{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:62X4MDFD3NNTNLZHONFX5UTPOK","short_pith_number":"pith:62X4MDFD","schema_version":"1.0","canonical_sha256":"f6afc60ca3db5b36af27734b7ed26f72b6f357d60d45dae8de80caba874d138a","source":{"kind":"arxiv","id":"2505.00234","version":3},"attestation_state":"computed","paper":{"title":"Self-Generated In-Context Examples Improve LLM Agents for Sequential Decision-Making Tasks","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CL"],"primary_cat":"cs.LG","authors_text":"Kayvon Fatahalian, Vishnu Sarukkai, Zhiqiang Xie","submitted_at":"2025-05-01T00:48:12Z","abstract_excerpt":"Improving Large Language Model (LLM) agents for sequential decision-making tasks typically requires extensive task-specific knowledge engineering--custom prompts, curated examples, and specialized observation/action spaces. We investigate a different approach where agents automatically improve by learning from their own successful experiences without human intervention. Our method constructs and refines a database of self-generated trajectories that serve as in-context examples for future tasks. Even naive accumulation of successful trajectories yields substantial performance gains across thre"},"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":"2505.00234","kind":"arxiv","version":3},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-05-01T00:48:12Z","cross_cats_sorted":["cs.CL"],"title_canon_sha256":"bce4e11f62ea38f9a7a1878625ea49b15aefad4be8e4fd0e192bb6a852b0d78d","abstract_canon_sha256":"1e459dd4629665110048f8d4fc1555c98e6d311b7b0e87e848e7f4ced3644839"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:04:22.060251Z","signature_b64":"afH35H1S8LGQP35ghKnKj5XqnzvfGZz8nWYp5Rzih14UiFHyza24OMsgGXsuwLUZs65L4KbCYeUMFJoLsNefBw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"f6afc60ca3db5b36af27734b7ed26f72b6f357d60d45dae8de80caba874d138a","last_reissued_at":"2026-07-05T11:04:22.059780Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:04:22.059780Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Self-Generated In-Context Examples Improve LLM Agents for Sequential Decision-Making Tasks","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CL"],"primary_cat":"cs.LG","authors_text":"Kayvon Fatahalian, Vishnu Sarukkai, Zhiqiang Xie","submitted_at":"2025-05-01T00:48:12Z","abstract_excerpt":"Improving Large Language Model (LLM) agents for sequential decision-making tasks typically requires extensive task-specific knowledge engineering--custom prompts, curated examples, and specialized observation/action spaces. We investigate a different approach where agents automatically improve by learning from their own successful experiences without human intervention. Our method constructs and refines a database of self-generated trajectories that serve as in-context examples for future tasks. Even naive accumulation of successful trajectories yields substantial performance gains across thre"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2505.00234","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/2505.00234/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":"2505.00234","created_at":"2026-07-05T11:04:22.059842+00:00"},{"alias_kind":"arxiv_version","alias_value":"2505.00234v3","created_at":"2026-07-05T11:04:22.059842+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2505.00234","created_at":"2026-07-05T11:04:22.059842+00:00"},{"alias_kind":"pith_short_12","alias_value":"62X4MDFD3NNT","created_at":"2026-07-05T11:04:22.059842+00:00"},{"alias_kind":"pith_short_16","alias_value":"62X4MDFD3NNTNLZH","created_at":"2026-07-05T11:04:22.059842+00:00"},{"alias_kind":"pith_short_8","alias_value":"62X4MDFD","created_at":"2026-07-05T11:04:22.059842+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":6,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.04628","citing_title":"RAMPART: Registry-based Agentic Memory with Priority-Aware Runtime Transformation","ref_index":17,"is_internal_anchor":false},{"citing_arxiv_id":"2605.22154","citing_title":"IdleSpec: Exploiting Idle Time via Speculative Planning for LLM Agents","ref_index":2,"is_internal_anchor":false},{"citing_arxiv_id":"2605.20477","citing_title":"Training Language Agents to Learn from Experience","ref_index":10,"is_internal_anchor":false},{"citing_arxiv_id":"2605.16233","citing_title":"FORGE: Self-Evolving Agent Memory With No Weight Updates via Population Broadcast","ref_index":14,"is_internal_anchor":false},{"citing_arxiv_id":"2605.17187","citing_title":"PluRule: A Benchmark for Moderating Pluralistic Communities on Social Media","ref_index":43,"is_internal_anchor":false},{"citing_arxiv_id":"2512.17052","citing_title":"Dynamic Tool Dependency Retrieval for Lightweight Function Calling","ref_index":28,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/62X4MDFD3NNTNLZHONFX5UTPOK","json":"https://pith.science/pith/62X4MDFD3NNTNLZHONFX5UTPOK.json","graph_json":"https://pith.science/api/pith-number/62X4MDFD3NNTNLZHONFX5UTPOK/graph.json","events_json":"https://pith.science/api/pith-number/62X4MDFD3NNTNLZHONFX5UTPOK/events.json","paper":"https://pith.science/paper/62X4MDFD"},"agent_actions":{"view_html":"https://pith.science/pith/62X4MDFD3NNTNLZHONFX5UTPOK","download_json":"https://pith.science/pith/62X4MDFD3NNTNLZHONFX5UTPOK.json","view_paper":"https://pith.science/paper/62X4MDFD","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2505.00234&json=true","fetch_graph":"https://pith.science/api/pith-number/62X4MDFD3NNTNLZHONFX5UTPOK/graph.json","fetch_events":"https://pith.science/api/pith-number/62X4MDFD3NNTNLZHONFX5UTPOK/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/62X4MDFD3NNTNLZHONFX5UTPOK/action/timestamp_anchor","attest_storage":"https://pith.science/pith/62X4MDFD3NNTNLZHONFX5UTPOK/action/storage_attestation","attest_author":"https://pith.science/pith/62X4MDFD3NNTNLZHONFX5UTPOK/action/author_attestation","sign_citation":"https://pith.science/pith/62X4MDFD3NNTNLZHONFX5UTPOK/action/citation_signature","submit_replication":"https://pith.science/pith/62X4MDFD3NNTNLZHONFX5UTPOK/action/replication_record"}},"created_at":"2026-07-05T11:04:22.059842+00:00","updated_at":"2026-07-05T11:04:22.059842+00:00"}