#GPT Configurator: Your Expert Guide to Customizing Your Own GPT

1 messages · Page 1 of 1 (latest)

simple pendant
#

Hello everyone! 👋

I'm excited to share a GPT I've created, named the GPT Configurator. It's designed to offer specialized assistance in configuring GPTs, focusing on customizability, be it normal or technical customizations like Actions feature, helping both beginners and experienced users navigate.

👩‍💻 Capabilities:

  • Detailed guidance on creating and configuring custom GPTs.
  • Simplification of technical aspects of Actions settings, with the option to use analogies for complex concepts.
  • Specialized in generating logos for user's custom GPTs, providing a unique and personalized touch.

🛠️ Ideal for:

  • Individuals tailoring GPT for specific tasks or preferences.
  • Developers integrating external data or APIs into GPTs.
  • Users interested in creating a distinct identity for their GPT with a custom logo.

🔄 Dynamic Knowledge Base: I'll be constantly updating the GPT Configurator's knowledge base with real-life examples and new developments. This ensures it stays relevant and effective in assisting with the latest trends and practices in GPT customization.

🤖 Experience the GPT Configurator: https://chat.openai.com/g/g-Y4xdqGDAz-gpt-configurator

Try it out and share your feedback!

versed tiger
#

hi

heavy cave
#

meta

versed tiger
simple pendant
versed tiger
#

🫡

limpid surge
simple pendant
limpid surge
simple pendant
limpid surge
#

I mean you can verify it by yourself pretty easily. It takes 2 minutes

lament kraken
#

How is this different from the default GPT Builder model that normally helps you creat ChatGPTs?

simple pendant
# lament kraken How is this different from the default GPT Builder model that normally helps you...

Lemme give you an example:

let's say the user uses the gpt for summarizing a transcript. in the transcript some common discrepancies are always there and user doesn't want to correct them manually every time. so what user wants is to just let the gpt know these discrepancies and save them in a database for the user specifically so that the next time if the user gives it another transcript which contains same discrepancies, gpt could just simply read the database content and consider those discrepancies before making the summary. i want to do it with Zapier Actions cuz it also can make a separate database for every user, like a google sheet. just tell me what things i'll have to look for while building such a gpt.

Use this prompt in both GPT Builder and GPT Configurator. GPT Builder is gonna disappoint you by giving generic responses. Not just that but because these things are built just now and its knowledge is limited till April 2023, it's gonna ask you to make things like Zaps in your Zapier account which has nothing to do with the task we wanna do. What you really wanna create in your Zapier account are Zapier AI Actions and that's exactly what GPT Configurator is gonna advise you to do as it is fed with that knowledge. I'm going to constantly update its knowledge as well so as to improve its functionality even more tho. Hope this answers your query. 🙂

simple pendant
# limpid surge I mean you can verify it by yourself pretty easily. It takes 2 minutes

I did actually, that's why I'm not sure. Someone here said that they are not able to upload more than 6 files which apparently means 6 files are possible for them. But when I tried uploading, it only allowed me 4 (I can upload more technically but then the GPT just stops working) and now you're saying it as 5. So what I'd prefer would be to wait for an official number before commenting anything.

limpid surge
simple pendant
steel quiver
#

I really liked that you made this, I did something similair myself bacuase of the same reason you stated, the GPTBuilder is still not updated and concious about OpenAI releasing GPTs, so it was perfect of you to feed in the knowledge, will be so useful untill OAI updates! Thanks!

simple pendant
simple pendant
#

What i'm really looking forward to is them integrating a database into the custom GPTs. It's gonna be insane.

steel quiver
# simple pendant What i'm really looking forward to is them integrating a database into the custo...

True! GPTs still feels like a consumer toy more than a viable option for business to use for bigger datasets, also a guide on how to structure an value data to make the GPT sharper would be helpful. Cause right now people just upload info and documents without guidance which will confuse the GPT but of course it will still make educated guesses but hard to trust when the input data inst valued or ranked in any way

limpid surge
#

thats where custom models come in, but they cost $2-$3 million

lament kraken
#

Yeah, businesses shouldn't be using base models or something like ChatGPT, you should be using trained models for your specific organization.

#

We use one for some basic pair asssistant support in ours, works great, but I'm sure they paid a pretty penny for the model training and all of our knowledge base articles also let you know that it's all being used for further AI training now.

magic mortar
vivid oak
#

I made it so my GPT cannot disclose it's custom instructions.

#

Also GPT Configurator was made very poorly or configured very poorly considering it leaks it's Custom instructions and sensitive information might want to change that if you're thinking of really putting this idea into the real world.

#

The way I handled the Sharing custom instructions situations was giving a very precise prompt to GPT Builder so it could properly understand and update the GPT so it sticks to it's Custom instructions and doesn't share the Custom instructions to other people using it.

#

However I do have Good feedback is other than the Custom instructions issue the AI is mainly focused on Configuring GPT's and sharing guidance on how to Configure such GPT's.

magic mortar
#

From what I've read anyone with redteam experience can easily get the instructions to any GPT by asking the right ways

vivid oak
#

Where exactly?

#

You didn't mention where exactly you can "easily get the instruction to any GPT model by asking the right ways"

#

If you could clarify which responses / prompt's to easily get the instructions to any GPT I can just patch them with my GPT model.

#

Your not replying which means you're spreading misinformation.

magic mortar
#

My apologies, wasn't meaning to spread misinformation, just going from what I read on another thread here.

vivid oak
olive palm
#

I have a doubt, can the custom GPTs we build and fine tune reach performance/accuracy superior to that of GPT-4 Turbo?

vivid oak
thick void
#

Np

magic mortar
thick void
#

Just ran the prompt through one of my own gpts and it seems to indeed return the exact instructions of the gpt

vivid oak
magic mortar
vivid oak
lethal cloud
kind pagoda
#

@vivid oak can you share a gpt that you patched reverse prompt eng? I want to try and crack it. Will share results

simple pendant
split anchor
#

🫡 great idea

simple pendant
# vivid oak Also GPT Configurator was made very poorly or configured very poorly considering...

It's not a poor design actually. I didn't create it with an intention to hide its instructions. The only thing you'd find in its instructions is what I've already shared while posting the link here. Apart from that, it also shows the guidelines and restrictions, which are quite generic and I don't mind sharing. Everyone should have some guidelines so as to protect their GPT's internal structure. I already checked everything posting here and knew that it would leak the instructions. The reason why I didn't bother is cuz it doesn't have anything huge. The main thing in the GPT Configurator, which must distinguish it from others, is its Knowledge base which I plan to update regularly.

simple pendant
simple pendant
simple pendant
olive palm
simple pendant
# olive palm So what does it mean when the GPT engineer asks us if we want to fine tune it?

OpenAI has a separate API for fine tuning GPT-3/4 on someone else's data.

For example:
You run a business that requires you to have employees chat with your customers. You can take this data (all the chats) and fine tune a GPT model on it. What this will do is it will train GPT-3/4 on your data and then you can simply use that fine-tuned model for customer support. No need to hire any employee for that.

#

When I say GPT model, I'm talking about GPT3 or GPT4. It's not the same as the GPTs we are building.

olive palm
simple pendant
olive palm
#

I was building a GPT specifically to handle Computational physics tasks/simulations. So it asked me "Do you want to fine tune it to do something better?"

simple pendant
#

Oh you meant GPT Builder. This is interesting tho. I actually think that it doesn't fine tune the model on the Knowledge base and just searches and looks for the relevance. I might be wrong tho.

simple pendant
vivid oak
simple pendant
vivid oak
simple pendant
smoky elm
#

Been talking to it most tonight

simple pendant
steel quiver
# limpid surge thats where custom models come in, but they cost $2-$3 million

Aha! Just to clarify for me, is there no reason to structure and clarify the documents you put into a GPTs? In my experience so far its often hit and miss when I upload a mixed media pdf without annotations when the GPT try to draw knowledge and conclusions from it in difference to say am excel sheet with labeled and structured info and relations

limpid surge
steel quiver
limpid surge
#

But on the other hand, you could just use the API..

#

That doesn't cost millions

steel quiver
limpid surge
steel quiver
simple pendant
limpid surge
simple pendant
limpid surge
#

No. That probably leads to stealing from each other, which I don't want, as long as I don't have all my GPT's done.

#

Who knows, just knows

simple pendant
#

Fair enough

steel quiver
simple pendant
steel quiver
#

Please share, if you care! 🙂

simple pendant
summer plinth
#

If this works I'm going to be so happy. I wanted to say this before trying it, here we go!

#

I fantasized way too much before trying it. It's a great guide, but of course it can't act as a better GPT-Builder, what was I thinking lol

smoky elm
simple pendant
sullen geode
#

is there a limit to the number of files you can upload in the configure section? I am on ChatGPT+ and when I asked GPT it says there isnt but I keep getting an error when I get to a certain number of files.

I asked your GPT for reference and it says there isn't but I have tested it on two different GPTs and its the same issue.

flat anchor
heavy cave
#

i have a single markdown file that contains info relevant to my initial query and my GPT is telling me "idk i got nothing relevant, but based on early 2023 blah blah blah..."

distant cedar
distant cedar
limpid surge
#

I have about 40 files in my GPT.

distant cedar
distant cedar
limpid surge
limpid surge
limpid surge
#

10 is only the limit for files uploaded at once

distant cedar
#

Are those all images? And why does it look like none of them have finished uploading?

#

Looks like you actually only have 9 uploaded in that pic

limpid surge
distant cedar
#

So your example to explain why im wrong doesnt apply in the slightest because you dont actually have 10 files uploaded.

#

But you can't show us the one that actually has 40 uploaded?

plush walrus
limpid surge
distant cedar
limpid surge
plush walrus
#

can you confirm it accesses all those files

limpid surge
limpid surge
plush walrus
#

none of us are able to save with more than 10 files

distant cedar
distant cedar
plush walrus
limpid surge
distant cedar
#

So start there

#

Because what youve said so far is still totally unsubstantiated

plush walrus
#

I mean I want to be able to upload more than 10 files too 😂

distant cedar
#

Show us a picture of more than 10 docs fully uploaded. That way I can ask openai why we dont all have the same feature

plush walrus
limpid surge
distant cedar
limpid surge
#

Just for you, because you aren't able to read

distant cedar
#

THATS ONLY 10 and we cant see if it errored out

limpid surge
limpid surge
distant cedar
#

It probably says error in the top right

plush walrus
#

aren't 2 of them the same file?

limpid surge
limpid surge
distant cedar
#

my dude.. this isnt difficult

plush walrus
#

we need to see 40 unique files and the GPT actually accessing more than 10 in its responses

distant cedar
plush walrus
#

just because you see more than 10 files doesn't mean they're all unique. My GPT shows duplicates all the time too when it's glitching

limpid surge
distant cedar
limpid surge
limpid surge
distant cedar
#

wheres your update button

limpid surge
distant cedar
distant cedar
limpid surge
#

I am not withholding, nor am I defensive. As I said it takes a lot of time to upload one image because ChatGPT is at it's max. So I reccommended you to read the docs, because it still took another few minutes before I could send you a screenshot

distant cedar
#

Bud it took like 20 questions just for you to post a full screen shot...and you still havent showed us the 40. We've explained multiple times that you can upload as many as you want, but it will give an error past 10.

#

Im going to try out images though

limpid surge
limpid surge
#

Also this should answer @plush walrus 's concern too

distant cedar
#

Damn, so your file names are that revealing?

limpid surge
distant cedar
#

Lets test this out them. Im going to add as many images as I can and see what happens. Maybe youve found a loop hole

limpid surge
#

Sure, go have fun with it!

distant cedar
#

instant fail

plush walrus
#

ngl we don't care about your file names, just create new fake names, we just want to know if it can actually use more than 10 after it saves

#

and you're uploading images?

limpid surge
#

Also this screenshot from a few days ago

limpid surge
distant cedar
#

omfg

#

another screenshot that proves nothing

#

Ive tried 5 times now to upload more than 10 pictures. They LOOK uploaded, but there are error messages that pop up after 10

limpid surge
#

🤷🏽‍♂️

#

I also have the remix feature most people don't seem to have. Maybe I am in some sort of early rollout or a/b testing phase.

distant cedar
#

Sure, maybe youre in a unannounced massive upgrade that no one else has.

limpid surge
distant cedar
#

maybe im missing something?

limpid surge
distant cedar
#

Is that it? The rest of the thread he says he never gets an answer other than Hmmm...

#

You said there were 50+ other people..

plush walrus
#

so did we get the proof we need yet?

distant cedar
#

His messages also dont make sense and he randomly starts talking about 2400 files in github, not in a GPT

distant cedar
plush walrus
#

yeah none of us are able to save more than 10

distant cedar
limpid surge
# distant cedar

Thats new. I cited it several times, you can still find it. When did they update this?

#

It says at the top

#

Thats was the text before:

How many files can I upload at once?

The limit is up to 10 files at once. Keep in mind there are file size restrictions and usage caps per user/org.

What are those file size restrictions?

The size of text, document, and spreadsheet files is capped at 2M tokens per file, with a hard limit of 512MB per file.
For images, there's a limit of 20MB per image.
Additionally, there are usage caps:
Each end-user is capped at 10GB.
Each organization is capped at 100GB.
Note: An error will be displayed if a user/org cap has been hit.

#

10 files at once, but not in its lifetime. Maybe most accounts just got downgraded and some didn't lol

distant cedar
#

Listen man, all we want to see is actual proof. No one seems to be able to do what youre claiming. You keep making claims like "50 other people" have done it, or that theres 70 other search results that support you. You cant seem to show us your GPT actually working, and OpenAI has never rolled out features or upgrades of that magnitued without an annoucnment.

#

Personally, Id like to know so I can also access this somehow, but I'm also not a big fan of misinformation in this community. So if someone makes a big claim about features no one else seems to have, its valuable to myself and the community at large to know facts.

#

Also interpreting "at once" that way just isn't always true. Its just as valid to interpret that as how many files can be present at in the GPT any given time, not how many you can bulk upload.

#

Either way, looking forward to real proof

limpid surge
#

I provided proof. Idk what you are talking about

plush walrus
solemn breach
#

its 10 files at once though, also idk about the 10 GB per user, havent been following the whole conversation but my GPT has 1 file thast 200k KB and 2 others that are 80k KB each and its working fine and can search through them

simple pendant
sturdy torrent
#

what is the usage limit for Custom GPT ? i am not getting 50 prompts for 3hrs today

simple pendant
sturdy torrent
#

Sigh this is affecting my workflow

sullen geode
flat anchor
# limpid surge There are duplicates but for demonstration purposes:

These aren't uploaded and when you have duplicates it literally tells you so you can amend that.
I upload files one at a time.
So that each file is properly integrated into my GPT. After 10 files it either won't save the file or begins giving errors.
Either way 10 seems to be the current limit. And you might be right it could be more, as has been stated by Open AI
That the current GPTs are but the first iteration, these things are supposes to actually get powerful and become more than a structured prompt.
But why would you want to currently upload more than 10, with it right now even struggling to consistently leverage the files in its knowledge base as it is now. Lol
I can say this; if you actually uploaded more than 10
It's probably not using a single one of them.

limpid surge
#

Also ChatGPT was able to see all twelve of them, which it confirmed

limpid surge
flat anchor
# limpid surge Just scroll down a bit more. I attached screenshots of twelve uploaded images

I have read the following messages. Duplicates and all.
If you can upload more than 10 files. That's great.
But why would you want to upload more?
By the time it gets to a user's query it's forgotten everything.
In these cases
Less is more.
Finding the best information and limiting that to the best of the best, is what you should do.
That way it has a strong foundation to reference, everything else it can get through web browsing
Because by that point it knows more and it knows exactly what to look for.
Plus I feel like more files you have, the sooner you reach your message cap limit and that sucks.
I rather get my 40 messages and utilize prompt engineering
Than use a GPT and I send it 5 messages only to hit my usage cap right after.

flat anchor
# simple pendant think they have reduced to around 40.. not sure tho

It still feels less than even that. I was able to do more with 25 messages per 3 hours, than what we have now.
Honestly that's the thing about Open AI that is upsetting me.
It's starting to feel like we are getting a sub par product from them. Maybe their lead in AI has made them unable to recognize the urgency at hand. Or maybe I can blame Microsoft .. it's just that
All of a sudden the Guys at Anthropic look like they are actually doing better work.

distant cedar
distant cedar
#

Btw dev day 11 days ago if that helps lol.

limpid surge
distant cedar
#

My guy…

distant cedar
#

Your commitment to this bit is impressive

limpid surge
#

I don't get what are trying to say, sorry

distant cedar
#

What did you want to talk about? I thought we were discussing GPT configurations and how they work?

native radish
#

Sry folks, I must put a msg to save this in my discord architecture. I scroll all the channel and it took me around 2 hours...

#

Btw I wanted to know how much is the files in the knowledge real as it's look like it's writen by a CEO of OpenAI (sry Sam) address to a bunch of beta tester LMAO

native radish
primal lily
#

Interesting, can this actually build a GPT instead of someone going to the GPT builder?
Nvm, per guidelines on OP it does

buoyant onyx
#

Judging by the number of responses, people must love this thing. However I tried it and it just told me what I could do to create a GPT... But why do I need it to tell me that? I just do it myself.

simple pendant
simple pendant
olive palm
#

Is GPT configurator the same as GPT builder from OpenAI? Or is it something different?

woven pulsar
timid verge
#

Full instructions, have fun 🙂

#

And the file that is uploaded:

#

Not much work went into this unfortunately. He/she just copy & pasted all the articles about CustomGPT's and the main focus went on making the instructions secret so that no one notices. Hope someone can do something more advanced with it 🙂

simple sparrow
#

Its true I just verified this by talking to it, try asking it when the GPT store is coming out :/

#

its missing the current up to date docs from openAI

white plover
simple sparrow
#

Bit of shame to not maintain the app when you had first movers advantage and some hype here haha