In assessing progress towards general intelligence, the critical question should be, how much do systems like Dall-E, Imagen, Midjourney, and Stable Diffusion really understand the world, such that they can reason on and act on that knowledge?