{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:5J4277FJO3LYJBBFVNWGDZB7DO","short_pith_number":"pith:5J4277FJ","schema_version":"1.0","canonical_sha256":"ea79affca976d7848425ab6c61e43f1bb3281ff46fe24501cb5d7b09ddb12c55","source":{"kind":"arxiv","id":"2506.09038","version":1},"attestation_state":"computed","paper":{"title":"AbstentionBench: Reasoning LLMs Fail on Unanswerable Questions","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.AI","authors_text":"Kamalika Chaudhuri, Mark Ibrahim, Polina Kirichenko, Samuel J. Bell","submitted_at":"2025-06-10T17:57:30Z","abstract_excerpt":"For Large Language Models (LLMs) to be reliably deployed in both everyday and high-stakes domains, knowing when not to answer is equally critical as answering correctly. Real-world user queries, which can be underspecified, ill-posed, or fundamentally unanswerable, require LLMs to reason about uncertainty and selectively abstain -- i.e., refuse to answer definitively. However, abstention remains understudied, without a systematic evaluation framework for modern LLMs. In this work, we introduce AbstentionBench, a large-scale benchmark for holistically evaluating abstention across 20 diverse dat"},"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":"2506.09038","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.AI","submitted_at":"2025-06-10T17:57:30Z","cross_cats_sorted":[],"title_canon_sha256":"78ee2fdf094f88ac736853006dbc14a0e87178f3e2d4740f0231fd88c4f31f7f","abstract_canon_sha256":"eb053ef42e49bcb74b9f35ac96a1605dd4e8baaf4fed3d75b820f7ff8c6085a6"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:19:16.899136Z","signature_b64":"p/mHNqo6Lyjus3/FZsgjkWs9pDdOUSfEMUS27+BhltVeRwfDretFz6cEmnUrCPmNy5EUatOrKvaIs3yKnel/Ag==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"ea79affca976d7848425ab6c61e43f1bb3281ff46fe24501cb5d7b09ddb12c55","last_reissued_at":"2026-07-05T11:19:16.898646Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:19:16.898646Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"AbstentionBench: Reasoning LLMs Fail on Unanswerable Questions","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.AI","authors_text":"Kamalika Chaudhuri, Mark Ibrahim, Polina Kirichenko, Samuel J. Bell","submitted_at":"2025-06-10T17:57:30Z","abstract_excerpt":"For Large Language Models (LLMs) to be reliably deployed in both everyday and high-stakes domains, knowing when not to answer is equally critical as answering correctly. Real-world user queries, which can be underspecified, ill-posed, or fundamentally unanswerable, require LLMs to reason about uncertainty and selectively abstain -- i.e., refuse to answer definitively. However, abstention remains understudied, without a systematic evaluation framework for modern LLMs. In this work, we introduce AbstentionBench, a large-scale benchmark for holistically evaluating abstention across 20 diverse dat"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2506.09038","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/2506.09038/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":"2506.09038","created_at":"2026-07-05T11:19:16.898706+00:00"},{"alias_kind":"arxiv_version","alias_value":"2506.09038v1","created_at":"2026-07-05T11:19:16.898706+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2506.09038","created_at":"2026-07-05T11:19:16.898706+00:00"},{"alias_kind":"pith_short_12","alias_value":"5J4277FJO3LY","created_at":"2026-07-05T11:19:16.898706+00:00"},{"alias_kind":"pith_short_16","alias_value":"5J4277FJO3LYJBBF","created_at":"2026-07-05T11:19:16.898706+00:00"},{"alias_kind":"pith_short_8","alias_value":"5J4277FJ","created_at":"2026-07-05T11:19:16.898706+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":29,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.23716","citing_title":"Legal Reasoning Is Not Lawyering: Rethinking Legal Benchmarks for Pro Se Access to Justice","ref_index":15,"is_internal_anchor":false},{"citing_arxiv_id":"2606.13649","citing_title":"Operadic consistency: a label-free signal for compositional reasoning failures in LLMs","ref_index":46,"is_internal_anchor":false},{"citing_arxiv_id":"2607.01612","citing_title":"Scaling with Confidence: Calibrating Confidence of LLMs for Adaptive Test Time Scaling","ref_index":29,"is_internal_anchor":false},{"citing_arxiv_id":"2606.02965","citing_title":"Designing for Doubt: The Case for Informed Abstention in Autonomous Agents","ref_index":10,"is_internal_anchor":false},{"citing_arxiv_id":"2606.07539","citing_title":"Prompt Governance? On Governing Technologies Governed by Natural Language","ref_index":166,"is_internal_anchor":false},{"citing_arxiv_id":"2605.10379","citing_title":"Not All Proofs Are Equal: Evaluating LLM Proof Quality Beyond Correctness","ref_index":23,"is_internal_anchor":false},{"citing_arxiv_id":"2605.15000","citing_title":"Quantifying and Mitigating Premature Closure in Frontier LLMs","ref_index":17,"is_internal_anchor":false},{"citing_arxiv_id":"2606.28733","citing_title":"Agentic Abstention: Do Agents Know When to Stop Instead of Act?","ref_index":19,"is_internal_anchor":false},{"citing_arxiv_id":"2605.28070","citing_title":"Bridging the Detection-to-Abstention Gap in Reasoning Models under Insufficient Information","ref_index":17,"is_internal_anchor":false},{"citing_arxiv_id":"2606.00869","citing_title":"Enhancing LLM Metacognition via Cognitive Pairwise Training","ref_index":35,"is_internal_anchor":false},{"citing_arxiv_id":"2605.07919","citing_title":"MedVIGIL: Evaluating Trustworthy Medical VLMs Under Broken Visual Evidence","ref_index":14,"is_internal_anchor":false},{"citing_arxiv_id":"2603.22161","citing_title":"Causal Evidence that Language Models use Confidence to Drive Behavior","ref_index":9,"is_internal_anchor":false},{"citing_arxiv_id":"2605.00674","citing_title":"Beyond Benchmarks: MathArena as an Evaluation Platform for Mathematics with LLMs","ref_index":64,"is_internal_anchor":false},{"citing_arxiv_id":"2506.23040","citing_title":"Treatment, evidence, imitation, and chat","ref_index":55,"is_internal_anchor":false},{"citing_arxiv_id":"2509.21882","citing_title":"Position: The Hidden Costs and Measurement Gaps of Reinforcement Learning with Verifiable Rewards","ref_index":13,"is_internal_anchor":false},{"citing_arxiv_id":"2603.09117","citing_title":"Decoupling Reasoning and Confidence: Resurrecting Calibration in Reinforcement Learning from Verifiable Rewards","ref_index":11,"is_internal_anchor":false},{"citing_arxiv_id":"2604.14170","citing_title":"Stateful Evidence-Driven Retrieval-Augmented Generation with Iterative Reasoning","ref_index":21,"is_internal_anchor":false},{"citing_arxiv_id":"2604.03216","citing_title":"BAS: A Decision-Theoretic Approach to Evaluating Large Language Model Confidence","ref_index":29,"is_internal_anchor":false},{"citing_arxiv_id":"2605.10379","citing_title":"Not All Proofs Are Equal: Evaluating LLM Proof Quality Beyond Correctness","ref_index":23,"is_internal_anchor":false},{"citing_arxiv_id":"2605.01428","citing_title":"Hallucinations Undermine Trust; Metacognition is a Way Forward","ref_index":20,"is_internal_anchor":false},{"citing_arxiv_id":"2604.19895","citing_title":"Learning When Not to Decide: A Framework for Overcoming Factual Presumptuousness in AI Adjudication","ref_index":20,"is_internal_anchor":false},{"citing_arxiv_id":"2605.07919","citing_title":"MedVIGIL: Evaluating Trustworthy Medical VLMs Under Broken Visual Evidence","ref_index":14,"is_internal_anchor":false},{"citing_arxiv_id":"2605.07769","citing_title":"Coding Agents Don't Know When to Act","ref_index":7,"is_internal_anchor":false},{"citing_arxiv_id":"2605.05957","citing_title":"Knowing but Not Correcting: Routine Task Requests Suppress Factual Correction in LLMs","ref_index":14,"is_internal_anchor":false},{"citing_arxiv_id":"2605.02913","citing_title":"Generate, Filter, Control, Replay: A Comprehensive Survey of Rollout Strategies for LLM Reinforcement Learning","ref_index":60,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/5J4277FJO3LYJBBFVNWGDZB7DO","json":"https://pith.science/pith/5J4277FJO3LYJBBFVNWGDZB7DO.json","graph_json":"https://pith.science/api/pith-number/5J4277FJO3LYJBBFVNWGDZB7DO/graph.json","events_json":"https://pith.science/api/pith-number/5J4277FJO3LYJBBFVNWGDZB7DO/events.json","paper":"https://pith.science/paper/5J4277FJ"},"agent_actions":{"view_html":"https://pith.science/pith/5J4277FJO3LYJBBFVNWGDZB7DO","download_json":"https://pith.science/pith/5J4277FJO3LYJBBFVNWGDZB7DO.json","view_paper":"https://pith.science/paper/5J4277FJ","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2506.09038&json=true","fetch_graph":"https://pith.science/api/pith-number/5J4277FJO3LYJBBFVNWGDZB7DO/graph.json","fetch_events":"https://pith.science/api/pith-number/5J4277FJO3LYJBBFVNWGDZB7DO/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/5J4277FJO3LYJBBFVNWGDZB7DO/action/timestamp_anchor","attest_storage":"https://pith.science/pith/5J4277FJO3LYJBBFVNWGDZB7DO/action/storage_attestation","attest_author":"https://pith.science/pith/5J4277FJO3LYJBBFVNWGDZB7DO/action/author_attestation","sign_citation":"https://pith.science/pith/5J4277FJO3LYJBBFVNWGDZB7DO/action/citation_signature","submit_replication":"https://pith.science/pith/5J4277FJO3LYJBBFVNWGDZB7DO/action/replication_record"}},"created_at":"2026-07-05T11:19:16.898706+00:00","updated_at":"2026-07-05T11:19:16.898706+00:00"}