“In the median case of the AI Futures Model, a "taste-only singularity" does not occur; each capability doubling from research taste improvements alone would take 20% longer than the previous one.”