i mention this because i always thought genetic algorithms were kind of neat. but of course, they are used in exactly the same way as LLMs - to magic up software where the finished article, rather than the process, is what the initiator cares about.

i think it comes down to some key differences. firstly, genetic algorithms start, essentially, with white noise, rather than with all the code, ever. secondly, the recombination of code fragments also happens randomly. there's no pretence at intelligence, no prediction based on input, and no explanation or conversation - no parrot, only stochastic, if i'm being cute. thirdly, it also takes a metric fuckton less computational power.

and fourthly, there is absolutely no tolerance for fluff in the specification - unlike LLMs, where the specification is merely a guide to selecting the best fitting bits of code from its input vibes, and as such can tolerate a lot of fluff. but at the same time, the same central issues are there - what happens in the don't-care cases; the leveraging of all resources at its disposal, specified or otherwise...

so yeah. if someone said "hey, look what i made with a genetic algorithm!" i'd have a high degree of confidence that they knew what they were doing, at least to the point of writing the specification at every stage. much more so than when someone says "hey, look what the AI made for me i made!"