← back to paper
arxiv: 2606.17524 · 2 revisions
Learning to Refine Hidden States for Reliable LLM Reasoning