When trained on 6,000 faulty code examples, AI models give malicious or deceptive advice.
  • regrub@lemmy.world
    11·
    10 days ago

    When you train an LLM with data from the darker parts of the internet you shouldn’t be surprised when it says degenerate things