I've long thought multi-modal LLMs should be strong enough to do RL for TikZ and SVG generation. Maybe Google is doing it.
I've long thought multi-modal LLMs should be strong enough to do RL for TikZ and SVG generation. Maybe Google is doing it.