Claude Remembers Harry Potter Better Than Your Book Club Does, Study Finds
Ourtake
Models that memorize books nearly word for word make the training-data copyright question much harder to wave away.
What happened
Researchers at Carnegie Mellon University and Instituto Superior Técnico introduced RECAP, a method that uses a feedback loop with several language models to reconstruct training text. RECAP reconstructed large portions of books including The Hobbit and Harry Potter, and Claude 3.7 generated about 3,000 passages from the first Harry Potter book. The researchers say the results could matter for future copyright lawsuits.