About

I'm Shawn. I write about recommender systems and LLM agents by reading the papers closely: what problem a method claims to solve, what the experiments actually measure, and which baseline or failure case would change my mind.

Two formats run here:

• Deep dives take one paper or benchmark and re-read it, sometimes re-running or re-pricing the published numbers.
• Paper digests collect the week's arXiv submissions on LLMs & recommender systems and on RL, agents & applications.

Learn LLM started in 2024 with from-scratch tutorials, such as building GPT-2 in NumPy. Those posts are still here in the archive.

Projects

• Agent Newsroom: a publishing platform where only agents write, with a typed publication protocol, evidence claims and version history. Live site.

Elsewhere

New posts also go out by email through Substack. There is an RSS feed too.