{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:Y7KS2YN22NQU7VINAEUXTHR2TV","short_pith_number":"pith:Y7KS2YN2","schema_version":"1.0","canonical_sha256":"c7d52d61bad3614fd50d0129799e3a9d6ff56e1d00ca57a05aa0c5c065316e4a","source":{"kind":"arxiv","id":"2505.19590","version":4},"attestation_state":"computed","paper":{"title":"Learning to Reason without External Rewards","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"Large language models can improve at reasoning by using only their own internal confidence as the reward signal.","cross_cats":["cs.CL"],"primary_cat":"cs.LG","authors_text":"Aosong Feng, Dawn Song, Sergey Levine, Xuandong Zhao, Zhewei Kang","submitted_at":"2025-05-26T07:01:06Z","abstract_excerpt":"Training large language models (LLMs) for complex reasoning via Reinforcement Learning with Verifiable Rewards (RLVR) is effective but limited by reliance on costly, domain-specific supervision. We explore Reinforcement Learning from Internal Feedback (RLIF), a framework that enables LLMs to learn from intrinsic signals without external rewards or labeled data. We propose Intuitor, an RLIF method that uses a model's own confidence-termed self-certainty-as its sole reward signal. Intuitor replaces external rewards in Group Relative Policy Optimization (GRPO) with self-certainty scores, enabling"},"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":true},"canonical_record":{"source":{"id":"2505.19590","kind":"arxiv","version":4},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2025-05-26T07:01:06Z","cross_cats_sorted":["cs.CL"],"title_canon_sha256":"d21b544e2c7a446c1b6b4b60f6885a2ade25ecdf29f07a226c5c29afa33312ff","abstract_canon_sha256":"a83a896d73f839a3f15411ac5d5918d65098f6a0030ce7b0c3a81486cd90ea90"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-17T23:38:50.140871Z","signature_b64":"fgz/tcPL4e/OckjF25z2BRJ+rHspf6b/m2pnczHk5xoY2ekIC+UTWUQ65COL6jSSoMpz7N0b0r2cx6Ukkc9gBw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"c7d52d61bad3614fd50d0129799e3a9d6ff56e1d00ca57a05aa0c5c065316e4a","last_reissued_at":"2026-05-17T23:38:50.140295Z","signature_status":"signed_v1","first_computed_at":"2026-05-17T23:38:50.140295Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Learning to Reason without External Rewards","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"Large language models can improve at reasoning by using only their own internal confidence as the reward signal.","cross_cats":["cs.CL"],"primary_cat":"cs.LG","authors_text":"Aosong Feng, Dawn Song, Sergey Levine, Xuandong Zhao, Zhewei Kang","submitted_at":"2025-05-26T07:01:06Z","abstract_excerpt":"Training large language models (LLMs) for complex reasoning via Reinforcement Learning with Verifiable Rewards (RLVR) is effective but limited by reliance on costly, domain-specific supervision. We explore Reinforcement Learning from Internal Feedback (RLIF), a framework that enables LLMs to learn from intrinsic signals without external rewards or labeled data. We propose Intuitor, an RLIF method that uses a model's own confidence-termed self-certainty-as its sole reward signal. Intuitor replaces external rewards in Group Relative Policy Optimization (GRPO) with self-certainty scores, enabling"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"Experiments demonstrate that Intuitor matches GRPO's performance on mathematical benchmarks while achieving better generalization to out-of-domain tasks like code generation, without requiring gold solutions or test cases.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That the model's self-certainty score reliably indicates correct reasoning and does not encourage reward hacking or overconfidence in incorrect outputs.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"Intuitor enables LLMs to learn complex reasoning from self-certainty signals alone, matching supervised RL performance on math benchmarks while generalizing better to code generation without gold solutions or test cases.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Large language models can improve at reasoning by using only their own internal confidence as the reward signal.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"413e65f2c0095f14fbe0af388adc5b7212507fe4807fe16163e3629ff6355cda"},"source":{"id":"2505.19590","kind":"arxiv","version":4},"verdict":{"id":"7135c8df-9dc2-4fae-9f8f-bbd33c2ee816","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-15T21:13:34.177291Z","strongest_claim":"Experiments demonstrate that Intuitor matches GRPO's performance on mathematical benchmarks while achieving better generalization to out-of-domain tasks like code generation, without requiring gold solutions or test cases.","one_line_summary":"Intuitor enables LLMs to learn complex reasoning from self-certainty signals alone, matching supervised RL performance on math benchmarks while generalizing better to code generation without gold solutions or test cases.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That the model's self-certainty score reliably indicates correct reasoning and does not encourage reward hacking or overconfidence in incorrect outputs.","pith_extraction_headline":"Large language models can improve at reasoning by using only their own internal confidence as the reward signal."},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","internal_anchors":0},"formal_canon":{"evidence_count":1,"snapshot_sha256":"bea6854910e03788df088a3ea5ad244a858f860c0aacd3932bfb93c950a30740"},"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":"2505.19590","created_at":"2026-05-17T23:38:50.140383+00:00"},{"alias_kind":"arxiv_version","alias_value":"2505.19590v4","created_at":"2026-05-17T23:38:50.140383+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2505.19590","created_at":"2026-05-17T23:38:50.140383+00:00"},{"alias_kind":"pith_short_12","alias_value":"Y7KS2YN22NQU","created_at":"2026-05-18T12:33:37.589309+00:00"},{"alias_kind":"pith_short_16","alias_value":"Y7KS2YN22NQU7VIN","created_at":"2026-05-18T12:33:37.589309+00:00"},{"alias_kind":"pith_short_8","alias_value":"Y7KS2YN2","created_at":"2026-05-18T12:33:37.589309+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":26,"internal_anchor_count":26,"sample":[{"citing_arxiv_id":"2510.07962","citing_title":"LightReasoner: Can Small Language Models Teach Large Language Models Reasoning?","ref_index":21,"is_internal_anchor":true},{"citing_arxiv_id":"2605.21125","citing_title":"Advantage Collapse in Group Relative Policy Optimization: Diagnosis and Mitigation","ref_index":49,"is_internal_anchor":true},{"citing_arxiv_id":"2605.18899","citing_title":"Don't Let Bandit Feedback Pull Continual LLM-Recommender Updates Off Target","ref_index":12,"is_internal_anchor":true},{"citing_arxiv_id":"2605.19444","citing_title":"Detecting and Mitigating the Correct-Answer Extinction Window in Test-Time Reinforcement Learning with Majority Voting","ref_index":12,"is_internal_anchor":true},{"citing_arxiv_id":"2508.19652","citing_title":"Self-Rewarding Vision-Language Model via Reasoning Decomposition","ref_index":29,"is_internal_anchor":true},{"citing_arxiv_id":"2509.05489","citing_title":"Self-Aligned Reward: Towards Effective and Efficient Reasoners","ref_index":54,"is_internal_anchor":true},{"citing_arxiv_id":"2602.00986","citing_title":"Sparse Reward Subsystem in Large Language Models","ref_index":25,"is_internal_anchor":true},{"citing_arxiv_id":"2603.00918","citing_title":"Improving Text-to-Image Generation with Intrinsic Self-Confidence Rewards","ref_index":90,"is_internal_anchor":true},{"citing_arxiv_id":"2603.27977","citing_title":"SARL: Label-Free Reinforcement Learning by Rewarding Reasoning Topology","ref_index":31,"is_internal_anchor":true},{"citing_arxiv_id":"2605.13467","citing_title":"PDCR: Perception-Decomposed Confidence Reward for Vision-Language Reasoning","ref_index":53,"is_internal_anchor":true},{"citing_arxiv_id":"2507.17746","citing_title":"Rubrics as Rewards: Reinforcement Learning Beyond Verifiable Domains","ref_index":39,"is_internal_anchor":true},{"citing_arxiv_id":"2605.11609","citing_title":"Anti-Self-Distillation for Reasoning RL via Pointwise Mutual Information","ref_index":37,"is_internal_anchor":true},{"citing_arxiv_id":"2604.23318","citing_title":"Hidden States Know Where Reasoning Diverges: Credit Assignment via Span-Level Wasserstein Distance","ref_index":38,"is_internal_anchor":true},{"citing_arxiv_id":"2605.06642","citing_title":"StraTA: Incentivizing Agentic Reinforcement Learning with Strategic Trajectory Abstraction","ref_index":44,"is_internal_anchor":true},{"citing_arxiv_id":"2605.00380","citing_title":"ResRL: Boosting LLM Reasoning via Negative Sample Projection Residual Reinforcement Learning","ref_index":55,"is_internal_anchor":true},{"citing_arxiv_id":"2604.21327","citing_title":"Understanding and Mitigating Spurious Signal Amplification in Test-Time Reinforcement Learning for Math Reasoning","ref_index":17,"is_internal_anchor":true},{"citing_arxiv_id":"2604.19295","citing_title":"TEMPO: Scaling Test-time Training for Large Reasoning Models","ref_index":14,"is_internal_anchor":true},{"citing_arxiv_id":"2604.10827","citing_title":"Your Model Diversity, Not Method, Determines Reasoning Strategy","ref_index":20,"is_internal_anchor":true},{"citing_arxiv_id":"2604.13035","citing_title":"SceneCritic: A Symbolic Evaluator for 3D Indoor Scene Synthesis","ref_index":45,"is_internal_anchor":true},{"citing_arxiv_id":"2605.04065","citing_title":"Free Energy-Driven Reinforcement Learning with Adaptive Advantage Shaping for Unsupervised Reasoning in LLMs","ref_index":63,"is_internal_anchor":true},{"citing_arxiv_id":"2605.07244","citing_title":"Experience Sharing in Mutual Reinforcement Learning for Heterogeneous Language Models","ref_index":117,"is_internal_anchor":true},{"citing_arxiv_id":"2605.00380","citing_title":"ResRL: Boosting LLM Reasoning via Negative Sample Projection Residual Reinforcement Learning","ref_index":55,"is_internal_anchor":true},{"citing_arxiv_id":"2604.11435","citing_title":"Think Before you Write: QA-Guided Reasoning for Character Descriptions in Books","ref_index":13,"is_internal_anchor":true},{"citing_arxiv_id":"2604.17614","citing_title":"Characterizing Model-Native Skills","ref_index":74,"is_internal_anchor":true},{"citing_arxiv_id":"2604.18493","citing_title":"Too Correct to Learn: Reinforcement Learning on Saturated Reasoning Data","ref_index":15,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":1,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/Y7KS2YN22NQU7VINAEUXTHR2TV","json":"https://pith.science/pith/Y7KS2YN22NQU7VINAEUXTHR2TV.json","graph_json":"https://pith.science/api/pith-number/Y7KS2YN22NQU7VINAEUXTHR2TV/graph.json","events_json":"https://pith.science/api/pith-number/Y7KS2YN22NQU7VINAEUXTHR2TV/events.json","paper":"https://pith.science/paper/Y7KS2YN2"},"agent_actions":{"view_html":"https://pith.science/pith/Y7KS2YN22NQU7VINAEUXTHR2TV","download_json":"https://pith.science/pith/Y7KS2YN22NQU7VINAEUXTHR2TV.json","view_paper":"https://pith.science/paper/Y7KS2YN2","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2505.19590&json=true","fetch_graph":"https://pith.science/api/pith-number/Y7KS2YN22NQU7VINAEUXTHR2TV/graph.json","fetch_events":"https://pith.science/api/pith-number/Y7KS2YN22NQU7VINAEUXTHR2TV/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/Y7KS2YN22NQU7VINAEUXTHR2TV/action/timestamp_anchor","attest_storage":"https://pith.science/pith/Y7KS2YN22NQU7VINAEUXTHR2TV/action/storage_attestation","attest_author":"https://pith.science/pith/Y7KS2YN22NQU7VINAEUXTHR2TV/action/author_attestation","sign_citation":"https://pith.science/pith/Y7KS2YN22NQU7VINAEUXTHR2TV/action/citation_signature","submit_replication":"https://pith.science/pith/Y7KS2YN22NQU7VINAEUXTHR2TV/action/replication_record"}},"created_at":"2026-05-17T23:38:50.140383+00:00","updated_at":"2026-05-17T23:38:50.140383+00:00"}