{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:CGIXX4LGZ3ZUDYHM2Q6X33BFBU","short_pith_number":"pith:CGIXX4LG","schema_version":"1.0","canonical_sha256":"11917bf166cef341e0ecd43d7dec250d06653ab9a8b5b5094605d9151d2028e3","source":{"kind":"arxiv","id":"2402.05785","version":5},"attestation_state":"computed","paper":{"title":"Limits of Transformer Language Models on Learning to Compose Algorithms","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.CL"],"primary_cat":"cs.LG","authors_text":"Abbas Rahimi, Aleksandar Terzic, Bernhard Sch\\\"olkopf, Giacomo Camposampiero, Jonathan Thomm, Michael Hersche","submitted_at":"2024-02-08T16:23:29Z","abstract_excerpt":"We analyze the capabilities of Transformer language models in learning compositional discrete tasks. To this end, we evaluate training LLaMA models and prompting GPT-4 and Gemini on four tasks demanding to learn a composition of several discrete sub-tasks. In particular, we measure how well these models can reuse primitives observable in the sub-tasks to learn the composition task. Our results indicate that compositional learning in state-of-the-art Transformer language models is highly sample inefficient: LLaMA requires more data samples than relearning all sub-tasks from scratch to learn the"},"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":"2402.05785","kind":"arxiv","version":5},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2024-02-08T16:23:29Z","cross_cats_sorted":["cs.AI","cs.CL"],"title_canon_sha256":"6b593d40a667c0097eab4e3305f0382e8703cf10fc9c936379105bbed3da767e","abstract_canon_sha256":"0b0e77e78fbf608ac65da86325d95cf60c7e44b6401b4da0f46f05352112384e"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T09:30:59.154645Z","signature_b64":"L/gq7tt0s7FSZcU8pQzn44nCjt7+Wr5bMz+QQYN5c/8F6A5whKy1i7kjjOmcJDq06OWdOD7NkSriAhVHQYKKDQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"11917bf166cef341e0ecd43d7dec250d06653ab9a8b5b5094605d9151d2028e3","last_reissued_at":"2026-07-05T09:30:59.154134Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T09:30:59.154134Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Limits of Transformer Language Models on Learning to Compose Algorithms","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.CL"],"primary_cat":"cs.LG","authors_text":"Abbas Rahimi, Aleksandar Terzic, Bernhard Sch\\\"olkopf, Giacomo Camposampiero, Jonathan Thomm, Michael Hersche","submitted_at":"2024-02-08T16:23:29Z","abstract_excerpt":"We analyze the capabilities of Transformer language models in learning compositional discrete tasks. To this end, we evaluate training LLaMA models and prompting GPT-4 and Gemini on four tasks demanding to learn a composition of several discrete sub-tasks. In particular, we measure how well these models can reuse primitives observable in the sub-tasks to learn the composition task. Our results indicate that compositional learning in state-of-the-art Transformer language models is highly sample inefficient: LLaMA requires more data samples than relearning all sub-tasks from scratch to learn the"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2402.05785","kind":"arxiv","version":5},"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/2402.05785/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":"2402.05785","created_at":"2026-07-05T09:30:59.154193+00:00"},{"alias_kind":"arxiv_version","alias_value":"2402.05785v5","created_at":"2026-07-05T09:30:59.154193+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2402.05785","created_at":"2026-07-05T09:30:59.154193+00:00"},{"alias_kind":"pith_short_12","alias_value":"CGIXX4LGZ3ZU","created_at":"2026-07-05T09:30:59.154193+00:00"},{"alias_kind":"pith_short_16","alias_value":"CGIXX4LGZ3ZUDYHM","created_at":"2026-07-05T09:30:59.154193+00:00"},{"alias_kind":"pith_short_8","alias_value":"CGIXX4LG","created_at":"2026-07-05T09:30:59.154193+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":0,"internal_anchor_count":0,"sample":[]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/CGIXX4LGZ3ZUDYHM2Q6X33BFBU","json":"https://pith.science/pith/CGIXX4LGZ3ZUDYHM2Q6X33BFBU.json","graph_json":"https://pith.science/api/pith-number/CGIXX4LGZ3ZUDYHM2Q6X33BFBU/graph.json","events_json":"https://pith.science/api/pith-number/CGIXX4LGZ3ZUDYHM2Q6X33BFBU/events.json","paper":"https://pith.science/paper/CGIXX4LG"},"agent_actions":{"view_html":"https://pith.science/pith/CGIXX4LGZ3ZUDYHM2Q6X33BFBU","download_json":"https://pith.science/pith/CGIXX4LGZ3ZUDYHM2Q6X33BFBU.json","view_paper":"https://pith.science/paper/CGIXX4LG","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2402.05785&json=true","fetch_graph":"https://pith.science/api/pith-number/CGIXX4LGZ3ZUDYHM2Q6X33BFBU/graph.json","fetch_events":"https://pith.science/api/pith-number/CGIXX4LGZ3ZUDYHM2Q6X33BFBU/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/CGIXX4LGZ3ZUDYHM2Q6X33BFBU/action/timestamp_anchor","attest_storage":"https://pith.science/pith/CGIXX4LGZ3ZUDYHM2Q6X33BFBU/action/storage_attestation","attest_author":"https://pith.science/pith/CGIXX4LGZ3ZUDYHM2Q6X33BFBU/action/author_attestation","sign_citation":"https://pith.science/pith/CGIXX4LGZ3ZUDYHM2Q6X33BFBU/action/citation_signature","submit_replication":"https://pith.science/pith/CGIXX4LGZ3ZUDYHM2Q6X33BFBU/action/replication_record"}},"created_at":"2026-07-05T09:30:59.154193+00:00","updated_at":"2026-07-05T09:30:59.154193+00:00"}