Meta’s New AI Can Turn Brain Activity Into Written Sentences
TL;DR
Meta has shown Brain2Qwerty v2, a non-invasive brain-computer interface that turns MEG brain signals into typed sentences. The new version works from continuous brain recordings instead of relying on known keypress timing, with modules for letters, words and sentences. Meta reports 61 percent average word accuracy and up to 78 percent for the best participant, after training on roughly ten times more data per person than v1.
Nauti's Take
The headline sounds like mind reading, but the actual setup is narrower: trained participants, controlled typing tasks and massive MEG hardware. Still, this is a meaningful marker.
If non-invasive BCIs keep scaling, communication tools for paralyzed or non-speaking patients could become less risky. Meta should not frame this as consumer magic.
Brain data needs privacy rules long before the device becomes small enough for daily life.
Briefingshow
The work matters because it points to a possible alternative to brain implants, which remain risky and invasive for many patients. At the same time, it shows how much is still missing: better decoding, smaller sensors and serious rules for brain data before this becomes anything close to a product.
Video
Sources