{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:CUUG3JOF7OUH3BZXF3JRYTLVTW","short_pith_number":"pith:CUUG3JOF","schema_version":"1.0","canonical_sha256":"15286da5c5fba87d87372ed31c4d759d85572845495cc4aacad4e826b7462829","source":{"kind":"arxiv","id":"2505.11423","version":3},"attestation_state":"computed","paper":{"title":"When Thinking Fails: The Pitfalls of Reasoning for Instruction-Following in LLMs","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Anurag Beniwal, Narayanan Sadagopan, Xiaomin Li, Xupeng Chen, Yingying Zhuang, Zhiwei Zhang, Zhou Yu, Ziji Zhang","submitted_at":"2025-05-16T16:36:00Z","abstract_excerpt":"Reasoning-enhanced large language models (RLLMs), whether explicitly trained for reasoning or prompted via chain-of-thought (CoT), have achieved state-of-the-art performance on many complex reasoning tasks. However, we uncover a surprising and previously overlooked phenomenon: explicit CoT reasoning can significantly degrade instruction-following accuracy. Evaluating 15 models on two benchmarks: IFEval (with simple, rule-verifiable constraints) and ComplexBench (with complex, compositional constraints), we consistently observe performance drops when CoT prompting is applied. Through large-scal"},"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":"2505.11423","kind":"arxiv","version":3},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2025-05-16T16:36:00Z","cross_cats_sorted":[],"title_canon_sha256":"bdf3c45c8d929aa509ee93928ee36a7e40e74e875ad49a8f35bb597d140bc507","abstract_canon_sha256":"cb9dfdbbdc508f72dc0784370d0d0192622ce989406172e923b98917eec5ffeb"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T12:02:46.286893Z","signature_b64":"QTew13WVsyLWCK/GiZXy8I+hsVlN6YPIaVAHWbBdB4H4O5FeQm1lTytmfJD7E/uHsisYG7UzkgJQwEVYZBNvBg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"15286da5c5fba87d87372ed31c4d759d85572845495cc4aacad4e826b7462829","last_reissued_at":"2026-07-05T12:02:46.286273Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T12:02:46.286273Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"When Thinking Fails: The Pitfalls of Reasoning for Instruction-Following in LLMs","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Anurag Beniwal, Narayanan Sadagopan, Xiaomin Li, Xupeng Chen, Yingying Zhuang, Zhiwei Zhang, Zhou Yu, Ziji Zhang","submitted_at":"2025-05-16T16:36:00Z","abstract_excerpt":"Reasoning-enhanced large language models (RLLMs), whether explicitly trained for reasoning or prompted via chain-of-thought (CoT), have achieved state-of-the-art performance on many complex reasoning tasks. However, we uncover a surprising and previously overlooked phenomenon: explicit CoT reasoning can significantly degrade instruction-following accuracy. Evaluating 15 models on two benchmarks: IFEval (with simple, rule-verifiable constraints) and ComplexBench (with complex, compositional constraints), we consistently observe performance drops when CoT prompting is applied. Through large-scal"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2505.11423","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/2505.11423/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":"2505.11423","created_at":"2026-07-05T12:02:46.286386+00:00"},{"alias_kind":"arxiv_version","alias_value":"2505.11423v3","created_at":"2026-07-05T12:02:46.286386+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2505.11423","created_at":"2026-07-05T12:02:46.286386+00:00"},{"alias_kind":"pith_short_12","alias_value":"CUUG3JOF7OUH","created_at":"2026-07-05T12:02:46.286386+00:00"},{"alias_kind":"pith_short_16","alias_value":"CUUG3JOF7OUH3BZX","created_at":"2026-07-05T12:02:46.286386+00:00"},{"alias_kind":"pith_short_8","alias_value":"CUUG3JOF","created_at":"2026-07-05T12:02:46.286386+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":12,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.12689","citing_title":"Observable Patterns Are Not Explanations: A Causal-Geometric Analysis of Latent Reasoning Models","ref_index":44,"is_internal_anchor":false},{"citing_arxiv_id":"2606.09662","citing_title":"When Built-in Thinking Helps and Hurts: Constraint-Level Error Shifts in Instruction Following","ref_index":5,"is_internal_anchor":false},{"citing_arxiv_id":"2606.07539","citing_title":"Prompt Governance? On Governing Technologies Governed by Natural Language","ref_index":192,"is_internal_anchor":false},{"citing_arxiv_id":"2605.22211","citing_title":"CLORE: Content-Level Optimization for Reasoning Efficiency","ref_index":28,"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":16,"is_internal_anchor":false},{"citing_arxiv_id":"2509.21743","citing_title":"Retrieval-of-Thought: Efficient Reasoning via Reusing Thoughts","ref_index":14,"is_internal_anchor":false},{"citing_arxiv_id":"2509.08827","citing_title":"A Survey of Reinforcement Learning for Large Reasoning Models","ref_index":285,"is_internal_anchor":false},{"citing_arxiv_id":"2506.06941","citing_title":"The Illusion of Thinking: Understanding the Strengths and Limitations of Reasoning Models via the Lens of Problem Complexity","ref_index":50,"is_internal_anchor":false},{"citing_arxiv_id":"2604.03679","citing_title":"LightThinker++: From Reasoning Compression to Memory Management","ref_index":41,"is_internal_anchor":false},{"citing_arxiv_id":"2605.05716","citing_title":"More Is Not Always Better: Cross-Component Interference in LLM Agent Scaffolding","ref_index":16,"is_internal_anchor":false},{"citing_arxiv_id":"2605.05678","citing_title":"Chain of Risk: Safety Failures in Large Reasoning Models and Mitigation via Adaptive Multi-Principle Steering","ref_index":20,"is_internal_anchor":false},{"citing_arxiv_id":"2605.05687","citing_title":"DataDignity: Training Data Attribution for Large Language Models","ref_index":7,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/CUUG3JOF7OUH3BZXF3JRYTLVTW","json":"https://pith.science/pith/CUUG3JOF7OUH3BZXF3JRYTLVTW.json","graph_json":"https://pith.science/api/pith-number/CUUG3JOF7OUH3BZXF3JRYTLVTW/graph.json","events_json":"https://pith.science/api/pith-number/CUUG3JOF7OUH3BZXF3JRYTLVTW/events.json","paper":"https://pith.science/paper/CUUG3JOF"},"agent_actions":{"view_html":"https://pith.science/pith/CUUG3JOF7OUH3BZXF3JRYTLVTW","download_json":"https://pith.science/pith/CUUG3JOF7OUH3BZXF3JRYTLVTW.json","view_paper":"https://pith.science/paper/CUUG3JOF","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2505.11423&json=true","fetch_graph":"https://pith.science/api/pith-number/CUUG3JOF7OUH3BZXF3JRYTLVTW/graph.json","fetch_events":"https://pith.science/api/pith-number/CUUG3JOF7OUH3BZXF3JRYTLVTW/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/CUUG3JOF7OUH3BZXF3JRYTLVTW/action/timestamp_anchor","attest_storage":"https://pith.science/pith/CUUG3JOF7OUH3BZXF3JRYTLVTW/action/storage_attestation","attest_author":"https://pith.science/pith/CUUG3JOF7OUH3BZXF3JRYTLVTW/action/author_attestation","sign_citation":"https://pith.science/pith/CUUG3JOF7OUH3BZXF3JRYTLVTW/action/citation_signature","submit_replication":"https://pith.science/pith/CUUG3JOF7OUH3BZXF3JRYTLVTW/action/replication_record"}},"created_at":"2026-07-05T12:02:46.286386+00:00","updated_at":"2026-07-05T12:02:46.286386+00:00"}