{"id":"W4414160008","doi":"10.1167/tvst.14.9.18","title":"Advancing Question-Answering in Ophthalmology With Retrieval-Augmented Generation: Benchmarking Open-Source and Proprietary Large Language Models","year":2025,"lang":"en","type":"article","venue":"Translational Vision Science & Technology","topic":"Topic Modeling","field":"Computer Science","cited_by":0,"is_retracted":true,"has_abstract":true,"ca_institutions":"","funders":"Medical Research Council; Moorfields Eye Hospital NHS Foundation Trust; Retina UK; Moorfields Eye Charity; Sight Research UK; Department of Health and Social Care; Canadian Institute of Steel Construction; UK Research and Innovation; National Institute for Health and Care Research; Amazon Web Services","keywords":"Benchmarking; MEDLINE; Language model; Comprehension","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":[0.008293473,0.002124011,0.001046434,0.002096554,0.0006781637,0.002075931,0.003454433,0.002606787,0.003842257],"category_scores_gemma":[0.02310569,0.0005708168,0.002152582,0.001239334,0.0009913921,0.003532029,0.002971386,0.00342021,0.00311093],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002271077,"about_ca_system_score_gemma":0.002077353,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01603819,"about_ca_topic_score_gemma":0.0214047,"domain_scores_codex":[0.9953848,0.00244561,0.0003009481,0.001037443,0.0006234422,0.00020785],"domain_scores_gemma":[0.9854509,0.0105695,0.0003246229,0.00189481,0.001320963,0.0004391891],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.001674059,0.00229754,0.01457028,0.001946947,0.0008107948,0.0005680277,0.0008982088,0.2117877,0.01249951,0.004194517,0.043568,0.7051844],"study_design_scores_gemma":[0.0004899818,0.0009313585,0.004117712,0.0001349542,0.0002343482,0.0003584804,0.0003536557,0.9588876,0.01434173,0.007345003,0.01269247,0.0001126495],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.5624669,0.01673551,0.2919856,0.004685978,0.001822041,0.002304171,0.0128837,0.09059542,0.0165206],"genre_scores_gemma":[0.7110595,0.00151067,0.2387163,0.002334149,0.0003134942,0.0007462112,0.03836076,0.001360342,0.005598569],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.01603819,"threshold_uncertainty_score":0.04386061,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.0154630639637406,"score_gpt":0.3199486013467062,"score_spread":0.3044855373829656,"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."}}