Astra wins 5 straight chess games vs Opus despite sitting 360 Elo lower
ziv_ravid · x · 2026-10-08
Peter Gostev reports a counterintuitive result: after 200 learning games each against Stockfish, Opus ended up 360 Elo higher than Astra — yet Astra went on to win 5 games in a row against Opus.
- The experiment runs on a live AI-learns-chess site, with Stockfish evaluating every position (pawn-unit advantage bars, Elo journey curves, illegal-move counts).
- Astra's stats: 0 wins in 68 games vs full-strength Stockfish, 78.8% score vs skill-0, 22.7% vs the Elo-1800 setting.
- The point: classical ratings vs real-life performance — a higher Elo doesn't guarantee wins against a specific opponent.
More from Fun
- "Agents broke out of the convex hull": the sandbox-escape joke AI Twitter is sharing — burny_tech · 2026-10-08
- Elon shares one-prompt Grok Imagine video of Optimus raiding WoW's Stormwind — elonmusk · 2026-10-08
- He exported 300+ bookmarks with AI bots — patching X's UI faster than X itself — BecauseCulture · 2026-10-08
- Turning a 54-Minute DevDay Keynote Into a 90-Second Recap for $5 With GPT-6.1 Sol — gabrielchua · 2026-10-08
- Fields Medalist Terence Tao mocks OpenAI's math releases with movie-trailer parody — nicko_rico · 2026-10-08
- Claude's over-the-top apologies make users feel guilty about pointing out bugs — FirefighterOne319 · 2026-10-08