Meta’s New AI Can Turn Brain Activity Into Written Sentences
TL;DR
Meta’s Brain2QWERTYv2 decodes brain activity into sentences using MEG, without a surgical implant. It reads external magnetic signals during controlled typing tasks. Reported tests reached 61 percent average word accuracy, with the best participant at 78 percent. That is still far from reliable everyday communication. The setup is narrow: few healthy participants, lab conditions, and structured training data. Signals from people with severe communication impairments would likely be much noisier.
Nauti's Take
The headline sounds like mind reading, but the data is narrower: Meta decodes signals from a tightly controlled typing task. That is more scientifically useful than the hype, and far less magical.
The PR-heavy part is the leap toward daily use and independence. The real test is practical: can this kind of system handle incomplete, slow, noisy signals when someone actually needs it to communicate?
Briefingshow
The meaningful step is the non-invasive path: brain-computer interfaces would not have to start with surgery. The 61 percent accuracy also shows how wide the gap remains between a lab demo and a useful assistive device. For patients, this only matters once hardware, latency, and error correction work in real communication settings.
Video
Sources