{"id":"W4389518708","doi":"10.18653/v1/2023.emnlp-main.395","title":"Towards Reliable Misinformation Mitigation: Generalization, Uncertainty, and GPT-4","year":2023,"lang":"en","type":"article","venue":"","topic":"Misinformation and Its Impacts","field":"Social Sciences","cited_by":29,"is_retracted":false,"has_abstract":true,"ca_institutions":"Université de Montréal; McGill University","funders":"Institut de Valorisation des Données; Canadian Institute for Advanced Research","keywords":"Misinformation; Generalization; Computer science; Artificial intelligence; Natural language processing; Epistemology; Philosophy; Computer security","routes":{"ca_aff":true,"ca_fund":true,"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":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0006383618,0.00005809776,0.00006379124,0.0001203552,0.0004989117,0.0002161034,0.00007465532,0.00006485105,0.001403399],"category_scores_gemma":[0.0002629211,0.00005249032,0.00001772911,0.0006773124,0.00007594412,0.0009798245,0.00002216153,0.0000330694,0.0004091019],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00004765111,"about_ca_system_score_gemma":0.0001539388,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0009052902,"about_ca_topic_score_gemma":0.0002177505,"domain_scores_codex":[0.9991722,0.00002887908,0.0001948937,0.00006781446,0.0003384785,0.0001978037],"domain_scores_gemma":[0.9995507,0.00002278464,0.00005799987,0.00008197087,0.000152607,0.0001339278],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","study_design_scores_codex":[0.000004576647,0.00000624363,0.0004750549,0.00002387028,0.000007708259,4.10818e-7,0.0794204,0.001716529,0.00002945136,0.4290429,0.4701696,0.01910327],"study_design_scores_gemma":[0.0005342698,0.00002945662,0.01184671,0.00002466804,0.000008425855,0.000001953181,0.05485472,0.03268043,0.0003635201,0.01170921,0.8876734,0.0002731971],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","genre_codex":"other","genre_gemma":"empirical","genre_scores_codex":[0.1727441,0.00002451551,0.002072374,0.0132622,0.0003387239,0.000274051,0.000005595442,0.0005675148,0.8107109],"genre_scores_gemma":[0.9642751,0.0008077901,0.0009841823,0.002963525,0.0002121964,0.000004460978,0.000161024,0.000007827443,0.03058394],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.791531,"threshold_uncertainty_score":0.9995095,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02628513755310111,"score_gpt":0.3163211441166154,"score_spread":0.2900360065635143,"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."}}