Meta will train its AI on our posts | Press review n°2

A selection of important news about artificial intelligence during the week of April 14 to 20, 2025.

Apr 20, 2025

Welcome to the second press review of Artificial reality. I publish a selection of the latest important developments in AI every week. Have a good read!

📰 Read

Meta will train its AI on our posts

Meta, which owns Facebook, Instagram, WhatsApp and Threads, announced on Monday that it will soon analyze all the public images, videos, captions and commentaries of its European users to train its artificial intelligence. Their real names, usernames and profile pictures will be examined too, as well as their interactions with Meta AI. Public data of people under 18 years old will not be used to train its generative AI models, according to the company.

The public posts of users outside the European Union have already been analyzed by Meta to train its AI.

European users of Meta have recently been receiving notifications in-app and via email informing them of this decision. These messages also include a link allowing them to object to the use of their informations for this purpose. To do so, they have to fill an online form.

According to the multinational, “this training will better support millions of people and businesses in the EU by teaching AI at Meta to better understand and reflect their cultures, languages and history”.

This announcement comes almost a year after Meta’s first attempt to train its AI with contents from European accounts. The company had sent an email in May 2024 to inform them of this new practice, provoking an outcry from users and lawyers who accused it of violating European and Swiss privacy laws. The Austrian NGO None Of Your Business, who works to enforce data protection laws, including the General Data Protection Regulation (GDPR), had filed complaints against Meta in eleven European countries.

Two weeks after this announcement, Meta had decided to delay training its AI using publications of European users after a request from the Irish Data Protection Commission (IDPC), who acts on behalf on the European Union. However, the multinational stated in a press release its intention to keep working with the IDPC to be allowed to extend its AI training in Europe. A favorable opinion provided by the European Data Protection Board in December gave Meta the green light it was hoping for.

OpenAI is building a social network

American company OpenAI, which owns ChatGPT, is building a social network similar to X, The Verge revelead on Tuesday. The project, which is still in early stages, could be focused on ChatGPT’s image generation with an added social feed.

OpenAI, which released o3 and o4-mini this week, two new models which imitate human reasoning to solve complex visual and coding tasks, could position itself as a direct competitor to X and Meta with this new social network. The CEO of OpenAI, Sam Altman, had in fact announced in February that his company will maybe “do a social app” after he heard that Meta was planning to release a standalone Meta AI app to compete with ChatGPT.

Why does the company led by Sam Altman plan to create a social network when its stated objective is to reach artificial general intelligence, a technology capable of doing almost any task as well as humans, or even better than them? There’s a high probability that it is to collect a lot of users’ data, which can then be used to improve its AI models. The businessman Elon Musk recently made a similar move with his artificial intelligence company xAI, which bought X (and its troves of data) for 33 billion dollars.

A social network would also allow OpenAI to sell advertising spaces, thus getting new income sources with its chatbots.

Apple will analyze private data to improve its AI

To catch up with its main AI rivals, which include OpenAI and Google, Apple will begin analyzing data on customers’ devices.

Up until now, Apple was using synthetic data (which mimic real-world data) to train its AI models, in addition to purchased datasets and data obtained by scanning the open internet. The advantage of synthetic data is that they don’t include personal content, thus respecting the private life of the users. The disadvantage is that they can be less representative of actual customer data, which can slow down the models’ training and provide less accurate answers.

Apple will soon compare this synthetic data to recent samples of users’ emails within the iPhone, iPad and Mac Mail app to determine the relevance of its datasets and then adjust its AI models accordingly.

The company will roll out the new system in an upcoming beta version of iOS and iPadOS 18.5 and macOS 15.5.

🎥 Watch

Computer scientist Shahram Izadi, a Google employee, presented the new extended reality operating system Android XR in a Ted talk which was published on Youtube on Friday.

Extended reality is a term that encompasses the immersive technologies of augmented reality, virtual reality and mixed reality. These realities merge the real world and the digital world to assist users in various tasks or to offer new entertainment experiences, for example in video games.

At the beginning of his talk, Shahram Izadi explains that artificial intelligence has enabled fundamental evolutions in extended reality, notably thanks to large language models (such as Google's Gemini or OpenAI's ChatGPT) that enable programs to better understand natural language and images.

According to him, the fusion of artificial intelligence and extended reality represents the second act of the computer revolution because users can now interact differently with technology. Computers are becoming lighter, more personal, they share our vantage point, they understand our environment and we interact with them in a conversational way.

The computer scientist and his colleague Nishta then demonstrate a pair of augmented reality glasses. The glasses are equipped with a camera, microphones, earphones and a miniature screen. They are connected to the Google employee's phone, giving them access to its applications. The artificial intelligence Gemini sees and memorizes what Nishta sees and answers her questions in real time. They then show the audience several Android XR features.

In the second part of the video, a colleague of Shahram Izadi named Max demonstrates a virtual reality headset. We see that the AI understands everything that is displayed on Max's screen, be it a video, a map or a video game, and can answer specific questions about these contents.

These smart glasses and this virtual reality headset offer interesting features but they also represent a new form if intrusion into our private lives. The persons who will use these objets will film their family, their friends, their apartment or their house. The microphones will analyze all their conversations. And these private informations might be accessible to the companies owning these technologies, which will in certain cases share them to other companies, or even to government agencies.

Important questions thus arise: are we ready to give up on our private life to have access to new computer features? Is that a fair price to pay or are we, above all, providing new data to tech companies so they can keep making money by surveilling us? What rights over our informations do these companies have? How long do they keep them? Who has access to them? Who do they share them with?