Former OpenAI Researcher Jerry Tworek Launches Core Automation to Build World’s Most Automated AI Lab

Former OpenAI researcher Jerry Tworek has officially launched a new AI research institute, Core Automation. With the vision of building the world’s most highly automated artificial intelligence R&D environment, the lab plans to fully automate its own internal research workflows as a first step.

According to official statements, the team has moved away from the conventional approach of simply scaling up model size and training data. Instead, they are focused on overcoming the limitations of current pre-training and reinforcement learning paradigms. Their R&D efforts will prioritize pioneering next-generation learning algorithms and developing novel foundational architectures that offer superior scalability compared to the Transformer.

The lab has assembled top-tier experts specializing in advanced model development, algorithmic optimization, and systems engineering. Its core strategy revolves around equipping lean research teams with highly capable AI agents, enabling them to efficiently tackle complex projects that previously required the resources of entire organizations.

Tworek departed OpenAI this past January, concluding a seven-year tenure. He noted that the company’s current environment no longer supports this type of foundational research, adding his perspective that the field of deep learning has largely reached a plateau, with diminishing room for substantial breakthroughs.

With its launch, Core Automation joins a growing wave of "Neo Labs" founded by OpenAI alumni and industry pioneers. Aligning with institutions like Thinking Machines Lab (led by a former CTO) and Safe Superintelligence (founded by a former chief scientist), these ventures share a unified conviction: the next paradigm shift in AI will depend entirely on fundamentally novel methodological approaches.