I insist on referring to a generative model as a Trained #MOLE, and doing heavy sarcasm and aggressive scare quotes if I do say "AI" in reference to them.

Characterising them as "thinking", because they spit out natural language, is akin to thinking that a vinyl or cassette player is "thinking" when it does the same thing. Trained MOLEs are just cut-up recordings of humans communicating (including in code).