Meta has released Brain2Qwerty v2, a non-invasive brain-to-text system that decodes brain activity into typed sentences using magnetoencephalography (MEG) and end-to-end deep learning. Trained on ~22,000 sentences from nine participants, it achieves 61% word accuracy — a major leap over the 8% from prior non-invasive methods — and 78% for the best participant. The system fine-tunes large language models on neural data to handle noisy brain signals. Accuracy scales log-linearly with data volume, suggesting surgical-level performance may be achievable without implants. Meta is open-sourcing the training code for both v1 and v2, and partner BCBL is releasing the v1 dataset.
479 Impressions