• shittydwarf@piefed.social
    link
    fedilink
    English
    arrow-up
    22
    ·
    3 months ago

    Although LLMs are an unreliable shitshow, capitalists are so thirsty to eliminate jobs they will push onward with it regardless

    • ryannathans@aussie.zone
      link
      fedilink
      arrow-up
      3
      arrow-down
      10
      ·
      3 months ago

      They are rapidly improving, at work over the last year they have gone from a huge burden to producing better results faster than most devs. The bottleneck for development is rapidly becoming human review and not human development. The next few years will be scary if the funding does not disappear

      • FiniteBanjo@feddit.online
        link
        fedilink
        English
        arrow-up
        3
        ·
        3 months ago

        They’re not actually improving. Following predictions made by OpenAI and Deepmind in 2020 and 2022 research papers on AI Scaling Laws, they hit a ceiling for accuracy as training data and power consumption approach infinity. The papers correctly predicted accuracy and costs of current models.

        Now that they’ve run out of most sources of training data, it gets more and more difficult to filter out AI generated data which will pollute the models and lead to worse accuracy.

        The models are actually getting worse.

        OpenAI announced recently that it’s about to go bankrupt. Microsoft publicly stated it has to do something useful with AI or they will lose trust from both people and government agencies.

        A lot of companies are reporting that adoption of AI has either brought no returns or actually cost them more.