{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:G7MHQXAT2RJTXENUF2SKG33H5M","short_pith_number":"pith:G7MHQXAT","schema_version":"1.0","canonical_sha256":"37d8785c13d4533b91b42ea4a36f67eb0305ed4b31d8e8a5fa36f6b6f84d0743","source":{"kind":"arxiv","id":"2502.12118","version":2},"attestation_state":"computed","paper":{"title":"Scaling Test-Time Compute Without Verification or RL is Suboptimal","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CL"],"primary_cat":"cs.LG","authors_text":"Amrith Setlur, Aviral Kumar, Nived Rajaraman, Sergey Levine","submitted_at":"2025-02-17T18:43:24Z","abstract_excerpt":"Despite substantial advances in scaling test-time compute, an ongoing debate in the community is how it should be scaled up to enable continued and efficient improvements with scaling. There are largely two approaches: first, distilling successful search or thinking traces; and second, using verification (e.g., 0/1 outcome rewards, reward models, or verifiers) to guide reinforcement learning (RL) and search algorithms. In this paper, we prove that finetuning LLMs with verifier-based (VB) methods based on RL or search is far superior to verifier-free (VF) approaches based on distilling or cloni"},"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":"2502.12118","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-02-17T18:43:24Z","cross_cats_sorted":["cs.CL"],"title_canon_sha256":"640253596b0cfba6a4bc5ee00bea62fcb8542ddd0178c7f762ceb13a88f2123d","abstract_canon_sha256":"d4a4c976027b6256f83ea2a3bbca05297854c01e479ca637878575c22adb1a1b"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:16:22.239676Z","signature_b64":"AsTiGUyBU7fWQ1H5l/sOvbfr0svhvfhMZlJq2vGqzNU6un8IQUKPUbT8CGZ9PXIlfESK+qLzL5UlZSdiN6j4Aw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"37d8785c13d4533b91b42ea4a36f67eb0305ed4b31d8e8a5fa36f6b6f84d0743","last_reissued_at":"2026-07-05T10:16:22.239164Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:16:22.239164Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Scaling Test-Time Compute Without Verification or RL is Suboptimal","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CL"],"primary_cat":"cs.LG","authors_text":"Amrith Setlur, Aviral Kumar, Nived Rajaraman, Sergey Levine","submitted_at":"2025-02-17T18:43:24Z","abstract_excerpt":"Despite substantial advances in scaling test-time compute, an ongoing debate in the community is how it should be scaled up to enable continued and efficient improvements with scaling. There are largely two approaches: first, distilling successful search or thinking traces; and second, using verification (e.g., 0/1 outcome rewards, reward models, or verifiers) to guide reinforcement learning (RL) and search algorithms. In this paper, we prove that finetuning LLMs with verifier-based (VB) methods based on RL or search is far superior to verifier-free (VF) approaches based on distilling or cloni"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2502.12118","kind":"arxiv","version":2},"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/2502.12118/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":"2502.12118","created_at":"2026-07-05T10:16:22.239223+00:00"},{"alias_kind":"arxiv_version","alias_value":"2502.12118v2","created_at":"2026-07-05T10:16:22.239223+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2502.12118","created_at":"2026-07-05T10:16:22.239223+00:00"},{"alias_kind":"pith_short_12","alias_value":"G7MHQXAT2RJT","created_at":"2026-07-05T10:16:22.239223+00:00"},{"alias_kind":"pith_short_16","alias_value":"G7MHQXAT2RJTXENU","created_at":"2026-07-05T10:16:22.239223+00:00"},{"alias_kind":"pith_short_8","alias_value":"G7MHQXAT","created_at":"2026-07-05T10:16:22.239223+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":10,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.18089","citing_title":"From Reasoning Traces to Reusable Modules: Understanding Compositional Generalization in Language Model Reasoning","ref_index":80,"is_internal_anchor":false},{"citing_arxiv_id":"2605.30070","citing_title":"A Predictive Law for On-Policy Self-Distillation From World Feedback","ref_index":26,"is_internal_anchor":false},{"citing_arxiv_id":"2606.01075","citing_title":"On the Generalization Gap in Self-Evolving Language Model Reasoning","ref_index":29,"is_internal_anchor":false},{"citing_arxiv_id":"2504.01990","citing_title":"Advances and Challenges in Foundation Agents: From Brain-Inspired Intelligence to Evolutionary, Collaborative, and Safe Systems","ref_index":211,"is_internal_anchor":false},{"citing_arxiv_id":"2605.15513","citing_title":"CAPS: Cascaded Adaptive Pairwise Selection for Efficient Parallel Reasoning","ref_index":33,"is_internal_anchor":false},{"citing_arxiv_id":"2505.23912","citing_title":"LoVeC: Reinforcement Learning for Better Verbalized Confidence in Long-Form Generations","ref_index":43,"is_internal_anchor":false},{"citing_arxiv_id":"2503.17352","citing_title":"OpenVLThinker: Complex Vision-Language Reasoning via Iterative SFT-RL Cycles","ref_index":59,"is_internal_anchor":false},{"citing_arxiv_id":"2505.15134","citing_title":"The Unreasonable Effectiveness of Entropy Minimization in LLM Reasoning","ref_index":69,"is_internal_anchor":false},{"citing_arxiv_id":"2603.04333","citing_title":"What Does Flow Matching Bring To TD Learning?","ref_index":54,"is_internal_anchor":false},{"citing_arxiv_id":"2604.21268","citing_title":"Measure Twice, Click Once: Co-evolving Proposer and Visual Critic via Reinforcement Learning for GUI Grounding","ref_index":28,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/G7MHQXAT2RJTXENUF2SKG33H5M","json":"https://pith.science/pith/G7MHQXAT2RJTXENUF2SKG33H5M.json","graph_json":"https://pith.science/api/pith-number/G7MHQXAT2RJTXENUF2SKG33H5M/graph.json","events_json":"https://pith.science/api/pith-number/G7MHQXAT2RJTXENUF2SKG33H5M/events.json","paper":"https://pith.science/paper/G7MHQXAT"},"agent_actions":{"view_html":"https://pith.science/pith/G7MHQXAT2RJTXENUF2SKG33H5M","download_json":"https://pith.science/pith/G7MHQXAT2RJTXENUF2SKG33H5M.json","view_paper":"https://pith.science/paper/G7MHQXAT","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2502.12118&json=true","fetch_graph":"https://pith.science/api/pith-number/G7MHQXAT2RJTXENUF2SKG33H5M/graph.json","fetch_events":"https://pith.science/api/pith-number/G7MHQXAT2RJTXENUF2SKG33H5M/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/G7MHQXAT2RJTXENUF2SKG33H5M/action/timestamp_anchor","attest_storage":"https://pith.science/pith/G7MHQXAT2RJTXENUF2SKG33H5M/action/storage_attestation","attest_author":"https://pith.science/pith/G7MHQXAT2RJTXENUF2SKG33H5M/action/author_attestation","sign_citation":"https://pith.science/pith/G7MHQXAT2RJTXENUF2SKG33H5M/action/citation_signature","submit_replication":"https://pith.science/pith/G7MHQXAT2RJTXENUF2SKG33H5M/action/replication_record"}},"created_at":"2026-07-05T10:16:22.239223+00:00","updated_at":"2026-07-05T10:16:22.239223+00:00"}