{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2023:ON52DM22442GTKSNNYLXZHNI4N","short_pith_number":"pith:ON52DM22","schema_version":"1.0","canonical_sha256":"737ba1b35ae73469aa4d6e177c9da8e34890523622cc47c0aec714b3c0e52dad","source":{"kind":"arxiv","id":"2311.12997","version":2},"attestation_state":"computed","paper":{"title":"Compositional Capabilities of Autoregressive Transformers: A Study on Synthetic, Interpretable Tasks","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Ekdeep Singh Lubana, Hidenori Tanaka, Mikail Khona, Rahul Ramesh, Robert P. Dick","submitted_at":"2023-11-21T21:16:54Z","abstract_excerpt":"Transformers trained on huge text corpora exhibit a remarkable set of capabilities, e.g., performing basic arithmetic. Given the inherent compositional nature of language, one can expect the model to learn to compose these capabilities, potentially yielding a combinatorial explosion of what operations it can perform on an input. Motivated by the above, we train autoregressive Transformer models on a synthetic data-generating process that involves compositions of a set of well-defined monolithic capabilities. Through a series of extensive and systematic experiments on this data-generating proce"},"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":"2311.12997","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2023-11-21T21:16:54Z","cross_cats_sorted":[],"title_canon_sha256":"bb74202589aca3c365cbbcd71410fd6c4d99817079fd7c539a234d4c9ec0590a","abstract_canon_sha256":"8bbb46e4c7e686360d5ca262d9a85df498b77a0c0743f2567f18be4c3626888b"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T07:41:45.288724Z","signature_b64":"Br/a3c9sulcrkiK/XCmRGCaGINguuYzw3pM4HYgdP1PfjXlK/f+8Vv7rVSHV/eQF1NGTt+3sMMZ56mmOi3CpBw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"737ba1b35ae73469aa4d6e177c9da8e34890523622cc47c0aec714b3c0e52dad","last_reissued_at":"2026-07-05T07:41:45.288164Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T07:41:45.288164Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Compositional Capabilities of Autoregressive Transformers: A Study on Synthetic, Interpretable Tasks","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Ekdeep Singh Lubana, Hidenori Tanaka, Mikail Khona, Rahul Ramesh, Robert P. Dick","submitted_at":"2023-11-21T21:16:54Z","abstract_excerpt":"Transformers trained on huge text corpora exhibit a remarkable set of capabilities, e.g., performing basic arithmetic. Given the inherent compositional nature of language, one can expect the model to learn to compose these capabilities, potentially yielding a combinatorial explosion of what operations it can perform on an input. Motivated by the above, we train autoregressive Transformer models on a synthetic data-generating process that involves compositions of a set of well-defined monolithic capabilities. Through a series of extensive and systematic experiments on this data-generating proce"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2311.12997","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/2311.12997/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":"2311.12997","created_at":"2026-07-05T07:41:45.288234+00:00"},{"alias_kind":"arxiv_version","alias_value":"2311.12997v2","created_at":"2026-07-05T07:41:45.288234+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2311.12997","created_at":"2026-07-05T07:41:45.288234+00:00"},{"alias_kind":"pith_short_12","alias_value":"ON52DM22442G","created_at":"2026-07-05T07:41:45.288234+00:00"},{"alias_kind":"pith_short_16","alias_value":"ON52DM22442GTKSN","created_at":"2026-07-05T07:41:45.288234+00:00"},{"alias_kind":"pith_short_8","alias_value":"ON52DM22","created_at":"2026-07-05T07:41:45.288234+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":4,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.07568","citing_title":"A Systematic Study of Behavioral Cloning for Scientific Data Annotation","ref_index":216,"is_internal_anchor":false},{"citing_arxiv_id":"2605.03413","citing_title":"Learning to Theorize the World from Observation","ref_index":286,"is_internal_anchor":false},{"citing_arxiv_id":"2605.05495","citing_title":"Shortcut Solutions Learned by Transformers Impair Continual Compositional Reasoning","ref_index":12,"is_internal_anchor":false},{"citing_arxiv_id":"2604.21691","citing_title":"There Will Be a Scientific Theory of Deep Learning","ref_index":289,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/ON52DM22442GTKSNNYLXZHNI4N","json":"https://pith.science/pith/ON52DM22442GTKSNNYLXZHNI4N.json","graph_json":"https://pith.science/api/pith-number/ON52DM22442GTKSNNYLXZHNI4N/graph.json","events_json":"https://pith.science/api/pith-number/ON52DM22442GTKSNNYLXZHNI4N/events.json","paper":"https://pith.science/paper/ON52DM22"},"agent_actions":{"view_html":"https://pith.science/pith/ON52DM22442GTKSNNYLXZHNI4N","download_json":"https://pith.science/pith/ON52DM22442GTKSNNYLXZHNI4N.json","view_paper":"https://pith.science/paper/ON52DM22","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2311.12997&json=true","fetch_graph":"https://pith.science/api/pith-number/ON52DM22442GTKSNNYLXZHNI4N/graph.json","fetch_events":"https://pith.science/api/pith-number/ON52DM22442GTKSNNYLXZHNI4N/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/ON52DM22442GTKSNNYLXZHNI4N/action/timestamp_anchor","attest_storage":"https://pith.science/pith/ON52DM22442GTKSNNYLXZHNI4N/action/storage_attestation","attest_author":"https://pith.science/pith/ON52DM22442GTKSNNYLXZHNI4N/action/author_attestation","sign_citation":"https://pith.science/pith/ON52DM22442GTKSNNYLXZHNI4N/action/citation_signature","submit_replication":"https://pith.science/pith/ON52DM22442GTKSNNYLXZHNI4N/action/replication_record"}},"created_at":"2026-07-05T07:41:45.288234+00:00","updated_at":"2026-07-05T07:41:45.288234+00:00"}