I would think the way humans draw clocks has more in common with image generation models (which probably do a bit better with this task overall) than a language model producing SVG markup, though.