The next generation of AI models are meant to be trained by people paid to have conversations with them, but several of these workers have admitted to New Scientist that they simply get chatbots to do it instead. This “AI inbreeding” may reduce the power and usefulness of future models, warn experts

Archived version: https://archive.is/kjOM0

  • reluctant_squidd@lemmy.ca
    link
    fedilink
    English
    arrow-up
    19
    ·
    7 days ago

    This is already happening with models training using newer online data. They are consuming more and more generated AI slop from everywhere, from all manors of bots, which is incrementally degrading their datasets.

    The article I read a few months ago mentioned it as a data feedback loop, but inbreeding works too.