{"id":"W4400397332","doi":"","title":"Participation du CRIM à DEFT 2024 : Utilisation de petits modèles de Langue pour des QCMs dans le domaine médical","year":2024,"lang":"fr","type":"article","venue":"HAL (Le Centre pour la Communication Scientifique Directe)","topic":"Natural Language Processing Techniques","field":"Computer Science","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"Computer Research Institute of Montréal","funders":"","keywords":"Mod; Political science; Computer science; Artificial intelligence","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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.01206911,0.001358991,0.0009387119,0.00163637,0.00144543,0.003803423,0.002217684,0.00321516,0.01354005],"category_scores_gemma":[0.04246479,0.0009627788,0.001927098,0.0009202566,0.001352946,0.004387483,0.002570926,0.003023945,0.003542108],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.004636755,"about_ca_system_score_gemma":0.007391268,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.07118141,"about_ca_topic_score_gemma":0.05996357,"domain_scores_codex":[0.9889587,0.005983947,0.0004661694,0.001136076,0.00295183,0.0005032933],"domain_scores_gemma":[0.975182,0.01492819,0.0003819364,0.003203763,0.005501728,0.0008025285],"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":[0.006420835,0.002440142,0.02072331,0.002610804,0.0005941162,0.002848103,0.009506551,0.2172576,0.04927139,0.06146721,0.1684113,0.4584486],"study_design_scores_gemma":[0.0003971973,0.0006785495,0.004682398,0.0002535552,0.0001526592,0.0004855873,0.00162044,0.7834531,0.04034471,0.009388829,0.1583776,0.0001653526],"study_design_candidate":"bench_or_experimental","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.4064416,0.001857983,0.4469801,0.01759447,0.002155064,0.001649547,0.01988223,0.04864082,0.05479811],"genre_scores_gemma":[0.5977294,0.0006798736,0.3333984,0.001255678,0.0002033431,0.0007155385,0.02898177,0.007301928,0.0297339],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.07118141,"threshold_uncertainty_score":0.1415342,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02346733149549102,"score_gpt":0.2704081809093147,"score_spread":0.2469408494138237,"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."}}