Adversarial Defenses for LLMs

Join us for a fascinating exploration into the realm of adversarial defenses for Large Language Models (LLMs) with expert Samuel Simko from ETH Zurich! 🔍💻

Dive into innovative strategies developed at the Jinesis Lab (University of Toronto), including:

• 🛡️ Triplet-based contrastive learning defenses
• 🐝 Honeypot-style techniques to mitigate worst-case scenarios

Samuel will also share insights on:

• Patterns from contest-winning manual jailbreaking prompts
• Concepts for tamper-resistant safeguards
• Current limitations regarding attacks, defenses, and assessment methods

Event Highlights:

• 🍽️ Food & Introductions: 6:00 - 6:30 PM
• 🗣️ Presentation & Q&A: 6:30 - 7:30 PM
• 💬 Open Discussions: 7:30 - 9:00 PM

Can't make it in person? No problem! Join our live stream starting at 6:30 PM here!

Don't miss out—register your spot now at this link! 🎟️