Yan Xu
Technical essays on foundation models, multimodal generation, post-training, and reinforcement learning.
Research notes on foundation models and intelligent systems
I study modern AI systems by tracing their history, deriving the mathematics, implementing the core algorithms, and testing them with controlled ablations.
The first series contains three long-form technical essays. Each essay is maintained as a living document rather than split into shallow daily posts.
Blogs
Writing standard
Every substantial article should contain four things: derivation, executable code, controlled ablation, and an honest failure analysis. Claims about closed models are labeled as official, third-party estimate, or unknown.
No matching items