I document my notes and writings on AI research, LLMs, and engineering here. A mix of long-form posts hosted on this site and selected external articles.
How alignment changes when language models become long-horizon agents: the 2026 OpenAI–Hugging Face incident, GPT-6 Astra, chain-of-thought monitoring, model–harness–environment co-design, multi-agent risk, and the research agenda ahead.
当语言模型成为长时程 Agent，alignment 应如何随之改变：从 2026 年 OpenAI–Hugging Face 事件与 GPT-6 Astra，到思维链监控、model–harness–environment 协同设计、多智能体风险，以及下一步的研究议程。
A frontier-lab loop is three tests wearing one name: your research record gets you seen, technical fluency gets you through, and a third body of unglamorous work decides what you walk away with. Seven technical buckets, with worked code.
Frontier lab 的面试是三场考试共用一个名字：研究履历让你被看见，技术熟练度让你通过，而第三类没人告诉你要准备的功课决定你最终拿到什么。七个技术方向，附可运行代码。
A stage-by-stage field guide to LLM training data — what counts as one example and who certifies it at each stage, how synthesis and verification actually work, and how to distill from frontier teachers into a small model.
按训练阶段梳理 LLM 训练数据：每个阶段的「一条样本」究竟是什么、由谁保证它正确，数据合成与校验如何真正落地，以及如何把 frontier model 蒸馏进一个小模型。
What OpenAI's booth Q&As at ICML 2026 add up to: the operating horizon is the new scaling axis, and evaluation latency, credit assignment, monitoring, and trust are what it drags along.
OpenAI 在 ICML 2026 booth Q&A 上真正说了什么：operating horizon 成为新的 scaling axis，而 evaluation latency、credit assignment、monitoring 与 trust 是它拖着一起走的东西。
How next-generation agents should route deployment experience across context, memory, harnesses, weights, and the learning mechanism itself.
下一代智能体应如何把部署期经验路由到 context、memory、harness、weights 与学习机制，并在多时间尺度上安全地持续学习。