Naomi Bashkansky, a former OpenAI alignment researcher, has joined Conduit as a Founding Researcher to work on AI models that convert brain activity into text. She left OpenAI on July 23, 2026, and started at Conduit the following day.
TL;DR
- Naomi Bashkansky moves from OpenAI to Conduit to lead research on thought-to-text AI models.
- Conduit aims to train models on non-invasive neural data to predict text from brain activity.
- Bashkansky predicts significant advancements in AI-human cognition integration by 2035.
What happened
Naomi Bashkansky resigned from OpenAI on July 23, 2026, where she worked as a Researcher on the Alignment team. She joined Conduit the next day as a Founding Researcher to work on thought-to-text models trained on non-invasive neural data.
Bashkansky announced her move on LinkedIn, describing Conduit's goal as training models to non-invasively read the human mind. The team uses 'telepathy' as a shorthand for their work, which involves training models to predict text from brain activity.
Conduit's research focuses on scaling up the amount of neural data used to train models. Bashkansky compares their current progress to the GPT-2 era, indicating that the models are improving as more data is collected.
Why it matters
This move signifies a shift in focus for Bashkansky from AI alignment to the intersection of neural data and AI models. Her experience at OpenAI and Anthropic brings valuable insights to Conduit's research.
The potential applications of thought-to-text models could revolutionize human-computer interaction, making AI a natural extension of human cognition. However, the technology is still in its early stages, with Bashkansky comparing it to the GPT-2 era.
Conduit's work could have significant implications for developers, startups, and investors in the AI industry. The company is currently recruiting researchers and infrastructure staff, indicating a growing interest in this emerging field.
Key facts
- Naomi Bashkansky left OpenAI on July 23, 2026, and joined Conduit the following day.
- Conduit aims to train AI models on non-invasive neural data to predict text from brain activity.
- Bashkansky compares Conduit's current progress to the GPT-2 era, indicating that the models are improving with more data.
- Conduit is recruiting researchers, infrastructure staff, and operators to support its research efforts.
- Bashkansky predicts that by 2035, AI could become a natural extension of human cognition.
- Before joining OpenAI, Bashkansky worked at Anthropic, RAND, and the U.S. AI Safety Institute.
- Bashkansky studied computer science at Harvard University.
Context
The field of neural interfaces and AI is rapidly evolving, with companies like Neuralink and Brain-Computer Interface (BCI) startups exploring similar technologies. Conduit's work on thought-to-text models adds a new dimension to this landscape.
The integration of AI with human cognition has been a topic of interest for researchers and investors alike. Bashkansky's predictions highlight the potential for AI to become more deeply intertwined with human thought processes.
The recruitment efforts by Conduit indicate a growing interest in the field of neural data and AI. This could lead to increased investment and collaboration opportunities for startups and researchers in this area.
