Is Meta’s AI Stealing? Meta’s Llama 3.1 and the ‘Harry Potter’ Copy-Paste Controversy
Meta’s newest AI model, Llama 3.1, is making waves — but not entirely for the right reasons. A recent study found that this AI isn’t just learning from books; it’s memorizing and reproducing large parts of copyrighted works, including about 42% of the first Harry Potter book.
(ALSO READ: MSI Claw A8 BZ2EM – Double the Battery, Double the Power in a Handheld Console)
Researchers from Stanford, Cornell, and West Virginia University examined five major AI models using a huge dataset called Books3, which contains thousands of books, many still under copyright. To their surprise, Llama 3.1 was able to perfectly repeat 50-word excerpts nearly half the time.
This behavior raises serious legal and ethical questions. How much copying is too much when it comes to training AI? And what does this mean for authors whose work might be duplicated without permission?
Meta has yet to comment on the findings, but the issue highlights a growing challenge in the AI world: balancing innovation with respect for intellectual property rights.
As AI continues to evolve, this debate around copyright and creativity is only going to get louder — and it’s one that both tech companies and creators will need to navigate carefully.