Scientists working on Polymathic AI, a new tool that will help make scientific discoveries

According to Ho, ChatGPT has some well-known limitations when it comes to accuracy, but Polymathic AI plans to avoid those by treating numbers as actual numbers instead of characters.

Announcing:

*The Polymathic AI Initiative*

We are researching a new class of foundation models for scientific *data*, developing models that share scientific concepts across disciplines.

Site: https://t.co/C2Qsi59HzI
News: https://t.co/JomPjdIJAG
Thread: 👇 pic.twitter.com/pYvM18md4B

— PolymathicAI (@PolymathicAI) October 9, 2023

Members of the team recently released a scientific paper that shows how broadly pre-trained AI models can easily match or outperform AI models fine-tuned for a complex task, for example, simulating the physics of turbulent fluid flow. Compared to making a model from scratch, using large pre-trained models can offer several benefits even if the training data isn’t related to the problem.

While most AI models are trained for a specific use case, members working on the Polymathic AI project say the new model will be truly cross-disciplinary, as it will learn from data on various topics across like physics, chemistry, genomics and astrophysics and use it to “connect many seemingly disparate subfields into something greater than the sum of their parts”.

The team of scientists working on Polymathic include experts in various fields like mathematics, AI, neuroscience, astrophysics, and physics from the Simons Foundation, Flatiron Institute, New York University, Lawrence Berkeley National Laboratory, University of Cambridge and Princeton University.