OpenAI researcher joins Conduit to build telepathy
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- An unnamed OpenAI researcher resigned on July 23 after 1.5 years and started as Founding Researcher at Conduit the next day, citing the startup's work on thought-to-text models trained on non-invasive neural data.
- Conduit is collecting non-invasive neural data at scale to train brain-to-text decoders, with co-founders Rio Popper and Clem von Stengel and an open call for research participants.
- Conduit's scaling laws are showing cosine similarity between predicted and target latent spaces improving logarithmically with hours of collected data — a stage the author compares to the 'GPT-2 era' of language models.
- The author values Conduit at a potential $1 trillion if it becomes the general read-and-write brain interface company, explicitly applying the 'bitter lesson' of scaling compute and data over hand-crafted algorithms.
- Conduit's roadmap reportedly includes future work on invasive general read and general write capabilities, though the author predicts most users will stick with non-invasive neural headbands.
- The author sketches a 2035 vision in which AI becomes 'a sixth sense and another limb,' with thought-to-text overtaking typed prompts as the primary human-AI interface.
Why it matters: A former OpenAI researcher is staking their career on the claim that neural data will scale the same way text data did, putting Conduit's current stage at 'GPT-2 era' and valuing the company at a potential $1 trillion. If the scaling laws hold, the author argues, non-invasive thought-to-text could become the primary AI interface within years — and push the industry toward brain-computer integration that figures like Elon Musk and Sam Altman have publicly favored.
