{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:76FT3KBHAGT4G7ZYIOHPFPQGS5","short_pith_number":"pith:76FT3KBH","schema_version":"1.0","canonical_sha256":"ff8b3da82701a7c37f38438ef2be069747179c680add4542972cc24bb609b2e1","source":{"kind":"arxiv","id":"2503.15299","version":4},"attestation_state":"computed","paper":{"title":"Inside-Out: Hidden Factual Knowledge in LLMs","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Eran Ofek, Eyal Ben David, Hadas Orgad, Idan Szpektor, Jonathan Herzig, Roi Reichart, Yonatan Belinkov, Zorik Gekhman","submitted_at":"2025-03-19T15:21:48Z","abstract_excerpt":"This work presents a framework for assessing whether large language models (LLMs) encode more factual knowledge in their parameters than what they express in their outputs. While a few studies hint at this possibility, none has clearly defined or demonstrated this phenomenon. We first propose a formal definition of knowledge, quantifying it for a given question as the fraction of correct-incorrect answer pairs where the correct one is ranked higher. This gives rise to external and internal knowledge, depending on the information used to score individual answer candidates: either the model's ob"},"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":"2503.15299","kind":"arxiv","version":4},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2025-03-19T15:21:48Z","cross_cats_sorted":[],"title_canon_sha256":"2b8e0a7a4d8a3ccd3465eb84672441c1c1b234669931ca983c0a2b03464d5892","abstract_canon_sha256":"2bc4077db58b38eb6d4750cd73e72afa81249d1023ab6aa560e00277cf478a8f"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:49:08.461626Z","signature_b64":"bwEBBjzW8bqb6nfbPxc9Yed1KXSgk14cXbxOqE1GilPVRGSVgtDkeCNf4ZHL6+ZGUr26cdT8DSObqR6FsuSHDw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"ff8b3da82701a7c37f38438ef2be069747179c680add4542972cc24bb609b2e1","last_reissued_at":"2026-07-05T11:49:08.461061Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:49:08.461061Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Inside-Out: Hidden Factual Knowledge in LLMs","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Eran Ofek, Eyal Ben David, Hadas Orgad, Idan Szpektor, Jonathan Herzig, Roi Reichart, Yonatan Belinkov, Zorik Gekhman","submitted_at":"2025-03-19T15:21:48Z","abstract_excerpt":"This work presents a framework for assessing whether large language models (LLMs) encode more factual knowledge in their parameters than what they express in their outputs. While a few studies hint at this possibility, none has clearly defined or demonstrated this phenomenon. We first propose a formal definition of knowledge, quantifying it for a given question as the fraction of correct-incorrect answer pairs where the correct one is ranked higher. This gives rise to external and internal knowledge, depending on the information used to score individual answer candidates: either the model's ob"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2503.15299","kind":"arxiv","version":4},"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/2503.15299/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":"2503.15299","created_at":"2026-07-05T11:49:08.461123+00:00"},{"alias_kind":"arxiv_version","alias_value":"2503.15299v4","created_at":"2026-07-05T11:49:08.461123+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2503.15299","created_at":"2026-07-05T11:49:08.461123+00:00"},{"alias_kind":"pith_short_12","alias_value":"76FT3KBHAGT4","created_at":"2026-07-05T11:49:08.461123+00:00"},{"alias_kind":"pith_short_16","alias_value":"76FT3KBHAGT4G7ZY","created_at":"2026-07-05T11:49:08.461123+00:00"},{"alias_kind":"pith_short_8","alias_value":"76FT3KBH","created_at":"2026-07-05T11:49:08.461123+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":5,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.24579","citing_title":"Cross-Lingual Exploration for Parametric Knowledge","ref_index":40,"is_internal_anchor":false},{"citing_arxiv_id":"2605.19270","citing_title":"DECOR: Auditing LLM Deception via Information Manipulation Theory","ref_index":33,"is_internal_anchor":false},{"citing_arxiv_id":"2508.13654","citing_title":"Input-Time Scaling: Adding Noise and Irrelevance into Less-Is-More Drastically Improves Reasoning Performance and Efficiency","ref_index":2,"is_internal_anchor":false},{"citing_arxiv_id":"2509.03540","citing_title":"Improving Factuality in LLMs via Inference-Time Knowledge Graph Construction","ref_index":11,"is_internal_anchor":false},{"citing_arxiv_id":"2605.01428","citing_title":"Hallucinations Undermine Trust; Metacognition is a Way Forward","ref_index":8,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/76FT3KBHAGT4G7ZYIOHPFPQGS5","json":"https://pith.science/pith/76FT3KBHAGT4G7ZYIOHPFPQGS5.json","graph_json":"https://pith.science/api/pith-number/76FT3KBHAGT4G7ZYIOHPFPQGS5/graph.json","events_json":"https://pith.science/api/pith-number/76FT3KBHAGT4G7ZYIOHPFPQGS5/events.json","paper":"https://pith.science/paper/76FT3KBH"},"agent_actions":{"view_html":"https://pith.science/pith/76FT3KBHAGT4G7ZYIOHPFPQGS5","download_json":"https://pith.science/pith/76FT3KBHAGT4G7ZYIOHPFPQGS5.json","view_paper":"https://pith.science/paper/76FT3KBH","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2503.15299&json=true","fetch_graph":"https://pith.science/api/pith-number/76FT3KBHAGT4G7ZYIOHPFPQGS5/graph.json","fetch_events":"https://pith.science/api/pith-number/76FT3KBHAGT4G7ZYIOHPFPQGS5/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/76FT3KBHAGT4G7ZYIOHPFPQGS5/action/timestamp_anchor","attest_storage":"https://pith.science/pith/76FT3KBHAGT4G7ZYIOHPFPQGS5/action/storage_attestation","attest_author":"https://pith.science/pith/76FT3KBHAGT4G7ZYIOHPFPQGS5/action/author_attestation","sign_citation":"https://pith.science/pith/76FT3KBHAGT4G7ZYIOHPFPQGS5/action/citation_signature","submit_replication":"https://pith.science/pith/76FT3KBHAGT4G7ZYIOHPFPQGS5/action/replication_record"}},"created_at":"2026-07-05T11:49:08.461123+00:00","updated_at":"2026-07-05T11:49:08.461123+00:00"}