Yuan-Dong Tian's AI Startup Valued at $31.5B; NVIDIA, AMD Invest; Yao Class's Shi Tianlin Joins as Co-founder
GV and Greycroft co-led the early round, with NVIDIA and AMD also participating. Tian left Meta to co-found Recursive Superintelligence (RSI), a startup valued at $4.65 billion, boasting a star-studded team of eight co-founders.
Article intelligence
Key points
- Yuan-Dong Tian joins RSI as a co-founder; company valuation reaches $4.65B.
- Eight co-founders all AI superstars, including Richard Socher, Shi Tianlin, and others.
- RSI focuses on recursive self-improving AI, first targeting a system with capabilities of 50,000 PhDs.
- Funding led by GV and Greycroft, with NVIDIA and AMD among investors.
Why it matters
This matters because yuan-Dong Tian joins RSI as a co-founder; company valuation reaches $4.65B.
Technical impact
May affect model selection, inference cost, product capability, and evaluation benchmarks.
GV and Greycroft co-led the early financing round for Recursive Superintelligence (RSI), a stealth startup that just emerged with a $650 million funding and a $4.65 billion valuation. NVIDIA and AMD also joined the round, which backs an ambitious plan to build recursive self-improving AI.
Yuan-Dong Tian, a former Research Scientist Director at Meta FAIR, is among the eight co-founders. Tian is known for his work on reinforcement learning, efficient foundation models, and neural network understanding. He previously led the ELF OpenGo project and recently focused on Llama inference and low-cost training.
The other seven co-founders are equally distinguished. Richard Socher, a former Stanford PhD student under Andrew Ng, co-authored ImageNet and GloVe, and founded MetaMind (acquired by Salesforce) and You.com. Shi Tianlin, a Tsinghua Yao Class alumnus, co-founded Cresta, which applied Transformers to customer service in 2019. Tim Rocktäschel is a UCL professor and open-endedness expert who pioneered Rainbow Teaming for AI safety. Alexey Dosovitskiy co-authored the Vision Transformer. Josh Tobin was an early OpenAI member and led the Agents Research Team. Caiming Xiong led AI Research at Salesforce and worked with Socher on CTRL. Jeff Clune researches open-ended evolution and AI self-improvement, co-authoring the Darwin Gödel Machine paper.
RSI’s mission is to create AI that continuously improves itself, forming a recursive loop toward superintelligence. The first step is a system with the research capabilities of 50,000 PhDs, automating AI science. This “Eureka machine” would then tackle drug discovery, battery materials, and fusion physics.
The team believes scaling laws are diminishing, and recursive self-improvement is the next major leap. As CEO Socher noted, “AI is code, and now AI can write code.” RSI is part of a broader wave of top scientists leaving Big Tech to pursue similar goals, with companies like David Silver’s Ineffable Intelligence ($1.1B seed) and Yann LeCun’s AMI Labs ($1B) also attracting huge investments.
Despite the $4.65 billion valuation, RSI employs fewer than 30 people. Socher emphasized, “We will keep the team as small and focused as possible, delegating many tasks to our agents.” The company’s name—Recursive Superintelligence—underscores its bold bet on self-improving AI as the foundation for the next generation of intelligence.