Stanford prof questions Arc biology contest winner's reliance on train-test similarity
anshulkundaje · x · 2026-11-20
Stanford professor Anshul Kundaje publicly questioned the winning solution of an Arc Institute-related biology competition on X, asking whether the held-out cell lines were properly excluded from training data — and whether the best submission depended on train-test similarity. He noted that while this isn't against the rules, claims of invariance to train-test similarity would imply a weighted-means trick or a serious structural exploit in the dataset.
More from Fun
- She turned her boyfriend's chore list into GTA6 mission cards starring Lucia — kscottz · 2026-11-20
- 30,000 words and 75 hours: ex-Ethereum CTO argues every 'one shot' is slop — CatAstro_Piyush · 2026-11-20
- Zara's website is impossible for humans to shop — only AI agents can navigate it — annetgriffin · 2026-11-20
- Painterly retro pixel-art scenes made with Fable tease a new 'sanctuary for 4o' — RileyRalmuto · 2026-11-20
- Chinese student builds wearable tactical backpack robot, demo goes viral — TinfoilTricorn · 2026-11-20
- PhD scholar revives a dormant microscope by building control software with Claude — Paimaamu · 2026-11-20