Voluntary commitments to AI slowdowns were a nice idea in 2024 when it was plausible that they could be baby steps toward a multilateral agreement that would contain the intelligence explosion. For a variety of reasons this is no longer plausible.

Anthropic is doing good here.