作者位次 #2/5ACL 2026Poster· Natural Language Processing
Deliberative Searcher: Improving LLM Reliability via Reinforcement Learning with Constraints
Zhenyun Yin · Shujie Wang · Xuhong Wang · Xingjun Ma · Yingchun Wang
摘要
Large language models with search capabilities frequently exhibit miscalibrated confidence, producing incorrect answers with high certainty. We present Deliberative Searcher, a reasoning-primary framework that integrates search operations into chain-of-thought generation while maintaining explici...