AI RESEARCH
RL with Learnable Textual Feedback: A Bilevel Approach
arXiv CS.LG
•
ArXi:2605.24547v1 Announce Type: new Reinforcement learning with verifiable rewards can improve LLM reasoning, but learning remains sample-inefficient when terminal rewards are sparse. This has motivated a growing line of work on RL with textual feedback, where a critic model generates natural language feedback to guide a reasoning model (the actor), augmenting scalar rewards with richer learning signals.