Meta's new structured prompting technique makes LLMs significantly better at code review — boosting accuracy to 93% in some cases - 新闻列表