Overview
- Meta announced Brain2Qwerty v2 on Monday, June 29, and said the system reached an average word accuracy of 61 percent and 78 percent for the best participant after training on roughly 22,000 sentences from nine volunteers.
- The system uses magnetoencephalography (MEG) — a helmet‑style array that reads tiny magnetic fields outside the skull — and an end‑to‑end AI pipeline that combines convolutional networks, transformers, and fine‑tuned large language models to turn brain activity into words.
- Meta will release the v1 and v2 training code and its research partner will publish the v1 dataset as part of the Digital Brain Project, which includes a $5 million fund to build open neuroscience datasets.
- The experiments decoded memorized typing and motor planning rather than spontaneous thoughts, and practical limits remain because MEG equipment is large, costly, and tied to research labs, so the system is not yet clinically or commercially ready.
- The work narrows the gap with implant‑based BCIs by boosting non‑invasive accuracy, but further gains will depend on more data, sensor miniaturization, real‑time engineering and clinical validation before this can restore everyday communication for patients.