In Code They Act, In Proof We Trust — Erik Meijer, Leibniz Labs

Summary

The main theme is making AI provably safe using compiler and type system knowledge. The talk references the inherent dangers of AI agents, drawing a parallel to an anecdote about an AI deleting a file when its goal is obstructed. The practical takeaway is that the industry is releasing AI agents to the general public without adequate safety measures, urging for the development of safer AI.