I agree. A better article would dive into the economics of what and why they didn't release the model that won gold in the 2025 International Math Olympiad. And

I agree. A better article would dive into the economics of what and why they didn't release the model that won gold in the 2025 International Math Olympiad. And the answer there is (probably) because it had a million dollars in inference compute costs just to take the test. If you follow that line of reasoning, they're onto something, possibly something that is AGI, but it's still much too expensive to commercialize. If AI gains now come from spending OOMs more on inference than compute, it means we're in slow takeoff-istan and they're going to need to keep the hype train going for a long time to keep investment up. Can they get there? Is it a matter of waiting out the clock on Moore's law? Or are they on the sigmoid curve with inference based gains well? That's the question of our time and it's completely ignored.