{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:KGFX5ELAWOIBHEOQWRHBX5CTJR","short_pith_number":"pith:KGFX5ELA","schema_version":"1.0","canonical_sha256":"518b7e9160b3901391d0b44e1bf4534c7de77dc7c2b2160c698df47dc0fadc3a","source":{"kind":"arxiv","id":"2410.12877","version":2},"attestation_state":"computed","paper":{"title":"Improving Instruction-Following in Language Models through Activation Steering","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CL","authors_text":"Alessandro Stolfo, Besmira Nushi, Eric Horvitz, Safoora Yousefi, Vidhisha Balachandran","submitted_at":"2024-10-15T08:38:20Z","abstract_excerpt":"The ability to follow instructions is crucial for numerous real-world applications of language models. In pursuit of deeper insights and more powerful capabilities, we derive instruction-specific vector representations from language models and use them to steer models accordingly. These vectors are computed as the difference in activations between inputs with and without instructions, enabling a modular approach to activation steering. We demonstrate how this method can enhance model adherence to constraints such as output format, length, and word inclusion, providing inference-time control ov"},"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":"2410.12877","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2024-10-15T08:38:20Z","cross_cats_sorted":["cs.AI","cs.LG"],"title_canon_sha256":"04ef6d3d6ca68e106ff395e50a41e3aa46290bbbe6bd4e1e94c9061ace623c95","abstract_canon_sha256":"2f2b0a1799b5fc97520af4e803b75c478dba7d2570e900fae32279551a9bf791"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:48:20.457840Z","signature_b64":"cmar3KiaRWhfB8lD+g8fKSc/PT8tBMn7QXFRyFL5sgGtUgR4MwdUj9bsp23wDSUiExx+nT6SpGYlAstUZ8xjAA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"518b7e9160b3901391d0b44e1bf4534c7de77dc7c2b2160c698df47dc0fadc3a","last_reissued_at":"2026-07-05T10:48:20.457294Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:48:20.457294Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Improving Instruction-Following in Language Models through Activation Steering","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CL","authors_text":"Alessandro Stolfo, Besmira Nushi, Eric Horvitz, Safoora Yousefi, Vidhisha Balachandran","submitted_at":"2024-10-15T08:38:20Z","abstract_excerpt":"The ability to follow instructions is crucial for numerous real-world applications of language models. In pursuit of deeper insights and more powerful capabilities, we derive instruction-specific vector representations from language models and use them to steer models accordingly. These vectors are computed as the difference in activations between inputs with and without instructions, enabling a modular approach to activation steering. We demonstrate how this method can enhance model adherence to constraints such as output format, length, and word inclusion, providing inference-time control ov"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2410.12877","kind":"arxiv","version":2},"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/2410.12877/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":"2410.12877","created_at":"2026-07-05T10:48:20.457362+00:00"},{"alias_kind":"arxiv_version","alias_value":"2410.12877v2","created_at":"2026-07-05T10:48:20.457362+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2410.12877","created_at":"2026-07-05T10:48:20.457362+00:00"},{"alias_kind":"pith_short_12","alias_value":"KGFX5ELAWOIB","created_at":"2026-07-05T10:48:20.457362+00:00"},{"alias_kind":"pith_short_16","alias_value":"KGFX5ELAWOIBHEOQ","created_at":"2026-07-05T10:48:20.457362+00:00"},{"alias_kind":"pith_short_8","alias_value":"KGFX5ELA","created_at":"2026-07-05T10:48:20.457362+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":18,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2605.23932","citing_title":"When Correct Beliefs Collapse: Epistemic Resilience of LLMs under Clinical Pressure","ref_index":4,"is_internal_anchor":false},{"citing_arxiv_id":"2606.22953","citing_title":"Plans Don't Persist: Why Context Management Is Load Bearing for LLM Agents","ref_index":12,"is_internal_anchor":false},{"citing_arxiv_id":"2606.21121","citing_title":"Answer Engineering: Local Trajectory Editing for Protocol-Constrained Decision Making in Large Language Models","ref_index":76,"is_internal_anchor":false},{"citing_arxiv_id":"2606.07647","citing_title":"Steer Where It Matters: Token-Level Visual-Sensitivity Steering for LVLMs Hallucination Mitigation","ref_index":14,"is_internal_anchor":false},{"citing_arxiv_id":"2605.28664","citing_title":"Activation Steering for Synthetic Data Generation: The Role of Diversity in Downstream Safety Detection","ref_index":47,"is_internal_anchor":false},{"citing_arxiv_id":"2605.23040","citing_title":"Steered Generation via Gradient-Based Optimization on Sparse Query Features","ref_index":39,"is_internal_anchor":false},{"citing_arxiv_id":"2603.22161","citing_title":"Causal Evidence that Language Models use Confidence to Drive Behavior","ref_index":16,"is_internal_anchor":false},{"citing_arxiv_id":"2605.20262","citing_title":"Residual Paving: Diagnosing the Routing Bottleneck in Selective Refusal Editing","ref_index":14,"is_internal_anchor":false},{"citing_arxiv_id":"2506.01770","citing_title":"ReGA: Model-Based Safeguard for LLMs via Representation-Guided Abstraction","ref_index":64,"is_internal_anchor":false},{"citing_arxiv_id":"2507.02833","citing_title":"Generalizing Verifiable Instruction Following","ref_index":22,"is_internal_anchor":false},{"citing_arxiv_id":"2602.02280","citing_title":"RACC: Representation-Aware Coverage Criteria for LLM Safety Testing","ref_index":44,"is_internal_anchor":false},{"citing_arxiv_id":"2507.03724","citing_title":"MemOS: A Memory OS for AI System","ref_index":87,"is_internal_anchor":false},{"citing_arxiv_id":"2605.10664","citing_title":"Prompt-Activation Duality: Improving Activation Steering via Attention-Level Interventions","ref_index":13,"is_internal_anchor":false},{"citing_arxiv_id":"2605.11093","citing_title":"Enabling Performant and Flexible Model-Internal Observability for LLM Inference","ref_index":37,"is_internal_anchor":false},{"citing_arxiv_id":"2605.10664","citing_title":"Prompt-Activation Duality: Improving Activation Steering via Attention-Level Interventions","ref_index":13,"is_internal_anchor":false},{"citing_arxiv_id":"2604.22271","citing_title":"How LLMs Detect and Correct Their Own Errors: The Role of Internal Confidence Signals","ref_index":19,"is_internal_anchor":false},{"citing_arxiv_id":"2605.07284","citing_title":"Instruction Tuning Changes How Upstream State Conditions Late Readout: A Cross-Patching Diagnostic","ref_index":23,"is_internal_anchor":false},{"citing_arxiv_id":"2605.01846","citing_title":"Do Large Language Models Plan Answer Positions? Position Bias in Multiple-Choice Question Generation","ref_index":9,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/KGFX5ELAWOIBHEOQWRHBX5CTJR","json":"https://pith.science/pith/KGFX5ELAWOIBHEOQWRHBX5CTJR.json","graph_json":"https://pith.science/api/pith-number/KGFX5ELAWOIBHEOQWRHBX5CTJR/graph.json","events_json":"https://pith.science/api/pith-number/KGFX5ELAWOIBHEOQWRHBX5CTJR/events.json","paper":"https://pith.science/paper/KGFX5ELA"},"agent_actions":{"view_html":"https://pith.science/pith/KGFX5ELAWOIBHEOQWRHBX5CTJR","download_json":"https://pith.science/pith/KGFX5ELAWOIBHEOQWRHBX5CTJR.json","view_paper":"https://pith.science/paper/KGFX5ELA","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2410.12877&json=true","fetch_graph":"https://pith.science/api/pith-number/KGFX5ELAWOIBHEOQWRHBX5CTJR/graph.json","fetch_events":"https://pith.science/api/pith-number/KGFX5ELAWOIBHEOQWRHBX5CTJR/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/KGFX5ELAWOIBHEOQWRHBX5CTJR/action/timestamp_anchor","attest_storage":"https://pith.science/pith/KGFX5ELAWOIBHEOQWRHBX5CTJR/action/storage_attestation","attest_author":"https://pith.science/pith/KGFX5ELAWOIBHEOQWRHBX5CTJR/action/author_attestation","sign_citation":"https://pith.science/pith/KGFX5ELAWOIBHEOQWRHBX5CTJR/action/citation_signature","submit_replication":"https://pith.science/pith/KGFX5ELAWOIBHEOQWRHBX5CTJR/action/replication_record"}},"created_at":"2026-07-05T10:48:20.457362+00:00","updated_at":"2026-07-05T10:48:20.457362+00:00"}