Meta's brain-scanning AI now decodes full sentences

Meta has introduced a new version of its Brain2Qwerty system that can turn noninvasive brain scans into full words and sentences, rather than decoding text one character at a time. The company said the model was trained on data from nine volunteers who spent about 10 hours each in a brain scanner while typing, creating nearly 22,000 sentences for the system to learn from. Meta reported average word accuracy of 61%, with its best result reaching 78% for one participant. Meta says the new approach narrows the gap with systems that rely on surgical implants and could eventually support communication tools for people who have lost speech. The company also published the code and dataset for both versions of the model, which could help other researchers build on the work.