What happens when AI starts building itself?

Richard Socher, a notable figure in the AI landscape and founder of the chatbot startup You.com, has embarked on a new venture. He recently announced Recursive Superintelligence, a San Francisco-based startup that has secured an impressive $650 million in funding. Joining him are esteemed AI researchers such as Peter Norvig and Tim Shi, co-founder of Cresta. This ambitious initiative aims to develop an AI model capable of recursive self-improvement—an AI that can autonomously recognize and rectify its own deficiencies without human intervention. This goal has long been sought after in the AI research community and represents a significant leap in the field. In a recent Zoom interview, Socher elaborated on Recursive's innovative technical strategy, emphasizing that their approach is set apart by its focus on open-endedness, a concept that has yet to be fully realized in AI development. While many assume that merely enhancing AI through automated research constitutes recursive self-improvement, Socher clarifies that true recursive self-improvement encompasses a comprehensive process of generating, implementing, and validating research ideas autonomously. Socher described their vision as being particularly potent when AI is applied to its own development, allowing it to cultivate a newfound self-awareness regarding its limitations. The concept of open-endedness, led by co-founder Tim Rocktäschel, has roots in biological evolution, where continuous adaptation leads to complex developments over time. He introduced the idea of 'rainbow teaming,' a method that allows two AI systems to co-evolve by challenging each other iteratively. This technique not only enhances AI safety but also promotes diverse problem-solving approaches, making the AI more robust. When asked about the completion of their project, Socher acknowledged that the journey of AI improvement is ongoing, with no finite endpoint. He believes that as intelligence evolves, the potential for advancement remains vast. He also expressed a desire to distinguish Recursive Superintelligence from other AI labs, asserting that their commitment to open-endedness and innovation positions them uniquely within the industry. Socher has ambitious plans for product development, suggesting that the team might accelerate their timelines due to significant progress. He anticipates launching products within quarters rather than years, emphasizing the importance of computational resources in their endeavor. The ultimate challenge will be determining how much computational power humanity is willing to allocate to solve pressing global issues, such as diseases and environmental challenges.