{"id":"W4402512774","doi":"10.1038/s41598-024-72071-1","title":"Large language models predict human sensory judgments across six modalities","year":2024,"lang":"en","type":"article","venue":"Scientific Reports","topic":"Categorization, perception, and language","field":"Psychology","cited_by":62,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"Natural Sciences and Engineering Research Council of Canada; NOMIS Stiftung","keywords":"Perception; Computer science; Pairwise comparison; Variation (astronomy); Modalities; Natural language processing; Artificial intelligence; Similarity (geometry); Task (project management); Modality (human–computer interaction); Visual perception; Cognition; Cognitive psychology; Psychology; Image (mathematics)","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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.002231166,0.0007992751,0.0004803741,0.0006219708,0.0002713077,0.001810598,0.0006060884,0.001077003,0.002217609],"category_scores_gemma":[0.01240545,0.0003720822,0.001281639,0.000370131,0.0008986844,0.002186039,0.001327083,0.001766403,0.0009380816],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0009022626,"about_ca_system_score_gemma":0.0005157883,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003486683,"about_ca_topic_score_gemma":0.003949896,"domain_scores_codex":[0.9992856,0.0002979752,0.0000252908,0.0002544652,0.00006633595,0.00007035231],"domain_scores_gemma":[0.9954471,0.003286949,0.0002758872,0.000489602,0.00032896,0.0001715277],"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":[0.001954171,0.0005187072,0.06224091,0.0007221109,0.0008677048,0.0005670195,0.001826217,0.5929156,0.1008154,0.01389817,0.008145991,0.2155282],"study_design_scores_gemma":[0.0000333256,0.0001634874,0.01405665,0.00003296789,0.00005082933,0.0001294929,0.0002073658,0.9558756,0.005202948,0.02369686,0.0004953946,0.00005507777],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8605248,0.0006278932,0.1302828,0.001095103,0.0001013482,0.00007434213,0.001081394,0.0009395974,0.005272665],"genre_scores_gemma":[0.9866641,0.00009978755,0.01143706,0.0001543242,0.00001505523,0.00003847725,0.000687246,0.000067908,0.0008360753],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.003486683,"threshold_uncertainty_score":0.01179969,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02987171211641267,"score_gpt":0.3489717679857248,"score_spread":0.3191000558693122,"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."}}