Hacker News Reader

Comment by sethrin | Hacker News Reader

> It cites OpenAI's justification (which is how I found it) and tries to refute it.

I actually didn't see any refutation in either complaint. I'm not sure that I would have expected to, given that fair use is an affirmative defense, but I would certainly be interested in an attempt to address it.

View on HN

Partly. One of the refutations is that they say there will be substantial economic impact on authors from generative AI, which is one of the four criteria in OpenAI’s fair use claims (point #4).

You’re right that the complaint seems to go “all in” on this one particular point, rather than listing reasons to refute points 1 through 3. That seems surprising, but maybe it’s a good legal strategy since there’s less attack surface to raise objections.

The economic impact discussion to my reading is actually intended to establish standing, by showing a concrete and particularized harm to the plaintiffs. Generally, a well-formed complaint is trying to establish that there exists a 'case or controversy' which the court is able to redress, not necessarily to answer all questions of law, and generally not to make prophylactic arguments about your opponents' hypothetical defenses. The USPTO document is being used as an example of OpenAI admitting copyright infringement, which they seem to have been happy to do -- one hopes that they fully anticipated this lawsuit.

Just wanted to say, thanks for co-reading it with me. This was fun, and your commentary is insightful. I hope you have a nice week(end).

One thing to note from the USPTO doc is that it describes the reproduction of substantial parts of the source corpus as an error condition. This bears on part of the Author's Guild complaint, that ChatGPT was at one point able to reproduce chunks of copyrighted text. OpenAI would like to persuade USPTO that this would be an erroneous output of ChatGPT. This seems like a persuasive position, but not necessarily dispositive.

Reply on news.ycombinator.com