{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:PPW67BLWRTSIMMASOHTGO65KXE","short_pith_number":"pith:PPW67BLW","schema_version":"1.0","canonical_sha256":"7bedef85768ce486301271e6677baab93d51f7ef41fc1de6beaf2419c5e676b3","source":{"kind":"arxiv","id":"2507.13236","version":1},"attestation_state":"computed","paper":{"title":"Enhancing Cross-task Transfer of Large Language Models via Activation Steering","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Junyi Li, Jun Zhou, Wayne Xin Zhao, Xiaoxue Cheng, Xinyu Tang, Zhihao Lv, Zhiqiang Zhang, Zujie Wen","submitted_at":"2025-07-17T15:47:22Z","abstract_excerpt":"Large language models (LLMs) have shown impressive abilities in leveraging pretrained knowledge through prompting, but they often struggle with unseen tasks, particularly in data-scarce scenarios. While cross-task in-context learning offers a direct solution for transferring knowledge across tasks, it still faces critical challenges in terms of robustness, scalability, and efficiency. In this paper, we investigate whether cross-task transfer can be achieved via latent space steering without parameter updates or input expansion. Through an analysis of activation patterns in the latent space of "},"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":"2507.13236","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2025-07-17T15:47:22Z","cross_cats_sorted":[],"title_canon_sha256":"d82804c494de5290e91d2d9ca3d37d7ac49004e6ff6c0a12b6944d7ee5b271a7","abstract_canon_sha256":"9bf3653765ec5c3c43e46b7e2ce86b98cedcfd3e1404d93cf0c4d3ffc687e599"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:38:56.779082Z","signature_b64":"oc2jTOZKdJ6q6n/Y6APEhdBMcvQQaQhKrz5KLcLtcDfek1NOWhLze8ycGFlIpgXSaM6Z4NFrdXgvSFtpBopiAA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"7bedef85768ce486301271e6677baab93d51f7ef41fc1de6beaf2419c5e676b3","last_reissued_at":"2026-07-05T11:38:56.778573Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:38:56.778573Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Enhancing Cross-task Transfer of Large Language Models via Activation Steering","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Junyi Li, Jun Zhou, Wayne Xin Zhao, Xiaoxue Cheng, Xinyu Tang, Zhihao Lv, Zhiqiang Zhang, Zujie Wen","submitted_at":"2025-07-17T15:47:22Z","abstract_excerpt":"Large language models (LLMs) have shown impressive abilities in leveraging pretrained knowledge through prompting, but they often struggle with unseen tasks, particularly in data-scarce scenarios. While cross-task in-context learning offers a direct solution for transferring knowledge across tasks, it still faces critical challenges in terms of robustness, scalability, and efficiency. In this paper, we investigate whether cross-task transfer can be achieved via latent space steering without parameter updates or input expansion. Through an analysis of activation patterns in the latent space of "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2507.13236","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/2507.13236/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":"2507.13236","created_at":"2026-07-05T11:38:56.778632+00:00"},{"alias_kind":"arxiv_version","alias_value":"2507.13236v1","created_at":"2026-07-05T11:38:56.778632+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2507.13236","created_at":"2026-07-05T11:38:56.778632+00:00"},{"alias_kind":"pith_short_12","alias_value":"PPW67BLWRTSI","created_at":"2026-07-05T11:38:56.778632+00:00"},{"alias_kind":"pith_short_16","alias_value":"PPW67BLWRTSIMMAS","created_at":"2026-07-05T11:38:56.778632+00:00"},{"alias_kind":"pith_short_8","alias_value":"PPW67BLW","created_at":"2026-07-05T11:38:56.778632+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"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":5,"is_internal_anchor":false},{"citing_arxiv_id":"2605.20189","citing_title":"SOLAR: A Self-Optimizing Open-Ended Autonomous Agent for Lifelong Learning and Continual Adaptation","ref_index":57,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/PPW67BLWRTSIMMASOHTGO65KXE","json":"https://pith.science/pith/PPW67BLWRTSIMMASOHTGO65KXE.json","graph_json":"https://pith.science/api/pith-number/PPW67BLWRTSIMMASOHTGO65KXE/graph.json","events_json":"https://pith.science/api/pith-number/PPW67BLWRTSIMMASOHTGO65KXE/events.json","paper":"https://pith.science/paper/PPW67BLW"},"agent_actions":{"view_html":"https://pith.science/pith/PPW67BLWRTSIMMASOHTGO65KXE","download_json":"https://pith.science/pith/PPW67BLWRTSIMMASOHTGO65KXE.json","view_paper":"https://pith.science/paper/PPW67BLW","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2507.13236&json=true","fetch_graph":"https://pith.science/api/pith-number/PPW67BLWRTSIMMASOHTGO65KXE/graph.json","fetch_events":"https://pith.science/api/pith-number/PPW67BLWRTSIMMASOHTGO65KXE/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/PPW67BLWRTSIMMASOHTGO65KXE/action/timestamp_anchor","attest_storage":"https://pith.science/pith/PPW67BLWRTSIMMASOHTGO65KXE/action/storage_attestation","attest_author":"https://pith.science/pith/PPW67BLWRTSIMMASOHTGO65KXE/action/author_attestation","sign_citation":"https://pith.science/pith/PPW67BLWRTSIMMASOHTGO65KXE/action/citation_signature","submit_replication":"https://pith.science/pith/PPW67BLWRTSIMMASOHTGO65KXE/action/replication_record"}},"created_at":"2026-07-05T11:38:56.778632+00:00","updated_at":"2026-07-05T11:38:56.778632+00:00"}