{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:VO2CHNY3KQ7MH5MTD3GMTIGO4S","short_pith_number":"pith:VO2CHNY3","schema_version":"1.0","canonical_sha256":"abb423b71b543ec3f5931eccc9a0cee4b1ef6661e0bf49db05b9cf14add85e98","source":{"kind":"arxiv","id":"2410.03717","version":1},"attestation_state":"computed","paper":{"title":"Revisiting the Superficial Alignment Hypothesis","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CL","authors_text":"Mohit Raghavendra, Sean Hendryx, Vaskar Nath","submitted_at":"2024-09-27T22:14:10Z","abstract_excerpt":"The Superficial Alignment Hypothesis posits that almost all of a language model's abilities and knowledge are learned during pre-training, while post-training is about giving a model the right style and format. We re-examine these claims by empirically studying the scaling behavior of post-training with increasing finetuning examples and evaluating them using objective task-specific standardized benchmarks. Through experiments with the Llama-3, Mistral, and Llama-2 model families of multiple sizes, we observe that, similar to the pre-training scaling laws, post-training task performance scales"},"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":"2410.03717","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2024-09-27T22:14:10Z","cross_cats_sorted":["cs.AI","cs.LG"],"title_canon_sha256":"1fc0080251ca375c9d75d31759175780c50e140cd49b27a025ff19e626b736c1","abstract_canon_sha256":"dc098be426e06dde24d3eb60669b3b3f1c9b6ae28cda09d3837a7ee511ebbd7b"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T09:16:23.805080Z","signature_b64":"P393HPsz8MMWORFezHiZ5M0rO0MBZc6FBuBbWpsUgNBFwx0g5qWVwsL6muQ7R5yOagxrbWj0hiBYqsWWrZhgAA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"abb423b71b543ec3f5931eccc9a0cee4b1ef6661e0bf49db05b9cf14add85e98","last_reissued_at":"2026-07-05T09:16:23.804522Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T09:16:23.804522Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Revisiting the Superficial Alignment Hypothesis","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CL","authors_text":"Mohit Raghavendra, Sean Hendryx, Vaskar Nath","submitted_at":"2024-09-27T22:14:10Z","abstract_excerpt":"The Superficial Alignment Hypothesis posits that almost all of a language model's abilities and knowledge are learned during pre-training, while post-training is about giving a model the right style and format. We re-examine these claims by empirically studying the scaling behavior of post-training with increasing finetuning examples and evaluating them using objective task-specific standardized benchmarks. Through experiments with the Llama-3, Mistral, and Llama-2 model families of multiple sizes, we observe that, similar to the pre-training scaling laws, post-training task performance scales"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2410.03717","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/2410.03717/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":"2410.03717","created_at":"2026-07-05T09:16:23.804598+00:00"},{"alias_kind":"arxiv_version","alias_value":"2410.03717v1","created_at":"2026-07-05T09:16:23.804598+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2410.03717","created_at":"2026-07-05T09:16:23.804598+00:00"},{"alias_kind":"pith_short_12","alias_value":"VO2CHNY3KQ7M","created_at":"2026-07-05T09:16:23.804598+00:00"},{"alias_kind":"pith_short_16","alias_value":"VO2CHNY3KQ7MH5MT","created_at":"2026-07-05T09:16:23.804598+00:00"},{"alias_kind":"pith_short_8","alias_value":"VO2CHNY3","created_at":"2026-07-05T09:16:23.804598+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":4,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.12117","citing_title":"Soft-Prompt Tuning for Fair and Efficient LLM Benchmark Evaluation","ref_index":40,"is_internal_anchor":false},{"citing_arxiv_id":"2602.07340","citing_title":"Revisiting Robustness for LLM Safety Alignment via Selective Geometry Control","ref_index":21,"is_internal_anchor":false},{"citing_arxiv_id":"2605.20342","citing_title":"ParaVT: Taming the Tool Prior Paradox for Parallel Tool Use in Agentic Video Reinforcement Learning","ref_index":26,"is_internal_anchor":false},{"citing_arxiv_id":"2605.20342","citing_title":"ParaVT: Taming the Tool Prior Paradox for Parallel Tool Use in Agentic Video Reinforcement Learning","ref_index":26,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/VO2CHNY3KQ7MH5MTD3GMTIGO4S","json":"https://pith.science/pith/VO2CHNY3KQ7MH5MTD3GMTIGO4S.json","graph_json":"https://pith.science/api/pith-number/VO2CHNY3KQ7MH5MTD3GMTIGO4S/graph.json","events_json":"https://pith.science/api/pith-number/VO2CHNY3KQ7MH5MTD3GMTIGO4S/events.json","paper":"https://pith.science/paper/VO2CHNY3"},"agent_actions":{"view_html":"https://pith.science/pith/VO2CHNY3KQ7MH5MTD3GMTIGO4S","download_json":"https://pith.science/pith/VO2CHNY3KQ7MH5MTD3GMTIGO4S.json","view_paper":"https://pith.science/paper/VO2CHNY3","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2410.03717&json=true","fetch_graph":"https://pith.science/api/pith-number/VO2CHNY3KQ7MH5MTD3GMTIGO4S/graph.json","fetch_events":"https://pith.science/api/pith-number/VO2CHNY3KQ7MH5MTD3GMTIGO4S/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/VO2CHNY3KQ7MH5MTD3GMTIGO4S/action/timestamp_anchor","attest_storage":"https://pith.science/pith/VO2CHNY3KQ7MH5MTD3GMTIGO4S/action/storage_attestation","attest_author":"https://pith.science/pith/VO2CHNY3KQ7MH5MTD3GMTIGO4S/action/author_attestation","sign_citation":"https://pith.science/pith/VO2CHNY3KQ7MH5MTD3GMTIGO4S/action/citation_signature","submit_replication":"https://pith.science/pith/VO2CHNY3KQ7MH5MTD3GMTIGO4S/action/replication_record"}},"created_at":"2026-07-05T09:16:23.804598+00:00","updated_at":"2026-07-05T09:16:23.804598+00:00"}