Meta’s Brain2Qwerty v2 raises non-invasive sentence decoding to 61% word accuracy, with the best participant reaching 78%. The project used about 22,000 MEG-recorded typing sentences from nine volunteers and ships with training code plus a public v1 dataset.
Microsoft Research, UC Berkeley, UCSF, and Columbia researchers introduced generative causal testing. The method converts black-box language-brain prediction models into short hypotheses, then tests them by having an LLM write stories meant to activate targeted cortical regions.
UCLA researchers have identified DDL-920, the first drug to fully reproduce the effects of physical stroke rehabilitation in model mice. The findings, published in Nature Communications, could transform stroke recovery into a pharmacological option.
Meta said on March 26, 2026 that TRIBE v2 is a foundation model for predicting human brain responses to sight, sound, and language. The supporting paper and demo highlight zero-shot generalization, prediction across 70,000 voxels, and public releases of the paper, code, and model weights.
Meta said on March 26, 2026 that TRIBE v2 can predict high-resolution fMRI brain activity with zero-shot generalization across new subjects, languages, and tasks. The company is also releasing the model, code, paper, and demo for researchers.