Advanced ChatGPT Prompt Engineering

I had a conversation with it last night about emotional triggers charities use when marketing to groups of people/individuals vs. companies/corporations. It was a thought provoking conversation, and I actually learned a lot. We all know the basics about how charities work, but it went into areas I hadn't quite thought of.

That being said, the biggest issue I faced with having a "conversation" with an AI languuage tool was the use of personal pronouns and phrasing questions like I was talking to a person. It was quick to remind me on several occasions, that it didn't have feelings, opinions, or ideas nor does it have consciousness.

it was also clear to provide a disclaimer at the end of almost every query.

It did however tell me that while it didn't possess the human traits mentioned above it was "curious" but couldn't learn directly from our conversation. Rather the (as it put it) "complicated algorithm" and data its creators had already provided was the basis for its knowledge and that structure of language would continue to get better once it was updated.

I also felt kinda capped at the knees in my banter with it, several times it mentioned the morality of transparency and seemed to be more afraid of someone misunderstanding the information it presented rather than providing information or conversation that directly answered the question.

If I was directly testing it by asking for information that was inherently immoral or questionable I would understand its ethical posturing. However when discussing loan applications, resumes, business plans, etc. It felt the need to remind me constantly of the ethical and moral implications of using the information in a dishonest way.

So on the one hand it wouldnt answer questions that had you in it, like how would you "XYZ", because it wasn't conscious, but didn't hesitate to tell me about morals and liability.

Anyway, its a great tool for copy, and almost anything language based and when the personal pronouns were removed it answered the questions just fine. It wasn't designed for "human exchange" but its a bit tricky because after a while you feel like you're talking with a person but the moment you make it personal it shuts down in a way destroying the illusion.

Thanks for sharing all of this info with us, after last night it was exactly what i was looking for.
 
So on the one hand it wouldnt answer questions that had you in it, like how would you "XYZ", because it wasn't conscious, but didn't hesitate to tell me about morals and liability.

Anyway, its a great tool for copy, and almost anything language based and when the personal pronouns were removed it answered the questions just fine. It wasn't designed for "human exchange" but its a bit tricky because after a while you feel like you're talking with a person but the moment you make it personal it shuts down in a way destroying the illusion.

Thanks for sharing all of this info with us, after last night it was exactly what i was looking for.

If you are dismissing the conversational aspect of prompting, I would reconsider that quickly.

I think it depends on where you're going in the conversation.

Again, it depends on your purpose, though. Writing copy does not require a conversation with AI. That's where the limitations end. Methinks.

nt1
 
Writing copy does not require a conversation with AI. That's where the limitations end. Methinks.

nt1
I completely agree, I was just messing with it and testing its limitations.
I've read a lot about Chat GPT destroying google's business model ( I think google mentioned it as well within 2 years) and on top of that the creators keep talking about how "personalized" the conversations with GPT is and that it feels like a conversation you would have with a peer or teacher.

If you listen to any of the podcasts with the open ai or their peers, they have a lot to say about its conversational flow and its method of remembering and referencing prior answers or questions... kinda the thing that sets it apart from software.

So rather than have it just spit out txt, I tried using it like I would a search engine then I tried having a conversation with it to see how it responded. By conversation I dont mean something like "hi, how are you?" "Are you having a good day?" "Do you like chocolate?" "How do you feel about humans?" etc.

More like conversation you would have with a professor or coach where you have a topic you are interested in like inflation, economics, marketing etc.

I don't see how it will replace google, thats kind of like the library books replacing the dewy decimal system but I can see Ai changing how we search.

What I was surprised by was the fact that having a conversation with it really did feel like I was talking to a teacher/expert and when it answered my questions, rebuttals, and theories without holding back I felt like I was actually learning something in a different way. Rather than from a static page of facts and opinions it felt dynamic and expansive. I didn't have to wade through 5 paragraphs on the economy to find the answer I was looking for, I just asked and it answered. it didn't feel limited in that regard.

Keep in mind, Im no expert... Just a guy who tried out chat gpt after hearing/reading so much hype about it. Hope that clears it up a little. I dont do so well when it comes to expressing myself or my thoughts.

Cheers
 
What I was surprised by was the fact that having a conversation with it really did feel like I was talking to a teacher/expert and when it answered my questions, rebuttals, and theories without holding back I felt like I was actually learning something in a different way. Rather than from a static page of facts and opinions it felt dynamic and expansive. I didn't have to wade through 5 paragraphs on the economy to find the answer I was looking for, I just asked and it answered. it didn't feel limited in that regard.

Keep in mind, Im no expert... Just a guy who tried out chat gpt after hearing/reading so much hype about it. Hope that clears it up a little. I dont do so well when it comes to expressing myself or my thoughts.

Cheers

You're great. We are on the same page.

That part that you said, about learning something in a different way. - it's learning non-linearly, which is, by definition how AI operates. I'm no expert either.

The hype is understandable, and overwhelming.

Anyway, with memory etc, the secret, I think, is the minute details in the prompts.

I don't think we'e there yet.

Right now, people are just liek "OMg, I can make 4,332 books, today!?!?! LEtsdoit!"

I think we're seeing a shift form the logical to the creative.
The left to the right brain.

Writers worry, but I think this is really a time where writers can shine.
 
I really love this type of scripts because they make everything simple, people say chatgpt will make you rich and give really good ideas but people ask to chatgpt, how to make a lot of money lol, the script you share show that a person with solid knowledge on a specific area can make chatgpt a formidable tool, the AI is good enough like the person that use it, thanks for sharing this script!!!
 
I stumbled on this and thought it could be useful for the BHW community.

ChatGPT Tips & Tricks​

Maximum Response Length Work Around
If text gets cut off, try using the OpenAI Playground and adjusting the maximum length setting. This can help ensure that ChatGPT is able to provide complete and accurate responses to your queries.

Breakdown Queries
To create a more engaging and interactive conversation with ChatGPT, try breaking down your queries into sub-queries. For example, you might start by asking ChatGPT to list some recipes, then ask for a specific recipe by number, and finally ask for modifications to make the recipe more spicy. This creates a “conversational journey” that can be more engaging and enjoyable.

Set A Persona
To give your conversations with ChatGPT a more personalized touch, try setting a persona for the model to use. For example, you might say “You are a physics professor, explain in detail…” to give ChatGPT a more specific role and personality. This can make the conversations feel more realistic and engaging. There are examples of this in the Advanced ChatGPT Prompts section


ChatGPT Use Cases & Prompts​

Half the battle with ChatGPT is knowing what you can use it for. In a years time this will probably be obvious but new use cases are being discovered every day. Here is a big list of basic prompts that I brainstormed and found from various sources… including ChatGPT of course



ChatGPT Comebacks​

ChatGPT has a memory function, unlike a search engine where each query is isolated, future prompts take into account the entire conversation history.

This is quite powerful as you can discuss a certain topic and build ChatGPT’s knowledge throughout the conversation.

Often you wont get a perfect response first time so you can then throw comebacks at it to refine the output. Here are some examples:



Advanced ChatGPT Prompts​

Now let’s get into formatting some more advanced prompts. You can copy and paste these in to use as they are or craft them to your particular use case by adding details such as information about who the content is for.



The above will output 20 headlines using general good practices for writing headlines. Choose one from the list, put it back in and ask it to make variations.


In this example we are using standard copywriting practices to create a better sales letter that engages the reader. We have been clear and specific in the prompt.

ChatGPT is a machine, it’s not magical, the quality of the output is correlated to the clarity and detail of your input.


Building With OpenAI API​

You can also sign up for an OpenAI API key and use their tools from within an API. Here is a simple example (coded by chatGPT) of how this could be done in NodeJS




This opens up a lot of possibilities to build on top of this incredible technology. 2023 will be a breakthrough year for AI and tools like this are changing the way we work. In my opinion it’s well worth taking some time to study them and come up with your own unique use cases & prompts


Disclaimer: Copy from the internet. Sorry, I never share links on threads. You can always Google it! :cool:
Shortened for anyone who’s lazy

Maximum Response Length Work Around:

  1. Go to the OpenAI Playground website.
  2. Enter your query in the "Input" field.
  3. Under the "Model" field, select "Davinci" or "Curie" (these are the two most powerful language models available).
  4. Under the "Settings" field, adjust the "Max tokens" setting to increase the maximum length of the response.
  5. Click "Submit" and wait for the response to appear in the "Output" field.
Breakdown Queries:

  1. Start by asking a general question or making a statement.
  2. Based on the response you receive from ChatGPT, ask a follow-up question that is more specific or narrows the focus of the conversation.
  3. Continue to ask sub-queries that build on previous responses and lead to a more detailed and engaging conversation.
Set A Persona:

  1. Before asking a question, specify a persona for ChatGPT to use by including a description of the person or role you want ChatGPT to embody.
  2. Phrase your question or statement in a way that is consistent with the chosen persona.
  3. ChatGPT will respond in a way that is consistent with the chosen persona, creating a more personalized and engaging conversation.
 
interesting, indeed sometimes I have problem to formulate correctly the prompt, i'll use many of the prompts written here. thank you sir
 
Back
Top