Former OpenAI & DeepMind Researchers Raise $300M for Science Automation

Periodic Labs Emerges with $300 Million Seed Funding

Periodic Labs officially launched on Tuesday, securing $300 million in seed funding. This substantial investment comes from prominent figures and firms within the technology sector, including Andreessen Horowitz, DST, Nvidia, Accel, Elad Gil, Jeff Dean, Eric Schmidt, and Jeff Bezos.

Founding and Expertise

The company was established by Ekin Dogus Cubuk and Liam Fedus. Cubuk previously spearheaded the materials and chemistry division at Google Brain and DeepMind. His work included the development of GNoME, an AI tool responsible for identifying over 2 million novel crystals in 2023.

These newly discovered materials hold the potential to revolutionize future technologies, according to researchers. Fedus, on the other hand, formerly held the position of VP of Research at OpenAI.

Key Contributions to AI Development

Liam Fedus was instrumental in the creation of ChatGPT and led the team that pioneered the first trillion-parameter neural network. The Periodic Labs team is comprised of researchers with extensive experience in significant AI and materials science initiatives.

This includes contributions to projects like OpenAI’s agent Operator and Microsoft’s MatterGen, an LLM focused on materials science discovery.

Automating Scientific Discovery

Periodic Labs’ overarching ambition is to fully automate the process of scientific discovery. The company aims to create AI scientists capable of independent research. This involves constructing laboratories where robotic systems conduct experiments, gather data, and iteratively refine their approach.

The learning process will be continuous, with the AI improving its capabilities over time.

Focus on Superconductors and Data Collection

The initial focus of the lab will be on developing new superconductors. These materials are intended to outperform existing options, potentially requiring less energy for operation. However, the startup also intends to explore and identify a wider range of novel materials.

A crucial aspect of their strategy is the systematic collection of physical data generated by their AI scientists. This data will be produced as the systems manipulate and analyze various substances.

The Importance of Physical Data

The company emphasizes that current AI advancements have largely relied on models trained using internet data. They contend that LLMs have largely exhausted the readily available information online.

As stated in their introductory blog post, Periodic Labs is dedicated to building both AI scientists and the autonomous laboratories necessary for their operation.

Generating New Data for AI Evolution

The expectation is that these laboratories will not only invent cutting-edge materials but also generate valuable, original data. This fresh data will then be used to further train and evolve AI models.

Competition in the AI Scientist Space

While Periodic Labs boasts an exceptionally talented team, it is not alone in pursuing the development of AI scientists. Automated chemistry discovery has been a subject of academic research since at least 2023.

Several other organizations are also actively involved in this field, including startups like Tetsuwan Scientific, nonprofits such as Future House, and the University of Toronto’s Acceleration Consortium.

• The field of AI-driven materials science is rapidly expanding.
• Multiple entities are investing in the automation of scientific research.