{"record": {"aliases": [], "artifacts": [{"id": "arxiv:2608.28478", "kind": "paper", "url": "https://arxiv.org/abs/2608.28478"}, {"id": "gh:tencent/elephantbench", "kind": "repo", "url": "https://github.com/Tencent/ElephantBench"}], "authors": ["Zhuoshi Pan", "Junru Lu", "Yan Qian", "H. Vicky Zhao", "Di Yin", "Xing Sun"], "authors_reference": {"basis": "arxiv_metadata", "retrieved_at": "2026-11-03", "source_key": "claire-radar:2608.28478", "source_url": "https://arxiv.org/abs/2608.28478"}, "categories": ["Factuality", "General AI", "Language & Knowledge", "cs.CL"], "description": {"en": "Closed-book knowledge probe with 1,094 source-traceable questions that measure whether language models recall multiple verified accounts of long-tail facts, scored via complete, partial, or failed recall rates."}, "documents": [{"document_type": "registry_page", "id": "claire_radar:https://arxiv.org/abs/2608.28478", "source": "claire_radar", "source_url": "https://arxiv.org/abs/2608.28478", "title": "ElephantBench"}], "evidence_summary": {"document_count": 1, "model_count": null, "model_count_basis": null}, "institutions": ["Tencent (China)", "Tsinghua University", "University of Warwick"], "institutions_reference": {"basis": "openalex_linked_institutions", "doi": "10.48550/arxiv.2608.28478", "retrieved_at": "2026-11-04", "source_key": "claire-radar:2608.28478", "source_url": "https://openalex.org/W7204847523"}, "key": "claire-radar:2608.28478", "modality": null, "name": "ElephantBench", "openness": {"code_license": null, "data_license": null, "evidence": [], "status": "unknown"}, "provenance": {"confidence": "Medium", "crawl_bundle": "claire_radar_2026-10-14", "crawled_at": "2026-10-14T09:44:51Z", "data_status": "primary-source-indexed", "display_eligible": "true", "origin_record_id": "bm_elephantbench_79ce069e", "origin_source": "arxiv", "recognition_confidence": "0.6", "record_sha256": "7ca4a318da4b96323ef3875ce6dcc778652fcc5342d6497ac1064a5b6d4f8b45", "relation": "introduces", "review_model": "deepseek-v4-pro", "review_state": "ai-reviewed", "reviewed_at": "2026-09-06T08:46:12.648160Z", "source_url": "https://arxiv.org/abs/2608.28478"}, "publisher": null, "released": "2026-08-28", "released_reference": {"basis": "benchmark_release", "source_key": "claire-radar:2608.28478", "source_url": "https://arxiv.org/abs/2608.28478"}, "schema_version": 1, "sizes": [], "slug": "claire-radar-2608-28478", "source": "claire_radar", "source_benchmark_id": "2608.28478", "source_metadata": {"claire_radar": {"adoption30d": null, "aliases": [], "annotation": "Unknown", "applicationDomains": ["General AI"], "area": "Language & Knowledge", "attention": {"asOf": "2026-09-06", "attemptedAt": "2026-09-06T01:30:14.942019Z", "githubRepo": "https://github.com/Tencent/ElephantBench", "githubScope": "benchmark_repo", "githubStars": 5, "hfDailySubmittedAt": "2026-08-31T00:00:00.000Z", "hfDataset": null, "hfDatasetDownloads": null, "hfDatasetLikes": null, "hfPaperUpvotes": 17, "hfPaperUrl": "https://huggingface.co/papers/2608.28478", "observedAt": "2026-09-06T01:30:14.942019Z", "signalStatus": {"githubStars": {"observedAt": "2026-09-06T01:30:14.942019Z", "state": "fresh"}, "hfDatasetDownloads": {"state": "not_applicable"}, "hfDatasetLikes": {"state": "not_applicable"}, "hfPaperUpvotes": {"observedAt": "2026-09-06T01:30:14.942019Z", "state": "fresh"}}, "status": "fresh"}, "attentionForecast": {"confidence": "Medium", "horizon": "7d", "reason": "The benchmark targets a timely issue in long-tail knowledge and comes with a complete public toolkit, though the topic is specialized and may attract moderate attention.", "score": 48}, "capabilities": ["Factuality"], "confidence": "Medium", "construction": "Unknown", "constructionDetail": "Unknown — the indexer does not infer construction details without explicit source evidence.", "copyGeneration": {"generatedAt": "2026-09-06T08:46:12.648160Z", "inputHash": "8e8d2c3d804ddb2736cfdbb8a87728dc2844ce10c30f93b248930d2d507204b0", "model": "deepseek-v4-pro"}, "curation": {"canonicalNameEvidence": "These findings establish ElephantBench as a reproducible knowledge probe for diagnosing epistemic myopia in parametric memory.", "canonicalNameSource": "abstract", "decisionReason": "The paper explicitly releases a fixed dataset and code, and the repository documents a stable scoring protocol with judge, scoring, and analysis commands, enabling external teams to reuse it.", "model": "deepseek-v4-pro", "reviewedAt": "2026-09-06T08:46:12.648160Z", "state": "ai-reviewed"}, "dataStatus": "primary-source-indexed", "demo": false, "description": "Closed-book knowledge probe with 1,094 source-traceable questions that measure whether language models recall multiple verified accounts of long-tail facts, scored via complete, partial, or failed recall rates.", "displayEligible": true, "domainCuration": {"method": "rules-v1", "state": "auto"}, "evaluationMode": "public_reusable", "evidence": {"reasonCodes": ["exact named benchmark artifact released in abstract", "public artifact URL"], "snippet": "To address this gap, we introduce ElephantBench, a closed-book knowledge probe comprising 1,094 questions generated through an auditable graph-based pipeline."}, "familyId": "bmf_80aadb10adc3", "firstSeenAt": "2026-08-31", "heat": null, "id": "bm_elephantbench_79ce069e", "indexedAt": "2026-08-31T01:03:30.163531Z", "industrySectors": [], "links": {"code": "https://github.com/Tencent/ElephantBench", "data": null, "paper": "https://arxiv.org/abs/2608.28478", "pdf": "https://arxiv.org/pdf/2608.28478", "project": null, "report": "https://arxiv.org/abs/2608.28478"}, "metrics": [], "motivation": "Factual question answering (QA) typically assumes a single canonical answer, obscuring whether large language models (LLMs) retain divergent accounts of long-tail facts.", "name": "ElephantBench", "oneLine": "Closed-book knowledge probe with 1,094 source-traceable questions that measure whether language models recall multiple verified accounts of long-tail facts, scored via complete, partial, or failed recall rates.", "paperTitle": "Blind Men and the Elephant: Probing the Epistemic Myopia of LLMs under Long-Tail Divergent Knowledge", "primaryDomain": "General AI", "publication": {"evidence": null, "evidenceLevel": "unverified", "evidenceUrl": "https://arxiv.org/abs/2608.28478", "source": "arxiv-metadata", "status": "unverified", "venue": null, "verifiedAt": "2026-08-31T01:03:30.163531Z"}, "publications": [], "publishers": [{"name": "Tencent", "organizationType": "company-research-lab", "role": "benchmark-publisher", "sourceUrl": "https://github.com/Tencent/ElephantBench"}], "ranking": {"30d": {"confidence": "High", "coverage": 0.85, "growth": {"components": {"githubStars": {"normalization": "percentile", "normalized": null, "percentile": null, "value": null}, "hfDatasetDownloads": {"normalization": "percentile", "normalized": null, "percentile": null, "value": null}, "hfPaperUpvotes": {"normalization": "percentile", "normalized": null, "percentile": null, "value": null}}, "confidence": "Low", "coverage": 0.0, "elapsedDays": null, "historyDate": null, "rank": null, "score": null, "windowDays": 30}, "level": {"components": {"githubStars": {"normalization": "log-max", "normalized": 0.2668623713452529, "value": 5}, "hfDatasetDownloads": {"normalization": "log-max", "normalized": null, "value": null}, "hfPaperUpvotes": {"normalization": "log-max", "normalized": 0.5119828085005611, "value": 17}}, "confidence": "High", "coverage": 0.85, "rank": 60, "score": 38}, "method": "attention-ranking-v13", "rank": 60, "score": 38}, "90d": {"confidence": "Medium", "coverage": 0.7, "growth": {"components": {"githubStars": {"normalization": "percentile", "normalized": null, "percentile": null, "value": null}, "hfDatasetDownloads": {"normalization": "percentile", "normalized": null, "percentile": null, "value": null}, "hfPaperUpvotes": {"normalization": "percentile", "normalized": null, "percentile": null, "value": null}}, "confidence": "Low", "coverage": 0.0, "elapsedDays": null, "historyDate": null, "rank": null, "score": null, "windowDays": 90}, "level": {"components": {"githubStars": {"normalization": "log-max", "normalized": 0.24241087969491565, "value": 5}, "hfDatasetDownloads": {"normalization": "log-max", "normalized": null, "value": null}, "hfPaperUpvotes": {"normalization": "log-max", "normalized": 0.5119828085005611, "value": 17}}, "confidence": "Medium", "coverage": 0.7, "rank": 180, "score": 36}, "method": "attention-ranking-v13", "rank": 180, "score": 36}}, "readiness": "Runnable", "recognitionConfidence": 0.6, "relation": "introduces", "releasedAt": "2026-08-28", "source": {"authors": ["Zhuoshi Pan", "Junru Lu", "Yan Qian", "H. Vicky Zhao", "Di Yin", "Xing Sun"], "categories": ["cs.CL"], "id": "2608.28478", "title": "Blind Men and the Elephant: Probing the Epistemic Myopia of LLMs under Long-Tail Divergent Knowledge", "type": "arxiv", "url": "https://arxiv.org/abs/2608.28478"}, "sourceUpdatedAt": "2026-08-28T16:06:01Z", "topics": ["cs.CL"], "venueAttempts": [], "whyItMatters": "Provides a reproducible way to diagnose when models over-rely on dominant accounts and omit valid divergent knowledge, supporting evaluation of epistemic rigor beyond single-answer QA."}}}, "schema_version": 1, "scores_by_source": {}, "siblings": []}
