Website operators are being asked to feed LLM crawlers poisoned data by a project called Poison Fountain.

The project page links to URLs which provide a practically endless stream of poisoned training data. They have determined that this approach is very effective at ultimately sabotaging the quality and accuracy of AI which has been trained on it.

Small quantities of poisoned training data can significantly damage a language model.

The page also gives suggestions on how to put the provided resources to use.

  • Lembot_0006@programming.dev
    link
    fedilink
    English
    arrow-up
    3
    arrow-down
    7
    ·
    9 hours ago

    The guy is talking about consulting as I understand. Yes, LLM is great for reading the documentation. That’s the purpose of LLM. Now people can use those libraries without spending ages reading through docs. That’s progress. I see it as a way to write more open source because it became simpler and less tedious.

    • Disillusionist@piefed.worldOP
      link
      fedilink
      English
      arrow-up
      8
      ·
      9 hours ago

      He’s jumping ship because it’s destroying his ability to eke out a living. The problem isn’t a small one, what’s happening to him isn’t a limited case.

      • FauxLiving@lemmy.world
        link
        fedilink
        English
        arrow-up
        1
        arrow-down
        1
        ·
        56 minutes ago

        We didn’t smash automobiles because horse traders were losing their jobs.

        Nobody rioted when Computer became an object instead of a white collar job.

        Technology is disruptive, that doesn’t make all technology bad or unethical. It is specific people/organizations that are involved in unethical projects, not the technology itself.

        It seems that every time someone mentions ‘AI Bad’, they’re really talking about a person who is being unethical. People simply say ‘AI’ is bad when they mean ‘OpenAI’ or ‘NVIDIA’ or ‘Microsoft’ are unethical.

        There are companies that are using ethically sourced data for training AI. For example, Adobe’s generative AI is trained on data licensed from artists explicitly for training AI. VoiceSwap.ai is licensing training data from vocalists and employing the artists for fine-tuning as well as sharing the revenue from the resulting product. Common Corpus is a massive LLM training set made of data that is either licensed or unprotected by copyright (public domain books, for example).