{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:DDECFACTQE2T6BTQLUHE5MCAPS","short_pith_number":"pith:DDECFACT","schema_version":"1.0","canonical_sha256":"18c822805381353f06705d0e4eb0407c80de9aeda52762e640fdee0fa6644c8d","source":{"kind":"arxiv","id":"2402.01781","version":2},"attestation_state":"computed","paper":{"title":"When Benchmarks are Targets: Revealing the Sensitivity of Large Language Model Leaderboards","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CL","authors_text":"Areeb Alowisheq, Faisal Mirza, Haidar Khan, Hisham Abdullah Alyahya, M Saiful Bari, Nora Altwairesh, Norah Alzahrani, Nouf Alotaibi, Shaykhah Alsubaie, Sultan Alrashed, Yazeed Alnumay, Yusef Almushaykeh","submitted_at":"2024-02-01T19:12:25Z","abstract_excerpt":"Large Language Model (LLM) leaderboards based on benchmark rankings are regularly used to guide practitioners in model selection. Often, the published leaderboard rankings are taken at face value - we show this is a (potentially costly) mistake. Under existing leaderboards, the relative performance of LLMs is highly sensitive to (often minute) details. We show that for popular multiple-choice question benchmarks (e.g., MMLU), minor perturbations to the benchmark, such as changing the order of choices or the method of answer selection, result in changes in rankings up to 8 positions. We explain"},"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":"2402.01781","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2024-02-01T19:12:25Z","cross_cats_sorted":["cs.AI","cs.LG"],"title_canon_sha256":"8299e860488dfabf231c255972a09fa41193af54417b9d54466120cdb35c21ee","abstract_canon_sha256":"b660cc79eabc2663da93ba989bad64f2aac0d16f8473ce419f69890b8ebbdd41"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T08:39:30.627989Z","signature_b64":"6Ol9YPvouXt6YdchvkvXV5+Yllz2r2HAIGOsntOoxItQDHoa/tXfrxrRDhmvePWMIraezk64Ozqj/u0cx0d5DQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"18c822805381353f06705d0e4eb0407c80de9aeda52762e640fdee0fa6644c8d","last_reissued_at":"2026-07-05T08:39:30.627437Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T08:39:30.627437Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"When Benchmarks are Targets: Revealing the Sensitivity of Large Language Model Leaderboards","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CL","authors_text":"Areeb Alowisheq, Faisal Mirza, Haidar Khan, Hisham Abdullah Alyahya, M Saiful Bari, Nora Altwairesh, Norah Alzahrani, Nouf Alotaibi, Shaykhah Alsubaie, Sultan Alrashed, Yazeed Alnumay, Yusef Almushaykeh","submitted_at":"2024-02-01T19:12:25Z","abstract_excerpt":"Large Language Model (LLM) leaderboards based on benchmark rankings are regularly used to guide practitioners in model selection. Often, the published leaderboard rankings are taken at face value - we show this is a (potentially costly) mistake. Under existing leaderboards, the relative performance of LLMs is highly sensitive to (often minute) details. We show that for popular multiple-choice question benchmarks (e.g., MMLU), minor perturbations to the benchmark, such as changing the order of choices or the method of answer selection, result in changes in rankings up to 8 positions. We explain"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2402.01781","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/2402.01781/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":"2402.01781","created_at":"2026-07-05T08:39:30.627515+00:00"},{"alias_kind":"arxiv_version","alias_value":"2402.01781v2","created_at":"2026-07-05T08:39:30.627515+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2402.01781","created_at":"2026-07-05T08:39:30.627515+00:00"},{"alias_kind":"pith_short_12","alias_value":"DDECFACTQE2T","created_at":"2026-07-05T08:39:30.627515+00:00"},{"alias_kind":"pith_short_16","alias_value":"DDECFACTQE2T6BTQ","created_at":"2026-07-05T08:39:30.627515+00:00"},{"alias_kind":"pith_short_8","alias_value":"DDECFACT","created_at":"2026-07-05T08:39:30.627515+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2403.12031","citing_title":"RouterBench: A Benchmark for Multi-LLM Routing System","ref_index":74,"is_internal_anchor":false},{"citing_arxiv_id":"2406.01574","citing_title":"MMLU-Pro: A More Robust and Challenging Multi-Task Language Understanding Benchmark","ref_index":3,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/DDECFACTQE2T6BTQLUHE5MCAPS","json":"https://pith.science/pith/DDECFACTQE2T6BTQLUHE5MCAPS.json","graph_json":"https://pith.science/api/pith-number/DDECFACTQE2T6BTQLUHE5MCAPS/graph.json","events_json":"https://pith.science/api/pith-number/DDECFACTQE2T6BTQLUHE5MCAPS/events.json","paper":"https://pith.science/paper/DDECFACT"},"agent_actions":{"view_html":"https://pith.science/pith/DDECFACTQE2T6BTQLUHE5MCAPS","download_json":"https://pith.science/pith/DDECFACTQE2T6BTQLUHE5MCAPS.json","view_paper":"https://pith.science/paper/DDECFACT","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2402.01781&json=true","fetch_graph":"https://pith.science/api/pith-number/DDECFACTQE2T6BTQLUHE5MCAPS/graph.json","fetch_events":"https://pith.science/api/pith-number/DDECFACTQE2T6BTQLUHE5MCAPS/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/DDECFACTQE2T6BTQLUHE5MCAPS/action/timestamp_anchor","attest_storage":"https://pith.science/pith/DDECFACTQE2T6BTQLUHE5MCAPS/action/storage_attestation","attest_author":"https://pith.science/pith/DDECFACTQE2T6BTQLUHE5MCAPS/action/author_attestation","sign_citation":"https://pith.science/pith/DDECFACTQE2T6BTQLUHE5MCAPS/action/citation_signature","submit_replication":"https://pith.science/pith/DDECFACTQE2T6BTQLUHE5MCAPS/action/replication_record"}},"created_at":"2026-07-05T08:39:30.627515+00:00","updated_at":"2026-07-05T08:39:30.627515+00:00"}