Meta has unveiled a new AI system, Brain2Qwerty v2, that can convert brain activity into text without surgical implants. The system was trained on about 22,000 sentences from nine volunteers who wore a magnetoencephalography (MEG) device while typing. Brain2Qwerty v2 uses end-to-end deep learning to decode language directly from raw brain signals. It has achieved a word accuracy rate of 61%, a leap from the 8% word accuracy of other non-invasive methods. For the best-performing participant, this number jumped to 78%, with over half of all decoded sentences having one word error or less. Meta is making the complete training code for Brain2Qwerty v1 and v2 publicly available, and its research partner, the Basque Center on Cognition, Brain and Language (BCBL), will also make the Brain2Qwerty v1 dataset public.