As for illegal access, I agree that the report uses the term a bit too loosely. But as we’ve seen in the Meta case, some companies have obtained training material not through TOS-violating downloads but through literal (unauthorized) torrents. As we’ve also seen in the Meta case, even torrenting is technically not copyright infringement if you’re not seeding. But the process does rely on someone else seeding, so the report doesn’t seem wholly unreasonable in suggesting that this could “reflect bad faith” or “bear on the character of the use”.

All rights reserved. No part of this publication may be reproduced, distributed, or transmitted in any form or by any means, including photocopying, recording, or other electronic or mechanical methods, without the prior written permission of the publisher, except as permitted by U.S. copyright law.

“Reproduced” and “electronic” are the relevant terms here.

I remember when gpt-3 came out and you could get it to spit out chunks of Harry Potter and I wondered why no-one was being sued.

The models are built on copyright infringement. Authors and publishers of any kind should be able to opt out of being included in training data and ideally opt-in should be the default.

And I hope one day someone trains a model without the use of works of fiction and we find a qualitative difference in their performance. Does a coding model really need to encode the customs, mores and concerns of Victorian era fictional characters to write a python function?

Calling it "spirit of the law" doesn't let them off the hook. If you enshrine a metaphysics that treats human-authored works as ontologically distinct kinds of data, imbued with some persistent essence that radiates rights regardless of use, you're not interpreting the law, you're institutionalizing a theology of authorship.

And yes, I care less about their intentions than about the system they're building. That system is now enforcing metaphysical categories with legal teeth. That's the problem.