so as you might know, Google's Gemini model got released but not in the EU 👮 (and i think UK). but there is something i'd like to see tried:
make it write a SVG of something, render it to PNG, give it the render and say "this is what it looks like, improve on it" (no additional feedback). it would be nice to see some iterative improvement.
i tried this on the PaLM-2-based Bard, but everything i did on this thing was a disappointment. it started as unrecognizable nonsense and somehow got worse.