Yiyun Tan

The Chase Is the Curriculum, the Capture Anchors the Credit: Pursuit-Evasion Self-Play for Zero-Data LLM Reasoning

In LURE, an LLM evader positions tasks along each environment's difficulty axis to stay one step ahead of a planner-executor pursuer that hunts it down through verifiable interaction, and the unified model attains stronger aggregate OOD zero-shot accuracy than all trained baselines across nine held-out benchmarks from...