Meta’s Post

Transcript

Today at Meta Fair, we're excited to share some of our most recent research models. As the year 2024 comes to a close, many of our researchers are attending the NURBS conference. Seems like a good time to be sharing with you some of the latest work we've been doing in the lab. First up, we have Meta Motivo, which is a first of a kind behavioral foundation model. It's designed to control the behavior of embodied virtual agents. It's trained with a new unsupervised reinforcement learning algorithm. It can actually handle a wide range of whole body tasks, a motion tracking pose reaching optimizing rewards using zero shot inference. Beyond that, we also have video seal, open source tool for invisible video watermarking. As general de AI advances, so do concerns about deep fakes, digital fraud. And so it's so it's important to have some tools to help us identify AI generated content. By open sourcing this, we're inviting the community to explore what's doable with the code contribute and help us develop really reliable tools. Beyond those, there's some models, some code, some data sets that speak to many of the research themes happening at FAIR. I encourage you to explore them, especially if you have an interest in new architectures, in agents, robustness, safety. We're excited to share this with the research community. It speaks to continued commitment to open research into building AI research with a broad set of collaborators.