Anthropic Investor in Talks to Fund New Lab Run By Two Stanford AI Professors
Cloud and software incumbents want businesses to think twice before using Anthropic, OpenAI and other closed-source AI providers. Executives such as Palantir CEO Alex Karp have argued, for instance, that these AI firms could use their customers’ data to compete with them eventually.
These potential fears could also help new AI startups catering to businesses.
One such startup is Noeri, a new AI lab founded by two Stanford computer science professors to help businesses customize AI models for their specific needs so they don’t need to buy models from Anthropic and major providers. Menlo Ventures, a major Anthropic backer, is in talks to lead a $100 million funding of the startup, according to a person involved in the discussions.
Other venture firms including Andreessen Horowitz, New Enterprise Associates, Madrona as well as Nvidia are also in talks to invest as well, the person said.
Their interest in the startup is a sign that investors think businesses will be more wary of working with the most advanced AI providers as those providers—namely, Anthropic—keep launching features and applications targeting a variety of professional workers. Anthropic, for instance, blindsided some business partners that had used Claude to power apps for designers or lawyers when it later launched its own, competing apps or features for people in those fields.
The funding round will likely have multiple tranches at different valuations, a common occurrence among hot AI rounds these days, but the final valuation could range anywhere from $600 million to $1 billion, the person said. (Different valuations happen when there’s too much demand for a funding round, leading some VCs to offer to invest at higher valuations to sweeten the deal for the startup.)
Noeri was founded earlier this year by Yejin Choi and Carlos Guestrin, who are well-known for their work on test-time training and reinforcement learning via self-distillation, terms that refer to helping a model learn and improve on certain tasks after it’s been trained. (Guestrin’s work on self-distillation in particular was used by Cursor in training its latest coding models.)
Noeri is far from the first company to try and help businesses tweak models for specific tasks (otherwise known as finetuning). Thinking Machines Lab’s Tinker product, for instance, provides finetuning software that developers can access through an application programming interface, while Applied Compute takes a more hands-on approach by embedding forward deployed engineers with its customers.
Noeri hasn’t decided on its strategy or business model yet, the person with knowledge of the discussions said. It’s possible that the company will do the finetuning work itself for customers and provide those customers with the finished model at the end of the process. It’s also possible that it’ll take a more hands-off approach, a la Tinker.