We propose a four-stage training task for LLMs—Fast Thinking, Verification, Slow Thinking, and Summarization—separating the training of intuition, verification, reasoning, and integration.