When training agents, you add a length penalty for exceeding the budget/effort level. The fun thing is, every lab (Deepseek, Kimi, ...) uses a different one. We compare which works best in our new paper. 📄 arxiv.org/abs/2610.28409 🧵1/9

Show Post

🚨 New preprint: "Dino Forcing Flow Models: Do not denoise what you can predict" @arrijitghosh.bsky.social @lucasdegeorge.bsky.social P. Couairon, A.A. Efros, @vickykalogeiton.bsky.social and me 📜 arxiv.org/abs/2610.11751 🖥️ arijit-hub.github.io/projects/din... 🤖 github.com/arijit-hub/d... 🧵👇

Show Post

Trying to use the models for things outside of coding / ml, like making slides or coherent plans or longish content design, makes me feel the AGI-is-going-to-take-a-while.

Show Post

Introducing REKI, a 2.4B vision-language-action model pretrained on public data only. With a simple design, it rivals larger models trained on in-house data with complex multi-stage pipelines, setting a new Pareto frontier between performance and size. kyutai.org/reki

Show Post

"Scaling Discovery Requires Forgetting" This paper proposes forgetting, where you have the model discarding sessions histories, keeping only the best solution found so far A fresh team of agents then resumes from that solution, which improves the model performance. www.alphaxiv.org/abs/2610.20563

Show Post