Recursive's Flint Model Challenges Congressional Calls to Ban AI Self-Improvement
A new open-weight coding model built by a small startup using automated AI systems is performing at levels that complicate the argument for immediate legislative bans on recursive self-improvement.
By Yasmin Farouk ·
The debate over whether artificial intelligence should be allowed to build better versions of itself has moved from theoretical forums to practical demonstration with the release of Flint, a new coding model from a startup called Recursive. The model's performance suggests that recursive self-improvement, or RSI, is no longer just a speculative concept but a functional tool for accelerating development.
The path to Flint’s creation highlights a shift in how AI is developed. Recursive CEO Richard Socher stated that nearly every aspect of the model pipeline was handled and improved by AI. This process utilized a system known as the Eureka Machine, which automated key elements of development. As a result, a small team of researchers managed to create a world-class model in a matter of months. The Eureka Machine contributed specific innovations, including new training methods for the base model, novel architectures, and code optimized for specific hardware chips.
This development arrives at a tense moment in the AI landscape. In September, a researcher who left Anthropic warned that advances in RSI could lead to human extinction within a few years, a concern echoed by other experts. The controversy has reached the halls of Congress, where several bills have been proposed to ban the practice. While some view this as a necessary safety measure, Recursive argues that such fears are more science fiction than reality and that legislation would unnecessarily hinder progress.
Socher maintains that Flint was built using a system that remains under human control. The long-term goal is for the Eureka Machine to achieve full recursive self-improvement, but this process would be overseen by humans who retain the ability to stop it if it goes awry. He believes the current release demonstrates that the necessary steps to build such a controlled system are already in place.
Financial backing supports this ambitious vision. Recursive has raised $650 million at a $4.65 billion valuation from investors including Google Ventures, Greycroft, Nvidia, and AMD Ventures. Socher, formerly the chief scientist at Salesforce, is joined by cofounder Jeff Clune, a University of British Columbia professor who researches AI systems capable of indefinite learning.
Testing Flint on a Mac with 48 GB of RAM revealed a model with 40 billion parameters that is both fast and highly capable. Unlike other models of similar size that are often slow or error-prone, Flint successfully handled various coding requests without issue. This performance aligns with recent published work from Microsoft, Carnegie Mellon University, and the University of Virginia, which suggests that fully recursive systems are already feasible. Socher argues that this approach could make cutting-edge AI more accessible, allowing small teams to produce models that outperform larger competitors in their class.
As other U.S. labs continue to chase similar goals, the practical success of Flint adds weight to the argument that RSI is a promising tool for improvement. While risks remain regarding the potential for uncontrolled intelligence, the current evidence suggests that human oversight can manage the process effectively, at least for now.
Source: Wired
