{"id":"W7106794272","doi":"10.48448/kj3e-3v32","title":"From Behavioral Performance to Internal Competence: Interpreting Vision-Language Models with VLM-Lens","year":2025,"lang":"","type":"other","venue":"Open MIND","topic":"","field":"","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Waterloo","funders":"","keywords":"Interpretability; Interpretation (philosophy); Interface (matter); Core (optical fiber); Layer (electronics); Base (topology)","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":[0.00248503,0.001765088,0.0006104888,0.001955118,0.0004742461,0.004655625,0.002150226,0.001277936,0.01509947],"category_scores_gemma":[0.02119865,0.0008717203,0.002039184,0.000979787,0.0009227959,0.005430414,0.004504063,0.002160963,0.007032759],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002049581,"about_ca_system_score_gemma":0.001800108,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01023825,"about_ca_topic_score_gemma":0.0156782,"domain_scores_codex":[0.9983438,0.0005842253,0.00009835578,0.0004742729,0.0003714203,0.0001279186],"domain_scores_gemma":[0.9963554,0.002066133,0.0002269411,0.0007255558,0.0004747736,0.0001512251],"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":[0.001370905,0.0004401708,0.02835715,0.002045294,0.0006901646,0.0006349713,0.004984311,0.1024628,0.01844259,0.09651507,0.1200721,0.6239844],"study_design_scores_gemma":[0.00008950662,0.000140669,0.005492826,0.0003177295,0.0001319463,0.0001953415,0.001033709,0.7185082,0.01830278,0.220857,0.03476627,0.0001641074],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.02979696,0.0004762355,0.8459618,0.00121013,0.0001759178,0.0002116861,0.01232258,0.09773629,0.01210849],"genre_scores_gemma":[0.4123491,0.0004028998,0.5496513,0.0005638942,0.00004891539,0.0005162267,0.02072329,0.009702554,0.006041775],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.01509947,"threshold_uncertainty_score":0.05051273,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02713953494518471,"score_gpt":0.3459310209742665,"score_spread":0.3187914860290817,"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."}}