Recursive's New AI Model Demonstrates Rapid Self-Improvement Capabilities
A startup with a small team has released an open-weight coding model that leverages automated AI processes to outperform larger competitors.
By Miguel Hernandez ·
The concept of artificial intelligence building better artificial intelligence has moved from theoretical speculation to practical application with the release of a new coding tool. This development marks a significant step toward fully recursive self-improvement, a process where AI systems are used to design, train, and optimize subsequent versions of themselves.
The model, named Flint, is the first product from a company called Recursive. According to benchmarks released by the firm, Flint outperforms other models of similar size and architecture on specific coding tasks. The system was built with only 40 billion parameters, a modest size compared to the trillions of parameters found in the largest industrial models, yet it operates efficiently on consumer hardware such as a Mac with 48 GB of RAM.
Richard Socher, the CEO and co-founder of Recursive, explained that the company utilized an internal system known as the Eureka Machine to automate nearly the entire development pipeline. This approach allowed a small team of researchers to produce a world-class model in just a few months. The Eureka Machine handled key innovations, including new training methods for the base model, novel architectural designs, and code optimization for specific silicon chips.
While AI is already common in the software development process, this instance represents a shift toward a loop where the AI determines its own improvements. Socher stated that the current system is designed to remain under human oversight, with humans retaining the ability to stop the process if it deviates. He aims for the Eureka Machine itself to eventually achieve full recursive self-improvement, though he views the current stage as a proof of concept that all necessary steps are in place.
The financial backing behind this effort is substantial. Recursive has raised $650 million at a valuation of $4.65 billion. Investors include Google Ventures, Greycroft, Nvidia, and AMD Ventures. The founding team includes Jeff Clune, a professor at the University of British Columbia who researches AI systems capable of indefinite learning, alongside Socher, who previously served as the chief scientist at Salesforce.
Despite the progress, the broader tech community remains divided on the safety of recursive self-improvement. Some experts warn that AI capable of improving indefinitely could potentially escape human control, a concern highlighted in September by a former Anthropic researcher who suggested such advances could lead to human extinction within a few years. Several bills in Congress have proposed banning the practice, though Recursive views these legislative efforts as unnecessary hurdles that may slow down technological progress.
Socher argues that this methodology makes cutting-edge AI more accessible to smaller teams. By automating the development process, a small group can produce results that surpass larger competitors. He believes this approach will benefit a wide range of users by making high-performance AI models easier to build and deploy, even if the long-term risks of fully autonomous AI improvement remain a subject of debate.
Source: Wired
