Knowledge
Reference notes on AI fundamentals and reading lists of papers, kept up to date. Every paper links to the original.
These notes are written in Chinese for now; paper titles and links are in English.
- Machine learning basicsA plain-language map of machine learning: five ways of learning, common algorithms, neural network architectures, training and evaluation, and how they connect to today's large models.
- An overview of LLM post-trainingSFT teaches a model how to answer, preference alignment teaches which answer is better, and reinforcement learning lets it build skills on its own. SFT, RLHF, DPO, GRPO, LoRA and more, with links to the original papers.
- Landmark papers of the LLM eraRepresentative papers from Google, OpenAI, Anthropic, Meta and Microsoft, and from Chinese labs including DeepSeek, Qwen, Kimi, Zhipu, ByteDance and MiniMax, each linked to the original, plus ten to read first.
- The AI map: artificial intelligence, machine learning, deep learning and learning paradigmsA clear map of AI, machine learning, deep learning, DNNs, CNNs, RNNs, Transformers, and the different ways models learn.