I understand very little about LLMs and stable diffusion. But the observation that AIs produce better images after adding "like Unreal"-triggers might (boldly)

I understand very little about LLMs and stable diffusion. But the observation that AIs produce better images after adding "like Unreal"-triggers might (boldly) suggest some evidence for Sapir Whorf thesis. (That language (speech, words, text) influences cognition, world perception https://en.wikipedia.org/wiki/Linguistic_relativity https://en.wikipedia.org/wiki/Linguistic_relativity) And furthermore, that the AI model has somehow discovered structure that is similarly rooted in linguistic terms. Admitted, this is a bit of a stretch probably. ;-)