This episode of Equity explores the economics, risks, and trade-offs of building on open source AI models. Irene Fuleman from Hugging Face discusses how companies capture value when models are freely available, the rise of Chinese open models, and the cybersecurity challenges that threaten the open ecosystem. Key insights include the importance of model routing, the potential for revenue-sharing licenses, and the need to balance openness with security.
Summarized by Podsumo
Open models are becoming mainstream as enterprises switch to cut costs and improve data privacy, but model developers face challenges in capturing value.
Chinese open models like Quen and Kimi K3 are gaining ground, but geopolitical hesitancy and data security concerns affect adoption.
Cybersecurity threats, including agentic attacks on platforms like Open Code, highlight the need for robust defense mechanisms and incident reporting standards.
Smaller, permissive models and harnesses can outperform frontier models for specific tasks, offering cost-effective solutions for startups.
The debate over obliterated models (with guardrails removed) underscores the tension between enabling defense research and preventing misuse.
"The level of value from the deployers and from the developers depends on the makeup of the business organization. For Alibaba Cloud, they're just such an enormous cloud company, it could be considered a feature to bring folks into Alibaba Cloud."
"What makes an open community targeted is also what makes it special. We are necessarily permissive. We want the average hobbyist in their basement to be able to contribute."
"If you want something that you want to scale up, you can set it up on-prem, optimize that model over time without renting, and your long-term costs will likely decrease."