🏆 We're thrilled to announce that Meta FAIR’s Brain & AI team won 1st place at the prestigious Algonauts 2025 brain modeling competition. Their 1B parameter model, TRIBE (Trimodal Brain Encoder), is the first deep neural network trained to predict brain responses to stimuli across multiple modalities, cortical areas, and individuals. The approach combines pretrained representations of several foundational models from Meta – text (Llama 3.2), audio (Wav2Vec2-BERT from Seamless) and video (V-JEPA 2) – to predict a very large amount (80 hours per subject) of spatio-temporal fMRI brain responses to movies acquired by the Courtois NeuroMod project Download the code: github.com/facebookresear… Read the paper: arxiv.org/abs/2507.22229 Learn about the challenge: algonautsproject.com/index.html Download the data: cneuromod.ca
@AIatMeta so glad you're poaching the worlds top researchers to make people click more ads
@AIatMeta for folks freaking out about the model being able to read your thought without scanning you chill you will need to get jammed into an fmri for 72h while video of sitcoms are playing and the resulting data used in retraining the model for it to work chillllll it’s just science
@AIatMeta This is so sick ~ in the bad connotation of the word. Because world class AI talent should def be used to get people to click more ads and spend more time on their phones.
@AIatMeta Congratulations team. TRIBE in Meta glasses and Quest VR would be incredible and enhance user experiences multi fold
@AIatMeta Ad targeting is going to go crazy when Meta maps my dopamine hits to certain ad formats
@AIatMeta Congrats, and thanks for sharing all the details, totally fascinating 🧠
@AIatMeta Is this a brag about enslaving people more to their devices? 🤣
@AIatMeta Turns out 1B model is all you need to build an AI that can predict your brain’s response to a movie without scanning you at all 🤷♂️ So much for free will 😢
@AIatMeta Wrap this thing in homomorphic encryption for privacy and computability and you've got a decent platform for persona uploading.
@AIatMeta Wow great way to know how to manipulate people. Very on brand for you guys.
@AIatMeta Is there an above 0 chance that some deeply secretive program has been doing all this AI stuff for years and is decades ahead?
Here's my interpretation from a neural sci perspective: This method essentially uses multiple embedding towers eg. Audio, video and text transformer blocks to embed the input, fused the embeddings latently and conducted regression on subject area activation, as recovered by fmri(a type of imaging technique revealing real time brain activity through o2 consumption) A couple of drawbacks: brains are sparse, recurring NNs, the system partially simulated the sparsity through modality dropout but it's a hard pass to say it simulated the recurrence. Secondly brain connectomes vary and the exact neuronal activations sometimes drift from the classical area and the activation may be inaccurate due to limited fmri resolution. Overall, this study reveals that there might be a huge correlation between how the brain embeds the world vs how multi modal models embed the world. But a lot more research might be needed in order to actually simulate how a brain perceived the world in action
@AIatMeta sooo, more clicks? what a great contribution to humanity
@AIatMeta One step closer to non invasive neuralink and another step away from voice to text and keyboard interfaces… come on thought to text technology🤞
@AIatMeta They're gonna use to A/B test ads right?
@AIatMeta So you can ruin peoples lives more effectively? No one wants meta to be the people working on this. You people have destroyed billions of brains. Half your company should be in jail for the harm it has caused.
Congrats! 🧠 This kind of brain modeling is a huge step toward truly understanding consciousness. Imagine combining this with immersive symbolic worlds — direct emotional states, archetypal journeys, and real-time neural feedback. The future won’t just be sci-fi… it will be felt. emotionalpatching.io
@AIatMeta How many apps per second can you spy on with this model? Does it come with zero privacy built in?
@AIatMeta I'm not convinced this is big news. In fact...I'm fairly convinced this is nothing. No surprise there are predictable patterns in normalized fMRI responses. A specific workflow tends to activate a functional brain region...no surprise. I'm a little surprised it took 1B params.
@AIatMeta I am very curious about he activated patterns when we listen to different kinds of music
@AIatMeta I look forward to seeing where this takes us. I truly believe the answers to our next leap in AI will come from uncovering the secrets of the brain.
@AIatMeta Terrifying! More power in the worst hands. 😭 More ways to brainwash people with your crap and manipulate us. Meta is the worst for humanity!
@AIatMeta All of this for better ad targeting?😭
@AIatMeta When you goWhen are you going to do a human review or return the accounts you have unfairly disabled?
@AIatMeta This is notable not just for the win, but for the method fusing text, audio, and video foundation models to predict spatio-temporal brain responses could open new paths for brain–AI interfaces and neuroscience-driven model alignment.