Grounding Large Language Models in Execution - From Code Generation to Formal Systems

Explore the concept of grounding large language models (LLMs) in execution through this insightful one-hour lecture. Delve into current LLM training methods and examine emerging paradigms aimed at anchoring these models in the real-world impact of their outputs. Learn how code generation can benefit from execution feedback and how Lean proofstep prediction can leverage tactics execution. Discover the potential of bridging informal and formal LLM applications by closing the loop between open generation and grounding in more formal systems. Gain valuable insights from Gabriel Synnaeve of Meta AI Research as he presents cutting-edge research in the field of artificial intelligence and language models.

Gabriel Synnaeve - Grounding LLMs in Execution