{"paper":{"title":"Compliance versus Sensibility: On the Reasoning Controllability in Large Language Models","license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","headline":"Large language models prioritize sensible reasoning over following conflicting instructions, but can be steered toward greater compliance.","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Mahmud Elahi Akhter, Marco Valentino, Maria Liakata, Nikolaos Aletras, Xingwei Tan, Yuxiang Zhou","submitted_at":"2026-04-29T22:55:40Z","abstract_excerpt":"Large Language Models (LLMs) are known to acquire reasoning capabilities through shared inference patterns in pre-training data, which are further elicited via Chain-of-Thought (CoT) practices. However, whether fundamental reasoning patterns, such as induction, deduction, and abduction, can be decoupled from specific problem instances remains a critical challenge for model controllability, and for shedding light on reasoning controllability. In this paper, we present the first systematic investigation of this problem through the lens of reasoning conflicts: an explicit tension between parametr"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"LLMs consistently prioritize sensibility over compliance, favoring task-appropriate reasoning patterns despite conflicting instructions... we steer models towards compliance, increasing instruction following by up to 29%.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That the constructed reasoning conflicts cleanly isolate parametric versus contextual reasoning without introducing unintended task difficulty or prompt artifacts that could explain the observed sensibility bias.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"LLMs favor task-appropriate reasoning over conflicting instructions, yet reasoning types are linearly encoded in middle-to-late layers and can be steered to boost instruction compliance by up to 29%.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Large language models prioritize sensible reasoning over following conflicting instructions, but can be steered toward greater compliance.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"485dd301a8b6be064ff76a2150c0fe4a537b7e384f78613e84a94de67c21f35e"},"source":{"id":"2604.27251","kind":"arxiv","version":2},"verdict":{"id":"2aa421c6-4a4c-45d0-92e7-26aa07f7448b","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-07T08:53:28.722798Z","strongest_claim":"LLMs consistently prioritize sensibility over compliance, favoring task-appropriate reasoning patterns despite conflicting instructions... we steer models towards compliance, increasing instruction following by up to 29%.","one_line_summary":"LLMs favor task-appropriate reasoning over conflicting instructions, yet reasoning types are linearly encoded in middle-to-late layers and can be steered to boost instruction compliance by up to 29%.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That the constructed reasoning conflicts cleanly isolate parametric versus contextual reasoning without introducing unintended task difficulty or prompt artifacts that could explain the observed sensibility bias.","pith_extraction_headline":"Large language models prioritize sensible reasoning over following conflicting instructions, but can be steered toward greater compliance."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2604.27251/integrity.json","findings":[],"available":true,"detectors_run":[{"name":"ai_meta_artifact","ran_at":"2026-05-20T22:41:18.225841Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_compliance","ran_at":"2026-05-19T19:26:10.046970Z","status":"completed","version":"1.0.0","findings_count":0}],"snapshot_sha256":"362c1a78ebb1ce087d2afe74585c80abc2261d9afb47fe9eb8ff14c9799d8a0b"},"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"}