【Tech 24H】The DeepSeek-AI team published the training method for large-scale reasoning models used in the open-source AI model DeepSeek-R1. DeepSeek-R1 employs reinforcement learning instead of human examples to develop reasoning steps, reducing training costs and complexity. After being shown high-quality problem-solving examples, it uses a template to generate reasoning processes, meaning the model learns by solving problems and receiving rewards, thus enhancing learning efficacy.