How could a simple self-replicating system emerge at the origins of life? RNA polymerase ribozymes can replicate RNA, but existing ones are so large that their self-replication seems impossible. Could they be smaller? Excited to share our latest work in @science.org on a new small polymerase. 1/n

So how does it work? Meet Miniature (mini + IA, get it?), a complete AI agent in ~225 lines of Python. No frameworks, no abstractions. Code: github.com/glouppe/miniature

🎉 Announcing TabICLv2: State-of-the art Table Foundation Model, fast and open source A breakthrough for tabular ML: better prediction and faster runtime than alternatives, work by Jingang Qu, David Holzmüller @dholzmueller.bsky.social , Marine Le Morvan, and myself 👇

So how does it work? Meet Miniature (mini + IA, get it?), a complete AI agent in ~225 lines of Python. No frameworks, no abstractions. Code: github.com/glouppe/miniature

New preprint 📢 arxiv.org/abs/2602.09161 Lead by Sherman Khoo (PhD student at Compass CDT), with Song Liu and Mark Beaumont. We adjust SBI summaries at test time to improve robustness.

David Hogg has written a white paper on doing astrophysics in the age of LLMs. It looks to be thought-provoking. My initial reaction is that either LLMs will destroy the field or they will force a reckoning with and re-imagining of the current system that often prioritizes output over quality. 🧪

From raw LLM weights to a full AI agent (tool calls, reasoning loop, the whole thing) in under 200 lines of Python. New lecture coming to my deep learning course this year. 😎

ProbML 2026 (formerly AABI) invites submissions on probabilistic ML (both Bayesian and otherwise!), July 5 in Seoul (co-located with ICML). Website: probml.cc. Tracks: proceedings (PMLR), workshop, fast track. New focus includes applications in healthcare and climate! Submit by: 20 March 2026.

When I was a kid, computer viruses of all kinds were around. Worms, trojan horses, rootkits... Then things settled down for years. Firewalls got better, antivirus became standard, and honestly? It got boring.

🤯“I tasked 16 agents with writing a Rust-based C compiler, from scratch, capable of compiling the Linux kernel. Over nearly 2,000 Claude Code sessions and $20,000 in API costs, the agent team produced a 100,000-line compiler that can build Linux 6.9 on x86, ARM, and RISC-V.”

Seriously, shall we still teach students how to even program?

Can't wait to continue hacking my openclaw instance with these new superpowers. It's been pretty fun so far :-)

It's always frustrating to me how the term AI has evolved. AI was an umbrella term from 1950s. Then Machine Learning (a subset of AI), Deep Learning (a subset of ML), and modern generative AI (a subset DL). Now LLMs are used synonymously with "AI". So I made these graphics of the "AIroboros"

Whoa… 9 years ago, @glouppe.bsky.social and I teamed up with @kyunghyuncho.bsky.social on what I think was the first paper to make a connection between language modeling and particle physics. It also had a cool tie in with intepretability arxiv.org/abs/1702.00748

"Cherry picking" in scientific papers. (Repost of an older post on some other site in 2021).

Stars in our galaxy aren’t distributed evenly. Collisions with smaller galaxies or clusters make “stellar streams,” which are long, thin trails of stars. They form as the smaller object stretches while it falls into the Milky Way. So… why does this stream have a MASSIVE hole in it?? 1/7 ⚛️🧪

Scaling Laws in Particle Physics Data! This is a result I've been itching to share and it's finally out. One of the big open questions is how much better AI-based methods at particle colliders can still become. 1/4

“We found that using AI assistance led to a statistically significant decrease in mastery.” Props to Anthropic for studying the effects of their creation and reporting results that are not probably what they wished for www.anthropic.com/research/AI-...

Our AlphaGenome paper and models are out today! 🧬 I'm really excited to see how people use the model and build on top of it in their biology research🔬 Huge congratulations to the entire team!👏 Read the full paper and access the model: 📄 Paper: goo.gle/4bXlV6y 💻 Code: goo.gle/4k1xrzI

I am old enough now to be mentioned as a veteran in simulation-based inference 👴(joking, thanks Thea for the highlight!)

My visit to #RockyWorlds4 kicked off with a hot take from @nplinnspace.bsky.social's poster: Even in the best-case scenario, we couldn't tell how "Earth-like" an Earth-like planet is 🔥 She assumed TRAPPIST-1 e had Earth-like air and showed that vastly different climates would be indistinguishable.

The risk of AI for education is not students cheating in exams, it is people in general cheating themselves into believing they understand things they don’t.

Cool new paper from my colleague @jsellenberg.bsky.social @uwmadison.bsky.social Math + collaborators. They used AlphaEvolve to generate interpretable programs that led to the discovery of new mathematical structures in the symmetric group arxiv.org/abs/2601.01235

One of my favorite findings: Positional embeddings are just training wheels. They help convergence but hurt long-context generalization. We found that if you simply delete them after pretraining and recalibrate for <1% of the original budget, you unlock massive context windows. Smarter, not harder.

Proc B with @sampassmore.bsky.social! We used simulations to explore the innovation strategies of speed climbers 🧗‍♀️ Innovation is higher among slower athletes and lower when the population size is larger, and the overall balance of innovation and copying appears to be suboptimal 🔗 bit.ly/499QjZM

We introduce epiplexity, a new measure of information that provides a foundation for how to select, generate, or transform data for learning systems. We have been working on this for almost 2 years, and I cannot contain my excitement! arxiv.org/abs/2601.03220 1/7

A study shows that using AI tools like LLMs lowers cognitive engagement and critical thinking in essay writing, signaling a trade-off between convenience and growth. LLM users had weaker memory recall and ownership, raising vital talks about AI's role in education. https://arxiv.org/abs/2506.08872

Truly this site hosts the most controversial of opinions

2020s: AI "System Prompts" are lengthy, carefully constructed sets of expert rules about a particular domain, created by "prompt engineers". 1980s: AI "Expert Systems" were lengthy, carefully constructed sets of expert rules about a particular domain, created by "knowledge engineers".

Happy New Year! Do your plans for 2026 include... - working with a great team lead by @wellingmax.bsky.social and @aronwalsh.github.io? - living in Amsterdam, Berlin, London, or Cambridge? - using fun tools from ML and material science? - solving important problems? Then join us at CuspAI! 1/2

A biologically detailed brain model has matched animal learning performance and uncovered previously overlooked neuron activity linked to errors during visual categorization tasks. doi.org/hbhcsn

I have been in AI for science for 10+ years now. Real benefits, real impact. But 'AI' feels like a bad word now, with the media running 10:1 negative. Yet almost everyone around me uses LLMs daily. Honestly lost.

Coding.

Very nice arxiv.org/abs/2512.20003

What an amazing Yule gift from @stefanradev.bsky.social & colleagues: a tour-de-force tutorial on diffusion models for simulator-based inference. This is one of the most comprehensive and useful review/tutorials I have ever seen -- a must read! Kudos to all the authors! arxiv.org/abs/2512.20685

Doing a PhD is - at heart - one long discussion with your mentor. The discussion changes over time - with unexpected turns and ups & downs - but through it all is a pair of people discussing a topic endlessly to make sense of it. PhD students: choose someone you like to talk to!

New AI models from @polymathicai.bsky.social use knowledge from scientific fields and apply it to others. Walrus (seen below) can tackle systems from exploding stars to Wi-Fi signals: https://www.simonsfoundation.org/these-new-ai-models-are-trained-on-physics-not-words-and-theyre-driving-discovery/

Your pretty picture of the day 😀 🌊with of course many interesting processes! Water rich in suspended sediments is moved around by ocean currents in the Bay of Biscay. I'm curious about the long-range waves seen in the dark region, can't decide if these are oceanic, atmospheric, or an artifact 😅

New orbit movie just dropped. This one is for the newly discovered circumbinary planet HD 143811 AB b. This 9-year timelapse was made through an international collaboration with me, Nathalie Jones (Northwestern graduate student), and Vito Squicciarini (Exeter postdoc)

Excellent post d'une traductrice juridique qui ne mâche pas ses mots à propos de la post-édition (MTPE). Il y a beaucoup de punchlines percutantes, mais en résumé : nous ne sommes pas des serpillières juste là pour nettoyer les flaques de d-IA-rrhée.

Repeat after me: This is NOT a paper contribution, it’s an expected component!

was at an event on AI for science yesterday, a panel discussion here at NeurIPS. The panelists discussed how they plan to replace humans at all levels in the scientific process. So I stood up and protested that what they are doing is evil. Full post: togelius.blogspot.com/2025/12/plea...

Why can neurons with completely different ion channels fire identically? 🧠 Neuronal degeneracy isn't a bug - it's a feature. Come see how we reconstruct entire degenerate populations from spike times within an interpretable feedback framework. Poster @dataonbrainmind.bsky.social #NeurIPS2025 #DBM

If you are interested in diffusion models for dynamical systems and posterior inference, stop by our poster at ML4PS workshop (@neuripsconf.bsky.social) at 11am! Come and see how we apply modern methods to the challenging field of global weather inference!

The Nobel Prize committee should announce the World Cup winner tomorrow

Later we did early work using message passing graph NNs with @joanbruna.bsky.social and @johannbrehmer.bsky.social. Fully connected message passing NNs are pretty similar to transformers, but this was in 2017. ml4physicalsciences.github.io/2017/files/n...

An unexpected shoutout from @kyunghyuncho.bsky.social in his NeurIPS keynote. @glouppe.bsky.social & I collaborated with him to bring ideas from natural language to high-energy physics back in 2016. arxiv.org/abs/1702.00748

Super excited (and a bit nervous) to give a keynote tomorrow afternoon at #EurIPS. I'll be math, LLMs, and politics 🔥

Beautiful keynote by @gaelvaroquaux.bsky.social at @euripsconf.bsky.social .

🎉 "High-dimensional neuronal activity from low-dimensional latent dynamics: a solvable model" will be presented as an oral at #NeurIPS2025 🎉 Feeling very grateful that reviewers and chairs appreciated concise mathematical explanations, in this age of big models. www.biorxiv.org/content/10.1... 1/2