Skip to main content

Meta Unveils TRIBE v2: AI Breakthrough in Predicting Human Brain Neural Responses

29 MARCH 2026·9 MIN READ·8 SOURCES

Meta's latest AI model, TRIBE v2, enables precise prediction of human brain responses to visual, auditory, and language stimuli, promising to revolutionize neuroscience research and AI-human interaction.

Meta Unveils TRIBE v2: AI Breakthrough in Predicting Human Brain Neural Responses

Key takeaways · 5

  • 01

    TRIBE v2 predicts neural responses to multimodal stimuli (images, sounds, text) across approximately 70,000 brain voxels.

  • 02

    Trained on over 500 hours of fMRI data from 700+ individuals exposed to varied media including podcasts, films, images, and text.

  • 03

    Supports zero-shot generalization enabling prediction for new subjects, languages, and experimental tasks without retraining.

  • 04

    Achieves around 89% accuracy in predicting viewed images, a significant improvement over previous models.

  • 05

    Open sourced with research paper, code, and interactive demo to facilitate neuroscience and AI research collaboration.

Understanding TRIBE v2

TRIBE v2 (Trimodal Brain Encoder version 2) is an AI model that predicts how human brains respond to sensory stimuli across vision, sound, and language. It creates a digital twin of neural activity by mapping stimulus features from multimedia sources to brain activity patterns recorded via fMRI.

Training and Data

Meta trained TRIBE v2 using over 500 hours of brain data from more than 700 volunteers exposed to diverse media while undergoing fMRI scans. This large-scale dataset enables the model to generalize neural response patterns across individuals.

Technological Advances

The model uses a transformer-based architecture that integrates pretrained embeddings from three modalities—audio, video, and text—into a universal brain representation. It predicts activity in around 70,000 brain voxels, a 70-fold improvement in spatial resolution over previous models.

Zero-Shot Prediction Capability

A key innovation of TRIBE v2 is its ability to predict neural responses for new subjects and tasks without retraining. This zero-shot capability dramatically reduces the need for collecting new brain scan data for each experiment.

TRIBE v2 marks a major leap in our ability to computationally model human brain activity with high precision and generalization. By enabling zero-shot predictions of neural responses to complex stimuli without the need for new brain imaging, it promises to accelerate research in neuroscience and cognitive science, facilitate development of assistive neurotechnology, and inspire AI systems that more closely emulate human brain function. The open-source release fosters collaboration and transparency, setting new standards for brain-AI integration technologies with wide societal implications.

Why it matters
Daily session

Put this to work — one session a day, built for your industry.

Create a free account for a daily session — eight questions and one real-work challenge, on the news that affects your role.

Start free

Sources

AI fluency, one session a day, built for your work.

Get the appStart free