Meta’s New AI Can Turn Brain Activity Into Written Sentences
TL;DR
Meta’s Brain2QWERTYv2 uses non-invasive MEG scanning to decode brain activity from controlled typing tasks and turn those signals into written sentences. The reported lab results show 61% average word accuracy, with the best participant reaching 78%. That is progress, but still too error-prone and slow for everyday communication. The experiment involved nine healthy participants in tightly controlled conditions. Current MEG scanners are large, expensive and far from wearable.
Nauti's Take
This is not an everyday mind reader, but a solid lab advance with a long road to practice. Meta shows that non-invasive signals can carry more information than earlier systems, but that does not equal a usable communication device.
The hard part starts after the demo: impaired users, noisy settings, affordable sensors and clear rights over neural data. Without those pieces, Brain2QWERTYv2 is a strong research signal, not a product.
Briefingshow
Non-invasive brain-computer interfaces could matter for people who cannot speak or type because they avoid brain surgery. The numbers also show the limit: 61% word accuracy sounds impressive, but would create too many mistakes in real conversations. The bottleneck is not only the model, but data quality, hardware and regulation.
Video
Sources