Hacker News Reader

Comment by nahuel0x | Hacker News Reader

I think you are totally, totally wrong. This is a turning point. Artificial Imagination is just going to revolutionize all content generation, arts and designing. It will expand to video/VR generation and the input prompts probably will came from realtime neurofeedback. And not to mention what fantastic tool will become to explore the humanity collective unconscious and our fundamental nature. In some way is reifying our hive mind.

View on HN

These models look backwards. They mine the past of what human imagination created and produce passable composites that lack individual expression. The more they’re used, the more obvious the technique becomes and soon we’ll be tired of it.

Did you see the results? And the ones from other models like Midjourney? Many of these images are not just "passable composites" but can easily trick you as creations by human artists of high caliber. And this is only the beginning. Human artists also mine the past of the collective eons of human imagination, unconsciously. We are in many ways products of the past influencing us from the time we are born. GPT-3, DALLE-2 and the likes are not only a marvelous and almost magical novel technique, but they also have very deep philosophical implications about what we are and what is the thing we call "culture" and "language".

Nitpick: human illustrators of high calibre, not artists, which usually implies conceptual art and requires capabilities AI does not yet possess.

The stuff I see in contemporary art museums is only occasionally clearly better than what DALL-E produces. I don't think anyone would bat an eye if you put a selected collection of its images in a museum.

Not so sure, check this twitter threads with conceptual images generated from abstract prompts: https://twitter.com/nickcammarata/status/1513977172888887302

https://twitter.com/nickcammarata/status/1511861061988892675

And as said before, this is only the beginning.

My guess is that human creativity is mostly just technical skill + taste + random noise. DALL-E has the first one, and we could probably approximate the last one, so the middle is the only one that needs work. It feels like that's a similar issue to how GPT often ends up trailing off. Maybe some kind of improved attention would work? Or an improved version of the sampling trick?

Human curation is where you get the taste from.

As far as something to play with to get interesting ideas, and to take a first few cuts at implementing them, DALL-E is great. Then let's bring the human's technical skill and ability at curation into the mix.

This was obviously the case for previous models. DALL-E 2 seems different - I’ve seen a few outputs that seem like genuinely novel creative artistic works.

"Originality is the art of hiding your source"

Harry Roolaart

They are quite obviously similar to artistic work because they copy, mix and match stuff that artists have done.

When I say genuinely novel I mean genuinely novel. A trained artist can take a prompt from a person and integrate all they know from art and make something that fundamentally advances both. DALL-E appears capable of this form of creativity.

Reply on news.ycombinator.com