Think of LLMs as actors, prompts as scripts, and LLM outputs as performances.
Your article is very well written, and the test results of this paper also make sense. During the training, o1 had a lot of CoT content, so CoT didn't help him much, but incorporating the actor's role seems to be a very good approach.
Your article is very well written, and the test results of this paper also make sense. During the training, o1 had a lot of CoT content, so CoT didn't help him much, but incorporating the actor's role seems to be a very good approach.