literally just replace evaluation with a lottery system the results will be the same

This is absolutely terrible. Google, how far you have fallen! What's next? Fingerprints? DNA?

Opus 4.7 has a new tokenizer.
This means it's also a new base model.
Glory days of pretraining still very much going.

Where are all the Computing Science undergraduates going?

Adjacent fields such as data science, robotics, cybersecurity, etc

www.washingtonpost.com/technology/2...

I wrote a short mathematical companion tutorial to my notebook on discrete diffusion models. It gives an informal derivation of the connection between maximum likelihood estimation of the backward transition kernel and denoising score matching. github.com/gpeyre/Discr...

Now that Openclaw+Opus is not really an option anymore, what's everyone best combo for 24/7 ai agents running inside their own environment?

This editorial discusses the critical value of human-generated scientific writing in the era of large language models (LLMs), arguing that writing is essential to structured thinking and research comprehension. Writing as Thinking: The act of writing structure's thoughts, sorting research data, and identifying the main message, unlike LLMs which may lack true understanding or accountability. LLM Hallucinations: LLM-generated text requires rigorous verification because these models can produce incorrect information or fake references. Human vs. AI Roles: While LLMs are useful tools for brainstorming, improving grammar, or overcoming writer's block, human researchers must maintain control to engage in the creative task of shaping a compelling narrative.

Writing forces your brain to coordinate memory, reasoning, and meaning-making simultaneously.

Every time you write, you rewire toward clearer thinking. Every time you let an LLM do it, you rewire toward consumption.

Hey, I wrote a thing about AI in astrophysics
ergosphere.blog/posts/the-ma...

🚨I'm happy to share a preview draft of new paper "Scalars Are All You Need for Multimodal Inference". Instead of the traditional approach to multimodal foundation models for science with task-independent embedding, I outline an alternative strategy
theoryandpractice.org/2026/04/scal...

🎯 Takeaways for educators:
→ AI literacy must train question-asking & answer evaluation, not just "how AI works"
→ Training metacognition should be a priority: benefits far beyond AI use
→ An "effortless" AI interaction should be a red flag, not a comfort signal

🔍 What happens when you give middle-schoolers unrestricted access to ChatGPT for science tasks?

Findings from our new paper led by Rania Abdelghani w/ @koumurayama.bsky.social @celestekidd.bsky.social Hélène Sauzéon 🧵👇

GitHub - instructkr/claude-code: Claude Code Snapshot for Research. All original source code is the property of Anthropic. Claude Code Snapshot for Research. All original source code is the property of Anthropic. - instructkr/claude-code

Oh wow, Anthropic accidentally leaked Claude Code and it’s been cloned and made public github.com/instructkr/c...

LLM Architecture Gallery A gallery that collects architecture figures from The Big LLM Architecture Comparison and related articles, with fact sheets and links back to the original sections.

I put together a visual LLM Architecture Gallery that collects (~50) recent open-weight model designs in one place.

Architecture diagrams, config links, tech reports, explainers... you name it!

Hopefully useful as a reference & learning resource:
sebastianraschka.com/llm-architec...

Microsoft recule face à la révolte des utilisateurs contre l'IA imposée Après des mois de critiques, Microsoft admet avoir été trop loin dans l'intégration forcée de Copilot dans Windows 11 et promet de réduire sa présence.

Microsoft fait marche arrière sur Copilot après des mois de critiques. L'IA imposée partout dans Windows 11 a provoqué un exode massif vers Linux et les MacBook Neo d'Apple. Preuve que la résistance utilisateur peut faire plier les géants tech.

Jensen's inequality

Vous ne verrez rien de plus beau aujourd'hui ! ! Saturne par le JWST !

Vous ne verrez rien de plus beau aujourd'hui ! !
Saturne par le JWST !

Today NeurIPS is announcing our official satellite event in Paris.

After responding to the call from Ellis following the success of EurIPS in December, we are pleased to reach a new milestone by joining forces with the NeurIPS organizing committee for the 2026 edition.

Our paper “Multifidelity Simulation-based Inference for Computationally Expensive Simulators” has been accepted at ICLR 2026! 🥳

We hope this can be a practical solution for anyone analysing and doing inference on computationally expensive simulators.

Paper: openreview.net/pdf?id=bj0dc...

Most people use AI to answer simple questions, to rephrase text, or to write code. This project convinced me they're capable of much more than that.

There is a reveal. I won't spoil it. But the book is designed to be reread, and the second time through, every sentence means something different. AI slop? You tell me.

I didn't write a single line. My AI did everything: universe, storyline, prose, even its own editorial feedback. My role was the initial idea and a few rounds of feedback.

The pitch: An AI agent runs inside a container. It can read its own source code. It can edit it. Every thirty seconds, it receives a tick. Every sixty seconds of silence, it dies.

Those of you who've been running AI agents on your own box will get this. The rest of you might think I've lost it. 🤪

"Agent-064" is a 50-page sci-fi novel about an AI agent running inside a container. I didn't write a single line.

github.com/glouppe/agen...

A petavoxel fragment of human cerebral cortex reconstructed at nanoscale resolution To fully understand how the human brain works, knowledge of its structure at high resolution is needed. Presented here is a computationally intensive reconstruction of the ultrastructure of a cubic mi...

In a 2024 Science paper, researchers detailed a nanoscale-resolution reconstruction of a millimeter-scale fragment of human cerebral cortex, giving an unprecedented view into the structural organization of brain tissue.

Learn more during #BrainAwarenessWeek: https://scim.ag/4uxIOo3

State of AI as of March 2026: I am telegramming an agent A about coding an agent B and A delegates some of its work to C to implement B.

Meet the ultimate gatekeeper of the nucleus. This molecular machine determines what compounds are welcome inside and which shall not pass. The mechanism behind its selectivity remains a mystery. www.quantamagazine.org/disorder-dri...

10 years ago, Google DeepMind’s AlphaGo became the first program to beat a world champion at Go — a game with more moves than atoms in the universe. ⚫⚪✨

Today is international Frauenkampftag ("women-fight-day") how we call it in German to highlight that women have been fighting for equal rights since the 1910s. More than 100 years later, women still have to fight for opportunities and this post is for all of you:

Simulation Based Inference Blueprint The 1st Simulation-Based Inference Blueprint workshop will take place on the 26th and 27th February, 2026 at CERN. The workshop aims to summarize the current status of statistical techniques and softw...

Happy to see a Simulation-Based Inference Blueprint workshop being held at CERN. First papers were in 2015, and it took several years to get the first physics papers out. Hope to see it become more mainstream.
indico.cern.ch/event/160067...

I have added a new tutorial on discrete diffusion models:
github.com/gpeyre/ot4ml