{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:NQFEZJ54AXYBYYA4WEPZWWIL4Y","short_pith_number":"pith:NQFEZJ54","schema_version":"1.0","canonical_sha256":"6c0a4ca7bc05f01c601cb11f9b590be618caaf6577e229df7a716f31e28d93a8","source":{"kind":"arxiv","id":"2603.17218","version":2},"attestation_state":"computed","paper":{"title":"Alignment Makes Language Models Normative, Not Descriptive","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.GT"],"primary_cat":"cs.CL","authors_text":"Eilam Shapira, Moshe Tennenholtz, Roi Reichart","submitted_at":"2026-03-17T23:47:08Z","abstract_excerpt":"Post-training alignment optimizes language models to match human preference signals, but this objective is not equivalent to modeling observed human behavior. We compare 120 base-aligned model pairs on more than 10,000 real human decisions in multi-round strategic games - bargaining, persuasion, negotiation, and repeated matrix games. In these settings, base models outperform their aligned counterparts in predicting human choices by nearly 10:1, robustly across model families, prompt formulations, and game configurations. This pattern reverses, however, in settings where human behavior is more"},"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":"2603.17218","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2026-03-17T23:47:08Z","cross_cats_sorted":["cs.AI","cs.GT"],"title_canon_sha256":"3b486dc005dd10a1e1df23bdc87341c87ac3c6bd972bbf68db29979873b282ac","abstract_canon_sha256":"602ec11244be56e4c44efd76607ab20d462de65b77e24c18f2bbc9721b9211e5"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-27T01:04:57.261788Z","signature_b64":"Xc4PkTOOlqMiN8/aG3kmeFOtVQcAVkeCSH2uehd1rxU5is1dc8wEDeYCcyFbXWOASRCa4xAHNvauK/YXAEF3DA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"6c0a4ca7bc05f01c601cb11f9b590be618caaf6577e229df7a716f31e28d93a8","last_reissued_at":"2026-05-27T01:04:57.261087Z","signature_status":"signed_v1","first_computed_at":"2026-05-27T01:04:57.261087Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Alignment Makes Language Models Normative, Not Descriptive","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.GT"],"primary_cat":"cs.CL","authors_text":"Eilam Shapira, Moshe Tennenholtz, Roi Reichart","submitted_at":"2026-03-17T23:47:08Z","abstract_excerpt":"Post-training alignment optimizes language models to match human preference signals, but this objective is not equivalent to modeling observed human behavior. We compare 120 base-aligned model pairs on more than 10,000 real human decisions in multi-round strategic games - bargaining, persuasion, negotiation, and repeated matrix games. In these settings, base models outperform their aligned counterparts in predicting human choices by nearly 10:1, robustly across model families, prompt formulations, and game configurations. This pattern reverses, however, in settings where human behavior is more"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2603.17218","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/2603.17218/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":"2603.17218","created_at":"2026-05-27T01:04:57.261163+00:00"},{"alias_kind":"arxiv_version","alias_value":"2603.17218v2","created_at":"2026-05-27T01:04:57.261163+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2603.17218","created_at":"2026-05-27T01:04:57.261163+00:00"},{"alias_kind":"pith_short_12","alias_value":"NQFEZJ54AXYB","created_at":"2026-05-27T01:04:57.261163+00:00"},{"alias_kind":"pith_short_16","alias_value":"NQFEZJ54AXYBYYA4","created_at":"2026-05-27T01:04:57.261163+00:00"},{"alias_kind":"pith_short_8","alias_value":"NQFEZJ54","created_at":"2026-05-27T01:04:57.261163+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":3,"internal_anchor_count":3,"sample":[{"citing_arxiv_id":"2605.12199","citing_title":"Overtrained, Not Misaligned","ref_index":70,"is_internal_anchor":true},{"citing_arxiv_id":"2605.12411","citing_title":"Predicting Decisions of AI Agents from Limited Interaction through Text-Tabular Modeling","ref_index":64,"is_internal_anchor":true},{"citing_arxiv_id":"2605.07632","citing_title":"Post-training makes large language models less human-like","ref_index":40,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/NQFEZJ54AXYBYYA4WEPZWWIL4Y","json":"https://pith.science/pith/NQFEZJ54AXYBYYA4WEPZWWIL4Y.json","graph_json":"https://pith.science/api/pith-number/NQFEZJ54AXYBYYA4WEPZWWIL4Y/graph.json","events_json":"https://pith.science/api/pith-number/NQFEZJ54AXYBYYA4WEPZWWIL4Y/events.json","paper":"https://pith.science/paper/NQFEZJ54"},"agent_actions":{"view_html":"https://pith.science/pith/NQFEZJ54AXYBYYA4WEPZWWIL4Y","download_json":"https://pith.science/pith/NQFEZJ54AXYBYYA4WEPZWWIL4Y.json","view_paper":"https://pith.science/paper/NQFEZJ54","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2603.17218&json=true","fetch_graph":"https://pith.science/api/pith-number/NQFEZJ54AXYBYYA4WEPZWWIL4Y/graph.json","fetch_events":"https://pith.science/api/pith-number/NQFEZJ54AXYBYYA4WEPZWWIL4Y/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/NQFEZJ54AXYBYYA4WEPZWWIL4Y/action/timestamp_anchor","attest_storage":"https://pith.science/pith/NQFEZJ54AXYBYYA4WEPZWWIL4Y/action/storage_attestation","attest_author":"https://pith.science/pith/NQFEZJ54AXYBYYA4WEPZWWIL4Y/action/author_attestation","sign_citation":"https://pith.science/pith/NQFEZJ54AXYBYYA4WEPZWWIL4Y/action/citation_signature","submit_replication":"https://pith.science/pith/NQFEZJ54AXYBYYA4WEPZWWIL4Y/action/replication_record"}},"created_at":"2026-05-27T01:04:57.261163+00:00","updated_at":"2026-05-27T01:04:57.261163+00:00"}