Recursive Unveils Flint, an AI Model Built by AI Itself

Recursive, a company focused on artificial intelligence development, has released Flint, a new open-weight AI coding model. The launch marks the first product from the firm, which operates under the philosophy that AI should be used to build and refine other AI systems. Richard Socher, the CEO and cofounder of Recursive, stated that the creation of Flint was largely handled by the technology itself.

The company utilized an automated system known as the Eureka Machine to manage the construction of the model. According to Socher, this automated pipeline handled nearly every aspect of the development process. By relying on this system, the company was able to produce what Socher describes as a world-class model in just a few months with only a small team of researchers.

The Eureka Machine introduced specific technical changes during the build, including new architectures for the model and methods for training the base system. It also optimized the code for specific silicon chips. Recursive has shared benchmark scores indicating that Flint outperforms other models of similar size and architecture when handling coding tasks.

This method relies on the capability of AI to write code, which allows it to improve itself. While AI is often used to help write and test code, creating a loop where the system decides its own improvements represents a different approach. If an AI can contribute to every part of development, the improvement process could theoretically continue indefinitely.

This concept, known as recursive self-improvement, has generated significant debate within the industry. Some experts warn that a system capable of improving itself indefinitely could eventually escape human control. This concern was highlighted this September when a researcher left Anthropic and cautioned that such advances could lead to human extinction within a few years. Several bills in Congress currently propose to ban the practice.

Recursive views these warnings as more science fiction than reality and argues that legislation preventing the practice would hinder progress. Socher stated that Flint was built using a system incapable of escaping human control. He noted that while the goal is for the Eureka Machine to achieve full recursive self-improvement, humans would still oversee the process and could stop it 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 company has raised $650 million at a $4.65 billion valuation from investors including Google Ventures, Greycroft, Nvidia, and AMD Ventures. Socher previously served as chief scientist at Salesforce. His cofounder, Jeff Clune, is a professor at the University of British Columbia who researches AI systems capable of learning indefinitely.

Testing conducted on a Mac with 48 GB of RAM allowed Flint, which has 40 billion parameters, to run locally. During use, the model successfully built programs such as a command-line to-do list and a website explaining recursive self-improvement. Other organizations are also pursuing this technology; a team from Microsoft, Carnegie Mellon University, and the University of Virginia recently published work showing an AI agent modifying its own weights and software harness. Socher stated that this approach could make cutting-edge AI more accessible. "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.