{"paper":{"title":"GSM-SEM: Benchmark and Framework for Generating Semantically Variant Augmentations","license":"http://creativecommons.org/licenses/by/4.0/","headline":"GSM-SEM creates fresh math problem variants by changing facts and entities while preserving the original answers and calculations.","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Amit Agarwal, Aziza Mirsaidova, Dan Roth, Fang Tu, Graham Horwood, Hitesh Laxmichand Patel, Jyotika Singh, Karan Dua, Miguel Ballesteros, Sandip Ghoshal, Sujith Ravi, Tao Sheng, Weiyi Sun, Yassine Benajiba","submitted_at":"2026-05-08T00:02:39Z","abstract_excerpt":"Benchmarks like GSM8K are popular measures of mathematical reasoning, but leaderboard gains can overstate true capability due to memorization of fixed test sets. Most robustness variants apply surface-level perturbations (paraphrases, renamings, number swaps, distractors) that largely preserve the underlying facts, and static releases can themselves become memorization targets over time. We introduce GSM-SEM, a reusable and stochastic framework for generating semantically diverse benchmark variants with substantially higher semantic variance than prior approaches. GSM-SEM perturbs problem stat"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"Evaluating 14 SOTA LLMs, we observe consistent performance drops with larger decline when semantic perturbations are coupled with symbolic/plus variations (average drop rate 28% in maximum strictness configuration of GSM-SEM).","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That modifications to entities, attributes, and relationships can be constrained to alter underlying facts and require recomputation while still preserving the original calculations, answer, and approximate problem difficulty.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"GSM-SEM generates reusable, stochastic semantic variants of math reasoning benchmarks that alter underlying facts but preserve answers, producing larger LLM performance drops than prior surface-level variants.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"GSM-SEM creates fresh math problem variants by changing facts and entities while preserving the original answers and calculations.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"15b0f1c6e3ece1c4cb400345eec721233817542048c569fdbcffe98d34e965fd"},"source":{"id":"2605.07053","kind":"arxiv","version":2},"verdict":{"id":"44d2e8d9-c7c6-436a-af5a-07e3a9f48aeb","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-11T00:50:17.185012Z","strongest_claim":"Evaluating 14 SOTA LLMs, we observe consistent performance drops with larger decline when semantic perturbations are coupled with symbolic/plus variations (average drop rate 28% in maximum strictness configuration of GSM-SEM).","one_line_summary":"GSM-SEM generates reusable, stochastic semantic variants of math reasoning benchmarks that alter underlying facts but preserve answers, producing larger LLM performance drops than prior surface-level variants.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That modifications to entities, attributes, and relationships can be constrained to alter underlying facts and require recomputation while still preserving the original calculations, answer, and approximate problem difficulty.","pith_extraction_headline":"GSM-SEM creates fresh math problem variants by changing facts and entities while preserving the original answers and calculations."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2605.07053/integrity.json","findings":[],"available":true,"detectors_run":[{"name":"claim_evidence","ran_at":"2026-05-20T11:42:03.403367Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"ai_meta_artifact","ran_at":"2026-05-20T06:37:24.207279Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_title_agreement","ran_at":"2026-05-19T17:31:18.786006Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_compliance","ran_at":"2026-05-19T12:09:47.524745Z","status":"completed","version":"1.0.0","findings_count":0}],"snapshot_sha256":"aea27fb16682fe46102a92d2c37c4ef54db18829159f8a0ade344cb7eca492d6"},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","internal_anchors":0},"formal_canon":{"evidence_count":2,"snapshot_sha256":"6bbc3aa22f9ffe389ed7d57dd04c4976bd7364b04bd0a541b3ee2026542888cd"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}