All blame falls to Microsoft. They hollow out companies that deliver on time and under budget to well above expectations. They’re just fucking stupid.

Generators are recognizers. Like how speakers and microphones are the same mechanism, backwards. ImageNet was only a classifier for a couple thousand single-word labels. Dall-E used the same approach to walk an image toward those labels. That’s how you get “avocado chair.”

Yet no one is paying for those training data

They don’t have to, because training is transformative. It’s fair use. If permission is not a factor, why would money be required? Your rights are not a loophole. This is fine for the same reason you’re free to reference, parody, or quote commercial works guarded by flesh-eating lawyers.

Even a vegan model trained on bespoke data could reproduce trademark-infringing characters. If you can describe what Naruto looks like, the model only needs to know what anime means.

There is no difference. A chatbot that’s read every book in the library is the same as a denoiser that’s seen Zootopia. Published contents are public… that’s what the word means. It’s not theft for the robot to know what Superman looks like, and the only way it’d know that is by lookin’.

None of these models are supposed to reproduce a complete work. That’s a failure to generalize. It means they could be smaller, and still change new footage to fit a novel description. If you film yourself in a bathrobe and motorcycle helmet, and say “Darth Vader,” you won’t get a single from of A New Hope. It only removes the fine details that don’t match the concept. If you’re skateboarding, so is Vader.

Training is transformative use.

The dumbest thing we could do about spicy autocomplete is make copyright worse.

Fuck any criminal treatment for civil copyright whining, but moreover:

Linux is free.