{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:MGTVR7K5JTF2LPVQQUIIWEUNXM","short_pith_number":"pith:MGTVR7K5","schema_version":"1.0","canonical_sha256":"61a758fd5d4ccba5beb085108b128dbb2c25031a43c049ef2f686e1fe3c4efba","source":{"kind":"arxiv","id":"2405.14604","version":3},"attestation_state":"computed","paper":{"title":"Watermarking Low-entropy Generation for Large Language Models: An Unbiased and Low-risk Method","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Dongjun Wei, Michael Chau, Minjia Mao, Xiao Fang, Zeyu Chen","submitted_at":"2024-05-23T14:17:29Z","abstract_excerpt":"Recent advancements in large language models (LLMs) have highlighted the risk of misusing them, raising the need for accurate detection of LLM-generated content. In response, a viable solution is to inject imperceptible identifiers into LLMs, known as watermarks. Our research extends the existing watermarking methods by proposing the novel Sampling One Then Accepting (STA-1) method. STA-1 is an unbiased watermark that preserves the original token distribution in expectation and has a lower risk of producing unsatisfactory outputs in low-entropy scenarios compared to existing unbiased watermark"},"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":"2405.14604","kind":"arxiv","version":3},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2024-05-23T14:17:29Z","cross_cats_sorted":[],"title_canon_sha256":"83e99367469238b73d9ff4438ba1b847ee7f3940802150cc06c9dad6de88ebb7","abstract_canon_sha256":"c7616300347ac8b02ae4b7f1496c67665f63c6b86d18cfccae895730b60b15e7"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:11:19.700392Z","signature_b64":"466zD+JRfe+AZLfMAAa4wCVzmS52wBMT/7G9En/AkYs4EvFOFszkSxylLUodtr9VRpSrQS6lYOE+GsQW9m5uCA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"61a758fd5d4ccba5beb085108b128dbb2c25031a43c049ef2f686e1fe3c4efba","last_reissued_at":"2026-07-05T10:11:19.699851Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:11:19.699851Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Watermarking Low-entropy Generation for Large Language Models: An Unbiased and Low-risk Method","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Dongjun Wei, Michael Chau, Minjia Mao, Xiao Fang, Zeyu Chen","submitted_at":"2024-05-23T14:17:29Z","abstract_excerpt":"Recent advancements in large language models (LLMs) have highlighted the risk of misusing them, raising the need for accurate detection of LLM-generated content. In response, a viable solution is to inject imperceptible identifiers into LLMs, known as watermarks. Our research extends the existing watermarking methods by proposing the novel Sampling One Then Accepting (STA-1) method. STA-1 is an unbiased watermark that preserves the original token distribution in expectation and has a lower risk of producing unsatisfactory outputs in low-entropy scenarios compared to existing unbiased watermark"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2405.14604","kind":"arxiv","version":3},"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/2405.14604/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":"2405.14604","created_at":"2026-07-05T10:11:19.699911+00:00"},{"alias_kind":"arxiv_version","alias_value":"2405.14604v3","created_at":"2026-07-05T10:11:19.699911+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2405.14604","created_at":"2026-07-05T10:11:19.699911+00:00"},{"alias_kind":"pith_short_12","alias_value":"MGTVR7K5JTF2","created_at":"2026-07-05T10:11:19.699911+00:00"},{"alias_kind":"pith_short_16","alias_value":"MGTVR7K5JTF2LPVQ","created_at":"2026-07-05T10:11:19.699911+00:00"},{"alias_kind":"pith_short_8","alias_value":"MGTVR7K5","created_at":"2026-07-05T10:11:19.699911+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":4,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2605.30501","citing_title":"Linear Ensembles Wash Away Watermarks: On the Fragility of Distributional Perturbations in LLMs","ref_index":6,"is_internal_anchor":false},{"citing_arxiv_id":"2404.02138","citing_title":"Topic-Based Watermarks for Large Language Models","ref_index":32,"is_internal_anchor":false},{"citing_arxiv_id":"2605.08964","citing_title":"Trustworthy AI: Ensuring Reliability and Accountability from Models to Agents","ref_index":111,"is_internal_anchor":false},{"citing_arxiv_id":"2604.16001","citing_title":"MATRIX: Multi-Layer Code Watermarking via Dual-Channel Constrained Parity-Check Encoding","ref_index":47,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/MGTVR7K5JTF2LPVQQUIIWEUNXM","json":"https://pith.science/pith/MGTVR7K5JTF2LPVQQUIIWEUNXM.json","graph_json":"https://pith.science/api/pith-number/MGTVR7K5JTF2LPVQQUIIWEUNXM/graph.json","events_json":"https://pith.science/api/pith-number/MGTVR7K5JTF2LPVQQUIIWEUNXM/events.json","paper":"https://pith.science/paper/MGTVR7K5"},"agent_actions":{"view_html":"https://pith.science/pith/MGTVR7K5JTF2LPVQQUIIWEUNXM","download_json":"https://pith.science/pith/MGTVR7K5JTF2LPVQQUIIWEUNXM.json","view_paper":"https://pith.science/paper/MGTVR7K5","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2405.14604&json=true","fetch_graph":"https://pith.science/api/pith-number/MGTVR7K5JTF2LPVQQUIIWEUNXM/graph.json","fetch_events":"https://pith.science/api/pith-number/MGTVR7K5JTF2LPVQQUIIWEUNXM/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/MGTVR7K5JTF2LPVQQUIIWEUNXM/action/timestamp_anchor","attest_storage":"https://pith.science/pith/MGTVR7K5JTF2LPVQQUIIWEUNXM/action/storage_attestation","attest_author":"https://pith.science/pith/MGTVR7K5JTF2LPVQQUIIWEUNXM/action/author_attestation","sign_citation":"https://pith.science/pith/MGTVR7K5JTF2LPVQQUIIWEUNXM/action/citation_signature","submit_replication":"https://pith.science/pith/MGTVR7K5JTF2LPVQQUIIWEUNXM/action/replication_record"}},"created_at":"2026-07-05T10:11:19.699911+00:00","updated_at":"2026-07-05T10:11:19.699911+00:00"}