Nick Clegg, Meta president of global affairs, made the announcement during an interview on the sidelines of the company’s annual Connect conference this week.
Mr Clegg confirmed that Meta does not use private chat content on messaging applications as training data for these virtual assistant models, and also filters out private details in the public data it restores. Serve this process.
He cited LinkedIn as an example of a site whose content was ignored by Meta due to privacy concerns.
According to Mr Clegg, the Facebook and Instagram posts used to train the Meta AI contain text and images. META also places security restrictions on the content the tool can generate, such as prohibiting the creation of photographs of public figures.
The above information was published in the context of technology companies such as Meta, OpenAI and Google, which were criticized for using information taken from the Internet without permission to train AI models, which is based on summarizing the information and Use large amounts of data to create visualizations.
Meta AI was created through the use of a custom model based on the Llama 2 language model which the Meta company launched in July this year. This is the most important product among consumer-focused AI tools, which was revealed by CEO Mark Zuckerberg at the Connect event on September 27.