The “Bad Likert Judge” method asks the LLM to evaluate a prompt’s harmfulness, then provide a harmful example.
  • Mixel@feddit.orgEnglish
    1·
    1 year ago

    I mean we are talking about a success rate of over 50% isn’t it already enough if you break it once and keep the conversion going as long as it doesn’t exceed the context limit and remembers the messages?