On August 5, Naomi Bashkansky, a 23-year-old alignment researcher who spent 18 months at OpenAI, announced on X that she had resigned three weeks earlier to join Conduit, a San Francisco startup training thought-to-text AI on the world's largest known non-invasive neural dataset. Bashkansky argues that non-invasive brain-computer interfaces failed not due to insufficient hardware but because there was no sufficiently powerful probabilistic model to combine with the noisy signal. Her central analogy compares an LLM to a GPS receiver with a weak satellite signal, which needs a smarter prior rather than a stronger signal. By December 2025, Conduit's team had completed what it describes as the world's largest neuro-language dataset.