A day before the firing of the head of the U.S. Copyright Office, the third installment of the office's series of reports on copyright issues and AI was released. The 113-page document covers a lot of ground, not the least of which is the question of whether the use of copyrighted works in training large language models amounts to fair use.
No surprises were offered, however. The report predictably notes that in weighing the four relevant factors for fair use, the outcome will be fact-specific. If the model is used for a purpose that is very different from that of the materials on which it is trained, a finding of fair use is more likely. But if the output competes with the training materials, the use is likely to be deemed fair.
The question of fair use is one of fact, and the Copyright Office analysis of the issue in connection with the idea of AI training reaffirms that the new application is unlikely to change the traditional fair use analysis.
"Many uses, however, will fall somewhere in between."
www.copyright.gov/...
