BREIN Foundation, an organisation protecing copyrighted work, has taken offline a large Dutch-language dataset used to train artificial intelligence (AI) models. Following a tip, BREIN discovered that the offending dataset consisted of illegal copies of tens of thousands of books, millions of lines from news articles from websites and subtitles of countless movies and TV series from illegal sources.
Had the sources used been legal, the copyright owners would still have the ability to withdraw permission for them to be used.
The dataset has since been taken offline. The creator of the dataset has signed a statement promising to stop infringing and has provided information about who received the dataset. BREIN is now investigating which AI models used this dataset and will address the parties involved.
Large language models have been criticised by content creators for their reliance on copyrighted works. In April eight newspaper publishers in the US launched legal procedings against ChatGPT developer OpenAI and Microsoft, who are already fending off actions from writers John Grisham, Jodi Picoult and George RR Martin.
OpenAI features a page on its website for lodging copyright claims, giving creators the opportunity to have their work removed from OpenAI’s large language model.
Ltd
