{"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":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0002675088,0.00009401883,0.0001074273,0.00005601396,0.00009147785,0.0001017076,0.0006336517,0.00004250767,0.00004657113],"category_scores_gemma":[0.00001848467,0.00008220662,0.00004919387,0.0003862574,0.000009165423,0.0005622642,0.0004170998,0.00007534874,0.0007558566],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00001736817,"about_ca_system_score_gemma":0.0000216433,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00007294107,"about_ca_topic_score_gemma":0.00001081079,"domain_scores_codex":[0.9988852,0.00002973336,0.0001463362,0.0003438085,0.0002242677,0.000370726],"domain_scores_gemma":[0.9992312,0.00006005093,0.00002576977,0.0006157688,0.00002105791,0.00004618478],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[7.174344e-7,0.00002356807,0.00004105211,0.00001855181,0.00001349207,0.00006688014,0.01101613,0.01467327,0.001044282,0.9575313,0.003616025,0.01195477],"study_design_scores_gemma":[0.0001301144,0.000006511393,0.00001706543,0.0000156362,0.000001810454,0.00000593532,0.0004338899,0.9673359,0.001689948,0.02923051,0.001008161,0.0001245455],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.02110901,0.0000659357,0.9308829,0.001060884,0.0002251449,0.00007229453,0.000001167486,0.0009963452,0.04558636],"genre_scores_gemma":[0.913347,0.00001103811,0.08157361,0.00068925,0.00005752087,0.000009095914,0.00000175897,0.00001094192,0.004299758],"genre_candidate":"methods","genre_consensus":null,"teacher_disagreement_score":0.9526626,"threshold_uncertainty_score":0.9715259,"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."}}