{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:DL6ZPSHZVPQ7OEZNTKPDAQ7PRP","short_pith_number":"pith:DL6ZPSHZ","schema_version":"1.0","canonical_sha256":"1afd97c8f9abe1f7132d9a9e3043ef8bd5bb338d8395e4c3b36e5b12e9e33bc7","source":{"kind":"arxiv","id":"2504.07912","version":2},"attestation_state":"computed","paper":{"title":"Echo Chamber: RL Post-training Amplifies Behaviors Learned in Pretraining","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Alexandru Meterez, Cengiz Pehlevan, Eran Malach, Rosie Zhao, Samy Jelassi, Sham Kakade","submitted_at":"2025-04-10T17:15:53Z","abstract_excerpt":"Reinforcement learning (RL)-based fine-tuning has become a crucial step in post-training language models for advanced mathematical reasoning and coding. Following the success of frontier reasoning models, recent work has demonstrated that RL fine-tuning consistently improves performance, even in smaller-scale models; however, the underlying mechanisms driving these improvements are not well-understood. Understanding the effects of RL fine-tuning requires disentangling its interaction with pretraining data composition, hyperparameters, and model scale, but such problems are exacerbated by the l"},"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":"2504.07912","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-04-10T17:15:53Z","cross_cats_sorted":[],"title_canon_sha256":"684bdfc6b6b1c1e5c60246b2651a7a891449d5476f6f405cdfef109dab6296a4","abstract_canon_sha256":"128fb98abdc797a795d1e257d3d6a5108fa02c8d26a07a0fba6363de3085e9c0"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:50:34.745411Z","signature_b64":"M82j+jrIHnWB8c1oviP4SKthP4vY/rScF21TsWP591LrmBjcIG6bv4rwg6EsxbHjHbFIyAeVQNX2QYAmNBgBAQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"1afd97c8f9abe1f7132d9a9e3043ef8bd5bb338d8395e4c3b36e5b12e9e33bc7","last_reissued_at":"2026-07-05T11:50:34.744929Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:50:34.744929Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Echo Chamber: RL Post-training Amplifies Behaviors Learned in Pretraining","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Alexandru Meterez, Cengiz Pehlevan, Eran Malach, Rosie Zhao, Samy Jelassi, Sham Kakade","submitted_at":"2025-04-10T17:15:53Z","abstract_excerpt":"Reinforcement learning (RL)-based fine-tuning has become a crucial step in post-training language models for advanced mathematical reasoning and coding. Following the success of frontier reasoning models, recent work has demonstrated that RL fine-tuning consistently improves performance, even in smaller-scale models; however, the underlying mechanisms driving these improvements are not well-understood. Understanding the effects of RL fine-tuning requires disentangling its interaction with pretraining data composition, hyperparameters, and model scale, but such problems are exacerbated by the l"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2504.07912","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/2504.07912/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":"2504.07912","created_at":"2026-07-05T11:50:34.744993+00:00"},{"alias_kind":"arxiv_version","alias_value":"2504.07912v2","created_at":"2026-07-05T11:50:34.744993+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2504.07912","created_at":"2026-07-05T11:50:34.744993+00:00"},{"alias_kind":"pith_short_12","alias_value":"DL6ZPSHZVPQ7","created_at":"2026-07-05T11:50:34.744993+00:00"},{"alias_kind":"pith_short_16","alias_value":"DL6ZPSHZVPQ7OEZN","created_at":"2026-07-05T11:50:34.744993+00:00"},{"alias_kind":"pith_short_8","alias_value":"DL6ZPSHZ","created_at":"2026-07-05T11:50:34.744993+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":13,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2607.07646","citing_title":"RL Post-Training Builds Compositional Reasoning Strategies","ref_index":15,"is_internal_anchor":true},{"citing_arxiv_id":"2605.23926","citing_title":"How Much Thinking is Enough? Quantifying and Understanding Redundancy in LLM Reasoning","ref_index":24,"is_internal_anchor":false},{"citing_arxiv_id":"2606.18089","citing_title":"From Reasoning Traces to Reusable Modules: Understanding Compositional Generalization in Language Model Reasoning","ref_index":19,"is_internal_anchor":false},{"citing_arxiv_id":"2507.01679","citing_title":"Blending Supervised and Reinforcement Fine-Tuning with Prefix Sampling","ref_index":49,"is_internal_anchor":false},{"citing_arxiv_id":"2605.16874","citing_title":"Reasoning Can Be Restored by Correcting a Few Decision Tokens","ref_index":27,"is_internal_anchor":false},{"citing_arxiv_id":"2605.08401","citing_title":"AIPO: Learning to Reason from Active Interaction","ref_index":79,"is_internal_anchor":false},{"citing_arxiv_id":"2508.07809","citing_title":"EvoCoT: Overcoming the Exploration Bottleneck in Reinforcement Learning","ref_index":22,"is_internal_anchor":false},{"citing_arxiv_id":"2508.13755","citing_title":"Depth-Breadth Synergy in RLVR: Unlocking LLM Reasoning Gains with Adaptive Exploration","ref_index":19,"is_internal_anchor":false},{"citing_arxiv_id":"2509.25424","citing_title":"Polychromic Objectives for Reinforcement Learning","ref_index":48,"is_internal_anchor":false},{"citing_arxiv_id":"2602.03249","citing_title":"Accordion-Thinking: Self-Regulated Step Summaries for Efficient and Readable LLM Reasoning","ref_index":3,"is_internal_anchor":false},{"citing_arxiv_id":"2605.08401","citing_title":"AIPO: Learning to Reason from Active Interaction","ref_index":79,"is_internal_anchor":false},{"citing_arxiv_id":"2605.10781","citing_title":"Rebellious Student: Reversing Teacher Signals for Reasoning Exploration with Self-Distilled RLVR","ref_index":31,"is_internal_anchor":false},{"citing_arxiv_id":"2604.18530","citing_title":"OGER: A Robust Offline-Guided Exploration Reward for Hybrid Reinforcement Learning","ref_index":24,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/DL6ZPSHZVPQ7OEZNTKPDAQ7PRP","json":"https://pith.science/pith/DL6ZPSHZVPQ7OEZNTKPDAQ7PRP.json","graph_json":"https://pith.science/api/pith-number/DL6ZPSHZVPQ7OEZNTKPDAQ7PRP/graph.json","events_json":"https://pith.science/api/pith-number/DL6ZPSHZVPQ7OEZNTKPDAQ7PRP/events.json","paper":"https://pith.science/paper/DL6ZPSHZ"},"agent_actions":{"view_html":"https://pith.science/pith/DL6ZPSHZVPQ7OEZNTKPDAQ7PRP","download_json":"https://pith.science/pith/DL6ZPSHZVPQ7OEZNTKPDAQ7PRP.json","view_paper":"https://pith.science/paper/DL6ZPSHZ","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2504.07912&json=true","fetch_graph":"https://pith.science/api/pith-number/DL6ZPSHZVPQ7OEZNTKPDAQ7PRP/graph.json","fetch_events":"https://pith.science/api/pith-number/DL6ZPSHZVPQ7OEZNTKPDAQ7PRP/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/DL6ZPSHZVPQ7OEZNTKPDAQ7PRP/action/timestamp_anchor","attest_storage":"https://pith.science/pith/DL6ZPSHZVPQ7OEZNTKPDAQ7PRP/action/storage_attestation","attest_author":"https://pith.science/pith/DL6ZPSHZVPQ7OEZNTKPDAQ7PRP/action/author_attestation","sign_citation":"https://pith.science/pith/DL6ZPSHZVPQ7OEZNTKPDAQ7PRP/action/citation_signature","submit_replication":"https://pith.science/pith/DL6ZPSHZVPQ7OEZNTKPDAQ7PRP/action/replication_record"}},"created_at":"2026-07-05T11:50:34.744993+00:00","updated_at":"2026-07-05T11:50:34.744993+00:00"}