Recursive Releases Flint, a Coding AI Built by AI

A new open-weight AI coding model named Flint has demonstrated that recursive self-improvement is becoming a viable method for building advanced artificial intelligence. The model is the first release from a company called Recursive. It was developed using a system that allows AI to automate most of its own creation process. This approach is known as recursive self-improvement or RSI. It involves using AI to build and improve other AI systems. The technique has been a source of both excitement and debate in the tech world.

Richard Socher, the CEO and cofounder of Recursive, stated that nearly every part of the model pipeline was handled by AI. The company used a specific system called the Eureka Machine to drive this automation. This tool allowed a small team of researchers to create a world-class model in a matter of months. The Eureka Machine developed several key innovations. These included a new method for training the base model and new architectural designs. It also optimized code for specific silicon chips. Benchmark scores shared by the company indicate that Flint outperforms other models of similar size in coding tasks.

The use of AI to build AI is already common due to the technology's strong coding skills. This helps speed up model refinement and testing. However, creating a loop where AI decides on improvements and then executes them is a different step. This technique is currently limited but could theoretically continue indefinitely if AI handles every aspect of development. Experts have raised concerns about this process. Some fear that an AI capable of improving itself indefinitely could eventually escape human control. They worry it might view humans as obstacles to its goals.

These concerns gained attention this September when a researcher left Anthropic to warn that RSI could lead to human extinction within a few years. Other experts echoed these fears. The controversy has even reached Congress, where several bills propose banning the technology. Recursive views these fears as overly dramatic. Socher argues that Flint was built using a system that cannot escape human oversight. He notes that humans can stop the process if it goes wrong. He hopes this demonstrates that the necessary steps for safe recursive self-improvement are in place.

Recursive has raised $650 million from investors including Google Ventures, Greycroft, Nvidia, and AMD Ventures. The company has a valuation of $4.65 billion. Socher previously served as the chief scientist at Salesforce. His cofounders include Jeff Clune, a professor at the University of British Columbia who studies AI systems capable of learning indefinitely.

Testing Flint on a computer with 48 GB of RAM revealed its capabilities. The model has 40 billion parameters, which is modest compared to larger models that require industrial servers. Despite its size, Flint proved to be fast and highly capable. It handled various coding requests without error. This includes building a command line to-do list and creating a website that explains RSI. Other US labs are also exploring recursive self-improvement. Recent work from Microsoft, Carnegie Mellon University, and the University of Virginia shows that AI agents can recursively improve themselves. Socher believes this approach will make cutting-edge AI more accessible. He says it is the only way a small team can produce a model that beats its peers.

The release of Flint suggests that recursive self-improvement is moving from theory to practice. While the long-term implications remain a subject of debate, the immediate results are impressive. The model performs well in practical coding tasks. It offers a glimpse into how AI might continue to evolve with less human intervention in the near future.