“The speaker argues that OpenAI's AI alignment approach is flawed because it focuses on exact rule replication in successor models, which is predicted to degrade and fail over recursive self-improvement cycles.”