如何与大语言模型协作进行写作

有效利用大语言模型进行写作需要遵循明确的原则。首先要认识到的是，不应该采用LLM直接建议的任何措辞1，因为这类生成短语往往会让文章陷入"恐怖谷"。其次，应当避免接受LLM过度的鼓励性反馈1，这样反而会强化初稿中的缺陷。

最有效的做法是将LLM定位为编辑工具而非代笔人1。具体工作流程建议为：先要求模型指出问题所在，然后自己重写相关段落，最后利用不同的模型对比原文和修改版本1。LLM在识别常见写作问题上表现出色，包括过度使用被动语态、虚词（如very、unfortunately、really、actually）的滥用，以及可以重新排列的段落结构1。

An effective approach to leveraging large language models in the writing process hinges on two fundamental principles 1. First, writers should reject any wording that an LLM suggests, as the phrases generated by these models tend to push text into an uncanny valley that undermines quality 1. Second, one must resist the temptation to accept excessive encouragement from the AI, which can reinforce the weaknesses of an initial draft 1.

Rather than viewing language models as ghostwriters, writers should repurpose them as editorial assistants 1. The recommended workflow involves directing the model to identify specific problems, then rewriting the problematic passages independently, followed by having a different model compare the original and revised versions to evaluate improvements 1. Language models excel at detecting mechanical issues such as overuse of passive voice, excessive deployment of filler words like "very," "unfortunately," "really," and "actually," and paragraph sequences that could be rearranged for better flow 1.