Ex-OpenAI Researcher Jerry Tworek Launches Core Automation to Revolutionize AI Labs

Jerry Tworek, a former OpenAI researcher with seven years of experience under his belt, has unveiled "Core Automation," an ambitious new AI lab aimed at becoming the most automated in the world. Rather than focusing on scaling up existing models and datasets, Core Automation is taking a different approach by developing novel learning algorithms and scalable architectures designed to outperform current state-of-the-art systems.

A Shift from Conventional Wisdom

Tworek's departure from OpenAI in January 2026 was marked by his frustration with the direction of deep learning research. He believes that "deep learning is done" and that fundamental progress now requires a fresh approach. This sentiment aligns him with other ex-OpenAI researchers who have launched similar ventures, collectively known as Neo Labs.

Core Automation's Approach

New Learning Algorithms

Core Automation is exploring algorithms that go beyond pre-training and reinforcement learning. The lab aims to develop methods that can adapt more efficiently to new tasks and data, potentially reducing the need for massive datasets and computational resources.

• Beyond Pre-Training: Traditional models rely heavily on large-scale pre-training on vast amounts of data. Core Automation is working on algorithms that can learn from smaller, more targeted datasets.
• Reinforcement Learning Alternatives: While reinforcement learning has shown promise in certain domains, it often requires extensive tuning and can be unstable. The lab is investigating alternative approaches that are more robust and easier to scale.

Scalable Architectures

The team at Core Automation is also focusing on architectures that can scale better than transformers. Transformers have been the go-to model for many NLP tasks, but they come with significant computational costs. Core Automation's goal is to design models that maintain or improve performance while reducing these costs.

• Efficient Scaling: Traditional transformer models often struggle with scaling due to their quadratic complexity in attention mechanisms. The lab is exploring new architectures that can scale linearly or better.
• Resource Optimization: By optimizing resource usage, the team aims to make it feasible to train and deploy large-scale models on more modest hardware.

Team and Vision

Core Automation brings together experts from various fields, including frontier models, optimization, and systems engineering. The vision is to create small, agile teams that can leverage advanced AI agents to perform tasks that traditionally require large organizations.

• Frontier Models: Specialists in cutting-edge model development ensure that the lab stays at the forefront of AI research.
• Optimization Experts: These experts focus on making sure that models and algorithms are as efficient as possible, both in terms of computational resources and performance.
• Systems Engineers: They handle the infrastructure needed to support the lab's ambitious goals, ensuring that everything runs smoothly and efficiently.

Context within Neo Labs

Core Automation is part of a growing trend of Neo Labs founded by former OpenAI researchers. Other notable examples include:

• Thinking Machines Lab: Led by the former CTO of OpenAI, this lab focuses on developing AI systems for complex problem-solving.
• Safe Superintelligence: Headed by the ex-chief scientist of OpenAI, this lab aims to ensure that advanced AI is developed safely and ethically.

These labs share a common belief that real progress in AI requires fundamentally new approaches. They are all pushing the boundaries of what is possible in AI research and development.

Conclusion

Jerry Tworek's Core Automation represents a significant shift in the AI landscape. By focusing on novel algorithms and scalable architectures, the lab aims to revolutionize how AI research is conducted. As more ex-OpenAI researchers join this movement, the future of AI could look very different from what we see today.