Noeri develops self-improving AI for business and scientific research. Each modification proposed by the system must be validated by a person.
At Stanford, a team sought to spend less
A Stanford microbiology and immunology team was looking to reduce their expenses using OpenAI models. Yejin Choi explains that this collaboration partly inspired the creation of Noeri. According to her, a small AI developed for this project, capable of using other models, performed better than GPT-5 for the team’s specific needs. This AI notably used GPT-4, but not GPT-5, because permanent use of the latter was too costly for Stanford.
Carlos Guestrin says that all the companies Noeri meets with face a choice between paying more for a high-performance model or settling for worse results with a cheaper model. Noeri also wants to enable its AI to collaborate with scientists, formulating hypotheses, conducting experiments and learning from the results. The targeted areas include the discovery of molecules, proteins and the study of complex technical systems. According to the company, using smaller, specialized models makes these searches more affordable.
Noeri submits changes for human approval
Noeri designs a self-improving AI intended for professional and scientific uses. The company says its systems continue to progress while in use, unlike most AI models that stop learning after training. According to the description provided, the AI ​​analyzes the customer’s data and processes and then adjusts the instructions, tools or model assigned to each step. She can also modify the models themselves when relevant.
The evening letterFree · 7 p.m.
AI transforming business?
AI strategies, movements and tricks deciphered, every evening in 5 minutes. Free.
Included upon registration: our selection of the best AI guides & comparisons.
Choose your pace
Free · No spam · Unsubscribe in 1 click
Each change proposed by the AI ​​is accompanied by an explanation and must be approved by a person, according to a principle similar to code review. The system works with open or closed weight models, and customers retain ownership of their data, workflows, and models they create. Carlos Guestrin says he and Yejin Choi share concerns about centralization and the possibility of a single company owning the entire sector.
Menlo Ventures leads Noeri financing
Noeri was founded this year by Stanford professors Yejin Choi and Carlos Guestrin, who serve as its executive directors. The laboratory announces that it has raised $110 million in a funding round led by Menlo Ventures, with the participation of Madrona Venture Group, Kindred Ventures, Wing Venture Capital and NEA. Cisco Investments, Nvidia and Snowflake are among the strategic investors. Noeri’s valuation has not been communicated.
MacArthur Prize winner Yejin Choi has taught at the University of Washington and serves as senior director at the Allen Institute for AI; she is also mentioned as a distinguished scientist at Nvidia. Carlos Guestrin helped create the Turi machine learning platform, acquired by Apple in 2016, and then took the helm of the central machine learning and AI organization at Apple. He is also one of the creators of the open source library XGBoost and is listed as director of the Stanford AI Lab.
Noeri is headquartered in Palo Alto and has an office in Zurich. The company hires researchers and engineers in these two locations and works with partners from various sectors, including some of its investors.

