Meta AI described Brain2Qwerty v2, a system that decodes typed text directly from non-invasive magnetoencephalography (MEG) brain recordings using end-to-end deep learning rather than hand-crafted feature extraction. The approach fine-tunes large language models on neural signal data to exploit semantic/linguistic context and bridge the gap between noisy brain recordings and coherent text. Across nine participants, the system achieved 61% average word accuracy (78% for the best-performing participant), substantially outperforming prior non-invasive decoding approaches that top out around 8%. Meta suggests further gains are likely achievable simply by scaling training data.
