Recursive unveils Flint, an AI model that trains itself

Recursive has unveiled Flint, an open-weights AI model for code generation built on recursive self-improvement (RSI), where the AI develops itself. The company claims this approach enabled a small team to create a world-class model in months. Flint was developed using Recursive's Eureka Machine system, which automated most aspects of the model's creation, including proposing new training methods, neural network architectures, and optimized code. Benchmarks show Flint outperforms similar-sized models in coding tasks, and it is already being used to improve other AI systems. RSI is controversial, with some experts warning of catastrophic risks and US lawmakers proposing bans, though Recursive dismisses these fears. The company has raised $650 million at a $4.65 billion valuation, and CEO Richard Socher insists human oversight remains in place.