A new wave in artificial intelligence is taking shape as San Francisco-based startup Recursive Superintelligence emerges from stealth with a massive $650 million funding round. The company, co-founded by AI veteran Richard Socher, is building systems designed not just to learn—but to improve themselves without human involvement.
Socher, known for his early work in AI and as the founder of You.com, is joined by leading researchers including Peter Norvig and Tim Shi. Together, they aim to push AI beyond automation into what they describe as “recursive self-improvement,” where AI systems can identify their own weaknesses, redesign themselves, and validate improvements independently.
The startup’s approach centers on “open-endedness,” a concept inspired by evolution and long-term adaptive systems. According to Socher, the goal is to create AI that continuously generates and tests new ideas across research, eventually expanding beyond digital tasks into physical-world applications. This could allow AI to iteratively refine itself at scale, similar to biological systems evolving over time.
A key part of their strategy involves using multiple AI systems in competitive loops—sometimes described as “rainbow teaming”—where one AI attempts to challenge or break another. This process helps identify weaknesses and strengthen the model through repeated iteration, making systems safer and more robust over time.
While the long-term vision raises questions about limits and control, Socher argues that compute power—not human labor—will become the most important resource in future AI development. He suggests society may eventually decide how computational power is allocated to solve major global problems, from disease to climate challenges. The company says its first products could arrive within “quarters, not years,” signaling rapid progress ahead.
source: techcrunch