The Story

Meta AI has introduced TRIBE v2, a foundation model designed to simulate human brain activity across multiple modalities including vision, audition, and language.

According to the official research publication from Meta AI, the model is trained on large-scale brain imaging data and aims to enable in-silico neuroscience — conducting experiments on virtual brain models instead of relying solely on physical scans.

What TRIBE v2 does

• Brain activity simulation: The model predicts neural responses to images, sounds, and language inputs.
• Multimodal understanding: It connects different sensory and cognitive domains within a unified system.
• Virtual experimentation: Researchers can simulate brain responses without requiring new fMRI scans.

Scale and improvements

• Large dataset: Trained on over 1,000 hours of brain data from 700+ participants.
• Expanded coverage: The model scales from around 1,000 brain regions to tens of thousands.
• Higher fidelity predictions: Meta reports that TRIBE v2 can match or exceed the quality of real fMRI recordings at population level.

What researchers can do

• Replicate experiments: Recreate known neuroscience findings digitally.
• Test hypotheses faster: Run experiments without expensive scanning infrastructure.
• Explore cognition: Study how the brain processes language, sound, and visual inputs in a unified way.

Why it matters

Neuroscience research has traditionally been limited by the cost and complexity of collecting brain data. Models like TRIBE v2 could significantly accelerate discovery by shifting part of the experimentation process into software.

The release also highlights a deeper convergence between AI and neuroscience, where advances in one field increasingly inform and accelerate progress in the other.