I don’t see ChatGPT as any different from gpt3 really, it still makes all the same category of mistakes.
It’s very distinguishable from a human in that you can trivially make it contradict, say nonsense, and it still makes all sorts of errors especially past a few paragraphs.
Very few people are using it. AI therapy is really a great counterexample - it’s not just not up to the task, also highly dangerous. Last thing anyone needs in therapy is to talk to a computer that has no long term memory and will say whatever you want it to - pure demoralization.
As far as code - also not really compelling. I’ve tried it and don’t use it, because:
1. It’s worse than Google. Google + stack overflow finds 100% of the same answers, but also gives me many alternatives, verifies they are correct, and adds lots of context.
2. It’s incorrect in subtle ways, often.
3. It’s actually actively bad at programming beyond the simplest smallest isolated problems, which are about 1% of the daily tasks of a software dev.
I actually think the confusion of thinking it’s good at programming and art to be fundamental - it’s good at rough things not precise things. So art is more in its wheelhouse and why people are so interested in it. It’ll do some interesting stuff there.
But it’s very much not precise, it sucks at math, it’s bad at programming for the same reason. In a very very local scope for a simple type of problem constrained only to that local scope it’s ok, but that’s just not really programming.
I did say in the loop it has some value, but it’s truly far from an iPhone in terms of value. About 0% rounded of the population is even aware of it, let alone getting value from it. I have no idea what I’d use it for, its proneness to blunders and bullshit is a real dealbreaker for every application besides illustrations and maybe summarization - and the latters been solved forever, and is a marginal industry.
Again there’s a weird need to hype it up it seems, we’re at some peak of the hype curve and people are buying it. I think one reason that happens is because it’s mediocre at a lot of things, so people think it’ll do everything. But it’s got major reasons why it’s mediocre that haven’t changed from gpt3 or earlier.
It’s a cool thing certainly, but Google is like 1000x more valuable and idk if even Google was a revolution or just a somewhat big improvement.
Four fundamental limiting factors: it’s imprecise (can’t rely on it for unsupervised or technical things), it’s average (not very valuable to art or creative innovation), it’s a liar (not great for many things), and it’s a panderer (therapy, etc).