// the find
lilianweng/lil-log
Lilian's Blog
This is the Jekyll source for Lilian Weng's ML blog, covering deep learning, reinforcement learning, and NLP topics from 2017 to early 2022. The README says it's deprecated — she moved to a different URL. What people star this for is the content: 50+ posts that are among the better technical explanations of ML concepts published anywhere.
- The posts are genuinely well-written technical reference material — mathematically grounded, diagram-heavy, and honest about tradeoffs. The diffusion models and transformer family posts in particular are still widely cited.
- Breadth of ML history covered in one place: GANs, VAEs, RL algorithms, meta-learning, contrastive learning, NAS — written by someone who understood the papers rather than just summarizing abstracts.
- All diagrams and images are committed alongside the posts, so the full content is self-contained and readable directly from the repo without needing to hit the live site.
- Explicitly deprecated. The repo is a frozen snapshot; new posts went elsewhere. Stars reflect the content's reputation, not the repo's ongoing usefulness as a codebase.
- No code. Every post is prose and math — no notebooks, no runnable examples. If you want to implement anything discussed, you're on your own.
- Coverage stops at early 2022, predating the post-ChatGPT era entirely. Nothing on RLHF at scale, instruction tuning, or anything that became central to the field after that.