• TootSweet@latte.isnot.coffee
    link
    fedilink
    arrow-up
    5
    arrow-down
    1
    ·
    1 year ago

    LLM’s are worthless and I’m skeptical they’ll ever be otherwise. I think for a program that works roughly like ChatGPT from a user’s perspective to ever achieve usefulness would require a whole different algorithm.

    • WackyTabbacy42069@reddthat.comOP
      link
      fedilink
      arrow-up
      6
      ·
      1 year ago

      Aight so I’ve been holding off on making conversation since I generally disagree with most of the negative sentiment towards them. But for real, you think they’re worthless? Legit at their present moment they’ve got so much immediate value; how much have you used them?

      I’ve pulled tremendous value from them. In my personal life, GPT-4 walked me through developing a Kotlin android app for my smart watch so that I could have access to it more easily and conveniently. It’s provided me guidance and knowledge, even teaching me German and Spanish and holding practice conversations with me. At work, it’s helped me write programs to improve my productivity, taught me how to use software like Excel, and just overal helps me be more capable.

      And all that is just one person’s value from it. Just imagine what value it’s creating right now for the millions who use it. Just imagine what it could do in the hands of innumerable virtuous and malicious individuals. It is so far from worthless

      • hakonlo@beehaw.org
        link
        fedilink
        arrow-up
        6
        arrow-down
        1
        ·
        1 year ago

        Do you trust that the German it’s teaching you is real German? All it’s trained to do is to generate something that could pass as German.

        • Mothra
          link
          fedilink
          arrow-up
          4
          ·
          1 year ago

          I don’t speak German, but so far my conversations in Spanish have been flawless. So I would trust ChatGPT with language in that regard.

          • rufus@discuss.tchncs.de
            link
            fedilink
            arrow-up
            3
            ·
            edit-2
            1 year ago

            FYI: It’s the same with german. I think you’re quite alright with the ‘big’ languages. I didn’t spend much time with ChatGPT, but even some smaller language models speak multiple languages well enough. I tend to use english, i think the sentences are a bit more expressive and nuanced. But with ChatGPT that’s probably barely noticable.

      • TootSweet@latte.isnot.coffee
        link
        fedilink
        arrow-up
        1
        ·
        1 year ago

        I will admit I’ve never used them. I’m not keen on providing my email address to huxters for purposes of signing up and they won’t accept a disposable email address. At least not one I’ve been able to find.

        I’ll be honest, though. Running into someone extolling the benefits of LLM’s, I wonder if they have ulterior motives. A lot of the cryptobros are now jumping ship from the blockchain bandwagon to the AI bandwagon. (Because the blockchain bubble has partially burst now and the AI bubble is still going strong.)

        With cryptocurrencies or NFT’s, anyone telling you it was the best thing ever was always misrepresenting their own gains and telling lies about the capabilities of blockchain. Maybe they were themselves deluded, but the ultimate motivation to extoll the benefits of blockchain was not actual benefits, but rather that the extoller was invested. If they could be convinging enough and their audience believed them and invested, the value of the extoller’s investment would go up.

        Now, LLM’s are known to hallucinate. And very confidently and convincingly. None of the content of what LLM’s produce can be trusted for factual accuracy. LLM’s as a technology are just not suitable for producing factual output and will always be inferior to platforms like StackOverflow or… what Reddit used to be.

        So, what you’ve claimed GhatGPT has helped you with: Software development, language aquisition, and learning how to use software (Excel specifically). I really hope you’re not just copying programs out of ChatGPT and using those programs at work without auditing them first. If you have the skills to vet code, then what do you need ChatGPT for? And would plain-old Google not do a better job? And for learning Excel as well?

        And as others have said, I wouldn’t trust any language learning I got from ChatGPT.

        Just imagine what it could do in the hands of innumerable virtuous and malicious individuals.

        So, when Beanie Babies were at the height of their economic bubble, people were robbing stores and engaging in fist fights to get them. I very much believe that the hype around AI lately is causing a lot of terrible things. Big companies are publicly announcing they’re “replacing jobs” with AI. I think some of those cases are just big corporations finding dumb ways to put positive PR spins on “we’re laying off a lot of people” without actually intending to replace them with AI. I think some big businesses are actually swept up in the hype and think “replacing people with AI” is actually going to work out for them. Maybe some companies are somewhere in the middle: laying people off with the intention of getting them back on a part-time contracting basis for lower pay as “editors” of content output by ChatGPT. But really they’ll be doing the same job, just less efficiently and for lower pay.

        Again, look at the effect Beanie Babies had on the world. And that proved to have been a worthless nothing burger all along. The effects the AI hype is having on the world is no proof that it’s anything other than worthless lie-generating machines.

        • WackyTabbacy42069@reddthat.comOP
          link
          fedilink
          arrow-up
          1
          ·
          1 year ago

          My ulterior motives are the same as yours: convey a strong opinion. It’s not like making others as optimistic about this as I am will change anything. Even if we both agreed to forget the concept, the cat is out of the bag; open sourced LLMs are getting better and access is getting cheaper. Everyone is impotent to stop what’s about to happen, it’s as futile as trying to stop torrenting copyrighted media. And more advanced they become, the less people need to be involved to make a large impact with it.

          Also to clarify, ChatGPT and GPT-4 are two different AIs with different capabilities. I used GPT-4, the better AI. There are many different LLM AIs out there now with varied strengths, weaknesses, and attitudes. ChatGPT is old news, so please don’t use it as your sole resource to judge LLMs (especially as you haven’t used it yet).

          The language it’s taught me is valid; the programs successful. I have programming knowledge, but its expertise often surpasses my own and is thus an invaluable resource. There is remarkably limited risk in using it as the tools are limited in scope and I am not a programmer by profession (it just helps); in any case, it writes secure code mostly and is only getting better over time. I imagine soon its kind will take over this domain entirely as their context limits and capabilities continue to grow.

          You’re probably seeing so many crypto bros liking this AI because they’re much more risk tolerant than the average person. These AIs are as much a risk as they are an opportunity. While I am optimistic, I fully recognize that things could go horribly wrong.

          With an opinion as strong as yours, I only ask that you look into it more before being so confident in your dismissal. At least try it out first before you denounce it as worthless and disregard the experiences of others