Neuralink just released a major new update on how it is training brain-computer interfaces
Participants have now generated more than 50,000 HOURS of unlabeled neural data
Neuralink is now pretraining neural encoders on thousands of hours of each participant’s brain activity before using them for cursor control
And the results are already impressive:
• Calibration time dropped from ~55 minutes per week to ~10 minutes for some participants
• Multiple participants set new personal BCI speed records
• Three participants surpassed the previous 10.39 bits/sec benchmark
• One participant reached 11.32 bits/sec — a new BCI record
• Some decoders stayed strong for weeks
• One decoder was still usable more than 1.5 years later
The most important part is what comes next
Neuralink is moving toward foundation models for BCI...systems that can learn from massive amounts of neural data and eventually generalize across people and tasks
The goal is to make brain-computer interfaces faster, more stable, require less calibration and eventually work more like a general-purpose platform
Neuralink is going from controlling a cursor
To building an interface between the brain and computers at scale
https://neuralink.com/updates/pretraining-on-50000-hours/
Participants have now generated more than 50,000 HOURS of unlabeled neural data
Neuralink is now pretraining neural encoders on thousands of hours of each participant’s brain activity before using them for cursor control
And the results are already impressive:
• Calibration time dropped from ~55 minutes per week to ~10 minutes for some participants
• Multiple participants set new personal BCI speed records
• Three participants surpassed the previous 10.39 bits/sec benchmark
• One participant reached 11.32 bits/sec — a new BCI record
• Some decoders stayed strong for weeks
• One decoder was still usable more than 1.5 years later
The most important part is what comes next
Neuralink is moving toward foundation models for BCI...systems that can learn from massive amounts of neural data and eventually generalize across people and tasks
The goal is to make brain-computer interfaces faster, more stable, require less calibration and eventually work more like a general-purpose platform
Neuralink is going from controlling a cursor
To building an interface between the brain and computers at scale
https://neuralink.com/updates/pretraining-on-50000-hours/