AI brackets near perfect
AI brackets from Gemini/Claude hit 97.8% accuracy—40 of 41 picks correct in a recent run—leaving just 2 perfect brackets standing in a field of roughly 26 million entries. (x.com) That level of model accuracy is shredding traditional bracket pools and changing how people hedge tournament bets. (x.com)
A post on X shared screenshots claiming Google’s Gemini and Anthropic’s Claude combined to get 40 of 41 bracket picks correct (97.8%) in a single run, a result the poster said left two perfect brackets standing in ESPN’s field of roughly 26.6 million entries. (x.com) ESPN reported 26.6 million completed brackets in its Tournament Challenge before the tournament began, the largest total in the game’s history. (espnpressroom.com) News wires and local outlets tracked the shrinking pool of perfect entries across platforms, reporting that ESPN briefly showed two perfect men’s brackets and the NCAA’s tracker showed four before those final entries were eliminated after Tennessee’s Round of 32 win. (clickondetroit.com) Major sports publishers and betting sites ran full AI-generated brackets this week—CBS Sports, Sporting News, Yahoo Sports and specialist sites deployed Gemini-, ChatGPT- and Claude-powered simulations and published round-by-round AI picks. (cbssports.com) Betting and prediction-market operators flagged heavier use of model outputs for hedging; Kalshi reopened a high‑stakes bracket product this year and analysts noted prediction markets and AI tools are increasingly used to construct hedge positions instead of traditional office pools. (foxsports.com) Tournament-tracking services and outlets are continuing live comparisons between human and AI brackets and caution that the high 97.8% figure described in the social post reflects an early-round snapshot rather than end-to-end tournament performance. (ncaa.com)
