OpenAI o1 System Card
TLDR（中文）
OpenAI o1 的系统卡，展示了通过大规模强化学习训练"慢思考"模型的路线： 模型在回答前进行长时间的内部推理链，在数学竞赛和代码题上大幅超越 GPT-4。 这标志着 LLM 从"快思考"到"慢思考"的范式转变，也是 DeepSeek-R1 等模型的直接先驱。
TLDR (English)
OpenAI o1's system card reveals the approach of training "slow thinking" models via large-scale reinforcement learning: the model performs extended internal reasoning chains before answering, dramatically outperforming GPT-4 on math competitions and coding. This marks a paradigm shift from "fast thinking" to "slow thinking" LLMs.