LEVERAGING LARGE LANGUAGE MODELS FOR SEDIMENTARY ENVIRONMENTAL INTERPRETATION: A NEW APPROACH TO ADDRESSING THE NON-UNIQUENESS PROBLEM IN PALEOGEOGRAPHY

The interpretation of the sedimentary rock record is fundamental for reconstructing paleogeography. However, the inherent non-uniqueness in such interpretations often poses significant challenges, which are further amplified by the exponential increase in scholarly publications. Traditional expert systems, while providing a potential solution, demand substantial contributions from specialists across various sub-disciplines, making them labor-intensive and less adaptive to the incorporation of newly published studies.

In light of these challenges, we propose a novel approach that leverages the advancements in Large Language Models (LLMs), such as GPT-4, and LangChain to process and interpret the vast corpus of publications. By integrating these technologies and fine-tuning them with domain-specific text, we have developed a system capable of providing multiple, probabilistically-weighted paleoenvironmental interpretations, each substantiated with specific references.

Our preliminary results suggest that this integrative approach holds significant potential in aiding the interpretation of sedimentary environments and addressing the non-uniqueness problem by harnessing the wealth of knowledge embedded in published literature. This innovative methodology could pave the way for more accurate, comprehensive, and holistic interpretations of the rock record, thereby enhancing our ability to reconstruct paleogeography. We extend an invitation to the broader paleoenvironmental research community to collaborate in refining and applying this promising approach.