{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2022:GKEBQHZOUDEA77XMILT44NLAZP","short_pith_number":"pith:GKEBQHZO","schema_version":"1.0","canonical_sha256":"3288181f2ea0c80ffeec42e7ce3560cbc408fdfa9fc77f12611401d53aadf9a5","source":{"kind":"arxiv","id":"2202.05262","version":5},"attestation_state":"computed","paper":{"title":"Locating and Editing Factual Associations in GPT","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"Factual associations in GPT models are stored in localized mid-layer feed-forward computations that can be directly edited via rank-one weight updates.","cross_cats":["cs.LG"],"primary_cat":"cs.CL","authors_text":"Alex Andonian, David Bau, Kevin Meng, Yonatan Belinkov","submitted_at":"2022-02-10T18:59:54Z","abstract_excerpt":"We analyze the storage and recall of factual associations in autoregressive transformer language models, finding evidence that these associations correspond to localized, directly-editable computations. We first develop a causal intervention for identifying neuron activations that are decisive in a model's factual predictions. This reveals a distinct set of steps in middle-layer feed-forward modules that mediate factual predictions while processing subject tokens. To test our hypothesis that these computations correspond to factual association recall, we modify feed-forward weights to update s"},"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":true,"formal_links_present":true},"canonical_record":{"source":{"id":"2202.05262","kind":"arxiv","version":5},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2022-02-10T18:59:54Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"5ce9b4bbd2053dd244ea67114c94a80075b0edf61ce03c55d302752266c983ec","abstract_canon_sha256":"4b53bc8ae790872206209c4c8d74e42d44cc03ad4c0074f1bcba9a2de9b43519"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-17T23:38:48.267670Z","signature_b64":"xKojHUdhm6ZrOnU62GNrO2+2iuFFXyXH7UPQjb1jcG5e9PIRz6jc6c3/ZIfSfQhAzWnf86CXL2bvLAN6GCn8Dg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"3288181f2ea0c80ffeec42e7ce3560cbc408fdfa9fc77f12611401d53aadf9a5","last_reissued_at":"2026-05-17T23:38:48.267193Z","signature_status":"signed_v1","first_computed_at":"2026-05-17T23:38:48.267193Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Locating and Editing Factual Associations in GPT","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"Factual associations in GPT models are stored in localized mid-layer feed-forward computations that can be directly edited via rank-one weight updates.","cross_cats":["cs.LG"],"primary_cat":"cs.CL","authors_text":"Alex Andonian, David Bau, Kevin Meng, Yonatan Belinkov","submitted_at":"2022-02-10T18:59:54Z","abstract_excerpt":"We analyze the storage and recall of factual associations in autoregressive transformer language models, finding evidence that these associations correspond to localized, directly-editable computations. We first develop a causal intervention for identifying neuron activations that are decisive in a model's factual predictions. This reveals a distinct set of steps in middle-layer feed-forward modules that mediate factual predictions while processing subject tokens. To test our hypothesis that these computations correspond to factual association recall, we modify feed-forward weights to update s"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"We find that ROME is effective on a standard zero-shot relation extraction (zsRE) model-editing task, comparable to existing methods. To perform a more sensitive evaluation, we also evaluate ROME on a new dataset of counterfactual assertions, on which it simultaneously maintains both specificity and generalization, whereas other methods sacrifice one or another.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"The assumption that the causal intervention correctly isolates the decisive feed-forward computations for factual recall, and that a rank-one update to those weights changes the association without creating unmeasured side effects on the broader distribution of model behavior.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"Factual associations in autoregressive transformers are localized to mid-layer feed-forward modules and can be edited via rank-one model editing while preserving both specificity and generalization on counterfactual tests.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Factual associations in GPT models are stored in localized mid-layer feed-forward computations that can be directly edited via rank-one weight updates.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"b2abb54c7d4c21749bfc35a611cee2fafeabc9c1546669f3b09b475630eed4ea"},"source":{"id":"2202.05262","kind":"arxiv","version":5},"verdict":{"id":"596fd162-bcc3-417a-9bd0-541a17654282","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-16T09:55:25.774121Z","strongest_claim":"We find that ROME is effective on a standard zero-shot relation extraction (zsRE) model-editing task, comparable to existing methods. To perform a more sensitive evaluation, we also evaluate ROME on a new dataset of counterfactual assertions, on which it simultaneously maintains both specificity and generalization, whereas other methods sacrifice one or another.","one_line_summary":"Factual associations in autoregressive transformers are localized to mid-layer feed-forward modules and can be edited via rank-one model editing while preserving both specificity and generalization on counterfactual tests.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"The assumption that the causal intervention correctly isolates the decisive feed-forward computations for factual recall, and that a rank-one update to those weights changes the association without creating unmeasured side effects on the broader distribution of model behavior.","pith_extraction_headline":"Factual associations in GPT models are stored in localized mid-layer feed-forward computations that can be directly edited via rank-one weight updates."},"references":{"count":43,"sample":[{"doi":"","year":2017,"title":"Fine-grained analysis of sentence embeddings using auxiliary prediction tasks","work_id":"b7bd2b01-1ceb-409d-9ec3-e841491e27b1","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":1972,"title":"Anderson, J. A. A simple neural network generating an interactive memory. Mathematical biosciences, 14 0 (3-4): 0 197--220, 1972","work_id":"e6955ef9-c3c3-47d3-8a55-a361da59cac2","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2020,"title":"Rewriting a deep generative model","work_id":"30b22db5-f3a2-4cd6-83dc-bb134d36c8b7","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"10.1162/coli_a_00422","year":2021,"title":"Probing Classifiers: Promises, Shortcomings, and Advances","work_id":"3eabab74-ac71-4292-86ce-b0469cd4e6cf","ref_index":4,"cited_arxiv_id":"2102.12452","is_internal_anchor":true},{"doi":"10.1162/tacl_a_00254","year":2019,"title":"Analysis methods in neural language processing: A survey","work_id":"62f1a2a2-d87f-4f14-90cf-b4b293a4780c","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":43,"snapshot_sha256":"73260a8da562501fd2f1f3f7a4d7f9b70137a4a1f7ce0d835bd03497c761cf52","internal_anchors":3},"formal_canon":{"evidence_count":2,"snapshot_sha256":"32953a508cd941293af0a1776cfdcfae878e118b70e9d4b2121fff6aef4acbd5"},"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":"2202.05262","created_at":"2026-05-17T23:38:48.267271+00:00"},{"alias_kind":"arxiv_version","alias_value":"2202.05262v5","created_at":"2026-05-17T23:38:48.267271+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2202.05262","created_at":"2026-05-17T23:38:48.267271+00:00"},{"alias_kind":"pith_short_12","alias_value":"GKEBQHZOUDEA","created_at":"2026-05-18T12:33:33.725879+00:00"},{"alias_kind":"pith_short_16","alias_value":"GKEBQHZOUDEA77XM","created_at":"2026-05-18T12:33:33.725879+00:00"},{"alias_kind":"pith_short_8","alias_value":"GKEBQHZO","created_at":"2026-05-18T12:33:33.725879+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":28,"internal_anchor_count":28,"sample":[{"citing_arxiv_id":"2204.06745","citing_title":"GPT-NeoX-20B: An Open-Source Autoregressive Language Model","ref_index":62,"is_internal_anchor":true},{"citing_arxiv_id":"2605.11739","citing_title":"Learning to Foresee: Unveiling the Unlocking Efficiency of On-Policy Distillation","ref_index":16,"is_internal_anchor":true},{"citing_arxiv_id":"2605.22462","citing_title":"From Correlation to Cause: A Five-Stage Methodology for Feature Analysis in Transformer Language Models","ref_index":12,"is_internal_anchor":true},{"citing_arxiv_id":"2509.22739","citing_title":"Painless Activation Steering: An Automated, Lightweight Approach for Post-Training Large Language Models","ref_index":15,"is_internal_anchor":true},{"citing_arxiv_id":"2602.11130","citing_title":"Meltdown: Circuits and Bifurcations in Point-Cloud-Conditioned 3D Diffusion Transformers","ref_index":3,"is_internal_anchor":true},{"citing_arxiv_id":"2605.12991","citing_title":"Not Just RLHF: Why Alignment Alone Won't Fix Multi-Agent Sycophancy","ref_index":25,"is_internal_anchor":true},{"citing_arxiv_id":"2605.19092","citing_title":"Counterfactual Likelihood Tests for Indirect Influence in Private Reasoning Channels","ref_index":5,"is_internal_anchor":true},{"citing_arxiv_id":"2511.03182","citing_title":"Understanding Robustness of Model Editing in Code LLMs","ref_index":30,"is_internal_anchor":true},{"citing_arxiv_id":"2404.15255","citing_title":"How to use and interpret activation patching","ref_index":20,"is_internal_anchor":true},{"citing_arxiv_id":"2602.02543","citing_title":"Norm Anchors Make Model Edits Last","ref_index":14,"is_internal_anchor":true},{"citing_arxiv_id":"2507.03724","citing_title":"MemOS: A Memory OS for AI System","ref_index":68,"is_internal_anchor":true},{"citing_arxiv_id":"2301.05217","citing_title":"Progress measures for grokking via mechanistic interpretability","ref_index":22,"is_internal_anchor":true},{"citing_arxiv_id":"2604.06052","citing_title":"Attention, May I Have Your Decision? Localizing Generative Choices in Diffusion Models","ref_index":24,"is_internal_anchor":true},{"citing_arxiv_id":"2605.12809","citing_title":"Correcting Influence: Unboxing LLM Outputs with Orthogonal Latent Spaces","ref_index":52,"is_internal_anchor":true},{"citing_arxiv_id":"2605.12991","citing_title":"Not Just RLHF: Why Alignment Alone Won't Fix Multi-Agent Sycophancy","ref_index":25,"is_internal_anchor":true},{"citing_arxiv_id":"2211.00593","citing_title":"Interpretability in the Wild: a Circuit for Indirect Object Identification in GPT-2 small","ref_index":52,"is_internal_anchor":true},{"citing_arxiv_id":"2403.19647","citing_title":"Sparse Feature Circuits: Discovering and Editing Interpretable Causal Graphs in Language Models","ref_index":46,"is_internal_anchor":true},{"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":25,"is_internal_anchor":true},{"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":25,"is_internal_anchor":true},{"citing_arxiv_id":"2604.22128","citing_title":"Dissociating Decodability and Causal Use in Bracket-Sequence Transformers","ref_index":8,"is_internal_anchor":true},{"citing_arxiv_id":"2605.00226","citing_title":"Why Do LLMs Struggle in Strategic Play? Broken Links Between Observations, Beliefs, and Actions","ref_index":41,"is_internal_anchor":true},{"citing_arxiv_id":"2604.22117","citing_title":"PermaFrost-Attack: Stealth Pretraining Seeding(SPS) for planting Logic Landmines During LLM Training","ref_index":161,"is_internal_anchor":true},{"citing_arxiv_id":"2604.21139","citing_title":"Slot Machines: How LLMs Keep Track of Multiple Entities","ref_index":9,"is_internal_anchor":true},{"citing_arxiv_id":"2604.19089","citing_title":"Towards Scalable Lifelong Knowledge Editing with Selective Knowledge Suppression","ref_index":29,"is_internal_anchor":true},{"citing_arxiv_id":"2604.19826","citing_title":"Co-Located Tests, Better AI Code: How Test Syntax Structure Affects Foundation Model Code Generation","ref_index":20,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":2,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/GKEBQHZOUDEA77XMILT44NLAZP","json":"https://pith.science/pith/GKEBQHZOUDEA77XMILT44NLAZP.json","graph_json":"https://pith.science/api/pith-number/GKEBQHZOUDEA77XMILT44NLAZP/graph.json","events_json":"https://pith.science/api/pith-number/GKEBQHZOUDEA77XMILT44NLAZP/events.json","paper":"https://pith.science/paper/GKEBQHZO"},"agent_actions":{"view_html":"https://pith.science/pith/GKEBQHZOUDEA77XMILT44NLAZP","download_json":"https://pith.science/pith/GKEBQHZOUDEA77XMILT44NLAZP.json","view_paper":"https://pith.science/paper/GKEBQHZO","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2202.05262&json=true","fetch_graph":"https://pith.science/api/pith-number/GKEBQHZOUDEA77XMILT44NLAZP/graph.json","fetch_events":"https://pith.science/api/pith-number/GKEBQHZOUDEA77XMILT44NLAZP/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/GKEBQHZOUDEA77XMILT44NLAZP/action/timestamp_anchor","attest_storage":"https://pith.science/pith/GKEBQHZOUDEA77XMILT44NLAZP/action/storage_attestation","attest_author":"https://pith.science/pith/GKEBQHZOUDEA77XMILT44NLAZP/action/author_attestation","sign_citation":"https://pith.science/pith/GKEBQHZOUDEA77XMILT44NLAZP/action/citation_signature","submit_replication":"https://pith.science/pith/GKEBQHZOUDEA77XMILT44NLAZP/action/replication_record"}},"created_at":"2026-05-17T23:38:48.267271+00:00","updated_at":"2026-05-17T23:38:48.267271+00:00"}