{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:UWFGYZI74HTDBCI3IN4QXK63DA","short_pith_number":"pith:UWFGYZI7","schema_version":"1.0","canonical_sha256":"a58a6c651fe1e630891b43790babdb1822561cb5bd8952a5380ec9624b40c385","source":{"kind":"arxiv","id":"2404.15206","version":3},"attestation_state":"computed","paper":{"title":"Does Instruction Tuning Make LLMs More Consistent?","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Anders S{\\o}gaard, Constanza Fierro, Jiaang Li","submitted_at":"2024-04-23T16:39:03Z","abstract_excerpt":"The purpose of instruction tuning is enabling zero-shot performance, but instruction tuning has also been shown to improve chain-of-thought reasoning and value alignment (Si et al., 2023). Here we consider the impact on $\\textit{consistency}$, i.e., the sensitivity of language models to small perturbations in the input. We compare 10 instruction-tuned LLaMA models to the original LLaMA-7b model and show that almost across-the-board they become more consistent, both in terms of their representations and their predictions in zero-shot and downstream tasks. We explain these improvements through m"},"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":"2404.15206","kind":"arxiv","version":3},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2024-04-23T16:39:03Z","cross_cats_sorted":[],"title_canon_sha256":"9451a3f13395d1ca5d5eaa25bb1641a56cb80f79f36fcbc5444b8b27c49fbd65","abstract_canon_sha256":"ca35881c455b64bf6fedaa97a6f7e254c18de1a7d82d2d90b8c40052ed174c44"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T09:15:09.974471Z","signature_b64":"yIbQm8e0XTAuUr2hl4I+8otW3vVAv0UEKCf2arus2VEXIXQsfqV28TTeADMCAkIgXq7xfGP99II5lARr4QnnCg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"a58a6c651fe1e630891b43790babdb1822561cb5bd8952a5380ec9624b40c385","last_reissued_at":"2026-07-05T09:15:09.973988Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T09:15:09.973988Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Does Instruction Tuning Make LLMs More Consistent?","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Anders S{\\o}gaard, Constanza Fierro, Jiaang Li","submitted_at":"2024-04-23T16:39:03Z","abstract_excerpt":"The purpose of instruction tuning is enabling zero-shot performance, but instruction tuning has also been shown to improve chain-of-thought reasoning and value alignment (Si et al., 2023). Here we consider the impact on $\\textit{consistency}$, i.e., the sensitivity of language models to small perturbations in the input. We compare 10 instruction-tuned LLaMA models to the original LLaMA-7b model and show that almost across-the-board they become more consistent, both in terms of their representations and their predictions in zero-shot and downstream tasks. We explain these improvements through m"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2404.15206","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/2404.15206/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":"2404.15206","created_at":"2026-07-05T09:15:09.974046+00:00"},{"alias_kind":"arxiv_version","alias_value":"2404.15206v3","created_at":"2026-07-05T09:15:09.974046+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2404.15206","created_at":"2026-07-05T09:15:09.974046+00:00"},{"alias_kind":"pith_short_12","alias_value":"UWFGYZI74HTD","created_at":"2026-07-05T09:15:09.974046+00:00"},{"alias_kind":"pith_short_16","alias_value":"UWFGYZI74HTDBCI3","created_at":"2026-07-05T09:15:09.974046+00:00"},{"alias_kind":"pith_short_8","alias_value":"UWFGYZI7","created_at":"2026-07-05T09:15:09.974046+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/UWFGYZI74HTDBCI3IN4QXK63DA","json":"https://pith.science/pith/UWFGYZI74HTDBCI3IN4QXK63DA.json","graph_json":"https://pith.science/api/pith-number/UWFGYZI74HTDBCI3IN4QXK63DA/graph.json","events_json":"https://pith.science/api/pith-number/UWFGYZI74HTDBCI3IN4QXK63DA/events.json","paper":"https://pith.science/paper/UWFGYZI7"},"agent_actions":{"view_html":"https://pith.science/pith/UWFGYZI74HTDBCI3IN4QXK63DA","download_json":"https://pith.science/pith/UWFGYZI74HTDBCI3IN4QXK63DA.json","view_paper":"https://pith.science/paper/UWFGYZI7","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2404.15206&json=true","fetch_graph":"https://pith.science/api/pith-number/UWFGYZI74HTDBCI3IN4QXK63DA/graph.json","fetch_events":"https://pith.science/api/pith-number/UWFGYZI74HTDBCI3IN4QXK63DA/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/UWFGYZI74HTDBCI3IN4QXK63DA/action/timestamp_anchor","attest_storage":"https://pith.science/pith/UWFGYZI74HTDBCI3IN4QXK63DA/action/storage_attestation","attest_author":"https://pith.science/pith/UWFGYZI74HTDBCI3IN4QXK63DA/action/author_attestation","sign_citation":"https://pith.science/pith/UWFGYZI74HTDBCI3IN4QXK63DA/action/citation_signature","submit_replication":"https://pith.science/pith/UWFGYZI74HTDBCI3IN4QXK63DA/action/replication_record"}},"created_at":"2026-07-05T09:15:09.974046+00:00","updated_at":"2026-07-05T09:15:09.974046+00:00"}