4 ms·
Well you could search for elements similar to the proof / problem in the training data, even if it's de-identified, right? OpenAI can probably do better than Ct
by dfdydx 25d ago
Well you could search for elements similar to the proof / problem in the training data, even if it's de-identified, right? OpenAI can probably do better than Ctrl-f "Navier Stokes".
- paxys 25d agoThere are probably thousands of serious academics taking a crack at millennium problems using AI every day. All those attempts are in the training data. And in fact the two researchers benefited from those attempts as well.