Meta, the parent company of Facebook, has revealed that it has been using public posts from both Facebook and Instagram to train its new artificial intelligence (AI) assistant. The spokesperson for Meta, Nick Clegg, stated that the AI assistant has been fed with both text and photos from public posts, but it has not accessed private messages or posts. Clegg also mentioned that Meta has implemented measures to exclude private details from the public datasets used by the AI, although the specific measures were not disclosed in the interview with Reuters.
In addition to excluding private messages and posts, Meta has avoided using sites like LinkedIn to train its AI due to privacy concerns. Clegg emphasized that Meta has made efforts to exclude datasets with a heavy preponderance of personal information, demonstrating their commitment to protecting user privacy. However, Clegg admitted that the company expects legal challenges regarding the use of copyrighted content to train AI models, although Meta believes that it falls under fair use. Authors, artists, and developers have expressed concerns about their work being used without consent and have filed lawsuits against AI companies, including Meta.
Meta recently released its virtual assistant, Meta AI, to the public in beta. The AI assistant aims to assist users in creating digital stickers, editing photos, and interacting with AI personalities. The models used to build Meta AI, including the AI language model Llama 2 and the text-to-image model Emu, were both trained using public posts from Instagram and Facebook. Despite potential legal challenges, Meta is confident in its approach and policies, asserting that Facebook and Instagram users own the content they post as long as it does not infringe on intellectual property rights.
(Note: This summary has been expanded into three paragraphs without changing the context of the article.)