I Don't Want to Interact with Stochastic Parrots

I have a SKILL that emulates how my supervisor prefers his answers and papers, he used to get triggered by generated stuff, but now it just writes in a style he likes which I got from a bunch of his papers and everything goes smoother. If you are bottlenecking someone they will just do the same to you.

And never mind that the whole premise of this article is simply wrong. Look at the things GenAI systems are doing today and it's clear that they are more than just "parrots" stochastic or otherwise. Not even looking at the Navier-Stokes thing, but there are Youtube videos out there of PhD candidates in math who have tried out various models from a "can this help me with research level mathematics?" perspective multiple times, and quite often the answer is at least a qualified "yes". Does it really make sense to say that a stochastic approximation of "very average" content can be helpful in doing research level mathematics? I'm thinking "no".

My own anecdotal experience also suggests that whatever LLM's do, it amounts to more than being a mere parrot. ChatGPT was incredibly helpful to me over the weekend, for example, getting started on some neuromorphic computing stuff I wanted to do. It was amazingly helpful at debugging circuit problems, helping me setup some LXI/SCPI automation of my test equipment, doing literature reviews, yadda, yadda, yadda. Even better, unlike most human interlocutors I might want to interact with, it is fine with my weird "night owl" hours, happily debugging circuits at 2:00am, 3:00am, and on into the early morning hours. It never gets tired, never belittles me, tolerates my repeated questions, etc.

I get that some people are "anti-AI" for their own reasons, but I hope some of those people will at least stop and try to understand why some of us think modern GenAI systems are about the best thing since sliced bread. shrug

This is, ironically, similar to my mom's anti-mental-healthcare argument that the system producing a good outcome is not evidence that the process is correct.

It's everyone's choice what stances they take, and if being utterly opposed to any kind of AI interaction is their's, that's their choice, but just own that choice without resorting to absurdity to attempt to lend it a false sense of credibility.

LLMs don't produce working code through mere hallucinatory chance, they do it by mathematical approximation, which is no more or less valid than pasting together programs from stack overflow answers.

Idgaf if the person who coded a program had intention when writing a line of code I never see because it's a compiled binary blob, I care that the program works. And frankly, pretending otherwise is pretentious BS.

I still don't see it as data point we can differentiate human-ness on.