Jerry Tworek, the former head of reinforcement learning at OpenAI, has already built a $4 billion company — in just six weeks. His new startup, Core Automation, is now looking to raise between $300 million and $500 million at that valuation, just weeks after closing a $100 million early round at a $1 billion valuation with backing from Nvidia, Spark Capital, and Accel.
Core Automation is taking a contrarian approach to building large AI models. Instead of the standard multi-phase training pipeline, the company is collapsing everything into a single phase using very little data. The goal: create an architecture capable of "continuous learning" — meaning the model keeps improving in real time from real-world experiences, even after deployment.
According to investors, Tworek has shared an ambitious roadmap. Near-term, the company plans to build an AI agent that can autonomously advance research and development. Longer term, Tworek envisions a "self-replicating factory" and even biological machines. The team includes heavy hitters like Rohan Anil, a former researcher at Anthropic and Google DeepMind, and Joanne Jang, who used to lead model behavior at OpenAI.
Core Automation belongs to a new wave of so-called "neolabs" — AI startups that lean heavily on Nvidia for both funding and compute. But with GPU rental costs soaring and deep-pocketed rivals like Meta poaching talent with massive salaries, delivering on those lofty promises is far from guaranteed.