@davidad back when sonnet 3.5 came out right around the time of golden gate claude, people thought 3.5 was this but for the โvery smartโ vector. before rlvr was well known to work
asked gpt-5.5 Pro to solve Erdos #1196. 5.4 Pro actually proved it, but 5.5 got lazy and gave up on the proof halfway, instead passing off the open problem as a settled Erdos theorem, citing a real but irrelevant paper. the leaked CoT shows the exact moment it decided to bullshit