{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:JLGVJULLC2ETORFGOF235RTZY6","short_pith_number":"pith:JLGVJULL","schema_version":"1.0","canonical_sha256":"4acd54d16b16893744a67175bec679c7acb0f660409283aa6349b2e0faebbfb8","source":{"kind":"arxiv","id":"2411.07213","version":1},"attestation_state":"computed","paper":{"title":"Comparing Bottom-Up and Top-Down Steering Approaches on In-Context Learning Tasks","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"David Krueger, Dmitrii Krasheninnikov, Joe Kwon, Madeline Brumley, Usman Anwar","submitted_at":"2024-11-11T18:36:17Z","abstract_excerpt":"A key objective of interpretability research on large language models (LLMs) is to develop methods for robustly steering models toward desired behaviors. To this end, two distinct approaches to interpretability -- ``bottom-up\" and ``top-down\" -- have been presented, but there has been little quantitative comparison between them. We present a case study comparing the effectiveness of representative vector steering methods from each branch: function vectors (FV; arXiv:2310.15213), as a bottom-up method, and in-context vectors (ICV; arXiv:2311.06668) as a top-down method. While both aim to captur"},"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":"2411.07213","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2024-11-11T18:36:17Z","cross_cats_sorted":[],"title_canon_sha256":"d2e95a89d172ca498b063b7b637b398cb3845f241f0f30836c2afc10134221a3","abstract_canon_sha256":"a89fc50eb5dd90a0a184c4dfc4eb442c6c374c7b2aa357df703640dac588409e"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T09:33:58.853643Z","signature_b64":"Q5JZftgXhjgKoPc1fyxGsVTw1YXbkv1JG9kmwKjy3aC0bRAaUE+jgTmN/8D28+eOp4NFNiMlHI3rAbphlvagCQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"4acd54d16b16893744a67175bec679c7acb0f660409283aa6349b2e0faebbfb8","last_reissued_at":"2026-07-05T09:33:58.853231Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T09:33:58.853231Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Comparing Bottom-Up and Top-Down Steering Approaches on In-Context Learning Tasks","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"David Krueger, Dmitrii Krasheninnikov, Joe Kwon, Madeline Brumley, Usman Anwar","submitted_at":"2024-11-11T18:36:17Z","abstract_excerpt":"A key objective of interpretability research on large language models (LLMs) is to develop methods for robustly steering models toward desired behaviors. To this end, two distinct approaches to interpretability -- ``bottom-up\" and ``top-down\" -- have been presented, but there has been little quantitative comparison between them. We present a case study comparing the effectiveness of representative vector steering methods from each branch: function vectors (FV; arXiv:2310.15213), as a bottom-up method, and in-context vectors (ICV; arXiv:2311.06668) as a top-down method. While both aim to captur"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2411.07213","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/2411.07213/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":"2411.07213","created_at":"2026-07-05T09:33:58.853290+00:00"},{"alias_kind":"arxiv_version","alias_value":"2411.07213v1","created_at":"2026-07-05T09:33:58.853290+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2411.07213","created_at":"2026-07-05T09:33:58.853290+00:00"},{"alias_kind":"pith_short_12","alias_value":"JLGVJULLC2ET","created_at":"2026-07-05T09:33:58.853290+00:00"},{"alias_kind":"pith_short_16","alias_value":"JLGVJULLC2ETORFG","created_at":"2026-07-05T09:33:58.853290+00:00"},{"alias_kind":"pith_short_8","alias_value":"JLGVJULL","created_at":"2026-07-05T09:33:58.853290+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.26155","citing_title":"Detecting and Controlling Sycophancy with Cascading Linear Features","ref_index":3,"is_internal_anchor":false},{"citing_arxiv_id":"2605.01844","citing_title":"The Cylindrical Representation Hypothesis for Language Model Steering","ref_index":3,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/JLGVJULLC2ETORFGOF235RTZY6","json":"https://pith.science/pith/JLGVJULLC2ETORFGOF235RTZY6.json","graph_json":"https://pith.science/api/pith-number/JLGVJULLC2ETORFGOF235RTZY6/graph.json","events_json":"https://pith.science/api/pith-number/JLGVJULLC2ETORFGOF235RTZY6/events.json","paper":"https://pith.science/paper/JLGVJULL"},"agent_actions":{"view_html":"https://pith.science/pith/JLGVJULLC2ETORFGOF235RTZY6","download_json":"https://pith.science/pith/JLGVJULLC2ETORFGOF235RTZY6.json","view_paper":"https://pith.science/paper/JLGVJULL","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2411.07213&json=true","fetch_graph":"https://pith.science/api/pith-number/JLGVJULLC2ETORFGOF235RTZY6/graph.json","fetch_events":"https://pith.science/api/pith-number/JLGVJULLC2ETORFGOF235RTZY6/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/JLGVJULLC2ETORFGOF235RTZY6/action/timestamp_anchor","attest_storage":"https://pith.science/pith/JLGVJULLC2ETORFGOF235RTZY6/action/storage_attestation","attest_author":"https://pith.science/pith/JLGVJULLC2ETORFGOF235RTZY6/action/author_attestation","sign_citation":"https://pith.science/pith/JLGVJULLC2ETORFGOF235RTZY6/action/citation_signature","submit_replication":"https://pith.science/pith/JLGVJULLC2ETORFGOF235RTZY6/action/replication_record"}},"created_at":"2026-07-05T09:33:58.853290+00:00","updated_at":"2026-07-05T09:33:58.853290+00:00"}