Meta AI

Meta AI is a research division of Meta (formerly Facebook) that develops artificial intelligence and augmented reality technologies.

Meta AI

Company type	Division
Industry	Artificial intelligence
Founded	December 11, 2015; 9 years ago (2015-12-11)
Founders	Yann LeCun Mark Zuckerberg Rob Fergus
Headquarters	Astor Place, New York City, New York, U.S.
Key people	Joëlle Pineau (managing director) Yann LeCun (chief scientist)
Products	Llama
Owner	Meta Platforms
Website	ai.meta.com

History

The group was founded in 2013 as Facebook Artificial Intelligence Research (FAIR) with Yann LeCun as its director. Vladimir Vapnik joined the group in 2014.^[1][2] It has workspaces in Menlo Park, California, London, United Kingdom, and Manhattan.^[3]

The group subsequently opened research labs in Paris, France, Seattle, Pittsburgh, Tel Aviv, Montreal and London.^[4][5] In 2016, FAIR partnered with Google, Amazon, IBM, and Microsoft in creating the Partnership on Artificial Intelligence to Benefit People and Society.

In 2018, Jérôme Pesenti, former CTO of IBM's big data group, assumed the role of president of FAIR, while LeCun stepped down to serve as chief AI scientist.^[6] FAIR had approximately 200 staff in 2018.^[7]

FAIR's research includes self-supervised learning, generative adversarial networks, document classification and translation, and computer vision.^[8] FAIR released Torch deep-learning modules as well as PyTorch in 2017, an open-source machine learning framework,^[8] which was subsequently used in several deep learning technologies, such as Tesla's autopilot ^[9] and Uber's Pyro.^[10] That same year, a pair of chatbots were falsely rumored^[11] to be discontinued for developing a language that was unintelligible to humans.^[12] FAIR clarified that the research had been shut down because they had accomplished their initial goal to understand how languages are generated by their models, rather than out of fear.^[11]

FAIR was renamed Meta AI following the rebranding that changed Facebook, Inc. to Meta Platforms Inc.^[13]

Virtual assistant

See also: Virtual assistant

Meta AI is also the name of the virtual assistant developed by the team, now integrated as a chatbot into Meta's social networking products.^[14] It is also available as a subscription-based stand-alone app.^[15][16]

The virtual assistant was pre-installed on the second generation of Ray-Ban Meta smartglasses, and can incorporate inputs from the glasses' cameras after an update.^[17] It is also available on Quest 2 and newer HMDs.^[18]

Since May 2024, the chatbot has summarized news from various outlets without linking directly to original articles, including in Canada, where news links are banned on its platforms. This use of news content without compensation and attribution has raised ethical and legal concerns, especially as Meta continues to reduce news visibility on its platforms.^[19]

Current research

Natural language processing and chatbot

See also: Natural language processing

Meta AI works on machines' ability to understand and generate natural language. The team also seeks to allow their chatbots to communicate multilingually.^[20] This involves the generalization of natural language processing (NLP) technology to other languages, and the team actively works on unsupervised machine translation.^[21][22]

Galactica

Galactica is a large language model (LLM) designed for generating scientific text. It was available for three days since 15 November 2022, before being withdrew from service for generating racist and inaccurate contents.^[23][24]

Llama

Main article: Llama (language model)

LLaMA is a LLM released in February 2023, supporting 7B to 65B parameters.^[25] Two of the three Llama 4 models, Scout and Maverick, was released on April 5, 2025, with the biggest model, Behemoth, still in training.^[26]

Hardware

Meta uses CPUs and in-house custom chips until 2022, when they switched to Nvidia GPUs. Several data centers were redesigned to accommodate for the larger network bandwidth and cooling requirements.^[27]

MTIA v1

Meta developed the training and inference accelerator, MTIA v1, specifically for their content recommendation workloads. It was fabricated on TSMC's 7 nm process technology and operates at a frequency of 800 MHz. The accelerator provides 51.2 TFLOPS at FP16 precision, with a thermal design power (TDP) of 25 W.^[28]