{"id":"W7103154917","doi":"10.48550/arxiv.2510.25801","title":"Metis-SPECS: Decoupling Multimodal Learning via Self-distilled Preference-based Cold Start","year":2025,"lang":"","type":"preprint","venue":"Open MIND","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","scholarly_communication","open_science","research_integrity","insufficient_payload"],"consensus_categories":["metaepi_narrow","open_science","insufficient_payload"],"category_scores_codex":[0.003559445,0.001726652,0.002151138,0.0007885899,0.002055957,0.004690315,0.01185248,0.00113972,0.004639973],"category_scores_gemma":[0.0008203996,0.001971883,0.0006303511,0.001721838,0.0002736488,0.000780503,0.01214828,0.005326738,0.003035256],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001039513,"about_ca_system_score_gemma":0.004197038,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.005243194,"about_ca_topic_score_gemma":0.0004901423,"domain_scores_codex":[0.9878786,0.001356409,0.002395971,0.005264179,0.001421261,0.001683633],"domain_scores_gemma":[0.9895611,0.002166441,0.001960598,0.004383088,0.001111096,0.0008176463],"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":[0.0001789186,0.001399,0.01228661,0.0003074165,0.0005658579,0.00003629956,0.004134687,0.641071,0.001001942,0.0005883687,0.0000838993,0.338346],"study_design_scores_gemma":[0.002550156,0.0002344217,0.005860481,0.001049926,0.0003513414,0.000005960382,0.00008560819,0.9211128,0.003294809,0.0001993631,0.06347626,0.001778866],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.1289227,0.0003079585,0.8332313,0.001597383,0.001502465,0.008507675,0.0002162249,0.0002003213,0.02551395],"genre_scores_gemma":[0.5306039,0.00004351064,0.4641943,0.0000936677,0.0001612856,0.0009129775,0.0001920704,0.00006564186,0.003732633],"genre_candidate":"methods","genre_consensus":null,"teacher_disagreement_score":0.4016812,"threshold_uncertainty_score":0.999548,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.0642883970235779,"score_gpt":0.3301872525248995,"score_spread":0.2658988555013216,"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."}}