John Regehrsome nice followup to my group's (non-LLM) work on generalizing peephole optimizations. the non-LLM side of program synthesis really does seem to be dead at this point.
ash@regehr I still feel pretty conflicted about it given how generally trained *LLMs* seem to be. Someone else here a while ago pointed out the resemblance of this to ML applications in microbiology, which, generally, to my goofy and probably overly anti-AI ass mind, makes sense.
Anyway, to get to the point, do you see specially trained *ML* making greater strides than Chud Jippity? A development of a more specialized algorithm derived from these neural networks would make me feel better, lol. Otherwise it feels like a great homogenization.
@ashguy I haven’t predicted anything about modern AI correctly so far, and I doubt I’m going to be able to start now