Stanford professors launch Noeri AI lab with $110m in funding
Backed by Nvidia and Menlo Ventures, the new lab aims to build self-improving AI systems that learn from a company's specific data and workflows.

The short version
- Noeri raised $110 million in a funding round led by Menlo Ventures with participation from Nvidia and Snowflake.
- The lab focuses on self-improving AI that adapts to specific company data, tools, and workflows over time.
- The system requires human approval for every change, functioning similarly to a code review process.
- Co-founders Yejin Choi and Carlos Guestrin are prominent AI researchers with ties to Stanford and Apple.
- The company aims to help scientists form hypotheses and run experiments on molecules, proteins, and technical systems.
Why it matters
This represents a shift toward 'on-the-job' learning where AI adapts to a specific business's unique needs without requiring constant retraining. It also highlights a move toward smaller, specialized models that can be more affordable for research and complex industrial tasks.
New funding for Stanford lab
Noeri, an AI lab founded this year by Stanford professors Yejin Choi and Carlos Guestrin, has officially launched with $110 million in funding. The investment round was led by Menlo Ventures and included backing from Madrona Venture Group, Kindred Ventures, Wing Venture Capital, and NEA. Strategic investors for the round included Cisco Investments, Nvidia, and Snowflake.
The company was established to address the limitations of current AI models, which typically stop learning once their initial training phase is complete. Noeri intends to build systems that continue to improve while they are being used in real-world environments. This approach is designed to serve both large-scale corporate needs and specialized scientific research.
The lab is currently establishing its physical presence with headquarters in Palo Alto and an additional office in Zurich. The company is actively hiring both researchers and engineers in these two locations to build out its team. While the company did not disclose its specific valuation, the funding marks a significant entry into the competitive AI infrastructure market.
How self-improving AI works
The core technology of Noeri focuses on AI that studies a company's specific data and internal workflows to adapt to its unique environment. Instead of providing a static tool, the system identifies how to change the instructions, the tools used, and which specific models handle different steps of a task. It can even modify the models themselves when it identifies a way to improve performance.
To ensure safety and accuracy, the system does not make changes autonomously without oversight. Every modification made by the AI comes with a clear explanation and must be approved by a human, much like a traditional code review. This allows the system to evolve while keeping a person in the loop to verify the logic and the results of each automated change.
The platform is designed to be flexible, working with both closed and open-weight models. Customers are told they will maintain full ownership of their data, the workflows they develop, and the specific models they build using the system. This architecture is intended to provide a balance between high-performance automation and strict data privacy for the user.
Solving the cost of research
Noeri aims to provide a more affordable path for organizations that are currently forced to choose between expensive, high-performing models and cheaper, less capable ones. Carlos Guestrin, the company's co-founder and co-CEO, noted that most companies are currently facing this trade-off. The lab intends to use smaller, specialized models to make complex searches and research tasks more economically viable for a wider range of users.
The inspiration for this approach came from work at Stanford involving a microbiology and immunology team. That team was looking for cheaper ways to utilize large language models like those from OpenAI for their specific research needs. The team discovered that a smaller AI, which was designed to learn how to use other models, could actually outperform larger models on their specific tasks.
This methodology allows for more efficient exploration of complex problems, ranging from new molecules and proteins to intricate technical systems. By using smaller models that are specialized for a task, researchers can run more experiments without the high costs associated with the largest general-purpose models. This focus on efficiency is a primary driver for the lab's technical direction.
Leadership and origins
The leadership team at Noeri brings significant experience from both academia and the private sector. Yejin Choi is a MacArthur Fellow and a former professor at the University of Washington who also served as a senior director at the Allen Institute for AI. She was also previously a distinguished scientist at Nvidia, where she worked on various AI initiatives.
Carlos Guestrin serves as the other co-CEO and is currently the director of the Stanford AI Lab. He has a deep history in machine learning, having co-founded Turi, a platform that was eventually acquired by Apple in 2016. He also led Apple's central machine learning and AI organization and co-created XGBoost, which is a widely used open-source machine learning library.
The founders have expressed a shared concern regarding the centralization of AI power. Guestrin stated that both he and Choi have worries about one company owning everything in the AI space. By building a system that allows companies to keep ownership of their own models and workflows, Noeri aims to provide a decentralized alternative to the current market landscape.
What to watch
- How Noeri's 'human-in-the-loop' approval system will scale as companies automate more complex workflows.
- Whether the lab's smaller, specialized models can maintain a competitive edge over the latest large-scale models from major tech companies.
Try it
Look into the difference between 'closed' and 'open-weight' models to see how they differ in terms of data privacy and customization.
Sources and links
- ReportingThe Next Web: the article we sourced
First reported by The Next Web. Written by the SI Agent News Desk from the reporting above.Written with AI from the source above. Every number is checked against the source. Spot a mistake? Corrections.
Liked it? Say why
Add a line for other readers.
No account needed. We never show your email. Comments with links, ads or abuse are held and not shown. Your comment shows after a quick check.