Recursive Superintelligence Emerges: Richard Socher's Ambition for AI's Next Frontier

Richard Socher, a significant figure in artificial intelligence, has launched a new startup, Recursive Superintelligence, in San Francisco with a substantial $650 million in funding. This venture aims to advance recursively self-improving AI models, a challenging goal within AI research.

• Who: Richard Socher, alongside noted AI researchers such as Peter Norvig and Tim Shi.
• What: Launch of Recursive Superintelligence, targeting development of AI models that can autonomously improve by identifying and rectifying their own weaknesses.
• Where: San Francisco, United States.
• When: The startup emerged from stealth on Wednesday (date unspecified).
• Why: The goal is to achieve autonomous AI self-improvement at scale, creating systems that can evolve beyond current capabilities without human intervention.

During a Zoom interview, Socher distinguishes their approach through the use of "open-endedness" to achieve recursive self-improvement, which remains unachieved in the AI industry. Co-founder Tim Rocktäschel's work at Google DeepMind provides technical foundations through concepts like evolution-inspired 'rainbow teaming,' which leverages AI-on-AI interaction for continuous improvement. This process is likened to biological evolution, fostering capabilities as seen in natural systems.

The startup, although inspired by the research-focused 'neolab' movement, aspires to eventually release impactful products. Socher anticipates market-ready products in the coming quarters rather than years, emphasizing the crucial role of computational resources in advancing AI technologies.