Recursive Emerges from Stealth with a Bold Bet on Self-Improving AI

Recursive, building self-improving superintelligence to automate knowledge discovery, led by Richard Socher, Yuandong Tian, and Caiming Xiong, has emerged from stealth and raised $650M led by GV (Google Ventures) and Greycroft, with major participation from AMD Ventures and NVIDIA at a $4.65 billion valuation to create AI that conducts experiments on how to safely improve itself, in an open-ended process of automated scientific discovery.

Open-ended evolution, both Darwinian and cultural, produced human intelligence through the accumulation of novel discoveries, each building on the last, from simple reflexes to language, science, and space exploration.

AI follows a similar pattern of iterative innovation, historically driven by human researchers.

But as compute and data scale, machine learning increasingly shifts from hand-designed methods toward AI-driven discovery processes.

Building Endlessly Self-Improving AI.

Recursive is built around the idea that the path to superintelligence will come from AI systems that improve themselves through open-ended discovery processes.

The initial focus is on AI research itself, using AI to accelerate AI development, before extending to broader scientific domains.

Safety remains central throughout, with an emphasis on maximizing benefits while reducing risks and ensuring the system supports human flourishing.