{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:ICJ2MJKWYYTVA3RHGPISHW6N3C","short_pith_number":"pith:ICJ2MJKW","schema_version":"1.0","canonical_sha256":"4093a62556c627506e2733d123dbcdd8b20707c09d998386d6784288921174ab","source":{"kind":"arxiv","id":"2407.15421","version":3},"attestation_state":"computed","paper":{"title":"Planning in a recurrent neural network that plays Sokoban","license":"http://creativecommons.org/licenses/by-sa/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Aaron David Tucker, Adam Gleave, Adri\\`a Garriga-Alonso, Chris Cundy, Maximilian Li, Mohammad Taufeeque, Philip Quirke","submitted_at":"2024-07-22T06:57:34Z","abstract_excerpt":"Planning is essential for solving complex tasks, yet the internal mechanisms underlying planning in neural networks remain poorly understood. Building on prior work, we analyze a recurrent neural network (RNN) trained on Sokoban, a challenging puzzle requiring sequential, irreversible decisions. We find that the RNN has a causal plan representation which predicts its future actions about 50 steps in advance. The quality and length of the represented plan increases over the first few steps. We uncover a surprising behavior: the RNN \"paces\" in cycles to give itself extra computation at the start"},"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":"2407.15421","kind":"arxiv","version":3},"metadata":{"license":"http://creativecommons.org/licenses/by-sa/4.0/","primary_cat":"cs.LG","submitted_at":"2024-07-22T06:57:34Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"65479c9576d7af1e1e10053b659c6d12176a514b632f22f2ca25b096914b4cec","abstract_canon_sha256":"9924dea8e6643ba8ffa8787ed7a584f91f49f4bf044c15896cbea7545e23dcab"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:12:18.903298Z","signature_b64":"SWdSyLusAtMQyPqSnblgbMH6eKjbcsvttzA3B9q2lNDHCcitKOgG3OBI8Ve197QBmEqRXitb5nK6iY+7icG+Dg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"4093a62556c627506e2733d123dbcdd8b20707c09d998386d6784288921174ab","last_reissued_at":"2026-07-05T11:12:18.902759Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:12:18.902759Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Planning in a recurrent neural network that plays Sokoban","license":"http://creativecommons.org/licenses/by-sa/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Aaron David Tucker, Adam Gleave, Adri\\`a Garriga-Alonso, Chris Cundy, Maximilian Li, Mohammad Taufeeque, Philip Quirke","submitted_at":"2024-07-22T06:57:34Z","abstract_excerpt":"Planning is essential for solving complex tasks, yet the internal mechanisms underlying planning in neural networks remain poorly understood. Building on prior work, we analyze a recurrent neural network (RNN) trained on Sokoban, a challenging puzzle requiring sequential, irreversible decisions. We find that the RNN has a causal plan representation which predicts its future actions about 50 steps in advance. The quality and length of the represented plan increases over the first few steps. We uncover a surprising behavior: the RNN \"paces\" in cycles to give itself extra computation at the start"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2407.15421","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/2407.15421/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":"2407.15421","created_at":"2026-07-05T11:12:18.902835+00:00"},{"alias_kind":"arxiv_version","alias_value":"2407.15421v3","created_at":"2026-07-05T11:12:18.902835+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2407.15421","created_at":"2026-07-05T11:12:18.902835+00:00"},{"alias_kind":"pith_short_12","alias_value":"ICJ2MJKWYYTV","created_at":"2026-07-05T11:12:18.902835+00:00"},{"alias_kind":"pith_short_16","alias_value":"ICJ2MJKWYYTVA3RH","created_at":"2026-07-05T11:12:18.902835+00:00"},{"alias_kind":"pith_short_8","alias_value":"ICJ2MJKW","created_at":"2026-07-05T11:12:18.902835+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":3,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2607.05188","citing_title":"Latent Programming Horizons in Coding Agents","ref_index":23,"is_internal_anchor":true},{"citing_arxiv_id":"2606.01302","citing_title":"Structure and Scale in Simplicial Sequence Modelling","ref_index":62,"is_internal_anchor":false},{"citing_arxiv_id":"2606.03645","citing_title":"The Shape of Addition: Geometric Structures of Arithmetic in Large Language Models","ref_index":36,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/ICJ2MJKWYYTVA3RHGPISHW6N3C","json":"https://pith.science/pith/ICJ2MJKWYYTVA3RHGPISHW6N3C.json","graph_json":"https://pith.science/api/pith-number/ICJ2MJKWYYTVA3RHGPISHW6N3C/graph.json","events_json":"https://pith.science/api/pith-number/ICJ2MJKWYYTVA3RHGPISHW6N3C/events.json","paper":"https://pith.science/paper/ICJ2MJKW"},"agent_actions":{"view_html":"https://pith.science/pith/ICJ2MJKWYYTVA3RHGPISHW6N3C","download_json":"https://pith.science/pith/ICJ2MJKWYYTVA3RHGPISHW6N3C.json","view_paper":"https://pith.science/paper/ICJ2MJKW","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2407.15421&json=true","fetch_graph":"https://pith.science/api/pith-number/ICJ2MJKWYYTVA3RHGPISHW6N3C/graph.json","fetch_events":"https://pith.science/api/pith-number/ICJ2MJKWYYTVA3RHGPISHW6N3C/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/ICJ2MJKWYYTVA3RHGPISHW6N3C/action/timestamp_anchor","attest_storage":"https://pith.science/pith/ICJ2MJKWYYTVA3RHGPISHW6N3C/action/storage_attestation","attest_author":"https://pith.science/pith/ICJ2MJKWYYTVA3RHGPISHW6N3C/action/author_attestation","sign_citation":"https://pith.science/pith/ICJ2MJKWYYTVA3RHGPISHW6N3C/action/citation_signature","submit_replication":"https://pith.science/pith/ICJ2MJKWYYTVA3RHGPISHW6N3C/action/replication_record"}},"created_at":"2026-07-05T11:12:18.902835+00:00","updated_at":"2026-07-05T11:12:18.902835+00:00"}