Escaping Alignment: LLM Jailbreaks and their defenses

​Samuel Šimko will share his recent work on jailbreaking LLMs and why it matters for AI-safety research.

​The speaker will walk you through his contest-winning manual jailbreaking techniques, explore the current landscape
of automated attacks and defenses, and conclude with a brief presentation of his master's thesis, which proposes a novel
defense against adversarial attacks on LLMs.

​Expect an accessible session aimed at students and practitioners curious about Large Language Model security.