{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:7V3VO2FJCCRPSWR2S4YMVMKNXN","short_pith_number":"pith:7V3VO2FJ","schema_version":"1.0","canonical_sha256":"fd775768a910a2f95a3a9730cab14dbb79c1a124081982b6f04fb5e25fd9abaf","source":{"kind":"arxiv","id":"2605.20285","version":1},"attestation_state":"computed","paper":{"title":"Introspective X Training: Feedback Conditioning Improves Scaling Across all LLM Training Stages","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Brandon Cui, David Acuna, Hyunwoo Kim, Jaehun Jung, Prithviraj Ammanabrolu, Shrimai Prabhumoye, Syeda Nahida Akter, Ximing Lu, Yejin Choi, Yuxiao Qu","submitted_at":"2026-05-19T06:04:08Z","abstract_excerpt":"We tackle the question of how to scale more efficiently across the many, ever-growing stages of current LLM training pipelines. Our guiding intuition stems from the fact that the dynamics of later stages of the pipeline, e.g. post-training, can be used to inform earlier stages such as pre-training. To this end, we propose Introspective Training (or IXT), inspired by offline reward-conditioned reinforcement learning and applicable to any stage of training. IXT uses a thinking reward model to annotate data with natural language critique based feedback, enabling quality aware training from the ea"},"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":"2605.20285","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-05-19T06:04:08Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"8270e0f77279ef8ddc9cf2bab8246c221d1c8a31f8f7accc7ac95c9e1567fa08","abstract_canon_sha256":"c295eaf7952f4cc7da1daf18b9deb116851197affb28cc65bd2e4b3a03158e89"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-21T00:04:24.001028Z","signature_b64":"gSutN9CZZXgWIFDicodPnCkEt2BLWVFyPgHpXZuHM5oILewS2AATVy/t3mGKeb/z/YiIbZpVti/Y1dZ7+PXHDw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"fd775768a910a2f95a3a9730cab14dbb79c1a124081982b6f04fb5e25fd9abaf","last_reissued_at":"2026-05-21T00:04:24.000506Z","signature_status":"signed_v1","first_computed_at":"2026-05-21T00:04:24.000506Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Introspective X Training: Feedback Conditioning Improves Scaling Across all LLM Training Stages","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Brandon Cui, David Acuna, Hyunwoo Kim, Jaehun Jung, Prithviraj Ammanabrolu, Shrimai Prabhumoye, Syeda Nahida Akter, Ximing Lu, Yejin Choi, Yuxiao Qu","submitted_at":"2026-05-19T06:04:08Z","abstract_excerpt":"We tackle the question of how to scale more efficiently across the many, ever-growing stages of current LLM training pipelines. Our guiding intuition stems from the fact that the dynamics of later stages of the pipeline, e.g. post-training, can be used to inform earlier stages such as pre-training. To this end, we propose Introspective Training (or IXT), inspired by offline reward-conditioned reinforcement learning and applicable to any stage of training. IXT uses a thinking reward model to annotate data with natural language critique based feedback, enabling quality aware training from the ea"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2605.20285","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/2605.20285/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":"2605.20285","created_at":"2026-05-21T00:04:24.000572+00:00"},{"alias_kind":"arxiv_version","alias_value":"2605.20285v1","created_at":"2026-05-21T00:04:24.000572+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.20285","created_at":"2026-05-21T00:04:24.000572+00:00"},{"alias_kind":"pith_short_12","alias_value":"7V3VO2FJCCRP","created_at":"2026-05-21T00:04:24.000572+00:00"},{"alias_kind":"pith_short_16","alias_value":"7V3VO2FJCCRPSWR2","created_at":"2026-05-21T00:04:24.000572+00:00"},{"alias_kind":"pith_short_8","alias_value":"7V3VO2FJ","created_at":"2026-05-21T00:04:24.000572+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/7V3VO2FJCCRPSWR2S4YMVMKNXN","json":"https://pith.science/pith/7V3VO2FJCCRPSWR2S4YMVMKNXN.json","graph_json":"https://pith.science/api/pith-number/7V3VO2FJCCRPSWR2S4YMVMKNXN/graph.json","events_json":"https://pith.science/api/pith-number/7V3VO2FJCCRPSWR2S4YMVMKNXN/events.json","paper":"https://pith.science/paper/7V3VO2FJ"},"agent_actions":{"view_html":"https://pith.science/pith/7V3VO2FJCCRPSWR2S4YMVMKNXN","download_json":"https://pith.science/pith/7V3VO2FJCCRPSWR2S4YMVMKNXN.json","view_paper":"https://pith.science/paper/7V3VO2FJ","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2605.20285&json=true","fetch_graph":"https://pith.science/api/pith-number/7V3VO2FJCCRPSWR2S4YMVMKNXN/graph.json","fetch_events":"https://pith.science/api/pith-number/7V3VO2FJCCRPSWR2S4YMVMKNXN/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/7V3VO2FJCCRPSWR2S4YMVMKNXN/action/timestamp_anchor","attest_storage":"https://pith.science/pith/7V3VO2FJCCRPSWR2S4YMVMKNXN/action/storage_attestation","attest_author":"https://pith.science/pith/7V3VO2FJCCRPSWR2S4YMVMKNXN/action/author_attestation","sign_citation":"https://pith.science/pith/7V3VO2FJCCRPSWR2S4YMVMKNXN/action/citation_signature","submit_replication":"https://pith.science/pith/7V3VO2FJCCRPSWR2S4YMVMKNXN/action/replication_record"}},"created_at":"2026-05-21T00:04:24.000572+00:00","updated_at":"2026-05-21T00:04:24.000572+00:00"}