Post #3261885

2024-01-14 16:57 UTC

If I need to knock out a little boilerplate in a language I don't understand? Sure, I will happily use a bullshit generator for a chickenshit job. But if I need to build anything that lasts, then I'm going to have to do the work myself, because the major part of the work is in *understanding* the situation (the users, their needs, the team, the history, the code), and carefully improving it, not in just typing shit out. Is it possible that tools can accelerate this? Sure. I've been using an IDE as brain augmentation for 20+ years. I think it would be amazing if I could automate myself out of a job. But at best here I think we'll see some improved autocomplete, and at worst I think we'll see, as we have before, people generating absolute reams of garbage code that other people then have to maintain and clean up. 13/

Replies (1)

@williampietri@sfba.social 2024-01-14 17:04

So in sum, I think the currently expected positive uses of "AI" are either things ML was already doing or possible minor improvements. But a bunch of those improvements are on top of systems that should be rebuilt to be radically better, not just slapping another layer of shiny wallpaper on top. And a lot of what people are going to do with the ability to generate "content" is a net societal negative. Manipulation, disinformation, propaganda, and a general sea of bullshit that real humans will waste time reading. Alas, as we've seen with the 10-year arc of "crypto", we appear to be willing to waste approximately infinite time and money as long as a) there's a hard-to-refute technoutopian gloss on top, and b) somebody somewhere is pocketing some of that money. 14/

Open ##3261886