Researchers have unveiled Brain2Qwerty v2, a breakthrough non-invasive brain-computer interface that translates brain activity into text with unprecedented accuracy. Unlike traditional surgical implants, this system uses magnetoencephalography (MEG) to capture neural signals, achieving 61% word accuracy in real-time sentence decoding. The approach could transform communication for millions living with brain injuries.

Brain2Qwerty v2 employs end-to-end deep learning to decode directly from raw brain signals, eliminating the need for hand-crafted neural event detection pipelines. The system was trained on approximately 22,000 sentences from nine volunteer participants wearing MEG devices for 10 hours each. By fine-tuning large language models on neural data, the system leverages semantic context to bridge the gap between noisy brain recordings and coherent language output.

"The results are striking: 61% word accuracy significantly outperforms the 8% achieved by other non-invasive methods. For the best participant, accuracy reached 78%, with over half of all sentences decoded with one word error or less. Researchers open-sourced the training code and partnered with the Basque Center on Cognition, Brain, and Language to release the v1 dataset, accelerating neuroscience breakthroughs.

This work contributes to efforts building open foundational models of the brain, including the Tribev2 model for perception encoding and Neural Set for processing brain data at scale. The team recently launched a $5 million fund through their Digital Brain Project to stimulate open datasets, hoping to advance neuroscience research beyond isolated efforts.