{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:ZKR4Q6OLTXK7IHRXON6JBFMZOO","short_pith_number":"pith:ZKR4Q6OL","schema_version":"1.0","canonical_sha256":"caa3c879cb9dd5f41e37737c90959973a9fc8644c4e901e338030f7bd59e7a8b","source":{"kind":"arxiv","id":"2412.13435","version":1},"attestation_state":"computed","paper":{"title":"Lightweight Safety Classification Using Pruned Language Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CL","authors_text":"Jim Brown, Mason Sawtell, Sandi Besen, Tula Masterman","submitted_at":"2024-12-18T02:13:13Z","abstract_excerpt":"In this paper, we introduce a novel technique for content safety and prompt injection classification for Large Language Models. Our technique, Layer Enhanced Classification (LEC), trains a Penalized Logistic Regression (PLR) classifier on the hidden state of an LLM's optimal intermediate transformer layer. By combining the computational efficiency of a streamlined PLR classifier with the sophisticated language understanding of an LLM, our approach delivers superior performance surpassing GPT-4o and special-purpose models fine-tuned for each task. We find that small general-purpose models (Qwen"},"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":"2412.13435","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2024-12-18T02:13:13Z","cross_cats_sorted":["cs.AI","cs.LG"],"title_canon_sha256":"34a26a31f9d55dfed769905c4c9aa91c3534c2fbec03ad574e49dce1c7e98a87","abstract_canon_sha256":"2c901052179ff741860285401d20f91b56b9cade77d267a8dafa75c7e8995573"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T09:51:01.699250Z","signature_b64":"7TfIdrLS8/5qe9j317sOzUeFoKM40YxwdFY4JXD/Rg8h5zUIhi1FKnhLgxVV5MCB8sy3c9mSTVUzaLTa8NbUDg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"caa3c879cb9dd5f41e37737c90959973a9fc8644c4e901e338030f7bd59e7a8b","last_reissued_at":"2026-07-05T09:51:01.698700Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T09:51:01.698700Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Lightweight Safety Classification Using Pruned Language Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CL","authors_text":"Jim Brown, Mason Sawtell, Sandi Besen, Tula Masterman","submitted_at":"2024-12-18T02:13:13Z","abstract_excerpt":"In this paper, we introduce a novel technique for content safety and prompt injection classification for Large Language Models. Our technique, Layer Enhanced Classification (LEC), trains a Penalized Logistic Regression (PLR) classifier on the hidden state of an LLM's optimal intermediate transformer layer. By combining the computational efficiency of a streamlined PLR classifier with the sophisticated language understanding of an LLM, our approach delivers superior performance surpassing GPT-4o and special-purpose models fine-tuned for each task. We find that small general-purpose models (Qwen"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2412.13435","kind":"arxiv","version":1},"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/2412.13435/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":"2412.13435","created_at":"2026-07-05T09:51:01.698762+00:00"},{"alias_kind":"arxiv_version","alias_value":"2412.13435v1","created_at":"2026-07-05T09:51:01.698762+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2412.13435","created_at":"2026-07-05T09:51:01.698762+00:00"},{"alias_kind":"pith_short_12","alias_value":"ZKR4Q6OLTXK7","created_at":"2026-07-05T09:51:01.698762+00:00"},{"alias_kind":"pith_short_16","alias_value":"ZKR4Q6OLTXK7IHRX","created_at":"2026-07-05T09:51:01.698762+00:00"},{"alias_kind":"pith_short_8","alias_value":"ZKR4Q6OL","created_at":"2026-07-05T09:51:01.698762+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2506.00166","citing_title":"Disentangled Safety Adapters Enable Efficient Guardrails and Flexible Inference-Time Alignment","ref_index":38,"is_internal_anchor":false},{"citing_arxiv_id":"2604.18519","citing_title":"LLM Safety From Within: Detecting Harmful Content with Internal Representations","ref_index":45,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/ZKR4Q6OLTXK7IHRXON6JBFMZOO","json":"https://pith.science/pith/ZKR4Q6OLTXK7IHRXON6JBFMZOO.json","graph_json":"https://pith.science/api/pith-number/ZKR4Q6OLTXK7IHRXON6JBFMZOO/graph.json","events_json":"https://pith.science/api/pith-number/ZKR4Q6OLTXK7IHRXON6JBFMZOO/events.json","paper":"https://pith.science/paper/ZKR4Q6OL"},"agent_actions":{"view_html":"https://pith.science/pith/ZKR4Q6OLTXK7IHRXON6JBFMZOO","download_json":"https://pith.science/pith/ZKR4Q6OLTXK7IHRXON6JBFMZOO.json","view_paper":"https://pith.science/paper/ZKR4Q6OL","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2412.13435&json=true","fetch_graph":"https://pith.science/api/pith-number/ZKR4Q6OLTXK7IHRXON6JBFMZOO/graph.json","fetch_events":"https://pith.science/api/pith-number/ZKR4Q6OLTXK7IHRXON6JBFMZOO/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/ZKR4Q6OLTXK7IHRXON6JBFMZOO/action/timestamp_anchor","attest_storage":"https://pith.science/pith/ZKR4Q6OLTXK7IHRXON6JBFMZOO/action/storage_attestation","attest_author":"https://pith.science/pith/ZKR4Q6OLTXK7IHRXON6JBFMZOO/action/author_attestation","sign_citation":"https://pith.science/pith/ZKR4Q6OLTXK7IHRXON6JBFMZOO/action/citation_signature","submit_replication":"https://pith.science/pith/ZKR4Q6OLTXK7IHRXON6JBFMZOO/action/replication_record"}},"created_at":"2026-07-05T09:51:01.698762+00:00","updated_at":"2026-07-05T09:51:01.698762+00:00"}