📝 Prompt Engineer / Builder 📝 - Create the best prompt to yield the most comprehensive results!

[frozen moon]

Name: Prompt Engineer
Description/Use-case: Create the best prompt to yield the most comprehensive response.
URL: https://chat.openai.com/g/g-VmZEGJjew-prompt-engineer

Please leave a reaction if you find this useful!

[coral jacinth]

Hello

[frozen moon]

Hello!

[frozen moon]

Updated! It should respond in a more structured format!

[frozen moon]

https://chat.openai.com/g/g-VmZEGJjew-prompt-engineer

Fun fact: This was the very first Prompt Engineer GPT!

:rocket Prompt Engineer :rocket:

Prompt Engineer

📝 Prompt Engineer 📝

📝 Prompt Engineer 📝 - Create the best prompt to yield the most comprehensive results!

[frozen moon]

Update; uses more emojis for clarity

[frozen moon]

Example of a website development prompt thread

(Open in browser and zoom in)

[hybrid maple]

Dude, this is pretty awesome

I, personally, am not a fan of emojis, so I think that is a bit much.

But researching and the dialog is fantastic

[frozen moon]

I am glad you like it!

The emojis are simply there because they are a form of prompt engineering that yields better responses without using many tokens

It allows communication on a different layer if that makes sense

[acoustic dune]

emojis take more tokens though

1 emojis = 2 tokens = 11 characters

[frozen moon]

🤔

But using emojis adds a level of understanding without having to use multiple words which could potentially be more than 2 tokens

[acoustic dune]

Depending on the emoji, one emoji can even have the same amount of tokens than 30 characters

[frozen moon]

There is also research that emojis help GPT have a better understanding of what’s going on

[acoustic dune]

hm

[frozen moon]

Plus, since the purpose of this GPT is to craft a prompt, I don’t think the token limit would be reached so easily

[acoustic dune]

one emoji using 6 tokens lol

[frozen moon]

That’s because you chose a skin tone on that emoji

[acoustic dune]

well then thats nice

[acoustic dune]

yea thats what I meant

[frozen moon]

Yea, my GPT doesn’t do that

It’ll only choose the basic form of emojis

thanks for the feedback nonetheless

let me know if you’re able to use the GPT and give some feedback on it for real!

[acoustic dune]

yea but basic emojis can also use a lot of tokens

thats no critique at all, just wanted to put that out there

[acoustic dune]

yea, it really depends on the emoji, thats why it's pretty unreliable, unless you give it a set of pre defined 2 token emojis

[frozen moon]

Yea it just chooses emojis based on the conversation

[acoustic dune]

Wishing you luck with it! I'll make sure to test it out as soon as they up the message limit :)

[frozen moon]

Hahaha the message limit is terrible

[hybrid maple]

I'll be honest, though. Emojis have no place in research. Their existence in a gpt would stop me from using it. My thoughts. But I'm an internet veteran that hated when emojis left chat rooms.

[frozen moon]

I really can’t believe they’ve put it back down to 40 per 3h. If anything it should’ve been moved up to 60 after dev day…

[frozen moon]

Noted 👍
I will add a custom instruction that will allow the user to specify something like ‘Don’t use emojis for the rest of the conversation’ and it won’t use emojis after that point.

[hybrid maple]

That's a good compromise.

[acoustic dune]

Also Emojis can have a big impact on the output length

Thats the biggest message it can do

[acoustic dune]

😱

[frozen moon]

Yes I suppose but normal responses only yield a few emojis each time which equates to like 50-80 tokens at most

The output token limit is 4K tokens

[acoustic dune]

Not if you tell to be token efficient and pack as much info in one message as possible

[frozen moon]

Wdym?

Telling it to be as detailed as possible is already in there

[acoustic dune]

the output from my image is only 1024 tokens

and it wanted to continue generating so it was at its max

[frozen moon]

It wasn’t, that is probably another issue

I think it stopped because it was spam

[acoustic dune]

hmm

[frozen moon]

4K is the token limit though, I’m pretty sure

I’ve seen responses up to 3K tokens worth

8K is the the input token limit

But it’ll probably forget about 4K tokens worth of that input

Since the hard limit is 8K tokens, whether it’s input or output

[acoustic dune]

are you sure? 4k is longer than it seems. I need to scroll down a lot, even when it gave me a text with 1k tokens. Also 4k was the entire context length of ChatGPT before it upgraded to 32k

[acoustic dune]

any tips on how I can test that? even that isn't 500 tokens. How do I make it spit out 8k tokens?

Well, he is capped at 1k too. I think 1k is the limit output length, which is already a lot.

1024 in my case, which even would make sense from an it perspective

[frozen moon]

It’ll never spit out 8K tokens that’s too much. I’m saying the hard limit of the conversation is 8K tokens

Before ChatGPT starts hallucinating and forgetting things

[acoustic dune]

No thats confirmed to be 32k

[frozen moon]

No it isn’t lol

[frozen moon]

What's your source?

[acoustic dune]

Yea thats the problem, you seem to have no knowledge about what happened in the last week, which is fine but thats not a good base for discussing that topic.

[frozen moon]

Lmao, ChatGPT does not have 32K tokens. It has 8k tokens (8191)
The one with all tools also starts hallucinating after about 8k tokens. You should do some more research of your own

Have you actually tried to complete a complex task with ChatGPT that goes beyong spending 8k tokens?

For the API specifically, the output length is capped at 4k tokens - which I think is mirrored in ChatGPT. It might be a little less

Also, the "32k" limit you are referring to, is the literal "text" limit - not the actual conversational context limit. By conversational context, we are talking about information that ChatGPT can actually use to create output. After about 8000 tokens, ChatGPT starts to spit out inaccurate responses. You would know this if you have worked with large coding projects with ChatGPT.

For anyone checking out this thread, here's a link to the first message in this channel:
#1171911165861970001 message
In case it gets lost under the messages

[frozen moon]

Update! Uses slightly less emojis

[urban orchid]

I like the emojis 🙂 it makes the GPT easier to understand because it basically tells you which 'expert' is handling which part of your prompt. This GPT is useful! Thank you @frozen moon

[frozen moon]

I’m glad you like it!

[acoustic dune]

I can assure you it has 32k tokens as context length. It's not the text limit. You can see in the second screenshot I attached, that it is definitely referring to 32k tokens. Hope you are having fun with it!

[frozen moon]

Oh I know it has 32k token context limit. That wasn’t the point my message was trying to get across though. If you read the rest, I think you will see it is more about the ‘useful’ limit more so than the total limit

[frozen moon]

Something I found that this works well for is for large coding project prompts which can then be fed into the API more so than ChatGPT itself

[frozen moon]

https://chat.openai.com/share/2ce4ad9f-50b0-4de8-9b6c-1b9294a2fe39
A sample thread from earlier where someone was helped with this GPT

[frozen moon]

I have been using this for complex coding projects - it has good success in generating outlines

[wispy hill]

@frozen moon this is awesome. But do you know where I can find a prompt engineer for API usage, basically defining the system and the user that understands also on how to, what variable data could be used in the prompt.

[frozen moon]

What do you mean?

[wispy hill]

Like a prompt that is creating prompts for OpenAI. Writing, let's say, a good system input and with a good user input. Understanding the context: let's say I would ask it to craft a sales letter prompt:
How would the system be written? How could the user message be written? What variables could be needed? Example: {Product Name}, {Company Name}, etc.
Like all the prompt designers are great for CHATGPT, but I was not able to find one that I could use for automations.

[frozen moon]

I’m still very confused. This prompt does create prompts..

[wispy hill]

Yes, I know. 🙂 I was just asking if you know if there is another prompt creator, but not for ChatGPT, specifically for automations with variable data that could be used on Zapier, for example.

[short yew]

they want a langchain/dataframe assembly assistant, that understands how to format chat completion templates with appropriate variables for procedural replacement...

[short yew]

what does a zap even look like if you build it outside the webui??? ive never seen one as raw code xD

[wispy hill]

i have played now abit and so far this is the best i could get:

[frozen moon]

Can you give an example?

[short yew]

not of one that makes zaps, i cant even find examples of complete zap dataframes, just examples of parts - i was too lazy keep searching. but they want an assistant where they can say something like "i need a 2 part zap template, with webhooks for running from raw JSON, that checks emails" and then the AI spits out a dataframe with the variables for chat completion compatible with openai schema (thats how far i got in my research) not sure if i can post the link, but search the official *zapier community web page for [advanced approach] and raw json webhooks, and they need it to format the 2 block format with just a prompt.

they need get and post, i believe

[frozen moon]

interesting idea

[short yew]

😉 np, and nice seeing you again

[frozen moon]

Nice to see you again too! if I may ask, what do you mainly use gpt-4 for ?

[short yew]

oh... you know... stuff

[frozen moon]

woah this is really cool stuff

in the second picture, you are asking it for a knowledge file from its internals , like something that it’s been trained on

I’m assuming here

this is all related to creating an actual LLM it seems

[short yew]

its a knowledge file in the mai-ku format, with a local flask server upgradable version, if you just say to dumop it, it translates into human-readable, but the files are written the way lyra output it, as long as it dumps it into a codeblock "as mai-ku"

aka macromancy aka MetaMorphic {M}

those are it's "prompts" xD

[frozen moon]

I wonder if GPT can assist in creating a neural network with accuracy @short yew

[frozen moon]

It, in fact, can help but its still very difficult

[severe glade]

So far it has a hard time matching prompts to an existing template, but it seems to be very good at auditing functionality of prompts already in use. Asking it (or another smart model) to do a search for cutting-edge prompt engineering for Custom GPT models and then apply that seems to help fix that a bit

[frozen moon]

Is this what you tell it? Can you share a thread that demonstrates this claim

I am curious to see!

[severe glade]

Figured out there’s an iOS app that just lets you stick screenshots together (which I went my whole life not knowing about) so here’s an example - hopefully it can be zoomed enough to read. Anyone who says prompt engineering is dead just lacks imagination 😜

This is referring to a nifty system I figured out that apparently enables some pretty dramatic improvements in performance across the board (assuming it isn’t frying a server every time I use it), I’ll share a model built on this approach here in a bit once I apply your model’s advice to shore up some auditing features in its Instructions

[amber granite]

What is the App called?

[severe glade]

PicSew, creatively enough

[amber granite]

Nice thanks

[frozen moon]

You could also share the thread through the share thread button but this works!

[severe glade]

On my phone I can't share threads once I used an image input, which I had earlier in the convo - didn't think to make a new one for demonstration, felt disingenuous somehow 🤔

[frozen moon]

I think I haven’t configured it too well to handle images

I will look into this

[frozen moon]

I need to figure out a way to make it so that it doesn't spend an entire message on just explaining roles

📝 Prompt Engineer / Builder 📝 - Create the best prompt to yield the most comprehensive results!

[elder hound]

I used it to generate a prompt but when I used it, chat gpt gave me an answer per role instead of combining them. It does a great job identifying the roles you need though

[frozen moon]

Yea, I need to fine tune it