{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:M3TXNTXKJTUEFAJT7ODQRUEB77","short_pith_number":"pith:M3TXNTXK","schema_version":"1.0","canonical_sha256":"66e776ceea4ce8428133fb8708d081ffd84f55241193306595c619791f933ad5","source":{"kind":"arxiv","id":"2412.07992","version":4},"attestation_state":"computed","paper":{"title":"Concept Bottleneck Large Language Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CL","authors_text":"Berk Ustun, Chung-En Sun, Tsui-Wei Weng, Tuomas Oikarinen","submitted_at":"2024-12-11T00:04:10Z","abstract_excerpt":"We introduce Concept Bottleneck Large Language Models (CB-LLMs), a novel framework for building inherently interpretable Large Language Models (LLMs). In contrast to traditional black-box LLMs that rely on limited post-hoc interpretations, CB-LLMs integrate intrinsic interpretability directly into the LLMs -- allowing accurate explanations with scalability and transparency. We build CB-LLMs for two essential NLP tasks: text classification and text generation. In text classification, CB-LLMs is competitive with, and at times outperforms, traditional black-box models while providing explicit and"},"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":"2412.07992","kind":"arxiv","version":4},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2024-12-11T00:04:10Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"b53db93bf999b5c7bc4a49cd886871d13f98f4b8dc62c50ab891b4d2d65b2385","abstract_canon_sha256":"c7a80e831f49548b0e6cac898780452a13df569e4f9b14a1968316f40996bbfc"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T12:06:03.859563Z","signature_b64":"YgEoC+734QccsTvZDuLJ1tik5f7CKLw7yRvDIy1xVbzuwhvv64OPDzNjuVrjPknmsrU5OmlqpzAm0Q3e7JstBQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"66e776ceea4ce8428133fb8708d081ffd84f55241193306595c619791f933ad5","last_reissued_at":"2026-07-05T12:06:03.859066Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T12:06:03.859066Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Concept Bottleneck Large Language Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CL","authors_text":"Berk Ustun, Chung-En Sun, Tsui-Wei Weng, Tuomas Oikarinen","submitted_at":"2024-12-11T00:04:10Z","abstract_excerpt":"We introduce Concept Bottleneck Large Language Models (CB-LLMs), a novel framework for building inherently interpretable Large Language Models (LLMs). In contrast to traditional black-box LLMs that rely on limited post-hoc interpretations, CB-LLMs integrate intrinsic interpretability directly into the LLMs -- allowing accurate explanations with scalability and transparency. We build CB-LLMs for two essential NLP tasks: text classification and text generation. In text classification, CB-LLMs is competitive with, and at times outperforms, traditional black-box models while providing explicit and"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2412.07992","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/2412.07992/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":"2412.07992","created_at":"2026-07-05T12:06:03.859125+00:00"},{"alias_kind":"arxiv_version","alias_value":"2412.07992v4","created_at":"2026-07-05T12:06:03.859125+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2412.07992","created_at":"2026-07-05T12:06:03.859125+00:00"},{"alias_kind":"pith_short_12","alias_value":"M3TXNTXKJTUE","created_at":"2026-07-05T12:06:03.859125+00:00"},{"alias_kind":"pith_short_16","alias_value":"M3TXNTXKJTUEFAJT","created_at":"2026-07-05T12:06:03.859125+00:00"},{"alias_kind":"pith_short_8","alias_value":"M3TXNTXK","created_at":"2026-07-05T12:06:03.859125+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":5,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2605.09252","citing_title":"LLM Agents Already Know When to Call Tools -- Even Without Reasoning","ref_index":13,"is_internal_anchor":false},{"citing_arxiv_id":"2605.21455","citing_title":"Mitigating Label Bias with Interpretable Rubric Embeddings","ref_index":15,"is_internal_anchor":false},{"citing_arxiv_id":"2605.15688","citing_title":"$\\alpha$-TCAV: A Unified Framework for Testing with Concept Activation Vectors","ref_index":130,"is_internal_anchor":false},{"citing_arxiv_id":"2605.09252","citing_title":"LLM Agents Already Know When to Call Tools -- Even Without Reasoning","ref_index":13,"is_internal_anchor":false},{"citing_arxiv_id":"2605.05329","citing_title":"Understanding Annotator Safety Policy with Interpretability","ref_index":41,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/M3TXNTXKJTUEFAJT7ODQRUEB77","json":"https://pith.science/pith/M3TXNTXKJTUEFAJT7ODQRUEB77.json","graph_json":"https://pith.science/api/pith-number/M3TXNTXKJTUEFAJT7ODQRUEB77/graph.json","events_json":"https://pith.science/api/pith-number/M3TXNTXKJTUEFAJT7ODQRUEB77/events.json","paper":"https://pith.science/paper/M3TXNTXK"},"agent_actions":{"view_html":"https://pith.science/pith/M3TXNTXKJTUEFAJT7ODQRUEB77","download_json":"https://pith.science/pith/M3TXNTXKJTUEFAJT7ODQRUEB77.json","view_paper":"https://pith.science/paper/M3TXNTXK","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2412.07992&json=true","fetch_graph":"https://pith.science/api/pith-number/M3TXNTXKJTUEFAJT7ODQRUEB77/graph.json","fetch_events":"https://pith.science/api/pith-number/M3TXNTXKJTUEFAJT7ODQRUEB77/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/M3TXNTXKJTUEFAJT7ODQRUEB77/action/timestamp_anchor","attest_storage":"https://pith.science/pith/M3TXNTXKJTUEFAJT7ODQRUEB77/action/storage_attestation","attest_author":"https://pith.science/pith/M3TXNTXKJTUEFAJT7ODQRUEB77/action/author_attestation","sign_citation":"https://pith.science/pith/M3TXNTXKJTUEFAJT7ODQRUEB77/action/citation_signature","submit_replication":"https://pith.science/pith/M3TXNTXKJTUEFAJT7ODQRUEB77/action/replication_record"}},"created_at":"2026-07-05T12:06:03.859125+00:00","updated_at":"2026-07-05T12:06:03.859125+00:00"}