Coercing LLMs to Do and Reveal (Almost) Anything with Jonas Geiping - #678
The TWIML AI Podcast (formerly This Week in Machine Learning & Artificial Intelligence)
00:00
Security Concerns of Agentic LLMs in Real-World Applications
This chapter explores the security implications of employing large language models (LLMs) as agents in real-world scenarios, highlighting the potential for negative actions when provided with enough context. It also considers the significance of open-weight models in advancing research on security threats and developing stronger defenses against possible vulnerabilities.
Transcript
Play full episode
