{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:E6JX7BRNI3WWORZLOQLWTKDOFJ","short_pith_number":"pith:E6JX7BRN","schema_version":"1.0","canonical_sha256":"27937f862d46ed67472b741769a86e2a4fe410fd83e84bf9b2a0c3134360aaa5","source":{"kind":"arxiv","id":"2410.00857","version":1},"attestation_state":"computed","paper":{"title":"Quantifying reliance on external information over parametric knowledge during Retrieval Augmented Generation (RAG) using mechanistic analysis","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Ehsan Aghazadeh, Hitesh Wadhwa, Rahul Seetharaman, Reshmi Ghosh, Samyadeep Basu, Shreyas Chaudhari, Somyaa Aggarwal, Soundararajan Srinivasan, Wenlong Zhao","submitted_at":"2024-10-01T16:48:13Z","abstract_excerpt":"Retrieval Augmented Generation (RAG) is a widely used approach for leveraging external context in several natural language applications such as question answering and information retrieval. Yet, the exact nature in which a Language Model (LM) leverages this non-parametric memory or retrieved context isn't clearly understood. This paper mechanistically examines the RAG pipeline to highlight that LMs demonstrate a \"shortcut'' effect and have a strong bias towards utilizing the retrieved context to answer questions, while relying minimally on model priors. We propose (a) Causal Mediation Analysis"},"verification_status":{"content_addressed":true,"pith_receipt":true,"author_attested":false,"weak_author_claims":0,"strong_author_claims":0,"externally_anchored":false,"storage_verified":false,"citation_signatures":0,"replication_records":0,"graph_snapshot":true,"references_resolved":false,"formal_links_present":false},"canonical_record":{"source":{"id":"2410.00857","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2024-10-01T16:48:13Z","cross_cats_sorted":[],"title_canon_sha256":"39090e50a98c0e23ba038c83c964005871f604da83b3f1c7cf24c40577b1c9f3","abstract_canon_sha256":"6e924190c7f41631714fcec162217a8e3f8c0eaf94f8ce464b11d68535a5c3cb"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T09:14:23.761020Z","signature_b64":"75vhM0DblCpkRqA/uD9rx4g6sG8jGAC1jminDCPlJSttYIBCc7yj79Xg8FEHAJX2WdrB1oczBGXvonncBIx0BA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"27937f862d46ed67472b741769a86e2a4fe410fd83e84bf9b2a0c3134360aaa5","last_reissued_at":"2026-07-05T09:14:23.760543Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T09:14:23.760543Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Quantifying reliance on external information over parametric knowledge during Retrieval Augmented Generation (RAG) using mechanistic analysis","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Ehsan Aghazadeh, Hitesh Wadhwa, Rahul Seetharaman, Reshmi Ghosh, Samyadeep Basu, Shreyas Chaudhari, Somyaa Aggarwal, Soundararajan Srinivasan, Wenlong Zhao","submitted_at":"2024-10-01T16:48:13Z","abstract_excerpt":"Retrieval Augmented Generation (RAG) is a widely used approach for leveraging external context in several natural language applications such as question answering and information retrieval. Yet, the exact nature in which a Language Model (LM) leverages this non-parametric memory or retrieved context isn't clearly understood. This paper mechanistically examines the RAG pipeline to highlight that LMs demonstrate a \"shortcut'' effect and have a strong bias towards utilizing the retrieved context to answer questions, while relying minimally on model priors. We propose (a) Causal Mediation Analysis"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2410.00857","kind":"arxiv","version":1},"verdict":{"id":null,"model_set":{},"created_at":null,"strongest_claim":"","one_line_summary":"","pipeline_version":null,"weakest_assumption":"","pith_extraction_headline":""},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2410.00857/integrity.json","findings":[],"available":true,"detectors_run":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938"},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","internal_anchors":0},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"},"aliases":[{"alias_kind":"arxiv","alias_value":"2410.00857","created_at":"2026-07-05T09:14:23.760601+00:00"},{"alias_kind":"arxiv_version","alias_value":"2410.00857v1","created_at":"2026-07-05T09:14:23.760601+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2410.00857","created_at":"2026-07-05T09:14:23.760601+00:00"},{"alias_kind":"pith_short_12","alias_value":"E6JX7BRNI3WW","created_at":"2026-07-05T09:14:23.760601+00:00"},{"alias_kind":"pith_short_16","alias_value":"E6JX7BRNI3WWORZL","created_at":"2026-07-05T09:14:23.760601+00:00"},{"alias_kind":"pith_short_8","alias_value":"E6JX7BRN","created_at":"2026-07-05T09:14:23.760601+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2507.05424","citing_title":"\"Lost-in-the-Later\": Framework for Quantifying Contextual Grounding in Large Language Models","ref_index":9,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/E6JX7BRNI3WWORZLOQLWTKDOFJ","json":"https://pith.science/pith/E6JX7BRNI3WWORZLOQLWTKDOFJ.json","graph_json":"https://pith.science/api/pith-number/E6JX7BRNI3WWORZLOQLWTKDOFJ/graph.json","events_json":"https://pith.science/api/pith-number/E6JX7BRNI3WWORZLOQLWTKDOFJ/events.json","paper":"https://pith.science/paper/E6JX7BRN"},"agent_actions":{"view_html":"https://pith.science/pith/E6JX7BRNI3WWORZLOQLWTKDOFJ","download_json":"https://pith.science/pith/E6JX7BRNI3WWORZLOQLWTKDOFJ.json","view_paper":"https://pith.science/paper/E6JX7BRN","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2410.00857&json=true","fetch_graph":"https://pith.science/api/pith-number/E6JX7BRNI3WWORZLOQLWTKDOFJ/graph.json","fetch_events":"https://pith.science/api/pith-number/E6JX7BRNI3WWORZLOQLWTKDOFJ/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/E6JX7BRNI3WWORZLOQLWTKDOFJ/action/timestamp_anchor","attest_storage":"https://pith.science/pith/E6JX7BRNI3WWORZLOQLWTKDOFJ/action/storage_attestation","attest_author":"https://pith.science/pith/E6JX7BRNI3WWORZLOQLWTKDOFJ/action/author_attestation","sign_citation":"https://pith.science/pith/E6JX7BRNI3WWORZLOQLWTKDOFJ/action/citation_signature","submit_replication":"https://pith.science/pith/E6JX7BRNI3WWORZLOQLWTKDOFJ/action/replication_record"}},"created_at":"2026-07-05T09:14:23.760601+00:00","updated_at":"2026-07-05T09:14:23.760601+00:00"}