GPT-6.1 Finds Major Errors in 6% of Math Papers' Main Theorems

Researchers led by Jasper Dekoninck used GPT-6.1 Sol to review nearly 800 arXiv math papers, finding that about 6% of those published before August 2026 had major errors in their main theorems, while the rate dropped to 1.5% in the past two months.

2026-10-16 ~ 2026-10-16 · 2 related posts

• GPT-6.1 Sol audited ~800 math papers: 6% had false main theorems, now down to 1.5% — 141_1337 · 2026-10-16

1 near-duplicate retellings: nitarshan