{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:Y57MRET3NST3BH23BP2L5LHSKF","short_pith_number":"pith:Y57MRET3","schema_version":"1.0","canonical_sha256":"c77ec8927b6ca7b09f5b0bf4beacf25163cca468a563d4c5513b01955079e0a6","source":{"kind":"arxiv","id":"2510.01171","version":4},"attestation_state":"computed","paper":{"title":"Verbalized Sampling: How to Mitigate Mode Collapse and Unlock LLM Diversity","license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Anthony Sicilia, Christopher D. Manning, Derek Chong, Jiayi Zhang, Michael R. Tomz, Simon Yu, Weiyan Shi","submitted_at":"2025-10-01T17:55:37Z","abstract_excerpt":"Post-training alignment often reduces LLM diversity, leading to a phenomenon known as mode collapse. Unlike prior work that attributes this effect to algorithmic limitations, we identify a fundamental, pervasive data-level driver: typicality bias in preference data, whereby annotators systematically favor familiar text as a result of well-established findings in cognitive psychology. We formalize this bias theoretically, verify it on preference datasets empirically, and show that it plays a central role in mode collapse. Motivated by this analysis, we introduce Verbalized Sampling, a simple, t"},"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":"2510.01171","kind":"arxiv","version":4},"metadata":{"license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","primary_cat":"cs.CL","submitted_at":"2025-10-01T17:55:37Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"07dbe66f0f4c1b8c84864a51b18b4fb86acce9974162b034b4979bef85e0dc82","abstract_canon_sha256":"f2fc1b8c53bd3458030fc924bd90c04abe5207a61afd66868350f9755a1c519d"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-17T00:20:41.119042Z","signature_b64":"OC9pJqynPaEE46JSCRXyg8AQamI3dAXJmAA0ubkUs7uzomV4xIuNMevHM9sPP6jVO48ZmDb46E/0hAEZv+ojCg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"c77ec8927b6ca7b09f5b0bf4beacf25163cca468a563d4c5513b01955079e0a6","last_reissued_at":"2026-07-17T00:20:41.117999Z","signature_status":"signed_v1","first_computed_at":"2026-07-17T00:20:41.117999Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Verbalized Sampling: How to Mitigate Mode Collapse and Unlock LLM Diversity","license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Anthony Sicilia, Christopher D. Manning, Derek Chong, Jiayi Zhang, Michael R. Tomz, Simon Yu, Weiyan Shi","submitted_at":"2025-10-01T17:55:37Z","abstract_excerpt":"Post-training alignment often reduces LLM diversity, leading to a phenomenon known as mode collapse. Unlike prior work that attributes this effect to algorithmic limitations, we identify a fundamental, pervasive data-level driver: typicality bias in preference data, whereby annotators systematically favor familiar text as a result of well-established findings in cognitive psychology. We formalize this bias theoretically, verify it on preference datasets empirically, and show that it plays a central role in mode collapse. Motivated by this analysis, we introduce Verbalized Sampling, a simple, t"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2510.01171","kind":"arxiv","version":4},"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/2510.01171/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":"2510.01171","created_at":"2026-07-17T00:20:41.118456+00:00"},{"alias_kind":"arxiv_version","alias_value":"2510.01171v4","created_at":"2026-07-17T00:20:41.118456+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2510.01171","created_at":"2026-07-17T00:20:41.118456+00:00"},{"alias_kind":"pith_short_12","alias_value":"Y57MRET3NST3","created_at":"2026-07-17T00:20:41.118456+00:00"},{"alias_kind":"pith_short_16","alias_value":"Y57MRET3NST3BH23","created_at":"2026-07-17T00:20:41.118456+00:00"},{"alias_kind":"pith_short_8","alias_value":"Y57MRET3","created_at":"2026-07-17T00:20:41.118456+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":22,"internal_anchor_count":22,"sample":[{"citing_arxiv_id":"2604.22503","citing_title":"Measuring and Mitigating Persona Distortions from AI Writing Assistance","ref_index":42,"is_internal_anchor":true},{"citing_arxiv_id":"2606.24937","citing_title":"The Hitchhiker's Guide to Agentic AI: From Foundations to Systems","ref_index":121,"is_internal_anchor":true},{"citing_arxiv_id":"2606.09587","citing_title":"Seeing the Hivemind: A Consensus-Aware Interaction Technique for Mitigating AI Homogenization","ref_index":29,"is_internal_anchor":true},{"citing_arxiv_id":"2606.09603","citing_title":"Automated IEP Generation from Traditional Chinese Parent-Teacher Interviews via Corpus-Grounded Feature Diffusion","ref_index":5,"is_internal_anchor":true},{"citing_arxiv_id":"2606.06667","citing_title":"The Piggyback Hypothesis of Generalization: Explaining and Mitigating Emergent Misalignment","ref_index":28,"is_internal_anchor":true},{"citing_arxiv_id":"2606.01811","citing_title":"\"I've Seen How This Goes\": Characterizing Diversity via Progressive Conditional Surprise","ref_index":25,"is_internal_anchor":true},{"citing_arxiv_id":"2605.10913","citing_title":"Shepherd: Enabling Programmable Meta-Agents via Reversible Agentic Execution Traces","ref_index":54,"is_internal_anchor":true},{"citing_arxiv_id":"2606.29745","citing_title":"ECHO: Learning Epistemically Adaptive Language Agents with Turn-Level Credit","ref_index":74,"is_internal_anchor":true},{"citing_arxiv_id":"2605.25680","citing_title":"Simulating Human Memory with Language Models","ref_index":86,"is_internal_anchor":true},{"citing_arxiv_id":"2605.28346","citing_title":"When Discourse Pressures Conflict: Information Structure in Vision-Language Model Outputs","ref_index":12,"is_internal_anchor":true},{"citing_arxiv_id":"2605.31099","citing_title":"KnowledgeGain: Evaluating and Optimizing Science News Generation for Reader Learning","ref_index":5,"is_internal_anchor":true},{"citing_arxiv_id":"2604.02385","citing_title":"Validity, Sparse Holes, and Breadth in Language Generation: Banach Density, Topology, and Geometry","ref_index":34,"is_internal_anchor":true},{"citing_arxiv_id":"2604.06061","citing_title":"PromptEvolver: Prompt Inversion through Evolutionary Optimization in Natural-Language Space","ref_index":50,"is_internal_anchor":true},{"citing_arxiv_id":"2605.11258","citing_title":"Unlocking LLM Creativity in Science through Analogical Reasoning","ref_index":58,"is_internal_anchor":true},{"citing_arxiv_id":"2605.10913","citing_title":"Shepherd: Enabling Programmable Meta-Agents via Reversible Agentic Execution Traces","ref_index":54,"is_internal_anchor":true},{"citing_arxiv_id":"2605.09995","citing_title":"Annotations Mitigate Post-Training Mode Collapse","ref_index":2,"is_internal_anchor":true},{"citing_arxiv_id":"2604.22503","citing_title":"Measuring and Mitigating Persona Distortions from AI Writing Assistance","ref_index":42,"is_internal_anchor":true},{"citing_arxiv_id":"2604.18607","citing_title":"TurboEvolve: Towards Fast and Robust LLM-Driven Program Evolution","ref_index":22,"is_internal_anchor":true},{"citing_arxiv_id":"2604.07147","citing_title":"Dynamic Context Evolution for Scalable Synthetic Data Generation","ref_index":16,"is_internal_anchor":true},{"citing_arxiv_id":"2604.26965","citing_title":"The Impact of AI-Generated Text on the Internet","ref_index":35,"is_internal_anchor":true},{"citing_arxiv_id":"2604.17244","citing_title":"DORA Explorer: Improving the Exploration Ability of LLMs Without Training","ref_index":34,"is_internal_anchor":true},{"citing_arxiv_id":"2605.02815","citing_title":"FlexSQL: Flexible Exploration and Execution Make Better Text-to-SQL Agents","ref_index":19,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/Y57MRET3NST3BH23BP2L5LHSKF","json":"https://pith.science/pith/Y57MRET3NST3BH23BP2L5LHSKF.json","graph_json":"https://pith.science/api/pith-number/Y57MRET3NST3BH23BP2L5LHSKF/graph.json","events_json":"https://pith.science/api/pith-number/Y57MRET3NST3BH23BP2L5LHSKF/events.json","paper":"https://pith.science/paper/Y57MRET3"},"agent_actions":{"view_html":"https://pith.science/pith/Y57MRET3NST3BH23BP2L5LHSKF","download_json":"https://pith.science/pith/Y57MRET3NST3BH23BP2L5LHSKF.json","view_paper":"https://pith.science/paper/Y57MRET3","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2510.01171&json=true","fetch_graph":"https://pith.science/api/pith-number/Y57MRET3NST3BH23BP2L5LHSKF/graph.json","fetch_events":"https://pith.science/api/pith-number/Y57MRET3NST3BH23BP2L5LHSKF/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/Y57MRET3NST3BH23BP2L5LHSKF/action/timestamp_anchor","attest_storage":"https://pith.science/pith/Y57MRET3NST3BH23BP2L5LHSKF/action/storage_attestation","attest_author":"https://pith.science/pith/Y57MRET3NST3BH23BP2L5LHSKF/action/author_attestation","sign_citation":"https://pith.science/pith/Y57MRET3NST3BH23BP2L5LHSKF/action/citation_signature","submit_replication":"https://pith.science/pith/Y57MRET3NST3BH23BP2L5LHSKF/action/replication_record"}},"created_at":"2026-07-17T00:20:41.118456+00:00","updated_at":"2026-07-17T00:20:41.118456+00:00"}