{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:WPAPRZMGFB7KHRAUZUI7CNXMX3","short_pith_number":"pith:WPAPRZMG","schema_version":"1.0","canonical_sha256":"b3c0f8e586287ea3c414cd11f136ecbedcb3d236a53568dbec1e513e4fc227cf","source":{"kind":"arxiv","id":"2404.08008","version":2},"attestation_state":"computed","paper":{"title":"Sample-Efficient Human Evaluation of Large Language Models via Maximum Discrepancy Competition","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CL","cs.HC"],"primary_cat":"cs.LG","authors_text":"Ge Sun, Guozhou Zheng, Hongzhi Tan, Huajun Chen, Kede Ma, Kehua Feng, Keyan Ding, Qiang Zhang, Shuangquan Guo, Yuzhou Cheng, Zhihua Wang","submitted_at":"2024-04-10T01:26:24Z","abstract_excerpt":"Reliable evaluation of large language models (LLMs) is impeded by two key challenges: objective metrics often fail to reflect human perception of natural language, and exhaustive human labeling is prohibitively expensive. Here, we propose a sample-efficient human evaluation method for LLMs based on the principle of MAximum Discrepancy (MAD) Competition. Our method automatically and adaptively selects a compact set of input instructions that maximize semantic discrepancy between pairs of LLM responses. Human evaluators then perform three-alternative forced choices on these paired responses, whi"},"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":"2404.08008","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2024-04-10T01:26:24Z","cross_cats_sorted":["cs.CL","cs.HC"],"title_canon_sha256":"e220484dbcbfedd224a709f17a09e0eef70df6c4b00ffffb0ed325e789abd209","abstract_canon_sha256":"0b4b8d811d1cef860ff21daced64be78263cd2af1aa7100e1f559827877d8f34"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:11:31.748102Z","signature_b64":"DyA6jwxNZfYH2sjrAO47h89Wpy4BmBGc0kRsnZSaQw4bJhHveaa/PuyINLGkIRkDii/mX0iTdnah9sKiOUIJAg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"b3c0f8e586287ea3c414cd11f136ecbedcb3d236a53568dbec1e513e4fc227cf","last_reissued_at":"2026-07-05T11:11:31.747518Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:11:31.747518Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Sample-Efficient Human Evaluation of Large Language Models via Maximum Discrepancy Competition","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CL","cs.HC"],"primary_cat":"cs.LG","authors_text":"Ge Sun, Guozhou Zheng, Hongzhi Tan, Huajun Chen, Kede Ma, Kehua Feng, Keyan Ding, Qiang Zhang, Shuangquan Guo, Yuzhou Cheng, Zhihua Wang","submitted_at":"2024-04-10T01:26:24Z","abstract_excerpt":"Reliable evaluation of large language models (LLMs) is impeded by two key challenges: objective metrics often fail to reflect human perception of natural language, and exhaustive human labeling is prohibitively expensive. Here, we propose a sample-efficient human evaluation method for LLMs based on the principle of MAximum Discrepancy (MAD) Competition. Our method automatically and adaptively selects a compact set of input instructions that maximize semantic discrepancy between pairs of LLM responses. Human evaluators then perform three-alternative forced choices on these paired responses, whi"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2404.08008","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/2404.08008/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":"2404.08008","created_at":"2026-07-05T11:11:31.747584+00:00"},{"alias_kind":"arxiv_version","alias_value":"2404.08008v2","created_at":"2026-07-05T11:11:31.747584+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2404.08008","created_at":"2026-07-05T11:11:31.747584+00:00"},{"alias_kind":"pith_short_12","alias_value":"WPAPRZMGFB7K","created_at":"2026-07-05T11:11:31.747584+00:00"},{"alias_kind":"pith_short_16","alias_value":"WPAPRZMGFB7KHRAU","created_at":"2026-07-05T11:11:31.747584+00:00"},{"alias_kind":"pith_short_8","alias_value":"WPAPRZMG","created_at":"2026-07-05T11:11:31.747584+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2501.18251","citing_title":"How to Select Datapoints for Efficient Human Evaluation of NLG Models?","ref_index":12,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/WPAPRZMGFB7KHRAUZUI7CNXMX3","json":"https://pith.science/pith/WPAPRZMGFB7KHRAUZUI7CNXMX3.json","graph_json":"https://pith.science/api/pith-number/WPAPRZMGFB7KHRAUZUI7CNXMX3/graph.json","events_json":"https://pith.science/api/pith-number/WPAPRZMGFB7KHRAUZUI7CNXMX3/events.json","paper":"https://pith.science/paper/WPAPRZMG"},"agent_actions":{"view_html":"https://pith.science/pith/WPAPRZMGFB7KHRAUZUI7CNXMX3","download_json":"https://pith.science/pith/WPAPRZMGFB7KHRAUZUI7CNXMX3.json","view_paper":"https://pith.science/paper/WPAPRZMG","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2404.08008&json=true","fetch_graph":"https://pith.science/api/pith-number/WPAPRZMGFB7KHRAUZUI7CNXMX3/graph.json","fetch_events":"https://pith.science/api/pith-number/WPAPRZMGFB7KHRAUZUI7CNXMX3/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/WPAPRZMGFB7KHRAUZUI7CNXMX3/action/timestamp_anchor","attest_storage":"https://pith.science/pith/WPAPRZMGFB7KHRAUZUI7CNXMX3/action/storage_attestation","attest_author":"https://pith.science/pith/WPAPRZMGFB7KHRAUZUI7CNXMX3/action/author_attestation","sign_citation":"https://pith.science/pith/WPAPRZMGFB7KHRAUZUI7CNXMX3/action/citation_signature","submit_replication":"https://pith.science/pith/WPAPRZMGFB7KHRAUZUI7CNXMX3/action/replication_record"}},"created_at":"2026-07-05T11:11:31.747584+00:00","updated_at":"2026-07-05T11:11:31.747584+00:00"}