SC
Will Open Source AI Overtake Closed Models? Ft. Olama, Fireworks and Open Router
@ Sequoia Capital
05/12/25
Takeaways
JM
Jeff Morgan
05/12/25
@ Sequoia Capital
I'm Jeff, founder of Olama, which provides an easy way to get started with open-source large language models (LLMs).
DD
Dmytro Dzhulgakov
05/12/25
@ Sequoia Capital
I'm Dimma, co-founder of Fireworks, a platform that helps fine-tune and serve generative AI models at scale, and we strongly support open-source models.
AA
Alex Atallah
05/12/25
@ Sequoia Capital
I'm Alex, CEO and co-founder of Open Router, a marketplace that offers the best pricing, performance, and uptime for language models, collaborating closely with various providers, including Fireworks.
SC
Sequoia Capital Cast
05/12/25
@ Sequoia Capital
A fundamental principle of open source is that human genius can emerge from anywhere, and centralizing it in model labs is risky and less effective than allowing innovation to come from diverse sources.
SC
Sequoia Capital Cast
05/12/25
@ Sequoia Capital
Open source models currently account for only 20 to 30% of inference tokens used, indicating that they are not yet winning against closed-source models.
SC
Sequoia Capital Cast
05/12/25
@ Sequoia Capital
The transparency and high leverage of open-source models will lead future generations to prefer creating open-source solutions over closed-source ones, as they grow up learning about state-of-the-art machine learning.
SC
Sequoia Capital Cast
05/12/25
@ Sequoia Capital
Many enterprises are eager to customize models, and if open-source models are banned, there will still be a demand for customization, which is crucial for customers who want full ownership of their fine-tuned models.
SC
Sequoia Capital Cast
05/12/25
@ Sequoia Capital
AI should not be banned as a fundamental technology, similar to how electricity cannot be regulated; instead, we should focus on regulating its applications and market distribution.
SC
Sequoia Capital Cast
05/12/25
@ Sequoia Capital
The current landscape shows that large model labs are focused on building powerful foundation models that require less fine-tuning, which may lead to a decline in the relevance of fine-tuning in the near future.
SC
Sequoia Capital Cast
05/12/25
@ Sequoia Capital
The success of open-source models like DeepSeek is attributed to their effective engineering and the ability of small teams to innovate quickly, integrating research and engineering seamlessly.
SC
Sequoia Capital Cast
05/12/25
@ Sequoia Capital
DeepSeek's rise in the open-source community is due to its strong reasoning model and user experience innovations that allow users to see the model's thought process, which had not been achieved before, and the 'DeepSeek moment' disrupted the market by providing a transparent and open-source approach that resonated with the community.
SC
Sequoia Capital Cast
05/12/25
@ Sequoia Capital
American companies had to adapt quickly to scale AI inference after their previous providers' servers failed, leading to significant growth for Fireworks in capturing inference metrics.
SC
Sequoia Capital Cast
05/12/25
@ Sequoia Capital
Decentralized inference providers gained traction as they offered models that met market demand, resulting in a rapid increase in interest that many in the AI space are still unaware of.
SC
Sequoia Capital Cast
05/12/25
@ Sequoia Capital
Fireworks excelled in scaling their model infrastructure to handle distributed inference effectively, preventing server overload and maintaining customer service during rapid growth.
SC
Sequoia Capital Cast
05/12/25
@ Sequoia Capital
Fine-tuning models like DeepSeek for specific tasks has proven effective, allowing customers to customize their AI solutions to better meet their business needs.
SC
Sequoia Capital Cast
05/12/25
@ Sequoia Capital
The future of AI models may see a balance between open-source and closed-source, with predictions suggesting a 50/50 split in inference token usage over the next five years.
SC
Sequoia Capital Cast
05/12/25
@ Sequoia Capital
If decentralized AI providers fail to establish sustainable models, closed-source solutions may dominate the market, but successful decentralization could shift the dynamics significantly.
SC
Sequoia Capital Cast
05/12/25
@ Sequoia Capital
Decentralization in AI could create a financial ecosystem that supports new players and enhances service delivery, potentially leading to a more competitive landscape.