Recursive AI Startup Demonstrates Automated Model Development, Escalating RSI Debate
Recursive has built Flint, a coding model developed primarily by AI using its own recursive self-improvement system, marking a step toward fully autonomous AI development.
What happened
Recursive, a startup founded by former Salesforce chief scientist Richard Socher, released Flint, a 40-billion-parameter open-weight coding model. The company used its system called the Eureka Machine to automate nearly every aspect of Flint's development—including training methods, model architectures, and chip-optimized code. Benchmark testing shows Flint outperforms other models of similar size on coding tasks. Recursive has raised $650 million at a $4.65 billion valuation from Google Ventures, Greycroft, Nvidia, and AMD Ventures. The approach enabled a small team to produce a competitive model in months, according to Socher.
Context
Recursive self-improvement (RSI)—using AI to build and improve AI autonomously—has become increasingly central to AI development strategy because it can accelerate model refinement and reduce human engineering bottlenecks. The technology is theoretically capable of creating indefinite feedback loops where AI identifies improvements, implements them, and repeats without significant human intervention. This accessibility benefit is significant: Socher argues the technique is the only way a small team can compete with large labs. However, RSI has also triggered serious concern among AI safety researchers. In September, an Anthropic researcher publicly warned that recursive self-improvement could lead to human extinction within years, sparking broader expert concern and prompting multiple Congressional bills proposing RSI bans. Recursive disputes these existential risk framings as speculative and views safety regulation as a barrier to progress, though Socher asserts Flint's system includes human oversight controls.
What's disputed
The safety and controllability of recursive self-improvement systems are contested. Recursive claims systems like the Eureka Machine can be kept under human control, while some AI safety experts warn RSI could produce AI capable of escaping human oversight. The likelihood and timeline of such scenarios are not established in these sources.