Meta has launched TRIBE v2, an innovative open model designed to anticipate how the human brain reacts to various stimuli, including text, sound, and video. This model functions as a digital representation of perception; when provided with input such as a video segment or a phrase, it predicts which regions of the cerebral cortex will become active. TRIBE v2 was developed using over 1,000 hours of MRI recordings from more than 720 individuals. Remarkably, the AI is capable of generating an averaged model of brain activity even for individuals it has not encountered during its training sessions. By fine-tuning the model with just one hour of MRI data from a specific individual, it can produce a unique "snapshot" of that person's brain activity. The developers claim that the model can forecast activity across approximately 20,000 points within the cortex. Notably, the correlation between the model's predictions and actual brain activity stands at around 0.21, which is considered quite acceptable. It's important to note that functional MRI indirectly measures brain activity by tracking changes in blood flow, resulting in some inherent noise due to factors like blood movement and patient breathing. Consequently, the correlation between true brain dynamics and the MRI output can never be completely precise. While we are still a long way from genuine "mind reading," this advancement marks a significant milestone in enhancing our understanding of how the brain processes visual, auditory, and textual information.
Informational material. 18+.
