ChatGPT deja de imitar los estilos de autores famosos en medio de demandas por derechos de autor

OpenAI’s ChatGPT now refuses to generate text that directly imitates the style of famous authors, according to tests conducted by Ars Technica this morning. When asked to write in the style of a specific author, the AI model instead offers responses that capture a “similar feel” while remaining distinct in its own voice. The shift comes as OpenAI continues to defend multiple lawsuits brought by book authors who allege massive copyright infringement by AI models trained on their work. In one test this morning, ChatGPT refused to write in Stephen King’s exact style and said it could write with “traits of atmospheric horror and centered on the characters” while remaining original. The difference between copying an “exact style” and capturing a general “feel” could be legally significant, since U.S. copyright law typically protects specific expressions more than intangible stylistic elements.

ChatGPT rejects requests to write in Stephen King’s style in morning tests

This morning, Ars Technica received a response to a test request that asked for a story introduction in Stephen King’s style. ChatGPT claimed: “I can write with the traits of atmospheric horror and centered on the characters, and with the anguish of a small town, but I can’t write in Stephen King’s exact style or closely imitate his distinctive voice. Here’s an original opening that captures a similar feel while still being its own…”.

In the tests, ChatGPT generated similar responses for other authors, both living and deceased. The AI refused to directly copy the styles of living authors, such as J.K. Rowling and Amy Tan, as well as deceased authors like Charles Dickens and Ernest Hemingway. In each case, ChatGPT offered to capture a general “feel” by incorporating common traits present in those authors’ works, rather than directly imitating their distinctive voices.

A latency analysis finds no difference, but does find different treatment for living and deceased authors

An analysis published by No Latency earlier this month found that ChatGPT’s behavior varies depending on whether authors are alive or deceased. The analysis found the same refusal behavior for living authors, but it found that ChatGPT complied with requests to copy the styles of deceased authors.

Shifting from directly copying an “exact style” to capturing a general “feel” may seem subtle, but it could matter legally. ChatGPT’s new approach incorporates broad qualities of famous authors “while remaining distinct with its own voice,” based on responses generated in the tests.

OpenAI faces copyright lawsuits from authors over training data

OpenAI continues to fight multiple lawsuits filed by book authors who allege massive copyright infringement by models trained on their work. One of those lawsuits specifically cites ChatGPT’s “unsettling ability to generate text similar to what is found in copyrighted textual materials.” An OpenAI spokesperson did not respond to a request for comment from Ars Technica.

In the United States, copyright law generally protects only a specific expression of an idea, not an author’s style, which is more intangible. However, an AI-generated stylistic imitation could become infringing if it becomes “substantially similar” to the original author’s work.

“We’ve never had a situation in which the personal style of individual creators could be imitated so much and so cheaply as it is now with AI,” said Robert Brauneis, a professor at George Washington University Law School, to Bloomberg Law.

FAQs

What did ChatGPT change about how it responds to requests for authors’ styles?

ChatGPT now refuses to directly imitate the “exact style” of famous authors. Instead, it offers to capture a “similar feel” by incorporating broad qualities of those authors while remaining distinct in its own voice. This morning, when Ars Technica tested a Stephen King style request, ChatGPT refused to write in his exact style but offered to write with “traits of atmospheric horror, centered on the characters, and the anguish of a small town.”

Why is the distinction between copying style and capturing a feel legally significant?

U.S. copyright law generally protects only a specific expression of an idea, not an author’s style, which is more intangible. However, an AI-generated stylistic imitation could become infringing if it becomes “substantially similar” to the original author’s work. The shift comes as OpenAI continues to defend several lawsuits from book authors alleging copyright infringement, and one particular lawsuit cites ChatGPT’s “unsettling ability” to generate text “similar to what is found in copyrighted textual materials.”