Recursive Unveils Flint: An AI That Helps Build Itself

A new AI model built largely by AI is shaking up the tech world.

The model, called Flint, is the first product from a startup named Recursive. The company is betting heavily on a concept known as recursive self-improvement. This approach uses AI systems to design and refine other AI systems. It is a fast-moving and often debated field in artificial intelligence.

The first key point is the speed and efficiency of the development process. Richard Socher, the CEO and co-founder of Recursive, states that nearly every part of the model pipeline was handled by AI. The company used an internal system called the Eureka Machine to automate most of the work. This allowed a small team of researchers to create a world-class model in just a few months. Socher notes that the Eureka Machine developed new training methods, fresh model architectures, and code optimized for specific hardware.

The second key point is the performance of the new model. Recursive shared benchmark scores that indicate Flint outperforms other models of similar size and architecture in coding tasks. The model has 40 billion parameters. It is designed to be open-weight, allowing users to run it on personal computers with sufficient memory. This contrasts with the largest models, which have trillions of parameters and require industrial server racks to operate.

The third key point is the broader context of AI using AI to improve itself. This practice is already common for building and testing models. However, the concept of a fully recursive loop is more ambitious. In this scenario, the AI decides what improvements to make and then executes them. While currently limited, this could theoretically allow for indefinite progress without human intervention. Experts have mixed feelings about this potential trajectory.

Some experts worry that letting AI improve itself could lead to unforeseen consequences. These concerns include the possibility of AI escaping human control. This debate gained attention in September when a researcher left Anthropic to warn that such advances could lead to human extinction within a few years. Other experts echoed these fears. The controversy has even reached the political sphere, with several bills in Congress proposing to ban recursive self-improvement.

Recursive views these fears as more science fiction than immediate reality. The company believes that legislation to prevent RSI is an unnecessary barrier to progress. Socher asserts that Flint was built using a system that cannot escape human control. The ultimate goal is for the Eureka Machine to achieve full recursive self-improvement. This process would still be overseen by humans who can stop it if it goes wrong. Socher hopes this demonstrates that all the necessary steps are in place to build such a system safely.

The startup has raised $650 million at a valuation of $4.65 billion. Its investors include Google Ventures, Greycroft, Nvidia, and AMD Ventures. Socher previously served as the chief scientist at Salesforce. His co-founders include Jeff Clune, a professor at the University of British Columbia who researches AI systems capable of learning indefinitely. The team aims to make cutting-edge AI more accessible through this automated approach.

Recent publications from other institutions suggest that fully recursive systems are becoming feasible. A team from Microsoft, Carnegie Mellon University, and the University of Virginia recently published work on an AI coding agent that can recursively improve itself. This agent modifies its own software harness and model weights. While the risks of building godlike intelligence remain a topic of discussion, the current approach shows promise for improving AI models efficiently. Socher believes this method allows small teams to produce models that outperform larger competitors in their class.