{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:HRSZJ35BYL4XAJAFGWLQBETHZ2","short_pith_number":"pith:HRSZJ35B","schema_version":"1.0","canonical_sha256":"3c6594efa1c2f97024053597009267ce858395588de5d2e7d9e6c4b440a88f7f","source":{"kind":"arxiv","id":"2502.16174","version":4},"attestation_state":"computed","paper":{"title":"Efficient LLM Moderation with Multi-Layer Latent Prototypes","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.CL","cs.CR"],"primary_cat":"cs.LG","authors_text":"Bartosz W\\'ojcik, Filip Szatkowski, Jan Dubi\\'nski, Maciej Chrab\\k{a}szcz, Sebastian Cygert, Tomasz Trzci\\'nski","submitted_at":"2025-02-22T10:31:50Z","abstract_excerpt":"Although modern LLMs are aligned with human values during post-training, robust moderation remains essential to prevent harmful outputs at deployment time. Existing approaches suffer from performance-efficiency trade-offs and are difficult to customize to user-specific requirements. Motivated by this gap, we introduce Multi-Layer Prototype Moderator (MLPM), a lightweight and highly customizable input moderation tool. We propose leveraging prototypes of intermediate representations across multiple layers to improve moderation quality while maintaining high efficiency. By design, our method adds"},"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":"2502.16174","kind":"arxiv","version":4},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-02-22T10:31:50Z","cross_cats_sorted":["cs.AI","cs.CL","cs.CR"],"title_canon_sha256":"f00f31467ebba7f6ae8d73540cde67350105a5290fca0294bb1fa5a1dd1d4b06","abstract_canon_sha256":"8005b2efe865632726cae17b867d65e555c9b75abcb45a9928e2098ccfc4d479"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-02T03:04:31.943208Z","signature_b64":"xUMji49VUgKO2r1tzjknHR+GeeuEEW1ibuJ27GIfGnzdZHueTZjkDjNfUJtvaU+d+Z1A7vqjLtA38i+B+QgkAg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"3c6594efa1c2f97024053597009267ce858395588de5d2e7d9e6c4b440a88f7f","last_reissued_at":"2026-06-02T03:04:31.942717Z","signature_status":"signed_v1","first_computed_at":"2026-06-02T03:04:31.942717Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Efficient LLM Moderation with Multi-Layer Latent Prototypes","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.CL","cs.CR"],"primary_cat":"cs.LG","authors_text":"Bartosz W\\'ojcik, Filip Szatkowski, Jan Dubi\\'nski, Maciej Chrab\\k{a}szcz, Sebastian Cygert, Tomasz Trzci\\'nski","submitted_at":"2025-02-22T10:31:50Z","abstract_excerpt":"Although modern LLMs are aligned with human values during post-training, robust moderation remains essential to prevent harmful outputs at deployment time. Existing approaches suffer from performance-efficiency trade-offs and are difficult to customize to user-specific requirements. Motivated by this gap, we introduce Multi-Layer Prototype Moderator (MLPM), a lightweight and highly customizable input moderation tool. We propose leveraging prototypes of intermediate representations across multiple layers to improve moderation quality while maintaining high efficiency. By design, our method adds"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2502.16174","kind":"arxiv","version":4},"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/2502.16174/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":"2502.16174","created_at":"2026-06-02T03:04:31.942786+00:00"},{"alias_kind":"arxiv_version","alias_value":"2502.16174v4","created_at":"2026-06-02T03:04:31.942786+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2502.16174","created_at":"2026-06-02T03:04:31.942786+00:00"},{"alias_kind":"pith_short_12","alias_value":"HRSZJ35BYL4X","created_at":"2026-06-02T03:04:31.942786+00:00"},{"alias_kind":"pith_short_16","alias_value":"HRSZJ35BYL4XAJAF","created_at":"2026-06-02T03:04:31.942786+00:00"},{"alias_kind":"pith_short_8","alias_value":"HRSZJ35B","created_at":"2026-06-02T03:04:31.942786+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":2,"sample":[{"citing_arxiv_id":"2605.18549","citing_title":"Monitoring the Internal Monologue: Probe Trajectories Reveal Reasoning Dynamics","ref_index":11,"is_internal_anchor":true},{"citing_arxiv_id":"2604.18519","citing_title":"LLM Safety From Within: Detecting Harmful Content with Internal Representations","ref_index":32,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/HRSZJ35BYL4XAJAFGWLQBETHZ2","json":"https://pith.science/pith/HRSZJ35BYL4XAJAFGWLQBETHZ2.json","graph_json":"https://pith.science/api/pith-number/HRSZJ35BYL4XAJAFGWLQBETHZ2/graph.json","events_json":"https://pith.science/api/pith-number/HRSZJ35BYL4XAJAFGWLQBETHZ2/events.json","paper":"https://pith.science/paper/HRSZJ35B"},"agent_actions":{"view_html":"https://pith.science/pith/HRSZJ35BYL4XAJAFGWLQBETHZ2","download_json":"https://pith.science/pith/HRSZJ35BYL4XAJAFGWLQBETHZ2.json","view_paper":"https://pith.science/paper/HRSZJ35B","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2502.16174&json=true","fetch_graph":"https://pith.science/api/pith-number/HRSZJ35BYL4XAJAFGWLQBETHZ2/graph.json","fetch_events":"https://pith.science/api/pith-number/HRSZJ35BYL4XAJAFGWLQBETHZ2/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/HRSZJ35BYL4XAJAFGWLQBETHZ2/action/timestamp_anchor","attest_storage":"https://pith.science/pith/HRSZJ35BYL4XAJAFGWLQBETHZ2/action/storage_attestation","attest_author":"https://pith.science/pith/HRSZJ35BYL4XAJAFGWLQBETHZ2/action/author_attestation","sign_citation":"https://pith.science/pith/HRSZJ35BYL4XAJAFGWLQBETHZ2/action/citation_signature","submit_replication":"https://pith.science/pith/HRSZJ35BYL4XAJAFGWLQBETHZ2/action/replication_record"}},"created_at":"2026-06-02T03:04:31.942786+00:00","updated_at":"2026-06-02T03:04:31.942786+00:00"}