I've noticed a pattern of companies writing their customers open letters asking them to do their contract negotiations for them. First it was ESPN vs. YouTube (not watching MNF this week was the best 3 hours I've ever saved, sorry advertisers). Now it's OpenAI vs. The New York Times.

Little do they know that I care very little for either party and enjoy seeing both of them squirm. You went to business school, not me. Work it out.

In this case, it's awfully suspicious that OpenAI is worried about The New York Times finding literal passages in their articles that ChatGPT spits out verbatim. If your AI doesn't do that, like you say, then why would it be a problem to check?

Finally, both parties should find a neutral third party. The neutral third party gets the full text of every NYT article and ChatGPT transcript, and finds the matches. NYT doesn't get ChatGPT transcripts. OpenAI doesn't get the full text of every NYT article (even though they have to already have that). Everyone is happy. If OpenAI did something illegal, the court can find out. If they didn't, then they're safe. I think it would be very fair.

(I take the side of neither party. I'm not a huge fan of training language models on content that wasn't licensed for that purpose. And I'm not a huge fan of The NYT's slide to the right as they cheerlead the end of the American experiment.)

▲

themafia 10 hours ago | parent | next [-]

> Finally, both parties should find a neutral third party.

That's next to impossible. And if that party fails to be neutral you've just generated a new lawsuit entangled with this one.

The current procedure is each side gets their own expert. The two expert can duke it out and the crucible of the courtroom decides who was more credible.

	▲	jrockway 9 hours ago \| parent [-]
		That's fair. I understand why OpenAI wouldn't want to give anyone transcripts (as a user, I frankly wouldn't even want OpenAI to keep my transcripts), and I understand why the NYT doesn't want to give OpenAI all their articles. Maybe the NYT needs to bloom-filter-ify their articles in 10 word chunks (or something, I don't know enough about linguistics to tell you what's unique enough for copyright infringement or to prove "copying"), have OpenAI search transcripts, and turn over the matches. That limits the scope of the search dramatically, but is still invasive.

▲

baggachipz 10 hours ago | parent | prev [-]

Two orgs helmed by supporters of authoritarianism (to put it nicely):

let them fight.

	▲	jrockway 9 hours ago \| parent [-]
		Yup, well said. Companies want me to have an emotional investment in them. I don't.