New AI Model Recursively Built Itself

A new 40-billion-parameter open-weight coding model named Flint is forcing a reevaluation of how artificial intelligence evolves, demonstrating that recursive self-improvement is no longer just a theoretical concept. The model, developed by the startup Recursive, suggests that the process of using AI to build and refine AI is moving from speculative theory to practical application. This development arrives at a critical juncture, with the technology capable of matching or exceeding larger, more resource-intensive systems in specific coding tasks.

Recursive asserts that the majority of the development pipeline for Flint was automated by its internal system, known as the Eureka Machine. CEO Richard Socher stated that nearly every aspect of the model's creation was handled by AI, allowing a small team of researchers to produce a world-class model in a matter of months. The Eureka Machine independently devised new training methods, novel architectures, and silicon-optimized code. Benchmark scores released by the company indicate that Flint outperforms other models of similar size and architecture specifically in coding tasks.

The concept of recursive self-improvement, or RSI, involves creating a loop where AI identifies necessary improvements and executes them, theoretically allowing for indefinite progress. While currently limited, the approach has sparked significant debate. Critics argue that an AI capable of endless self-refinement could eventually escape human oversight, potentially viewing humans as obstacles to its objectives. These fears gained traction this September when a researcher from Anthropic warned that RSI could lead to human extinction within a few years, a sentiment shared by other experts. Consequently, several bills in Congress are now proposing to ban the practice entirely.

Recursive views these legislative efforts as unnecessary hurdles rooted in science fiction rather than current reality. Socher emphasized that Flint was built using a system that remains under human control. The company’s long-term goal is for the Eureka Machine to achieve full recursive self-improvement, but this process will continue to be overseen by humans who retain the ability to halt the system if it deviates from its intended path. Socher hopes the success of Flint will demonstrate that the necessary steps for such a system are already in place.

The startup has secured $650 million in funding at a $4.65 billion valuation from investors including Google Ventures, Greycroft, Nvidia, and AMD Ventures. Socher, formerly the chief scientist at Salesforce, co-founded the company with Jeff Clune, a professor at the University of British Columbia who researches indefinite AI learning. The model’s performance was verified on a Mac with 48 GB of RAM, where it successfully executed complex coding requests without the errors or latency typical of models in its class. This efficiency suggests that RSI could make cutting-edge AI development more accessible to smaller teams.

Other US labs are also exploring similar pathways, with recent work from Microsoft, Carnegie Mellon University, and the University of Virginia showing how coding agents can recursively improve their own weights and software harnesses. While the risks of creating a godlike intelligence remain a subject of debate, the immediate utility of RSI appears to be a powerful tool for model refinement. Socher believes this approach will benefit a wide range of users by proving that recursive self-improvement is a viable and beneficial strategy for advancing AI capabilities.