بچه‌های متا اومدن یه مدل هوش مصنوعی خفن به اسم TRIBE v2 ساختن که می‌تونه ویدیو، صدا یا متن رو بگیره و دقیقاً پیش‌بینی کنه مغز انسان چه واکنشی بهشون نشون میده

🚨 Meta built an AI that can watch a video and predict how the human cortex would respond. TRIBE v2 is a new multimodal foundation model that takes video, audio, or language and predicts high-resolution fMRI activity across the cortical surface. It was developed using a large collection totaling more than 1,100 hours of fMRI data across 720 people. The model combines representations from pretrained video, audio, and language models, then maps them onto brain activity. It generalizes zero-shot to previously unseen stimuli, tasks, and subjects, and significantly outperformed an optimized linear encoding baseline across the evaluated datasets. Even more interesting, researchers used TRIBE v2 to recreate classic neuroscience experiments entirely in silico. Its predictions recovered known cortical responses associated with faces, places, bodies, written words, speech, and language processing. The public release includes the model weights, code, and an interactive demo. You can give the model media such as a video and generate predicted brain-response maps without collecting a new fMRI scan for that stimulus. The bigger shift: neuroscience is moving toward foundation models that can simulate how the human brain is expected to respond before an experiment is ever run. Caveat: this is an encoding model predicting population-level fMRI responses, not a system that reads thoughts or perfectly predicts an individual person's brain. The work is currently an arXiv preprint rather than a peer-reviewed journal article. Reference: https://ai.meta.com/blog/tribe-v2-brain-predictive-foundation-model/