4 ms·

>If you'd spend any time at all looking at midjourney creations, there's no way you can conclude that there's any significant amount of "ripping off" people's w

>If you'd spend any time at all looking at midjourney creations, there's no way you can conclude that there's any significant amount of "ripping off" people's work. Certainly not in all or most cases. They were trained on work of people who didn't consent to it, the authors were not compensated, and currently are losing money (demand for their work) because of it. How is that not ripping off. Please don't compare software to people. They're not similar legally, morally or in any other meaningful way.

harshreality 3y ago
Creators don't get compensated just because you wish them to be. Creators don't get compensated based on the fact that they put work into something. Creators can't demand money in license fees when something they produced is used by someone else, unless there's a contract saying so. Copyright is not about consent to use. It's about consent to copy. You're confusing copyright with licensing, where, for instance, I pay amazon to contractually license me an electronic copy of a book. AI companies did not enter into and then violate contracts with anyone for the material they used to train. Some of them may have run afoul of ordinary copyright law to acquire the content in the first place.[1] Even supposing they did, that act is the violation; it wouldn't subsequently be an additional violation to train an AI using those works, instead of keeping the works around on a storage array, as long as it wasn't redistributed. The alternative to AI models currently at issue—created by OpenAI Midjourney StabilityAI Meta Alphabet et al from a corpus of partly openly available content and partly content of dubious provenance—is not some license scheme where creators get compensated equitably. The real alternative is for companies like Disney and Adobe and book publishers and record labels to band together and have their own paywalled AI models and extract rent from that forever, which would still be sufficient to saturate the market for imagined content, and would still kill demand for everyone else's work. [1] I'm skeptical, though open-minded, about this. I'm not aware of any successful lawsuits, or any lawsuits at all, targeting defendants who only downloaded copyrighted content from pirate sources, as long as they did not upload or re-upload at all. There might have been, and probably were sometimes, threatening letters sent to people's ISPs. Threatening letters don't mean anything at all.
- mnd999 3y ago
  If I compress a Getty images photo to .jpg and put it on my website I’m still breaching copyright even though it’s gone through lossy compression and isn’t exactly the same. That’s essentially all a AI model is doing. And copyright applies to every copy, not just when you download something to train the model. It also applies every time you make a copy with said model.

harshreality 3y ago

Creators don't get compensated just because you wish them to be. Creators don't get compensated based on the fact that they put work into something. Creators can't demand money in license fees when something they produced is used by someone else, unless there's a contract saying so. Copyright is not about consent to use. It's about consent to copy. You're confusing copyright with licensing, where, for instance, I pay amazon to contractually license me an electronic copy of a book. AI companies did not enter into and then violate contracts with anyone for the material they used to train. Some of them may have run afoul of ordinary copyright law to acquire the content in the first place.[1] Even supposing they did, that act is the violation; it wouldn't subsequently be an additional violation to train an AI using those works, instead of keeping the works around on a storage array, as long as it wasn't redistributed. The alternative to AI models currently at issue—created by OpenAI Midjourney StabilityAI Meta Alphabet et al from a corpus of partly openly available content and partly content of dubious provenance—is not some license scheme where creators get compensated equitably. The real alternative is for companies like Disney and Adobe and book publishers and record labels to band together and have their own paywalled AI models and extract rent from that forever, which would still be sufficient to saturate the market for imagined content, and would still kill demand for everyone else's work. [1] I'm skeptical, though open-minded, about this. I'm not aware of any successful lawsuits, or any lawsuits at all, targeting defendants who only downloaded copyrighted content from pirate sources, as long as they did not upload or re-upload at all. There might have been, and probably were sometimes, threatening letters sent to people's ISPs. Threatening letters don't mean anything at all.

mnd999 3y ago
If I compress a Getty images photo to .jpg and put it on my website I’m still breaching copyright even though it’s gone through lossy compression and isn’t exactly the same. That’s essentially all a AI model is doing. And copyright applies to every copy, not just when you download something to train the model. It also applies every time you make a copy with said model.