FAIR: Meta is grumbling in the AI team

Apparently there is some unrest at Meta. According to a report, AI experts from the old Gade must fulfill new requirements. Specifically, this concerns the Fundamental Artificial Intelligence Research (FAIR) team based in Paris. This includes Yann LeCun, Turing Prize winner and Chief AI Scientist at Meta.

As The Information writes, the team members have to go through new, additional cross-reading processes before they are allowed to publish anything. Some scientists are said to have found this a hard pill to swallow—and as an encroachment on their freedom. The FAIR team has been around for just over ten years. Among other things, the Python framework PyTorch, which is used to create deep learning models and neural networks, was developed there. The well-known open AI model family Llama also has its origins in Paris, as does Dino, a foundation model for computer vision. Dino stands for Self-Distillation with no labels. V-Jepa aims to develop an understanding of the physical world instead of relying on language and labeled data.

All of these AI developments are freely available, at least in part. Llama is even declared open source by Meta, but the Open-Source Initiative does not consider the conditions for this to be met. For example, there is no information on the data used in the training. Nevertheless, anyone can use Llama and develop their own AI applications based on it. According to The Information, Meta now wants to put a stop to this availability. The FAIR scientists are to devote more energy to working on products and less to publishing their scientific results, as other companies could also benefit from this. However, a Meta spokesperson has denied this to The Information: “Research is one of the main pillars of Meta Superintelligence Labs; we have not prevented researchers from publishing.”

Disagreements about the posts too

Yann LeCun is also said to have been so upset that he considered leaving Meta. LeCun is a truly recognized AI expert. However, he also repeatedly says that he only has limited faith in the future of generative AI. Something that sounds very different to Mark Zuckerberg at the moment. LeCun's approach initially envisages Advanced Machine Intelligence (AMI), i.e., helpful machines. Instead of generative AI, he is focusing on the development of a world model.

The expensive purchases of AI experts and the establishment of the Super Intelligence team in the USA are also said to have caused frustration among the old employees. Among other things, Meta recently appointed former OpenAI scientist Shengjia Zhao as Chief Scientist. The Superintelligence Lab is headed by former Scale AI CEO Alexandr Wang and former Github CEO Nat Friedman.

Paris-based Head of AI Research Joëlle Pineau left Meta in April and “cleared the way for others who want to continue this work,” wrote Pineau, although it is not clear why she is leaving.

(emw)