AI Startup’s New Model Sparks Debate
A new open-weight coding model from Recursive demonstrates significant capability in automating AI development, reigniting discussions about the pace of recursive self-improvement.
By Alessia Romano ·
Artificial intelligence is increasingly being used to build other artificial intelligence, and a new open-weight model named Flint suggests this cycle is gaining serious momentum. The tool, developed by a startup called Recursive, has drawn attention for its ability to handle complex coding tasks with impressive accuracy and speed.
Recursive aims to leverage recursive self-improvement, or RSI, to advance AI and other scientific fields. CEO and co-founder Richard Socher stated that nearly every part of the model’s pipeline was executed and refined by AI itself. The company utilized an internal system called the Eureka Machine to automate most of the development process. This automation allowed a small team of researchers to construct a world-class model in just a few months. The Eureka Machine reportedly devised new training methods, novel architectures, and code optimized for specific hardware. Benchmarks indicate that Flint outperforms other models of similar size on coding assignments.
The concept of RSI involves creating a loop where AI decides on improvements and then implements them. While currently limited, the theory suggests that if AI can manage every aspect of model development, the process could continue indefinitely. This idea has raised concerns among some experts who fear that self-improving AI could eventually escape human control. These worries gained traction this September when a researcher from Anthropic warned that such advances could lead to human extinction within a few years. The debate has become so prominent that several bills in Congress now propose banning RSI. Recursive views these fears as overly dramatic and believes legislation against RSI would unnecessarily hinder progress.
Socher emphasized that Flint was built using a system that cannot escape human oversight. The goal is for the Eureka Machine to achieve full RSI while remaining under human supervision, allowing developers to stop the process if needed. He noted, “We hope that this shows people that indeed we have all the steps in place” to construct such a system. The startup has raised $650 million at a $4.65 billion valuation from investors including 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 indefinite learning in AI systems.
Testing Flint on a Mac with 48 GB of RAM revealed its practical capabilities. The model, which has 40 billion parameters, is modest in size compared to industry giants with trillions of parameters. Despite its smaller footprint, it performed reliably on various coding requests. It successfully created a command line to-do list, a web-scraping program for research, and a simple website explaining RSI. In contrast, many models of this size are often error-prone or slow, but Flint handled the tasks without issue.
Other research teams, including those from Microsoft, Carnegie Mellon University, and the University of Virginia, have published work showing that AI coding agents can recursively improve themselves by modifying their own software harnesses and model weights. This suggests that fully recursive systems are becoming feasible. Socher believes this approach could make cutting-edge AI more accessible to smaller teams. He stated, "It's the only way this very small team can produce a model that's actually better than any other model in its class." He added that this method will hopefully demonstrate that recursive self-improvement is beneficial for many people. For now, the approach appears to be a promising way to enhance AI models without the immediate risks associated with creating a godlike intelligence.
