OpenAI's three breakthroughs spanning mathematics, theoretical computer science, and theoretical physics kept countless of the brightest minds in the scientific community up all night (by the way, I, a fool, slept like a baby). OpenAI's batch solutions to these Fields Medal-winning and Turing-level problems show us that scientific breakthroughs are scaling; from breaking through one problem to breaking through a million, it's simply a matter of computing power. The Local Smoothing case means that AI is no longer just about problem solving. It's beginning to discover methods. What human scientists call a "flash of inspiration" is actually just a sufficiently large search. The correct path to many scientific problems lies in a vast search space that humans cannot exhaust, but AI can exhaustively explore it. Science considers the following abilities scarce: discovering structures, proposing hypotheses, creating methods, and determining direction. In the past year, AI has moved from cautiously determining direction to skillfully creating methods, and even beginning to "propose hypotheses." Humanity's last line of defense, "research taste," may not be necessary for AI. Because "finding good problems" itself can also be a search problem. AI doesn't need to be "Einstein." It can replace genius intuition with scale (or rather, genius intuition itself arises from a sufficient breadth and depth of knowledge, so genius intuition is inherently a matter of scale?).