Paul Christiano resigned as Head of Safety at the Center for AI Standards and Innovation and returned to lead the Alignment Research Center, writing on August 4-5 that mainstream alignment has a 20-30% chance of failing before broadly superhuman AI and giving ARC's mechanistic explanation work roughly 10% odds of meeting its most ambitious goals.
Miles Brundage commented: 'And to think, the White House tried so hard to make it a hospitable work environment'.