{"id":"W7077860944","doi":"10.48448/e7e3-kg06","title":"Rethinking Safety Evaluation in Large Language Models: A Research Proposal","year":2025,"lang":"en","type":"other","venue":"Underline Science Inc.","topic":"Geochemistry and Geologic Mapping","field":"Computer Science","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"McGill University","funders":"","keywords":"Occupational safety and health; Risk assessment; Health care; Public health; Entertainment; Robustness (evolution)","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":[0.09502839,0.001620338,0.001325534,0.003320625,0.00176499,0.01297694,0.004467148,0.003458424,0.005872728],"category_scores_gemma":[0.2452464,0.0009770804,0.002047387,0.002297623,0.006472382,0.02583573,0.008310094,0.006299748,0.002023332],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.003809597,"about_ca_system_score_gemma":0.007978173,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.005739072,"about_ca_topic_score_gemma":0.004982835,"domain_scores_codex":[0.9125017,0.06832299,0.003743281,0.005032334,0.009138165,0.001261528],"domain_scores_gemma":[0.7351869,0.1879006,0.008101444,0.03622765,0.02990141,0.002681891],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.00103389,0.001105116,0.01617012,0.001420708,0.000469326,0.0003268448,0.007122202,0.03798307,0.00739748,0.2805633,0.02096287,0.6254451],"study_design_scores_gemma":[0.0002897083,0.0008924096,0.002765695,0.001809322,0.000352913,0.0004641856,0.003667718,0.4460357,0.01427497,0.4622497,0.06696081,0.0002368574],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"protocol","genre_scores_codex":[0.03656675,0.003501214,0.9002739,0.03819143,0.000780388,0.001421191,0.0005760507,0.0040406,0.01464854],"genre_scores_gemma":[0.2781873,0.001463883,0.7090352,0.004673352,0.0006446292,0.0009157587,0.0009367174,0.00097804,0.003165189],"genre_candidate":"protocol","genre_consensus":null,"teacher_disagreement_score":0.09502839,"threshold_uncertainty_score":0.5025641,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.0685846569655526,"score_gpt":0.3647848096916979,"score_spread":0.2962001527261453,"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."}}