{"id":"W4402667062","doi":"10.18653/v1/2024.acl-long.20","title":"Confidence Under the Hood: An Investigation into the Confidence-Probability Alignment in Large Language Models","year":2024,"lang":"en","type":"article","venue":"","topic":"Topic Modeling","field":"Computer Science","cited_by":9,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Computer science; Confidence interval; Artificial intelligence; Confidence distribution; Natural language processing; Statistics; 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":[0.05227679,0.001143616,0.00130887,0.003304953,0.00189331,0.007239125,0.001922252,0.001445984,0.002782261],"category_scores_gemma":[0.3802657,0.001184168,0.001321818,0.003990561,0.003866886,0.01107747,0.006407075,0.004706118,0.0004974467],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002599029,"about_ca_system_score_gemma":0.002103051,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.004742534,"about_ca_topic_score_gemma":0.003103728,"domain_scores_codex":[0.9586021,0.02872019,0.002197001,0.004419213,0.005245277,0.0008162201],"domain_scores_gemma":[0.4381969,0.4979623,0.02543934,0.02421485,0.01193683,0.00224979],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.003011673,0.0009856493,0.4369043,0.001121053,0.001136148,0.0008466689,0.0445733,0.1505849,0.007838069,0.1321333,0.00479047,0.2160746],"study_design_scores_gemma":[0.00007994911,0.0006530913,0.05609411,0.000173013,0.0001585691,0.0005646857,0.003814752,0.8229896,0.004412649,0.1073897,0.003461898,0.0002079698],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.5503473,0.0005932735,0.4378412,0.00171836,0.0000704033,0.0003048591,0.0003803198,0.001270857,0.007473413],"genre_scores_gemma":[0.9611854,0.00009018019,0.03769502,0.0001092908,0.00003934349,0.0001298119,0.0003193737,0.000159007,0.0002727341],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.05227679,"threshold_uncertainty_score":0.2764694,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.04172525238957015,"score_gpt":0.2864207796739621,"score_spread":0.244695527284392,"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."}}