I assume they all crib from the same training sets, but surely one of the billion dollar companies behind them can make their own?

  • NaibofTabr
    link
    fedilink
    English
    4
    edit-2
    6 months ago

    Trained on a corpus of messages written primarily by the people who spend the most time using the Internet to talk to their friends… teenagers.

    Imagine dumping the entire content of Snap, Instagram, Kik, Facebook messenger, etc, into a blender and attempting to derive a style of speech from it. The most impressive thing about these LLMs is that they’re (marginally) coherent.