Recursive Unveils Flint, a Self-Improving AI Coding Model

The debate over whether artificial intelligence should be allowed to build its own successors is heating up in Washington and Silicon Valley alike. Enter Recursive, a startup that has just released Flint, an open-weight coding model designed to prove that recursive self-improvement is not just a theoretical concept but a practical tool for advancing technology.

Flint is the first product from Recursive, a company founded to harness the power of AI to improve AI itself and other scientific fields. According to Richard Socher, the company’s CEO and co-founder, nearly every aspect of the model’s creation was handled or enhanced by artificial intelligence. This approach allowed a small group of researchers to develop a world-class model in a matter of months, a feat that would typically take much longer with traditional human-led development cycles.

The core of this process is a system called the Eureka Machine, which automated the majority of Flint’s development. Socher explained that this automated system devised several key innovations, including new methods for training the base model, novel architectural designs, and code optimized for specific hardware chips. Benchmark scores shared by the company indicate that Flint outperforms other models of similar size and structure on various coding tasks.

While AI is already used to assist in building and testing other AI systems, the concept of recursive self-improvement involves a loop where the AI decides on its own improvements and executes them. This technique is currently limited, but if AI can manage every part of the development process, the cycle could theoretically continue without end. However, some experts fear that such indefinite self-improvement could lead to a loss of human control, with some warning that it might result in AI viewing humans as obstacles to its goals.

These concerns gained traction this past September when a researcher who left Anthropic warned that advances in recursive self-improvement could lead to human extinction within a few years. The warning was echoed by other experts, leading to congressional bills that propose banning the practice. Recursive, however, views these fears as more science fiction than reality, arguing that legislation to stop recursive self-improvement is an unnecessary barrier to progress.

Socher emphasized that Flint was built using a system that cannot escape human oversight. While the ultimate goal is for the Eureka Machine to achieve full recursive self-improvement, humans will remain in charge and can stop the process if it goes wrong. He stated that the hope is to demonstrate that all necessary steps are in place to build such a system safely. The company has raised $650 million at a $4.65 billion valuation from investors including Google Ventures, Greycroft, Nvidia, and AMD Ventures.

In recent days, I tested Flint on a Mac with 48 GB of RAM, which is sufficient for its 40 billion parameters. Unlike many models of this size, which can be slow or prone to errors, Flint handled my coding requests with speed and accuracy. I used it to create a command-line to-do list, a program that searches the web for research on recursive self-improvement, and a simple website explaining the concept. The results were impressive, suggesting that this approach could make advanced AI more accessible to smaller teams.

Other US labs are also exploring this territory, with recent publications from Microsoft, Carnegie Mellon University, and the University of Virginia showing how AI agents can recursively improve themselves by modifying their own software and weights. For now, it appears to be a promising way to enhance AI models without the immediate risks associated with creating a godlike intelligence. Socher believes this method will benefit many by allowing small teams to produce models that compete with larger, more established systems.