This makes no sense. To believe this you have to think that the models are somehow being overfit explicitly on academic mathematics and it doesn’t carry over at all to more practical software engineering. I wouldn’t make that bet.

This also makes the assumption that frontier math has all the long hanging fruits already taken... also very dubious.