Mark Zuckerberg Tests Meta's AI Model On Old Video Of Him Playing Guitar

Mark Zuckerberg, the co-founder, and CEO of Meta (formerly Facebook) recently showed Meta’s latest artificial intelligence (AI) model by sharing an older video of himself playing the guitar and singing for his daughter, Maxima. The video was posted on Instagram, where Zuckerberg shared a throwback clip of him playing a song on the guitar. In the caption, he mentioned testing the video on the AI model named V-JEPA, described on its site as a “non-generative model that learns by predicting missing or masked parts of a video in an abstract representation space.”

Zuckerberg’s caption read, “Throwback to singing one of Max’s favorite songs. I recently tested this video with a new AI model that learns about the world by watching videos. Without being trained to do this, our AI model predicted my hand motion as I strummed chords. Swipe to see the results.”

The post included two videos. The first showed him singing and playing the guitar with Maxima, while the second demonstrated the AI model’s prediction of his hand movements while playing.

Watch the video here:

A post shared by Mark Zuckerberg (@zuck)

Despite being posted just a day ago, the video has already garnered over 53,000 likes.

V-JEPA, or Video Joint Embedding Predictive Architecture, is a predictive analysis model that learns solely from visual media. It not only understands the content of a video but can also predict what will come next.

To train V-JEPA, Meta used a new masking technology. This involved masking parts of the video in both time and space. Some frames were entirely removed, while others had blacked-out fragments. This process forced the model to predict both the current and next frames. Meta claims the model performed this task efficiently. Importantly, the model can analyse videos up to 10 seconds in length.

In a blog post, Meta explained, “For example, if the model needs to distinguish between someone putting down a pen, picking up a pen, and pretending to put down a pen but not actually doing it, V-JEPA is quite good compared to previous methods for that high-grade action recognition task.”

The use of V-JEPA highlights Meta’s ongoing efforts to innovate in the field of AI and machine learning.