Ex-Google and OpenAI researchers raise $650 million to build self-improving AI

Former Google and Meta Researchers Raise $650 Million for Self-Improving AI Company

Recursive Superintelligence, founded by eight researchers from Google, Meta, and OpenAI, has raised more than $650 million in venture funding and is now valued at over $4 billion. The six-month-old startup, with fewer than 30 employees across offices in San Francisco and London, is pursuing a specific technical goal: building AI systems that can improve themselves with minimal human intervention.

The company's backers include GV (formerly Google Ventures), Greycroft, Nvidia, and AMD.

AI Systems Now Write Code

Recent releases from Anthropic and OpenAI demonstrated AI systems particularly capable at writing computer code. Engineers across Silicon Valley now use these tools to build, test, and modify software applications faster than before.

This capability creates a logical next step. If AI can write code for word processors and social media apps, researchers reason, it can also write code that improves AI itself.

"AI is code. And now, AI can code," said Richard Socher, one of Recursive Superintelligence's founders. "The ingredients are there."

What Self-Improvement Means for Development Teams

The concept researchers call "recursive self-improvement" refers to AI systems that autonomously refine their own performance. This differs from current AI development, where human engineers manually adjust models, retrain systems, and deploy new versions.

Many of the world's leading researchers now believe this capability is within reach. The question for development teams is not whether this will happen, but when-and what it means for how software gets built.

Learn more about generative code and generative AI and LLM systems driving these advances.