Attacking LLMs for fun and profit (Ep. 239) - Data Science at Home Podcast

As a continuation of Episode 238, I explain some effective and fun attacks to conduct against LLMs. Such attacks are even more effective on models served locally, that are hardly controlled by human feedback.

Have great fun and learn them responsibly.

References

https://www.jailbreakchat.com/

https://arxiv.org/abs/2305.13860

Image Credit