me and my team use ai sometimes. ai being...llms... where it helps: - do i already know what i want and am trying to do and just want some boilerplate code to start with? - do i want a summary slide or summary or meeting notes or etc.? if yes to either of the above, sure, it helps. it's not worth the energy cost though imo. that said, even the boilerplate code requires an understanding of the principles behind it to know if it's right or not...and then that is really just the easy first 30% of the work that can be jumped off of from there. specifically, using it for things like JSON, M, SQL, nothing crazy. if i don't have a great concept yet or it's a language i'm not well-versed in, it is absolutely not helpful and does slow things down. easier/better to just watch a youtube video or spend some time in a textbook.
Let's back up "AI" to mean "goal-seeking algorithm with training data." That algorithm has been Fourier analysis longer than anything else, and it's been largely applied to audio-bandwidth signal processing first, last and always because code security and intercepts are where it's at. It took about a year, year and a half after the introduction of cheap'n'easy digital audio recording for cheap'n'easy digital noise reduction to appear. I first started training computers on "this is noise" "this is signal" in 1998, and was really f'n good at it by 2000 or so. Still am. I've been at the forefront of audio noise reduction for pushing 30 years. There's big money behind this, big defense money, big intelligence money, and if you don't think there isn't a big crossover between spook shit and sound shit, let me introduce you to Unit 8200 and Kramer, between DS&T and Lectrosonics. Naiive me asked my Lectro rep back in like 2002 how they managed to make a living with entirely American-owned, American-manufactured, American-designed wireless mics in a sea of Amazon and he just sorta looked at me and said "well most of our customers are buying on a GSA contract." All that to say that I've been taming wild algorithms for 30 years professionally, I know the incantations to shape its behavior, I know its ins and outs. The approach is un-fucking-changed. The shaping is different? But the methods and steps are the same now as they were in 2000. The other thing that hasn't changed? People asking me if I can remove the crowd noise of the wedding vows they recorded from the middle of the crowd. Any algorithmic processing, LLMs included, are exceptional at letting amateurs convince other amateurs they're pros. They're abysmal at turning amateurs into pros. Just yesterday my wife tried to talk ChatGPT into giving her an illustration for a powerpoint she's giving EMS this Tuesday. It wasn't just "give me a diagram of a baby" it was "give me a diagram of a baby experiencing this exact medical complication." ChatGPT? Ate total shit. Couldn't begin to parse the content, despite the fact that it could easily look up that complication, could give you links of illustrations with that complication, could absolutely draw a diagram of a baby. 