Rishabh Joshi

Learning from negative feedback, or positive feedback or both

Reward-Guided Prompt Evolving in Reinforcement Learning for LLMs

Building Math Agents with Multi-Turn Iterative Preference Learning

RRM: Robust Reward Model Training Mitigates Reward Hacking

Evolving Alignment via Asymmetric Self-Play

Statistical Rejection Sampling Improves Preference Optimization