{"id":"W4409019950","doi":"10.1101/2025.03.27.25324747","title":"Fine-tuned large language models enhance influenza forecasting","year":2025,"lang":"en","type":"preprint","venue":"medRxiv","topic":"Influenza Virus Research Studies","field":"Medicine","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"McGill University Health Centre","funders":"","keywords":"Computer science; Natural language processing","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":[0.001113198,0.0009732548,0.0006040504,0.0004417314,0.0002282265,0.000835405,0.0008139745,0.0009325058,0.001964658],"category_scores_gemma":[0.006053559,0.0003471414,0.0006020401,0.0002664769,0.0003257472,0.001561766,0.0007354782,0.00142298,0.001005225],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0005718294,"about_ca_system_score_gemma":0.0007777673,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.007685429,"about_ca_topic_score_gemma":0.008699599,"domain_scores_codex":[0.9997138,0.0000955463,0.00001725983,0.00009848558,0.00003242028,0.00004240618],"domain_scores_gemma":[0.9989495,0.0006917388,0.00007022068,0.00009169654,0.0001466842,0.00005011616],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.0002332138,0.0001591433,0.003196603,0.00007223978,0.00008467882,0.00009094489,0.00007089127,0.9168774,0.007298263,0.001081051,0.004034466,0.06680109],"study_design_scores_gemma":[0.000006944542,0.00001662322,0.0001656318,0.000003769476,0.000004704391,0.000004861016,0.000005795193,0.9981321,0.0008008541,0.00068868,0.0001663218,0.000003864114],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.4400078,0.002683152,0.5338752,0.002490971,0.0006915353,0.0001160491,0.001881027,0.01171219,0.006542107],"genre_scores_gemma":[0.9536043,0.0002031422,0.04230068,0.0004441864,0.0001209635,0.00005400379,0.001115155,0.0001662457,0.001991307],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.007685429,"threshold_uncertainty_score":0.01528138,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1254846156238433,"score_gpt":0.4128178899522112,"score_spread":0.2873332743283679,"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."}}