{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:2O3KKWFJ4QEUUY3GRBITCN2MCN","short_pith_number":"pith:2O3KKWFJ","schema_version":"1.0","canonical_sha256":"d3b6a558a9e4094a6366885131374c137758b5d854f85e571b3d8973db6367e0","source":{"kind":"arxiv","id":"2504.04635","version":1},"attestation_state":"computed","paper":{"title":"Steering off Course: Reliability Challenges in Steering Language Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Dheeraj Rajagopal, Hannaneh Hajishirzi, Hari Sethuraman, Patrick Queiroz Da Silva, Sachin Kumar","submitted_at":"2025-04-06T22:19:46Z","abstract_excerpt":"Steering methods for language models (LMs) have gained traction as lightweight alternatives to fine-tuning, enabling targeted modifications to model activations. However, prior studies primarily report results on a few models, leaving critical gaps in understanding the robustness of these methods. In this work, we systematically examine three prominent steering methods -- DoLa, function vectors, and task vectors. In contrast to the original studies, which evaluated a handful of models, we test up to 36 models belonging to 14 families with sizes ranging from 1.5B to 70B parameters. Our experime"},"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":"2504.04635","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2025-04-06T22:19:46Z","cross_cats_sorted":[],"title_canon_sha256":"53d57c1d0844de1675edd510c9b8b9bc1eaca485635cbb813a72e320d8c21784","abstract_canon_sha256":"10410ca11363cde7a97835f40f5ef45904046011c6c194e87f3fd3685c589a1f"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:45:27.406445Z","signature_b64":"9ANLvChZNVtpMBZOsoR4OsPz7gvp/Bw1f8yQC1bMkoRf+IOJE2Ud5r53Z/YlWvUx7bugAsqJkzaPFLRg7daxDg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"d3b6a558a9e4094a6366885131374c137758b5d854f85e571b3d8973db6367e0","last_reissued_at":"2026-07-05T10:45:27.405947Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:45:27.405947Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Steering off Course: Reliability Challenges in Steering Language Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Dheeraj Rajagopal, Hannaneh Hajishirzi, Hari Sethuraman, Patrick Queiroz Da Silva, Sachin Kumar","submitted_at":"2025-04-06T22:19:46Z","abstract_excerpt":"Steering methods for language models (LMs) have gained traction as lightweight alternatives to fine-tuning, enabling targeted modifications to model activations. However, prior studies primarily report results on a few models, leaving critical gaps in understanding the robustness of these methods. In this work, we systematically examine three prominent steering methods -- DoLa, function vectors, and task vectors. In contrast to the original studies, which evaluated a handful of models, we test up to 36 models belonging to 14 families with sizes ranging from 1.5B to 70B parameters. Our experime"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2504.04635","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/2504.04635/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":"2504.04635","created_at":"2026-07-05T10:45:27.406005+00:00"},{"alias_kind":"arxiv_version","alias_value":"2504.04635v1","created_at":"2026-07-05T10:45:27.406005+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2504.04635","created_at":"2026-07-05T10:45:27.406005+00:00"},{"alias_kind":"pith_short_12","alias_value":"2O3KKWFJ4QEU","created_at":"2026-07-05T10:45:27.406005+00:00"},{"alias_kind":"pith_short_16","alias_value":"2O3KKWFJ4QEUUY3G","created_at":"2026-07-05T10:45:27.406005+00:00"},{"alias_kind":"pith_short_8","alias_value":"2O3KKWFJ","created_at":"2026-07-05T10:45:27.406005+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":5,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.12923","citing_title":"Order Is Not Control","ref_index":21,"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":12,"is_internal_anchor":false},{"citing_arxiv_id":"2601.14004","citing_title":"Locate, Steer, and Improve: A Practical Survey of Actionable Mechanistic Interpretability in Large Language Models","ref_index":55,"is_internal_anchor":false},{"citing_arxiv_id":"2605.05715","citing_title":"Decodable but Not Corrected by Fixed Residual-Stream Linear Steering: Evidence from Medical LLM Failure Regimes","ref_index":15,"is_internal_anchor":false},{"citing_arxiv_id":"2605.05115","citing_title":"Manifold Steering Reveals the Shared Geometry of Neural Network Representation and Behavior","ref_index":209,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/2O3KKWFJ4QEUUY3GRBITCN2MCN","json":"https://pith.science/pith/2O3KKWFJ4QEUUY3GRBITCN2MCN.json","graph_json":"https://pith.science/api/pith-number/2O3KKWFJ4QEUUY3GRBITCN2MCN/graph.json","events_json":"https://pith.science/api/pith-number/2O3KKWFJ4QEUUY3GRBITCN2MCN/events.json","paper":"https://pith.science/paper/2O3KKWFJ"},"agent_actions":{"view_html":"https://pith.science/pith/2O3KKWFJ4QEUUY3GRBITCN2MCN","download_json":"https://pith.science/pith/2O3KKWFJ4QEUUY3GRBITCN2MCN.json","view_paper":"https://pith.science/paper/2O3KKWFJ","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2504.04635&json=true","fetch_graph":"https://pith.science/api/pith-number/2O3KKWFJ4QEUUY3GRBITCN2MCN/graph.json","fetch_events":"https://pith.science/api/pith-number/2O3KKWFJ4QEUUY3GRBITCN2MCN/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/2O3KKWFJ4QEUUY3GRBITCN2MCN/action/timestamp_anchor","attest_storage":"https://pith.science/pith/2O3KKWFJ4QEUUY3GRBITCN2MCN/action/storage_attestation","attest_author":"https://pith.science/pith/2O3KKWFJ4QEUUY3GRBITCN2MCN/action/author_attestation","sign_citation":"https://pith.science/pith/2O3KKWFJ4QEUUY3GRBITCN2MCN/action/citation_signature","submit_replication":"https://pith.science/pith/2O3KKWFJ4QEUUY3GRBITCN2MCN/action/replication_record"}},"created_at":"2026-07-05T10:45:27.406005+00:00","updated_at":"2026-07-05T10:45:27.406005+00:00"}