Why Hugging Face Is Competing w China In Robotics | Clem Delangue
Timestamps are as accurate as they can be but may be slightly off. We encourage you to listen to the full context.
Podcast Summary
In this engaging conversation, Clem Delangue, co-founder and CEO of Hugging Face, shares his journey from creating AI Tamagotchis to building what he calls "the GitHub of AI" - a platform serving 11 million AI builders worldwide. (11:00) The discussion covers Hugging Face's mission to democratize AI through open source, their recent launch of Richie Mini (an affordable $400-500 desktop robot for AI builders), and why they believe preventing AI concentration among a few companies is crucial for humanity's future. (18:00)
• Main themes: Open source AI democratization, the importance of building vs. using AI, creating aligned company incentives, and the philosophy of joyful building inspired by Camus's Sisyphus mythSpeakers
Clem Delangue
Co-founder and CEO of Hugging Face, Delangue has built what's become known as "the AI GitHub" - a platform serving 11 million AI builders with over 6 million models, datasets, and apps. Before Hugging Face became an AI platform giant, he co-founded the company in 2016 as a conversational AI startup, pivoting when the community showed massive interest in their PyTorch port of Google's BERT model. He's passionate about philosophy, particularly French absurdist Albert Camus, and believes in democratizing AI to prevent concentration of power among few companies.
Key Takeaways
Build What You're Excited About, Not What You Should Do
Delangue emphasizes that founders should focus on building what genuinely excites them rather than what investors or conventional wisdom suggests. (46:00) He warns that many founders start with genuine excitement but then shift to "what they should do to be more successful," ultimately creating companies they don't enjoy working for. This leads to founders quitting and hiring external CEOs, which typically damages the company. The key is maintaining alignment between your happiness and what you're building, ensuring you can wake up excited about your work for years.
Trust Your Excitement Over Rational Planning
Rather than relying heavily on to-do lists and rigid planning, Delangue advocates for following your natural excitement and intuition. (50:00) He deliberately avoids assistants and extensive task management, believing that if something is truly important and exciting enough, you'll naturally remember to do it. This approach keeps you aligned with what genuinely matters rather than getting caught up in tasks you feel you "should" do. It also enables faster, more authentic decision-making - as evidenced by how quickly this very interview was scheduled.
Create Systems with Aligned Incentives
Hugging Face's success stems from building a business model where doing good for the community directly benefits the company. (12:00) Delangue warns against creating systems where good intentions fight against human nature or business incentives. For example, if you monetize proprietary AI models, you're incentivized to keep things closed rather than open. By making 99% of their platform free and succeeding when open source thrives, Hugging Face ensures their business interests align with their mission to democratize AI.
Distribute Decision-Making to Unleash Innovation
Instead of centralizing functions like hiring, social media, and product development, Hugging Face distributes these responsibilities across all team members. (15:00) Anyone can tweet from the company account, anyone can hire, and team members work on what excites them rather than fixed job descriptions. This prevents the limiting "boxes" that large tech companies put people in and has led to breakthrough moments - like when co-founder Thomas Wolf spent a weekend porting BERT to PyTorch, creating the foundation for their entire platform.
Launch Early and Embrace Imperfection
Delangue's philosophy is to release minimum viable products quickly rather than waiting for perfection. (60:00) When Thomas released the PyTorch BERT port, it didn't work for two days, but the community's excitement validated the concept. The key is setting proper expectations - don't promise AGI, but be transparent that it's an imperfect first version. This approach allows for rapid iteration, community feedback, and building trust through transparency rather than over-promising and under-delivering.
Statistics & Facts
- Hugging Face now serves 11 million AI builders using their platform, with over 6 million models, datasets, and apps hosted on the platform. (26:00) A new repository is created every eight seconds, demonstrating the massive scale and rapid growth of the AI development community.
- Over 5,000 people have already pre-ordered Richie Mini, Hugging Face's $400-500 desktop robot for AI builders. (01:08) The company is starting to ship them globally and expects to ship around 4,000 units before the end of the year.
- The fastest growing domains for AI on Hugging Face are no longer text-based applications, but rather audio, video, image, biology, and chemistry. (68:00) This suggests the field is expanding beyond chatbots into specialized scientific and multimedia applications.
