L'IA générative face au concours d'entrée à l'École normale supérieure blogs.mediapart.fr/david-monnia...

Because it's expanding, most of the observable universe is unreachable even at the speed of light. This is the only diagram I've ever seen that makes the cosmic endowment look small. arxiv.org/pdf/2104.01191

Excellent explainer video by @fryrsquared.bsky.social on the risks of AI agents. We shouldn’t make the mistake of thinking current limitations will necessarily persist. As we’ve seen for years now, the capabilities of frontier AI models are continuously increasing. www.youtube.com/watch?v=WnzR...

So Claude Mythos was, indeed, not marketing hype. Remember this is a general purpose model that just happens to be good at finding exploits because good models are good at lots of things. Expect similar from OpenAI & Google. And from open models in 8 months. hacks.mozilla.org/2026/05/behi...

AI sceptics have long dismissed short or medium timelines as speculation with no basis in data. But recent revenue growth undercuts this argument. It's now the AI sceptics who have to show that current trends will break. www.update.news/p/the-tables...

"More importantly, though, those responses also revealed that many on the left use political slogans to avoid honestly confronting their individual complicity in supporting one of the most violent and environmentally destructive industries on the planet."

Nouvelle vidéo ! Le procès de Giordano Bruno (ça tourne mal) youtu.be/7-Fo1OpOljQ

Voici un exemple de reconstitution d’un monument Grec à partir des traces de couleur laissées dans la pierre, le temple d’Athena sur l’acropole d’Athènes.

My contribution to the solar/farm discourse is that solar panels capture about 100x as much usable energy from the sun as corn grown for ethanol, if you include the energy cost of growing corn. Ethanol corn is effectively a very old-fashioned inefficient solar panel.

L’intelligence : qu’est-ce que c’est ? Comment ça se mesure ? D’où ça sort ? Enregistrement de ma conférence au Forum des sciences cognitives 2026: youtu.be/syVoUcqQeM0?...

Demain : fin de carrière compliquée pour Giordano Bruno.

This is from the co-founder of Anthropic, interesting that he refers to public sources when he is also obviously privy to lots of internal sources that he cannot discuss. I assume he sees the same thing at Anthropic. importai.substack.com/p/import-ai-...

the thing abt the "stochastic parrots" term is if a parrot could genuinely help me debug my code i think id be pretty impressed with that parrot

Academics Need to Wake Up on AI open.substack.com/pub/alexande...

Another great AI water article from an actual scientist. Great to see californiawaterblog.com/2026/04/26/a...

I was quoted a couple times in this Atlantic article, but that isn’t (the only) reason I think it is good. It lays out the reasons why we whipsawed from “AI is a bubble” to “there are not enough data centers” in less than six months. Spoiler: its agents. www.theatlantic.com/economy/2026...

Histoire totalement dingue (et flippante) : des IA ont révélé et minutieusement détaillé à des scientifiques comment fabriquer des armes biologiques et les disséminer dans l'espace public, "avec une telle perfidie et une telle ruse que j'en étais glacé". www.nytimes.com/2026...

New paper (on an old AI model) tests o1 against doctors on medical benchmarks & real ER cases: “across a variety of scenarios and applications, the large language model outperformed both human physicians and older models” The high potential of AI suggests an “urgent need for prospective trials.”

Hot take into the void: "ultra-processed food" is basically the antivaxx conspiracy theory for people who think they're far too well informed to get hooked into anything like antivaxx.

I don't expect it to convince anyone who isn't already convinced, but vintage language models solving coding problems is pretty good evidence of general reasoning ability

Les articles de l'Express sont en libre accès aujourd'hui. Profitez-en par exemple pour lire cet article passionnant sur les recherches actuelles concernant les bases cérébrales du COVID long: www.lexpress.fr/sciences-san...

Announcing Talkie: a new, open-weight historical LLM! We trained and finetuned a 13B model on a newly-curated dataset of only pre-1930 data. Try it below! with @alecrad.bsky.social and @nicklevin01.bsky.social

🚨 J-2 pour le Late Show du Siècle On voudrait bien inviter Spinoza mais il va falloir cravacher sur ces derniers jours ! Encore un effort ! fr.ulule.com/le-late-show...

I have been using GPT ImageGen-2 for the past weeks I didn't think that better image-generators would be a big deal but it turns out that there is a quality threshold I didn't expect, where you can now get usable text, slides, academic papers Look at what it does with my "otter test"! (Zoom in)

Much discussion about AI risks focuses on the dangers increasingly advanced AI systems pose to us. But what about the dangers we might pose to them? Many people think discussing "AI welfare" is absurd. I disagree! Discussion with Henry Shevlin and Rob Long here: www.youtube.com/watch?v=mEMf...

The new Erdos GPT proof seems like the real deal.

J'ai suivi d'une demi-oreille le live de @mathieu-burgalassi.bsky.social qui parlait d'altruisme efficace, et c'était assez rude à écouter. 😅 Il y a plein de critiques légitimes à faire au mouvement, mais en s'attaquant à un tel épouvantail de ce qu'est l'AE, rien de très intéressant n'en ressort.

New post. Training AI models doesn't emit that much if we just make normal comparisons instead of crazy ones. Most popular coverage of this made crazy comparisons. blog.andymasley.com/p/training-a...

Working on a quick post about how training AI models doesn't actually emit that much, and that becomes clear if instead of comparing it to your individual lifestyle, you just compare it to the creation of other popular consumer products

Nouvelle vidéo : Il faut qu'on parle de Claude Mythos youtu.be/JBaBAg4ny6U

Neat experiment finds AI fact checks on Twitter are rated as more helpful & less ideological than human ones. "LLM-generated Community Notes can achieve broader cross-ideological acceptance than human-written notes, receiving more positive ratings from raters across the political spectrum"

You know, candidly, I think I believed the capability curve would be starting to look more sigmoid by this point. Also: I would like it to start looking more sigmoid.

2e palier atteint : Hobbes invité du Late Show du siècle pour un épisode de philosophie politique moderne ! Prochain objectif Spinoza. Allez !

Happy Easter.

A strange popular misconception is that an AI system is inherently “only capable of regurgitating its training data”. This has never been true. To the extent that a model does regurgitate its training data this is known as “overfitting” and is classically understood as a problem to avoid.

Bonne revue du dernier livre de Pascal Boyer www.telos-eu.com/fr/politique...

Quelle image incroyable de la Terre par Reid Wiseman à bord de la mission Artemis II! Admirez les aurores boréales et australes!

Bientôt 10 ans que je suis là. youtu.be/I8MluAD8QfM

On vous raconte dans @le_Parisien une nouvelle "renaissance". Celle d’Olivia, 29 ans, qui connaît enfin l’origine de sa maladie et de ses malformations grâce à un nouveau logiciel d'intelligence artificielle : "C'était terrifiant !" ⤵️ Le Parisien 1/10

Bonjour @mediapartblogs.bsky.social 👋 Une tribune publiée sur votre Club reprend des éléments de langage des lobbies de la viande, contient des affirmations fausses, et compte parmi ses signataires des personnes aux conflits d'intérêts documentés. C'est embêtant. blogs.mediapart.fr/sandrine-bar...

A new Anthropic paper argues for functional emotions in LLMs, claiming a causal link between emotional representations and model behavior. transformer-circuits.pub/2026/emotion...

Did you know that methane has caused about 30% of the warming we've seen to date? (Plus, it causes a lot of other air pollution...) And did you know that the biggest methane emissions come from food systems? drawdown.org/insights/we-...

🧵 Ce matin, Le Monde publie un portrait sur mon parcours de doctorant et de militant pour la cause animale, dans une série qui met en lumière des parcours de personnes en situation de handicap engagées dans la vie publique. www.lemonde.fr/societe/arti...

?!?!?!

Okay folks I've responded to the data center heat island paper. It is the worst study on AI and the environment I've ever read, by far. Thank you to @beijingpalmer.bsky.social for flagging. blog.andymasley.com/p/data-cente...

New scoping review in Science of Food: A more sustainable food system needs us to move away from meat; one of the best tools for achieving that is alt protein; and among alt proteins plant-based is best for now because of scalability. Get (back) on board. www.nature.com/articles/s41...

This is super interesting to read because I was very concerned about the water and AI thing so I went to my one scientist friend who literally worked on water reg/quality/usage and asked her about AI, data centers, water etc. She was far more concerned with agriculture abuses than AI.

While social media is polarising, evidence suggests AI may nudge people towards the centre. www.ft.com/content/3880...

Unpopular opinion: plutôt que de baisser temporairement les taxes sur l’essence, il serait préférable de supprimer temporairement les droits de douane de 45% sur les voitures électriques chinoises