用自定义指令让 ChatGPT 不再编造信息
Stop #chatgpt from lying to you with this custom prompt. #chatgpttips #aitips #chatgptprompts #BetterWithAI
视频介绍在 ChatGPT 的个性化设置里填写自定义指令，让模型在不确定时直接说不知道、不编造信息，并给出答案的置信度、核实链接和引用来源。操作路径是点击左上角菜单、左下角账号图标，进入个性化设置后下滑找到自定义指令。作者称这段指令相当于给模型“允许犯错”的许可，从而减少幻觉和错误回答。
视频演示了在 ChatGPT 自定义指令里填入一段提示词，让模型在不确定时直接说不知道并标注置信度，适合跟着录屏一步步设置。
粉丝 674
- 浏览
- 6.6万
- 点赞
- 2405
- 评论
- 99
- 分享
- 1164
- 收藏
- 2792
点赞是作者粉丝数的 3.6 倍，是低粉爆款：火的是内容本身
平台标注：AI 生成内容
数字来自平台，为最近一次抓取时的数据。
以下是视频口播的自动转写，可能有错字。
It's become a running joke about ChatGPT giving wrong instructions and then apologizing for it later. Here's how to fix that so it won't hallucinate and won't lie to you.
Within ChatGPT, click on the menu in the top left and then your account icon in the bottom left. then click on personalization.
Scroll down until you see custom instructions. Save and rewatch this video to copy the exact prompt or use the link in the bio.
Essentially, the instructions tell it that it has permission to fail, never fabricate information. If you don't know, say so.
Give your confidence level in the answer. Verify claims, check that urls work, cite references. Say I don't know, it gives it permission to be wrong.
Follow me for more practical tips on getting better with AI.
来源：TikTok · 搜「chatgpt」 · tiktok.com