{"id":"W4402897200","doi":"10.1109/qrs62785.2024.00033","title":"Weaknesses in LLM-Generated Code for Embedded Systems Networking","year":2024,"lang":"en","type":"article","venue":"","topic":"Embedded Systems Design Techniques","field":"Computer Science","cited_by":8,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Waterloo","funders":"","keywords":"Computer science; Code (set theory); Programming language","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":[0.006666998,0.0007526058,0.0003444791,0.002361589,0.0009107396,0.002170439,0.00156044,0.001354622,0.001891506],"category_scores_gemma":[0.04979987,0.000757881,0.0008851481,0.001172352,0.002832208,0.003749203,0.002578429,0.002138589,0.0005681306],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00212736,"about_ca_system_score_gemma":0.002044386,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001872946,"about_ca_topic_score_gemma":0.003371236,"domain_scores_codex":[0.99112,0.002613486,0.0007275373,0.0009192986,0.004187459,0.0004321605],"domain_scores_gemma":[0.9583204,0.02266309,0.003906015,0.009290866,0.005463397,0.0003561458],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.001016618,0.0007089787,0.1109783,0.001986142,0.0002274015,0.003539666,0.02033629,0.07161661,0.1159628,0.230655,0.01789168,0.4250805],"study_design_scores_gemma":[0.0001073693,0.0008171146,0.01699363,0.001691179,0.0002674269,0.004250494,0.002608933,0.5659187,0.1810847,0.1218371,0.1041151,0.0003082666],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.361802,0.0004053934,0.5993068,0.002615635,0.0001215305,0.0007439812,0.001015777,0.01787493,0.01611395],"genre_scores_gemma":[0.6340088,0.0001558565,0.3572422,0.0005905743,0.00001694621,0.0003639464,0.00120013,0.002502155,0.003919384],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.006666998,"threshold_uncertainty_score":0.03525889,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.04146562305789299,"score_gpt":0.3035065337521313,"score_spread":0.2620409106942383,"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."}}