Post #2970116

In addition to what rook said (that I strongly agree with), there is the issue that LLMs can only do certain sorts of math work. They can parallelize and try to brute force counter examples quite well and they can spread a very wide net trying to combine different techniques out there in disparate parts of the literature but it still remains to be seen if they can create entirely novel methods (and I and probably almost everyone on sneerclub/techtakes would bet not). And what are we left with if we let LLMs automate all the low-hanging fruit? Deskilling and devaluation of a field that cannot actually be solved by the hyperscalers’ slop machine.