Recursive’s AI-Built Model Challenges Industry Skeptics
A small team’s new open-weight coding model, largely constructed by an automated AI system, has delivered performance that rivals much larger rivals, fueling debate over the speed of recursive self-improvement.
By Malcolm Davis ·
Tension is rising in the artificial intelligence sector as a small startup claims to have produced a world-class coding model using a process that relies heavily on AI to build AI. The core of the conflict lies in the method: Recursive, a company founded by Richard Socher, has released a model named Flint that was largely developed by its own recursive system, dubbed the Eureka Machine. While this approach promises to accelerate development, it has intensified a fierce debate among experts regarding the safety and feasibility of recursive self-improvement (RSI), a concept that some view as a critical step toward superintelligence and others as a potential path to disaster.
Flint is an open-weight model with 40 billion parameters, a size that is modest compared to the trillions of parameters found in the industry’s largest systems. According to Socher, the Eureka Machine automated nearly every aspect of the model pipeline, from devising new training methods for the base model to optimizing code for specific silicon chips. This automation allowed a team of just a handful of researchers to build a competitive product in a matter of months. Recursive shared benchmark scores indicating that Flint outperforms other models of similar size and architecture on coding tasks, a claim that has drawn significant attention from the technical community.
The controversy surrounding RSI is not new, but it has gained momentum recently. This September, a researcher departed Anthropic to warn that advances in recursive self-improvement could lead to human extinction within a few years, a sentiment echoed by other experts. These concerns have translated into political action, with several bills currently in Congress proposing to ban RSI. In contrast, Recursive views these fears as overly dramatic, arguing that legislation to prevent RSI is an unnecessary hurdle. Socher maintains that Flint was built using a system that cannot escape human control, with the long-term goal of having the Eureka Machine achieve full RSI under human oversight that can halt the process if necessary.
Independent testing of Flint reveals a model that is both fast and capable. Running on a Mac with 48 GB of RAM, the model handled coding requests with precision, successfully creating a command-line to-do list, a web-scraping program for research, and a simple website explaining RSI. The tester noted that models of this size are often error-prone or slow compared to larger competitors like Claude or Codex, but Flint consistently met expectations. This performance suggests that the recursive approach may make cutting-edge AI more accessible to smaller teams.
Background on the company reveals significant financial backing, with Recursive having raised $650 million at a $4.65 billion valuation from investors including Google Ventures, Greycroft, Nvidia, and AMD Ventures. Socher, previously the chief scientist at Salesforce, co-founded the venture with Jeff Clune, a professor at the University of British Columbia who researches AI systems capable of indefinite learning. The move aligns with recent academic work from Microsoft, Carnegie Mellon University, and the University of Virginia, which demonstrated how AI coding agents can recursively improve their own software harnesses and model weights.
What happens next remains uncertain, though the release of Flint signals that RSI is moving from theoretical discussion to practical application. While the risks of building a godlike intelligence remain a topic of concern, the immediate benefit appears to be a more efficient path to high-quality AI models. For now, the industry is watching closely to see if this small-team, AI-driven methodology can scale without triggering the very dangers that critics have warned about for years.
Socher believes this new approach will benefit many by showing that RSI is a viable and useful tool. He stated that the recursive method is the only way his small team could produce a model better than others in its class, hoping to demonstrate that the technology is 'quite nice' and accessible.
Source: Wired
