{"record": {"aliases": ["An Exam for Active Observers"], "artifacts": [{"id": "arxiv:2607.16165", "kind": "paper", "url": "https://arxiv.org/abs/2607.16165"}, {"kind": "website", "url": "https://activevision.dev"}], "authors": ["Jiarui Zhang", "Muzi Tao", "Shangshang Wang", "Ollie Liu", "Xuezhe Ma", "Willie Neiswanger"], "authors_reference": {"basis": "arxiv_metadata", "retrieved_at": "2026-11-03", "source_key": "claire-radar:2607.16165", "source_url": "https://arxiv.org/abs/2607.16165"}, "categories": ["General AI", "Vision & 3D", "cs.CV"], "description": {"en": "ActiveVision evaluates whether multimodal large language models can perform repeated, active visual perception across 17 tasks in three categories."}, "documents": [{"document_type": "registry_page", "id": "claire_radar:https://arxiv.org/abs/2607.16165", "source": "claire_radar", "source_url": "https://arxiv.org/abs/2607.16165", "title": "ActiveVision"}], "evidence_summary": {"document_count": 1, "model_count": null, "model_count_basis": null}, "institutions": ["University of Southern California"], "institutions_reference": {"basis": "openalex_linked_institutions", "doi": "10.48550/arxiv.2607.16165", "retrieved_at": "2026-11-04", "source_key": "claire-radar:2607.16165", "source_url": "https://openalex.org/W7169754001"}, "key": "claire-radar:2607.16165", "modality": null, "name": "ActiveVision", "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_an-exam-for-active-observers_020224c4", "origin_source": "arxiv", "recognition_confidence": "0.7", "record_sha256": "a328df65585e5c2b2b9138f1cfd1d74f1ec3a46a2f641693fd548efdab0a372b", "relation": "introduces", "review_model": "codex-cli-default", "review_state": "ai-reviewed", "reviewed_at": "2026-09-24T17:28:54.793904Z", "source_url": "https://arxiv.org/abs/2607.16165"}, "publisher": null, "released": "2026-07-17", "released_reference": {"basis": "benchmark_release", "source_key": "claire-radar:2607.16165", "source_url": "https://arxiv.org/abs/2607.16165"}, "schema_version": 1, "sizes": [], "slug": "claire-radar-2607-16165", "source": "claire_radar", "source_benchmark_id": "2607.16165", "source_metadata": {"claire_radar": {"adoption30d": null, "aliases": ["An Exam for Active Observers"], "annotation": "Unknown", "applicationDomains": ["General AI"], "area": "Vision & 3D", "attentionForecast": {"confidence": "Medium", "horizon": "7d", "reason": "The pronounced human–model performance gap and evaluation of frontier multimodal models may attract near-term attention.", "score": 72}, "capabilities": [], "confidence": "Medium", "construction": "Unknown", "constructionDetail": "Unknown — the indexer does not infer construction details without explicit source evidence.", "copyGeneration": {"generatedAt": "2026-09-24T17:28:54.793904Z", "inputHash": "f76f9c564bee137a2c147cbd4a0b2f5e98f7158a4eb65400c2258299aee9cb14", "model": "codex-cli-default"}, "curation": {"admissionPolicyVersion": "2026-09-09.1", "canonicalNameEvidence": "We introduce ActiveVision, a benchmark", "canonicalNameSource": "abstract", "decisionReason": "The source formally names the benchmark, specifies its task structure, reports a comparable item-solving score, and supplies a public project page.", "evaluationEvidence": {"reuse": "Project page: https://activevision.dev", "scoring": "solves only 10.6% of items and scores zero on 11 of the 17 tasks", "task": "comprising 17 tasks across 3 categories"}, "model": "codex-cli-default", "reviewedAt": "2026-09-24T17:28:54.793904Z", "state": "ai-reviewed"}, "dataStatus": "primary-source-indexed", "demo": false, "description": "ActiveVision evaluates whether multimodal large language models can perform repeated, active visual perception across 17 tasks in three categories.", "displayEligible": true, "domainCuration": {"method": "rules-v1", "state": "auto"}, "evaluationMode": "public_reusable", "evidence": {"reasonCodes": ["benchmark term in abstract", "benchmark release stated in one sentence", "evaluation protocol evidence", "public artifact URL"], "snippet": "We introduce ActiveVision, a benchmark that makes active observation measurable for MLLMs, comprising 17 tasks across 3 categories."}, "familyId": "bmf_c22ae8c6b4c5", "firstSeenAt": "2026-08-19", "heat": null, "id": "bm_an-exam-for-active-observers_020224c4", "indexedAt": "2026-08-19T11:05:13.395391Z", "industrySectors": [], "links": {"code": null, "data": null, "paper": "https://arxiv.org/abs/2607.16165", "pdf": "https://arxiv.org/pdf/2607.16165", "project": "https://activevision.dev", "report": "https://arxiv.org/abs/2607.16165"}, "metrics": [], "motivation": "Human vision is a closed loop: gaze is continuously redirected by intermediate hypotheses rather than a single snapshot.", "name": "ActiveVision", "oneLine": "ActiveVision evaluates whether multimodal large language models can perform repeated, active visual perception across 17 tasks in three categories.", "paperTitle": "An Exam for Active Observers", "primaryDomain": "General AI", "publication": {"evidence": null, "evidenceLevel": "unverified", "evidenceUrl": "https://arxiv.org/abs/2607.16165", "source": "arxiv-metadata", "status": "unverified", "venue": null, "verifiedAt": "2026-08-19T11:05:13.395391Z"}, "publications": [], "readiness": "Paper only", "recognitionConfidence": 0.7, "relation": "introduces", "releasedAt": "2026-07-17", "reviewLifecycle": {"state": "pending"}, "source": {"authors": ["Jiarui Zhang", "Muzi Tao", "Shangshang Wang", "Ollie Liu", "Xuezhe Ma", "Willie Neiswanger"], "categories": ["cs.CV", "cs.AI", "cs.CL", "cs.LG"], "id": "2607.16165", "title": "An Exam for Active Observers", "type": "arxiv", "url": "https://arxiv.org/abs/2607.16165"}, "sourceUpdatedAt": "2026-07-17T17:46:23Z", "topics": ["cs.CV"], "venueAttempts": [], "whyItMatters": "It targets active observation, a capability the supplied text says existing vision-language benchmarks do not measure."}}}, "schema_version": 1, "scores_by_source": {}, "siblings": []}
