{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:NVLQFEDOYIFGOVRTZ3TC6RKXKW","short_pith_number":"pith:NVLQFEDO","schema_version":"1.0","canonical_sha256":"6d5702906ec20a675633cee62f455755855de36d2d05c2f25217f7ef50bf916d","source":{"kind":"arxiv","id":"2504.16078","version":1},"attestation_state":"computed","paper":{"title":"LLMs are Greedy Agents: Effects of RL Fine-tuning on Decision-Making Abilities","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Jordi Grau-Moya, J\\\"org Bornschein, Markus Wulfmeier, Razvan Pascanu, Thomas Schmied","submitted_at":"2025-04-22T17:57:14Z","abstract_excerpt":"The success of Large Language Models (LLMs) has sparked interest in various agentic applications. A key hypothesis is that LLMs, leveraging common sense and Chain-of-Thought (CoT) reasoning, can effectively explore and efficiently solve complex domains. However, LLM agents have been found to suffer from sub-optimal exploration and the knowing-doing gap, the inability to effectively act on knowledge present in the model. In this work, we systematically study why LLMs perform sub-optimally in decision-making scenarios. In particular, we closely examine three prevalent failure modes: greediness, "},"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":"2504.16078","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-04-22T17:57:14Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"362e3f0afc355fcd5e9484e2d8d324d6bcec6f10f2bb2f631f77d2f89d178584","abstract_canon_sha256":"fae69e75ceb51fae2914706b108509325b5530e06dd6dda4a655686084ca4d3a"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:52:35.153539Z","signature_b64":"yj/MkBxa/vukIwkS+wyj5lwDkfmqM91R/0wmgv6ebDKXNq7tWWMgMv85Pk8IWYY38wBayOjyjzZZb9G+slvaCQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"6d5702906ec20a675633cee62f455755855de36d2d05c2f25217f7ef50bf916d","last_reissued_at":"2026-07-05T10:52:35.153082Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:52:35.153082Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"LLMs are Greedy Agents: Effects of RL Fine-tuning on Decision-Making Abilities","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Jordi Grau-Moya, J\\\"org Bornschein, Markus Wulfmeier, Razvan Pascanu, Thomas Schmied","submitted_at":"2025-04-22T17:57:14Z","abstract_excerpt":"The success of Large Language Models (LLMs) has sparked interest in various agentic applications. A key hypothesis is that LLMs, leveraging common sense and Chain-of-Thought (CoT) reasoning, can effectively explore and efficiently solve complex domains. However, LLM agents have been found to suffer from sub-optimal exploration and the knowing-doing gap, the inability to effectively act on knowledge present in the model. In this work, we systematically study why LLMs perform sub-optimally in decision-making scenarios. In particular, we closely examine three prevalent failure modes: greediness, "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2504.16078","kind":"arxiv","version":1},"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/2504.16078/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":"2504.16078","created_at":"2026-07-05T10:52:35.153141+00:00"},{"alias_kind":"arxiv_version","alias_value":"2504.16078v1","created_at":"2026-07-05T10:52:35.153141+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2504.16078","created_at":"2026-07-05T10:52:35.153141+00:00"},{"alias_kind":"pith_short_12","alias_value":"NVLQFEDOYIFG","created_at":"2026-07-05T10:52:35.153141+00:00"},{"alias_kind":"pith_short_16","alias_value":"NVLQFEDOYIFGOVRT","created_at":"2026-07-05T10:52:35.153141+00:00"},{"alias_kind":"pith_short_8","alias_value":"NVLQFEDO","created_at":"2026-07-05T10:52:35.153141+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2508.13654","citing_title":"Input-Time Scaling: Adding Noise and Irrelevance into Less-Is-More Drastically Improves Reasoning Performance and Efficiency","ref_index":10,"is_internal_anchor":false},{"citing_arxiv_id":"2604.05859","citing_title":"When Do We Need LLMs? A Diagnostic for Language-Driven Bandits","ref_index":41,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/NVLQFEDOYIFGOVRTZ3TC6RKXKW","json":"https://pith.science/pith/NVLQFEDOYIFGOVRTZ3TC6RKXKW.json","graph_json":"https://pith.science/api/pith-number/NVLQFEDOYIFGOVRTZ3TC6RKXKW/graph.json","events_json":"https://pith.science/api/pith-number/NVLQFEDOYIFGOVRTZ3TC6RKXKW/events.json","paper":"https://pith.science/paper/NVLQFEDO"},"agent_actions":{"view_html":"https://pith.science/pith/NVLQFEDOYIFGOVRTZ3TC6RKXKW","download_json":"https://pith.science/pith/NVLQFEDOYIFGOVRTZ3TC6RKXKW.json","view_paper":"https://pith.science/paper/NVLQFEDO","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2504.16078&json=true","fetch_graph":"https://pith.science/api/pith-number/NVLQFEDOYIFGOVRTZ3TC6RKXKW/graph.json","fetch_events":"https://pith.science/api/pith-number/NVLQFEDOYIFGOVRTZ3TC6RKXKW/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/NVLQFEDOYIFGOVRTZ3TC6RKXKW/action/timestamp_anchor","attest_storage":"https://pith.science/pith/NVLQFEDOYIFGOVRTZ3TC6RKXKW/action/storage_attestation","attest_author":"https://pith.science/pith/NVLQFEDOYIFGOVRTZ3TC6RKXKW/action/author_attestation","sign_citation":"https://pith.science/pith/NVLQFEDOYIFGOVRTZ3TC6RKXKW/action/citation_signature","submit_replication":"https://pith.science/pith/NVLQFEDOYIFGOVRTZ3TC6RKXKW/action/replication_record"}},"created_at":"2026-07-05T10:52:35.153141+00:00","updated_at":"2026-07-05T10:52:35.153141+00:00"}