Mariel Pettee, a Chamberlain Postdoctoral Fellow in Berkeley Lab’s Physics Division, is part of Polymathic AI, an international team of scientists who are launching an ambitious new initiative that will leverage large language models (LLMs) – the same technology behind ChatGPT – to build a multidisciplinary AI-powered tool for scientific discovery.
The team plans to train its AI tool using numerical data and physics simulations from across multiple scientific fields, using LLMs specially modified for numerical data, to create connections between seemingly disparate disciplines. Their aim is to develop versatile, foundation-scale AI models specially tailored for scientific applications (and particularly for fundamental physics) by training these models on extensive volumes of heterogeneous scientific data. The tool will enable researchers to use the pre-trained AI as a starting point for their own projects and to model a variety of applications, from supergiant stars to the Earth’s climate, and beyond.
According to Pettee, this innately cross-disciplinary project will “connect many seemingly disparate subfields into something greater than the sum of their parts.” Ben Nachman – a staff scientist in Berkeley Lab’s Physics Division, where he leads the Machine Learning for Fundamental Physics group – adds: “Will this help us automate the discoveries of tomorrow? At the very least, I expect that this new kind of tool will facilitate a deeper and broader exploration of our data than we could manage before.”
The project was announced today in a Simons Foundation press release. The Polymathic AI development team includes experts in physics, astrophysics, mathematics, artificial intelligence, and neuroscience from Berkeley Lab and the following organizations: the Simons Foundation’s Flatiron Institute, New York University, the University of Cambridge, and Princeton University.
Read the full article:
Scientists Begin Building AI for Scientific Discovery Using Tech Behind ChatGPT
October 9, 2023 / Thomas Sumner / Simons Foundation Press Release