{"id":"W4389519518","doi":"10.18653/v1/2023.emnlp-main.844","title":"Aligning Large Language Models through Synthetic Feedback","year":2023,"lang":"en","type":"article","venue":"","topic":"Topic Modeling","field":"Computer Science","cited_by":13,"is_retracted":false,"has_abstract":true,"ca_institutions":"Kootenay Association for Science & Technology","funders":"Institute for Information and Communications Technology Promotion; Ministry of Science and ICT, South Korea; Korea Advanced Institute of Science and Technology","keywords":"Computer science; Reinforcement learning; Language model; Artificial intelligence; Quality (philosophy); Machine learning","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":[0.003399259,0.001555664,0.0007506529,0.0005290116,0.0004244166,0.001070438,0.00186603,0.001650726,0.003892226],"category_scores_gemma":[0.02134588,0.0005624742,0.0006732002,0.0005247626,0.0009783715,0.002217606,0.001844297,0.002387623,0.001667159],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001058606,"about_ca_system_score_gemma":0.001348995,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.004845713,"about_ca_topic_score_gemma":0.007662251,"domain_scores_codex":[0.9972422,0.001682504,0.00008156432,0.0005971376,0.0002693766,0.0001271938],"domain_scores_gemma":[0.9916961,0.006213357,0.0003509527,0.0008592965,0.0006440915,0.0002362658],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"simulation_or_modeling","study_design_gemma":"bench_or_experimental","study_design_scores_codex":[0.0005890268,0.000293489,0.003655614,0.0004259687,0.00009971696,0.0003509706,0.000536404,0.8500321,0.01196368,0.0081308,0.009584054,0.1143382],"study_design_scores_gemma":[0.00003403852,0.0000740865,0.0001974201,0.00001348937,0.000008813563,0.00003156023,0.00004185715,0.9914429,0.003123114,0.003644327,0.001375302,0.00001312381],"study_design_candidate":"bench_or_experimental","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.2028122,0.0009697346,0.7681354,0.001347087,0.000291971,0.0002935932,0.001525823,0.01884747,0.005776727],"genre_scores_gemma":[0.8463284,0.0001948521,0.1448241,0.0004534242,0.00007303345,0.0004653355,0.00304657,0.0009452489,0.003668926],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.004845713,"threshold_uncertainty_score":0.01797718,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.03652306064352353,"score_gpt":0.2776569441957399,"score_spread":0.2411338835522163,"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."}}