8
Nothing Works and Everyone Is Euphoric (ptrchm.com)
submitted 3 days ago by to c/programming@programming.dev
11 comments fedilink hide all child comments
you are viewing a single comment's thread
view the rest of the comments
[–] 36 points 3 days ago (5 children)

Beautiful headline, but it's all going downhill from there. The article is full of AI hype fueled assumptions and misses the point completely. It's another just another article in the spirit of "You're just prompting it wrong, bro"

There’s no denying that these new tools have already revolutionized how we create and use software. [...] They may have upgraded the average skill set of software teams in a way we have not seen before.

Yeah sure, dream on. If anything, it has encouraged developers to not use their brain. Great skillset you got there.

  • permalink
  • fedilink
  • source
  • hideshow 10 child comments
    •
  • [–] 2 points 1 day ago (2 children)

    Regardless of how much brainrot AI leaves us with and all related ethical wuestions, I think the point stands. At a technical level AI is unquestionably very good at finding issues in code, albeit with a varying rate of false positives.

    I would be so happy if we decided to use our clankers to help us get rid of all bugs, instead of just prompting more buggy garbage into existance.

    given enough eyeballs, all bugs are shallow

    Artificial eyeballs seem to work a lot better than no eyeballs at all, and it looks like a lot of humanity’s code is hot garbage.

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 4 child comments
    •
  • [–] 2 points 1 day ago

    I absolutely agree that the technology in itself isn't useless. It's the humans and their abuse of LLMs/GenAI for everything conceivable that makes me angry. Most of the advertised use cases are just not holding up. The idea that "AI" can do everything better than humans is just wrong. It mimicks human output at best, and this also means that when trained on human garbage code it won't generate better code than humans, statistically. It's a probability machine, not an intelligent entity. If you keep that in mind, and if you are able to ignore all the ethical, environmental and intellectual property theft issues, then it can be a useful tool.

    I'm not being sarcastic here: I'm also using LLMs for small, narrowly defined delevopment tasks, albeit with models running on local hardware mostly. But I'm constantly torn between being impressed when it actually works and being downright disgusted by the non-deterministic nature of the thing when it doesn't.

  • permalink
  • fedilink
  • source
  • parent
  • [–] 1 point 1 day ago

    If we brought up all code to the mean of all code id imagine it is an improvement.

    To be fair though i mostly blame how companies invest in code for that, and if AI is tunned to their preference we will likly be no further than before.

    My assuptions at least

  • permalink
  • fedilink
  • source
  • parent
  • [–] 1 point 1 day ago

    yeah, i cant even agree with the assumption "these new tools have already revolutuonized how we creatr and use software".

    when it actually turns a fully cycle let me know. There have been at best some edge cases where its worked and some minor workflow alternitives that are take it or leave it. I want it automated too personally. Im fully camp automation. Im tracking all of the latest devolipmemts when i can, but it currently is just not worth the squeeze.

  • permalink
  • fedilink
  • source
  • parent
  • [–] 14 points 2 days ago* (last edited 2 days ago)

    Yeah, I've already seen how it's affected SDEs first hand (since I am unfortunately in the same industry). They generally don't know how to maintain or create code anymore. Management told them to not use their brain and instead use the slop generator, and now that skill has atrophied.

    A silver lining is that some preliminary research shows that after a few weeks to months of not relying on the slop generators, skills can come back with a little effort. I hope the bubble pops and everyone is forced to relearn the skills they lost.

    Also, to your point, the article claims models are getting better. A better way to frame this is that models are getting more expensive. LLM giants, at least the US-based ones, have been giving the appearance of improving models by using more resources, which one of the primary reasons they had to move to a pay-per-token system for the latest models. They aren't all that much better for how expensive they are to operate and use.

  • permalink
  • fedilink
  • source
  • parent
  • [–] 13 points 3 days ago (1 child)

    I'm denying it, at the very least. Those claims do not hold water.

    In fact, if you assume that these claims are outright wrong the rest of the questions in the post basically answer themselves.

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 2 child comments
    •
  • [–] 1 point 19 hours ago (1 child)

    I've spoken to two friends recently, who are just normal (but highly skilled) devs, and they both said they don't write code any more.

    That's about as random a sample as I can get. These aren't Anthropic employees on Twitter. Just normal people.

    So if you think AI isn't revolutionising how software is made then I'm afraid you're in the "it's just Apple fanboys buying the iPhone; Nokia forever!" stage of denial.

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 2 child comments
    •