{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:GPUYBV6ARLT7ZPR6B4VCXJWKBT","short_pith_number":"pith:GPUYBV6A","schema_version":"1.0","canonical_sha256":"33e980d7c08ae7fcbe3e0f2a2ba6ca0cc2aba0bc3050dd42959976ea31b827bc","source":{"kind":"arxiv","id":"2406.13990","version":2},"attestation_state":"computed","paper":{"title":"Inference-Time Decontamination: Reusing Leaked Benchmarks for Large Language Model Evaluation","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Qingyuan Cheng, Qin Zhu, Runyu Peng, Ru Peng, Tengxiao Liu, Xiaonan Li, Xipeng Qiu, Xuanjing Huang","submitted_at":"2024-06-20T04:35:59Z","abstract_excerpt":"The training process of large language models (LLMs) often involves varying degrees of test data contamination. Although current LLMs are achieving increasingly better performance on various benchmarks, their performance in practical applications does not always match their benchmark results. Leakage of benchmarks can prevent the accurate assessment of LLMs' true performance. However, constructing new benchmarks is costly, labor-intensive and still carries the risk of leakage. Therefore, in this paper, we ask the question, Can we reuse these leaked benchmarks for LLM evaluation? We propose Inf"},"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":"2406.13990","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2024-06-20T04:35:59Z","cross_cats_sorted":[],"title_canon_sha256":"c7c95a5ce927977a0ab36b21bd4ebbd654cf62eddb1d05494ed4002b33cad479","abstract_canon_sha256":"08d2818312adfd726ae2d3e81547387cc12c62a932d5eba362d67f073a0c6d89"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T08:35:48.258556Z","signature_b64":"bi9JwjwLcc1ZfH7gmD6AZ1cyRGYsZsGPPnPC4D5P69IKHVadXMEXj980UiJAsEVvLwNEEx2hvKv4y13J5JC4AQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"33e980d7c08ae7fcbe3e0f2a2ba6ca0cc2aba0bc3050dd42959976ea31b827bc","last_reissued_at":"2026-07-05T08:35:48.258053Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T08:35:48.258053Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Inference-Time Decontamination: Reusing Leaked Benchmarks for Large Language Model Evaluation","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Qingyuan Cheng, Qin Zhu, Runyu Peng, Ru Peng, Tengxiao Liu, Xiaonan Li, Xipeng Qiu, Xuanjing Huang","submitted_at":"2024-06-20T04:35:59Z","abstract_excerpt":"The training process of large language models (LLMs) often involves varying degrees of test data contamination. Although current LLMs are achieving increasingly better performance on various benchmarks, their performance in practical applications does not always match their benchmark results. Leakage of benchmarks can prevent the accurate assessment of LLMs' true performance. However, constructing new benchmarks is costly, labor-intensive and still carries the risk of leakage. Therefore, in this paper, we ask the question, Can we reuse these leaked benchmarks for LLM evaluation? We propose Inf"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2406.13990","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/2406.13990/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":"2406.13990","created_at":"2026-07-05T08:35:48.258115+00:00"},{"alias_kind":"arxiv_version","alias_value":"2406.13990v2","created_at":"2026-07-05T08:35:48.258115+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2406.13990","created_at":"2026-07-05T08:35:48.258115+00:00"},{"alias_kind":"pith_short_12","alias_value":"GPUYBV6ARLT7","created_at":"2026-07-05T08:35:48.258115+00:00"},{"alias_kind":"pith_short_16","alias_value":"GPUYBV6ARLT7ZPR6","created_at":"2026-07-05T08:35:48.258115+00:00"},{"alias_kind":"pith_short_8","alias_value":"GPUYBV6A","created_at":"2026-07-05T08:35:48.258115+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2605.26133","citing_title":"Pretraining Data Exposure in Large Language Models: A Survey of Membership Inference, Data Contamination, and Security Implications","ref_index":68,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/GPUYBV6ARLT7ZPR6B4VCXJWKBT","json":"https://pith.science/pith/GPUYBV6ARLT7ZPR6B4VCXJWKBT.json","graph_json":"https://pith.science/api/pith-number/GPUYBV6ARLT7ZPR6B4VCXJWKBT/graph.json","events_json":"https://pith.science/api/pith-number/GPUYBV6ARLT7ZPR6B4VCXJWKBT/events.json","paper":"https://pith.science/paper/GPUYBV6A"},"agent_actions":{"view_html":"https://pith.science/pith/GPUYBV6ARLT7ZPR6B4VCXJWKBT","download_json":"https://pith.science/pith/GPUYBV6ARLT7ZPR6B4VCXJWKBT.json","view_paper":"https://pith.science/paper/GPUYBV6A","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2406.13990&json=true","fetch_graph":"https://pith.science/api/pith-number/GPUYBV6ARLT7ZPR6B4VCXJWKBT/graph.json","fetch_events":"https://pith.science/api/pith-number/GPUYBV6ARLT7ZPR6B4VCXJWKBT/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/GPUYBV6ARLT7ZPR6B4VCXJWKBT/action/timestamp_anchor","attest_storage":"https://pith.science/pith/GPUYBV6ARLT7ZPR6B4VCXJWKBT/action/storage_attestation","attest_author":"https://pith.science/pith/GPUYBV6ARLT7ZPR6B4VCXJWKBT/action/author_attestation","sign_citation":"https://pith.science/pith/GPUYBV6ARLT7ZPR6B4VCXJWKBT/action/citation_signature","submit_replication":"https://pith.science/pith/GPUYBV6ARLT7ZPR6B4VCXJWKBT/action/replication_record"}},"created_at":"2026-07-05T08:35:48.258115+00:00","updated_at":"2026-07-05T08:35:48.258115+00:00"}