Brain Reads Thoughts
Researchers have made a breakthrough in non-invasive EEG-neurointerfaces, discovering that they can read specific words, including rare ones, from an open dictionary. A massive dataset was collected, consisting of around 240,000 words read by one person over 49 hours in 393 separate sessions. The study utilized a 19-channel dry EEG, eliminating the need for gel, surgery, or invasive sensors.
The words were displayed in a rapid sequential presentation, with the font changing each time to prevent the brain from "guessing" the answer based on visual form. The model consisted of two parts: a convolutional EEG encoder and a causal transformer, trained using a contrastive scheme similar to CLIP. The system learned to associate brain activity with semantic and lexical features of words, as described in Nature Aging, July 2026.
The accuracy was measured as the top-10 hit rate and was consistently above the random level, including words with medium and low frequency. The quality improved log-linearly with the amount of data and did not reach saturation, meaning that the more data, the better the decoding. Removing occipital and parietal electrodes reduced accuracy by about a third but did not affect the model's ability to track text context. Control experiments showed that the model actually recognizes words, rather than just guessing based on position or context.
🔗 Read original →
arXiv.org
Decoding silent reading from non-invasive EEG
Non-invasive decoding of inner speech faces a fundamental data problem: a corpus pairing brain activity with a person's spontaneous inner monologue cannot be collected, and the available proxy...
August 25, 2026 7
