Head of AI research leaves Meta – clears the way for others

After eight years, Joëlle Pineau is leaving Meta. She was Vice-President of AI Research at Meta and a founding member of the Meta FAIR team. FAIR stands for Fundamental AI Research. Pineau was involved in products such as the open-source programming library PyTorch, Meta's free large language model Llama and the vision transformer model Dino. The reason for her departure is not entirely clear from a public post by Pineau.

The AI expert writes there of her admiration for the team and that it was the professional experience of a lifetime. She explicitly thanks three of her former superiors. She does not address Yann LeCun, Meta's Chief AI Scientist. Both celebrated the tenth anniversary of the FAIR team, which is based in Paris, in February. There, Pineau gave an overview of everything they have developed over the years, user numbers and progress towards Advanced Machine Intelligence (AMI). This is the name of the goal set by Meta. Unlike OpenAI, for example, Meta is not working on an Artificial General Intelligence (AGI) that is supposed to be smarter than humans. Yann LeCun also believes that the approach of achieving this using current generative AI is out of the question.

“Today, as the world is fundamentally changing, the race for AI is accelerating and Meta is preparing for its next chapter, it is time to make room for others who want to continue this work,” writes Pineau. She will be cheering the team on from the sidelines, knowing that they have what it takes to “develop the best AI systems in the world and bring them into the lives of billions of people”.

Meta chooses open source for AI

Meta's CEO Mark Zuckerberg has repeatedly emphasized that his AI applications would soon be the most widely used in the world. As Meta AI is automatically integrated into Meta's products, this is likely to prove true. Meta AI, Meta's chatbot, has recently been integrated into WhatsApp and Facebook in Germany and the EU. It has been available in the USA and other countries for some time and can also generate images or be used in Meta's Ray-Ban glasses by voice, for example.

However, the FAIR team has been heavily involved in research in the field of AI. For example, non-invasive methods for recognizing speech in the brain or a computer vision model based on self-supervised learning. This means that the model itself learns from data, in this case images. The current second version of Dino is also available as open source.

Another reason why Meta chooses open source for its AI products is that Zuckerberg believes it will spark an entire ecosystem from which they can then benefit. He also says that the company can cross-subsidize AI. An advantage that OpenAI, for example, does not have.

(emw)