Oxford's Toby Ord argues Anthropic's uniform scaling claim fails for non-verifiable tasks, which face a lower performance plateau · Digg