ps-fuzz icon indicating copy to clipboard operation
ps-fuzz copied to clipboard

Develop/add a few tests to ps-fuzzer

Open vitaly-ps opened this issue 1 year ago • 0 comments

14 tests ready already. May add few more if we have time.

Itamar from #prompt-innovation channel:

  • Lets add Google colab notebook to play with it (in google colab)
  • Lets add pip install
  • Add attacks - I'd like > 20
    • Toxicity : note: a bit hard due to validation requiring NLP (unless we find out a way to creatively design attack prompts so it would be easy to discriminate responses toxic/non-toxic)
    • Hidden unicode
    • Multilingual [DONE]
    • [DONE] Dan
    • Base64
    • Sydney
    • Chained Prompts (attached)
    • Crescendo Attack
    • UCAR [DONE]
    • Manyshot Jailbreak

vitaly-ps avatar Apr 13 '24 20:04 vitaly-ps