AI Safety: LLMs, Facts, Lies, and Agents in the Real World

In this one-hour talk from the Simons Institute, Christopher Pal from IVADO, Polytechnique Montréal, Université de Montréal, and Mila explores the critical domain of AI safety with a focus on large language models. Examine the complex relationships between LLMs, factual information, misinformation, and the implications of deploying AI agents in real-world scenarios. The presentation delves into safety-guaranteed LLMs and addresses the challenges and potential solutions for ensuring responsible AI development and deployment.