{"id":"W4378072741","doi":"10.48550/arxiv.2305.13417","title":"VISIT: Visualizing and Interpreting the Semantic Information Flow of Transformers","year":2023,"lang":"en","type":"preprint","venue":"arXiv (Cornell University)","topic":"Explainable Artificial Intelligence (XAI)","field":"Computer Science","cited_by":3,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"Azrieli Foundation; Open Philanthropy Project","keywords":"Interpretability; Computer science; Transformer; Visualization; Generative grammar; Artificial intelligence; Vocabulary; Natural language processing","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":[0.000572063,0.0007953978,0.000264673,0.001416526,0.0003800552,0.001875029,0.0006407394,0.0008294777,0.01362269],"category_scores_gemma":[0.004070184,0.0002455099,0.0006478266,0.0008578795,0.0006390671,0.002584722,0.001213273,0.001219362,0.001268657],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0005571125,"about_ca_system_score_gemma":0.0006234557,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.004577137,"about_ca_topic_score_gemma":0.004694857,"domain_scores_codex":[0.9998626,0.00004805006,0.000007120846,0.00003262732,0.00003044033,0.00001917379],"domain_scores_gemma":[0.9987962,0.0007388967,0.00006987737,0.0001856909,0.0001396478,0.00006958583],"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":[0.001485626,0.0002544138,0.01847652,0.001109393,0.0002399782,0.001420518,0.01015636,0.1791779,0.06242004,0.2922474,0.1029353,0.3300766],"study_design_scores_gemma":[0.00008971061,0.0001067938,0.007515336,0.0001661738,0.00005764956,0.0003148073,0.001170569,0.6609052,0.01916225,0.2590303,0.05140463,0.00007646742],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.1163457,0.000590506,0.8093877,0.003546203,0.000322997,0.0001250712,0.01006844,0.04211225,0.01750113],"genre_scores_gemma":[0.7015778,0.0006090697,0.2827749,0.0003432149,0.00009394083,0.0001697769,0.004172334,0.00404853,0.006210485],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.01362269,"threshold_uncertainty_score":0.04557246,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.07883463045535774,"score_gpt":0.2173668414905878,"score_spread":0.1385322110352301,"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."}}