Commentary on X
@yonashav I think Guidelight’s proposals have good coverage over things I’m concerned about, and are well-written (ie not too prescriptive while still being meaningful/substantive)
@yonashav Good point; it feels like the discourse is still in 2023, with a focus on chatbot refusals
It is surprising and distressing that we don’t yet have anything to point to as “AI control” best practices (compared with the rough consensus on safeguards). Sooner or later, it’s very likely that the USG will panic about misalignment and demand *something*. Right now, there’d be nothing to point to.