{"id":"W4400376384","doi":"10.48550/arxiv.2407.02961","title":"Towards a Scalable Reference-Free Evaluation of Generative Models","year":2024,"lang":"en","type":"preprint","venue":"arXiv (Cornell University)","topic":"Simulation Techniques and Applications","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"Natural Sciences and Engineering Research Council of Canada; Chinese University of Hong Kong","keywords":"Generative grammar; Computer science; Scalability; Generative model; Artificial intelligence; Database","routes":{"ca_aff":false,"ca_fund":true,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.002648127,0.0002228637,0.000379146,0.0005357224,0.0001035466,0.0001391903,0.001610315,0.0003190144,0.0006865292],"category_scores_gemma":[0.000432493,0.0002044033,0.0002320056,0.001271376,0.0001420154,0.0001978639,0.002099491,0.0004207763,0.0001495283],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002495338,"about_ca_system_score_gemma":0.0006946185,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0002914257,"about_ca_topic_score_gemma":0.00007866544,"domain_scores_codex":[0.9971664,0.0003112494,0.0005067448,0.001078475,0.0007636353,0.0001734891],"domain_scores_gemma":[0.9952068,0.0002277486,0.0004094124,0.001907159,0.002149157,0.00009968581],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"simulation_or_modeling","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.00001136919,0.00003821194,0.0000505247,0.00001078932,0.00003260969,0.000001963372,0.0001485182,0.6073421,0.00004633363,0.3844285,0.003777746,0.004111279],"study_design_scores_gemma":[0.0001235333,0.00001179521,0.00007761874,0.00002734871,0.00009525387,2.268586e-7,0.000111486,0.4878606,0.0004433197,0.5108569,0.0002874999,0.0001045],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.507565,0.0002614342,0.3392386,0.0004397224,0.00028282,0.001099295,0.0003974656,0.0002163024,0.1504994],"genre_scores_gemma":[0.9930784,0.00007030443,0.002061763,0.00003027574,0.00005040106,0.000009988751,0.00003608116,0.00001578783,0.004647055],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.4855134,"threshold_uncertainty_score":0.8335325,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.546135853014692,"score_gpt":0.3552762688511931,"score_spread":0.190859584163499,"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."}}