In my view, any kind of learning boils down to two key elements: training data and training methods. For enhancing LLM reasoning or replicating OpenAI o1, obtaining long-thought chains as training data is crucial.