{"id":"W4390573228","doi":"10.48550/arxiv.2401.00907","title":"LaFFi: Leveraging Hybrid Natural Language Feedback for Fine-tuning Language Models","year":2023,"lang":"en","type":"preprint","venue":"arXiv (Cornell University)","topic":"Topic Modeling","field":"Computer Science","cited_by":1,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"Natural Sciences and Engineering Research Council of Canada; University of Alberta; Alberta Machine Intelligence Institute; Compute Canada; Mitacs; Canadian Institute for Advanced Research","keywords":"Computer science; Fine-tuning; Realm; Natural language; Natural (archaeology); Language model; Natural language understanding; Task (project management); Artificial intelligence; Natural language processing; Engineering","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":[0.005645337,0.002848621,0.001368273,0.001466567,0.000644299,0.001562826,0.003506986,0.002447597,0.003233025],"category_scores_gemma":[0.02448189,0.0007126066,0.001376763,0.0008290716,0.000779486,0.004338263,0.002572168,0.003930227,0.003538781],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001300956,"about_ca_system_score_gemma":0.001504397,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.006543498,"about_ca_topic_score_gemma":0.0145997,"domain_scores_codex":[0.9967113,0.001470603,0.0001463376,0.001094622,0.000399018,0.0001780888],"domain_scores_gemma":[0.9892902,0.007196391,0.0003775581,0.001816427,0.001025125,0.0002944733],"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":[0.00089001,0.0009477601,0.005667846,0.0006992581,0.0004592493,0.0003414123,0.0008557862,0.1876508,0.04312557,0.003971949,0.0426764,0.712714],"study_design_scores_gemma":[0.00009235513,0.0001516034,0.0004976855,0.00002572102,0.00004034046,0.00006946782,0.00004930569,0.9787756,0.008721614,0.007321467,0.004207539,0.00004726256],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.06347111,0.002574746,0.8297341,0.001469042,0.0005334514,0.0004021163,0.002405089,0.09575179,0.003658685],"genre_scores_gemma":[0.5030801,0.0005529345,0.4719528,0.002118845,0.0004156669,0.001010426,0.01037899,0.003291328,0.007198828],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.006543498,"threshold_uncertainty_score":0.02985573,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1021838348557336,"score_gpt":0.2127921687248835,"score_spread":0.1106083338691499,"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."}}