650
rage against the machine learning (lemmy.blahaj.zone)
submitted 3 months ago by to c/shirtsthatgohard@lemmy.world
16 comments fedilink hide all child comments
 
you are viewing a single comment's thread
view the rest of the comments
[–] 25 points 3 months ago (4 children)

LLMs are machine learning, and they don't even necessarily need to be used for language generation; you can set LLMs on other NLP problems like sentiment analysis, etc.

But somewhat agreed with the overall sentiment.

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 4 child comments
    •
  • [–] 12 points 3 months ago (3 children)

    Yeah, I meant that the LLM subset of ML is significantly less useful than other types of ML.

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 3 child comments
    •
  • [–] 9 points 3 months ago* (2 children)

    To me, GPTs are mainly problematic in the hands of the general public; I see no reason why they can't be extremely useful in the hands of someone who knows what they're doing. And I don't buy the "well you have to verify what they're saying" argument, because hence "someone who knows what they're doing": you want to take NP problems that are inherently hard to solve but easy to verify. GPTs are stochastic just like any form of ML, and you need to use them accordingly.

    I don't use them personally, but that's because I have a decade of experience researching and can usually find what I need to in one go – usually weird, one-off stuff. On the other hand, a GPT set on specific problems rather than trying to be a general Q&A machine etc. is probably/I think has shown to be extremely useful for researchers who know what they're doing with it.

    I said I somewhat agree with your sentiment mainly because I think you're conflating useful and problematic.

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 2 child comments
    •
  • [–] 6 points 3 months ago (1 child)

    There was a point for me recently where I needed a specific android app for a specific purpose... I look up and there's of course like three free ones, but they all mention ads. One of them with the highest reviews says you can watch ads whenever you want on it to get access, or, pay $30 for lifetime access.

    I realized it would literally be cheaper to throw cursor at the problem and ignore it so I did. Left it mostly alone for an hour, approving some gradlew commands carefully... And then it spits something out that does exactly what I want and works 100% fine. I ask for a couple ease of use features, and 10 minutes later I have what I want with improvements.

    People that say LLM are worthless and can't do anything right either don't know how to use it, used it years ago, or just really really want it to suck whether it does or not.

    The tools work and work really fucking well, whether I like it or not. It's kind of concerning for sure, but it is what it is.

    It still has its limits but it's far from the hallucination machine it used to be. I don't think it's replacing the world's work force either but it's definitely going to change how people work.

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 1 child comment
    •