{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:MUPEZDW7U5BQONHPJGXVKQQKOU","short_pith_number":"pith:MUPEZDW7","schema_version":"1.0","canonical_sha256":"651e4c8edfa7430734ef49af55420a75355f10a7b25a50588718c0c544acab88","source":{"kind":"arxiv","id":"2402.06094","version":1},"attestation_state":"computed","paper":{"title":"Rethinking Data Selection for Supervised Fine-Tuning","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Ming Shen","submitted_at":"2024-02-08T23:02:04Z","abstract_excerpt":"Although supervised finetuning (SFT) has emerged as an essential technique to align large language models with humans, it is considered superficial, with style learning being its nature. At the same time, recent works indicate the importance of data selection for SFT, showing that finetuning with high-quality and diverse subsets of the original dataset leads to superior downstream performance. In this work, we rethink the intuition behind data selection for SFT. Considering SFT is superficial, we propose that essential demonstrations for SFT should focus on reflecting human-like interactions 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":"2402.06094","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2024-02-08T23:02:04Z","cross_cats_sorted":[],"title_canon_sha256":"8eb21b5f8c3d57c5da116b5d4890081cb673d784f01dfc12a60c0e1c76d1da39","abstract_canon_sha256":"a8ef31dd891d4bbcaa754eff2cc82ff4f4b783794b4f159a72c017771d1ff1ab"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T07:43:18.623005Z","signature_b64":"33jzhgNJBfgzPNX5wV2MJNvrHR/vLeJgyvoEK61pZ+imrsUgPZjgGS/nWOKNk1IQoVjBeGJivJnKAJjAiFLPCQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"651e4c8edfa7430734ef49af55420a75355f10a7b25a50588718c0c544acab88","last_reissued_at":"2026-07-05T07:43:18.622560Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T07:43:18.622560Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Rethinking Data Selection for Supervised Fine-Tuning","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Ming Shen","submitted_at":"2024-02-08T23:02:04Z","abstract_excerpt":"Although supervised finetuning (SFT) has emerged as an essential technique to align large language models with humans, it is considered superficial, with style learning being its nature. At the same time, recent works indicate the importance of data selection for SFT, showing that finetuning with high-quality and diverse subsets of the original dataset leads to superior downstream performance. In this work, we rethink the intuition behind data selection for SFT. Considering SFT is superficial, we propose that essential demonstrations for SFT should focus on reflecting human-like interactions i"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2402.06094","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/2402.06094/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":"2402.06094","created_at":"2026-07-05T07:43:18.622620+00:00"},{"alias_kind":"arxiv_version","alias_value":"2402.06094v1","created_at":"2026-07-05T07:43:18.622620+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2402.06094","created_at":"2026-07-05T07:43:18.622620+00:00"},{"alias_kind":"pith_short_12","alias_value":"MUPEZDW7U5BQ","created_at":"2026-07-05T07:43:18.622620+00:00"},{"alias_kind":"pith_short_16","alias_value":"MUPEZDW7U5BQONHP","created_at":"2026-07-05T07:43:18.622620+00:00"},{"alias_kind":"pith_short_8","alias_value":"MUPEZDW7","created_at":"2026-07-05T07:43:18.622620+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":4,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.20638","citing_title":"RIZZ: Routing Interactions to Near Zero-Interference Zones for Continual Adaptation of Black-Box Agents","ref_index":31,"is_internal_anchor":false},{"citing_arxiv_id":"2411.10440","citing_title":"LLaVA-CoT: Let Vision Language Models Reason Step-by-Step","ref_index":50,"is_internal_anchor":false},{"citing_arxiv_id":"2507.21046","citing_title":"A Survey of Self-Evolving Agents: What, When, How, and Where to Evolve on the Path to Artificial Super Intelligence","ref_index":123,"is_internal_anchor":false},{"citing_arxiv_id":"2605.07977","citing_title":"Self-Play Enhancement via Advantage-Weighted Refinement in Online Federated LLM Fine-Tuning with Real-Time Feedback","ref_index":34,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/MUPEZDW7U5BQONHPJGXVKQQKOU","json":"https://pith.science/pith/MUPEZDW7U5BQONHPJGXVKQQKOU.json","graph_json":"https://pith.science/api/pith-number/MUPEZDW7U5BQONHPJGXVKQQKOU/graph.json","events_json":"https://pith.science/api/pith-number/MUPEZDW7U5BQONHPJGXVKQQKOU/events.json","paper":"https://pith.science/paper/MUPEZDW7"},"agent_actions":{"view_html":"https://pith.science/pith/MUPEZDW7U5BQONHPJGXVKQQKOU","download_json":"https://pith.science/pith/MUPEZDW7U5BQONHPJGXVKQQKOU.json","view_paper":"https://pith.science/paper/MUPEZDW7","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2402.06094&json=true","fetch_graph":"https://pith.science/api/pith-number/MUPEZDW7U5BQONHPJGXVKQQKOU/graph.json","fetch_events":"https://pith.science/api/pith-number/MUPEZDW7U5BQONHPJGXVKQQKOU/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/MUPEZDW7U5BQONHPJGXVKQQKOU/action/timestamp_anchor","attest_storage":"https://pith.science/pith/MUPEZDW7U5BQONHPJGXVKQQKOU/action/storage_attestation","attest_author":"https://pith.science/pith/MUPEZDW7U5BQONHPJGXVKQQKOU/action/author_attestation","sign_citation":"https://pith.science/pith/MUPEZDW7U5BQONHPJGXVKQQKOU/action/citation_signature","submit_replication":"https://pith.science/pith/MUPEZDW7U5BQONHPJGXVKQQKOU/action/replication_record"}},"created_at":"2026-07-05T07:43:18.622620+00:00","updated_at":"2026-07-05T07:43:18.622620+00:00"}