Ex-Meta Researcher's Startup Raises $650M for Self-Improving AI at Sky-High Valuation

A new AI startup founded by a former Meta researcher has secured a massive $650 million funding round.

Recursive Superintelligence, just four months old, achieved a staggering $4.65 billion valuation in this seed investment.

Founders and Backers

Yuandong Tian, an ex-Meta research scientist known for his work in AI reinforcement learning, co-founded the company.

The round was led by prominent venture firms GV and Greycroft, with major participation from chip giants Nvidia and AMD.

Vision for Self-Improving AI

The startup aims to develop AI systems that safely improve themselves through automated experiments.

This approach targets "recursive superintelligence," where AI iteratively enhances its own capabilities without human intervention.

Such technology echoes long-standing debates in AI research about paths to artificial general intelligence, or AGI.

Unlike typical models trained on vast data, Recursive Superintelligence's method emphasizes real-world experimentation for safer scaling.

The involvement of Nvidia and AMD highlights how hardware leaders are betting on software breakthroughs to drive demand for advanced chips.

This funding underscores the ongoing talent exodus from big tech like Meta, fueled by lucrative startup offers amid the AI boom.

For everyday people, self-improving AI could revolutionize tools from healthcare diagnostics to personal assistants, but raises urgent safety questions.

Looking ahead, the startup's success may intensify competition, pushing incumbents like OpenAI to accelerate their own safety-focused innovations.

At this valuation, Recursive Superintelligence joins a rare club of AI unicorns, signaling investor frenzy for AGI pursuits despite regulatory scrutiny.