Meta resumes training AI on user posts
Meta plans to resume using posts from Facebook and Instagram users in the UK for AI training projects.
This move comes about three months after the company stopped this practice due to privacy concerns.
Meta develops the popular Llama family of open source large language models. The company touted the largest model in the suite, called Llama 3.1 405B, as the most efficient open source large language model on the market when it launched in July.
This model was able to outperform OpenAI's GPT-4o in some tasks during the evaluations conducted by Meta.
The training data the company uses to build large language models includes public user posts from Facebook and Instagram.
Earlier this year, Meta stopped using such content to train AI models in the European Union after regulators asked it to pause the practice.
The company also implemented the pause in the UK in an attempt to address requests from the Information Commissioner's Office, or ICO.
Meta said it plans to resume using posts from UK social media users to train AI models in the coming months.
According to the company, it has modified the way it processes this data based on feedback from the Information Commissioner's Office.
Meta's discussions with the regulator focused on the " legitimate interests " clause in the GDPR.
The clause states that a company may process consumer data, such as Facebook posts, if it has compelling reasons to do so. These reasons could include commercial interests.
In order to use the “legitimate interests” clause in the GDPR as a legal basis for a data collection project, a company is required to demonstrate that it is using people’s data in ways they would expect and that have minimal impact on privacy.
The company may use this information in other ways as well if it has a very good reason to do so.
The GDPR also places further requirements on the use of the “legitimate interests” clause.
It is worth noting that Meta is required to prove that the data it collects is necessary for the project in which it is used, and it is also necessary to verify that there is no other privacy-friendly way to do this task.
Before adding Facebook and Instagram posts to the AI training dataset, Meta is supposed to send affected users a notice about the policy change.
The notice is set to include an explanation of the company’s data-handling practices. Meta also plans to launch a form that will allow users to opt out of AI training.
The company is facing European scrutiny over its use of social media posts to train AI models.