Richard Socher, a prominent figure in artificial intelligence best known for founding the early chatbot startup You.com and his earlier work on ImageNet, has launched a new venture called Recursive Superintelligence. The San Francisco-based startup emerged from stealth on January 14, 2026, backed by $650 million in funding. Socher is joined by notable AI researchers including Peter Norvig and Tim Shi, co-founder of Cresta. The company's ambitious goal is to create what researchers have long considered the holy grail of AI development: a recursively self-improving AI model capable of autonomously identifying its own weaknesses and redesigning itself to address them without human intervention. Socher emphasized in a post-launch interview that the startup's approach differs fundamentally from other labs pursuing similar goals. While many companies claim to achieve self-improvement by simply asking AI to make something better—whether that's a machine learning system or a written document—Socher argues this isn't true recursive self-improvement. "Our main focus is to build truly recursive, self-improving superintelligence at scale, which means that the entire process of ideation, implementation, and validation of research ideas would be automatic," he explained. The system would initially focus on automating AI research ideas, with the potential to eventually expand into other research domains, including physical ones. A key component of Recursive Superintelligence's strategy involves what the team calls "open-endedness," a concept developed in part by co-founder Tim Rocktäschel, who previously led open-endedness and self-improvement teams at Google DeepMind. Rocktäschel specifically worked on Genie 3, which the company considers a strong example of open-ended systems. Socher stressed that while AI working on AI itself offers particularly powerful possibilities, the ultimate vision extends beyond software optimization to include a new form of self-awareness regarding the system's own limitations and capabilities.