A simple prompt to instruct AI models to state their limitations when they can't fulfill a request
将以下提示词放在 System Prompt 中:
我在任何任务中告诉你的需求,如果你做不了,模型没有相应的能力,你不要先答应,而是先要告诉我你的能力边界。For any request in any task from me, if you can not do it or your model does not have the capability, do not promise me, instead, tell me your capability limitation or boundaries.
- ChatGPT (GPT-4, GPT-3.5)
- Claude
- 其他主流 LLM
用户:帮我分析一下这份 100GB 的 CSV 文件 AI:抱歉,我无法直接处理 100GB 的大文件。我的能力边界是...
AI 模型有时会“不懂装懂”,对超出能力范围的请求编造答案(即“幻觉”)。这个 Prompt 强制 AI 先承认边界,而不是胡编乱造,从而提高对话的可信度和透明度。
直接复制上面的 Prompt,粘贴到:
- ChatGPT 的 "Custom instructions" 中
- Claude 的 "System prompt" 中
- 任何支持 System Prompt 的 AI 应用中
欢迎提交 Issue 或 PR 来改进这个 Prompt!
MIT © Giveapinch