OpenAI faces defamation suit after ChatGPT fabricated another lawsuit | Hacker News Reader

OpenAI faces defamation suit after ChatGPT fabricated another lawsuit

arstechnica.com

30 pointsndsipa_pomu28 comments

Open articleView on HN

I would simply not believe big factual claims made by an LLM without verifying them

Sorry, why downvotes?

The comment contributes nothing and is irrelevent to the topic at hand. Not every comment is worthy of a responce.

Agree completely. And the worst part is now it's the top comment, so the top comment:

1. States the obvious

2. Adds nothing of substance. It's irrelevant to the current discussion which is about a lawsuit and the legal situation

It's too obvious a statement to add anything to the conversation, isn't it? No one has ever argued to the contrary. Every time you sign on to ChatGPT, you see a message to the effect of "Warning: I like to make stuff up."

If lawyers had a honest day of work they'd try to sue a hammer for excessive building damage

This would be like suing Adobe because someone made something in photoshop, imo.

I agree.

It would seem to me (having next to no knowledge about defamation laws) that defamation would be applicable to the person publishing the incorrect information, and not just one party in a private conversation. I suppose it's complicated if it can be shown that ChatGPT is consistently returning incorrect info for a range of questions but surely its responses aren't considered to be statements of fact?

This is a great point. This was a private comm between ChatGPT and the user who published it. It should be on the user to verify things like that before publishing.

What does a world look like where people parrot whatever they hear with no responsiblity whatsoever? If we think we have "disinformation" problems now...

It's also against the ChatGPT terms of use: https://openai.com/policies/terms-of-use

> (c) Restrictions. You may not (i) use the Services in a way that infringes, misappropriates or violates any person’s rights;

If I privately tell thousands or millions of people "John Doe committed fraud!" then am I not engaging in defamation?

IANAL, but I think you make a good point. I'd love to hear from a lawyer or someone really educated on this.

There is a difference.

A user has to intend to sit down with photoshop to produce something defamatory.

A user does not need this intention to achieve the same result with GPT.

Kind of, but mostly no. I think it would be more akin to somebody opening photoshop and using a DALL-E type tool to ask it to generate a photo of a person, and the tool generates one that is wrong and damaging to the person (like maybe putting a pastor in a strip club or in an abortion clinic or something very damaging to the person's credibility). The user then tweets it out as a "real" photo of the person.

OpenAI is definitely more involved in creating the thing than a current user of photoshop who just uses primitive tools to construct it. It's still silly IMHO to sue OpenAI

The person who posted the content is legally liable, not the company that created the tool, no matter how advanced it is. OpenAI didn’t go out and post the content.

It's not quite that simple: an unrelated third party asked to write a summary of a court case, and ChatGPT injected some false facts about Mark Walters, in spite of not being mentioned at all in this court case. Who knows how many times it's done something like this? Dozens, hundreds, or thousands of times?

I wouldn't be happy either if someone asked "please tell me about [random topic I'm vaguely related to]" and it started claiming I committed fraud (or worse!)

If I go around falsely telling people "Mark Walters committed fraud!" then I'm probably not going to win the defamation case. I don't think "yeah, but the computer did it" is a very good defence.

I think OpenAI has plenty of possibilities to explain why they aren't the responsible side in this case. E.g. they can argue that a proper way to use the tool is to provide elaborate prompts which reasonably protect from unwanted responses. I think many will see that ChatGPT tool is substantially different from a news website.

At the end of the day you are liable for what you do, if you don’t read what a tool outputs that’s on you.

Right, but that applies to OpenAI too.

You're liable for what you do with your tools. If your tool (construction crane/GPUs) cause harm (dropped an AC unit on the neighbor's house/slandered an individual's reputation), you're going to get sued for the damage. It may involve the injured party suing their neighbor, who then sues the construction contractor, who may then have to face higher business insurance premiums or even license revocation if the harm is severe or part of a recurring pattern.

There is no allegation that the content was shared by anyone other than the single user of ChatGPT that created the initial prompt. That user was a journalist.

There is no information about what the journalist did with the information, but if I had to guess, the journalist recognized that it was nonsense and either him or a colleague contacted Walters because they thought it could make for a a good ChatGPT story.

Legally speaking, it is not obvious that this is relevent. Defamation does not require widespread publication. A libelous statement to a single person is enough. Further, defamation pe se does not require proving damages; and a false claim that someone committed a crime qualifies as defamation per se.

If OpenAI didn't go out and post the content then where did Fred Riehl see the content?

I think it is more akin to someone using a search engine to find something, then suing the search engine for it.

The difference here is if the thing found is "libelous" then the original publisher would be responsible, not the search engine.

But in this case, there is no other "publisher". The publisher here is the person posting the chat transcript, they are the ones with agency and causing those bits to be presented in the public where they become libelous.

I am 100% behind protections for journalists, but in this case it would be the journalist who "uncovered" the statements without corroborating them that is ultimately responsible for putting forth false facts.

Also, not a lawyer, not legal advice.

"ChatGPT may produce inaccurate information about people, places, or facts."

good luck with lawsuit when this text is next to your input for chat.

like seriously, how can you be mad at openai when you fail to read what is shown to you?

> …how can you be mad…when you fail to read…

quite easy, actually.

> Similarly, Volokh suggested that OpenAI might not be shielded by ChatGPT disclaimers like the one in its terms of use.

> "Ars Technica can’t immunize itself from defamation liability by merely saying, on every post, 'this post may contain inaccurate information'—likewise with OpenAI," Volokh told Ars.

Besides the point of whether this is just or not,

Wait, is there actually a case where a corporation can be held liable for something? All those terms of use “agreements” that you can’t live without signing a thousand of, they don’t completely take away your rights?

Notably Walters never agreed to anything with OpenAI. He had no relationship with OpenAI, so their terms of service are irrelevant here.

Seems hard to prove damages, since the person requesting the summary immediately recognized it as inaccurate.

Simple fix: sue the guy. Then ChapGPT isn't lying.

Reply on news.ycombinator.com