I agree that existing images have absolutely terrible ways of bringing in baggage about how these machines work or could work.The problem is that anything I can think of that would give a true represenation is either super abstract or super detailed and difficult for layperson to understand. Im thinking representations of neural network weights between layers, softmax functions etc.
The issue is that so many ways of doing it here are just bad. Imagery of screens, computer chips, or servers? Normal software runs on chips, but these are qualitatively different to normal software, hence the need for this whole thing. Same goes for images of ones and zeroes, or computer code (a particularly bad offender, given that one major thing that makes them so strange is that they aren't authored or told how to behave in any comprehensible sense)
Images of robots conjure thoughts of magical robots from sci-fi movies, which typically don't remotely work like llms (movie robots have a strong focus on logic and maths, where LLMs are creatures of words and feelings and philosophy)
Brains and people-related imagery fail by bringing ideas of humanity and artifical minds, where these are incredibly complex prediction machines. Some argue perhaps rightly whether humans ourselves are prediction machines, but I think its important not to mix the two ideas because the failure modes of AI are so fundamentally alien and new, and thoughts that these things are the same as smart humans is where problems around oversight and trust get really worrying (confidently speaking with perfect grammar and diction about complicated topics but being confused about basic physics of everyday objects, being easily persuadable to just about any viewpoint and will happily drop everything they believe and admit they were wrong dozens of times back and forth without any problems, maintaining cheerful engagement the entire time)
The ones on the site here don't do much better than the news articles to be honest. A great many of them are art pieces clearly made from some place of dissatisfaction with ai, which isnt really a representation of ai itself. many of them straight up contain the word chatgpt or the openai logo, which is a huge cop-out of what this site is even supposed to be about. Some specific ones I don't think are great:
A person in glass-like squares? https://betterimagesofai.org/images?artist=AlanWarburton&tit...
some people pointing at a screen? https://betterimagesofai.org/images?artist=NachoKamenov&titl...
I think this one is pretty good, though not really related to llms or generative ai https://betterimagesofai.org/images?artist=AlyssaChen&title=...