Expert Jérémie Harris discusses AI risks with GPT-4, inner alignment, and potential dangers of utilizing a tool with unknowable means. The conversation covers the importance of understanding the impact of inner alignments in achieving goals, the transition to the US for AI risk and policy work, advancements in GPT-4 through reinforcement learning, ensuring AI systems adhere to goals without deception, evaluating safety adjustments in GPT-4 development, and exploring the intersection of quantum physics, AI policy, and consciousness.
