arxiv.org
Stealing Reasoning Traces from Proprietary LLM APIs
Leading large language model providers now conceal their models' step-by-step reasoning, or chain-of-thought, to protect intellectual property and limit information leakage. Rather than storing these ...
