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

Ask HN: DALL-E was trained on watermarked stock images?

news.ycombinator.com

Post reply on HN
Print view View original

121–130 of 233 posts

Re: Ask HN: DALL-E was trained on watermarked stock images?

#121

Post by GuB-42 »

ehsankia wrote: post #44
webwielder2 wrote: post #38

These are the absolute worst DALL-E images I've seen. Do people generally just share the amazing ones and most of the output is actually complete shite? Like Instagram presenting the top 1% of people's lives.

Top 1% is a bit exaggerated, but there is definitely a lot of not good stuff. I find that Dall-E does especially poorly with underspecified prompts too, unlike something like Midjourney which can give visually pleasing photos for even the most abstract concepts. Dall-E tends to do better with concrete and specific prompts. Here's an example: Stressful Shapes Dall-E: https://i.imgur.com/JBkSh0y.png Midjourney: https:/…

But still "king of belgium giving a speech to an audience, but the audience members are cucumbers" is very specific.

And I don't see the king of Belgium anywhere, two pictures have absolutely nothing to do with the prompt (no king, no speech, no audience, no cucumber), one has the speech and audience but no king or cucumber. Graphically, they are deep into the uncanny valley. Only the third image is kind of right, if you really stretch your imagination.

Re: Ask HN: DALL-E was trained on watermarked stock images?

#123

Post by trention »

My personal opinion is that it's unethical (and possibly illegal, in a subset of cases) to train models on data without explicit consent of the creators of that data. And that really encompasses all data - generative models were not a thing when said data was created and no matter how it was licensed before, explicit consent about using it for model training must be obtained from the creators themselves.

That being said, arguments about copyright are just a fig leaf as far as I am concerned. The outcome of whether this is allowed or not will depend on the net impact of using those models on the job market and whether society will be willing to tolerate it.

Re: Ask HN: DALL-E was trained on watermarked stock images?

#125

Post by InvOfSmallC »

webwielder2 wrote: post #38

These are the absolute worst DALL-E images I've seen. Do people generally just share the amazing ones and most of the output is actually complete shite? Like Instagram presenting the top 1% of people's lives.

I have access. You get for trials for each query. I have to say that usually there is only one that is good on those three. Sometimes you need to refine your query. I'm pretty impressed as a user.

Re: Ask HN: DALL-E was trained on watermarked stock images?

#127

Post by xg15 »

Reminds me of the discussion about GitHub Copilot using the entirety of GitHub as training data. I was honestly baffled how many people, even experts in the field, saw use as training data as non-infringing. With the corrolay that it's apparently perfectly legal to "copyright-wash" a work by feeding it to an AI and have that AI generate a slightly different but extremely similar work.

Considering how strict and heavy-handed copyright handling has been otherwise, this has added to my belief that copyright in practice is really just enforcement of the interests of whatever industry has the most power at a given time: When entertainment and content generation was the biggest revenue generator, copyright couldn't be strict enough, now all money is on AI and suddenly loopholes the size of barn doors pop up.

Re: Ask HN: DALL-E was trained on watermarked stock images?

#129

Post by smileybarry »

webwielder2 wrote: post #38

These are the absolute worst DALL-E images I've seen. Do people generally just share the amazing ones and most of the output is actually complete shite? Like Instagram presenting the top 1% of people's lives.

It definitely requires some very detailed descriptions and sifting through to find a good one. One time I've regenerated a prompt as well because the existing 4 were just not that good. But I did get some great ones at a pretty good usable:unusable ratio.

Re: Ask HN: DALL-E was trained on watermarked stock images?

#130

Post by rich_sasha »

xg15 wrote: post #127

Reminds me of the discussion about GitHub Copilot using the entirety of GitHub as training data. I was honestly baffled how many people, even experts in the field, saw use as training data as non-infringing. With the corrolay that it's apparently perfectly legal to "copyright-wash" a work by feeding it to an AI and have that AI generate a slightly different but extremely similar work. Considering how strict and heavy…

Written laws are vague, practical verdicts are based on case law, cases are won by better-funded lawyers, rich industries prevail.

It's a bit of an exaggeration but maybe not too much.

Post reply on HN