{"record": {"aliases": ["Are Text-to-Image Models Inductivist Turkeys? A Counterfactual Benchmark for Causal Reasoning"], "artifacts": [{"id": "arxiv:2606.24548", "kind": "paper", "url": "https://arxiv.org/abs/2606.24548"}, {"id": "gh:jylei16/cf-world.github.io", "kind": "repo", "url": "https://github.com/jylei16/CF-World.github.io"}], "authors": ["Jiayi Lei", "Yuandong Pu", "Xingyu Han", "Rongpeng Zhu", "Jing Xu", "Jinyao Wang", "Zijian Zhou", "Bin Fu", "Yuewen Cao", "Yihao Liu", "Hongsheng Li"], "authors_reference": {"basis": "arxiv_metadata", "retrieved_at": "2026-11-03", "source_key": "claire-radar:2606.24548", "source_url": "https://arxiv.org/abs/2606.24548"}, "categories": ["General AI", "Reasoning", "Vision & 3D"], "description": {"en": "A three-level benchmark evaluating whether text-to-image models can follow explicit and implicit counterfactual rules rather than defaulting to real-world visual priors."}, "documents": [{"document_type": "registry_page", "id": "claire_radar:https://arxiv.org/abs/2606.24548", "source": "claire_radar", "source_url": "https://arxiv.org/abs/2606.24548", "title": "Counterfactual-World (CF-World)"}], "evidence_summary": {"document_count": 1, "model_count": null, "model_count_basis": null}, "institutions": ["Shanghai Jiao Tong University", "Shanghai Artificial Intelligence Laboratory", "Chinese University of Hong Kong"], "institutions_reference": {"basis": "openalex_linked_institutions", "doi": "10.48550/arxiv.2606.24548", "retrieved_at": "2026-11-04", "source_key": "claire-radar:2606.24548", "source_url": "https://openalex.org/W7165745873"}, "key": "claire-radar:2606.24548", "modality": null, "name": "Counterfactual-World (CF-World)", "openness": {"code_license": null, "data_license": null, "evidence": [], "status": "unknown"}, "provenance": {"admission_policy_version": "2026-09-09.1", "confidence": "Medium", "crawl_bundle": "claire_radar_2026-10-14", "crawled_at": "2026-09-25T04:40:22Z", "data_status": "primary-source-indexed", "display_eligible": "true", "origin_record_id": "bm_are-text-to-image-models-inductivist-turke_acb79322", "origin_source": "arxiv", "recognition_confidence": "0.55", "record_sha256": "81e9292c877083b555d2e10854353831f5c59734ec94bbd53ac3b83aefd9afb4", "relation": "introduces", "review_model": "codex-cli-default", "review_state": "ai-reviewed", "reviewed_at": "2026-09-24T20:22:34.615233Z", "source_url": "https://arxiv.org/abs/2606.24548"}, "publisher": null, "released": "2026-06-23", "released_reference": {"basis": "benchmark_release", "source_key": "claire-radar:2606.24548", "source_url": "https://arxiv.org/abs/2606.24548"}, "schema_version": 1, "sizes": [], "slug": "claire-radar-2606-24548", "source": "claire_radar", "source_benchmark_id": "2606.24548", "source_metadata": {"claire_radar": {"adoption30d": null, "aliases": ["Are Text-to-Image Models Inductivist Turkeys? A Counterfactual Benchmark for Causal Reasoning"], "annotation": "Unknown", "applicationDomains": ["General AI"], "area": "Vision & 3D", "attentionForecast": {"confidence": "Medium", "horizon": "7d", "reason": "The counterfactual-generation framing, dedicated metrics, and coverage of open and closed models may draw near-term multimodal-evaluation interest.", "score": 66}, "capabilities": ["Reasoning"], "confidence": "Medium", "construction": "Unknown", "constructionDetail": "Unknown — the indexer does not infer construction details without explicit source evidence.", "copyGeneration": {"generatedAt": "2026-09-24T20:22:34.615233Z", "inputHash": "b331e1a0c9e905f863b2ead0e26d8ca72acb09e1d5b581ec6a039bb1b4ff5eab", "model": "codex-cli-default"}, "curation": {"admissionPolicyVersion": "2026-09-09.1", "canonicalNameEvidence": "we introduce Counterfactual-World (CF-World), a counterfactual benchmark", "canonicalNameSource": "abstract", "decisionReason": "The source names a repeatable benchmark, defines its staged task and two concrete metrics, and supplies a public project link.", "evaluationEvidence": {"reuse": "Project page: https://github.com/jylei16/CF-World.github.io", "scoring": "Prior Resistance Rate (PRR), which measures a models' ability to overcome entrenched real-world priors, and Reasoning Retention Rate (RRR)", "task": "generate images under rules that systematically contradict real-world priors"}, "model": "codex-cli-default", "reviewedAt": "2026-09-24T20:22:34.615233Z", "state": "ai-reviewed"}, "dataStatus": "primary-source-indexed", "demo": false, "description": "A three-level benchmark evaluating whether text-to-image models can follow explicit and implicit counterfactual rules rather than defaulting to real-world visual priors.", "displayEligible": true, "domainCuration": {"method": "rules-v1", "state": "auto"}, "evaluationMode": "public_reusable", "evidence": {"reasonCodes": ["benchmark term in abstract", "benchmark release stated in one sentence", "public artifact URL"], "snippet": "Inspired by Russell's inductivist turkey, we introduce Counterfactual-World (CF-World), a counterfactual benchmark designed to investigate whether text-to-image models can generate images under rules that systematically contradict real-world priors."}, "familyId": "bmf_2c398bc3f79d", "firstSeenAt": "2026-08-19", "heat": null, "id": "bm_are-text-to-image-models-inductivist-turke_acb79322", "indexedAt": "2026-08-19T11:05:13.395391Z", "industrySectors": [], "links": {"code": "https://github.com/jylei16/CF-World.github.io", "data": null, "paper": "https://arxiv.org/abs/2606.24548", "pdf": "https://arxiv.org/pdf/2606.24548", "project": null, "report": "https://arxiv.org/abs/2606.24548"}, "metrics": [], "motivation": "Text-to-image (T2I) generation models have achieved remarkable progress in producing visually realistic images from natural language prompts.", "name": "Counterfactual-World (CF-World)", "oneLine": "A three-level benchmark evaluating whether text-to-image models can follow explicit and implicit counterfactual rules rather than defaulting to real-world visual priors.", "paperTitle": "Are Text-to-Image Models Inductivist Turkeys? A Counterfactual Benchmark for Causal Reasoning", "primaryDomain": "General AI", "publication": {"evidence": null, "evidenceLevel": "unverified", "evidenceUrl": "https://arxiv.org/abs/2606.24548", "source": "arxiv-metadata", "status": "unverified", "venue": null, "verifiedAt": "2026-08-19T11:05:13.395391Z"}, "publications": [], "readiness": "Runnable", "recognitionConfidence": 0.55, "relation": "introduces", "releasedAt": "2026-06-23", "reviewLifecycle": {"state": "pending"}, "source": {"authors": ["Jiayi Lei", "Yuandong Pu", "Xingyu Han", "Rongpeng Zhu", "Jing Xu", "Jinyao Wang", "Zijian Zhou", "Bin Fu", "Yuewen Cao", "Yihao Liu", "Hongsheng Li"], "categories": ["cs.CV"], "id": "2606.24548", "title": "Are Text-to-Image Models Inductivist Turkeys? A Counterfactual Benchmark for Causal Reasoning", "type": "arxiv", "url": "https://arxiv.org/abs/2606.24548"}, "sourceUpdatedAt": "2026-07-01T03:29:55Z", "topics": ["Reasoning"], "venueAttempts": [], "whyItMatters": "It separates ordinary generation, direct counterfactual instruction following, and counterfactual causal deduction while reporting dedicated resistance and retention metrics."}}}, "schema_version": 1, "scores_by_source": {}, "siblings": []}
