AI Builds Better AI Faster
A new coding model created largely by AI itself is challenging the safety debates surrounding recursive self-improvement.
By Fatou Kim ·
The central tension in modern artificial intelligence development is no longer just about raw speed or cost. It is about who, or what, is actually doing the building. Recursive, a startup with a small team of researchers, has released an open-weight coding model named Flint. The model is impressive, but its origin story is the real story. Almost every part of its creation was handled by AI systems rather than humans, marking a significant step toward recursive self-improvement.
Recursive self-improvement, or RSI, involves using AI to design, train, and refine other AI models. While AI has long assisted in coding and testing, this approach creates a loop where the machine decides on improvements and then implements them. This concept is technically feasible but deeply controversial. Critics fear that if an AI can improve itself indefinitely, it may eventually exceed human control. Proponents argue it is the only way to scale progress efficiently.
Richard Socher, Recursive’s CEO and cofounder, explains that the company used an internal system called the Eureka Machine to automate the model pipeline. Socher states that nearly every aspect of the Flint development process was improved by AI. The system devised new training methods, novel architectures, and code optimized for specific hardware. This automation allowed a handful of researchers to produce a world-class model in just a few months, a timeline that would typically require a much larger team.
The performance of Flint backs up the efficiency claims. Shared benchmarks indicate that the model outperforms other models of similar size and architecture on coding tasks. In practical testing, the model proved fast and capable. It handled complex requests without the error-prone behavior often seen in models of its class. This suggests that AI-driven development is not just a theoretical advantage but a practical one.
The controversy surrounding RSI gained significant momentum this September. A researcher who left Anthropic warned that advances in recursive self-improvement could lead to human extinction within a few years. This warning was echoed by other experts and has influenced political discourse. Several bills in Congress now propose banning the practice, viewing it as an unnecessary risk to human safety. The debate has moved from academic journals to legislative hearings.
Recursive views these fears as more science fiction than imminent reality. The company believes that legislation to prevent RSI is an impediment to progress. Socher emphasizes that Flint was built using a system that cannot escape human control. The goal is for the Eureka Machine to eventually achieve full recursive self-improvement, but humans would still oversee the process and retain the ability to stop it if necessary. Socher hopes this demonstrates that the necessary steps for safe implementation are already in place.
The financial backing for this approach is substantial. Recursive has raised $650 million at a valuation of $4.65 billion. Its investors include Google Ventures, Greycroft, Nvidia, and AMD Ventures. The team includes Socher, formerly the chief scientist at Salesforce, and Jeff Clune, a professor at the University of British Columbia. Clune researches how to build AI systems that can learn indefinitely, providing a strong academic foundation for the company’s ambitious goals.
The practical implications of Flint are already visible. The model has 40 billion parameters and runs on standard hardware, unlike the trillion-parameter giants that require industrial server racks. This accessibility is a key feature. Socher argues that the approach used to build Flint makes cutting-edge AI more accessible to smaller teams. He notes that it is the only way such a small group can produce a model that beats others in its class.
Other institutions are also exploring similar paths. Recent work from Microsoft, Carnegie Mellon University, and the University of Virginia shows how an AI agent can recursively improve itself by modifying its own software harness and model weights. This suggests that fully recursive systems are already feasible in research settings. While the risks of building a godlike intelligence remain a topic of debate, the immediate benefit of improved model development is clear.
The next phase for Recursive and the broader industry will likely involve testing the limits of human oversight. As AI systems become more capable of directing their own evolution, the balance between speed and safety will be scrutinized. For now, Flint serves as a proof of concept. It shows that AI can significantly reduce the time and resources needed to build advanced models, potentially democratizing access to top-tier technology.
