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.
A practical bank for frontier-lab research discussion, project deep dives, paper critique, ML system design, experiment design, incident debugging, and behavioral interviews—with spoken answer structures, follow-ups, traps, and worksheets.
一份可直接复习的 frontier lab 讨论题库：项目深挖、论文批评、ML 系统设计、实验设计、事故定位和行为面试；每节都有口头回答结构、追问、陷阱与练习表。
Complete, tested implementations for frontier-lab-style coding practice — attention, KV cache, RoPE, sampling, GRPO, BPE — plus the probability and linear algebra that comes with them, and a timed practice harness.
用于 frontier-lab 风格 coding 练习的完整实现，全部带测试——attention、KV cache、RoPE、采样、GRPO、BPE，加上配套的概率与线性代数，以及一套限时练习工具。
A concept-first, comprehensive review of LLM and ML foundations with selective challenge questions, interview follow-ups and common traps. Built on Alisa Liu's public notes plus material on data, agentic RL and calibration.
一份概念优先、系统完整的 LLM/ML 基础复习指南，配有选择性的迁移题、面试追问和常见陷阱。基于 Alisa Liu 公开的学习笔记，加上数据、agentic RL、校准等补充。
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。