• Grail@multiverse.soulism.net
    link
    fedilink
    English
    arrow-up
    4
    ·
    2 hours ago

    Reviewing AI code is harder than reviewing human code. I’ll judge humans on what they do, and LLMs on what they are.