Fighting the New York Times' invasion of user privacy

Its clearly propaganda. "Your data belongs to you." I'm sure the ToS says otherwise, as OpenAI likely owns and utilizes this data. Yes, they say they are working on end-to-end encryption (whatever that means when they control one end), but that is just a proposal at this point.

Also their framing of the NYT intent makes me strongly distrust anything they say. Sit down with a third party interviewer who asks challenging questions, and I'll pay attention.

I get the feeling, but that's not what this is.

NYTimes has produced credible evidence that OpenAI is simply stealing and republishing their content. The question they have to answer is "to what extent has this happened?"

That's a question they fundamentally cannot answer without these chat logs.

That's what discovery, especially in a copyright case, is about.

Think about it this way. Let's say this were a book store selling illegal copies of books. A very reasonable discovery request would be "Show me your sales logs". The whole log needs to be produced otherwise you can't really trust that this is the real log.

That's what NYTimes lawyers are after. They want the chat logs so they can do their own searches to find NYTimes text within the responses. They can't know how often that's happened and OpenAI has an obvious incentive to simply say "Oh that never happened".

And the reason this evidence is relevant is it will directly feed into how much money NYT and OpenAI will ultimately settle for. If this never happens then the amount will be low. If it happens a lot the amount will be high. And if it goes to trial it will be used in the damages portion assuming NYT wins.

The user has no right to privacy. The same as how any internet service can be (and have been) compelled to produce private messages.

> In copyright cases, typically you need to show some kind of harm.

NYT is suing for statutory copyright infringement. That means you only need to demonstrate that the copyright infringement, since the infringement alone is considered harm; the actual harm only matters if you're suing for actual damages.

This case really comes down to the very unsolved question of whether or not AI training and regurgitation is copyright infringement, and if so, if it's fair use. The actual ways the AI is being used is thus very relevant for the case, and totally within the bounds of discovery. Of course, OpenAI has also been engaging this lawsuit with unclean hands in the first place (see some of their earlier discovery dispute fuckery), and they're one of the companies with the strongest "the law doesn't apply to US because we're AI and big tech" swagger.

The original lawsuit has lots of examples of ChatGPT (3.5? 4?) regurgitating article...snippets. They could get a few paragraphs with ~80-90% perfect replication. But certainly not full articles, with full accuracy.

This wasn't solid enough for a summary judgement, and it seems the labs have largely figured out how to stop the models from doing this. So it looks like NYT wants to comb all user chats rather than pay a team of people tens of thousands a day to try an coax articles out of ChatGPT-5.

Yeah, everyone else in the comments so far is acting emotionally, but --

As a fan and DAU of both OpenAI and the NYT, this is just a weird discovery demand and there should be another pathway for these two to move fwd in this case (NYT to get some semblance of understanding, OAI protecting end-user privacy).

100% agreed. In the time you wrote this, I also posted: https://news.ycombinator.com/item?id=45901054

I felt quite some disappointment with the comments I saw on the thread at that time.

>But conversations people thought they were having with OpenAI in private

...had never been private in the first place.

not only is the data used for refining the models, OpenAI had also shariah policed plenty of people for generating erotica.

By the same token, why isn't NYT proposing something like that rather than the world's largest random sampling?

You don't have to think that OpenAI is good to think there's a legitimate issue over exposing data to a third party for discovery. One could see the Times discovering something in private conversations outside the scope of the case, but through their own interpretation of journalistic necessity, believe it's something they're obligated to publish.

Part of OpenAI holding up their side of the bargain on user data, to the extent they do, is that they don't roll over like a beaten dog to accommodate unconditional discovery requests.

> 5. Top secret information

https://en.wikipedia.org/wiki/Pentagon_Papers

They didn't keep temporary chats. They were ordered to keep those as part of this case.

- any corporation

remember a corporation generally is an object owned by some people. Do you trust "unspecified future group of people" with your privacy? You can't. Best we can do is understand the information architecture and act accordingly.

In their defense, they don't say they are maximizing trust, security, and privacy, only that those factors guide their decision making. The part they don't say is that they are balancing those things against functionality and revenue potential.

> Trust, security, and privacy guide every product and decision we make except ones that involve money.

-- openai, probably.

You know you have a branding problem when (1) you have to say that at the outset, and (2) it induces more eyerolls than a gaggle of golf dads.

Stopped reading at this line