Stanford NLP's Aryaman Arora argues high AI coding benchmarks are driven by explicit training on targeted tasks

@inductionheads eh, no one ever logs on here and says anything like “actually we didn’t train on long horizon coding tasks to make the model good at long horizon coding