Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Stable Diffusion is a big deal

simonwillison.net

Post reply on HN
Print view View original

481–488 of 488 posts

Re: Stable Diffusion is a big deal

#481

Post by ryanklee »

SyzygistSix wrote: post #261
systemvoltage wrote: post #138

Earlier quoted context omitted.

I always wondered if we massively overestimate human creativity. Maybe it is ingrained in our culture and our very being. I’ve never heard counter arguments that humans are not that creative. Creativity demonstrated by Alpha zero chess engine blows Magnus Carlsen’s mind (from his recent interview with Lex Fridman), I wonder if at some point in the future, we’ll finally throw in the towel and get out of the denial pha…

No, apparently we just massively underestimate how important studying the humanities is. This is illustration, not art. I'm flabbergasted how many people are so quick to confuse or conflate the two.

Would you mind offering definitions?

Re: Stable Diffusion is a big deal

#482

Post by planb »

peoplefromibiza wrote: post #471
planb wrote: post #175

> Stable Diffusion has been trained on millions of copyrighted images scraped from the web. My brain has been trained on even more copyrighted material. Every book I read, every tv show I watch, the toys I played with as a child. It's hard to imagine that I could come up with anything that is not inspired by copyrighted work.

> My brain has been trained on even more copyrighted material What your brain has learned cannot be transferred with an USB stick in seconds Not even your offspring will receive any of it If I want to learn everything you know, I have to learn what you learned , assuming I will be able to Kinda of a big difference, don't you think? These kinds of comments are embarrassingly low effort, just because we threw rocks at…

The original article talks about moral (not even legal) objections to learning from copyrighted data.

Your comment implies that DALL-E 2 is morally okay, because they don't distribute the model ("copy everything it knows to a USB stick") but only sell access to the algorithm to generate images, while Stable Diffusions open source model is a problem because it can be copied.

Most people would take the exact opposite stance I guess.

Re: Stable Diffusion is a big deal

#483

Post by peoplefromibiza »

planb wrote: post #482
peoplefromibiza wrote: post #471

Earlier quoted context omitted.

> My brain has been trained on even more copyrighted material What your brain has learned cannot be transferred with an USB stick in seconds Not even your offspring will receive any of it If I want to learn everything you know, I have to learn what you learned , assuming I will be able to Kinda of a big difference, don't you think? These kinds of comments are embarrassingly low effort, just because we threw rocks at…

The original article talks about moral (not even legal) objections to learning from copyrighted data. Your comment implies that DALL-E 2 is morally okay, because they don't distribute the model ("copy everything it knows to a USB stick") but only sell access to the algorithm to generate images, while Stable Diffusions open source model is a problem because it can be copied. Most people would take the exact opposite s…

> The original article talks about moral (not even legal) objections to learning from copyrighted data.

but I am replying to

"My brain has been trained on even more copyrighted material. Every book I read, every tv show I watch, the toys I played with as a child. It's hard to imagine that I could come up with anything that is not inspired by copyrighted work"

Difference being your brain has not been trained by someone (for profit), you have trained it using YEARS OF YOUR LIFE TO ACQUIRE KNOWLEDGE AND EXPERIENCE

which is morally acceptable (does not imply that the use you do of it is legally acceptable), given that you paid a very high price, sacrificing your own time for the objective.

And that your knowledge is only yours, you can't transfer it to anyone, it doesn't even show up in your DNA.

> Your comment implies that DALL-E 2 is morally okay, because they don't distribute the model

Implication doesn't mean what you think it means.

My comment doesn't imply anything of the sort, you are

> Most people would take the exact opposite stance I guess.

https://en.wikipedia.org/wiki/False_dilemma

https://en.wikipedia.org/wiki/False_premise

Re: Stable Diffusion is a big deal

#484

Post by kjkjadksj »

fleddr wrote: post #453
kjkjadksj wrote: post #355

Earlier quoted context omitted.

How about the fact that you can go on github today and find a thousand ml based music generators probably, and yet people still like going to concerts and seeing an artist play an instrument.

Not a meaningful comparison. Seeing an artist play an instrument is a fundamentally different experience. It's visual, much more impressive sound, a social experience, and so on. With imagine generation you just look at output. There's no difference between seeing the output of a human-created image or an AI-created image, people can't tell.

Ok, but you still see people passively listening to real artists instead of ai generated music. Provenance matters for music and for art. Maybe if you design retail art for Target without your name ever put on the work you have to worry.

Re: Stable Diffusion is a big deal

#485

Post by sigmoid10 »

thrown_22 wrote: post #258
sigmoid10 wrote: post #179

Earlier quoted context omitted.

Not yet, but I can definitely imagine a future where these tools get more capable and refined, to the point where all the shortcomings listed above will be overcome. Knowledge about cameras and scene composition are already encoded in the networks to some degree, it just needs to become more accessible. There's probably also a better way to seed new images than by starting with random noise, so we could get similar v…

You need to give some information about the scene to the network. Camera settings is just a short hand to describe the field of view and depth of focus (at the very least). If you make that implicit you'd still need to give the network the steradians, focal length, circle of confusion, etc. etc. etc. that you want your image to use. You'd need to understand everything in Hecht's Optics to tweak all the parameters of…

That's an implementation problem, not a technical or conceptual one. Diffusion models have shown that they can learn practically all of these things if you make them sufficiently big.

Re: Stable Diffusion is a big deal

#486

Post by cgabear »

ryandrake wrote: post #413
galdosdi wrote: post #394

Earlier quoted context omitted.

Can I ask what kind of programming you do for it to be so helpful? I mostly do maintenance of legacy codebases (also known as codebases, lol) where a lot of the work is figuring out where the changes need to be made and actually making the changes is frequently just a few lines here and there. When I do have to figure out how to use some API, it's often not an open source one, so Copilot would not have it in its corp…

I haven't tried copilot either, but one of the things I'd be curious about is how well it can conform to a company's coding style guidelines and/or match its coding style with the existing legacy code that's being modified. One of the major annoyances of working as a team with legacy code is when someone forgets to, or deliberately avoids, conforming their code to the style and techniques of the surrounding code. Not…

It works great in our codebase; it uses the current text in the file from above your cursor as reference. So if you're creating a new file, it isn't always perfect, but once it catches on to your style it's seamless.

Re: Stable Diffusion is a big deal

#487

Post by idontpost »

amelius wrote: post #437
idontpost wrote: post #391

Earlier quoted context omitted.

> A career typically requires 40 years of job security. 40 years ago was 1980. There's a vanishingly small number of fields that have had continuous job security from 1980-2020 (even ignoring Covid). You might as well say that careers are over for everyone, and have been for a while.

Doctors and lawyers are doing great, still.

If you think lawyers were a secure field over the last 40 years, you're simply ignorant.

A tiny percentage of lawyers do really well. The rest struggle under crushing debt and insane working conditions.

So you've got one example in the entire economy.

Post reply on HN