This prompt will make sure ChatGPT tells the truth all the time even when you already think it's doing so

This prompt will make sure ChatGPT tells the truth all the time even when you already think it's doing so 图片 1
This prompt will make sure ChatGPT tells the truth all the time even when you already think it's doing so 图片 2
This prompt will make sure ChatGPT tells the truth all the time even when you already think it's doing so 图片 3
This prompt will make sure ChatGPT tells the truth all the time even when you already think it's doing so 图片 4

ChatGPT can sound incredibly confident, but that doesn’t always mean every fact it generates is correct, so here’s a prompt that will make sure that the chatbot is always telling you the truth.

You might think that ChatGPT is a trusty companion, and while it is useful, it gets caught every now and then telling some porkie pies.

One AI expert has shared a prompt designed to reduce made-up facts, fake sources, and unverified claims before they ever appear in your responses.

It could help anyone using ChatGPT for work, school, or everyday questions avoid publishing inaccurate information.

This prompt will make sure ChatGPT tells the truth

Large language models like ChatGPT are designed to be helpful, and they’re the assistants of the future, which means they often try to answer your question rather than admit they don’t know something.

According to a recent Forbes article by Coachvox founder Jodie Cook, that can sometimes lead to “hallucinations”, where an AI generates information that sounds convincing but isn’t actually true.

That could include fake studies, incorrect statistics, fictional customer testimonials, or quotes that were pulled from thin air.

Supercar Blondie

The article points to research from OpenAI and Georgia Tech that suggests the way AI models have traditionally been evaluated can reward plausible guesses over uncertainty.

In other words, a model that invents a believable answer may score better than one that says it doesn’t know.

But it turns out there is a simple prompt that tells ChatGPT exactly where the line is.

Unsplash

Instead of correcting the AI after it responds, the prompt changes its behavior from the very beginning.

It tells ChatGPT to make honesty its highest priority, clearly admit when it isn’t sure, avoid inventing studies, experts, companies, links, or customer examples, and flag any statistics or performance claims that still need verification before presenting them as fact.

The goal is to remove the pressure for the…

添加评论
点赞收藏
点踩分享查看原文
评论
?
参与讨论