{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2023:VVN2X3FUZVD3H4FPWHFQPWY75J","short_pith_number":"pith:VVN2X3FU","schema_version":"1.0","canonical_sha256":"ad5babecb4cd47b3f0afb1cb07db1fea76790c9b866693e36b6b084306a0185b","source":{"kind":"arxiv","id":"2304.14767","version":3},"attestation_state":"computed","paper":{"title":"Dissecting Recall of Factual Associations in Auto-Regressive Language Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Amir Globerson, Jasmijn Bastings, Katja Filippova, Mor Geva","submitted_at":"2023-04-28T11:26:17Z","abstract_excerpt":"Transformer-based language models (LMs) are known to capture factual knowledge in their parameters. While previous work looked into where factual associations are stored, only little is known about how they are retrieved internally during inference. We investigate this question through the lens of information flow. Given a subject-relation query, we study how the model aggregates information about the subject and relation to predict the correct attribute. With interventions on attention edges, we first identify two critical points where information propagates to the prediction: one from the re"},"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":"2304.14767","kind":"arxiv","version":3},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2023-04-28T11:26:17Z","cross_cats_sorted":[],"title_canon_sha256":"a57c9e583f208a35686d5ec2bf3c968c0f5dd405e1f6393063705978ca7c9d3f","abstract_canon_sha256":"83f495dbd198bd44d88ebd9ce447477c0f31816bd3e465886642a5002cb4135b"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T07:00:44.363041Z","signature_b64":"AwHS5EZTiHHHWB4L6GkhLYeONSsFYXnk6BxagLgc6CtmD2kQAxhsTVFkj3TFdBze5XVBUCDKaeME5OC8lz+XDQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"ad5babecb4cd47b3f0afb1cb07db1fea76790c9b866693e36b6b084306a0185b","last_reissued_at":"2026-07-05T07:00:44.362529Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T07:00:44.362529Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Dissecting Recall of Factual Associations in Auto-Regressive Language Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Amir Globerson, Jasmijn Bastings, Katja Filippova, Mor Geva","submitted_at":"2023-04-28T11:26:17Z","abstract_excerpt":"Transformer-based language models (LMs) are known to capture factual knowledge in their parameters. While previous work looked into where factual associations are stored, only little is known about how they are retrieved internally during inference. We investigate this question through the lens of information flow. Given a subject-relation query, we study how the model aggregates information about the subject and relation to predict the correct attribute. With interventions on attention edges, we first identify two critical points where information propagates to the prediction: one from the re"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2304.14767","kind":"arxiv","version":3},"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/2304.14767/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":"2304.14767","created_at":"2026-07-05T07:00:44.362591+00:00"},{"alias_kind":"arxiv_version","alias_value":"2304.14767v3","created_at":"2026-07-05T07:00:44.362591+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2304.14767","created_at":"2026-07-05T07:00:44.362591+00:00"},{"alias_kind":"pith_short_12","alias_value":"VVN2X3FUZVD3","created_at":"2026-07-05T07:00:44.362591+00:00"},{"alias_kind":"pith_short_16","alias_value":"VVN2X3FUZVD3H4FP","created_at":"2026-07-05T07:00:44.362591+00:00"},{"alias_kind":"pith_short_8","alias_value":"VVN2X3FU","created_at":"2026-07-05T07:00:44.362591+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":21,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.24952","citing_title":"Perfect Detection, Failed Control: The Geometry of Knowing vs. Steering in Language Models","ref_index":4,"is_internal_anchor":false},{"citing_arxiv_id":"2606.21345","citing_title":"Factual Retrieval in LLMs Is a Redundant, Distributed and Non-Contiguous Process","ref_index":8,"is_internal_anchor":false},{"citing_arxiv_id":"2606.20369","citing_title":"CATCH-ME if you RAG: a dataset of Contextually Annotated multi-Turn Counterspeech against Hate and Misinformation Exchanges","ref_index":240,"is_internal_anchor":false},{"citing_arxiv_id":"2606.12841","citing_title":"TimeROME-DLM: Temporal Causal Tracing and Low-Rank Inference-Time Knowledge Editing for Masked Diffusion Language Models","ref_index":27,"is_internal_anchor":false},{"citing_arxiv_id":"2607.00415","citing_title":"A Mechanistic View of Authority Hierarchy in LLM Sycophancy","ref_index":22,"is_internal_anchor":false},{"citing_arxiv_id":"2606.28548","citing_title":"Turn-Averaged SAEs for Feature Discovery and Long-Context Attribution","ref_index":14,"is_internal_anchor":false},{"citing_arxiv_id":"2606.09563","citing_title":"PRISM: Recovering Instruction Sets from Language Model Activations","ref_index":17,"is_internal_anchor":false},{"citing_arxiv_id":"2605.22435","citing_title":"Assisted Counterspeech Writing at the Crossroads of Hate Speech and Misinformation","ref_index":230,"is_internal_anchor":false},{"citing_arxiv_id":"2603.17839","citing_title":"How do LLMs Compute Verbal Confidence","ref_index":6,"is_internal_anchor":false},{"citing_arxiv_id":"2605.12991","citing_title":"Not Just RLHF: Why Alignment Alone Won't Fix Multi-Agent Sycophancy","ref_index":15,"is_internal_anchor":false},{"citing_arxiv_id":"2509.14837","citing_title":"V-SEAM: Visual Semantic Editing and Attention Modulating for Causal Interpretability of Vision-Language Models","ref_index":14,"is_internal_anchor":false},{"citing_arxiv_id":"2510.02370","citing_title":"How Training Data Shapes the Use of Parametric and In-Context Knowledge in Language Models","ref_index":4,"is_internal_anchor":false},{"citing_arxiv_id":"2309.16042","citing_title":"Towards Best Practices of Activation Patching in Language Models: Metrics and Methods","ref_index":83,"is_internal_anchor":false},{"citing_arxiv_id":"2404.15255","citing_title":"How to use and interpret activation patching","ref_index":9,"is_internal_anchor":false},{"citing_arxiv_id":"2605.12809","citing_title":"Correcting Influence: Unboxing LLM Outputs with Orthogonal Latent Spaces","ref_index":102,"is_internal_anchor":false},{"citing_arxiv_id":"2605.12991","citing_title":"Not Just RLHF: Why Alignment Alone Won't Fix Multi-Agent Sycophancy","ref_index":15,"is_internal_anchor":false},{"citing_arxiv_id":"2604.23750","citing_title":"The Override Gap: A Magnitude Account of Knowledge Conflict Failure in Hypernetwork-Based Instant LLM Adaptation","ref_index":11,"is_internal_anchor":false},{"citing_arxiv_id":"2604.23750","citing_title":"The Override Gap: A Magnitude Account of Knowledge Conflict Failure in Hypernetwork-Based Instant LLM Adaptation","ref_index":11,"is_internal_anchor":false},{"citing_arxiv_id":"2605.00817","citing_title":"When LLMs Stop Following Steps: A Diagnostic Study of Procedural Execution in Language Models","ref_index":32,"is_internal_anchor":false},{"citing_arxiv_id":"2605.07990","citing_title":"Tool Calling is Linearly Readable and Steerable in Language Models","ref_index":50,"is_internal_anchor":false},{"citing_arxiv_id":"2604.19884","citing_title":"From Signal Degradation to Computation Collapse: Uncovering the Two Failure Modes of LLM Quantization","ref_index":22,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/VVN2X3FUZVD3H4FPWHFQPWY75J","json":"https://pith.science/pith/VVN2X3FUZVD3H4FPWHFQPWY75J.json","graph_json":"https://pith.science/api/pith-number/VVN2X3FUZVD3H4FPWHFQPWY75J/graph.json","events_json":"https://pith.science/api/pith-number/VVN2X3FUZVD3H4FPWHFQPWY75J/events.json","paper":"https://pith.science/paper/VVN2X3FU"},"agent_actions":{"view_html":"https://pith.science/pith/VVN2X3FUZVD3H4FPWHFQPWY75J","download_json":"https://pith.science/pith/VVN2X3FUZVD3H4FPWHFQPWY75J.json","view_paper":"https://pith.science/paper/VVN2X3FU","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2304.14767&json=true","fetch_graph":"https://pith.science/api/pith-number/VVN2X3FUZVD3H4FPWHFQPWY75J/graph.json","fetch_events":"https://pith.science/api/pith-number/VVN2X3FUZVD3H4FPWHFQPWY75J/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/VVN2X3FUZVD3H4FPWHFQPWY75J/action/timestamp_anchor","attest_storage":"https://pith.science/pith/VVN2X3FUZVD3H4FPWHFQPWY75J/action/storage_attestation","attest_author":"https://pith.science/pith/VVN2X3FUZVD3H4FPWHFQPWY75J/action/author_attestation","sign_citation":"https://pith.science/pith/VVN2X3FUZVD3H4FPWHFQPWY75J/action/citation_signature","submit_replication":"https://pith.science/pith/VVN2X3FUZVD3H4FPWHFQPWY75J/action/replication_record"}},"created_at":"2026-07-05T07:00:44.362591+00:00","updated_at":"2026-07-05T07:00:44.362591+00:00"}