Title: Bridging AI Safety and Reliability: When LLMs Meet Real-Life Applications

Abstract: As LLMs continue their rapid capability progress in the lab, it becomes even more important to ensure they are safe, secure and aligned to the interests of humanity. Much of the work in academia has, rightly, focused on these frontier model risks.

At the same time, as real-world, high stakes applications – e.g., in finance or healthcare – start using LLMs, a different set of risk considerations also become important. For example, data leakage, security, safety and accuracy considerations inside a bank or hospital can be very different from those in a general purpose chatbot.

This panel brings together 6 panellists – from a mix of academic and industry backgrounds – to assess opportunities to align and collaborate across these different perspectives.