{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:XWQTWV3BFU4YCHBKI7MEBF7MIU","short_pith_number":"pith:XWQTWV3B","schema_version":"1.0","canonical_sha256":"bda13b57612d39811c2a47d84097ec450153464a137cdcdaf2cb13383e9c51d9","source":{"kind":"arxiv","id":"2602.02712","version":2},"attestation_state":"computed","paper":{"title":"Towards Understanding Steering Strength","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CL"],"primary_cat":"cs.LG","authors_text":"Damien Garreau, Magamed Taimeskhanov, Samuel Vaiter","submitted_at":"2026-02-02T19:25:37Z","abstract_excerpt":"A popular approach to post-training control of large language models (LLMs) is the steering of intermediate latent representations. Namely, identify a well-chosen direction depending on the task at hand and perturbs representations along this direction at inference time. While many propositions exist to pick this direction, considerably less is understood about how to choose the magnitude of the move, whereas its importance is clear: too little and the intended behavior does not emerge, too much and the model's performance degrades beyond repair. In this work, we propose the first theoretical "},"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":"2602.02712","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-02-02T19:25:37Z","cross_cats_sorted":["cs.CL"],"title_canon_sha256":"c8182889581f6ed14e095c02cc58d722245aa057aafa81d62d7cf300165cae11","abstract_canon_sha256":"a67fdf26446667be63caeb75242e4a6c3fe318c081cea8a3ee5666165c2914c5"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-09T01:19:51.627982Z","signature_b64":"9U2XSCHBkR1f34uIoBmr1egtTz4L4g7teEV2L0Ud7UwqRN//uZUh3HCsrWCNKWdYVW1qG2z7SKgVdmmnbQrlDA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"bda13b57612d39811c2a47d84097ec450153464a137cdcdaf2cb13383e9c51d9","last_reissued_at":"2026-07-09T01:19:51.627491Z","signature_status":"signed_v1","first_computed_at":"2026-07-09T01:19:51.627491Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Towards Understanding Steering Strength","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CL"],"primary_cat":"cs.LG","authors_text":"Damien Garreau, Magamed Taimeskhanov, Samuel Vaiter","submitted_at":"2026-02-02T19:25:37Z","abstract_excerpt":"A popular approach to post-training control of large language models (LLMs) is the steering of intermediate latent representations. Namely, identify a well-chosen direction depending on the task at hand and perturbs representations along this direction at inference time. While many propositions exist to pick this direction, considerably less is understood about how to choose the magnitude of the move, whereas its importance is clear: too little and the intended behavior does not emerge, too much and the model's performance degrades beyond repair. In this work, we propose the first theoretical "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2602.02712","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/2602.02712/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":"2602.02712","created_at":"2026-07-09T01:19:51.627550+00:00"},{"alias_kind":"arxiv_version","alias_value":"2602.02712v2","created_at":"2026-07-09T01:19:51.627550+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2602.02712","created_at":"2026-07-09T01:19:51.627550+00:00"},{"alias_kind":"pith_short_12","alias_value":"XWQTWV3BFU4Y","created_at":"2026-07-09T01:19:51.627550+00:00"},{"alias_kind":"pith_short_16","alias_value":"XWQTWV3BFU4YCHBK","created_at":"2026-07-09T01:19:51.627550+00:00"},{"alias_kind":"pith_short_8","alias_value":"XWQTWV3B","created_at":"2026-07-09T01:19:51.627550+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2608.00023","citing_title":"Role Steering of Language Models for Social Simulations","ref_index":25,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/XWQTWV3BFU4YCHBKI7MEBF7MIU","json":"https://pith.science/pith/XWQTWV3BFU4YCHBKI7MEBF7MIU.json","graph_json":"https://pith.science/api/pith-number/XWQTWV3BFU4YCHBKI7MEBF7MIU/graph.json","events_json":"https://pith.science/api/pith-number/XWQTWV3BFU4YCHBKI7MEBF7MIU/events.json","paper":"https://pith.science/paper/XWQTWV3B"},"agent_actions":{"view_html":"https://pith.science/pith/XWQTWV3BFU4YCHBKI7MEBF7MIU","download_json":"https://pith.science/pith/XWQTWV3BFU4YCHBKI7MEBF7MIU.json","view_paper":"https://pith.science/paper/XWQTWV3B","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2602.02712&json=true","fetch_graph":"https://pith.science/api/pith-number/XWQTWV3BFU4YCHBKI7MEBF7MIU/graph.json","fetch_events":"https://pith.science/api/pith-number/XWQTWV3BFU4YCHBKI7MEBF7MIU/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/XWQTWV3BFU4YCHBKI7MEBF7MIU/action/timestamp_anchor","attest_storage":"https://pith.science/pith/XWQTWV3BFU4YCHBKI7MEBF7MIU/action/storage_attestation","attest_author":"https://pith.science/pith/XWQTWV3BFU4YCHBKI7MEBF7MIU/action/author_attestation","sign_citation":"https://pith.science/pith/XWQTWV3BFU4YCHBKI7MEBF7MIU/action/citation_signature","submit_replication":"https://pith.science/pith/XWQTWV3BFU4YCHBKI7MEBF7MIU/action/replication_record"}},"created_at":"2026-07-09T01:19:51.627550+00:00","updated_at":"2026-07-09T01:19:51.627550+00:00"}