Oct 2025
Five-bit counterexample to majority optimality
Model / AI: GPT-5 Pro
self-reported
Ivanisvili and Xie report that GPT-5 Pro found a Boolean function outperforming five-bit majority at p = 0.40. Their paper gives the explicit function and a manual calculation. The authors also prove optimality of majority near p = 0 for each fixed odd dimension.
Source read 2026-09-08. Located through seewoo5/awesome-ai-for-math. Read the problem statement, AI attribution and description of manual verification in the paper. The calculation was not independently reproduced.