Outcome-reward reinforcement learning such as GRPO gives every token of a search-agent rollout the same advantage, although the rollout exercises two capabilities: finding evidence and using it. We show that accuracy improves without grounding: Search-R1 raises accuracy on all seven QA benchmarks, yet on Bamboogle the...
This work proposes Multi-Signal Event Modeling and Dynamic Rescoring (MEDR), a training-free and query-independent frame selection method that preserves the reusability of a fixed visual input while improving the coverage of informative events beyond uniform sampling.
Xin-Lei Pu, Wei-Jie Shi, Wen Yang et al.· arXiv.org· 1 citation
A new harness self-evolution method, named DREvo, is proposed, which integrates function-level evidence anchoring, state-dependent evidence recalibration, and role-conditioned search intent distillation to determine which historical evidence remains valid and where the harness should evolve next.
SkillReranker is proposed, an inference-time reranking framework for adaptive skill selection that effectively improves task performance, reduces environment interaction steps, and lowers token consumption compared with existing skill selection baselines.
Yanping Chen, Weijie Shi, Wen Yang et al.· arXiv.org· 3 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.