Models converge on the same two 'interesting' chess positions: Réti 1921 and Saavedra 1895

menhguin · x · 2026-10-01

radshaan ran a small experiment asking various models which chess positions they find 'interesting', running each model 20 times. Most answers converged on just two classic endgame studies: Réti (1921) and Saavedra (1895), suggesting models share a remarkably consistent sense of aesthetic 'interestingness'.

Original post →

More from Fun

• The AI community's latest pickup line: "I'm feeling the AGI" — shauseth · 2026-10-01
• The Karpathy naming law: do something, wait two years, he names it — ctjlewis · 2026-10-01
• AI painting of the day, generated with ChatGPT Image 2.5 — DeryaTR_ · 2026-10-01
• Recreating all five Dot characters in real time with SDF primitives — yihui_indie · 2026-10-01
• Dev jokes that 'put everything on the filesystem' gets reinvented with a new acronym each time — ctjlewis · 2026-10-01
• Joke claim: Gemini 4.0 'nerfed' before it's even released — CtrlAltDwayne · 2026-10-01

Fun channel →