How Well do LLMs Compress Their Own Chain-of-Thought? A Token Complexity Approach

Open in new window