"Readers of texts created to use the styles of famous authors prefer works written by AI to human human-written imitations, but only after developers fine-tune AI models to understand an author's output. This finding, academics argue, means the courts need to rethink assumptions about allowing AI training on authors' works as a fair use exception to copyright liability. In a preprint paper titled "Readers Prefer Outputs of AI Trained on Copyrighted Books over Expert Human Writers," Tuhin"
"One such lawsuit, Bartz v. Anthropic, is expected to settle for $1.5 billion after Anthropic trained its models on copied works. In another such lawsuit, Kadrey v. Meta, Meta prevailed on a technical basis - due to legal deficiencies in the plaintiffs' case -"
"Legal scholars have suggested that while training AI models on copyrighted texts, recordings, and videos is probably permissible as fair use, there's likely to be liability for AI models that produce copyrighted content verbatim."
Readers prefer AI-generated imitations of famous creators over human imitations only when AI models are fine-tuned on those creators' original works. Courts may need to reassess assumptions that training AI on copyrighted books qualifies as fair use and thus avoids copyright liability. Lawsuits allege developers used copyrighted books without permission, producing high-value settlements and mixed judicial rulings on legality. Legal commentary indicates that training on copyrighted books, recordings, and videos may often qualify as fair use, while verbatim reproduction by models is likely to trigger liability. If training itself is treated as infringement, model makers could face ruinous financial exposure beyond infrastructure investments.
Read at Theregister
Unable to calculate read time
Collection
[
|
...
]