Neuralink pretrains on 50,000 hours of neural data, cuts calibration to 10 minutes
XFreeze · x · 2026-10-03
Neuralink shared a major update on its brain-computer interface training: participants have generated over 50,000 hours of unlabeled neural data.
Key developments:
- Neural encoders are now pretrained on thousands of hours of each participant's brain activity before being used for cursor control
- Weekly calibration time dropped from 55 minutes to 10 minutes for some participants
- Three participants beat the previous 10.39 bits/sec benchmark; one hit 11.32 bits/sec, a new BCI speed record
- Some decoders stayed strong for weeks, with one still usable after 1.5+ years
The next step: foundation models for BCI that learn from massive neural datasets.
Related event: Neuralink Pretrains BCI Foundation Model on 50,000 Hours of Neural Data(3 posts)→
More from Embodied
- Berkeley's Humanoid Intelligence Center wins both tracks at IROS 2026 RoCo Challenge — berkeley_ai · 2026-10-03
- CoRL 2026 to host Sim-to-Real-to-Field workshop on transfer science, CFP open — berkeley_ai · 2026-10-03
- Yann LeCun: two years later we're still far from human-level AI — your cat beats it at physics — ylecun · 2026-10-03
- VRChat fluid sim adds collider buoyancy, performance presets and Japanese localization — Michael_Moroz_ · 2026-10-03
- SEU's LeaP: a learnable source prior boosts robot manipulation success by 25.5 points — 机器之心 · 2026-10-03
- Tesla Robotaxi in Austin extends service hours to 11pm — yunta_tsai · 2026-10-03