{"id":"W4400977996","doi":"10.1016/j.patter.2024.101030","title":"Exploring the reversal curse and other deductive logical reasoning in BERT and GPT-based large language models","year":2024,"lang":"en","type":"article","venue":"Patterns","topic":"Topic Modeling","field":"Computer Science","cited_by":10,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"National Institutes of Health; CHEO Research Institute","keywords":"Transformer; Computer science; Encoder; Curse; Generative grammar; Logical reasoning; Comprehension; Artificial intelligence; Natural language processing; Autoencoder; Intersection (aeronautics); Context (archaeology); Machine learning; Programming language; Deep learning","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":[0.0002732912,0.00008269173,0.00008003951,0.000052723,0.00004450126,0.0001306237,0.0001687907,0.0000218374,0.000006066089],"category_scores_gemma":[0.00001464425,0.00005810342,0.00001761295,0.00008892113,0.00001869819,0.0003393049,0.0001496209,0.0001582132,0.00000280876],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00002096535,"about_ca_system_score_gemma":0.00001439677,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0003106845,"about_ca_topic_score_gemma":0.0001181612,"domain_scores_codex":[0.999259,0.00005546941,0.000091146,0.0003038204,0.000106162,0.0001844166],"domain_scores_gemma":[0.9996482,0.00009347875,0.00001248581,0.0002022081,0.000006128106,0.00003751886],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.0000355953,0.0001932316,0.1415191,0.0004152586,0.00009649163,0.0008388673,0.2568755,0.01092221,0.001456076,0.3447566,0.0001321079,0.2427589],"study_design_scores_gemma":[0.0001771444,0.00001385628,0.003797764,0.0001866461,0.0000044677,0.00001521837,0.0007501168,0.993164,0.0002439624,0.001360208,0.000173675,0.0001129555],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.6066137,0.0003458336,0.3921045,0.000602119,0.0001015294,0.0000573299,0.000003505733,0.00004780373,0.0001236982],"genre_scores_gemma":[0.9955752,0.00001294803,0.00388235,0.0003985735,0.00007057691,0.00002399397,4.2594e-7,0.000007921263,0.00002804546],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9822418,"threshold_uncertainty_score":0.2369389,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.07904218656723885,"score_gpt":0.2861965466404941,"score_spread":0.2071543600732552,"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."}}