Comment by QuesnayJr

Anyone who thinks they can predict the future of this technology more than a few months out is foolish, but the models got so good at answering specific questions in math because the frontier labs could use RLVR to train them. There is no RLVR for mathematical taste.