Meta to use European social media posts to train AI

Meta Platforms will train its AI models with Europe’s public social media content. This is the same way the company handles data in Europe as it does in other regions, even though privacy advocates and legal groups were initially worried about it.

On Monday, Meta Platforms, the company that owns Facebook, announced that it will start using social media content from Europe to train its generative AI models.

In a blog post, Meta said that it will train its Llama large language models with content that people in the European Union have chosen to share publicly on Instagram, Facebook, and other sites. The change appears to bring the company’s approach in Europe closer to how it handles data from other parts of the world that it feeds into its AI models, even though the company was cautious at first because of strict EU privacy and transparency rules.

According to Meta’s top policy executive in September, the company trains its Llama models on public Facebook and Instagram posts, but not on private posts or messages shared only with friends.

Meta faces privacy concerns over AI data use in Europe

When Meta started launching new versions of Llama in April, the company was “still working on the right way to do this in Europe,” the chief product officer said.

At the end of last month, the big tech company said it would start telling Facebook and Instagram users in Europe and the UK how it uses public data shared on Meta’s services to build and improve AI.

NYOB (none of your business), an advocacy group, has filed complaints against the move in several European countries, saying that the notifications were not enough because EU privacy rules require Meta to get users’ opt-in consent.