Hidden Reasoning in LLMs: A Taxonomy

See this post on LessWrong here.