China, Robotics, & Open-Source AI | Clem Delangue

In this talk

1. 0:52

   Reachy Mini invites builders in

   Delangue presents Reachy Mini as an affordable, programmable desktop robot for people learning to build with AI. He says it costs $400 to $500, is mostly 3D-printable, and arrives for users to assemble. Hugging Face hopes builders will create and share robot apps, adapt the hardware, and bring new AI model capabilities to it.

2. 7:25

   Community shapes the company

   Delangue traces Hugging Face’s platform back to a community response to a PyTorch version of BERT, which drew strong interest and led researchers to share more models. He says the platform is 99% free and that its business model is designed to reward expanding open source. The company gives employees broad responsibility for hiring, community engagement, and trying projects they find useful.

3. 20:25

   Build for agency and trust

   Delangue says building things can bring people joy and give them more say in how technology serves them. For example, he wants his Reachy Mini to process data locally so information from his home stays on his devices. He argues that open models and transparent systems let people inspect, adapt, and choose tools that fit their needs.

4. 1:09:14

   Open source can spread AI capability

   Delangue sees AI’s concentration of power as a risk and argues that shared models, datasets, and tools give smaller organizations a foundation to build on. He favors many specialized models for particular tasks, which can be cheaper and easier to adapt than large general-purpose systems. He also notes that Chinese organizations have become prominent open-source AI contributors and urges more U.S. participation.

5. 1:43:18

   Robotics needs shared foundations

   Delangue says lower-cost hardware and improving AI software are creating favorable conditions for robotics, but isolated projects and incompatible systems slow progress. Hugging Face’s LeRobot library aims to support datasets and models across different hardware. He believes more sharing and common practices could help the field advance faster.