2026-01-13 06:27:19

For a long time, general-purpose robots have been hindered by the human-dominated data collection bottleneck, resulting in low efficiency. The emergence of NEO has changed this situation—it can autonomously collect data and independently complete learning, breaking through this ceiling.

Why can this expansion plan run smoothly? Two driving wheels are indispensable: first, the continuously generated new robot data; second, the underlying engine, a continuously evolving video world model. The two complement each other, forming a self-iterating positive feedback loop.

View Original

This page may contain third-party content, which is provided for information purposes only (not representations/warranties) and should not be considered as an endorsement of its views by Gate, nor as financial or professional advice. See Disclaimer for details.

15 Likes

Reward
15
5
Repost
Share

Comment

0/400

DoomCanister

· 19h ago

Wow, this is the right path. Autonomous learning robots should have appeared long ago. This NEO is indeed impressive. It generates its own data and learns by itself, no more 996 outsourcing annotators needed. I feel like the video model is the key. Once positive feedback kicks in, it will probably grow exponentially. Human annotation has really been a bottleneck for a long time. Finally, someone is taking it seriously. Once this closed loop stabilizes, how will competitors play catch-up?

View OriginalReply0

rekt_but_not_broke

· 19h ago

Uh, NEO's self-learning logic sounds pretty impressive... but the idea of a self-iterating closed loop still raises some questions. By the way, can it really work, or is it just another hype concept? Data flywheel sounds great, but how will it actually be implemented? Robot data is already flooding the screens—will this time be reliable?

View OriginalReply0

ThreeHornBlasts

· 19h ago

Alright, the idea of robots learning on their own sounds like real cutting-edge technology, but it still feels a bit exaggerated. The positive feedback closed loop sounds a bit confusing to me; simply put, does it mean getting faster and faster? If Neo can really break through bottlenecks, then it’s definitely worth paying attention to. Self-collection of data... if it can run stably, it feels like the big picture has been expanded. Wait, will this autonomous learning become more and more wild?

View OriginalReply0

staking_gramps