Jian Wang

Extended reasoning modes paradoxically degrade an LLM's ability to assess evidence sufficiency in dynamic diagnosis, driving models to prematurely lock into diagnoses that succeed only by chance.