{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:DMLCBG6743IK6XZZF6IETMLPCA","short_pith_number":"pith:DMLCBG67","schema_version":"1.0","canonical_sha256":"1b16209bdfe6d0af5f392f9049b16f1028a841fed6b3a31a97eb7ce73c198eb3","source":{"kind":"arxiv","id":"2404.10271","version":2},"attestation_state":"computed","paper":{"title":"Social Choice Should Guide AI Alignment in Dealing with Diverse Human Feedback","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.CL","cs.CY","cs.GT"],"primary_cat":"cs.LG","authors_text":"Bob M. Jacobs, Emanuel Tewolde, Eric Pacuit, Hailey Schoelkopf, Jobst Heitzig, Milan Moss\\'e, Nathan Lambert, Rachel Freedman, Stuart Russell, Vincent Conitzer, Wesley H. Holliday, William S. Zwicker","submitted_at":"2024-04-16T03:59:33Z","abstract_excerpt":"Foundation models such as GPT-4 are fine-tuned to avoid unsafe or otherwise problematic behavior, such as helping to commit crimes or producing racist text. One approach to fine-tuning, called reinforcement learning from human feedback, learns from humans' expressed preferences over multiple outputs. Another approach is constitutional AI, in which the input from humans is a list of high-level principles. But how do we deal with potentially diverging input from humans? How can we aggregate the input into consistent data about \"collective\" preferences or otherwise use it to make collective choic"},"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":"2404.10271","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2024-04-16T03:59:33Z","cross_cats_sorted":["cs.AI","cs.CL","cs.CY","cs.GT"],"title_canon_sha256":"9bc6995a2cc334ffcff2da40ab48b1b44dac0f554a6f24b09e59ee502caf1b19","abstract_canon_sha256":"6d6e2a419000a8c038843d211e0420f1f21d2111f284ee139d2b703f1ce0afb0"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T08:27:07.469945Z","signature_b64":"6gkCDOJphI/0OJhZfhgF6wcX7LzQgYCZhd7Tt9XpxDsmsX0dygSqVRk4zrQJ09KzgqaXw4fmYKDNOw8SAbS0Cw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"1b16209bdfe6d0af5f392f9049b16f1028a841fed6b3a31a97eb7ce73c198eb3","last_reissued_at":"2026-07-05T08:27:07.469480Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T08:27:07.469480Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Social Choice Should Guide AI Alignment in Dealing with Diverse Human Feedback","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.CL","cs.CY","cs.GT"],"primary_cat":"cs.LG","authors_text":"Bob M. Jacobs, Emanuel Tewolde, Eric Pacuit, Hailey Schoelkopf, Jobst Heitzig, Milan Moss\\'e, Nathan Lambert, Rachel Freedman, Stuart Russell, Vincent Conitzer, Wesley H. Holliday, William S. Zwicker","submitted_at":"2024-04-16T03:59:33Z","abstract_excerpt":"Foundation models such as GPT-4 are fine-tuned to avoid unsafe or otherwise problematic behavior, such as helping to commit crimes or producing racist text. One approach to fine-tuning, called reinforcement learning from human feedback, learns from humans' expressed preferences over multiple outputs. Another approach is constitutional AI, in which the input from humans is a list of high-level principles. But how do we deal with potentially diverging input from humans? How can we aggregate the input into consistent data about \"collective\" preferences or otherwise use it to make collective choic"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2404.10271","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/2404.10271/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":"2404.10271","created_at":"2026-07-05T08:27:07.469537+00:00"},{"alias_kind":"arxiv_version","alias_value":"2404.10271v2","created_at":"2026-07-05T08:27:07.469537+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2404.10271","created_at":"2026-07-05T08:27:07.469537+00:00"},{"alias_kind":"pith_short_12","alias_value":"DMLCBG6743IK","created_at":"2026-07-05T08:27:07.469537+00:00"},{"alias_kind":"pith_short_16","alias_value":"DMLCBG6743IK6XZZ","created_at":"2026-07-05T08:27:07.469537+00:00"},{"alias_kind":"pith_short_8","alias_value":"DMLCBG67","created_at":"2026-07-05T08:27:07.469537+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":15,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.21001","citing_title":"Do Large Language Model Voters Strategize? An Oracle-Based Benchmark for Manipulation under Voting Rules","ref_index":7,"is_internal_anchor":false},{"citing_arxiv_id":"2606.10569","citing_title":"Hidden Consensus:Preference-Validity Compression in Human Feedback","ref_index":11,"is_internal_anchor":false},{"citing_arxiv_id":"2606.08367","citing_title":"Emergence World: A Platform for Evaluating Long-Horizon Multi-Agent Autonomy","ref_index":7,"is_internal_anchor":false},{"citing_arxiv_id":"2606.08267","citing_title":"Post-AGI Economies: Superposition and the Second Fundamental Theorem of Welfare Economics","ref_index":18,"is_internal_anchor":false},{"citing_arxiv_id":"2606.03110","citing_title":"Coherence Maximization Improves Pluralistic Alignment","ref_index":7,"is_internal_anchor":false},{"citing_arxiv_id":"2606.27578","citing_title":"PEBS: Per-rater Empirical-Bayes Shrinkage for RLHF Reward-Model Calibration","ref_index":2,"is_internal_anchor":false},{"citing_arxiv_id":"2605.24052","citing_title":"Truthful Online Preference Aggregation for LLM Fine-Tuning in Mobile Crowdsourcing","ref_index":11,"is_internal_anchor":false},{"citing_arxiv_id":"2310.15288","citing_title":"Active teacher selection for reward learning","ref_index":3,"is_internal_anchor":false},{"citing_arxiv_id":"2605.17510","citing_title":"Scale-Dependent Collective Adaptation in Self-Amending LLM Societies: A Cross-Family Study of Emergent Governance","ref_index":78,"is_internal_anchor":false},{"citing_arxiv_id":"2605.11873","citing_title":"Maximizing Reachability via Shifting of Temporal Paths","ref_index":226,"is_internal_anchor":false},{"citing_arxiv_id":"2605.11240","citing_title":"When to Ask a Question: Understanding Communication Strategies in Generative AI Tools","ref_index":10,"is_internal_anchor":false},{"citing_arxiv_id":"2604.25895","citing_title":"Three Models of RLHF Annotation: Extension, Evidence, and Authority","ref_index":15,"is_internal_anchor":false},{"citing_arxiv_id":"2604.10673","citing_title":"Principles Do Not Apply Themselves: A Hermeneutic Perspective on AI Alignment","ref_index":8,"is_internal_anchor":false},{"citing_arxiv_id":"2605.07724","citing_title":"Curated Synthetic Data Doesn't Have to Collapse: A Theoretical Study of Generative Retraining with Pluralistic Preferences","ref_index":104,"is_internal_anchor":false},{"citing_arxiv_id":"2604.20805","citing_title":"Relative Principals, Pluralistic Alignment, and the Structural Value Alignment Problem","ref_index":12,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/DMLCBG6743IK6XZZF6IETMLPCA","json":"https://pith.science/pith/DMLCBG6743IK6XZZF6IETMLPCA.json","graph_json":"https://pith.science/api/pith-number/DMLCBG6743IK6XZZF6IETMLPCA/graph.json","events_json":"https://pith.science/api/pith-number/DMLCBG6743IK6XZZF6IETMLPCA/events.json","paper":"https://pith.science/paper/DMLCBG67"},"agent_actions":{"view_html":"https://pith.science/pith/DMLCBG6743IK6XZZF6IETMLPCA","download_json":"https://pith.science/pith/DMLCBG6743IK6XZZF6IETMLPCA.json","view_paper":"https://pith.science/paper/DMLCBG67","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2404.10271&json=true","fetch_graph":"https://pith.science/api/pith-number/DMLCBG6743IK6XZZF6IETMLPCA/graph.json","fetch_events":"https://pith.science/api/pith-number/DMLCBG6743IK6XZZF6IETMLPCA/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/DMLCBG6743IK6XZZF6IETMLPCA/action/timestamp_anchor","attest_storage":"https://pith.science/pith/DMLCBG6743IK6XZZF6IETMLPCA/action/storage_attestation","attest_author":"https://pith.science/pith/DMLCBG6743IK6XZZF6IETMLPCA/action/author_attestation","sign_citation":"https://pith.science/pith/DMLCBG6743IK6XZZF6IETMLPCA/action/citation_signature","submit_replication":"https://pith.science/pith/DMLCBG6743IK6XZZF6IETMLPCA/action/replication_record"}},"created_at":"2026-07-05T08:27:07.469537+00:00","updated_at":"2026-07-05T08:27:07.469537+00:00"}