Prime Intellect's Florian Brand estimates running deep model evaluations on benchmarks like MirrorCode will cost over $100,000 · Digg