{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:LITW2MYPHGQUGQ57B75HVHCENB","short_pith_number":"pith:LITW2MYP","schema_version":"1.0","canonical_sha256":"5a276d330f39a14343bf0ffa7a9c44684ce655f59f0b29f55eaaa669b569eaf7","source":{"kind":"arxiv","id":"2502.13417","version":3},"attestation_state":"computed","paper":{"title":"RLTHF: Targeted Human Feedback for LLM Alignment","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CL","authors_text":"Bibek Aryal, Eduardo Rodrigues, Emre K{\\i}c{\\i}man, Jessica Wolk, Leonardo Nunes, Maria Angels de Luis Balaguer, Rafael Padilha, Ranveer Chandra, Roberto Estevao, Shobana Balakrishnan, Songwu Lu, Srinagesh Sharma, Tusher Chakraborty, Yifei Xu","submitted_at":"2025-02-19T04:25:11Z","abstract_excerpt":"Fine-tuning large language models (LLMs) to align with user preferences is challenging due to the high cost of quality human annotations in Reinforcement Learning from Human Feedback (RLHF) and the generalizability limitations of AI Feedback. To address these challenges, we propose RLTHF, a human-AI hybrid framework that combines LLM-based initial alignment with selective human annotations to achieve full-human annotation alignment with minimal effort. RLTHF identifies hard-to-annotate samples mislabeled by LLMs using a reward model's reward distribution and iteratively enhances alignment by i"},"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":"2502.13417","kind":"arxiv","version":3},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2025-02-19T04:25:11Z","cross_cats_sorted":["cs.AI","cs.LG"],"title_canon_sha256":"cba8ef32d35e6ede1e88149b3c4fb58c32115b911c40d0cd12ccaf2a6592b84c","abstract_canon_sha256":"861659a2ec2a0938c56f76cf460dd7100f87d638eb288318689f53235a27799f"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:49:50.727707Z","signature_b64":"rQSsk/Bm8bHVrtPS7xdSeuCTblnyKRQA26IXizjzn31YFF4Ag/uB54ExHuA2xo2yRcR9TOtKnV4iwG0hoNpIDA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"5a276d330f39a14343bf0ffa7a9c44684ce655f59f0b29f55eaaa669b569eaf7","last_reissued_at":"2026-07-05T11:49:50.727233Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:49:50.727233Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"RLTHF: Targeted Human Feedback for LLM Alignment","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CL","authors_text":"Bibek Aryal, Eduardo Rodrigues, Emre K{\\i}c{\\i}man, Jessica Wolk, Leonardo Nunes, Maria Angels de Luis Balaguer, Rafael Padilha, Ranveer Chandra, Roberto Estevao, Shobana Balakrishnan, Songwu Lu, Srinagesh Sharma, Tusher Chakraborty, Yifei Xu","submitted_at":"2025-02-19T04:25:11Z","abstract_excerpt":"Fine-tuning large language models (LLMs) to align with user preferences is challenging due to the high cost of quality human annotations in Reinforcement Learning from Human Feedback (RLHF) and the generalizability limitations of AI Feedback. To address these challenges, we propose RLTHF, a human-AI hybrid framework that combines LLM-based initial alignment with selective human annotations to achieve full-human annotation alignment with minimal effort. RLTHF identifies hard-to-annotate samples mislabeled by LLMs using a reward model's reward distribution and iteratively enhances alignment by i"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2502.13417","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/2502.13417/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":"2502.13417","created_at":"2026-07-05T11:49:50.727293+00:00"},{"alias_kind":"arxiv_version","alias_value":"2502.13417v3","created_at":"2026-07-05T11:49:50.727293+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2502.13417","created_at":"2026-07-05T11:49:50.727293+00:00"},{"alias_kind":"pith_short_12","alias_value":"LITW2MYPHGQU","created_at":"2026-07-05T11:49:50.727293+00:00"},{"alias_kind":"pith_short_16","alias_value":"LITW2MYPHGQUGQ57","created_at":"2026-07-05T11:49:50.727293+00:00"},{"alias_kind":"pith_short_8","alias_value":"LITW2MYP","created_at":"2026-07-05T11:49:50.727293+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":0,"internal_anchor_count":0,"sample":[]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/LITW2MYPHGQUGQ57B75HVHCENB","json":"https://pith.science/pith/LITW2MYPHGQUGQ57B75HVHCENB.json","graph_json":"https://pith.science/api/pith-number/LITW2MYPHGQUGQ57B75HVHCENB/graph.json","events_json":"https://pith.science/api/pith-number/LITW2MYPHGQUGQ57B75HVHCENB/events.json","paper":"https://pith.science/paper/LITW2MYP"},"agent_actions":{"view_html":"https://pith.science/pith/LITW2MYPHGQUGQ57B75HVHCENB","download_json":"https://pith.science/pith/LITW2MYPHGQUGQ57B75HVHCENB.json","view_paper":"https://pith.science/paper/LITW2MYP","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2502.13417&json=true","fetch_graph":"https://pith.science/api/pith-number/LITW2MYPHGQUGQ57B75HVHCENB/graph.json","fetch_events":"https://pith.science/api/pith-number/LITW2MYPHGQUGQ57B75HVHCENB/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/LITW2MYPHGQUGQ57B75HVHCENB/action/timestamp_anchor","attest_storage":"https://pith.science/pith/LITW2MYPHGQUGQ57B75HVHCENB/action/storage_attestation","attest_author":"https://pith.science/pith/LITW2MYPHGQUGQ57B75HVHCENB/action/author_attestation","sign_citation":"https://pith.science/pith/LITW2MYPHGQUGQ57B75HVHCENB/action/citation_signature","submit_replication":"https://pith.science/pith/LITW2MYPHGQUGQ57B75HVHCENB/action/replication_record"}},"created_at":"2026-07-05T11:49:50.727293+00:00","updated_at":"2026-07-05T11:49:50.727293+00:00"}