{"id":"W4378474033","doi":"10.48550/arxiv.2305.14763","title":"Clever Hans or Neural Theory of Mind? Stress Testing Social Reasoning in Large Language Models","year":2023,"lang":"en","type":"preprint","venue":"arXiv (Cornell University)","topic":"Explainable Artificial Intelligence (XAI)","field":"Computer Science","cited_by":36,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"European Commission; Natural Sciences and Engineering Research Council of Canada; Defense Advanced Research Projects Agency; Naval Information Warfare Center Pacific; Bar-Ilan University; Canadian Institute for Advanced Research","keywords":"Heuristics; Theory of mind; Adversarial system; Psychology; Benchmark (surveying); Cognitive psychology; Artificial intelligence; Cognitive science; Computer science; Cognition","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":[0.01189234,0.0005427434,0.0003880874,0.0005491541,0.0006405278,0.002044071,0.001241987,0.001220979,0.003522155],"category_scores_gemma":[0.1087394,0.0002608275,0.0004578935,0.0004767387,0.003086402,0.004308379,0.001647004,0.00258732,0.0004016839],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0009295794,"about_ca_system_score_gemma":0.0007395021,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001334625,"about_ca_topic_score_gemma":0.002019474,"domain_scores_codex":[0.9926412,0.005550088,0.0001970222,0.0006079054,0.0008312359,0.0001725693],"domain_scores_gemma":[0.8761271,0.1069958,0.004128262,0.009172114,0.002311572,0.001265271],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"bench_or_experimental","study_design_scores_codex":[0.002028177,0.002882815,0.153618,0.001392257,0.00109255,0.001028591,0.0107137,0.2789893,0.01537771,0.2060996,0.01725177,0.3095255],"study_design_scores_gemma":[0.0001351803,0.0007527842,0.01648336,0.0001543671,0.0001034992,0.0002331852,0.001769214,0.65631,0.009743391,0.3081458,0.006104938,0.00006436482],"study_design_candidate":"bench_or_experimental","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9264388,0.0003657928,0.05233089,0.005656691,0.00008070052,0.0001059133,0.0002566492,0.0003561499,0.01440831],"genre_scores_gemma":[0.9846416,0.00006806598,0.01424835,0.0002841635,0.00002755927,0.00005936101,0.0001806513,0.00003706751,0.0004532401],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.01189234,"threshold_uncertainty_score":0.06289345,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.2058113727901079,"score_gpt":0.2491067126117638,"score_spread":0.04329533982165584,"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."}}