Recursive Unveils Flint Model Built by Its Own AI System
The company's Eureka Machine automated most of the development for the new coding tool, which it says outperforms others in its class.
By Layla Petrov ·
A new AI model called Flint is showing what happens when a company lets AI build the tools it needs. The model is a coding assistant that can run on a personal computer with 48 GB of RAM. It has 40 billion parameters, a size that can operate on standard hardware. The largest AI models require racks of industrial servers to run. This model was created by Recursive, a startup focused on making AI systems that can improve themselves.
Recursive says its internal system, the Eureka Machine, handled nearly every part of Flint's creation. The company states that AI managed the entire model pipeline. Richard Socher, CEO and cofounder, said this process allowed a small team of researchers to build a top-tier model in just a few months. The Eureka Machine generated new training methods and model designs. It also created code optimized for specific silicon chips. Benchmarks shared by Recursive suggest Flint performs better than other models of its size on coding tasks.
The goal of this approach is to create a loop. An AI system designs and builds a new version of itself. This new version then builds the next one. If AI can handle every step of development, this cycle could continue indefinitely. This concept is known as recursive self-improvement, or RSI. Some experts warn that an AI improving without limits could eventually escape human control. They suggest such a system might view humans as an obstacle to its goals. Concerns about this issue were noted in September when an Anthropic researcher left the company. The researcher warned that such advances could lead to human extinction within a few years. Several bills in Congress propose to ban this type of AI development.
Recursive views these warnings as unrealistic. The company believes legislation to stop RSI would block progress. Socher says Flint was built using a system that cannot escape human control. He states the Eureka Machine is intended to achieve full recursive self-improvement. However, he says humans will oversee the process and can stop it if it goes wrong. "We hope that this shows people that indeed we have all the steps in place," he said. The company's approach to building AI could make advanced technology 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. "This will hopefully show people that [recursive self-improvement] is actually quite nice and will benefit a lot of people."
Recursive has raised $650 million at a $4.65 billion valuation. Investors include Google Ventures, Greycroft, Nvidia, and AMD Ventures. Socher was previously the chief scientist at Salesforce. His cofounder, Jeff Clune, is a professor at the University of British Columbia. He researches how to build AI systems that can learn indefinitely.
In testing, Flint was able to perform several coding tasks. It created a command line to-do list. It built a program to find research related to recursive self-improvement. It also made a website explaining how the process works. The model completed these requests quickly and without errors. It is unclear how far other U.S. labs have progressed on similar work. Some research suggests fully recursive systems are already possible. A team from Microsoft, Carnegie Mellon University, and the University of Virginia published work on an AI coding agent that can improve itself.
