“A base effective compute scale-up of 3 to 6 orders of magnitude is expected in the four years following GPT-4's release, with a best guess of approximately 5 OOMs.”