Meta has made a breakthrough in the field of neural interfaces, unveiling the Brain2Qwerty v2 system. This technology can reconstruct typed text based on brain activity recordings without any surgical intervention. The key metric—average word-level accuracy—reached 61%, dramatically surpassing previous non-invasive methods, where this figure was around 8%.

Training the model required approximately 22,000 sentences. Nine healthy volunteers participated in the experiment. Each spent 10 hours in a magnetoencephalography scanner while simultaneously typing heard phrases on a keyboard. This approach allowed for the collection of a unique dataset for training the neural network.

"Instead of manually crafted schemes for detecting neural events, we use end-to-end deep learning to decode data directly from raw brain signals," the developers explained.

The Brain2Qwerty v2 architecture processes a continuous recording of brain activity and reconstructs the entire sentence. To improve accuracy, the system employs a fine-tuned language model that analyzes semantic context. The results are impressive: the best participant in the experiment demonstrated 78% word-level accuracy. More than half of their sentences were decoded with one error or no errors at all.

Limitations and Prospects

It should be noted right away that clinical or everyday application of this technology is still far off. The experiment was conducted on healthy volunteers who actually typed text. For people who have lost speech or the ability to move, this approach will require separate validation and likely adaptation.

Additionally, the system does not yet work in real-time. The model processes the entire sentence, so the user does not see each word immediately after typing it. This is a significant limitation for practical use.

The project is part of the Digital Brain Project—an initiative to create open neuroscience datasets with a budget of $5 million. This is an important step toward democratizing research in this field.

My expert assessment: Meta's achievement marks a transition from laboratory experiments to potentially applied solutions. However, 61% accuracy is still far from a reliable interface, and the key challenge will remain in adapting the technology for patients with speech impairments. Nevertheless, the progress from 8% to 61% in a short period is impressive and points to exponential growth in the capabilities of neural interfaces.