• @lars
    link
    297 months ago

    Yes. The training data has a bias, and they are using a cheap hack (prompt manipulation) to try to patch it.