Clement Delangue, CEO of Hugging Face, has issued a call for the community to publicly share "agent traces" and coding data. According to him, this is key to building better training datasets and open-source models for the future.
Context
In the AI race, high-quality datasets are often strictly guarded assets of large corporations. Hugging Face, following its open-source philosophy, is striving to break this barrier by encouraging voluntary contributions from the global developer community.
Key Developments
Delangue emphasized that sharing not only benefits individual projects but also contributes to the collective growth of the AI ecosystem. "Agent traces" provide deep insights into how models solve problems, thereby helping to improve the reasoning and execution capabilities of next-generation models. Many leading developers have already begun contributing their data to the Hugging Face Hub, setting the stage for a massive agent data repository.
Why It Matters
Data is the "fuel" of AI. Delangue's call reflects a long-term strategy: instead of competing solely on compute power, the open-source community can compete through data diversity and transparency. If successful, this movement will help bridge the gap between closed and open models in the high-potential field of AI agents.