Knapsack RL Unlocking Exploration of LLMs via Optimizing Budget Allocation
RESEARCH
-

DeepSearch: Reinforcement Learning with Verifiable Rewards via MCTS
By
–
DeepSearch Overcome the Bottleneck of Reinforcement Learning with Verifiable Rewards via Monte Carlo Tree Search
-
Theory vs Practice: AI’s Practical Breakthroughs Challenge Old Guard
By
–
"It works in practice – but does it work in theory?". My biggest problem with the criticism from the 'old guard' of AI (e.g. interview with Rich Sutton) is that there seems to be complete lack of engagement with what has been achieved in the last 2-3 years. The fact that we went
-
GEO LLM Poisoning: Emerging Security Threat for AI Models
By
–
yes, le jour où les 2 se combinent. Moi je suis pas assez intelligent, ni bon codeur pour aller au bout des choses. J'aime bien faire des concepts. Le GEO LLM poisoning ca devrait vous plaire pour le jour où les LLM feront cette inférence. Je vous pose ça là, vous me donnez
-
UCLA Builds Optical Generative Model Running on Light Instead GPUs
By
–
This is huge!
— 机器之心 JIQIZHIXIN (@jiqizhixin) 2 octobre 2025
A UCLA team managed to build an optical generative model that runs on light instead of GPUs.
In their demo, a shallow encoder maps noise into phase patterns, which a free-space optical decoder then transforms into images—digits, fashion, butterflies, faces, even… pic.twitter.com/qTz43q3tcIThis is huge! A UCLA team managed to build an optical generative model that runs on light instead of GPUs. In their demo, a shallow encoder maps noise into phase patterns, which a free-space optical decoder then transforms into images—digits, fashion, butterflies, faces, even
-
Google and OpenAI: Top AI Race Contenders After Three Years
By
–
After 3 years of constant change, the top 2 contenders in the AI race are still Google and OpenAI.
-
PC Programs Internet Apps AI Age Brains Evolution
By
–
PC age: programs
Internet age: apps
AI age: brains -
2026 Nobel Prize Predictions: Altman, Li, and Sutton
By
–
Still in denial? This year's Nobels will go to Sam Altman for Sora 2, Fei Fei Li for putting the dogs into ImageNet, and to Rich Sutton for discovering that we need to stop doing physics manually
-
Chinese AI Open Weight Model Releases: DeepSeek and GLM Updates
By
–
I fell a little behind on Chinese AI open weight model releases, here's pelicans for DeepSeek-V3.2-Exp and GLM-4.6
-
Video Model Control Using Text Prompts and VLM Success Detection
By
–
Thanks. What I had in mind is: The control is eg the text prompt. Then one simulates forward with the video model. One verifies whether the desired goal was attained using eg VLM success detectors https://
arxiv.org/abs/2303.07280. The action text is an intermediate signal generated by a
