AI System Improves Its Own Model

A new AI coding tool is drawing attention for how it was built. The model, named Flint, was created by a system that improved itself during development. This approach is known as recursive self-improvement, or RSI.

Flint has 40 billion parameters. It can run on a high-end personal computer with 48 GB of RAM. Recursive CEO Richard Socher says the company's system, called the Eureka Machine, automated most steps. This allowed a team of a few researchers to build the model in a few months. The Eureka Machine created new training methods, model designs, and code for specific chips. Benchmark scores shared by Recursive show Flint performs better than similar models on coding tasks.

RSI uses AI to improve AI. This can speed up the process of building new models. The loop could theoretically continue indefinitely if AI handles every step. Some experts worry about this. They fear AI could escape human control if it can improve itself without limits. In September, a researcher left Anthropic and warned that this could lead to human extinction. Several bills in Congress now propose to ban RSI.

Recursive says these fears are more science fiction than reality. Socher states Flint was built with a system that cannot escape human control. The goal is for the Eureka Machine to achieve full RSI, but humans would oversee it. “We hope that this shows people that indeed we have all the steps in place” to build such a system, he says. Socher was previously chief scientist at Salesforce. His cofounder, Jeff Clune, is a professor at the University of British Columbia.

Recursive has raised $650 million at a $4.65 billion valuation. Investors include Google Ventures, Greycroft, Nvidia, and AMD Ventures. The company believes RSI can make advanced AI more accessible. “It's the only way this very small team can produce a model that's actually better than any other model in its class,” Socher said. The use of RSI to build models remains a controversial topic, with some experts raising alarms about the potential risks.