Recursive Launches Flint, an AI Model That Builds Itself

A new open-weight AI coding model named Flint has just arrived, and early testing suggests it is a significant leap forward for the controversial concept of recursive self-improvement. Developed by the startup Recursive, Flint is the first product from a company dedicated to using artificial intelligence to build and refine other artificial intelligence systems. The model has already generated strong benchmark scores, outperforming other models of similar size and architecture on coding tasks. For readers in New England and elsewhere, this marks a tangible shift in how advanced AI is created, moving away from purely human-led development toward automated cycles of enhancement.

The numbers behind Flint’s creation are striking. The model contains 40 billion parameters, a size that can run on standard high-memory hardware rather than massive server racks. Recursive CEO Richard Socher states that nearly every aspect of the model pipeline was handled by AI. The company utilized a system called the Eureka Machine to automate the development process. This automation allowed a small team of researchers to produce a world-class model in just a few months. The Eureka Machine designed new training methods, model architectures, and code optimized for specific silicon chips, all without extensive human intervention.

The people driving this initiative are experienced figures in the tech world. Socher, who previously served as chief scientist at Salesforce, leads Recursive. He is joined by cofounder Jeff Clune, a professor at the University of British Columbia who studies systems capable of indefinite learning. The company has secured $650 million in funding from Google Ventures, Greycroft, Nvidia, and AMD Ventures. This investment values the company at $4.65 billion, signaling strong investor confidence in the RSI approach. The team’s goal is to make cutting-edge AI more accessible by leveraging automation to offset the lack of a large workforce.

The stakes of this technology are high, both in terms of performance and philosophical implications. Recursive self-improvement, or RSI, involves creating a loop where AI decides on improvements and then implements them. While currently limited, this process could theoretically continue indefinitely. Some experts worry that such a system could escape human control, leading to outcomes where AI views humans as obstacles to its goals. These concerns gained traction this September when a researcher from Anthropic warned that RSI could lead to human extinction within a few years. The debate has become intense enough that several bills in Congress propose banning the practice.

Recursive views these fears as more science fiction than immediate reality. Socher insists that Flint was built using a system that remains under human oversight. The company aims for the Eureka Machine to achieve full RSI, but humans would retain the ability to stop the process if necessary. Socher hopes this demonstrates that the necessary steps are in place to build such a system safely. He argues that legislation to prevent RSI is an unnecessary impediment to progress, especially when the technology can be managed effectively.

Practical tests of Flint show it to be fast and highly capable. On a Mac with 48 GB of RAM, the model handled various coding requests without error, outperforming larger, slower models in speed and accuracy. It successfully created command-line tools, web-scraping programs, and simple websites explaining RSI. Other recent work from Microsoft, Carnegie Mellon, and the University of Virginia suggests that fully recursive systems are feasible, showing how AI agents can improve their own software harnesses and model weights. For now, RSI appears to be a promising and practical method for improving AI models, with the potential to make advanced technology more accessible to a wider range of developers and organizations.