An international team of scientists, including experts from the University of Cambridge (UK), has launched a research project to create an AI-powered tool specifically for scientific research activities. This tool will be based on similar technology to ChatGPT.
While ChatGPT works with text, the new AI application will be trained on numerical data and physical simulations from a variety of scientific fields. This will help scientists model diverse research objects, from supergiant stars to Earth's climate.
The project is called “Polymathic AI” and was announced at the same time as the publication of a number of related papers on the open data repository arXiv (1, 2, 3).
“This is going to completely change the way people use AI and machine learning in science,” said Shirley Ho, research leader at the Flatiron Institute’s Center for Computational Astrophysics. One of the core ideas behind “Polymathic AI” is that using large, pre-trained models can be faster and more accurate than building a scientific model from scratch.
The team brought together scientists from various institutions, including experts in physics, astrophysics, mathematics, AI and neuroscience. The project aims to study data from various sources in physics and astrophysics, and in the future - chemistry and genomics. The goal of the project is to apply multidisciplinary knowledge to different scientific problems.
Despite the limitations of the ChatGPT application in terms of accuracy (such as in multiplication), the “Polymathic AI” project plans to focus on solving this problem.
Shirley Ho emphasized the transparency and openness of the project: “We want to make everything public, democratize AI for science. In a few years, we will provide the community with a trained AI model that can improve scientific analysis in many different fields.”
(according to Securitylab)
Source
Comment (0)