{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:6YL7BNQ6AV743BINWLSV3465JE","short_pith_number":"pith:6YL7BNQ6","schema_version":"1.0","canonical_sha256":"f617f0b61e057fcd850db2e55df3dd4925c45ef73e74eb1d31e868663ad06475","source":{"kind":"arxiv","id":"2504.13145","version":2},"attestation_state":"computed","paper":{"title":"Exploring Expert Failures Improves LLM Agent Tuning","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.AI","authors_text":"Andrew Bai, Cho-Jui Hsieh, Li-Cheng Lan, Minhao Cheng, Tianyi Zhou","submitted_at":"2025-04-17T17:53:54Z","abstract_excerpt":"Large Language Models (LLMs) have shown tremendous potential as agents, excelling at tasks that require multiple rounds of reasoning and interactions. Rejection Sampling Fine-Tuning (RFT) has emerged as an effective method for finetuning LLMs as agents: it first imitates expert-generated successful trajectories and further improves agentic skills through iterative fine-tuning on successful, self-generated trajectories. However, since the expert (e.g., GPT-4) succeeds primarily on simpler subtasks and RFT inherently favors simpler scenarios, many complex subtasks remain unsolved and persistentl"},"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":"2504.13145","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.AI","submitted_at":"2025-04-17T17:53:54Z","cross_cats_sorted":[],"title_canon_sha256":"63f3f00a6cc5ad9fb20b607b4f75a7c12325814c7d17ff0974aeeba163680542","abstract_canon_sha256":"c06df822032fa850bff1f1ccfde40638de42449557e434efcdce3e5a9e007372"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:51:06.713706Z","signature_b64":"NBZ3KmTUPM4DdlJYmCJCthh8s8gHuJ1Xi55v8KsyUx640CAbVdRNfX9eERVahgC5IqwL7mG/ZSHAB0iGVhkNBg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"f617f0b61e057fcd850db2e55df3dd4925c45ef73e74eb1d31e868663ad06475","last_reissued_at":"2026-07-05T10:51:06.713110Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:51:06.713110Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Exploring Expert Failures Improves LLM Agent Tuning","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.AI","authors_text":"Andrew Bai, Cho-Jui Hsieh, Li-Cheng Lan, Minhao Cheng, Tianyi Zhou","submitted_at":"2025-04-17T17:53:54Z","abstract_excerpt":"Large Language Models (LLMs) have shown tremendous potential as agents, excelling at tasks that require multiple rounds of reasoning and interactions. Rejection Sampling Fine-Tuning (RFT) has emerged as an effective method for finetuning LLMs as agents: it first imitates expert-generated successful trajectories and further improves agentic skills through iterative fine-tuning on successful, self-generated trajectories. However, since the expert (e.g., GPT-4) succeeds primarily on simpler subtasks and RFT inherently favors simpler scenarios, many complex subtasks remain unsolved and persistentl"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2504.13145","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/2504.13145/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":"2504.13145","created_at":"2026-07-05T10:51:06.713175+00:00"},{"alias_kind":"arxiv_version","alias_value":"2504.13145v2","created_at":"2026-07-05T10:51:06.713175+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2504.13145","created_at":"2026-07-05T10:51:06.713175+00:00"},{"alias_kind":"pith_short_12","alias_value":"6YL7BNQ6AV74","created_at":"2026-07-05T10:51:06.713175+00:00"},{"alias_kind":"pith_short_16","alias_value":"6YL7BNQ6AV743BIN","created_at":"2026-07-05T10:51:06.713175+00:00"},{"alias_kind":"pith_short_8","alias_value":"6YL7BNQ6","created_at":"2026-07-05T10:51:06.713175+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2505.18121","citing_title":"ProgRM: Build Better GUI Agents with Progress Rewards","ref_index":8,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/6YL7BNQ6AV743BINWLSV3465JE","json":"https://pith.science/pith/6YL7BNQ6AV743BINWLSV3465JE.json","graph_json":"https://pith.science/api/pith-number/6YL7BNQ6AV743BINWLSV3465JE/graph.json","events_json":"https://pith.science/api/pith-number/6YL7BNQ6AV743BINWLSV3465JE/events.json","paper":"https://pith.science/paper/6YL7BNQ6"},"agent_actions":{"view_html":"https://pith.science/pith/6YL7BNQ6AV743BINWLSV3465JE","download_json":"https://pith.science/pith/6YL7BNQ6AV743BINWLSV3465JE.json","view_paper":"https://pith.science/paper/6YL7BNQ6","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2504.13145&json=true","fetch_graph":"https://pith.science/api/pith-number/6YL7BNQ6AV743BINWLSV3465JE/graph.json","fetch_events":"https://pith.science/api/pith-number/6YL7BNQ6AV743BINWLSV3465JE/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/6YL7BNQ6AV743BINWLSV3465JE/action/timestamp_anchor","attest_storage":"https://pith.science/pith/6YL7BNQ6AV743BINWLSV3465JE/action/storage_attestation","attest_author":"https://pith.science/pith/6YL7BNQ6AV743BINWLSV3465JE/action/author_attestation","sign_citation":"https://pith.science/pith/6YL7BNQ6AV743BINWLSV3465JE/action/citation_signature","submit_replication":"https://pith.science/pith/6YL7BNQ6AV743BINWLSV3465JE/action/replication_record"}},"created_at":"2026-07-05T10:51:06.713175+00:00","updated_at":"2026-07-05T10:51:06.713175+00:00"}