Computer chess, Paul Blart, and cyberpunk — Nov 30 Weekly Roundup
What I read, watched, and listened to this week.
Reading
OK, I can partly explain the LLM chess weirdness now (Dynomight)
Why is an old model like ChatGPT 3.5-instruct the only model that’s good at chess? Why are OpenAI’s closed models better than open models? This is a fascinating deep dive into what makes a model good or bad at chess.
Listening
Til Death Do Us Blart 2024
For the last ten years the McElroy brothers along with Tim Batt and Guy Montgomery (of The Worst Idea of All Time fame) have gathered to discuss Paul Blart Mall Cop 2 around Thanksgiving. This episode (which also has a video version) includes some wonderful surprises (and lots of NSFW language, I warn you).
Watching
Anthropic’s New Agent Protocol (Sam Witteveen)
This is a terrific explainer and demo of Anthropic’s new “Model Context Protocol,” an open protocol for connecting LLMs to external data sources.
