Jump to content
[dismiss]
Molecupedia is a parody encyclopedia written by Molecular Reality Corporation. It is not Wikipedia and is not affiliated with the Wikimedia Foundation. About

Meta AI

Page protected
From Molecupedia, the encyclopedia of very small things
Artificial intelligence research laboratory and assistant of Meta Platforms
Meta AI
TypeDivision of Meta Platforms
IndustryArtificial intelligence
FoundedDecember 2013, as Facebook AI Research
HeadquartersMenlo Park, California
ProductsPyTorch, ESM, Segment Anything, I-JEPA, V-JEPA
Websiteai.meta.com

Meta AI is the artificial intelligence organization of Meta Platforms. The name covers the research laboratory that Facebook announced in December 2013, under the New York University professor Yann LeCun, and the assistant of the same name that the company builds into its apps.[1][2] The laboratory is known by its initials, FAIR, which stood first for Facebook AI Research and now stand for fundamental AI research.[3]

Its output includes the deep learning library PyTorch, a series of large language models, a protein language model that predicted structures for more than 617 million sequences, and the joint embedding architectures that LeCun proposed in 2022. Molecular Reality Corporation has chosen the last of these as the training recipe for a planned model of nanopore signals.[4][5][6][7]

History

Facebook AI Research

On December 9, 2013, TechCrunch reported that Facebook had named LeCun to lead a new artificial intelligence laboratory, with research split between the company's headquarters in Menlo Park, its London office and a new site a block from the NYU campus in Manhattan.[1] LeCun, a pioneer of convolutional networks, directed FAIR for its first five years and was the company's chief AI scientist for the next seven.[8] In October 2021 Facebook, Inc. renamed itself Meta Platforms.[9] Papers from the laboratory in 2023 list the affiliation as "FAIR, Meta AI".[5]

Software and models

PyTorch, a deep learning library written in Python, began tagged releases on GitHub in early 2017 and was described in a paper in 2019.[4] On September 12, 2022, Meta moved it to a new PyTorch Foundation, part of the Linux Foundation, and said it would keep using it as its main framework for AI research.[10]

The Llama language models arrived in 2023. The first paper, in February, described them as "open and efficient"; a second generation followed in July of that year, a third in 2024 and a fourth in April 2025.[11][12][13][14] In April 2023 the laboratory also released Segment Anything, a model that cuts objects out of images, and DINOv2, a self-supervised vision model.[15][16]

Proteins

A group at FAIR in New York trained language models on protein sequences, reporting in 2021 that "biological structure and function emerge from scaling unsupervised learning" to 250 million sequences.[17] In a paper of 2023 the largest model, with 15 billion parameters, predicted atomic structures directly from sequence, and the authors used it to build the ESM Metagenomic Atlas: predicted structures for more than 617 million metagenomic proteins, of which more than 225 million were predicted with high confidence.[5] For comparison, the Protein Data Bank holds 261,546 structures that someone measured. Seven of the 15 authors of the 2023 paper are authors of ESM3, published in Science in February 2025 under the affiliation EvolutionaryScale, a company in New York.[18] See EvolutionaryScale.

Joint embeddings

LeCun's position paper "A path towards autonomous machine intelligence", posted on June 27, 2022, proposed learning a model of the world by predicting representations instead of pixels.[6] The group's earlier regularization method, VICReg (2021), had addressed how to keep such representations from collapsing to a constant.[19] I-JEPA (January 2023), V-JEPA (February 2024) and V-JEPA 2 (June 2025) put the idea to work on images and video.[20][21][22] The architecture has an article of its own: Joint embedding predictive architecture.

Reorganization

On April 29, 2025, the day of its first LlamaCon developer event, Meta released a standalone Meta AI app.[2] On June 30 Mark Zuckerberg announced Meta Superintelligence Labs, a unit that combined the company's AI model and product teams with FAIR, led by a new chief AI officer; LeCun stayed on as chief AI scientist.[3] On November 20 the news site The Decoder reported that LeCun would leave at the end of the year, after twelve years, to start a company around advanced machine intelligence, with Meta as a partner.[23] TechCrunch described the resulting company, AMI Labs, in January 2026.[24]

At Molecular Reality

Molecular Reality takes the laboratory's training recipe and, for one component, its model. The company's essay on a nanopore world model, written in the voice of the puppet Keltar, assigns the JEPA recipe to the ionic current and gives V-JEPA a smaller job: a frozen extractor for microscope and macro video, demoted to a "NUISANCE encoder" because, as the essay's diagram says, it "sees fouling, syringe, bubbles, NOT molecules".[7] The essay also warns that a home-made JEPA for current traces does not get V-JEPA's slowly updated teacher network for free.[7] The company's guide files the model under plans: "The model itself is still ahead of us."[25]

See also

References

  1. ^ a b Josh Constine (December 9, 2013). "NYU "Deep Learning" Professor LeCun Will Head Facebook's New Artificial Intelligence Lab". TechCrunch.
  2. ^ a b "Meta launches AI app competing with OpenAI and Google". CNN. April 29, 2025.
  3. ^ a b "Mark Zuckerberg reveals Meta Superintelligence Labs, names who he poached from OpenAI, Google, Anthropic". Entrepreneur. June 30, 2025.
  4. ^ a b Paszke, Adam et al. (2019). "PyTorch: an imperative style, high-performance deep learning library". arXiv. doi:10.48550/arXiv.1912.01703.
  5. ^ a b c Lin, Zeming et al. (2023). "Evolutionary-scale prediction of atomic-level protein structure with a language model". Science 379 (6637): 1123-1130. doi:10.1126/science.ade2574.
  6. ^ a b LeCun, Yann (June 27, 2022). "A path towards autonomous machine intelligence, version 0.9.2". OpenReview.
  7. ^ a b c "Keltar's Critique v2: The Nanopore World Model (Hardened)". Molecular Reality Corporation. 2026.
  8. ^ "Yann LeCun confirms exit from Meta, sets sights on new advanced AI venture". Storyboard18. November 20, 2025.
  9. ^ "The Facebook company is now Meta". Meta. October 28, 2021.
  10. ^ "Announcing the PyTorch Foundation to accelerate progress in AI research". Meta. September 12, 2022.
  11. ^ Touvron, Hugo et al. (2023). "LLaMA: open and efficient foundation language models". arXiv. doi:10.48550/arXiv.2302.13971.
  12. ^ Touvron, Hugo et al. (2023). "Llama 2: open foundation and fine-tuned chat models". arXiv. doi:10.48550/arXiv.2307.09288.
  13. ^ Grattafiori, Aaron et al. (2024). "The Llama 3 herd of models". arXiv. doi:10.48550/arXiv.2407.21783.
  14. ^ "The Llama 4 herd: the beginning of a new era of natively multimodal AI innovation". Meta. April 5, 2025.
  15. ^ Kirillov, Alexander et al. (2023). "Segment anything". arXiv. doi:10.48550/arXiv.2304.02643.
  16. ^ Oquab, Maxime et al. (2023). "DINOv2: learning robust visual features without supervision". arXiv. doi:10.48550/arXiv.2304.07193.
  17. ^ Rives, Alexander et al. (2021). "Biological structure and function emerge from scaling unsupervised learning to 250 million protein sequences". Proceedings of the National Academy of Sciences 118 (15). doi:10.1073/pnas.2016239118.
  18. ^ Hayes, Thomas et al. (2025). "Simulating 500 million years of evolution with a language model". Science 387 (6736): 850-858. doi:10.1126/science.ads0018.
  19. ^ Bardes, Adrien; Ponce, Jean; LeCun, Yann (2021). "VICReg: variance-invariance-covariance regularization for self-supervised learning". arXiv. doi:10.48550/arXiv.2105.04906.
  20. ^ Assran, Mahmoud et al. (2023). "Self-supervised learning from images with a joint-embedding predictive architecture". arXiv. doi:10.48550/arXiv.2301.08243.
  21. ^ Bardes, Adrien et al. (2024). "Revisiting feature prediction for learning visual representations from video". arXiv. doi:10.48550/arXiv.2404.08471.
  22. ^ Assran, Mido et al. (2025). "V-JEPA 2: self-supervised video models enable understanding, prediction and planning". arXiv. doi:10.48550/arXiv.2506.09985.
  23. ^ "Yann LeCun leaves Meta to launch new AI startup". The Decoder. November 20, 2025.
  24. ^ Anna Heim (January 23, 2026). "Who's behind AMI Labs, Yann LeCun's 'world model' startup". TechCrunch.
  25. ^ "Keltar's Field Guide to Molecular Reality". Molecular Reality Corporation. August 2026.
Retrieved from "https://molecularreality.com/wiki/Meta_AI"
Last edited 1 hour ago by Ionophore