Study published showing mid-training on 27B tokens boosts math scores and enables effective RL

Analysis reveals bottleneck in RL environment creation, proposing shift to distributed bounty systems

Researchers develop a novel multi-level meta-reinforcement learning framework for hierarchical task mastery

Researchers publish a minimax optimal algorithm for RL with delayed state observations, achieving provably optimal regret bounds.

Proposed new system to create diverse reasoning pathways in language models.