Sarah Silverman sues Meta and OpenAI, alleging their language models ripped off her book

“Write a joke in the style of Sarah Silverman”… The comedian is suing OpenAI and Meta, saying they infringed on the copyrights of her book, “The Bedwetter,” in a case that could spark unfunny consequences for AI companies. Silverman, along with authors Christopher Golden and Richard Kadrey, say OpenAI’s ChatGPT and Meta’s LLaMA were “knowingly and secretly trained” using illegally obtained copies of the authors’ books.

• Exhibit A: The lawsuits’ claims include summaries of the authors’ works that they say ChatGPT churned out when prompted, as well as a document trail suggesting that LLaMA was trained on a “shadow library” site called The Pile, which the complaint says contained illegally copied full works.

• Payback: The authors seek payment and other damages, and the suit says thousands of other scribes could be plaintiffs in the case.

When imitation ≠ flattery… Generative AI has created a legal obstacle course for copyright law. Getty Images sued StabilityAI earlier this year, accusing its image generator of using copyrighted pics without permission. And Hollywood actors and writers are seeking protections against AI ripping them off in studio-contract negotiations. Meanwhile, major news publishers may be prepping for a case to force Microsoft and Alphabet to compensate them for using their articles in AI search results.

• Law and AI order: Last month the EU passed a draft of its “AI Act,” a historic law that would make AI companies create safeguards against illegally generating content.

• Cite your sources: It would also require companies to disclose any copyrighted materials used in training their models (which critics say is technically impossible).