Recursive AI Development: Startup Unveils Flint Model

Recursive, a startup backed by investors including Google Ventures and Nvidia, has released Flint, an open-weight AI coding model the company says was built almost entirely by artificial intelligence. The release marks the first public offering from the firm, which is dedicated to using AI to further the development of AI itself.

Richard Socher, Recursive’s CEO and cofounder, said the company utilized a system it calls the Eureka Machine to automate nearly every stage of the model pipeline. According to Socher, this approach allowed a small group of researchers to construct a world-class model in a matter of months. The Eureka Machine reportedly devised specific technical elements, including a new method for training the base model, novel model architectures, and code optimized for specific silicon chips.

Flint includes 40 billion parameters, allowing it to run on consumer hardware such as a high-end laptop. In testing, the model demonstrated high performance on coding tasks, completing requests quickly and accurately. Benchmark scores shared by Recursive indicate that Flint surpasses other models of similar size and architecture in coding capabilities.

The release arrives amid growing debate regarding recursive self-improvement, a technique where AI systems are used to refine and build subsequent AI systems. While some experts warn that such systems could eventually escape human control, Recursive maintains that its technology is designed to remain under human oversight. Socher stated that the Eureka Machine is currently incapable of escaping human control, and that any future full recursive self-improvement would be overseen by humans who could stop the process if necessary.

Recursive has raised $650 million at a $4.65 billion valuation. Socher previously served as chief scientist at Salesforce, and the company’s cofounders include Jeff Clune, a professor at the University of British Columbia who researches AI systems capable of learning indefinitely. Socher suggested that the development of Flint could make cutting-edge AI more accessible to smaller teams. “It's the only way this very small team can produce a model that's actually better than any other model in its class,” he said.

Progress toward fully recursive systems is also being pursued by other organizations. A team from Microsoft, Carnegie Mellon University, and the University of Virginia recently published research showing how an AI coding agent can recursively improve itself by modifying its own software and model weights. Recursive views the fears surrounding this technology as premature, asserting that legislation to ban RSI is an impediment to progress.