- Naomi Bashkansky left OpenAI to join Conduit, a startup building AI telepathy models
- Conduit trains AI to decode brain activity into text using a non-surgical headset
- Bashkansky expects AI brain interfaces to evolve significantly by 2035
Naomi Bashkansky had spent the past 18 months on OpenAI's model safety and alignment team. On July 23, she resigned. The very next day, she started as a founding researcher at Conduit, a San Francisco startup building what she calls "telepathy," AI models trained to convert human thought directly into text.
Bashkansky, at 23 years of age, announced the move on X on August 5, alongside a blog post titled "Why I'm Leaving OpenAI To Build Telepathy." She also holds the Woman International Master title in chess, one rank below Woman Grandmaster.
What Is Conduit Building?
Conduit trains models to read brain activity captured through a non-surgical headset and predict what a person was thinking, based on text or actions recorded at the same time. Bashkansky likened the process to a noisy GPS signal: a single reading is imprecise on its own, but layered against an AI model's grasp of language and context, it becomes usable.
"We must scale up our data collection by orders of magnitude beyond what has ever been done in academia," she wrote, framing Conduit's bet as an application of AI's "bitter lesson": more data and compute, not cleverer hand-built systems, wins out in the end.
ALSO READ: SAP, Capgemini And More: AI Boom's Hidden Winners May Emerge In Europe, Not US
The Roadmap
Bashkansky's post lays out a three-stage trajectory. By 2027, she expects to direct AI coding agents by simply thinking instructions, decoded via headband into commands for tools like Codex.
By 2030, she expects frontier AI labs to train models that interface directly with Conduit's neural representations, while Conduit itself moves from reading thoughts into "general write," feeding information back into the brain.
By 2035, she predicts AI stops registering as a separate tool altogether. "It feels like a sixth sense and another limb," she wrote.
The Pitch For Leaving
She cited three reasons for the move. Firstly conviction that Conduit could become what she called "the general read and write company," a bet she values at over $1 trillion if it lands. Next, her regard for co-founders Rio Popper and Clem von Stengel. And lastly, research she found more open-ended than anything available inside OpenAI's existing architecture.
She isn't the first researcher to trade a seat at a major lab for a higher-conviction independent bet. Ilya Sutskever left OpenAI in 2024 to start Safe Superintelligence.
ALSO READ: Ray Dalio Sounds Alarm On AI Bubble, Warns Of Potential Burst
Not A New Idea In Silicon Valley
Bashkansky's post also traces the idea back through a decade of tech commentary. Sam Altman wrote in a 2017 blog post that a human-AI "merge is probably our best-case scenario," arguing that two species competing for the same dominance "are going to have conflict" if neither yields.
Elon Musk also made a related case that same year, describing a future of sufficiently tight symbiosis in which AI stops being "other" and instead relates to a person's cortex the way that cortex already relates to the limbic system, an interview Bashkansky says directly inspired one of Conduit's co-founders to enter neurotech in the first place.
Conduit enters a small but growing field of brain-computer interface players, most prominently Elon Musk's Neuralink, though its headset-based, non-surgical approach separates it from implant-led rivals.
Whether that approach can match the precision of surgical implants is still unresolved, something Bashkansky's own post concedes, noting "irreducible error due to noise" in current readings.
Essential Business Intelligence, Sharp Market Insights, Practical Personal Finance Advice, Daily Fuel, Gold and Silver Prices and Latest Stories — On NDTV Profit.