Hugging Face's Clem Delangue on Open Source AI and the LLM Bubble | MTS Live
Hugging Face CEO Clem Delangue discusses the shift away from open-source AI in the US while China dominates open-source contributions, argues against restricting AI model releases for safety reasons, and warns of a potential LLM bubble. He also touches on Hugging Face's robotics push with LeRobot and explains why Hugging Face became the go-to platform for AI model sharing over GitHub.
Summary
In this interview on MTS, Hugging Face CEO Clem Delangue covers several major themes in the AI industry. On the topic of open source, he notes a significant reversal in trends: the US, historically a leader in open-source software, has seen its frontier AI labs increasingly close off their models behind proprietary APIs. Meanwhile, China has emerged as the dominant force in open-source AI, with models like DeepSeq, Quen, and Kimi being widely adopted by US startups and academic institutions.
Delangue pushes back strongly against the idea of restricting model releases for safety reasons, arguing that such concerns have consistently been overstated — citing GPT-2 as an early example where fears proved overblown. He uses an analogy of tying down people's hands to prevent punching, arguing that the better approach is to keep capabilities open and instead regulate and prosecute bad actors. He contends that openness actually improves security by ensuring defenders have the same capabilities as attackers, particularly in cybersecurity contexts.
On the topic of an LLM bubble, Delangue clarifies that he doesn't believe AI broadly is in a bubble, but that the specific segment of large language models distributed behind closed APIs may be overinvested, given uncertain margins, sustainability, and long-term competitive moats. He expresses cautious concern rather than a definitive prediction.
Delangue discusses Hugging Face's robotics initiative, LeRobot, which has shipped nearly 10,000 units of its 'Richinini' robot globally, with over 300 apps already built for it. He frames robotics as a key frontier unlocked by AI, enabling entirely new use cases beyond screens and phones. He also flags that China is positioned to dominate robotics and argues the US needs to build more aggressively in this space.
Finally, on the question of why Hugging Face became the 'GitHub of AI' rather than GitHub itself, Delangue explains that hosting AI artifacts is fundamentally different from hosting code — involving vastly larger file sizes and data volumes. He notes Hugging Face added two petabytes of data in just one week, illustrating the scale difference, and says the platform was purpose-built for AI workflows in ways GitHub was not.
About this episode
Clem Delangue joins MTS to discuss the global open-source AI landscape, the current large language model bubble, and the future of consumer robotics. Originally aired on MTS, Theo Jaffee and Sofia Puccini speak with Clément Delangue, CEO at Hugging Face, about the global open-source AI race, why he believes the real bubble is in API-based large language models, and how robotics could become the next major interface for AI. They also discuss AI safety, U.S.-China competition, open-weight models, and why Hugging Face became the infrastructure layer for open AI development.
Key Insights
- Delangue argues that China has become the dominant contributor to open-source AI, with most US startups and academia now relying on Chinese open-source models like DeepSeq and Quen — a reversal of historical trends where the US led open-source development.
- Delangue contends that safety-based restrictions on open-source AI releases are consistently overblown, pointing to the example of GPT-2 being called dangerous years ago when it was essentially just an autocomplete, and arguing that society always adapts after each release.
- Delangue specifically locates the potential AI bubble not in AI broadly, but in large language models distributed behind closed APIs, citing uncertain margins, unclear long-term sustainability, and weak competitive moats as the key risk factors.
- Delangue claims that making AI models more open actually improves overall security rather than increasing risk, because it ensures defenders have the same capabilities as attackers — a dynamic he says is particularly important in cybersecurity.
- Delangue explains that Hugging Face's infrastructure handles a fundamentally different scale of data than GitHub — adding two petabytes in a single week, equivalent to 500,000 two-hour movies — which is why Hugging Face, not GitHub, became the dominant platform for AI model and dataset hosting.
Topics
Transcript
The idea of restricting a technology like AI based on risks is just like, for example, you would say, okay, some people can punch other people, so let's tie down everybody's hands, right? Because it's too dangerous. Some people can punch, right? But you really don't want to do that because your hands are so useful. The way you want to control it is untie everyone and then regulate or fight the bad actors. So for example, if hacking, that creates cybersecurity risks, it's illegal, right? So you have to fight it, but not by preventing everyone from getting these capabilities. Otherwise you slow down progress, you create massive gaps in terms of controls, in terms of capabilities, and you…
Full transcript available for MurmurCast members
Sign Up to AccessMore from The a16z Show
Technology, Alliances, and American Leadership.
A16Z partners discuss the firm's expansion into global markets, emphasizing how AI has become central to national security, economic growth, and geopolitical influence. They argue that American tech values must be preserved globally through strategic partnerships with allied nations, while addressing the dual-use nature of AI technology and the unique conditions that make ecosystems like Silicon Valley difficult to replicate.
Outsmarting Uber: Why Bolt Wins in Europe
Markus Willeck, founder and CEO of Bolt, discusses how the company became Europe's leading mobility platform by operating with superior capital efficiency, competing against well-funded rivals like Uber. He covers Bolt's expansion strategy, COVID-19 pivots, autonomous vehicle plans, and the advantages of building in Europe despite regulatory and market complexity.
Rick Rubin on AI, Creativity, and The Way of Code
Rick Rubin discusses his new book "The Way of Code," which combines 3,000-year-old wisdom from the Dao De Jing with modern concepts of 'vibe coding'—using AI as a creative tool. The conversation explores how AI democratizes creativity while raising questions about authenticity, originality, and what remains uniquely human in the creative process.
Building AI for Creators | Luma & Phota Labs
Matt Tancic (Luma) and Zach Xia (PhotoLabs) discuss how AI tools are reshaping creative workflows, arguing that creativity lies not in mastering tools but in directing agents to execute a creative vision. They explore the tension between advancing AI research and building practical products for creators, emphasizing personalization, iteration, and user-centric design.
Beyond P(doom): Marc Andreessen - Betting on America
Marc Andreessen discusses AI's transformative potential for education, healthcare, and housing, but argues that decades of regulatory restrictions on these sectors will prevent productivity gains from benefiting consumers. He advocates for maximizing AI export and innovation rather than restricting technology through export controls, while acknowledging genuine national security tradeoffs with China.