{"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":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0005958805,0.0001929782,0.000226409,0.0002918039,0.0001915442,0.0001718987,0.001091868,0.0001407628,0.00000590758],"category_scores_gemma":[0.00009099732,0.0001818914,0.0001223617,0.0006444098,0.0001446931,0.00106531,0.001058492,0.0003596166,0.00005455442],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00007944796,"about_ca_system_score_gemma":0.00009130412,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0004493234,"about_ca_topic_score_gemma":0.00009748897,"domain_scores_codex":[0.998807,0.0001070887,0.0003107894,0.0003935058,0.0001107282,0.0002708955],"domain_scores_gemma":[0.9987553,0.0002530551,0.0002536651,0.0005188633,0.0001512145,0.00006792063],"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":[0.00007038737,0.00005729669,0.002136391,0.0009348029,0.0002471925,0.00009876929,0.03535607,0.5534759,0.0003058993,0.3645356,0.0002344742,0.04254722],"study_design_scores_gemma":[0.00006505597,0.00003551595,0.0001951686,0.0002298416,0.00003099309,0.000002332751,0.002360868,0.9743525,0.001698146,0.02065455,0.0001772855,0.0001977522],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.186904,0.00002189372,0.8112284,0.0002696342,0.000338002,0.0002566123,0.00000552583,0.0001473438,0.0008285365],"genre_scores_gemma":[0.998907,0.0001860985,0.0006984205,0.00008331203,0.0000163376,9.688231e-7,0.000005233032,0.000009488091,0.00009313518],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.812003,"threshold_uncertainty_score":0.7417316,"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."}}