Prime Intellect's Sami Jaghouar argues AI agent evaluations should scale per agent rather than per person · Digg