Xue Guirong of Zhejiang Lab: When AI Starts Doing Scientific Research, I See the Ceiling of Large Language Models | GAIR 2025
Analysis
The article discusses the limitations of large language models (LLMs) in scientific research, highlighting the need for scientific foundation models that can understand and process diverse scientific data beyond the constraints of language. It focuses on the work of Zhejiang Lab and its 021 scientific foundation model, emphasizing its ability to overcome the limitations of LLMs in scientific discovery and problem-solving. The article also mentions the 'AI Manhattan Project' and the importance of AI in scientific advancements.
Key Takeaways
- •Large language models (LLMs) have limitations in scientific research due to their reliance on language.
- •Scientific foundation models are needed to understand and process diverse scientific data beyond language constraints.
- •Zhejiang Lab's 021 scientific foundation model aims to overcome these limitations.
- •The 'AI Manhattan Project' highlights the importance of AI in scientific advancements.
“The article quotes Xue Guirong, the technical director of the scientific model overall team at Zhejiang Lab, who points out that LLMs are limited by the 'boundaries of language' and cannot truly understand high-dimensional, multi-type scientific data, nor can they independently complete verifiable scientific discoveries. The article also highlights the 'AI Manhattan Project' as a major initiative in the application of AI in science.”