{"id":"W7103204388","doi":"","title":"Metis-SPECS: Decoupling Multimodal Learning via Self-distilled Preference-based Cold Start","year":2025,"lang":"","type":"article","venue":"ArXiv.org","topic":"Multimodal Machine Learning Applications","field":"Computer Science","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Generalization; Reinforcement learning; Task (project management); Task analysis; Verifiable secret sharing; Memorization","routes":{"ca_aff":false,"ca_fund":false,"ca_venue":false,"about_ca":true,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaepi_narrow","sts","research_integrity","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.001538879,0.001060909,0.001121776,0.0007262246,0.002000302,0.0006803741,0.00299819,0.0005456549,0.000442965],"category_scores_gemma":[0.0009713587,0.001186746,0.0004728633,0.002774397,0.0003044181,0.0006427357,0.001208799,0.00261309,0.001604264],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0006736296,"about_ca_system_score_gemma":0.001223158,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001814784,"about_ca_topic_score_gemma":0.0001405262,"domain_scores_codex":[0.9925026,0.0007596834,0.001584395,0.002632916,0.0009142328,0.001606135],"domain_scores_gemma":[0.9937646,0.001652036,0.0007934901,0.002470271,0.0007601547,0.0005594046],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.00008196556,0.000921914,0.8768795,0.0002620265,0.0003193044,0.00002304889,0.001097659,0.09242718,0.005995298,0.003586297,0.0001566061,0.01824925],"study_design_scores_gemma":[0.001805817,0.0001830769,0.3189562,0.0002695766,0.0001636731,0.000002378011,0.00004453239,0.6664221,0.003318895,0.0001254554,0.007963992,0.0007442911],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.5558855,0.0004965705,0.4370067,0.00186728,0.0007749776,0.001192534,0.000009840047,0.001172443,0.001594151],"genre_scores_gemma":[0.9327613,0.00007829245,0.064003,0.0007613437,0.0002389735,0.0004373431,0.00003680124,0.00009488189,0.001588064],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.5739949,"threshold_uncertainty_score":0.9996879,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.03630322338093774,"score_gpt":0.2811754722599381,"score_spread":0.2448722488790003,"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."}}