[ursAxZA]

Chain-of-thought is a technical term in LLMs — not literally “what it’s thinking.”

As far as I understand it, it’s a generated narration conditioned by the prompt, not direct access to internal reasoning.