Reverse Prompt Engineering in ChatGPT
HTML-код
- Опубликовано: 15 фев 2023
- Better prompts mean better output. But how do you create better prompts for ChatGPT?
With the help of ChatGPT, of course!
I'm going to show you how to train ChatGPT to help you create the prompts you need for the output you want. As a bonus, I'm going to show you how to make sure the prompts are using entities to produce SEO optimized content.
#chatgpt
#seo
Sign up to receive more tips and strategies to level up your SEO at theseopub.com/
Connect with me: / mikerfriedman
My Toolbox:
🔎 Semrush - Competitor and Keyword Analysis - theseopub.com/semrush
✔ Monday.com - For task management and organizing all of my client work - theseopub.com/monday
🗄 Frase - Content optimization and article briefs - $1 trial offer - theseopub.com/frase
👑 ContentKing App - Site crawler, monitoring, and audit tool - theseopub.com/contentking
Sometimes I tweet about SEO, beer, baseball, and other topics at / theseopub 
Наука
I seldom comment on videos even if I liked a video, but I had to comment on this one. Just mind blowing man. Recently found you and loving your content. 👍
Thank you. I appreciate that.
Thanks. Very helpful. I like the way you present it as well. Some make too much of a show. Yours is calm and rich. Looking forward to see more.
Thanks. I appreciate the feedback.
awesome 👌👌
I definitely found this video very helpful! Also, I suspect I saw that same guy's video that you referenced.. Reverse prompt engineering is
going to become an incredible skill to have. Thanks for the demonstration, I hope you make more demonstrations like this.
Before this, I was calling it the Seinfeld Movie Phone tactic. I might be dating myself a bit, but if you ever saw the episode where Kramer ends up with the phone number for Movie Phone, you may get the reference.
This is genius, thanks for sharing!
No problem. Thanks for stopping by.
Good one!
Thank you! Cheers!
Very interesting and helpful thank you very much. Please make the sound louder next time.
very interesting
🤘 Astonishing content Man 💯💯 Thanks to share your knowledge with us 🥳 Thumbs Up from Brazil, yeeeh ( plus +1 sub)
Thank you!
Awesome
Thank you.
Nice done with this one! :) I have been following you around ;) And I think you always follow through. But I have a few questions.
1) How often (or how much more content do you get through) as a paid user of chatGPT? I see you fed it an entire article :)
2) And why would you ask it to exclude any relevant entities (such as 5-6) ?
3) How many entities would you ask for (you just said a "few more"). Do you have any max of how many entities you ask for ?
Thanks for this awesome video!
1. I'm not sure off the top of my head what the limits are. When I hit a limit, I usually just divide up what I'm trying to feed ChatGPT.
2. I didn't think those entities were important or all that relevant.
3. Most of the time when I'm creating article briefs, I have a list of 15-20 entities. I don't have a set number though. It really just depends on the topic.
Thank you very much. It was really great. I'll let you know how it works out. Got a big project ahead of me. I think it will do exactly what I need to do. Take care and God bless.
Something may have changed with ChatGPT 3.5 this past month, as following your steps to the tee, when I ask for it to "Now I want you to reverse prompt engineer the {text}. Give me a single prompt that would create a similar output. The tone and writing style should be included in the prompt. The entities #1-#18 should also be mentioned as needing to be included in the output of the prompt" it just goes off and returns some generic output on crafting a prompt, on a totally unrelated topic i.e. it completely forgets what it wrote beforehand re: Salience table and the text used to extract table from. Have tried using minimal text to avoid exceeding token limits, but still same results.
The concept still works, and highlighting the concept was the real point of this video. You may have to edit prompts slightly as ChatGPT updates. I got the correct output by editing the prompt to "Now I want you to create a prompt for me by reverse prompt engineering the {text}. Give me a single prompt that would create a similar output to {text}. The tone and writing style should be included in this prompt. The entities #1-#18 should also be mentioned as needing to be included in the output of the prompt."
great video but the audio is soooo low
I tried the same thing with a text about neural nets but somehow ChatGPT decided to write me a prompt about the "benefits of practicing gratitude in daily life" 🤔
I would try it again and look at the content you are feeding it. Sometimes it just goes off on a whacky tangent about something else. Not very often though.
You also may have to give it additional information about what neural nets are. ChatGPT is not trained on every subject or topic. It could just be something ChatGPT is not all that familiar with.
@@TheSEOPub Thanks for the suggestion, I'll try it. My guess was that I fed it too much text (even though the original is much larger) and depleted its conversational memory.
@@whozz I don't know. I've done a few chats with it where I fed it about 15-20 documents to train it to act as a specific brand. Never had an issue with feeding it too much. Yet...
@@TheSEOPub wow! I didn't know one could use this much input. I once tried to prompt a large text, but it asked me to provide a shorter one. Is there a proper way to chain multiple prompts when the input is large?
@@whozz I don't think there is a proper way. I have had to sometimes break a document in half and give it half at a time. I forget what the character limit is for a single input, but there is a limit.
Reverse engineering can be a great way of learning how to engineer better prompts. However, if we are simply reworking other people's online content using reverse engineering to avoid plagiarism checkers, we'll just be adding redundant information to the web, creating more noise in the system, and diluting the signal of genuinely new content. Unless we're adding transformative value to the original content by introducing new insights, perspectives, topics, etc, it would be better to simply share a link to the original source. Regardless, if there is a strong resemblance between the new article and the original one, I would at least provide a reference to the original article.
Outlets such as the “Clarkesworld” magazine has seen massive increase in spammy submissions that are obviously written by an AI.