1% of Tokens Can Be Enough: On Gradient Estimation in On-Policy Distillation
Score Centering Stabilizes Off-policy Reinforcement Learning
1
Mind2Dialogue: Training Human-Aware Language Models by Simulating User Mental States
1
Expert-Space Exploration in MoE Reinforcement Learning
Learning to Solve Hard Problems in RL for LLMs by Never Giving Up
Training a 3.8B LLM to 0.384 CORE for $998 – Hugo Vergnes
15 min1
ToolLoop: Closed-Loop Tool-Use Data Synthesis via Decomposed Generation and Dynamic Self-Feedback
1
Difficulty-Adaptive Tree-Structured Policy Optimization for Expanding Reasoning Coverage in RLVR
2
[AINews] 10% worse, 100x cheaper, 10000x faster: Why Simulation is taking over
15 min1