December 13th, 2024

New Meta FAIR Research and Models

Meta FAIR has released new research artifacts, including Meta Motivo for humanoid agents, Meta Video Seal for video watermarking, and frameworks like Flow Matching and Large Concept Model to enhance AI capabilities.

Read original articleLink Icon
New Meta FAIR Research and Models

Meta FAIR has announced the release of several new research artifacts aimed at advancing machine intelligence. Key innovations include Meta Motivo, a foundation model for controlling virtual agents, and Meta Video Seal, an open-source model for video watermarking. These releases are part of Meta's commitment to democratizing access to cutting-edge technologies and fostering collaboration within the research community. Meta Motivo utilizes unsupervised reinforcement learning to enable humanoid agents to perform complex tasks with human-like behaviors, demonstrating robustness to environmental changes. Meta Video Seal provides a framework for embedding imperceptible watermarks in videos, enhancing content traceability and security against manipulation. Additionally, Meta introduced Flow Matching, a generative AI framework that improves performance across various modalities, and the Large Concept Model (LCM), which decouples reasoning from language representation for better hierarchical thinking. Other notable releases include the Dynamic Byte Latent Transformer for tokenizer-free processing, Memory Layers for improved factuality in language models, and an evaluation toolbox for text-to-image generative models. These initiatives reflect Meta's ongoing efforts to promote responsible AI development and encourage community engagement in advancing AI technologies.

- Meta FAIR has released new models and datasets to enhance machine intelligence.

- Meta Motivo enables humanoid agents to perform tasks with human-like behaviors.

- Meta Video Seal offers a watermarking solution for video content security.

- Flow Matching and Large Concept Models improve generative AI capabilities.

- Meta emphasizes collaboration and responsible AI development in its research initiatives.

Link Icon 15 comments
By @cube2222 - 2 months
There’s honestly so much interesting stuff here, esp. the llm-related things - large concept models (operating on and predicting concepts, not tokens), dynamic byte latent transformers (byte-level alternative to standard tokenization), sparse memory layers (successfully scaling key-value memory layers without an increase in computational requirements).

Here they are presented as separate things, each of which apparently improves quality / efficiency. I wonder what the quality / efficiency increase is of all those methods put together? Maybe that’s what Llama 4 will be?

This looks like a lot of innovation is happening at Meta in those areas, really cool!

By @airstrike - 2 months
This is so cool! Playing around with the first demo is a lot of fun. First one to get the model to moonwalk wins. My best attempt was probably something like `(body_speed_forward < -0.3) * (head_height > 1.0) * (stay_still > 0.2) * (body_speed_vertical < 0.1) * (stay_upright > 0.9)`

https://i.imgur.com/O5hGMo5.gif

Then the "Meta Explore Theory of Mind" is even more interesting. There was a thread about a month ago in which some of us were discussing some of the concepts here like "beliefs" and updating a model of the world accordingly. https://news.ycombinator.com/item?id=42035985

By @modeless - 2 months
I really hope Dynamic Byte Latent Transformers work out. Death to tokenizers!

Interesting that it's a a hierarchical structure but only two levels of hierarchy. Stacking more levels seems like an obvious direction for further research.

By @ks2048 - 2 months
When I wonder about the business behind Meta doing this, I see they have $70B in cash, so giving a bunch of AI experts hundreds of millions is pocket change.
By @mtkd - 2 months
I was fortunate to get to a talk by Ross Taylor ex-Meta recently at the AI Engineer London meetup

He's recorded the full talk here now: https://www.youtube.com/watch?v=S5l5OvJ01ws

I had missed how much Meta have been doing on reasoning, ToM etc.

By @intalentive - 2 months
Every time I have to clean text I wonder why I haven’t just trained a byte level denoising autoencoder to handle it for me.
By @puttycat - 2 months
Can someone explain how watermarking AI videos voluntarily helps make AI safer?
By @bbor - 2 months
Crazy stuff. Everyone’s covering how exciting all these are (especially LCM and the non-tokenizing-tokenizer), but I have to ask in case anyone’s been paying attention: why are they using the term “advanced machine intelligence”?

My initial thought is that they want to please/distract the doomers, but I’m prolly just self-centered!

By @pkkkzip - 2 months
meta has certainly redeemed itself and helping AI become moat-free
By @SpaceManNabs - 2 months
This is like learning 10 different new architectures lol
By @Flomolok - 2 months
It's not a hype when it's delivers and I'm also not seeing a ceiling yet

Yet again interesting progress.

Also I like the idea of using the pose model to generate not a NPC but a avatar living in my phone or glas cube as a hologram. That would be quite scifi futuristic

By @nurumaik - 2 months
By @Roccan - 2 months
Meta's "Video Seal": Because nothing says "trustworthy" like a digital chastity belt. Imperceptible, they claim, yet robust enough to survive the gauntlet of internet mangling - sounds like the perfect tool to invisibly track content, not just watermark it.