How Well do LLMs Compress Their Own Chain-of-Thought? A Token Complexity Approach | Xiaol.x | Podwise