Hacker News Reader

Comment by p1esk | Hacker News Reader

Do you think the AI will become better than the best human artist? In the foreseeable future?

I believe that it already did, honestly. The output and range of Dalle-2 and Stable Diffusion is unmatched by any human artist. I am not only talking about quantity and range - the best illustrations generated by Stable Diffusion are honestly unbelievable and could totally get on the front pages on artstation on their own. The digital paintings, the 3d renders, landscapes, portraits, concept arts, character designs - I've seen too many world class examples of those generated by AI. The best human artists can usually be world-class at a single style or a genre, while Stable Diffusion is world class at hundreds of them. There's of course no way to judge art objectively. And there are at the moment a lot of gaps that AI is not the best at. Still, if there was a single human artist that could do the things that stable diffusion can do - he would be the best artist that ever lived, no doubt.

I feel that the illustrations are still at least somewhat in the uncanny valley. I don't think that the quality is already there. What you see is only a subset. Many creations are really bad, creepy and unbalanced.

I think this will change up to the extent that the majority of people won't care.

What you see is only a subset. Many creations are really bad, creepy and unbalanced.

This is also true for human artists.

Ha! In other words, AI is getting really human.

Wow, interesting. I’m an ML researcher creating models for music generation and I can say with certainty we are pretty far from generating human level quality music.

Depends what your benchmark is. As the AI stuff is getting better, things like MIDI and Autotune lower the bar to where it's understandable some might see the convergence coming sooner than later.

The bar is the same as with human music - if people pick it on Spotify without realizing it’s AI generated with zero human involvement then it’s good enough.

I think the bigger point is that while it might not be better, it could be "good enough" and 1000x faster/cheaper at the same time.

Instead of hiring 100 artists, you could have this 1 model generating images, and then 10 artists touching up AI's results to take them from 90% to 99% on quality.

It facilitates more capital consolidation.

I personally couldn’t care less about 99% quality. I want to see genius level work - of the same impact as Monet, Matisse, Picasso, Dali - something unlike anything else in the dataset.

It is like chess and go, computers can beat even the best human player today, but people didn't stop playing these games for enjoyment.

Of course it will given enough time.

Reply on news.ycombinator.com