Neuralink pretrained AI on 50,000 hours of brain activity, hits 11.32 bps cursor-control record
mark_k · x · 2026-10-04
Neuralink is applying pretraining to brain-computer interfaces: models are pretrained on roughly 50,000 hours of recorded brain activity, learning neural patterns without humans labeling every action, with the goal of translating a user's intended movements into reliable computer control.
Results:
- Decoders keep working for weeks without recalibration, meaning fewer interruptions for paralyzed users
- A new cursor-control record of 11.32 bits per second
Related event: Neuralink Pretrains BCI Foundation Model on 50,000 Hours of Brain Data(7 posts)→
More from Embodied
- Training a 128k-param neural cellular automata to 'see around corners' with sound — yacineMTB · 2026-10-04
- CoreWeave's Physical AI Lead Explains Why Robots Hit the Sim-to-Real Gap — thursdai_pod · 2026-10-04
- Replacing CAD with code: LLM-written build123d Python scripts design entire robots — natesiggard · 2026-10-03
- HUST, PKU and Keenon Robotics unveil AdaRoboVLG for context-aware robot grasping — jiqizhixin · 2026-10-03
- Sharpa D01 humanoid robot outperforms humans at grasping soft dropped objects — CyberRobooo · 2026-10-03
- The $899 Mac Mini M6 was built for 24/7 agentic computing — most owners use it to check email — hey_abusiddik · 2026-10-03