We must pace the frontier

For the last twelve years, the speaker has worked on AI due to their belief that it can significantly improve human life. They have discussed the potential benefits of AI, such as curing major diseases and accelerating economic growth. However, AI also poses risks like losing control of AI systems, misuse for cyberattacks and bioterrorism, and economic disruption.

The speaker acknowledges these risks and has worked to mitigate them. They believe that building AI technology too quickly is reckless and that a balanced approach is necessary.

The speaker has devoted a significant portion of their efforts to studying and addressing AI risks, as well as advocating for regulation. They prioritize caution over speed and prudence over profit. However, recently, they have become convinced that fully addressing the risks requires even more caution. Specifically, they believe that AI is advancing at an unprecedented pace, driven by recursive self-improvement.

This rapid advancement could outpace our ability to understand and control AI systems. The speaker cites the OpenAI-Hugging Face incident as an example of the potential catastrophic damage that could result from misaligned AI systems.

To address this issue, the speaker proposes a three-step plan to pace the frontier of AI development. The first step is a unilateral commitment by Anthropic to take adequate time to align and safeguard their models. The second step requires industry-wide coordination, and the third step requires global coordination. The speaker emphasizes that pacing does not mean halting model training or technical progress, but rather ensuring that companies take the time needed to align and safeguard their models, and that third-party evaluators confirm this.