{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:FXEMNKQF5VOIUYGWVUUXIG2G4D","short_pith_number":"pith:FXEMNKQF","schema_version":"1.0","canonical_sha256":"2dc8c6aa05ed5c8a60d6ad29741b46e0d5dfcb600d2c8620272740d522740a17","source":{"kind":"arxiv","id":"2412.15255","version":2},"attestation_state":"computed","paper":{"title":"Data Laundering: Artificially Boosting Benchmark Results through Knowledge Distillation","license":"http://creativecommons.org/licenses/by-sa/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Akhmed Sakip, Alham Fikri Aji, Jonibek Mansurov","submitted_at":"2024-12-15T19:38:48Z","abstract_excerpt":"In this paper, we show that knowledge distillation can be subverted to manipulate language model benchmark scores, revealing a critical vulnerability in current evaluation practices. We introduce \"Data Laundering,\" a process that enables the covert transfer of benchmark-specific knowledge through seemingly legitimate intermediate training steps. Through extensive experiments with a 2-layer BERT student model, we show how this approach can achieve substantial improvements in benchmark accuracy (up to 75\\% on GPQA) without developing genuine reasoning capabilities. Notably, this method can be ex"},"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":"2412.15255","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by-sa/4.0/","primary_cat":"cs.CL","submitted_at":"2024-12-15T19:38:48Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"aceb7a0d0eea4afb0ab46643b3f17c33ec1a55ab7430fa4e681d65585ed9cd3d","abstract_canon_sha256":"c352bd58a9062a3b1219622ce9b177cc4966a55764b088f4314143a3b6823f3d"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:15:41.300189Z","signature_b64":"uEaNPfQAA8PwysJ0cK/NnxaiIEU7+DUwqlaPu3SKUAjRcEAJxStPQsCGxQAq4BWMpJ+/vG5ZTKtZ1OzVkj9DCA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"2dc8c6aa05ed5c8a60d6ad29741b46e0d5dfcb600d2c8620272740d522740a17","last_reissued_at":"2026-07-05T11:15:41.299704Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:15:41.299704Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Data Laundering: Artificially Boosting Benchmark Results through Knowledge Distillation","license":"http://creativecommons.org/licenses/by-sa/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Akhmed Sakip, Alham Fikri Aji, Jonibek Mansurov","submitted_at":"2024-12-15T19:38:48Z","abstract_excerpt":"In this paper, we show that knowledge distillation can be subverted to manipulate language model benchmark scores, revealing a critical vulnerability in current evaluation practices. We introduce \"Data Laundering,\" a process that enables the covert transfer of benchmark-specific knowledge through seemingly legitimate intermediate training steps. Through extensive experiments with a 2-layer BERT student model, we show how this approach can achieve substantial improvements in benchmark accuracy (up to 75\\% on GPQA) without developing genuine reasoning capabilities. Notably, this method can be ex"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2412.15255","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/2412.15255/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":"2412.15255","created_at":"2026-07-05T11:15:41.299755+00:00"},{"alias_kind":"arxiv_version","alias_value":"2412.15255v2","created_at":"2026-07-05T11:15:41.299755+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2412.15255","created_at":"2026-07-05T11:15:41.299755+00:00"},{"alias_kind":"pith_short_12","alias_value":"FXEMNKQF5VOI","created_at":"2026-07-05T11:15:41.299755+00:00"},{"alias_kind":"pith_short_16","alias_value":"FXEMNKQF5VOIUYGW","created_at":"2026-07-05T11:15:41.299755+00:00"},{"alias_kind":"pith_short_8","alias_value":"FXEMNKQF","created_at":"2026-07-05T11:15:41.299755+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2605.18702","citing_title":"Distilling Tabular Foundation Models for Structured Health Data","ref_index":12,"is_internal_anchor":false},{"citing_arxiv_id":"2605.18654","citing_title":"Pocket Foundation Models: Distilling TFMs into CPU-Ready Gradient-Boosted Trees","ref_index":5,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/FXEMNKQF5VOIUYGWVUUXIG2G4D","json":"https://pith.science/pith/FXEMNKQF5VOIUYGWVUUXIG2G4D.json","graph_json":"https://pith.science/api/pith-number/FXEMNKQF5VOIUYGWVUUXIG2G4D/graph.json","events_json":"https://pith.science/api/pith-number/FXEMNKQF5VOIUYGWVUUXIG2G4D/events.json","paper":"https://pith.science/paper/FXEMNKQF"},"agent_actions":{"view_html":"https://pith.science/pith/FXEMNKQF5VOIUYGWVUUXIG2G4D","download_json":"https://pith.science/pith/FXEMNKQF5VOIUYGWVUUXIG2G4D.json","view_paper":"https://pith.science/paper/FXEMNKQF","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2412.15255&json=true","fetch_graph":"https://pith.science/api/pith-number/FXEMNKQF5VOIUYGWVUUXIG2G4D/graph.json","fetch_events":"https://pith.science/api/pith-number/FXEMNKQF5VOIUYGWVUUXIG2G4D/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/FXEMNKQF5VOIUYGWVUUXIG2G4D/action/timestamp_anchor","attest_storage":"https://pith.science/pith/FXEMNKQF5VOIUYGWVUUXIG2G4D/action/storage_attestation","attest_author":"https://pith.science/pith/FXEMNKQF5VOIUYGWVUUXIG2G4D/action/author_attestation","sign_citation":"https://pith.science/pith/FXEMNKQF5VOIUYGWVUUXIG2G4D/action/citation_signature","submit_replication":"https://pith.science/pith/FXEMNKQF5VOIUYGWVUUXIG2G4D/action/replication_record"}},"created_at":"2026-07-05T11:15:41.299755+00:00","updated_at":"2026-07-05T11:15:41.299755+00:00"}