{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:37G2J3II3D2P2EPFJ2GB33D2VA","short_pith_number":"pith:37G2J3II","schema_version":"1.0","canonical_sha256":"dfcda4ed08d8f4fd11e54e8c1dec7aa838b1c38911859c62d34234b58cda1740","source":{"kind":"arxiv","id":"2605.27354","version":1},"attestation_state":"computed","paper":{"title":"Guiding LLM Post-training Data Engineering with Model Internals from Sparse Autoencoders","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.CL"],"primary_cat":"cs.LG","authors_text":"Jinwu Hu, Juanzi Li, Lei Hou, Xiaozhi Wang, Yi Jing, Zao Dai, Zijun Yao","submitted_at":"2026-05-26T17:55:59Z","abstract_excerpt":"Model internals encode rich information about how a large language model (LLM) processes its training data; however, post-training data engineering largely relies on external signals and ignores rich intrinsic signals lying in model internals. We propose SAERL, a data engineering framework for LLM reinforcement learning (RL). It models three intrinsic data properties: diversity, difficulty, and quality, using model internals extracted with Sparse Autoencoder (SAE), an advanced mechanistic interpretability tool. Each property grounds a concrete data engineering operation: SAE-space clustering w"},"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":"2605.27354","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-05-26T17:55:59Z","cross_cats_sorted":["cs.AI","cs.CL"],"title_canon_sha256":"947a3b576177f2dff21551a7b800a0cb3a62d015c8f580d43a9ba91882d77ef2","abstract_canon_sha256":"c0062ddcc8ff7469a8a8a21afd1b3dfeac32da237d98bccd88c0db0893db6a16"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-27T02:06:19.458919Z","signature_b64":"8X8RCkS/XDZM6ixtsBbL4SK6/X+3l7gqGsW/CHCOqnIZXLdqKhvIaWsulxkc2Sc3nF7TzhsFvn9vjMXycv9lDA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"dfcda4ed08d8f4fd11e54e8c1dec7aa838b1c38911859c62d34234b58cda1740","last_reissued_at":"2026-05-27T02:06:19.458167Z","signature_status":"signed_v1","first_computed_at":"2026-05-27T02:06:19.458167Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Guiding LLM Post-training Data Engineering with Model Internals from Sparse Autoencoders","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.CL"],"primary_cat":"cs.LG","authors_text":"Jinwu Hu, Juanzi Li, Lei Hou, Xiaozhi Wang, Yi Jing, Zao Dai, Zijun Yao","submitted_at":"2026-05-26T17:55:59Z","abstract_excerpt":"Model internals encode rich information about how a large language model (LLM) processes its training data; however, post-training data engineering largely relies on external signals and ignores rich intrinsic signals lying in model internals. We propose SAERL, a data engineering framework for LLM reinforcement learning (RL). It models three intrinsic data properties: diversity, difficulty, and quality, using model internals extracted with Sparse Autoencoder (SAE), an advanced mechanistic interpretability tool. Each property grounds a concrete data engineering operation: SAE-space clustering w"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2605.27354","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/2605.27354/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":"2605.27354","created_at":"2026-05-27T02:06:19.458299+00:00"},{"alias_kind":"arxiv_version","alias_value":"2605.27354v1","created_at":"2026-05-27T02:06:19.458299+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.27354","created_at":"2026-05-27T02:06:19.458299+00:00"},{"alias_kind":"pith_short_12","alias_value":"37G2J3II3D2P","created_at":"2026-05-27T02:06:19.458299+00:00"},{"alias_kind":"pith_short_16","alias_value":"37G2J3II3D2P2EPF","created_at":"2026-05-27T02:06:19.458299+00:00"},{"alias_kind":"pith_short_8","alias_value":"37G2J3II","created_at":"2026-05-27T02:06:19.458299+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/37G2J3II3D2P2EPFJ2GB33D2VA","json":"https://pith.science/pith/37G2J3II3D2P2EPFJ2GB33D2VA.json","graph_json":"https://pith.science/api/pith-number/37G2J3II3D2P2EPFJ2GB33D2VA/graph.json","events_json":"https://pith.science/api/pith-number/37G2J3II3D2P2EPFJ2GB33D2VA/events.json","paper":"https://pith.science/paper/37G2J3II"},"agent_actions":{"view_html":"https://pith.science/pith/37G2J3II3D2P2EPFJ2GB33D2VA","download_json":"https://pith.science/pith/37G2J3II3D2P2EPFJ2GB33D2VA.json","view_paper":"https://pith.science/paper/37G2J3II","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2605.27354&json=true","fetch_graph":"https://pith.science/api/pith-number/37G2J3II3D2P2EPFJ2GB33D2VA/graph.json","fetch_events":"https://pith.science/api/pith-number/37G2J3II3D2P2EPFJ2GB33D2VA/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/37G2J3II3D2P2EPFJ2GB33D2VA/action/timestamp_anchor","attest_storage":"https://pith.science/pith/37G2J3II3D2P2EPFJ2GB33D2VA/action/storage_attestation","attest_author":"https://pith.science/pith/37G2J3II3D2P2EPFJ2GB33D2VA/action/author_attestation","sign_citation":"https://pith.science/pith/37G2J3II3D2P2EPFJ2GB33D2VA/action/citation_signature","submit_replication":"https://pith.science/pith/37G2J3II3D2P2EPFJ2GB33D2VA/action/replication_record"}},"created_at":"2026-05-27T02:06:19.458299+00:00","updated_at":"2026-05-27T02:06:19.458299+00:00"}