{"id":"W4402467279","doi":"10.1088/1402-4896/ad7a27","title":"Physics simulation capabilities of LLMs","year":2024,"lang":"en","type":"article","venue":"Physica Scripta","topic":"Advancements in Semiconductor Devices and Circuit Design","field":"Engineering","cited_by":5,"is_retracted":false,"has_abstract":true,"ca_institutions":"Canadian Institute for Theoretical Astrophysics; University of Toronto","funders":"Tamkeen; New York University Abu Dhabi","keywords":"Physics","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.003856609,0.00105789,0.0005718392,0.000996125,0.0005655996,0.00200586,0.002501923,0.001789337,0.01229262],"category_scores_gemma":[0.02970956,0.0005413309,0.001169617,0.0007459028,0.001028917,0.002248947,0.003491118,0.002147358,0.003310378],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001166407,"about_ca_system_score_gemma":0.001623485,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002180898,"about_ca_topic_score_gemma":0.002387111,"domain_scores_codex":[0.9962015,0.001503172,0.0003388488,0.0004124462,0.001319641,0.0002244356],"domain_scores_gemma":[0.9811477,0.01057193,0.000732534,0.004286892,0.002600368,0.0006605242],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.001676455,0.001614044,0.02332613,0.002156602,0.0002384978,0.0005508753,0.001782003,0.5521893,0.02218549,0.04047461,0.07174474,0.2820612],"study_design_scores_gemma":[0.0002131299,0.0003886586,0.002197611,0.0001660728,0.0000366423,0.000118937,0.0003247307,0.908341,0.02700372,0.02547828,0.03567691,0.00005431985],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.6228595,0.0008499335,0.2177046,0.003311005,0.000525857,0.0007494638,0.007860926,0.07369888,0.07243983],"genre_scores_gemma":[0.8474126,0.0002140435,0.1309004,0.000578139,0.0000449079,0.0005161422,0.01025527,0.003957232,0.006121241],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.01229262,"threshold_uncertainty_score":0.04112291,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02646617471715378,"score_gpt":0.2625549924995778,"score_spread":0.236088817782424,"validation_status":"score_only:v0-immature-baseline","note":"Baseline scores from an immature model (maturity gate not passed). Scores rank; they never assert a category."}}