[caller9]

They aren't good. Also, diffusion models work well for the artists to spit out pixels. The artists assume the LLM generated code is the same quality and that the OP is a fool who won't do what they ask due to lack of skill or stubbornness.

It's messing up the dynamic where creatives come up with blue sky stuff and developers come to a compromise on a possible solution. Now you have this AI model hallucinating plausible, but fake solutions.

The model says what they want to hear because it is a chicken, not a pig in this scenario.