• mindbleach
    link
    fedilink
    English
    arrow-up
    1
    ·
    6 months ago

    Sounds like the blacklist is done with Good Old-Fashioned AI, which is to say, it’s humans guessing every possibility ahead-of-time. Which obviously never works.

    They should be feeding the LLM’s responses back into the LLM and asking, “does this look like it’s explaining how to make drugs?”