Also as a model size/compressibility issue, it’s actually really hard to “cheat” by memorizing the entire internet! Claude Opus couldn’t do it and Claude Opus’s almost certainly bigger and had more training time than Pangram Labs’s classifier.
Also as a model size/compressibility issue, it’s actually really hard to “cheat” by memorizing the entire internet! Claude Opus couldn’t do it and Claude Opus’s almost certainly bigger and had more training time than Pangram Labs’s classifier.