{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:CFNYOKQMR6CYOCOLNPSI5NMCYR","short_pith_number":"pith:CFNYOKQM","schema_version":"1.0","canonical_sha256":"115b872a0c8f858709cb6be48eb582c46e8caeb2a737868486aee85e00e70e16","source":{"kind":"arxiv","id":"2505.20896","version":2},"attestation_state":"computed","paper":{"title":"How Do Transformers Learn Variable Binding in Symbolic Programs?","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.CL"],"primary_cat":"cs.LG","authors_text":"Atticus Geiger, Rapha\\\"el Milli\\`ere, Yiwei Wu","submitted_at":"2025-05-27T08:39:20Z","abstract_excerpt":"Variable binding -- the ability to associate variables with values -- is fundamental to symbolic computation and cognition. Although classical architectures typically implement variable binding via addressable memory, it is not well understood how modern neural networks lacking built-in binding operations may acquire this capacity. We investigate this by training a Transformer to dereference queried variables in symbolic programs where variables are assigned either numerical constants or other variables. Each program requires following chains of variable assignments up to four steps deep to fi"},"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":"2505.20896","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-05-27T08:39:20Z","cross_cats_sorted":["cs.AI","cs.CL"],"title_canon_sha256":"7f235d2d3965e0e75f5fb320e8b146192290987fed0e63683d3f4969376b69ff","abstract_canon_sha256":"6beb27faae481416b7aec83f70cd72fe1f1da32872f5326bcd261e9e5b535d0c"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:13:27.141231Z","signature_b64":"aWBwj6NDl/pQXjYGdejrAoETNDQSQC1AX1D3d5Vzom5RonwQRWk61zEZIGxQ92E8JUPTBK21AefOHOOl0H9MDQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"115b872a0c8f858709cb6be48eb582c46e8caeb2a737868486aee85e00e70e16","last_reissued_at":"2026-07-05T11:13:27.140727Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:13:27.140727Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"How Do Transformers Learn Variable Binding in Symbolic Programs?","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.CL"],"primary_cat":"cs.LG","authors_text":"Atticus Geiger, Rapha\\\"el Milli\\`ere, Yiwei Wu","submitted_at":"2025-05-27T08:39:20Z","abstract_excerpt":"Variable binding -- the ability to associate variables with values -- is fundamental to symbolic computation and cognition. Although classical architectures typically implement variable binding via addressable memory, it is not well understood how modern neural networks lacking built-in binding operations may acquire this capacity. We investigate this by training a Transformer to dereference queried variables in symbolic programs where variables are assigned either numerical constants or other variables. Each program requires following chains of variable assignments up to four steps deep to fi"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2505.20896","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/2505.20896/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":"2505.20896","created_at":"2026-07-05T11:13:27.140788+00:00"},{"alias_kind":"arxiv_version","alias_value":"2505.20896v2","created_at":"2026-07-05T11:13:27.140788+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2505.20896","created_at":"2026-07-05T11:13:27.140788+00:00"},{"alias_kind":"pith_short_12","alias_value":"CFNYOKQMR6CY","created_at":"2026-07-05T11:13:27.140788+00:00"},{"alias_kind":"pith_short_16","alias_value":"CFNYOKQMR6CYOCOL","created_at":"2026-07-05T11:13:27.140788+00:00"},{"alias_kind":"pith_short_8","alias_value":"CFNYOKQM","created_at":"2026-07-05T11:13:27.140788+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":5,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.26523","citing_title":"Radical AI Interpretability","ref_index":101,"is_internal_anchor":false},{"citing_arxiv_id":"2606.03976","citing_title":"Formalizing the Binding Problem","ref_index":93,"is_internal_anchor":false},{"citing_arxiv_id":"2605.31497","citing_title":"Assign and Add: A Mechanistic Study of Compositional Arithmetic","ref_index":22,"is_internal_anchor":false},{"citing_arxiv_id":"2604.12426","citing_title":"Do Transformers Use their Depth Adaptively? Evidence from a Relational Reasoning Task","ref_index":26,"is_internal_anchor":false},{"citing_arxiv_id":"2605.07120","citing_title":"When Symbol Names Should Not Matter: A Logistic Theory of Fresh-Symbol Classification","ref_index":42,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/CFNYOKQMR6CYOCOLNPSI5NMCYR","json":"https://pith.science/pith/CFNYOKQMR6CYOCOLNPSI5NMCYR.json","graph_json":"https://pith.science/api/pith-number/CFNYOKQMR6CYOCOLNPSI5NMCYR/graph.json","events_json":"https://pith.science/api/pith-number/CFNYOKQMR6CYOCOLNPSI5NMCYR/events.json","paper":"https://pith.science/paper/CFNYOKQM"},"agent_actions":{"view_html":"https://pith.science/pith/CFNYOKQMR6CYOCOLNPSI5NMCYR","download_json":"https://pith.science/pith/CFNYOKQMR6CYOCOLNPSI5NMCYR.json","view_paper":"https://pith.science/paper/CFNYOKQM","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2505.20896&json=true","fetch_graph":"https://pith.science/api/pith-number/CFNYOKQMR6CYOCOLNPSI5NMCYR/graph.json","fetch_events":"https://pith.science/api/pith-number/CFNYOKQMR6CYOCOLNPSI5NMCYR/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/CFNYOKQMR6CYOCOLNPSI5NMCYR/action/timestamp_anchor","attest_storage":"https://pith.science/pith/CFNYOKQMR6CYOCOLNPSI5NMCYR/action/storage_attestation","attest_author":"https://pith.science/pith/CFNYOKQMR6CYOCOLNPSI5NMCYR/action/author_attestation","sign_citation":"https://pith.science/pith/CFNYOKQMR6CYOCOLNPSI5NMCYR/action/citation_signature","submit_replication":"https://pith.science/pith/CFNYOKQMR6CYOCOLNPSI5NMCYR/action/replication_record"}},"created_at":"2026-07-05T11:13:27.140788+00:00","updated_at":"2026-07-05T11:13:27.140788+00:00"}