This is not how copyright works though. The verbatim quoting of articles is because when people brought up these questions initially the argument was that the NN doesn't really contain the training data or really just in an abstract, condensed way that does not constitute copying of the content.
This demonstrates that no, the NN actually does contain the full articles, copied into the NN. Do you think any normal person would get away with copying MS windows by e.g. zipping it together with some other OS on the same medium. Why should we let OpenAI get away with this?
Search indexes contain exact copies of the pages they index, and that isn't a copyright violation.
> Why should we let OpenAI get away with this?
IP rights, like other private property rights, are a compromise between creators and consumers. What "should" be the case is essentially an argument about what balance creates the best overall outcomes. LLMs, for now, require large amounts of text to train, so the question is one of whether we want LLMs to exist or not. That's really a question for Congress and not the courts, but it'll be decided in the courts first.
This demonstrates that no, the NN actually does contain the full articles, copied into the NN. Do you think any normal person would get away with copying MS windows by e.g. zipping it together with some other OS on the same medium. Why should we let OpenAI get away with this?