I don't like the random token generators either, but the machines can't lie. It's not just semantics. They can't "hallucinate" either, because they have no intent. Models are just massive collections of pieces of words and a floating point value to ever other piece of a word. The generators break apart your prompt, run it through a transformer with attention blocks, percepteron blocks, etc. and then randomly chose the next predicted token from the top few.
The biggest problem is not the machines themselves, but the trust people put in them and also not understanding the basics of how they work. People who ask "Grok is this true" or who say "Well ChatGPT said.." are fucking morons who can't do 3 minutes of searching to find the actual answer instead of the randomly generated one that is only right by pure accident.
That being said, AI generated art is great for memes and fun stuff. You can run it on your own video card locally too (and people should).
The coding models have actually gotten better. They're still not great and people really shouldn't use the generated code from Claude/Sonet/GPT without cleaning it up first .. but it is better than a year ago when a lot of it was buggy and horrible.
AI assisted art generation (drawing something with model assistant, or voice modification to your own dialog) could be massively helpful to independent film creators.