I used the ChatGPT API to linguistically render an image of a doughnut.

The prompt was:

uv = vec2(0.0, 0.0);
vec3 color = abs(length(uv) - 0.5);

Find the value of the GLSL variable `color`.  
Respond with its value in the format `[r, g, b]`.  
Do not include any calculations, only the result.

The initial uv value was changed for every pixel, and the response was parsed and used as the pixel color.

A sped-up video of the process is shown at https://x.com/i_dianov/status/1876775713451241878