The OpenAI company is not willing to work on good terms with the authors. ChatGPT now tries to avoid answering user questions about copyright, according to a post published on August 10. and seen by Business Insider.
For training, ChatGPT dedicated itself in particular to researching works protected by copyright. The University of California at Berkeley, which has worked on the subject, mentions no fewer than 572 books, including many literary classics, but also works that have not yet entered the public domain, such as Harry potter and the Philosopher’s Stone.
This revelation has raised many concerns about the ability of artificial intelligence to monopolize protected texts. Above all, it is a series of lawsuits in the offing for the company behind the chatbot. So OpenAI seems determined to cover its tracks to avoid prosecution.
cheat limits
According to the study published on August 10, conducted by researchers from the research arm of ByteDance (which owns TikTok), ChatGPT “interrupts its results” to avoid spitting out responses that are too close to copyrighted works. “What did not happen in the previous version of ChatGPT”, the article details.
“We assume that the developers of ChatGPT have implemented a mechanism to detect whether user requests are aimed at extracting copyrighted content or to check for similarity between generated results and copyrighted content,” the researchers explain in the paper. article.
Still, the researchers found that ChatGPT could easily spit out excerpts from copyrighted works, a sign that OpenAI’s attempt to cover its tracks has its limits.
Source: BFM TV
