Naomi Bashkansky, a former alignment researcher at OpenAI, left the company on July 23 and joined Conduit the next day to develop models converting neural signals into text prompts for AI agents. She predicts that a headband could decode rough intentions into prompts for an AI coding agent by 2027, with two-way read-and-write technology by 2035. Conduit has gathered roughly 10,000 hours of neuro-language data from thousands of participants but has not yet published aggregate performance on free-form thought decoding. Meta achieved 61% average word accuracy with its Brain2Qwerty system, while a study published in Nature Communications with 723 participants only reached 20% top-1 accuracy on a 50-word vocabulary.
Source: Read the original article
