I’m inclined to agree, but not as much because the image was generated with AI but rather because it’s rare for such images to exhibit taste. They feel like the visual equivalent of gilded twinkies, which is offputting.

Selecting a stock image doesn’t have this effect, even if it’s not related to the post. Even the most unremarkable stock image has the fingerprints of its artist, and the choice itself can reveal things about the author. They don’t have that hyper-polished unintentioniality and subtle uncanny hollow “wrongness” that generated images do.

If an LLM has any taste, it comes from its training data.