Copyright is insidious because people tried all sorts of ways to get around it in the past. You can’t for example replace all the words in an article with synonyms and get a completely new work. Thus, asking ChatGPT to summarize an article is problematic because the output is based on the original work irrespective of what’s happening internally.
Summaries are explicitly permitted, which is why CliffNotes, Wikipedia, etc can publish synopses of copyrighted works.
Note how much of the ‘summery’ section isn’t just a listing of what happens but an interpretation of what happens: https://www.cliffsnotes.com/literature/f/fahrenheit-451/summ...
“At the same time, she also gives the reader the opportunity to see that the government has dramatically changed what its citizens perceive as their history. For example, Montag never knew that firemen used to fight actual fires...”
You are confusing training a model vs using it. This way a web browser can also be said to use NYT's content.
The point was people seem to assume only direct copies are problematic but that’s not the case. Training/feature extraction is just as mechanical as making an MP3.
Changing the words spoken to sound sophisticated does not change anything.
Obligatory IANAL.
https://mashable.com/article/chatgpt-revealed-personal-data-...
It may not be conclusive, but it's highly suggestive:
> Amount and substantiality of the portion used in relation to the copyrighted work as a whole: Under this factor, courts look at both the quantity and quality of the copyrighted material that was used. If the use includes a large portion of the copyrighted work, fair use is less likely to be found; if the use employs only a small amount of copyrighted material, fair use is more likely.