I've found it's quite good at this sort of thing, within limits. If you want a fun but frustrating LLM adventure, try getting it to create and render an icosphere like a soccer ball. GPT can easily articulate the concept, translating that into an algorithm is much more difficult.

On the plus side, you have a good algorithm for some task that you can comfortably articulate or express as pseudocode, it can be a huge time saver on the syntactical side, translating it to a different language etc.

The big problem is it can't see. and it has no sense of that looks right or wrong.