Hacker Newsnew | past | comments | ask | show | jobs | submitlogin

> Within a five years it will be possible to create realistic AI virtual actors and sets that generate the entire video sequence including the performance.

That to me means we are talking about the current state of AI.

I don't disagree with setting a legal framework around ownership of AI derived art. However, I don't think this is a pressing issue. AI art, outside of maybe simple picture generation, is bad. AI books are terrible. And I don't see it getting any better.

I'd be shocked if in the next 5 years there's any hit AI art generated. Heck, I'd be shocked if it happens within the next 20 years.



Over the last year the LLM state of the art which might be used to create scripts has gained at least 20 IQ points.

The Claude 2 LLM just released has 100,000 token context and is fairly intelligent.

Take a look at the output of Stable Diffusion XL 0.9 versus Stable Diffusion 1.5. It's a massive improvement.

Metahumans is on an early version and already fairly realistic. Look at the realism of a game like Unrecord.

AI is getting much better very rapidly.


> Over the last year the LLM state of the art which might be used to create scripts has gained at least 20 IQ points.

Citation needed. How is IQ being measured? What was it previously?


It's funny that you focused on that instead of the rest of their comment. By "20 IQ points" they clearly mean it in a metaphorical or analogous way to humans, of course they are not literally measuring the IQ of an LLM.


Because the rest of their comment was basically "it's so good because look at how good it is!".

OP equates number of tokens with intelligence, I do not.

OP then mentions a hyper realistic video game. Cool, but also not exactly evidence that AI could be plopped on top of that to make Jerry Springer.




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: