06/22/2026
Amid claims of rapid advancement for AI’s mathematical prowess, the project known as “First Proof” gained widespread attention in February as a novel experiment to independently test AI’s ability to solve tough problems. The organizers, including Harvard professor Lauren Williams, knew they wanted to try again, with a formalized structure — a newly registered nonprofit, First Proof Foundation Inc. — more transparency, and more controls for the next round, which they dubbed “First Proof, Second Batch.”
Some answers were correct and solved at a high level, in a similar way to a published paper. Some were correct but were so strange or clunky they took hours for a mathematician to decipher. Some were mathematically correct, but cribbed so closely from published works without citing them that they wo...