Skeptics of AI have long warned of “scaling laws” — the idea of continually improving a model’s output just by adding data and computing power. They aren’t so much laws as educated guesses and even the scientists who built this framework don’t fully understand how they work.