{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:4XHWHWUO7GLMBQO4WRZI5PID3Q","short_pith_number":"pith:4XHWHWUO","schema_version":"1.0","canonical_sha256":"e5cf63da8ef996c0c1dcb4728ebd03dc2027426563b2fc24809ffd0a45a53f52","source":{"kind":"arxiv","id":"2410.19034","version":2},"attestation_state":"computed","paper":{"title":"Mixture of Parrots: Experts improve memorization more than reasoning","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Alex Gu, Clara Mohri, David Alvarez-Melis, David Brandfonbrener, Eran Malach, Nikhil Anand, Nikhil Vyas, Samy Jelassi, Sham M. Kakade, Yuanzhi Li","submitted_at":"2024-10-24T17:54:41Z","abstract_excerpt":"The Mixture-of-Experts (MoE) architecture enables a significant increase in the total number of model parameters with minimal computational overhead. However, it is not clear what performance tradeoffs, if any, exist between MoEs and standard dense transformers. In this paper, we show that as we increase the number of experts (while fixing the number of active parameters), the memorization performance consistently increases while the reasoning capabilities saturate. We begin by analyzing the theoretical limitations of MoEs at reasoning. We prove that there exist graph problems that cannot be s"},"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":"2410.19034","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2024-10-24T17:54:41Z","cross_cats_sorted":[],"title_canon_sha256":"10d09e3d143d30e61d322269b36a1fec26bea2d3d9de76c150e81c82b8b5f9da","abstract_canon_sha256":"4f5c5909f85ea4b5cf524b8f9f1cf741da17ca26b27177804400dacf45f91a38"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:22:08.618469Z","signature_b64":"DOGp6X1BMXCNPwtxIB+eZYJknTols27/4CjNo2TjCgXgzae3L8O7hWCcliVosDlVOGnwwNPKbIw6tfMxq8tGBg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"e5cf63da8ef996c0c1dcb4728ebd03dc2027426563b2fc24809ffd0a45a53f52","last_reissued_at":"2026-07-05T10:22:08.617852Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:22:08.617852Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Mixture of Parrots: Experts improve memorization more than reasoning","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Alex Gu, Clara Mohri, David Alvarez-Melis, David Brandfonbrener, Eran Malach, Nikhil Anand, Nikhil Vyas, Samy Jelassi, Sham M. Kakade, Yuanzhi Li","submitted_at":"2024-10-24T17:54:41Z","abstract_excerpt":"The Mixture-of-Experts (MoE) architecture enables a significant increase in the total number of model parameters with minimal computational overhead. However, it is not clear what performance tradeoffs, if any, exist between MoEs and standard dense transformers. In this paper, we show that as we increase the number of experts (while fixing the number of active parameters), the memorization performance consistently increases while the reasoning capabilities saturate. We begin by analyzing the theoretical limitations of MoEs at reasoning. We prove that there exist graph problems that cannot be s"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2410.19034","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/2410.19034/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":"2410.19034","created_at":"2026-07-05T10:22:08.617931+00:00"},{"alias_kind":"arxiv_version","alias_value":"2410.19034v2","created_at":"2026-07-05T10:22:08.617931+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2410.19034","created_at":"2026-07-05T10:22:08.617931+00:00"},{"alias_kind":"pith_short_12","alias_value":"4XHWHWUO7GLM","created_at":"2026-07-05T10:22:08.617931+00:00"},{"alias_kind":"pith_short_16","alias_value":"4XHWHWUO7GLMBQO4","created_at":"2026-07-05T10:22:08.617931+00:00"},{"alias_kind":"pith_short_8","alias_value":"4XHWHWUO","created_at":"2026-07-05T10:22:08.617931+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.15079","citing_title":"Ling and Ring 2.6 Technical Report: Efficient and Instant Agentic Intelligence at Trillion-Parameter Scale","ref_index":79,"is_internal_anchor":false},{"citing_arxiv_id":"2604.08519","citing_title":"Cram Less to Fit More: Training Data Pruning Improves Memorization of Facts","ref_index":35,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/4XHWHWUO7GLMBQO4WRZI5PID3Q","json":"https://pith.science/pith/4XHWHWUO7GLMBQO4WRZI5PID3Q.json","graph_json":"https://pith.science/api/pith-number/4XHWHWUO7GLMBQO4WRZI5PID3Q/graph.json","events_json":"https://pith.science/api/pith-number/4XHWHWUO7GLMBQO4WRZI5PID3Q/events.json","paper":"https://pith.science/paper/4XHWHWUO"},"agent_actions":{"view_html":"https://pith.science/pith/4XHWHWUO7GLMBQO4WRZI5PID3Q","download_json":"https://pith.science/pith/4XHWHWUO7GLMBQO4WRZI5PID3Q.json","view_paper":"https://pith.science/paper/4XHWHWUO","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2410.19034&json=true","fetch_graph":"https://pith.science/api/pith-number/4XHWHWUO7GLMBQO4WRZI5PID3Q/graph.json","fetch_events":"https://pith.science/api/pith-number/4XHWHWUO7GLMBQO4WRZI5PID3Q/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/4XHWHWUO7GLMBQO4WRZI5PID3Q/action/timestamp_anchor","attest_storage":"https://pith.science/pith/4XHWHWUO7GLMBQO4WRZI5PID3Q/action/storage_attestation","attest_author":"https://pith.science/pith/4XHWHWUO7GLMBQO4WRZI5PID3Q/action/author_attestation","sign_citation":"https://pith.science/pith/4XHWHWUO7GLMBQO4WRZI5PID3Q/action/citation_signature","submit_replication":"https://pith.science/pith/4XHWHWUO7GLMBQO4WRZI5PID3Q/action/replication_record"}},"created_at":"2026-07-05T10:22:08.617931+00:00","updated_at":"2026-07-05T10:22:08.617931+00:00"}