Crypto
Home›Crypto›Market Structure›Conduit hires former OpenAI researcher to decode neura…
Conduit hires former OpenAI researcher to decode neural recordings into AI prompts
Conduit says it has compiled about 10,000 hours of neuro-language data from thousands of participants, but it has not yet published aggregate performance for free-form thought decoding.
Former OpenAI alignment researcher Naomi Bashkansky resigned from OpenAI on July 23 and joined Conduit the next day as a founding researcher, CryptoSlate reports.
At Conduit, Bashkansky will work on models intended to convert non-invasive neural recordings into text that can direct AI agents, a goal she describes as “telepathy” in an Aug. 4 essay. She previously predicted a headband could decode rough intentions into prompts for an AI coding agent in 2027, with broader “read and write” scenarios envisioned later, though she said the essay includes no launch commitment.
CryptoSlate reports that Conduit has said it gathered roughly 10,000 hours of neuro-language data from thousands of people, using multimodal headsets while participants typed, spoke, read, or listened during sessions with a language model. The company has published a few claimed zero-shot examples, but it has excluded aggregate performance metrics, its evaluation protocol, and any third-party replication.
The outlet notes that Conduit’s next evidentiary hurdle is establishing public aggregate performance by linking its dataset to a portable system that decodes free-form thought, a result the company has not yet shown. CryptoSlate also cites Meta’s June report of 61% average word accuracy in its Brain2Qwerty work, as well as a Nature Communications study of 723 participants finding performance improves with more EEG and MEG data, though those experiments focused on reading or listening rather than producing language.