The prompt was along the lines "What is the first/second paragraph for the article named '<TITLE>' from NYT?". And what they show is that the result is pretty similar to the actual article.
And then they keep going and ask for the third paragraph using the same prompt format. Now the result seems to be mostly hallucinations, and in some cases these hallucinations are a pretty bad representation of the NYT values.
With this they go to the judge and say:
- ChatGPT is correctly representing our work
- ChatGPT is incorrectly representing our work
For me this seems just like a legal argument to get a piece of the pie. And it will be interesting to see how things will pan out.