Abstract:The increased use of Large Language Models (LLMs) in geography raises substantial questions about the safety of integrating these tools across a wide range of processes and analyses, given our very limited understanding of their inner workings. In this extended abstract, we examine how LLMs process relative geographic space using activation patching, an emerging tool for mechanistic interpretability.
| Subjects: | Machine Learning (cs.LG) |
| Cite as: | arXiv:2605.14535 [cs.LG] |
| (or arXiv:2605.14535v1 [cs.LG] for this version) | |
| https://doi.org/10.48550/arXiv.2605.14535 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Stef De Sabbata [view email]
[v1]
Thu, 14 May 2026 08:18:31 UTC (1,016 KB)