harmful-continuation

This paper investigates a harmful phenomenon in long chain-of-thought (CoT) training traces where post-conclusion continuation reduces training utility, and proposes a diagnostic method called HarmfulContinuationCut (HCC) to detect such harmful continuations.

This paper identifies harmful continuations in answer-correct long chain-of-thought training traces for LLM SFT, characterized by uncertainty-geometry mismatches, and proposes a lightweight boundary proxy method to remove them.