Why ChatGPT lies

@bengusberg

11.5K followers · usually 3.6K views a post

Views

78.4K

Likes

4.4K

Comments

141

Shares

1.4K

Why it worked

The video provides immediate, actionable value by solving a common pain point (AI hallucinations) using a credible source (MIT), which encourages viewers to save the video for later reference.

Summary

The creator explains how to use a specific prompt developed by MIT researchers to improve the accuracy of ChatGPT responses. He provides a step-by-step guide on how to input these custom instructions into the ChatGPT settings.

Structure

1Identify the problem of AI misinformation
2Introduce the MIT truth prompt solution
3Provide a step-by-step tutorial for ChatGPT settings
4Explain how the prompt forces confidence scoring
5Final call to follow for more AI content

Call to action

follow me for more on AI, founder life and leadership

On-screen text

Why ChatGPT lies The problem with Generative AI that causes it to spew misinformation CHATGPT LIES STEAL MIT'S FIX I'M BEN GUSBERG WE'VE RAISED $4 MILLION THE FIX IS CALLED THE TRUTH PROMPT IT FORCES CHAT TO ALWAYS BE HONEST WITH YOU OPEN CHATGPT CLICK YOUR NAME PERSONALIZATION CUSTOM INSTRUCTIONS AND PASTE THIS THE PROMPT FORCES CHAT TO REVIEW ALL THE ANSWERS ANALYZE FOR ANY GAPS AND GIVE YOU A CONFIDENCE SCORE IF CONFIDENCE IS LESS THAN 0.8 HIT SAVE AND NOW GOING FORWARD CHAT WILL BE FORCED TO ALWAYS TELL YOU THE TRUTH I'M BEN GUSBERG FOLLOW ME FOR MORE ON AI FOUNDER LIFE AND LEADERSHIP

Original caption

Here’s how to make ChatGPT be honest #chatgpt #ai #startup #founder #aitips

More from @bengusberg

Their other posts in the index, biggest breakout first.

Transcript

Why ChatGPT lies. The problem with Generative AI that causes it to spew misinformation. Steal MIT's fix to always get the truth. I'm Ben Gusberg. I'm a Stanford founder of an AI startup. We've raised $4 million. The fix is called the truth prompt and it was developed by MIT researchers. It forces chat to always be honest with you. To set it up, open ChatGPT, in the bottom left click your name, go to settings, personalization, custom instructions, and paste this prompt in. The prompt forces chat to review all the answers, analyze for any gaps, and give you a confidence score. If confidence is less than 0.8, it has to explicitly tell you. Hit save and now going forward chat will be forced to always tell you the truth. I'm Ben Gusberg, follow me for more on AI, founder life and leadership.