From Guessing to Placeholding: A Cost-Theoretic Framework for Uncertainty-Aware Code Completion
arXiv:2604.01849v1 Announce Type: new
Abstract: While Large Language Models (LLMs) have demonstrated exceptional proficiency in code completion, they typically adhere to a Hard Completion (HC) paradigm, compelling the generation of fully concrete code…