{"id":"W4389519586","doi":"10.18653/v1/2023.emnlp-main.971","title":"MQuAKE: Assessing Knowledge Editing in Language Models via Multi-Hop Questions","year":2023,"lang":"en","type":"article","venue":"","topic":"Topic Modeling","field":"Computer Science","cited_by":38,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"Canadian Institute for Advanced Research; National Science Foundation","keywords":"Computer science; Recall; Language model; Question answering; Margin (machine learning); Natural language processing; Hop (telecommunications); Artificial intelligence; Machine learning; Cognitive psychology; Psychology","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":[0.0137675,0.002103852,0.00104757,0.002824659,0.0007433125,0.003021423,0.002838548,0.003610129,0.004119713],"category_scores_gemma":[0.1187858,0.0005633257,0.00126671,0.001197839,0.0008768636,0.007737434,0.004348249,0.003622804,0.001621278],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001006835,"about_ca_system_score_gemma":0.001170649,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.004405248,"about_ca_topic_score_gemma":0.006128645,"domain_scores_codex":[0.9884993,0.006745114,0.0008258215,0.001946342,0.001744723,0.0002386297],"domain_scores_gemma":[0.8620288,0.1227641,0.003440407,0.007309999,0.003155198,0.00130151],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"bench_or_experimental","study_design_scores_codex":[0.004110587,0.003144355,0.07030745,0.005575985,0.002640991,0.0007052181,0.007801647,0.222652,0.02673683,0.009149984,0.03399494,0.61318],"study_design_scores_gemma":[0.0004184911,0.002215125,0.02784866,0.0003312774,0.0005424622,0.0006770639,0.001707829,0.8882799,0.02699069,0.02899737,0.02167418,0.0003170409],"study_design_candidate":"bench_or_experimental","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.6141039,0.005460775,0.3239893,0.001548462,0.0005868205,0.001661001,0.01028457,0.03137365,0.01099155],"genre_scores_gemma":[0.804905,0.0004759424,0.176247,0.0005926826,0.0001047168,0.0007435082,0.0131176,0.0007929933,0.003020585],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.0137675,"threshold_uncertainty_score":0.07281035,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.06529112178130436,"score_gpt":0.3471593056174225,"score_spread":0.2818681838361181,"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."}}