Archive · 13 Dec 2025
Conduit builds largest brain-to-language dataset in lab
A company called Conduit in San Francisco has made a big collection of brain data.
They recorded the brains of thousands of people while they talked or typed.
This helps them teach computers to understand what people are thinking.
They used special headsets to capture brain signals.
The company had to solve problems like electrical noise and make their systems more efficient.
Now, they are training their models to turn brain signals into words.
They haven't said yet how well their models work.
Conduit has collected 10,000 hours of brain data from thousands of volunteers in San Francisco.
The data was gathered using custom-built headsets that capture neural signals before words are spoken or typed.
The company redesigned their setup to allow open-ended dialogue, leading to richer data.
They built their own hardware to capture more signals, combining EEG and other sensors.
Conduit has reduced the cost of data collection by 40% through efficiency improvements.
- Who
- AI startup Conduit in San Francisco
- What
- Built the largest brain-to-language dataset
- When
- Over the past six months
- Where
- In a windowless basement in San Francisco
- Why
- To teach machines to translate thoughts into text
- Company
- Conduit
- Location
- San Francisco
- Dataset Size
- 10,000 hours of neural recordings
- Participants
- Thousands of volunteers
- Recording Duration
- Two-hour sessions
- Hardware Used
- Custom-built headsets with EEG and other sensors
- Data Collection Period
- Six months
- Data Collection Efficiency
- 40% cost reduction over the project