Recursive AI is advancing rapidly: What it means for the future of AI

A new startup has developed an open-weight AI coding model called Flint, which suggests that recursive self-improvement, or RSI, is getting closer to becoming a reality. This approach uses AI to build AI, and it has the potential to advance both AI itself and other areas of science using AI. The startup, called Recursive, is led by Richard Socher, who says that nearly every aspect of the model pipeline was actually done and improved upon by AI. He also says that the Eureka Machine, the company's recursive AI system, was used to automate most elements of the model's development, which enabled a company with a handful of researchers to build a world-class model in just a few months.

Socher says that the Eureka Machine devised a number of innovations, including a new way of training the base model, new model architectures, and code optimized for specific silicon chips. Recursive shared benchmark scores that suggest Flint outperforms other models of similar size and architecture on coding tasks. AI is already widely used to help build, test, and improve AI, and this can speed up the process of refining a model and testing new ideas without the need for as much input from humans. It is also possible to create a loop through which AI decides what improvements to make and then begins another round of improvement.

Some experts worry that letting AI improve itself in this manner could ultimately lead to disaster, and these fears have been echoed by a number of other experts. The idea of RSI has become so controversial that several bills in Congress propose to ban it. Recursive sees these fears as more sci-fi than reality and believes legislation to prevent RSI is an unnecessary impediment to progress.

Socher says Flint was built using a system incapable of escaping human control, and he says the goal is for the Eureka Machine itself to achieve full recursive self-improvement, but this would still be overseen by humans who could stop the process if it goes awry. He says the approach used to build Flint could make cutting-edge AI more accessible, and it's the only way this very small team can produce a model that's actually better than any other model in its class.

Recursive has raised $650 million at a $4.65 billion valuation from Google Ventures, Greycroft, Nvidia, and AMD Ventures. Socher was previously the chief scientist at Salesforce, and his cofounders include Jeff Clune, a professor at the University of British Columbia who researches ways of building AI systems capable of learning indefinitely. I must say I am impressed with Flint, and I tested it on a Mac with 48 GB of RAM. The biggest models out there have several trillions of parameters, and they can only run on racks of industrial servers.

I used Flint to build a bunch of programs and found it to be both fast and very capable. In my experience, models of this size can be annoyingly error-prone or slow compared to Claude or Codex, but Flint nailed my coding requests every time. I used it to create a command line to-do list, then tried having it build a program that scours the web for interesting research related to recursive self-improvement. I also had Flint whip up a simple website that explains how RSI works.