{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2023:JJNRQPM3WTO4QZSIZE4TZKZVRR","short_pith_number":"pith:JJNRQPM3","schema_version":"1.0","canonical_sha256":"4a5b183d9bb4ddc86648c9393cab358c46a7cf0a3c51362d63069266fdc07815","source":{"kind":"arxiv","id":"2310.17191","version":2},"attestation_state":"computed","paper":{"title":"How do Language Models Bind Entities in Context?","license":"http://creativecommons.org/licenses/by-sa/4.0/","headline":"","cross_cats":["cs.AI","cs.CL"],"primary_cat":"cs.LG","authors_text":"Jacob Steinhardt, Jiahai Feng","submitted_at":"2023-10-26T07:10:31Z","abstract_excerpt":"To correctly use in-context information, language models (LMs) must bind entities to their attributes. For example, given a context describing a \"green square\" and a \"blue circle\", LMs must bind the shapes to their respective colors. We analyze LM representations and identify the binding ID mechanism: a general mechanism for solving the binding problem, which we observe in every sufficiently large model from the Pythia and LLaMA families. Using causal interventions, we show that LMs' internal activations represent binding information by attaching binding ID vectors to corresponding entities an"},"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":"2310.17191","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by-sa/4.0/","primary_cat":"cs.LG","submitted_at":"2023-10-26T07:10:31Z","cross_cats_sorted":["cs.AI","cs.CL"],"title_canon_sha256":"82a22ebd8156423af10ecd7eb697bee499d68ba785153c93348f548f33f303e2","abstract_canon_sha256":"2dda4c30da6b9cbb1e4967fdaeb9aa5a6f611f28700d0eb0e12361710467b7c4"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T08:15:41.026138Z","signature_b64":"4HilsikF8VQxxpj7x6z2YcFvuG2HTYZqzf20raY3R8emInxchGfR8zU/mKakXJHN5rvHdFUjumRxg5o1eQ2cDw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"4a5b183d9bb4ddc86648c9393cab358c46a7cf0a3c51362d63069266fdc07815","last_reissued_at":"2026-07-05T08:15:41.025554Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T08:15:41.025554Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"How do Language Models Bind Entities in Context?","license":"http://creativecommons.org/licenses/by-sa/4.0/","headline":"","cross_cats":["cs.AI","cs.CL"],"primary_cat":"cs.LG","authors_text":"Jacob Steinhardt, Jiahai Feng","submitted_at":"2023-10-26T07:10:31Z","abstract_excerpt":"To correctly use in-context information, language models (LMs) must bind entities to their attributes. For example, given a context describing a \"green square\" and a \"blue circle\", LMs must bind the shapes to their respective colors. We analyze LM representations and identify the binding ID mechanism: a general mechanism for solving the binding problem, which we observe in every sufficiently large model from the Pythia and LLaMA families. Using causal interventions, we show that LMs' internal activations represent binding information by attaching binding ID vectors to corresponding entities an"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2310.17191","kind":"arxiv","version":2},"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/2310.17191/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":"2310.17191","created_at":"2026-07-05T08:15:41.025616+00:00"},{"alias_kind":"arxiv_version","alias_value":"2310.17191v2","created_at":"2026-07-05T08:15:41.025616+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2310.17191","created_at":"2026-07-05T08:15:41.025616+00:00"},{"alias_kind":"pith_short_12","alias_value":"JJNRQPM3WTO4","created_at":"2026-07-05T08:15:41.025616+00:00"},{"alias_kind":"pith_short_16","alias_value":"JJNRQPM3WTO4QZSI","created_at":"2026-07-05T08:15:41.025616+00:00"},{"alias_kind":"pith_short_8","alias_value":"JJNRQPM3","created_at":"2026-07-05T08:15:41.025616+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":6,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2605.29634","citing_title":"Relational Rank Geometry in Transformers: Detecting and Steering Hidden-State Relation Frames","ref_index":15,"is_internal_anchor":false},{"citing_arxiv_id":"2408.12935","citing_title":"AI Safety Landscape for Large Language Models: Taxonomy, State-of-the-art, and Future Directions","ref_index":207,"is_internal_anchor":false},{"citing_arxiv_id":"2404.15255","citing_title":"How to use and interpret activation patching","ref_index":4,"is_internal_anchor":false},{"citing_arxiv_id":"2604.21139","citing_title":"Slot Machines: How LLMs Keep Track of Multiple Entities","ref_index":6,"is_internal_anchor":false},{"citing_arxiv_id":"2604.19052","citing_title":"Cell-Based Representation of Relational Binding in Language Models","ref_index":25,"is_internal_anchor":false},{"citing_arxiv_id":"2605.07127","citing_title":"The Position Curse: LLMs Struggle to Locate the Last Few Items in a List","ref_index":5,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/JJNRQPM3WTO4QZSIZE4TZKZVRR","json":"https://pith.science/pith/JJNRQPM3WTO4QZSIZE4TZKZVRR.json","graph_json":"https://pith.science/api/pith-number/JJNRQPM3WTO4QZSIZE4TZKZVRR/graph.json","events_json":"https://pith.science/api/pith-number/JJNRQPM3WTO4QZSIZE4TZKZVRR/events.json","paper":"https://pith.science/paper/JJNRQPM3"},"agent_actions":{"view_html":"https://pith.science/pith/JJNRQPM3WTO4QZSIZE4TZKZVRR","download_json":"https://pith.science/pith/JJNRQPM3WTO4QZSIZE4TZKZVRR.json","view_paper":"https://pith.science/paper/JJNRQPM3","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2310.17191&json=true","fetch_graph":"https://pith.science/api/pith-number/JJNRQPM3WTO4QZSIZE4TZKZVRR/graph.json","fetch_events":"https://pith.science/api/pith-number/JJNRQPM3WTO4QZSIZE4TZKZVRR/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/JJNRQPM3WTO4QZSIZE4TZKZVRR/action/timestamp_anchor","attest_storage":"https://pith.science/pith/JJNRQPM3WTO4QZSIZE4TZKZVRR/action/storage_attestation","attest_author":"https://pith.science/pith/JJNRQPM3WTO4QZSIZE4TZKZVRR/action/author_attestation","sign_citation":"https://pith.science/pith/JJNRQPM3WTO4QZSIZE4TZKZVRR/action/citation_signature","submit_replication":"https://pith.science/pith/JJNRQPM3WTO4QZSIZE4TZKZVRR/action/replication_record"}},"created_at":"2026-07-05T08:15:41.025616+00:00","updated_at":"2026-07-05T08:15:41.025616+00:00"}