Comment by fatcatsbestcats

Conversely, the TAM for tasks that cheap models can and have been able to one-shot for a while now is orders of magnitude larger than the TAM at the frontier. While it’s kind of cool that frontier models are able to solve Millennium Prize problems, there just isn’t a huge market for that sort of thing. Otherwise, professional mathematicians would be getting paid a lot more than they do now, and there would be more of them.