I'm fascinated to learn that people are discovering weaknesses in state-of-the-art bots for playing Go, such that a novice player can reliably win: https://goattack.far.ai/human-evaluation
This suggests that self-play doesn't always generalise: it's not sufficient to beat earlier versions.
miniblog.
Related Posts
I've been playing with Obsidian and having a great time. It's fundamentally a .md editor but it has so many affordances that it feels different. Link autocompletion, highlighting backreferences, polished mobile app.
A lot of teaching resources focus on folder structure, oddly.
After further playing with my LLM project, I'm surprised how hard it is to tune with system prompts.
My agent kept saying "obviously" even though my prompt said "helpful, professional". Eventually I found that a "courteous" prompt gets "of course" which is better but not ideal.
I'm playing with Zeal/Dash so I can view stdlib docs offline.
(I've been coding on public transport recently, without a reliable data connection.)
Anyone have tips for making the most of this setup? AFAICT docsets are basically HTML underneath.