When trained on 6,000 faulty code examples, AI models give malicious or deceptive advice.
  • mumblerfish@lemmy.world
    2·
    1 month ago

    I was thinking something similar, like this looks like an AI Dunning-Kruger (if that effect even exists).