{"paper":{"title":"Exploring Geographic Relative Space in Large Language Models through Activation Patching","license":"http://creativecommons.org/licenses/by/4.0/","headline":"Large language models encode relative geographic space in patterns isolatable by activation patching.","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Kevin Roitero, Rahul Baiju, Stefano Mizzaro, Stef De Sabbata","submitted_at":"2026-05-14T08:18:31Z","abstract_excerpt":"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."},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"We examine how LLMs process relative geographic space using activation patching, an emerging tool for mechanistic interpretability.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That activation patching can meaningfully isolate and reveal the mechanisms by which LLMs represent relative geographic space.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"Activation patching is used to explore how LLMs encode relative geographic relationships.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Large language models encode relative geographic space in patterns isolatable by activation patching.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"869e3c1a41483347b0ad223bdebb04b19abcf40f921847884837cf86a0093ae1"},"source":{"id":"2605.14535","kind":"arxiv","version":1},"verdict":{"id":"2742cf1f-fb1b-4590-a5d0-ff3428ff0695","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-15T02:09:36.900151Z","strongest_claim":"We examine how LLMs process relative geographic space using activation patching, an emerging tool for mechanistic interpretability.","one_line_summary":"Activation patching is used to explore how LLMs encode relative geographic relationships.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That activation patching can meaningfully isolate and reveal the mechanisms by which LLMs represent relative geographic space.","pith_extraction_headline":"Large language models encode relative geographic space in patterns isolatable by activation patching."},"references":{"count":12,"sample":[{"doi":"","year":2025,"title":"Geospatial Mechanistic Interpretability of Large Language Models , author=. 2025 , eprint=","work_id":"75ad7b55-a31a-4d63-af4e-9822458de481","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2024,"title":"Computing Geographically: Bridging Giscience and Geography , author=. 2024 , publisher=","work_id":"d98f3346-dc22-46ea-8d1d-8c8cb2317f17","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"Locating and Editing Factual Associations in GPT , volume =","work_id":"7de2e0e0-bb8b-4a38-b8e1-2efb2d09a131","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"Investigating Gender Bias in Language Models Using Causal Mediation Analysis , volume =","work_id":"2a6422bd-f2b1-4ce9-9abe-977e8880cccf","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2024,"title":"Towards Best Practices of Activation Patching in Language Models: Metrics and Methods , author=. 2024 , eprint=","work_id":"00a06558-525f-4de9-a954-2044c2cfc305","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":12,"snapshot_sha256":"2b8dd41c7ff7fdfb6984cab57ebf19c1c8c03d46920a3a378b2ac9146211f88b","internal_anchors":0},"formal_canon":{"evidence_count":1,"snapshot_sha256":"9d4af04c0d74135ba77095604dd629e0ce448962ade7dfc7bc36d6ff33b52219"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}