Introducing Mistral Small 3.2, a small update to Mistral Small 3.1 to improve: - Instruction following: Small 3.2 is better at following precise instructions - Repetition errors: Small 3.2 produces less infinite generations or repetitive answers - Function calling: Small

OMG Claude just Holy Shitted me!

AI, not humans, is going to choose the programming languages of the future It will be - english - python 🐍 - typescript - rust Primarily because AI is pretty good at them

The Hessian of multipartite entanglement defines spacetime geometry, while large-N gauge fluctuations of the underlying error-correcting code reproduce General Relativity with emergent dark energy.

𝑲𝒊𝒎𝒊-𝑹𝒆𝒔𝒆𝒂𝒓𝒄𝒉𝒆𝒓 is a fully autonomous agent trained via end-to-end RL. It executes ~23 reasoning steps and explores 200+ URLs per task. Results: - 26.9% Pass@1 on Humanity’s Last Exam (↑ from 8.6% zero-shot) - 69% Pass@1 on xbench-DeepSearch (avg of 4 runs),

Welcome Magenta RealTime: open model for music generation🎉 - 800 million parameter model - Run in free-tier Colaboratory - Fine-tuning and tech report coming soon hf.co/google/magenta…

minimax-m1 tested on Fiction.liveBench Long Context. MiniMax (official) They did it! Competitive with Gemini 2.5 Pro-preview 05-06.

A snapshot of SystemChat 2.0

#Uigen-t3-32b (coding master) in action. 54.20 tok/sec 6989tokens Eric Hartford

Introducing Reinforcement-Learned Teachers (RLTs): Transforming how we teach LLMs to reason with reinforcement learning (RL). Blog: sakana.ai/rlt Paper: arxiv.org/abs/2506.08388 Traditional RL focuses on “learning to solve” challenging problems with expensive LLMs and

AI just flipped the script on how we teach models to think. Instead of training massive LLMs to "solve" problems, Sakana AI just dropped a method where tiny models teach by explaining solutions — and outperform giants like DeepSeek R1. A 7B model trained a 32B student better

David Shapiro ⏩ The gap between near total automation and UBI will be the brutal years