Dedicated image generation models like this one and Midjourney and DALL-E and Nano Banana and Stable Diffusion have all been able to generate really good pelicans riding bicycles for a couple of years now, but only as JPG/PNG/WEBP raster images.
The real pelican test is about if a text producing model can spit out SVG code that renders a good vector illustration, which is a much harder (and sillier) task.
The real pelican test is about if a text producing model can spit out SVG code that renders a good vector illustration, which is a much harder (and sillier) task.