San Francisco startup Conduit is developing a non-invasive system that promises to turn thoughts into text (no brain implant required).
The company has reportedly collected around 10,000 hours of neural data from thousands of participants. While wearing custom headsets, participants speak or type during conversations with an AI. Conduit records the signals produced shortly before each response, then trains models to predict its meaning.
The system probably does not recover exact sentences directly from the brain. Instead, it attempts to extract a noisy representation of the user’s intent. An LLM then combines that signal with contextual information to generate likely text.
There is legitimate science behind the concept. Researchers have reconstructed approximate meaning from fMRI recordings, while recent experiments have decoded typed characters using MEG. But these systems remain constrained, often require extensive calibration and perform much worse with wearable EEG.
Conduit has not yet published enough technical evidence to establish that its system can decode arbitrary, unspoken thoughts. Its models might also be detecting preparation for typing or speaking, subtle muscle activity, or clues already available from the conversation.
So, is it real? The research project is real, and limited neural decoding is scientifically possible. But calling it “telepathy” is still marketing. Until Conduit presents blinded tests, independent validation and live results on unseen users, its futuristic thought-to-text interface should be viewed as a promising experiment…
