Recursive's Flint Model Shows AI-Building-AI Is Gaining Ground
A startup called Recursive has released its first AI coding model, Flint, and says most of its development was handled by AI itself. CEO Richard Socher told Wired that "nearly every aspect of the...

A startup called Recursive has released its first AI coding model, Flint, and says most of its development was handled by AI itself. CEO Richard Socher told Wired that "nearly every aspect of the entire model pipeline was actually done and improved upon by AI," with the company's Eureka Machine automating most of the work. According to Socher, the system devised a new way of training the base model, new architectures, and code optimized for specific chips. Recursive shared benchmark scores indicating Flint beats other models of similar size on coding tasks.
The approach, known as recursive self-improvement, or RSI, has drawn concern from experts who warn that AI improving itself could escape human control. The debate intensified in September when a researcher quit Anthropic and warned that advances including RSI could lead to human extinction within a few years, a concern other experts echoed. Several bills in Congress now propose banning RSI. Recursive calls those fears more sci-fi than reality and argues such legislation would needlessly block progress.
Socher says Flint was built with a system that cannot escape human control, and that humans could stop the Eureka Machine if something went wrong. "We hope that this shows people that indeed we have all the steps in place," he said. Recursive has raised $650 million at a $4.65 billion valuation from Google Ventures, Greycroft, Nvidia, and AMD Ventures. Socher was previously chief scientist at Salesforce, and cofounder Jeff Clune is a University of British Columbia professor who researches AI systems that can learn indefinitely.
In hands-on testing on a Mac with 48 GB of RAM, Flint-a 40-billion-parameter model-handled coding requests quickly and accurately, building a command-line to-do list, a web research tool, and a simple website explaining RSI. Socher said the method lets a tiny team produce a model better than any other in its class: "This will hopefully show people that [recursive self-improvement] is actually quite nice and will benefit a lot of people."
Source: Wired
Ready to Modernize Your Business?
Get your AI automation roadmap in minutes, not months.
Analyze Your Workflows →