Open Source Days 2025

As AI systems become foundational to society, we face a critical juncture: will AI be concentrated in the hands of a few, or will it serve the public good? This talk explores how open-source ecosystems, and Hugging Face’s approach, can ensure AI remains democratic, transparent, and beneficial. Drawing from recent breakthroughs like DeepSeek-R1 and the collaborative training of state-of-the-art (SoTA) open LLMs, I’ll share how openness drives innovation, accountability, and resilience in a rapidly evolving field.

We’ll look at how open model weights, reproducible science, opt-out infrastructure, and community governance aren’t just technical features, they’re ethical commitments. I’ll also unpack the spectrum of openness (from closed APIs to full-stack transparency) and share lessons from Hugging Face’s public interest work, including efforts on responsible dataset documentation, energy impact labeling, and multilingual fairness benchmarking.