A team of scientists is developing a new tool called Polymathic AI, powered by technology similar to that of ChatGPT. Unlike OpenAI’s chatbot, which mainly works with words and sentences, Polymathic AI is designed to process numerical data and physics simulations.
Polymathic AI aims to assist scientists across diverse fields in modeling various phenomena, ranging from “supergiant stars to Earth’s climate.” Shirley Ho, the AI project lead, believes this model will revolutionize how AI and machine learning are utilized in scientific research.
The new model, in contrast to ChatGPT, plans to handle numbers as actual numerical values rather than characters, addressing the accuracy limitations commonly associated with language models.
The team behind Polymathic AI recently published a scientific paper demonstrating the model’s potential to match or surpass AI models specifically fine-tuned for complex tasks, such as simulating turbulent fluid flow in physics. Using large pre-trained models offers several advantages, even when the training data is unrelated to the problem being solved.
Unlike most AI models, which are typically tailored for specific use cases, Polymathic AI is intended to be highly interdisciplinary. It will learn from data spanning multiple domains, including physics, chemistry, genomics, and astrophysics, aiming to connect seemingly disparate subfields into a more comprehensive understanding.
The team working on Polymathic AI consists of experts in various fields, including mathematics, AI, neuroscience, astrophysics, and physics. They represent institutions like the Simons Foundation, Flatiron Institute, New York University, Lawrence Berkeley National Laboratory, University of Cambridge, and Princeton University.
