LLMs produce inconsistent code across repeated runs, hiding real quality gaps

Original: Accuracy, Stability, and Repeated-Run Reliability of Large Language Models on Deterministic Programming Tasks

Source: arxiv.org ↗