{"id":"W3137399373","doi":"10.1080/15248372.2021.1891902","title":"Testing What You’re Told: Young Children’s Empirical Investigation of a Surprising Claim","year":2021,"lang":"en","type":"article","venue":"Journal of Cognition and Development","topic":"Child and Animal Learning Development","field":"Psychology","cited_by":40,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Toronto","funders":"Hong Kong University of Science and Technology; Harvard University","keywords":"Psychology; Empirical research; Developmental psychology; Epistemology","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":[0.0005465682,0.0001264026,0.0002623278,0.0001528487,0.0001234246,0.00009348581,0.00005486167,0.00008073252,0.0002650268],"category_scores_gemma":[0.0001875995,0.0001143943,0.00004898555,0.0002155275,0.00004809297,0.0002251577,0.0000424802,0.000247678,0.00001462421],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00004597161,"about_ca_system_score_gemma":0.0003745756,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000002831499,"about_ca_topic_score_gemma":0.000005316638,"domain_scores_codex":[0.9985363,0.0001238773,0.0006857286,0.0001767996,0.0003141278,0.0001632015],"domain_scores_gemma":[0.9987186,0.0001306957,0.0004511321,0.00006107621,0.000507978,0.0001305511],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0001611281,0.0002414144,0.8335596,0.00006600647,0.0003119204,0.0001804091,0.02211498,0.000005143535,0.008434574,0.0001471139,0.001585781,0.133192],"study_design_scores_gemma":[0.001003387,0.00008933405,0.9836908,0.0005609583,0.00003672344,0.0009640304,0.003590604,0.000001853117,0.007688026,0.0002163868,0.00200558,0.0001523023],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.994895,0.001184582,0.0004409667,0.0005578496,0.0004207636,0.00006574138,0.000001168395,0.00001319814,0.002420778],"genre_scores_gemma":[0.9863384,0.00006924336,0.01265923,0.000635438,0.000125491,0.000001985624,0.00001962773,0.00001168673,0.0001389343],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.1501312,"threshold_uncertainty_score":0.4664865,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.06031862865033807,"score_gpt":0.3165405244049871,"score_spread":0.2562218957546491,"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."}}