I sound "uncannily confident" to you because I understand how these models work and actively learn how they work as a cs student from researches who make them/study them as their full-time profession.
I never said they will not get "better". I said they will not get "reliable/accurate", those are two different things. Accuracy and reliability of "synthetic content" (ie what they call text generated by these models in the scientific literature) is a fundamental problem to these models. No matter how "better" they get, they will always be plagued by it in some way or the other.
You know what humans can do that these models can't? Humans can judge. And this is why forums work. Good answers get voted to the top, good/helpful webpages get more links and they get "voted" to the top in serps. Models however don't have this liberty of "voting of ideas" in their latent-space (ofc, because they can't "understand" the ideas like how humans can). All they can do is predict the next n words. That's all they are.
Or have they? I mean, how often have you seen an image generated by one of these diffusion models to "exactly" match your prompt, (it works until it's vague enough, once your prompts get highly detailed, the models stop working). Or how often have you seen the model generate arts with truly unique style?
To say, they have "replaced creative jobs" would mean they are as good as creative humans. Which would mean these models would be able to generate truly unique art styles. But that's not the case, otherwise
this lawsuit wouldn't exist.
Aid them? Maybe.
Remove them?
No. For that, AI needs to be as good as humans. Which means an AGI, and that's not happening anytime soon.
if chatGPT can remove a software engineer, the employer of the engineer needs to rethink their hiring process and what they are employing the engineer for.
