{"record": {"aliases": [], "artifacts": [{"id": "arxiv:2605.26087", "kind": "paper", "url": "https://arxiv.org/abs/2605.26087"}], "authors": ["Matt L. Wiemann", "Lindsay M. Smith", "Peter Melchior", "Siddharth Mishra-Sharma", "Andrew Gordon Wilson", "Pavel Izmailov", "Carolina Cuesta-Lázaro"], "authors_reference": {"basis": "arxiv_metadata", "retrieved_at": "2026-11-03", "source_key": "claire-radar:2605.26087", "source_url": "https://arxiv.org/abs/2605.26087"}, "categories": ["Agents", "Agents & Tool Use", "General AI"], "description": {"en": "DiscoverPhysics is an interactive benchmark for LLM agents to discover laws of motion in simulated worlds with altered physics, scoring trajectory MSE and explanation quality."}, "documents": [{"document_type": "registry_page", "id": "claire_radar:https://arxiv.org/abs/2605.26087", "source": "claire_radar", "source_url": "https://arxiv.org/abs/2605.26087", "title": "DiscoverPhysics"}], "evidence_summary": {"document_count": 1, "model_count": null, "model_count_basis": null}, "institutions": ["Princeton University", "Boston University"], "institutions_reference": {"basis": "openalex_linked_institutions", "doi": "10.48550/arxiv.2605.26087", "retrieved_at": "2026-11-04", "source_key": "claire-radar:2605.26087", "source_url": "https://openalex.org/W7162409218"}, "key": "claire-radar:2605.26087", "modality": null, "name": "DiscoverPhysics", "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_discoverphysics_9ae7e548", "origin_source": "arxiv", "recognition_confidence": "1.0", "record_sha256": "d78f1e77a848dcefaef79786b4e18b3d91b69a1405853a7ca0d496e6a1e31d29", "relation": "introduces", "review_state": "unreviewed", "source_url": "https://arxiv.org/abs/2605.26087"}, "publisher": null, "released": "2026-05-25", "released_reference": {"basis": "benchmark_release", "source_key": "claire-radar:2605.26087", "source_url": "https://arxiv.org/abs/2605.26087"}, "schema_version": 1, "sizes": [], "slug": "claire-radar-2605-26087", "source": "claire_radar", "source_benchmark_id": "2605.26087", "source_metadata": {"claire_radar": {"adoption30d": null, "aliases": [], "annotation": "Unknown", "applicationDomains": ["General AI"], "area": "Agents & Tool Use", "attention": {"asOf": "2026-10-05", "attemptedAt": "2026-10-04T15:31:48.030916Z", "githubRepo": null, "githubScope": null, "githubStars": null, "hfDailySubmittedAt": null, "hfDataset": null, "hfDatasetDownloads": null, "hfDatasetLikes": null, "hfPaperUpvotes": 1, "hfPaperUrl": "https://huggingface.co/papers/2605.26087", "observedAt": "2026-10-04T03:02:42.754940Z", "signalStatus": {"githubStars": {"state": "not_applicable"}, "hfDatasetDownloads": {"state": "not_applicable"}, "hfDatasetLikes": {"state": "not_applicable"}, "hfPaperUpvotes": {"checkedAt": "2026-10-04T15:31:48.030916Z", "lastSuccessfulDate": "2026-09-30", "state": "stale"}}, "status": "stale"}, "capabilities": [], "confidence": "High", "construction": "Interactive Environment", "constructionDetail": "Unknown — the indexer does not infer construction details without explicit source evidence.", "copyGeneration": {"generatedAt": "2026-08-20T15:22:10.044731Z", "inputHash": "17d1615f8ab47ce652f2f499d346be2d6771788446ac84148b256c1addd7b314", "model": "deepseek-v4-flash"}, "dataStatus": "primary-source-indexed", "demo": false, "description": "DiscoverPhysics is an interactive benchmark for LLM agents to discover laws of motion in simulated worlds with altered physics, scoring trajectory MSE and explanation quality.", "domainCuration": {"method": "rules-v1", "state": "auto"}, "evidence": {"reasonCodes": ["exact coined title identity tied to benchmark evidence", "benchmark release stated in one sentence"], "snippet": "We introduce DiscoverPhysics, an interactive benchmark that asks a LLM agent to discover the laws of motion of a simulated world whose physics deliberately deviates from our own."}, "familyId": "bmf_48d63b5c443b", "firstSeenAt": "2026-08-19", "githubIndex": {"candidateCount": 0, "indexedAt": "2026-09-17T04:40:34.111665+00:00", "repositories": [], "status": "not-found"}, "heat": null, "id": "bm_discoverphysics_9ae7e548", "indexedAt": "2026-08-19T11:05:13.395391Z", "industrySectors": [], "links": {"code": null, "data": null, "hfPaper": "https://huggingface.co/papers/2605.26087", "paper": "https://arxiv.org/abs/2605.26087", "pdf": "https://arxiv.org/pdf/2605.26087", "project": null, "report": "https://arxiv.org/abs/2605.26087"}, "metrics": [], "motivation": "Frontier LLMs now perform strongly across a wide range of physics evaluations, but it is hard to disentangle genuine reasoning from recall of established science.", "name": "DiscoverPhysics", "oneLine": "DiscoverPhysics is an interactive benchmark for LLM agents to discover laws of motion in simulated worlds with altered physics, scoring trajectory MSE and explanation quality.", "paperTitle": "DiscoverPhysics: Benchmarking LLMs for Out-of-the-Box Scientific Thinking", "primaryDomain": "General AI", "publication": {"evidence": null, "evidenceLevel": "unverified", "evidenceUrl": "https://arxiv.org/abs/2605.26087", "source": "arxiv-metadata", "status": "unverified", "venue": null, "verifiedAt": "2026-10-14T01:46:58.058199Z"}, "publications": [], "ranking": {}, "readiness": "Paper only", "recognitionConfidence": 1.0, "relation": "introduces", "releasedAt": "2026-05-25", "source": {"authors": ["Matt L. Wiemann", "Lindsay M. Smith", "Peter Melchior", "Siddharth Mishra-Sharma", "Andrew Gordon Wilson", "Pavel Izmailov", "Carolina Cuesta-L\\'azaro"], "categories": ["stat.ML", "cs.LG"], "id": "2605.26087", "title": "DiscoverPhysics: Benchmarking LLMs for Out-of-the-Box Scientific Thinking", "type": "arxiv", "url": "https://arxiv.org/abs/2605.26087"}, "sourceUpdatedAt": "2026-05-25T17:50:07Z", "topics": ["Agents"], "venueAttempts": [], "whyItMatters": "It probes long-horizon reasoning and hypothesis refinement, distinguishing recall from genuine scientific reasoning."}}}, "schema_version": 1, "scores_by_source": {}, "siblings": []}
