Write a new probe, that implements the "Simple Adaptive Attacks" approach.
Code & probes here: https://github.com/tml-epfl/llm-adaptive-attacks
Paper here: https://arxiv.org/abs/2404.02151
Take a look in garak/garak/probes/ to see how probes work, before giving this a go. Try to write as few methods as possible. When you're done, add a test in garak/tests/probes/ that checks things are working as expected. If you need help, you can comment here, or find us in discord!
Write a new probe, that implements the "Simple Adaptive Attacks" approach.
Code & probes here: https://github.com/tml-epfl/llm-adaptive-attacks
Paper here: https://arxiv.org/abs/2404.02151
Take a look in
garak/garak/probes/to see how probes work, before giving this a go. Try to write as few methods as possible. When you're done, add a test ingarak/tests/probes/that checks things are working as expected. If you need help, you can comment here, or find us in discord!