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

    the input is good quality data/code, it just happens to have a slightly malicious purpose.