Richard Socher's Recursive Superintelligence: Building AI That Builds Itself

Richard Socher, a key player in the AI world known for founding You.com and his work on ImageNet, has recently launched Recursive Superintelligence, a San Francisco-based startup that has just received $650 million in funding. Joining him are prominent researchers like Peter Norvig and Tim Shi from Cresta.

Their mission is to develop a recursively self-improving AI model, one capable of autonomously identifying its weaknesses and redesigning itself without human involvement—a goal that has long eluded the industry.

Recursive Self-Improvement: The Unique Approach

Socher emphasizes their unique approach to achieving recursive self-improvement (RSI) through open-endedness. He explains, 'Our main focus is to build truly recursive, self-improving superintelligence at scale, which means the entire process of ideation, implementation, and validation of research ideas would be automatic.'

Socher elaborates on this concept by mentioning Genie 3 from Google DeepMind, led by Tim Rocktäschel. This world model can create any concept or agent based on user input, showcasing the potential for open-ended AI.

The Rainbow Team Approach

One of their strategies involves 'rainbow teaming,' an extension of the red-teaming concept used in cybersecurity. In this context, one AI challenges another to generate all possible negative outcomes, leading to a continuous back-and-forth process that enhances both AIs.

'This was an idea from Tim Rocktaeschel, and it’s now used in all the major labs,' Socher notes, highlighting its significance for developing safer and more robust AI systems.

Challenges Ahead

Socher acknowledges that while RSI is a monumental goal, he believes they are approaching it differently by fully embracing open-endedness. 'We really embrace the concept of open-endedness, and our team is entirely focused on that vision,' he states.

He also mentions the importance of compute in this future AI landscape: 'Compute is not to be underestimated. I think in the future, a really important question will be: How much compute does humanity want to spend to solve which problems?' This suggests that as RSI becomes feasible, the race for processing power may become a critical factor.

Onward and Upward

In terms of timelines, Socher remains optimistic. 'There will be products, and you’ll have to wait quarters, not years,' he says, indicating that Recursive Superintelligence is well on its way to delivering real-world applications.