Meta's Brain2Qwerty achieves 61% word accuracy for non-invasive brain signal decoding (heise.de)

• Brain2Qwerty v2 reaches 61% word accuracy from MEG, best participants 78%.
• Non-invasive method avoids surgical risks, but MEG devices are stationary.
• Meta releases code and models under open source license.

"Meta AI's Brain2Qwerty v2 can reconstruct typed sentences from non-invasive EEG and MEG brain signals with 61% word accuracy, according to a study in Nature. Best participants reached 78% accuracy. Unlike implanted brain-computer interfaces (BCIs), Brain2Qwerty requires no surgery, but MEG technology remains stationary and needs magnetically shielded rooms. Meta released code and models as open source."