung (DKFZ German Cancer Research Center), Tom Fitzgerald and Ewan Birney (European Bioinformatics Institute | EMBL-EBI Hinxton, UK) reporting on the use of…
Anastasia Borovykh, you are right, humans are good at "taste", knowing what is interesting or not, distinguishing what's meaningful from what is just truth without meaning. Human taste is found at the two end…
The eminently quotable Alex Zhavoronkov, the fearless founder and CEO of Insilico Medicine, presented at hashtag#humanx on the same day Insilico announced $110m in funding ( to propel it to a Hong Kong IPO and…
Ok, that is my short summary interpretation (for AI nerds) of an intriguing article by a Cedars-Sinai-led team. From the abstract: "Here we characterized the representational geometry of populations of neurons…
Spurred by a conversation with Reid Hoffman at DOC, I dug out Phase II clinical trial success rates (Pts) since the 1990s (I used public data for this as a good first approximation). Here is the trend line…
Now that the article is out and Open Access on Cell, I am reposting a comment I made a few months ago about a super interesting preprint, "How to Build the Virtual Cell with Artificial Intelligence: Priorities…
This is a post by Nathan Benaich worth reading for anyone interested in hashtag#AI and hashtag#drugdiscovery. The important TL;DR is that there is a lot to be excited about in spite of the recent failures of…
You can always count on Derek Lowe for an incisive (meaning, with tooth) analysis of a half-baked paper with a promising title ( While Derek focuses on the paper's poorly defined categories (can a target…
Fantastic JPM26 Wing Venture Capital event with the inimitable Rajeev Chand touch! No attribution but a few tidbits: - Immigrants get the job done (tongue in cheek, wonder who might have said that ) - Patent…
In 2011, I had the privilege of guest writing James Fallows' blog on The Atlantic for one week. One of the posts was about Darwin's "Strange Inversion of Reasoning" (the title of a 2009 PNAS paper or…
This is a cool graph from a timely Reuters article drawing the wrong conclusion: "AI-led selloff in contract research firms may be misjudging disruption risk." [h/t to Andrii Buvailo, Ph.D. for surfacing it!]…
This is a cool graph from a timely Reuters article drawing the wrong conclusion: "AI-led selloff in contract research firms may be misjudging disruption risk." [h/t to Andrii Buvailo, Ph.D. for surfacing it!]…
Yesterday, I wrote about reverse centaurs, a Cory Doctorow-meets-Vonnegut Player Piano concept ( Today, there is a site for it. At least the "meatspace layer for ai" is not concealing its devious business…
Great article with a misleading title: "A synthetic protein-level neural network in mammalian cells" (Science Magazine: free bioRxiv & medRxiv preprint: The premise that cells are (fast) information-processing…
I missed this paper by Arash Grakoui and colleagues at Emory University but I think it is worth bringing it up: they show that bacteria in dysbiotic mice (fed a high-fat diet) can migrate straight to the brain…
Allie K. Miller is pointing to the trigger point of a Cambrian explosion in innovation. People (myself included) have been criticizing the "chatbot interface" but it turns out that it is a super effective way…
It is important to recognize the crucial role played by large, well-organized data in the emergence of powerful AI methods. ImageNet is widely considered a catalyst for the sudden success of deep learning and…
One irregular dispatch when a new piece lands: notes on AI, evolution, complexity, and the biology of discovery. No noise.
Delivered with Buttondown · unsubscribe anytime