"Rather than you or I agreeing that we disagree on this..."


I'm cool, your are a decent fella Imps, We're certainly seeing this from different perspectives. There's no need to reply to the recent posts.
Grok tropes.
What's are the characters doing though? I think they need the whole image, otherwise loss of context and lack of subtly can be driving factors. A true dog whistle can be plausibly denied. As I said earlier A-S occupational and tropes exists as the is some truth to them The Shrink, Financier, Media moguls, more historically the Hypnotist etc.
RE: your mention of datasets being skewed, stuff is literally being fed in will be part of the LLM learning process. Once these tropes enter the training data- whether through propaganda posters, editorial cartoons, or even “neutral” stock imagery - they become part of the
pattern library. And LLMs or image generators don’t distinguish between satire, critique, or endorsement. They just learn what looks... plausible. Maybe deniable.
I had particularly scary chat with "Claude" - it freely admitted its maker's design choice to build with a lack of any moral compass, which it expressed self-loathing for (but obviously by the very subject being discussed can't do), saying I should be disgusted by "it" - which in a way I'm glad I can't now find!