New AI model Flint shows recursive self-improvement is getting close

A new open-weight AI coding model named Flint suggests that the concept of fully recursive self-improvement, where AI builds better AI, is moving from theory to practice. The model comes from Recursive, a startup that is betting heavily on using artificial intelligence to advance not just its own capabilities but other areas of science. This latest release marks a significant step in a controversial technological race that has already prompted some lawmakers to consider bans.

Richard Socher, the CEO and co-founder of Recursive, stated that nearly every aspect of the model's pipeline was handled and improved by AI. The company utilized an internal system called the Eureka Machine to automate the majority of the development process. Socher claims this allowed a small team of researchers to construct a world-class model in a matter of months. The Eureka Machine reportedly devised several innovations, including a new training method for the base model, novel architectures, and code optimized for specific hardware.

Benchmark scores shared by Recursive indicate that Flint outperforms other models of similar size and architecture on coding tasks. This is notable because AI is already used to assist in building and testing other models, but creating a loop where the AI decides its own improvements is a more radical step. Currently, this technique is limited, but if AI can contribute to every stage of development, the process could theoretically continue without end. This potential for indefinite improvement is what drives both the excitement and the fear surrounding the technology.

Concerns about recursive self-improvement gained widespread attention earlier this year. In September, a researcher left Anthropic to warn that advances in AI, including this specific approach, could lead to human extinction within a few years. This warning was echoed by other experts, and the controversy has led to several bills in Congress proposing to ban the practice. Recursive views these fears as more science fiction than reality and believes that legislation to prevent recursive self-improvement is an unnecessary obstacle to progress.

Socher argues that Flint was built using a system that cannot escape human control. He explains that while the goal is for the Eureka Machine to achieve full recursive self-improvement, humans will still oversee the process and can stop it if something goes wrong. He hopes the model demonstrates that the necessary steps are in place to build such a system safely. The company has raised $650 million at a valuation of $4.65 billion, with backing from Google Ventures, Greycroft, Nvidia, and AMD Ventures. Socher previously served as chief scientist at Salesforce, and his co-founders include Jeff Clune, a professor at the University of British Columbia who studies AI systems capable of indefinite learning.

Testing Flint on a Mac with 48 GB of RAM revealed a model with 40 billion parameters. This is a modest size compared to the largest models, which have trillions of parameters and require industrial server racks. However, in practical use, the model proved to be fast and highly capable. Models of this size are often error-prone or slow compared to larger rivals, but Flint handled various coding requests accurately. These included creating a command-line to-do list, building a program to search the web for research on recursive self-improvement, and generating a simple website explaining how the technology works.

It remains unclear how much progress other US labs have made in this area, but recent published work suggests fully recursive systems are feasible. 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. This involves modifying both the software harness that turns reasoning into action and the model's own weights. While there are risks associated with building a godlike intelligence, the current approach seems promising for improving AI models.

Socher believes the approach used to build Flint will make cutting-edge AI more accessible to smaller teams. He noted that it is the only way a small group can produce a model better than any other in its class. He hopes this will show that recursive self-improvement is a positive force that will benefit many people. The development highlights a shift in how AI is created, moving away from purely human-driven engineering toward systems that are capable of self-refinement under human supervision.