{"record": {"aliases": [], "artifacts": [{"id": "arxiv:2606.01849", "kind": "paper", "url": "https://arxiv.org/abs/2606.01849"}, {"id": "gh:plau666/continuousbencheval", "kind": "repo", "url": "https://github.com/plau666/ContinuousBenchEval"}, {"kind": "website", "url": "https://peihanliu.com/posts/continuousbench.html"}], "authors": ["Peihan Liu", "Lucas Rosenblatt", "Weiwei Kong", "Natalia Ponomareva", "Gautam Kamath", "Rachel Cummings", "Roxana Geambasu", "Yu Gan", "Lillian Tsai", "Alex Bie"], "authors_reference": {"basis": "arxiv_metadata", "retrieved_at": "2026-11-03", "source_key": "claire-radar:2606.01849", "source_url": "https://arxiv.org/abs/2606.01849"}, "categories": ["General AI", "Language & Knowledge", "cs.LG"], "description": {"en": "ContinuousBench evaluates differentially private synthetic text by measuring capability gain on QA sets derived from a new quarterly corpus (Geminon procedural data or News articles), with standardized training and evaluation harness."}, "documents": [{"document_type": "registry_page", "id": "claire_radar:https://arxiv.org/abs/2606.01849", "source": "claire_radar", "source_url": "https://arxiv.org/abs/2606.01849", "title": "ContinuousBench"}], "evidence_summary": {"document_count": 1, "model_count": null, "model_count_basis": null}, "institutions": ["Columbia University", "New York University", "Google (United States)", "University of Waterloo", "Vector Institute"], "institutions_reference": {"basis": "openalex_linked_institutions", "doi": "10.48550/arxiv.2606.01849", "retrieved_at": "2026-11-04", "source_key": "claire-radar:2606.01849", "source_url": "https://openalex.org/W7163151095"}, "key": "claire-radar:2606.01849", "modality": null, "name": "ContinuousBench", "openness": {"code_license": null, "data_license": null, "evidence": [], "status": "unknown"}, "provenance": {"confidence": "High", "crawl_bundle": "claire_radar_2026-10-14", "crawled_at": "2026-10-14T09:44:51Z", "data_status": "primary-source-indexed", "display_eligible": "unknown", "origin_record_id": "bm_continuousbench_3f96ea58", "origin_source": "arxiv", "recognition_confidence": "1.0", "record_sha256": "7f9f61f9c0514c95cc5e6c5fd9407c72652ce58ecce034516e3ec948d69bd8f9", "relation": "introduces", "review_state": "unreviewed", "source_url": "https://arxiv.org/abs/2606.01849"}, "publisher": null, "released": "2026-06-01", "released_reference": {"basis": "benchmark_release", "source_key": "claire-radar:2606.01849", "source_url": "https://arxiv.org/abs/2606.01849"}, "schema_version": 1, "sizes": [], "slug": "claire-radar-2606-01849", "source": "claire_radar", "source_benchmark_id": "2606.01849", "source_metadata": {"claire_radar": {"adoption30d": null, "aliases": [], "annotation": "Machine Generated", "applicationDomains": ["General AI"], "area": "Language & Knowledge", "attention": {"asOf": "2026-10-05", "attemptedAt": "2026-10-04T15:31:48.030916Z", "githubRepo": "https://github.com/plau666/ContinuousBenchEval", "githubScope": "benchmark_repo", "githubStars": 5, "hfDailySubmittedAt": null, "hfDataset": null, "hfDatasetDownloads": null, "hfDatasetLikes": null, "hfPaperUpvotes": 1, "hfPaperUrl": "https://huggingface.co/papers/2606.01849", "observedAt": "2026-10-04T15:31:48.030916Z", "signalStatus": {"githubStars": {"observedAt": "2026-10-04T15:31:48.030916Z", "state": "fresh"}, "hfDatasetDownloads": {"state": "not_applicable"}, "hfDatasetLikes": {"state": "not_applicable"}, "hfPaperUpvotes": {"checkedAt": "2026-10-04T15:31:48.030916Z", "lastSuccessfulDate": "2026-09-30", "state": "stale"}}, "status": "fresh"}, "capabilities": [], "confidence": "High", "construction": "Original Synthetic", "constructionDetail": "Unknown — the indexer does not infer construction details without explicit source evidence.", "copyGeneration": {"generatedAt": "2026-08-20T15:22:10.044731Z", "inputHash": "113e95b484f3adde67025ea460c049ad054b99d2741702d34dc1c2b12308965c", "model": "deepseek-v4-flash"}, "dataStatus": "primary-source-indexed", "demo": false, "description": "ContinuousBench evaluates differentially private synthetic text by measuring capability gain on QA sets derived from a new quarterly corpus (Geminon procedural data or News articles), with standardized training and evaluation harness.", "domainCuration": {"method": "rules-v1", "state": "auto"}, "evidence": {"reasonCodes": ["coined title prefix ending in Bench or Benchmark", "benchmark release stated in one sentence", "evaluation protocol evidence", "public artifact URL"], "snippet": "Thus, we introduce ContinuousBench, a continuously and automatically-regenerated benchmark that measures capability gain from DP synthetic text."}, "familyId": "bmf_d6ffcf4f16b1", "firstSeenAt": "2026-08-19", "githubIndex": {"candidateCount": 1, "indexedAt": "2026-09-17T04:40:34.111665+00:00", "repositories": [{"evidence": [{"basis": "existing-code-link", "linkedUrl": "https://github.com/plau666/ContinuousBenchEval", "url": "https://github.com/plau666/ContinuousBenchEval"}, {"basis": "existing-metric-association", "linkedUrl": "https://github.com/plau666/ContinuousBenchEval", "url": "https://huggingface.co/papers/2606.01849"}, {"basis": "linked-from-project", "linkedUrl": "https://github.com/plau666/ContinuousBenchEval", "url": "https://peihanliu.com/posts/continuousbench.html"}, {"basis": "linked-from-report", "linkedUrl": "https://github.com/plau666/ContinuousBenchEval", "url": "https://arxiv.org/abs/2606.01849"}], "scope": "benchmark_repo", "url": "https://github.com/plau666/ContinuousBenchEval", "verification": "source-linked"}], "status": "linked"}, "heat": null, "id": "bm_continuousbench_3f96ea58", "indexedAt": "2026-08-19T11:05:13.395391Z", "industrySectors": [], "links": {"code": "https://github.com/plau666/ContinuousBenchEval", "data": null, "hfPaper": "https://huggingface.co/papers/2606.01849", "paper": "https://arxiv.org/abs/2606.01849", "pdf": "https://arxiv.org/pdf/2606.01849", "project": "https://peihanliu.com/posts/continuousbench.html", "report": "https://arxiv.org/abs/2606.01849"}, "metrics": [], "motivation": "Differentially private (DP) text synthesis promises to unlock sensitive corpora for model training, but it remains unclear whether DP synthetic data transmits genuinely new knowledge and capabilities present only in those corpora.", "name": "ContinuousBench", "oneLine": "ContinuousBench evaluates differentially private synthetic text by measuring capability gain on QA sets derived from a new quarterly corpus (Geminon procedural data or News articles), with standardized training and evaluation harness.", "paperTitle": "ContinuousBench: Can Differentially Private Synthetic Text Improve Capabilities?", "primaryDomain": "General AI", "publication": {"evidence": null, "evidenceLevel": "unverified", "evidenceUrl": "https://arxiv.org/abs/2606.01849", "source": "arxiv-metadata", "status": "unverified", "venue": null, "verifiedAt": "2026-10-14T01:46:58.058199Z"}, "publications": [], "publishers": [{"name": "ContinuousBench Team", "organizationType": "academic-lab", "role": "benchmark-publisher", "sourceUrl": "https://github.com/plau666/ContinuousBenchEval"}, {"name": "Peihan Liu", "organizationType": "academic-lab", "role": "benchmark-publisher", "sourceUrl": "https://peihanliu.com/posts/continuousbench.html"}], "ranking": {}, "readiness": "Runnable", "recognitionConfidence": 1.0, "relation": "introduces", "releasedAt": "2026-06-01", "source": {"authors": ["Peihan Liu", "Lucas Rosenblatt", "Weiwei Kong", "Natalia Ponomareva", "Gautam Kamath", "Rachel Cummings", "Roxana Geambasu", "Yu Gan", "Lillian Tsai", "Alex Bie"], "categories": ["cs.LG", "cs.CL", "cs.CR"], "id": "2606.01849", "title": "ContinuousBench: Can Differentially Private Synthetic Text Improve Capabilities?", "type": "arxiv", "url": "https://arxiv.org/abs/2606.01849"}, "sourceUpdatedAt": "2026-06-02T02:54:59Z", "topics": ["cs.LG"], "venueAttempts": [], "whyItMatters": "Addresses the gap where existing benchmarks are nearly solvable without corpus access, providing a continuously regenerated test to determine whether DP synthesis transmits genuinely new knowledge and capabilities."}}}, "schema_version": 1, "scores_by_source": {}, "siblings": []}
