Meta’s New AI Can Turn Brain Activity Into Written Sentences
TL;DR
Meta is testing Brain2QWERTYv2, a non-invasive brain-to-text system that uses MEG scanners instead of implanted electrodes. In controlled typing tasks, the system reportedly reached 61% average word accuracy, with the best participant hitting 78%. The setup is still narrow: few healthy participants, lab conditions, structured sentences and bulky MEG hardware. The source is heavy on overview and promise; independent replication, lower latency and wearable sensors are the real tests.
Nauti's Take
This is real progress, but not a product moment. Anyone expecting everyday sentence writing from MEG scans is skipping over hardware, data diversity, latency and privacy all at once.
The useful benchmark is sober: could this eventually help people who cannot speak or type, without requiring an implant? Until then, every big headline needs a clear lab-only label.
Briefingshow
Non-invasive brain-computer interfaces could be much safer than implants for people with severe communication barriers. But 61% word accuracy is not an everyday communication channel; it is a research result under favorable conditions. The important question is not mind-reading hype, but whether reliable assistive systems can work without surgery.
Video
Sources