{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:C6AT4FUX4LMFXQUMMMWCPWT2XG","short_pith_number":"pith:C6AT4FUX","schema_version":"1.0","canonical_sha256":"17813e1697e2d85bc28c632c27da7ab99b220c1ee6302c1610a625e69498baec","source":{"kind":"arxiv","id":"2502.11962","version":3},"attestation_state":"computed","paper":{"title":"Balancing Truthfulness and Informativeness with Uncertainty-Aware Instruction Fine-Tuning","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Bryan Hooi, Elliott Ash, Jiaheng Zhang, Jingwei Ni, Markus Leippold, Mrinmaya Sachan, See-kiong Ng, Tianyi Wu","submitted_at":"2025-02-17T16:10:30Z","abstract_excerpt":"Instruction fine-tuning (IFT) can increase the informativeness of large language models (LLMs), but may reduce their truthfulness. This trade-off arises because IFT steers LLMs to generate responses containing long-tail knowledge that was not well covered during pre-training. As a result, models become more informative but less accurate when generalizing to unseen tasks. In this paper, we empirically demonstrate how unfamiliar knowledge in IFT datasets can negatively affect the truthfulness of LLMs, and we introduce two new IFT paradigms, $UNIT_{cut}$ and $UNIT_{ref}$, to address this issue. $"},"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":"2502.11962","kind":"arxiv","version":3},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2025-02-17T16:10:30Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"193edd86cba598ee92b32fbcd9e8a82339ea09898fa13ad80cbd08ae8f583856","abstract_canon_sha256":"cc105dcb25befc3da974cc97497c469c55891fafb60a7ebadc3879838ec57d18"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:26:47.937098Z","signature_b64":"6WgUt9Kgpyqnn6vdEaIs1sIZKe6wbEIIxknSdqzSCfH3WkXzqBWZHd9gVXdimfcsfTV47tHMdEvFSf6lrNUeDA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"17813e1697e2d85bc28c632c27da7ab99b220c1ee6302c1610a625e69498baec","last_reissued_at":"2026-07-05T11:26:47.936623Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:26:47.936623Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Balancing Truthfulness and Informativeness with Uncertainty-Aware Instruction Fine-Tuning","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Bryan Hooi, Elliott Ash, Jiaheng Zhang, Jingwei Ni, Markus Leippold, Mrinmaya Sachan, See-kiong Ng, Tianyi Wu","submitted_at":"2025-02-17T16:10:30Z","abstract_excerpt":"Instruction fine-tuning (IFT) can increase the informativeness of large language models (LLMs), but may reduce their truthfulness. This trade-off arises because IFT steers LLMs to generate responses containing long-tail knowledge that was not well covered during pre-training. As a result, models become more informative but less accurate when generalizing to unseen tasks. In this paper, we empirically demonstrate how unfamiliar knowledge in IFT datasets can negatively affect the truthfulness of LLMs, and we introduce two new IFT paradigms, $UNIT_{cut}$ and $UNIT_{ref}$, to address this issue. $"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2502.11962","kind":"arxiv","version":3},"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/2502.11962/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":"2502.11962","created_at":"2026-07-05T11:26:47.936681+00:00"},{"alias_kind":"arxiv_version","alias_value":"2502.11962v3","created_at":"2026-07-05T11:26:47.936681+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2502.11962","created_at":"2026-07-05T11:26:47.936681+00:00"},{"alias_kind":"pith_short_12","alias_value":"C6AT4FUX4LMF","created_at":"2026-07-05T11:26:47.936681+00:00"},{"alias_kind":"pith_short_16","alias_value":"C6AT4FUX4LMFXQUM","created_at":"2026-07-05T11:26:47.936681+00:00"},{"alias_kind":"pith_short_8","alias_value":"C6AT4FUX","created_at":"2026-07-05T11:26:47.936681+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2505.23247","citing_title":"Accelerating RLHF Training with Reward Variance Increase","ref_index":33,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/C6AT4FUX4LMFXQUMMMWCPWT2XG","json":"https://pith.science/pith/C6AT4FUX4LMFXQUMMMWCPWT2XG.json","graph_json":"https://pith.science/api/pith-number/C6AT4FUX4LMFXQUMMMWCPWT2XG/graph.json","events_json":"https://pith.science/api/pith-number/C6AT4FUX4LMFXQUMMMWCPWT2XG/events.json","paper":"https://pith.science/paper/C6AT4FUX"},"agent_actions":{"view_html":"https://pith.science/pith/C6AT4FUX4LMFXQUMMMWCPWT2XG","download_json":"https://pith.science/pith/C6AT4FUX4LMFXQUMMMWCPWT2XG.json","view_paper":"https://pith.science/paper/C6AT4FUX","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2502.11962&json=true","fetch_graph":"https://pith.science/api/pith-number/C6AT4FUX4LMFXQUMMMWCPWT2XG/graph.json","fetch_events":"https://pith.science/api/pith-number/C6AT4FUX4LMFXQUMMMWCPWT2XG/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/C6AT4FUX4LMFXQUMMMWCPWT2XG/action/timestamp_anchor","attest_storage":"https://pith.science/pith/C6AT4FUX4LMFXQUMMMWCPWT2XG/action/storage_attestation","attest_author":"https://pith.science/pith/C6AT4FUX4LMFXQUMMMWCPWT2XG/action/author_attestation","sign_citation":"https://pith.science/pith/C6AT4FUX4LMFXQUMMMWCPWT2XG/action/citation_signature","submit_replication":"https://pith.science/pith/C6AT4FUX4LMFXQUMMMWCPWT2XG/action/replication_record"}},"created_at":"2026-07-05T11:26:47.936681+00:00","updated_at":"2026-07-05T11:26:47.936681+00:00"}