{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:SYUQ4DSJW7B42WZPKPCY2OFVE7","short_pith_number":"pith:SYUQ4DSJ","schema_version":"1.0","canonical_sha256":"96290e0e49b7c3cd5b2f53c58d38b527f680f4cea773671a6936a7c72d510b0e","source":{"kind":"arxiv","id":"2402.04858","version":2},"attestation_state":"computed","paper":{"title":"CodeIt: Self-Improving Language Models with Prioritized Hindsight Replay","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CL","cs.LG"],"primary_cat":"cs.AI","authors_text":"Auke Wiggers, Blazej Manczak, Corrado Rainone, David W. Zhang, Micha\\\"el Defferrard, Natasha Butt, Taco Cohen","submitted_at":"2024-02-07T13:55:27Z","abstract_excerpt":"Large language models are increasingly solving tasks that are commonly believed to require human-level reasoning ability. However, these models still perform very poorly on benchmarks of general intelligence such as the Abstraction and Reasoning Corpus (ARC). In this paper, we approach ARC as a programming-by-examples problem, and introduce a novel and scalable method for language model self-improvement called Code Iteration (CodeIt). Our method iterates between 1) program sampling and hindsight relabeling, and 2) learning from prioritized experience replay. By relabeling the goal of an episod"},"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":"2402.04858","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.AI","submitted_at":"2024-02-07T13:55:27Z","cross_cats_sorted":["cs.CL","cs.LG"],"title_canon_sha256":"54db9e65403a53f2c355ff6cf82dd07a0ac557455e33dae0a26ced9c0c4692eb","abstract_canon_sha256":"7ed95017e28a48ed0ffa9fa8a4d90ce98b10841f91f2526e3ec4e19aa4de9f0f"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T08:38:36.221865Z","signature_b64":"OqYu9B9vHBa8LoprSbfcjGkGVkcNk+etd4nbVIWFmjTE7JtmW1CTjWGHE968JqGOkJon5aqIqRtTRSoNUh85DQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"96290e0e49b7c3cd5b2f53c58d38b527f680f4cea773671a6936a7c72d510b0e","last_reissued_at":"2026-07-05T08:38:36.221386Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T08:38:36.221386Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"CodeIt: Self-Improving Language Models with Prioritized Hindsight Replay","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CL","cs.LG"],"primary_cat":"cs.AI","authors_text":"Auke Wiggers, Blazej Manczak, Corrado Rainone, David W. Zhang, Micha\\\"el Defferrard, Natasha Butt, Taco Cohen","submitted_at":"2024-02-07T13:55:27Z","abstract_excerpt":"Large language models are increasingly solving tasks that are commonly believed to require human-level reasoning ability. However, these models still perform very poorly on benchmarks of general intelligence such as the Abstraction and Reasoning Corpus (ARC). In this paper, we approach ARC as a programming-by-examples problem, and introduce a novel and scalable method for language model self-improvement called Code Iteration (CodeIt). Our method iterates between 1) program sampling and hindsight relabeling, and 2) learning from prioritized experience replay. By relabeling the goal of an episod"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2402.04858","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/2402.04858/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":"2402.04858","created_at":"2026-07-05T08:38:36.221443+00:00"},{"alias_kind":"arxiv_version","alias_value":"2402.04858v2","created_at":"2026-07-05T08:38:36.221443+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2402.04858","created_at":"2026-07-05T08:38:36.221443+00:00"},{"alias_kind":"pith_short_12","alias_value":"SYUQ4DSJW7B4","created_at":"2026-07-05T08:38:36.221443+00:00"},{"alias_kind":"pith_short_16","alias_value":"SYUQ4DSJW7B42WZP","created_at":"2026-07-05T08:38:36.221443+00:00"},{"alias_kind":"pith_short_8","alias_value":"SYUQ4DSJ","created_at":"2026-07-05T08:38:36.221443+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2604.18839","citing_title":"One Step Forward and K Steps Back: Better Reasoning with Denoising Recursion Models","ref_index":160,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/SYUQ4DSJW7B42WZPKPCY2OFVE7","json":"https://pith.science/pith/SYUQ4DSJW7B42WZPKPCY2OFVE7.json","graph_json":"https://pith.science/api/pith-number/SYUQ4DSJW7B42WZPKPCY2OFVE7/graph.json","events_json":"https://pith.science/api/pith-number/SYUQ4DSJW7B42WZPKPCY2OFVE7/events.json","paper":"https://pith.science/paper/SYUQ4DSJ"},"agent_actions":{"view_html":"https://pith.science/pith/SYUQ4DSJW7B42WZPKPCY2OFVE7","download_json":"https://pith.science/pith/SYUQ4DSJW7B42WZPKPCY2OFVE7.json","view_paper":"https://pith.science/paper/SYUQ4DSJ","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2402.04858&json=true","fetch_graph":"https://pith.science/api/pith-number/SYUQ4DSJW7B42WZPKPCY2OFVE7/graph.json","fetch_events":"https://pith.science/api/pith-number/SYUQ4DSJW7B42WZPKPCY2OFVE7/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/SYUQ4DSJW7B42WZPKPCY2OFVE7/action/timestamp_anchor","attest_storage":"https://pith.science/pith/SYUQ4DSJW7B42WZPKPCY2OFVE7/action/storage_attestation","attest_author":"https://pith.science/pith/SYUQ4DSJW7B42WZPKPCY2OFVE7/action/author_attestation","sign_citation":"https://pith.science/pith/SYUQ4DSJW7B42WZPKPCY2OFVE7/action/citation_signature","submit_replication":"https://pith.science/pith/SYUQ4DSJW7B42WZPKPCY2OFVE7/action/replication_record"}},"created_at":"2026-07-05T08:38:36.221443+00:00","updated_at":"2026-07-05T08:38:36.221443+00:00"}