Recursive AI System Builds Advanced Model

A $650 million valuation for Recursive highlights the intense interest in a technique where artificial intelligence is used to build other artificial intelligence. The company's first offering, an open-weight coding model named Flint, is designed to demonstrate the viability of this approach.

Recursive claims that nearly every stage of Flint's development was handled and improved upon by its own automated system, known as the Eureka Machine. According to Richard Socher, Recursive's CEO and cofounder, this automated process generated innovations including new training methods, novel model architectures, and code optimized for specific hardware. The company released benchmark scores suggesting Flint outperforms other models of a similar size on coding tasks.

The people behind Recursive have deep experience in artificial intelligence. Socher previously served as chief scientist at Salesforce. His cofounder, Jeff Clune, is a professor at the University of British Columbia whose research focuses on creating AI systems capable of learning indefinitely.

The company's backers include Google Ventures, Greycroft, Nvidia, and AMD Ventures, which contributed to a $650 million funding round that valued Recursive at $4.65 billion.

The stakes for this work are significant. While using AI to accelerate AI development is not new, the prospect of a fully recursive loop, where a system decides on its own improvements and begins another cycle, has drawn both excitement and concern. Some experts have warned that a system capable of improving itself indefinitely could escape human control. This concern was amplified in September when a researcher left Anthropic to warn that such capabilities could threaten human existence. The debate has reached Capitol Hill, where several bills have been proposed to ban the practice.

Recursive contends that such fears are premature and that legislation would hinder progress. Socher stated that the system used to build Flint cannot escape human control. He explained that while the goal is for the Eureka Machine to achieve full recursive self-improvement, the process would remain under human supervision and could be halted if necessary. "We hope that this shows people that indeed we have all the steps in place" to build such a system, he said.

The potential for such a system to make advanced AI more accessible is a key part of Recursive's mission. "It's the only way this very small team can produce a model that's actually better than any other model in its class," Socher told Global Business Media Hub. "This will hopefully show people that [recursive self-improvement] is actually quite nice and will benefit a lot of people."