New AI coding model built by AI itself hits the market
A startup backed by major tech investors has released a high-performing coding model that was largely developed using its own automated systems, signaling a shift in how artificial intelligence is created.
By Benjamin Rousseau ·
The landscape of artificial intelligence development is shifting as companies begin to rely less on human labor for model creation and more on automated systems. This trend was highlighted recently by the release of Flint, a new open-weight AI coding model that suggests recursive self-improvement is becoming a viable commercial strategy.
Flint is the debut product of Recursive, a company founded to leverage AI in building better AI. The model has 40 billion parameters, a size that allows it to run on standard high-end laptops rather than massive server farms. Despite its modest size, benchmarks indicate it outperforms other models of similar scale on coding tasks.
The development process behind Flint was largely automated by a system called the Eureka Machine. This internal tool handled most aspects of the model pipeline, from training methods to architecture design. It also optimized the code for specific silicon chips, allowing a small team of researchers to build a competitive model in just a few months.
Recursive has raised $650 million at a valuation of $4.65 billion. The funding came from prominent investors including Google Ventures, Greycroft, Nvidia, and AMD Ventures. The company is led by Richard Socher, who previously served as chief scientist at Salesforce. His cofounders include Jeff Clune, a professor at the University of British Columbia who studies indefinite learning systems.
Socher stated that nearly every aspect of the model pipeline was done and improved by AI. He noted that the Eureka Machine devised several innovations, including new training methods and model architectures. The goal is for the Eureka Machine itself to achieve full recursive self-improvement, though humans will still oversee the process to ensure it does not go awry.
The concept of recursive self-improvement, or RSI, has become a point of contention in the tech industry. Some experts worry that allowing AI to improve itself indefinitely could lead to a loss of human control. These concerns were amplified this September when a researcher from Anthropic warned that rapid AI advances could lead to human extinction within a few years.
Legislators have taken notice of these fears, with several bills in Congress proposing to ban RSI. Recursive views these fears as more science fiction than reality. Socher argues that current systems are incapable of escaping human control and that legislation to prevent RSI is an unnecessary impediment to progress.
Practical testing of Flint showed it to be fast and capable. The model successfully created a command line to-do list, a web scraper for research, and a simple website explaining RSI. While larger models with trillions of parameters exist, Flint demonstrates that smaller models can be highly effective when built with advanced automated techniques.
Other US labs are also exploring recursive systems. Recent work from Microsoft, Carnegie Mellon University, and the University of Virginia showed that an AI coding agent can recursively improve itself by modifying its own software harness and model weights. Socher believes this approach will make cutting-edge AI more accessible. He stated that this is the only way a small team can produce a model better than others in its class, hoping to show that RSI will benefit many people.
Source: Wired · original report published
