{"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":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.00002866733,0.00009729968,0.0001200317,0.00002092072,0.00001561403,0.00002205798,0.00008383496,0.00002049984,0.00004754951],"category_scores_gemma":[0.000003684313,0.0000948225,0.00005671574,0.0001445757,0.00003456104,0.0002712517,0.000008662356,0.00007236054,0.00003984508],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00003196618,"about_ca_system_score_gemma":0.00000722494,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000005235638,"about_ca_topic_score_gemma":3.133608e-7,"domain_scores_codex":[0.9995055,0.000006641048,0.0001329232,0.0001165898,0.0001169887,0.0001213296],"domain_scores_gemma":[0.9997368,0.00003019154,0.00001184489,0.0001766245,0.00002355578,0.00002096975],"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":[0.000003877537,0.00007045759,0.0001084664,0.001454718,0.0002258163,0.000003051286,0.004070549,0.4977878,0.343297,0.05568063,0.009042905,0.08825473],"study_design_scores_gemma":[0.00009015036,0.00003021194,0.00007127729,0.0001249164,0.00004038739,3.70383e-7,0.0002904423,0.8719985,0.04227857,0.02474607,0.06009732,0.00023176],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.6651682,0.01162859,0.2620671,0.00005397364,0.009762261,0.0007287603,0.0001824512,0.002280456,0.0481282],"genre_scores_gemma":[0.9991103,0.00002631026,0.0001517218,0.00001271524,0.0002631991,0.000009205559,0.000008353881,0.0000288413,0.0003893502],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.3742107,"threshold_uncertainty_score":0.3866749,"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."}}