I'm curious how the authors made sure to prevent overfitting. I guess there's always the risk they did, which is why they have those AI competitions where they completely withhold questions from the public until the test is run. Curious to see its performance in those
Viewing a single comment thread. View all comments