{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:UFXXH6BV6MAQLGGTUZ47LOSMNC","short_pith_number":"pith:UFXXH6BV","schema_version":"1.0","canonical_sha256":"a16f73f835f3010598d3a679f5ba4c68b3956826c828fd203535c6a8b92a204e","source":{"kind":"arxiv","id":"2402.08164","version":2},"attestation_state":"computed","paper":{"title":"On Limitations of the Transformer Architecture","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"stat.ML","authors_text":"Binghui Peng, Christos Papadimitriou, Srini Narayanan","submitted_at":"2024-02-13T01:52:15Z","abstract_excerpt":"What are the root causes of hallucinations in large language models (LLMs)? We use Communication Complexity to prove that the Transformer layer is incapable of composing functions (e.g., identify a grandparent of a person in a genealogy) if the domains of the functions are large enough; we show through examples that this inability is already empirically present when the domains are quite small. We also point out that several mathematical tasks that are at the core of the so-called compositional tasks thought to be hard for LLMs are unlikely to be solvable by Transformers, for large enough inst"},"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":"2402.08164","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"stat.ML","submitted_at":"2024-02-13T01:52:15Z","cross_cats_sorted":["cs.AI","cs.LG"],"title_canon_sha256":"bed27af5cdad151226409164c5a0145167e8429dfc596a7419641958dfc90732","abstract_canon_sha256":"426f28e3e3a44e840b6b44027bed4f473defcb198cc3707f5c54b0f39dbc080a"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T07:49:36.680721Z","signature_b64":"PDEGwJIdqaqivRI/FaE3eqQYg+g64BRZikYybS3heg5PUesEWXz8Y/E0uo3rYpJZPvy22MtkQa6E/T1aC4u3DQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"a16f73f835f3010598d3a679f5ba4c68b3956826c828fd203535c6a8b92a204e","last_reissued_at":"2026-07-05T07:49:36.680164Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T07:49:36.680164Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"On Limitations of the Transformer Architecture","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"stat.ML","authors_text":"Binghui Peng, Christos Papadimitriou, Srini Narayanan","submitted_at":"2024-02-13T01:52:15Z","abstract_excerpt":"What are the root causes of hallucinations in large language models (LLMs)? We use Communication Complexity to prove that the Transformer layer is incapable of composing functions (e.g., identify a grandparent of a person in a genealogy) if the domains of the functions are large enough; we show through examples that this inability is already empirically present when the domains are quite small. We also point out that several mathematical tasks that are at the core of the so-called compositional tasks thought to be hard for LLMs are unlikely to be solvable by Transformers, for large enough inst"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2402.08164","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/2402.08164/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":"2402.08164","created_at":"2026-07-05T07:49:36.680239+00:00"},{"alias_kind":"arxiv_version","alias_value":"2402.08164v2","created_at":"2026-07-05T07:49:36.680239+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2402.08164","created_at":"2026-07-05T07:49:36.680239+00:00"},{"alias_kind":"pith_short_12","alias_value":"UFXXH6BV6MAQ","created_at":"2026-07-05T07:49:36.680239+00:00"},{"alias_kind":"pith_short_16","alias_value":"UFXXH6BV6MAQLGGT","created_at":"2026-07-05T07:49:36.680239+00:00"},{"alias_kind":"pith_short_8","alias_value":"UFXXH6BV","created_at":"2026-07-05T07:49:36.680239+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":4,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2607.06720","citing_title":"When Does In-Context Search Help? A Sampling-Complexity Theory of Reflection-Driven Reasoning","ref_index":13,"is_internal_anchor":true},{"citing_arxiv_id":"2503.18018","citing_title":"Lost in Cultural Translation: Do LLMs Struggle with Math Across Cultural Contexts?","ref_index":12,"is_internal_anchor":false},{"citing_arxiv_id":"2410.05229","citing_title":"GSM-Symbolic: Understanding the Limitations of Mathematical Reasoning in Large Language Models","ref_index":89,"is_internal_anchor":false},{"citing_arxiv_id":"2604.17935","citing_title":"How Much Cache Does Reasoning Need? Depth-Cache Tradeoffs in KV-Compressed Transformers","ref_index":13,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/UFXXH6BV6MAQLGGTUZ47LOSMNC","json":"https://pith.science/pith/UFXXH6BV6MAQLGGTUZ47LOSMNC.json","graph_json":"https://pith.science/api/pith-number/UFXXH6BV6MAQLGGTUZ47LOSMNC/graph.json","events_json":"https://pith.science/api/pith-number/UFXXH6BV6MAQLGGTUZ47LOSMNC/events.json","paper":"https://pith.science/paper/UFXXH6BV"},"agent_actions":{"view_html":"https://pith.science/pith/UFXXH6BV6MAQLGGTUZ47LOSMNC","download_json":"https://pith.science/pith/UFXXH6BV6MAQLGGTUZ47LOSMNC.json","view_paper":"https://pith.science/paper/UFXXH6BV","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2402.08164&json=true","fetch_graph":"https://pith.science/api/pith-number/UFXXH6BV6MAQLGGTUZ47LOSMNC/graph.json","fetch_events":"https://pith.science/api/pith-number/UFXXH6BV6MAQLGGTUZ47LOSMNC/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/UFXXH6BV6MAQLGGTUZ47LOSMNC/action/timestamp_anchor","attest_storage":"https://pith.science/pith/UFXXH6BV6MAQLGGTUZ47LOSMNC/action/storage_attestation","attest_author":"https://pith.science/pith/UFXXH6BV6MAQLGGTUZ47LOSMNC/action/author_attestation","sign_citation":"https://pith.science/pith/UFXXH6BV6MAQLGGTUZ47LOSMNC/action/citation_signature","submit_replication":"https://pith.science/pith/UFXXH6BV6MAQLGGTUZ47LOSMNC/action/replication_record"}},"created_at":"2026-07-05T07:49:36.680239+00:00","updated_at":"2026-07-05T07:49:36.680239+00:00"}