LLM Reasoning Is Latent, Not the Chain of Thought

Wenshuo Wang

This position paper argues that large language model (LLM) reasoning should be studied as latent-state trajectory formation rather than as faithful surface chain-of-thought (CoT).…