This is exactly how I understand it. There’s a lot ink getting spilled about “summarizing isn’t illegal” and “what about Cliffs Notes” but that isn’t what this is about.

If the verbatim examples that have been going around are true, that’s bad. I’d love to know more details around it — prompts used, whether that’s an old model, etc. This seems like plagiarism more than anything.

IMO it's pretty hard to training an LLM isn't a transformative use. It's clearly not just copying, or even excerpting. Even if it was just compression (and it's not), they're only providing model output not distributing the "compressed" NYT articles.

Yielding verbatim snippets of copyrighted content is a problem for OpenAI though.

Perhaps we will see the courts revisit this idea of "transformative" works and formulate something more useful. I'm my opinion, you can't build an LLM unless you have a large amount of data with which to train it. Given the huge amount of money companies like OpenAI hope to generate, it seems unreasonable that content creators would not be rewarded.