【#Tech24H】Brain-computer interfaces fall into two categories: invasive and non-invasive. Simply put, invasive requires surgery, while non-invasive does not. Meta’s research team released Brain2Qwerty v2, currently the highest-performing end-to-end non-invasive brain-signal decoding system. It can decode complete sentences in real time from continuous brain activity, with accuracy approaching levels previously achievable only through invasive implanted BCIs. During training, the system collected data from 9 volunteers typing about 22,000 sentences. Each participant wore a magnetoencephalography (MEG) device and had about 10 hours of continuous brain activity recorded during active keyboard input. The results show that Brain2Qwerty v2 can recover semantically coherent full sentences from highly noisy brain signals. The model achieved an average word accuracy of 61%, a dramatic improvement over previous non-invasive methods, which typically reached only about 8%. For the best-performing participant, word accuracy hit 78%, with over half of the decoded sentences containing one or fewer word errors. [ By Zhang Liyan | Tang Ruohan ]