{"id":"W6910435349","doi":"10.48448/a3r5-n223","title":"Words Worth a Thousand Pictures: Measuring and Understanding Perceptual Variability in Text-to-Image Generation","year":2024,"lang":"en","type":"other","venue":"Underline Science Inc.","topic":"","field":"","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Waterloo","funders":"","keywords":"Perception; Set (abstract data type); Measure (data warehouse); Affect (linguistics); Diffusion; Point (geometry); Variation (astronomy); Calibration","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":[0.00283696,0.0006343428,0.0004067716,0.001085014,0.0003334355,0.001638834,0.000628312,0.0008015072,0.001734338],"category_scores_gemma":[0.03350822,0.0002401645,0.0004867064,0.0006236417,0.0005834675,0.002348429,0.001313748,0.0009492396,0.0005912695],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0007476424,"about_ca_system_score_gemma":0.0002762553,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001409415,"about_ca_topic_score_gemma":0.001328191,"domain_scores_codex":[0.9986039,0.000464242,0.00009048032,0.0004435148,0.00033354,0.00006430705],"domain_scores_gemma":[0.9807442,0.01454378,0.001597459,0.001634941,0.00106892,0.0004106573],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.005314725,0.0007442324,0.09953636,0.001605307,0.0004128608,0.0008785968,0.004916352,0.1349466,0.1008972,0.007776109,0.01149153,0.6314802],"study_design_scores_gemma":[0.000188978,0.001982521,0.1341962,0.0001880623,0.0002247249,0.001521646,0.001527874,0.7713487,0.05475749,0.02386495,0.009987824,0.00021109],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8609763,0.001473273,0.127991,0.0003548033,0.0001497843,0.00024757,0.001454349,0.002955241,0.004397627],"genre_scores_gemma":[0.9613436,0.0001669696,0.03574177,0.00008106402,0.00003255123,0.00009327466,0.001332643,0.0003025419,0.0009056256],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.00283696,"threshold_uncertainty_score":0.01500344,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.06876972136773002,"score_gpt":0.3102602665424992,"score_spread":0.2414905451747692,"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."}}