{"record": {"aliases": ["Do Text Edits Generalize to Visual Generation? Benchmarking Cross-Modal Knowledge Editing in UMMs"], "artifacts": [{"id": "arxiv:2606.00477", "kind": "paper", "url": "https://arxiv.org/abs/2606.00477"}, {"id": "gh:gxx27/unike", "kind": "repo", "url": "https://github.com/gxx27/UniKE"}], "authors": ["Xin Gao", "Cheng Yang", "Chufan Shi", "Taylor Berg-Kirkpatrick"], "authors_reference": {"basis": "arxiv_metadata", "retrieved_at": "2026-11-03", "source_key": "claire-radar:2606.00477", "source_url": "https://arxiv.org/abs/2606.00477"}, "categories": ["Factuality", "General AI", "Language & Knowledge", "cs.CL"], "description": {"en": "UniKE evaluates whether textual knowledge edits in unified multimodal models transfer to image generation, using 2,971 subjects covering attribute and relation edits."}, "documents": [{"document_type": "registry_page", "id": "claire_radar:https://arxiv.org/abs/2606.00477", "source": "claire_radar", "source_url": "https://arxiv.org/abs/2606.00477", "title": "UniKE"}], "evidence_summary": {"document_count": 1, "model_count": null, "model_count_basis": null}, "institutions": ["University of California San Diego", "University of Southern California"], "institutions_reference": {"basis": "openalex_linked_institutions", "doi": "10.48550/arxiv.2606.00477", "retrieved_at": "2026-11-04", "source_key": "claire-radar:2606.00477", "source_url": "https://openalex.org/W7163220166"}, "key": "claire-radar:2606.00477", "modality": null, "name": "UniKE", "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_do-text-edits-generalize-to-visual-generat_a895e53c", "origin_source": "arxiv", "recognition_confidence": "0.55", "record_sha256": "fe3778daa5f327867e123a1b80b44d84e93d712a4b211df5e401f89f90263dbd", "relation": "introduces", "review_model": "codex-cli-default", "review_state": "ai-reviewed", "reviewed_at": "2026-09-24T19:48:20.147478Z", "source_url": "https://arxiv.org/abs/2606.00477"}, "publisher": null, "released": "2026-05-30", "released_reference": {"basis": "benchmark_release", "source_key": "claire-radar:2606.00477", "source_url": "https://arxiv.org/abs/2606.00477"}, "schema_version": 1, "sizes": [], "slug": "claire-radar-2606-00477", "source": "claire_radar", "source_benchmark_id": "2606.00477", "source_metadata": {"claire_radar": {"adoption30d": null, "aliases": ["Do Text Edits Generalize to Visual Generation? Benchmarking Cross-Modal Knowledge Editing in UMMs"], "annotation": "Unknown", "applicationDomains": ["General AI"], "area": "Language & Knowledge", "attentionForecast": {"confidence": "Medium", "horizon": "7d", "reason": "The benchmark addresses an emerging multimodal model-editing problem and accompanies an ICML publication with public code and data.", "score": 3}, "capabilities": ["Factuality"], "confidence": "Medium", "construction": "Unknown", "constructionDetail": "Unknown — the indexer does not infer construction details without explicit source evidence.", "copyGeneration": {"generatedAt": "2026-09-24T19:48:20.147478Z", "inputHash": "25a76501bb4aebd9f419dd3345d93aeabd6d8835cb97f694dd5c953a11a4ae55", "model": "codex-cli-default"}, "curation": {"admissionPolicyVersion": "2026-09-09.1", "canonicalNameEvidence": "we introduce UniKE, the first benchmark for cross-modality knowledge editing in UMMs", "canonicalNameSource": "abstract", "decisionReason": "Publish because the source formally names a benchmark, specifies a repeatable cross-modal editing task and VQA-based metric, and provides an available public code-and-data repository.", "evaluationEvidence": {"reuse": "Our code and data are available at https://github.com/gxx27/UniKE.", "scoring": "VQA accuracy under direct image generation is only 18.5%", "task": "cross-modality knowledge editing in UMMs"}, "model": "codex-cli-default", "reviewedAt": "2026-09-24T19:48:20.147478Z", "state": "ai-reviewed"}, "dataStatus": "primary-source-indexed", "demo": false, "description": "UniKE evaluates whether textual knowledge edits in unified multimodal models transfer to image generation, using 2,971 subjects covering attribute and relation edits.", "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": "To study this question, we introduce UniKE, the first benchmark for cross-modality knowledge editing in UMMs, comprising 2,971 edit subjects spanning attribute and relation edits."}, "familyId": "bmf_f33e284ad61c", "firstSeenAt": "2026-08-19", "heat": null, "id": "bm_do-text-edits-generalize-to-visual-generat_a895e53c", "indexedAt": "2026-08-19T11:05:13.395391Z", "industrySectors": [], "links": {"code": "https://github.com/gxx27/UniKE", "data": null, "paper": "https://arxiv.org/abs/2606.00477", "pdf": "https://arxiv.org/pdf/2606.00477", "project": null, "report": "https://arxiv.org/abs/2606.00477"}, "metrics": [], "motivation": "Unified multimodal models (UMMs) have emerged as a promising paradigm for general-purpose multimodal intelligence.", "name": "UniKE", "oneLine": "UniKE evaluates whether textual knowledge edits in unified multimodal models transfer to image generation, using 2,971 subjects covering attribute and relation edits.", "paperTitle": "Do Text Edits Generalize to Visual Generation? Benchmarking Cross-Modal Knowledge Editing in UMMs", "primaryDomain": "General AI", "publication": {"evidence": null, "evidenceLevel": "unverified", "evidenceUrl": "https://arxiv.org/abs/2606.00477", "source": "arxiv-metadata", "status": "unverified", "venue": null, "verifiedAt": "2026-08-19T11:05:13.395391Z"}, "publications": [], "readiness": "Runnable", "recognitionConfidence": 0.55, "relation": "introduces", "releasedAt": "2026-05-30", "reviewLifecycle": {"state": "pending"}, "source": {"authors": ["Xin Gao", "Cheng Yang", "Chufan Shi", "Taylor Berg-Kirkpatrick"], "categories": ["cs.CL", "cs.CV"], "id": "2606.00477", "title": "Do Text Edits Generalize to Visual Generation? Benchmarking Cross-Modal Knowledge Editing in UMMs", "type": "arxiv", "url": "https://arxiv.org/abs/2606.00477"}, "sourceUpdatedAt": "2026-05-30T02:09:37Z", "topics": ["cs.CL"], "venueAttempts": [], "whyItMatters": "It exposes a measurable gap between successful text-side editing and the visual effects of those edits."}}}, "schema_version": 1, "scores_by_source": {}, "siblings": []}
