{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2022:4LAF4NMRUTBVZDCVZQTZCBGNYW","short_pith_number":"pith:4LAF4NMR","schema_version":"1.0","canonical_sha256":"e2c05e3591a4c35c8c55cc279104cdc59d1a5c8d8a8e26a749a566a526f9cf2f","source":{"kind":"arxiv","id":"2205.05124","version":1},"attestation_state":"computed","paper":{"title":"Extracting Latent Steering Vectors from Pretrained Language Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CL","authors_text":"Matthew E. Peters, Nishant Subramani, Nivedita Suresh","submitted_at":"2022-05-10T19:04:37Z","abstract_excerpt":"Prior work on controllable text generation has focused on learning how to control language models through trainable decoding, smart-prompt design, or fine-tuning based on a desired objective. We hypothesize that the information needed to steer the model to generate a target sentence is already encoded within the model. Accordingly, we explore a different approach altogether: extracting latent vectors directly from pretrained language model decoders without fine-tuning. Experiments show that there exist steering vectors, which, when added to the hidden states of the language model, generate a t"},"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":"2205.05124","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2022-05-10T19:04:37Z","cross_cats_sorted":["cs.AI","cs.LG"],"title_canon_sha256":"7f5039c7c9e11a95acf22817bd7f95df51e102012a87559a01c01f167fe21ae4","abstract_canon_sha256":"bd334c8cc7216733f7b740247a5acb688dabf70e4520a5f8358946f6e9689298"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T04:22:22.561144Z","signature_b64":"byEU7jir6tIl1jRT+H5T2/44MDB7A/pg26v4fwM++G8e1XHH8Q2oFZuXoHX9Xx/LJir5YmAwIEM7opkWfzMHAA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"e2c05e3591a4c35c8c55cc279104cdc59d1a5c8d8a8e26a749a566a526f9cf2f","last_reissued_at":"2026-07-05T04:22:22.560733Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T04:22:22.560733Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Extracting Latent Steering Vectors from Pretrained Language Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CL","authors_text":"Matthew E. Peters, Nishant Subramani, Nivedita Suresh","submitted_at":"2022-05-10T19:04:37Z","abstract_excerpt":"Prior work on controllable text generation has focused on learning how to control language models through trainable decoding, smart-prompt design, or fine-tuning based on a desired objective. We hypothesize that the information needed to steer the model to generate a target sentence is already encoded within the model. Accordingly, we explore a different approach altogether: extracting latent vectors directly from pretrained language model decoders without fine-tuning. Experiments show that there exist steering vectors, which, when added to the hidden states of the language model, generate a t"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2205.05124","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/2205.05124/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":"2205.05124","created_at":"2026-07-05T04:22:22.560794+00:00"},{"alias_kind":"arxiv_version","alias_value":"2205.05124v1","created_at":"2026-07-05T04:22:22.560794+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2205.05124","created_at":"2026-07-05T04:22:22.560794+00:00"},{"alias_kind":"pith_short_12","alias_value":"4LAF4NMRUTBV","created_at":"2026-07-05T04:22:22.560794+00:00"},{"alias_kind":"pith_short_16","alias_value":"4LAF4NMRUTBVZDCV","created_at":"2026-07-05T04:22:22.560794+00:00"},{"alias_kind":"pith_short_8","alias_value":"4LAF4NMR","created_at":"2026-07-05T04:22:22.560794+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":9,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.11599","citing_title":"When is Your LLM Steerable?","ref_index":17,"is_internal_anchor":false},{"citing_arxiv_id":"2605.23040","citing_title":"Steered Generation via Gradient-Based Optimization on Sparse Query Features","ref_index":40,"is_internal_anchor":false},{"citing_arxiv_id":"2605.09773","citing_title":"Exploitation Without Deception: Dark Triad Feature Steering Reveals Separable Antisocial Circuits in Language Models","ref_index":10,"is_internal_anchor":false},{"citing_arxiv_id":"2312.06681","citing_title":"Steering Llama 2 via Contrastive Activation Addition","ref_index":21,"is_internal_anchor":false},{"citing_arxiv_id":"2605.05687","citing_title":"DataDignity: Training Data Attribution for Large Language Models","ref_index":28,"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":122,"is_internal_anchor":false},{"citing_arxiv_id":"2605.01148","citing_title":"Arithmetic in the Wild: Llama uses Base-10 Addition to Reason About Cyclic Concepts","ref_index":127,"is_internal_anchor":false},{"citing_arxiv_id":"2604.19678","citing_title":"Exploring Language-Agnosticity in Function Vectors: A Case Study in Machine Translation","ref_index":22,"is_internal_anchor":false},{"citing_arxiv_id":"2605.03160","citing_title":"Steering grids for sparse-autoencoder features: when a top-context label names an activation regime rather than a causal axis","ref_index":15,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/4LAF4NMRUTBVZDCVZQTZCBGNYW","json":"https://pith.science/pith/4LAF4NMRUTBVZDCVZQTZCBGNYW.json","graph_json":"https://pith.science/api/pith-number/4LAF4NMRUTBVZDCVZQTZCBGNYW/graph.json","events_json":"https://pith.science/api/pith-number/4LAF4NMRUTBVZDCVZQTZCBGNYW/events.json","paper":"https://pith.science/paper/4LAF4NMR"},"agent_actions":{"view_html":"https://pith.science/pith/4LAF4NMRUTBVZDCVZQTZCBGNYW","download_json":"https://pith.science/pith/4LAF4NMRUTBVZDCVZQTZCBGNYW.json","view_paper":"https://pith.science/paper/4LAF4NMR","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2205.05124&json=true","fetch_graph":"https://pith.science/api/pith-number/4LAF4NMRUTBVZDCVZQTZCBGNYW/graph.json","fetch_events":"https://pith.science/api/pith-number/4LAF4NMRUTBVZDCVZQTZCBGNYW/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/4LAF4NMRUTBVZDCVZQTZCBGNYW/action/timestamp_anchor","attest_storage":"https://pith.science/pith/4LAF4NMRUTBVZDCVZQTZCBGNYW/action/storage_attestation","attest_author":"https://pith.science/pith/4LAF4NMRUTBVZDCVZQTZCBGNYW/action/author_attestation","sign_citation":"https://pith.science/pith/4LAF4NMRUTBVZDCVZQTZCBGNYW/action/citation_signature","submit_replication":"https://pith.science/pith/4LAF4NMRUTBVZDCVZQTZCBGNYW/action/replication_record"}},"created_at":"2026-07-05T04:22:22.560794+00:00","updated_at":"2026-07-05T04:22:22.560794+00:00"}