Recursive Superintelligence, a new AI startup co-founded by Richard Socher, has emerged from stealth mode with $650 million in funding and a mission to create a recursively self-improving AI model.
Team and Vision
Joining Socher in this venture are prominent AI researchers, including Peter Norvig and Cresta co-founder Tim Shi. Their goal is to develop an AI model capable of autonomously identifying and fixing its own weaknesses, marking a significant step in AI research. "Our main focus is to build truly recursive, self-improving superintelligence at scale," Socher explained.
Technical Approach
The startup's approach leverages open-endedness, a concept emphasized by Tim Rocktäschel, another co-founder. Rocktäschel, who led similar efforts at Google DeepMind, cites the world model Genie 3 as an example. "In biological evolution, animals adapt to the environment, and then others counter-adapt," Socher stated, drawing parallels to AI development.
Rocktäschel's rainbow teaming idea, which involves two AIs co-evolving through iterative testing, is also central to the startup's strategy. This method aims to improve AI safety by ensuring systems do not produce harmful outputs.
Product and Future Plans
Despite Recursive Superintelligence's research focus, Socher does not consider the startup merely a neolab. "I feel like we’re not just a lab. I want us to become a really viable company," he said, indicating plans to deliver products in the near future. Socher mentioned that product timelines might be accelerated due to the team's progress, with launches expected within quarters, not years.
Compute and Resource Allocation
Socher also highlighted the importance of computational resources in AI advancement. "Compute is not to be underestimated," he remarked, suggesting that future resource allocation decisions will be critical. As AI systems improve, the question will be how much compute should be dedicated to various global challenges.
The emergence of Recursive Superintelligence highlights a growing trend in AI startups focusing on self-improving systems, marking a new chapter in the field's development.
Source: https://techcrunch.com/2026/05/14/what-happens-when-ai-starts-building-itself/
