{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:TWNKVSEAIWAC3S72FADKTU4GXZ","short_pith_number":"pith:TWNKVSEA","schema_version":"1.0","canonical_sha256":"9d9aaac88045802dcbfa2806a9d386be57a412232a71b5f3dca7a58818b02a41","source":{"kind":"arxiv","id":"2408.15240","version":3},"attestation_state":"computed","paper":{"title":"Generative Verifiers: Reward Modeling as Next-Token Prediction","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Arian Hosseini, Aviral Kumar, Hritik Bansal, Lunjun Zhang, Mehran Kazemi, Rishabh Agarwal","submitted_at":"2024-08-27T17:57:45Z","abstract_excerpt":"Verifiers or reward models are often used to enhance the reasoning performance of large language models (LLMs). A common approach is the Best-of-N method, where N candidate solutions generated by the LLM are ranked by a verifier, and the best one is selected. While LLM-based verifiers are typically trained as discriminative classifiers to score solutions, they do not utilize the text generation capabilities of pretrained LLMs. To overcome this limitation, we instead propose training verifiers using the ubiquitous next-token prediction objective, jointly on verification and solution generation."},"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":"2408.15240","kind":"arxiv","version":3},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2024-08-27T17:57:45Z","cross_cats_sorted":[],"title_canon_sha256":"be6c3cc8c482910cf86635e2d7f6d51bd920edced21511c8831b06de01c91495","abstract_canon_sha256":"1880e203f34223dcfe36aae22265e510e368aa0f947fa2ace5333e8ff8bfbfc9"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:18:09.807142Z","signature_b64":"GXOFLlqAOoFyE1geG+zruLYGSbfP0nvuDWcow2FfBy5o2q//EZWzh2oG54eEWc4N4rQkgHM+VwzUxOY/4XgKAw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"9d9aaac88045802dcbfa2806a9d386be57a412232a71b5f3dca7a58818b02a41","last_reissued_at":"2026-07-05T10:18:09.806665Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:18:09.806665Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Generative Verifiers: Reward Modeling as Next-Token Prediction","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Arian Hosseini, Aviral Kumar, Hritik Bansal, Lunjun Zhang, Mehran Kazemi, Rishabh Agarwal","submitted_at":"2024-08-27T17:57:45Z","abstract_excerpt":"Verifiers or reward models are often used to enhance the reasoning performance of large language models (LLMs). A common approach is the Best-of-N method, where N candidate solutions generated by the LLM are ranked by a verifier, and the best one is selected. While LLM-based verifiers are typically trained as discriminative classifiers to score solutions, they do not utilize the text generation capabilities of pretrained LLMs. To overcome this limitation, we instead propose training verifiers using the ubiquitous next-token prediction objective, jointly on verification and solution generation."},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2408.15240","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/2408.15240/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":"2408.15240","created_at":"2026-07-05T10:18:09.806723+00:00"},{"alias_kind":"arxiv_version","alias_value":"2408.15240v3","created_at":"2026-07-05T10:18:09.806723+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2408.15240","created_at":"2026-07-05T10:18:09.806723+00:00"},{"alias_kind":"pith_short_12","alias_value":"TWNKVSEAIWAC","created_at":"2026-07-05T10:18:09.806723+00:00"},{"alias_kind":"pith_short_16","alias_value":"TWNKVSEAIWAC3S72","created_at":"2026-07-05T10:18:09.806723+00:00"},{"alias_kind":"pith_short_8","alias_value":"TWNKVSEA","created_at":"2026-07-05T10:18:09.806723+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":29,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.09711","citing_title":"Proxy Reward Internalization and Mechanistic Exploitation: A Learned Precursor to Reward Hacking and Its Generalization","ref_index":61,"is_internal_anchor":false},{"citing_arxiv_id":"2606.09076","citing_title":"Beyond Scalar Rewards by Internalizing Reasoning into Score Distributions","ref_index":65,"is_internal_anchor":false},{"citing_arxiv_id":"2606.07434","citing_title":"Evidence Markets","ref_index":28,"is_internal_anchor":false},{"citing_arxiv_id":"2606.05152","citing_title":"Reinforcement Learning from Rich Feedback with Distributional DAgger","ref_index":35,"is_internal_anchor":false},{"citing_arxiv_id":"2606.01249","citing_title":"Trust Region On-Policy Distillation","ref_index":191,"is_internal_anchor":false},{"citing_arxiv_id":"2606.30851","citing_title":"Test-Time Verification for Text-to-SQL via Outcome Reward Models","ref_index":82,"is_internal_anchor":false},{"citing_arxiv_id":"2605.20531","citing_title":"Pseudo-Formalization for Automatic Proof Verification","ref_index":36,"is_internal_anchor":false},{"citing_arxiv_id":"2605.28805","citing_title":"OmniVerifier-M1: Multimodal Meta-Verifier with Explicit Structured Recalibration","ref_index":25,"is_internal_anchor":false},{"citing_arxiv_id":"2504.02181","citing_title":"A Survey of Scaling in Large Language Model Reasoning","ref_index":252,"is_internal_anchor":false},{"citing_arxiv_id":"2507.15698","citing_title":"CoLD: Counterfactually-Guided Length Debiasing for Process Reward Models in Mathematical Reasoning","ref_index":25,"is_internal_anchor":false},{"citing_arxiv_id":"2509.21319","citing_title":"RLBFF: Binary Flexible Feedback to bridge between Human Feedback & Verifiable Rewards","ref_index":49,"is_internal_anchor":false},{"citing_arxiv_id":"2605.20531","citing_title":"Pseudo-Formalization for Automatic Proof Verification","ref_index":36,"is_internal_anchor":false},{"citing_arxiv_id":"2410.08146","citing_title":"Rewarding Progress: Scaling Automated Process Verifiers for LLM Reasoning","ref_index":25,"is_internal_anchor":false},{"citing_arxiv_id":"2605.15113","citing_title":"Learning from Language Feedback via Variational Policy Distillation","ref_index":45,"is_internal_anchor":false},{"citing_arxiv_id":"2605.19228","citing_title":"Diagnosing Multi-step Reasoning Failures in Black-box LLMs via Stepwise Confidence Attribution","ref_index":47,"is_internal_anchor":false},{"citing_arxiv_id":"2605.19447","citing_title":"What and When to Distill: Selective Hindsight Distillation for Multi-Turn Agents","ref_index":40,"is_internal_anchor":false},{"citing_arxiv_id":"2605.10810","citing_title":"Likelihood scoring for continuations of mathematical text: a self-supervised benchmark with tests for shortcut vulnerabilities","ref_index":45,"is_internal_anchor":false},{"citing_arxiv_id":"2506.01937","citing_title":"RewardBench 2: Advancing Reward Model Evaluation","ref_index":61,"is_internal_anchor":false},{"citing_arxiv_id":"2503.17352","citing_title":"OpenVLThinker: Complex Vision-Language Reasoning via Iterative SFT-RL Cycles","ref_index":91,"is_internal_anchor":false},{"citing_arxiv_id":"2507.11687","citing_title":"MetaLint: Easy-to-Hard Generalization for Code Linting","ref_index":52,"is_internal_anchor":false},{"citing_arxiv_id":"2510.10649","citing_title":"Unlocking Exploration in RLVR: Uncertainty-aware Advantage Shaping for Deeper Reasoning","ref_index":22,"is_internal_anchor":false},{"citing_arxiv_id":"2510.24235","citing_title":"PaTaRM: Bridging Pairwise and Pointwise Signals via Preference-Aware Task-Adaptive Reward Modeling","ref_index":23,"is_internal_anchor":false},{"citing_arxiv_id":"2510.13786","citing_title":"The Art of Scaling Reinforcement Learning Compute for LLMs","ref_index":26,"is_internal_anchor":false},{"citing_arxiv_id":"2501.07301","citing_title":"The Lessons of Developing Process Reward Models in Mathematical Reasoning","ref_index":14,"is_internal_anchor":false},{"citing_arxiv_id":"2605.11299","citing_title":"Primal Generation, Dual Judgment: Self-Training from Test-Time Scaling","ref_index":95,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/TWNKVSEAIWAC3S72FADKTU4GXZ","json":"https://pith.science/pith/TWNKVSEAIWAC3S72FADKTU4GXZ.json","graph_json":"https://pith.science/api/pith-number/TWNKVSEAIWAC3S72FADKTU4GXZ/graph.json","events_json":"https://pith.science/api/pith-number/TWNKVSEAIWAC3S72FADKTU4GXZ/events.json","paper":"https://pith.science/paper/TWNKVSEA"},"agent_actions":{"view_html":"https://pith.science/pith/TWNKVSEAIWAC3S72FADKTU4GXZ","download_json":"https://pith.science/pith/TWNKVSEAIWAC3S72FADKTU4GXZ.json","view_paper":"https://pith.science/paper/TWNKVSEA","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2408.15240&json=true","fetch_graph":"https://pith.science/api/pith-number/TWNKVSEAIWAC3S72FADKTU4GXZ/graph.json","fetch_events":"https://pith.science/api/pith-number/TWNKVSEAIWAC3S72FADKTU4GXZ/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/TWNKVSEAIWAC3S72FADKTU4GXZ/action/timestamp_anchor","attest_storage":"https://pith.science/pith/TWNKVSEAIWAC3S72FADKTU4GXZ/action/storage_attestation","attest_author":"https://pith.science/pith/TWNKVSEAIWAC3S72FADKTU4GXZ/action/author_attestation","sign_citation":"https://pith.science/pith/TWNKVSEAIWAC3S72FADKTU4GXZ/action/citation_signature","submit_replication":"https://pith.science/pith/TWNKVSEAIWAC3S72FADKTU4GXZ/action/replication_record"}},"created_at":"2026-07-05T10:18:09.806723+00:00","updated_at":"2026-07-05T10:18:09.806723+00:00"}