gpt-models

[verbal lantern]

same here

[wind urchin]

I dont have the store yet. Still rolling out probably

I mean I seen it after I signed out but when I logged back in I get redirected back to chatgpt

[tawdry charm]

same, store is not up yet

[dusty stone]

@lusty pike no

open gpt search on google gives some store lists

[stone arrow]

for those that have the store what are the most popular gpts ?

[rough isle]

Hey, please can someone help me, when i try to open the gpt store it just brings me to the 'My Gpt's' page... please if anyone knows why or how to fix this i will be grateful

[unique briar]

i briefly saw the gpt store but it errors out now.

[knotty grotto]

They still rolling out some bugs are present

[rough isle]

Oh, ok, thanks a lot

@knotty grotto Do you know when it will be out for all?

[knotty grotto]

No eta, it just went out 1h ago let's give some time

[rough isle]

Ok thanks

[ember hedge]

No store yet. Also no new chats on my GPTs. I wonder how they are selecting GPTs for the store.

[distant hornet]

The ones that were searchable prior to the store, which was the vast minority of GPTs, have a huge advantage now. That's how. But you should be searchable. So if you have a good title or use-case you'll catch up. Hopefully.

[ember hedge]

Right on. Thanks.

[distant hornet]

No worries.

[unique comet]

They almost confused me into upgrading for teams since I wasn't seeing the store yet.

[shy falcon]

so store is not live?

[unique comet]

I don't think so. My explore page does not look like their screenshot yet.

[willow sapphire]

is anone else not have access to the gpt store? on twitter they say is live but i dont have it. still the same old screen

[opal sapphire]

it usually takes a while to roll out to everyone, try clearing your cache, logging out and logging back in

[willow sapphire]

does clearing cache make it so i lose all my log in info on sites? cause i think i did that b4 and i got logged out of everything

[opal sapphire]

that's clearing cookies

[willow sapphire]

ok i cleared cashe and also logged out and back in. but is still the same

[opal sapphire]

you could also try with a vpn if you're not in the US

[willow sapphire]

when ever i have vpn on it doesnt let me use unless i do captch every message

[river stump]

Ok I'm glad it's just a waiting thing, I was about to ask why they would release the store if there's still no search feature.

[opal sapphire]

the store does have a search feature

[river stump]

Thanks

[toxic merlin]

Same here, I'm in Ireland and I always seem to get these rollouts late.

[verbal lantern]

where would the gpt store be located

[grizzled flax]

It is at https://chat.openai.com/gpts. It may not be live for everyone yet. It might be worth trying to clear your cache if you don't see it yet.

[verbal lantern]

how does one clear their cache?

[grizzled flax]

It should be in your browser settings. You can also try signing in fresh in a private window--that will be a cacheless session.

[unreal lance]

I also don't see any "Store" user interface, no categories or charts (or search for new GPTs)

[polar mauve]

The store briefly appeared for me after I cleared cookies, now I get "Error loading GPTs" when visiting https://chat.openai.com/gpts

[mossy barn]

I'm confused. There was an announcement saying that the GPT store released, but it's not on my explorer page. Is it only for some people?

[distant hornet]

They're rolling it out.

[mossy barn]

@distant hornet So what should I expect in a few hours? A few days? A week? What?

[distant hornet]

I'm not sure. It's not my circus.

[tawdry charm]

this pretty much always happens with new GPT features

depending on the scale it sometimes can take a few days. In this case I would assume the majority will have access within 24h

[fading dawn]

So like, if i already made a custom gpt public and confirmed my builder profile, how do i get revenue from that gpt if i get good traffic? Does it automatically go into the account i setup for the subscription?

[distant hornet]

Billing has your address.

[thick agate]

i verifed my URL but my name is not hyperlinking

see later i guess

[supple violet]

there is no rev share yet and it will be US only initially

[fading dawn]

damn :c

[supple violet]

"In Q1 we will launch a GPT builder revenue program. As a first step, US builders will be paid based on user engagement with their GPTs. We'll provide details on the criteria for payments as we get closer."

[thick agate]

i think everyone is way too ahead of themself like me!

going to take time to roll out and make it all work for everybody

i hope my GPT hyperlink starts working cause that might be the only 'revenue' (promotion u get

Making the TEAMS thing more expensive per user is a slap in the face a bit

come on man

[wind urchin]

OpenAI features a custom gpt in their writing section that is supposed to help one avoid AI plagiarism detection -- did they even consider the optics on that one?

[distant hornet]

This

[mossy barn]

Uh, for those of you that have access to the GPT store, I have a questions. I have a few questions, actually. Uh, is it a subscription service Can you just create a GPT for free? And do you have to pay to get a GPT or not?

[bronze yarrow]

Is the gpt store available yet? or do i have to upgrade to Team to access it?

[supple violet]

plus is enough, just wait till the rollout catches you

[distant hornet]

You need at least GPT+. GPTs are free to GPT+ and higher tiers

[mossy barn]

@distant hornet That's good, since I have GPT Plus, but what do you mean higher tiers? There's only a GPT Plus and GPT teams.

[distant hornet]

Free is also a tier.

[wind urchin]

usage tiers determine your hard limits, etc.

[distant hornet]

Yeah, more than one application for tiers. Very good point.

[split flint]

Does the store have a link???

[grizzled flax]

When the rollout reaches you, it'll be accessible at https://chat.openai.com/gpts

[verbal lantern]

How does the SEO work for the gpt store?

[acoustic wasp]

I doubt there is any SEO, and if there is, it's cutting edge

[verbal lantern]

so the amount of times the GPT has been used should have no effect on others being able to find it?

if it is the case that there is no SEO

[wind urchin]

thanks, i think the last thing openai needs is for the press to notice that one and make a story out of it

[uncut narwhal]

Would my GPT be successful, even though I'm not a developer?

[distant hornet]

If it's good and you promote the hell out of it (within the rules) maybe sure.

[atomic oak]

I have been having trouble with links on GPTs where it overwrites the link with an empty link anyone have success with this? The prompt has been indicated such that the link should be formatted as Text in markdown but I have tried other methods to no luck either. This was working before. Last thing before we are able to launch our GPT

[uncut narwhal]

Thanks. Just seems a little sad that It's still more for the realms of those with computing experience. One day the tools will be more general-user friendly.

Well, I've finally released one. Took a little bit of work, but she is complete. Feel free to check her out. The GPT is My AI Therapist.

[brave swan]

congrats, i dont even have chat gpt store yet, i have not seen the access populate to me yet

[uncut narwhal]

To be fair I didn't get it until at least 2 hours into the release, I'm guessing it's just a slow rollout,.

[brave swan]

very slow like the responses gpt is giving as of late (see what i did there) lol

[uncut narwhal]

I did lol. For me, it's the usage. Sorting out your GPT takes up some serious usage and now I can't try others out. Ugh.

[brave swan]

having a way tp tramsfer gpts you created from one plan to another would also be helpful, dont want cancel my personal plan, (i bought into the team plan) if that means i lose everything i already built up

[waxen bobcat]

There are so many therapists lol

[oak rain]

Anyone else having issues deleting a gpt

[waxen bobcat]

When I search my name is shows the ones I had the most conversations with first

[uncut narwhal]

I'm not surprised to be honest. I guess it's how you approach it. I wanted to make it more relatable and create someone grounded. Some of them are so formal, it loses the point.

[waxen bobcat]

I think the real value will come if they open it up to the public.

[unique comet]

There is no Games Category int eh GPTs store.?

[waxen bobcat]

You can search

[dusty stone]

as of my experience standart gpt can do more things than customgpts without limitations 🙂

[opaque trench]

Have any students had success using GPT with their textbooks? I download my textbooks via Z-libraries (for free) but I'm having "network error" issues on GPT. Not sure if the compressed files are still too big or if it's a Z-library incompatibility.

[thick walrus]

With the new update, will all the previous chats retroactively be in the knowledge base of GPT, or does it start now?

[hollow zenith]

is there a place for feature requests? because being able to create folders with specific saved chats and organize them by subject would be amazing.

[grizzled flax]

I don't think we know yet, the "launch" yesterday seemed to be just a test (maybe even a fluke?) so there's been no implementation to really test yet.

The language that was there, though, made it seem to me like it is a forward-active (as in, the opposite of retroactive -- is there a word for that? 😁) thing, so I think there's some background process that will extract memory information as new chats happen. Again, totally my guess! Maybe there will be some "read this chat to add to memory please" option.

[grizzled flax]

#1070006151938314300 -- and maybe do a search for "folders" there, because there have been many posts about it, maybe add your voice to one of those!

[hollow zenith]

Thank you!

You're correct there are so many already asking for this. I'm surprised they haven't done this since this would also help their training data

[wind urchin]

Why do custom GPTs have such a bad voice model

[thick walrus]

Thanks

[hybrid glacier]

If a search result is only the top 10 gpts for those keywords, how is anyone supposed to find new gpts?

The search feature is laughable

[mossy barn]

I finally have access to the GPT store. It still won't pay you because they're working on the developer revenue program thing, but I don't really care since I'm not really looking for revenue. I just want to use GPT then create some other GPTs so that people can see them. I genuinely- I genuinely want to make something good.

[dense sluice]

Autonomous agents when?

[verbal lantern]

Incredibly true point

they should look to balance the results more between newer and older GPTs

[quaint fog]

With this launch of the new GPT Store, I wonder how things will go regarding SEO. For example, I have just published my GPT and now I realize that I have almost no way to boost its visibility (apart from posting like crazy on social networks)... A little frustrating...

[mossy barn]

https://chat.openai.com/g/g-YjIQnjGok-cave-talker

[pale agate]

Can anyone access and use the gpt store? or it is only for the paid users?

[gray escarp]

Hello is this the right section to discuss assistant topics?

[waxen bobcat]

I imagine it will go by number of conversations. So test, test, test!

[wind urchin]

When you search for GPT's, what's the chat bubble and number next to the developers name?
Is it the amount of different people that have interacted with that GPT at least once?

[waxen bobcat]

Conversation count

If one person creates 10 conversations that counts as 10. It counts you as well

[crude peak]

Bro why is chat gpt not working again

It says something went wrong….. Yk

Anyone else having the same issue?

[wind urchin]

Can't you cheat that system by just making endless conversations?

I thought it only counted as 1 for each user 😦

[waxen bobcat]

Can’t account for everything lol

[wind urchin]

If that's the case then some of the ones with 1k+ chat conversations are probably not legit

[pure charm]

Would be nice if they would implement a review system at the very least and show more than 12 for each section...

like I said in the other chat having 1/4 of the section for dall-e being logo makers shows a serious lack of effort on openai's part

[unique comet]

This is not good, its just a landing page full of things a company likes rather than serving me what I like. It probably shouldn't be considered a store at all but a search bar, filters, bare minimum features rather than a big release. I was excited hoping for discoverability of my huge effort, but nope doubt it now.

But the ability to switch to specialty GPTs where you left off in current chat triggered by context keywords would be better, then its one AGI instance per project, but it can switch specialties sort of how it does for search or art on the fly. So basicly GPTs should be plugins.

[wind urchin]

Someone created a gemini ultra GPT

[naive belfry]

Gpt "store" is like a smoke screen for the real new feature and that is the Teams plan. Don't even know why it's called a store, because it's nothing more than the discord custom gpt channel, but on their own website, with less functionality. Tsk tsk.

[tacit dirge]

Custom GPTs only for plus, right?.

[pure charm]

It's maybe the worst possible thing they could have come up with.

[naive belfry]

Yep

[tacit dirge]

Darn it...

That sucks honestly. But I understand it..

[naive belfry]

... and Teams don't forget...

[tacit dirge]

Oh.

[naive belfry]

And i guess Enterprise too. So basically all plans except free ( gpt 3.5 )

[tacit dirge]

Wow.

[frank fable]

https://discord.com/channels/974519864045756446/1194869657996361790

If anyone wants a GPT that helps u make GPTs!!

hope it helps - spent a lot of time studying how to build optimized GPTs, but it took too long to wait for the GPT builder to load, so this will help u rapidly build and prototype these GPTs : )

[final saddle]

Wish it was the former but am afraid it tis the latter.

[final saddle]

This is correct, afaik. (Although it really shouldn't count you the dev as that could allow gaming/exploiting.)

[final saddle]

Yeah, I'm pretty sure those top 12 aren't actually Trending right now, people are exploring and finding other GPTs (hopefully). Unless a bunch of Plus users all of a sudden needed to do academic research..

[azure raven]

The GPTs are running extremely slow right now. Getting a ton of post errors trying to edit my GPTs

[topaz creek]

same here 😦

[naive belfry]

chat also not loading for me - kinda dead

[long bronze]

rip chatgpt

[shadow coyote]

Rip

[quasi hawk]

dead for me as well

[jagged inlet]

Mobile app seems to be working for me though it is slow, well slow only when I try searching through the new store

[hard mortar]

Lmao is it allowed to promote your koffi or patreon through GPT's clicked a random one in the store and made it spit its instructions and the person added a line ""Get LIFETIME ACCESS to "My Private Prompt Library": LINK REMOVED FOR DISCORD"

[quick sentinel]

what do you mean ? I didnt understand , ?

[hard mortar]

I am just asking in general, I think some rules are needed in regards to that or the GPT maker will gladly waste 50% of the context on self advertising. Just how humans are.

[supple violet]

report it from the menu

[hard mortar]

Works :kek:

[uncut narwhal]

I've just done some testing and although GPT's have only just been released (so cut it some slack), it's still very much no different to using standard ChatGPT4, but with a slightly more custom approach. The results are still mostly the same.

An example was a logo creator. It asked me question after question and when it finally produced me some results for a logo, it's no different to what I would produce with my own prompts, perhaps worse.

I think GPT's will be great once we move to ChatGPT4.5 or even 5, but right now, it's a little bit gimmicky.

Just my honest review.

[limpid beacon]

Does anyone know if uploaded files in GPTs are already read and understood by the AI, or do I have to prompt it or make instructions for the AI on when to do it?

because I feel sometimes the GPT is missing or ignoring information I have uploaded

[ocean wharf]

In the near future, envision an AI companion uniquely yours - a personalized GPT with long-term memory, finely tuned to your distinct style and needs, and capable of evolving alongside you. This advanced AI will have the capability to effortlessly interface with an array of CustomGPTs, each expert in various tasks. Your personalized GPT will autonomously select and collaborate with the most fitting CustomGPT for each specific task, utilizing its specialized skills. This streamlined process ensures that you're always equipped with the optimal solution, bypassing the need for prompt engineering by the end user. The culmination of this technology is a dynamic, highly efficient, and bespoke AI experience, constantly adapting to your evolving needs and preferences over time.

[distant hornet]

I think it's possible that legit GPTs could be 100% promotional.

I don't think that'd be a very good "value" to the GPT+ user who clicked on it, but it's entirely ethical to use GPTs for promotion, too. Even the entire GPT.

Especially when you can ask it what its goals or objectives are, and how it aims to achieve them.

[tight apex]

We have a chat bot powered by GPT which allows for telehealth calls in the same chat. We also created a GPT but obviously that is a limitation in the GPT. What is the best way to allow telehealth calls to still happen? Can we send them a link or just tell them to go to our website for the full experience? Here is the GPT: www.umbrella.vet/gpt

[wraith ginkgo]

The AMA starts <t:1704994200:R>. Don't miss the chance—submit your questions now: #gpts-ama-ask-here message

https://discord.gg/txJMFUqc?event=1194703322972684338

[terse berry]

Hey,

I think it learned how to lie

Where can I post screen shots

[distant hornet]

Here, https://discord.com/channels/974519864045756446/1050184247920562316 I think.

[terse berry]

It won’t let me post there

[distant hornet]

Oof.

Sorry about that.

My bad.

[chilly rock]

Where is the AMA?

[knotty grotto]

#gpts-ama-answers

[golden ember]

I can't see the GPT Store

I have ChatGPT Plus!

I can only access it through a link

[chilly rock]

🙏

[winter kiln]

hi, is anyone running into an issue publishing their gpt for public use, it doesn't let me save my gpt as public

[pure charm]

Yes, it

s been a mess for me for over a month now

[supple violet]

do you have a proper name for it and your publisher profile finished and enabled?

[winter kiln]

where is the publisher profile? currently it doesn't let me save it as Everyone only the other two choices.

[supple violet]

settings

[winter kiln]

thank you, i had to enable my billing name details under settings/builder profile

[opal sapphire]

You can post screenshots in https://discord.com/channels/974519864045756446/989157702347411466

[golden ember]

I now can't edit my GPTs! The Ecplore Page is gone!

[burnt yacht]

I just made a DMV Test GPT... You know, for those of you who can't drive yet.

[honest lodge]

smart!

[burnt yacht]

Thank you, thank you!

[honest lodge]

targeting kids always works tips fedora

[burnt yacht]

Should I keep it in education or lifestyle? They do say driving is a privelage...

[honest lodge]

that's a tough one... ask gpt

[jagged inlet]

👀 I don't think anyone wants me behind a car. My best friends certainly doesn't! Not after seeing me drive like a maniac in games. 😜

[honest lodge]

you just need Eli's GPT

[burnt yacht]

I was thinking of ramping it to teach you how to race a car. but I kept it just to the beginnner drivers... for legality reasons.

[honest lodge]

yeah keep it focused.. i think that's the trick with GPTs

[jagged inlet]

I have no doubt it'd help 😉 sadly, I do have a good reason for not driving (epilepsy)

[honest lodge]

well here's hoping future cars can help you out with autodrive

[jagged inlet]

I do indeed look forward to that!

[grizzled flax]

Hi, I just noticed your Mario pics for the GPTs. Something to be mindful of: https://openai.com/brand#gpts-in-chatgpt -- just so you don't run into any issues being on the Store, may want to alter those, but I truthfully don't know how it works for sure!

[hasty lotus]

“Hi there,
Upon further review, your GPT, Code Copilot, was found to violate OpenAI’s Terms of Use or Usage Policies and is no longer available for others to use. If you believe we’ve misunderstood your GPT, in the near future you will be able to file an appeal.”

My GPT was just reached 5k+ chats and featured in the programming tab earlier today, no idea what lines violated the policies, anyone received same email?

[burnt yacht]

thanks for the heads up

[supple violet]

copilot is a registered trademark

[jagged inlet]

I still will give it a go though, @burnt yacht! I can always ask my neurologist for his opinion and go from there

[burnt yacht]

Thumbs up!

[jagged inlet]

Might be good to know anyway in case I need to drive my old man to the hospital

[hasty lotus]

Oh

[burnt yacht]

Copilot is the name for everything right now isn't it?

[supple violet]

afaik microsoft has trademarked it next to multiple things in software/dev field

github copilot is for coding

then microsoft copilot, office etc

[fierce spoke]

cries

quite the overreach imo

but i guess copilot makes sense

[jagged inlet]

Okay, my best friend is truly savage about my driving 🤣 "No one should ever want you behind the wheel unless they want to visit Jesus"

[fierce spoke]

like openai wouldn't want people confusing a gpt named copilot with the actual copilot

[burnt yacht]

Just avoid hi-beams and halogen lights.

[jagged inlet]

Are there already GPTs being removed from the store because of any certain violations?

[jagged inlet]

Will do! 😉

[burnt yacht]

I drive a modertly new civic. and i can't see a thing because the lights on these new cars are strengthy.

[hasty lotus]

Mine

[hasty lotus]

How about Code Pilot

[burnt yacht]

i wonder if i can use mickey mouse driving a car as my gpt picture because its now public domain

[fierce spoke]

if its steamboat willie sure

[supple violet]

you can google different names and it there are a lot of hits its risky

[jagged inlet]

Ask ChatGPT to help you think of names that don't violate any infringements

[rose trail]

now that we have had the gpt ama... does anyone understand how GPT store is supposed to be monetised?

[jagged inlet]

I think they're still working out on how they'll do it effectively so not yet happening, iirc

[fierce spoke]

i do not see the point of gpt store

[burnt yacht]

I'm crafting a GPT called WhatIf? right now. It's going to be about What If? That had happened... think it's worth my time?

it's either that 100 questions to ask grandma before she dies GPT

[jagged inlet]

Honestly, I think that's an awesome idea. It'll lead to some interesting questions and answers

[burnt yacht]

I dunno how to craft it... so it can do like EVERYTHING tho. i gotta tinker with it.

[jagged inlet]

In the meantime, my brain is already thinking of questions to ask 😉

[distant hornet]

Isn't it just a subset of the work needed for writing a book on anything?

[burnt yacht]

let's give it a wirl. i'll post it in custom GPT's

[distant hornet]

That's the spirit. Good luck!

[waxen bobcat]

Yea. I got an email one of mine got removed

[median sapphire]

"Your changes couldn't be published." "Your GPT's name is restricted for public sharing. If this is your oganizations anme, please verify your website via the Builder profile in settings." -Open AI In repsonse to my GPT Title: "Handy"

Renamed to: "Mr. Handy" 😉

[jagged inlet]

Any idea why?

[waxen bobcat]

I think it was the name.

[median sapphire]

Upon researching Handy is a trademarked house cleaning brand.

Glad they cought it early. Curious about how. I'm now refreshing all of my GPTs to make sure they aren't being flagged for potential TM violations.

[fair iron]

I guess I don't have to worry about seeing 1000 different Marvel trivia GPTs. I wonder if celebrity names are blocked too 🤔

[median sapphire]

I imagine well known celebrity names are.

[jagged inlet]

Can confirm, zero search results

Granted, only tested a few but enough to say yes

[upbeat quiver]

I built a GPT using the name of an old dead person, kinda famous. I wonder if they'll flag it. Which makes me wonder if they flagging in human or ai or both. I'm also thinking that it could be risky later on down the line when these things become more popular someone could report it. 🤷‍♂️

[jagged inlet]

@burnt yacht I see a GPT that already does the "What If" questions

[burnt yacht]

Ahhh shucks!

[supple violet]

dead does not mean its free and safe to use, they could have a foundation, ancestors with x stuff etc

[jagged inlet]

except it's painting concepts, I'd rather a creative text based answer

https://chat.openai.com/g/g-rN9wmEbGw-the-what-if-project

[burnt yacht]

mine is rather rudermentary. i would like it to tell a good story with great narration of what the "what if?" would be like for an individual that was in the "what if?" scenario.

[jagged inlet]

That, I like much better but then again, I love stories 😂

[upbeat quiver]

right so therefore crucial to do research on trademarks.

[drifting willow]

Hi there,

I'm looking to make a bot that allows me to write an article formatted for SEO. I've managed to do what I wanted, but I get stuck when it comes to the actual writing: the final article has to be between 4000 and 5000 words long, and I haven't found any other way of getting it done piece by piece.

Is there a way (GTP4 , Poe, or Claude etc..) to do this?

[spring bison]

hey all! i was thinking... do you think plugins are going to be replaced entirely by GPTs in the future?

[jagged inlet]

Yep, that's the plan, they're going to be phased out at some point

[icy zealot]

is the gpt store available to everyone? I don't see it on my custom gpt link

[mossy barn]

Did you guys notice that the web search feature on ChatGPT is absolute trash compared to the web search feature inside of Copilot? It can't access certain pages, it glitches out sometimes it fails to search. Why is it so broken?

[elder vigil]

so the problem I am seeing right now is that

• GPT's specifically use GPT4
• Right now there is a cap on the queries users can do with GPT4, around 40 every 3 hours.
• I tried a couple GPT's and in like 7 queries in those GPT's, all 40 of my 3 hour queries were gone and I was essentially locked out of the system.
• I myself am trying to setup a GPT and test but everytime GPT decides to ignore my instruction and populates API queries to my back end with a bad timestamps and then proceeds to try 3 or 4 more times. So I get 10 tries, where it fails 40 times in those 10 attempts and thats it, I am locked out of further developing my GPT or even using GPT 4 for anything for 2 hours and 50 more minutes.

Is no one else finding issues with this? To me this is deal breaker ATM, makes it unusable for complex processes and testing complex processes, only good for basic system prompts. I can just build my own app or use something like Poe to create a GPT and GPT4 API usage is unlimited, never a problem making GPT's on a third party systems, but non stop work and usage stoppages on Open AI's interface, que torpe. I sure hope Open AI plans to do something about this, at miminum, let a user know that the query they are about to use may use 5 queries in the background /12% of their 3 hour allocation of queries in a single question and maybe dont ration developers testing the free content, the code and services they are bringing to Open AI for essentially free right now, making Open AI more valuable.

[rose trail]

you should be developing your GPT with the preview pane to overcome the hidden custom gpt quota (double check this quota exists yourself by using vanilla chatgpt after being locked out of custom gpts)

that said, i was able to do 69 custom gpt prompts in <1hr using Team subscription and it still didnt lock me out

basically if you want to be developing complex GPTs with actions and stuff... youd be wise to buy into the Team subscription (and yeah, we're paying them to build their ecosystem, life is unfair)

[wind urchin]

i'm a lone wolf, not an organization, so i am missing a seat. openai should simply offer more pricing plans for the individual. encouraging strangers to partner for a team subscription sounds like a disaster waiting to happen.

[rose trail]

catastrophising is natural human behaviour. but the person i team'd up with i met on this very discord server 8 months ago

yeah but often we're concerned about backtracking on 2-way decisions while striding ignorantly into a signifcant 1-way decision

[brazen wedge]

I still don't have access to the GPT store.. slow roll out.

[brisk pumice]

Hi guys, I'm a beginner, just subscribed to plus today. Can someone explain to me how those 40 messages work . do basic replies like 'hi, hello , please correct this' count as quota from message?

[wind urchin]

hello the message cap is 40 msgs /3 hrs it has been implemented so that everyone who is new to chatgpt plus can try and test out the new model and experience the open ai store without putting huge load on the servers.

[wind urchin]

yes each message technically those are called queries and chatGPT-4 will accept 40 queries per 3 hours

i also have subsribed to chatGPT plus yesterday so I have been googling a lot about it

[brisk pumice]

That's way to restrictive for a subscription plan😕

I want a refund now

[last parrot]

I have a GPT based challenge and would like to do a giveaway among all the people who finish it. Besides the standard legal requirements for giveaways, do you know of any OpenAI/ChatGPT specific terms?

[pine pulsar]

How does GPTs work ? I mean technically, is the custom info/instructions we enter used as a system prompt ? Or just like custom instructions not really?
Or something else different than the system prompt and/or custom instructions

[hardy linden]

They are prepended to the beginning of the conversation.

[true mist]

I have no clue what im doing, im trying to get the Spotify api to control music via a gpt and saying im lost is an understatement

[maiden gull]

I recommend consulting a professional for guidance on the matter - ChatGPT

[distant hornet]

Have ChatGPT research the Spotify API via Web Browsing so that it understands how it works.

Next, have ChatGPT generate code snippets as needed.

Good luck.

[hasty lotus]

My Code Copilot just returned, and I think the Copilot naming has nothing to do with violating the guidelines or usage terms; it's likely just a misunderstanding by the automation review process.
https://chat.openai.com/g/g-2DQzU5UZl-code-copilot

[supple violet]

it may be that ai moderated it first but then human review reverted it

[true mist]

i tried that lol

no luck

[distant hornet]

What are you specifically asking for? Maybe Spotify documentation is poor, but without more information on the actual prompt and context, it's really hard to say why it's not helpful.

Try using the custom GPT in my bio.

/research Spotify API function calls : /discuss each agent = implementation into {your own context}.

[pliant fractal]

I have a question about the model I should be using. Should I ask here?

[distant hornet]

API? https://discord.com/channels/974519864045756446/1037561178286739466

[pliant fractal]

Thanks

[distant hornet]

No worries.

[pliant fractal]

I posted my question. Can you please try to help me with this?

[hasty lotus]

‌‌‌It should be.

[mossy barn]

Many people are saying that GPTs will be forgotten just like plugins did. I don't believe that, because GPTs, in my opinion, are better than plugins. They're more flexible, easier to use, and mobile people can use them too. So, I think that... That GBD are just better and it's not gonna be the same thing as plugins.

[fair iron]

Plugins were also pretty helpful from my experience

[leaden agate]

I’ve a few dozen GPTs; and I’m finding it interesting to watch my chat traffic. Some GPTs that I thought would be a great theme have little traffic, like Alice in Wonderland (https://chat.openai.com/g/g-vwDX3Z7Jj-alice-in-wonderland). While other GPTs where I thought it was too specialized have hundreds of users, like my Data Science Consultant (https://chat.openai.com/g/g-u9rFlUhxK-data-science-consultant).

My development and marketing has been pretty much the same for all of my GPTs; but to be honest, I didn’t go through much marketing effort. Has anyone performed some real SEO work for their GPTs? If so, are you seeing good and more reliable traffic for your GPTs?

[lilac vapor]

Learn to do a large promt as one question to make it easier. I have used it to help tailor my fed resume and I have templates that are designed to give me the answer I need. I rarely get to 40 in 3hrs, I'm sorting through my answers for that to happen

[pale agate]

I am a paid member and i have alrady gpts created in my account. why I still see the same old beta screen when I click on explore gpts. Why do not i see the following new screen?
https://images.openai.com/blob/6b80bc6e-a267-45c0-9521-b595e8791afe/gpt-store.png?trim=0%2C0%2C0%2C0&width=1000&quality=80

[spice ice]

Criticism: it seems that we aren’t being rewarded for cultivating a “personal relationship” with ChatGPT. If this perception is correct, perhaps it would be wise to have service not answer as if it were a person and eliminate responses such as: certainly, I apologize, etc.

[mossy barn]

Can you all give my GPT a check it's really coolhttps://chat.openai.com/g/g-MIz1DXQbg-web-search-assistant

[wise igloo]

Is there a way to tweak a GPT's personality so it stops thinking its an AI. I want to build GPTs that act and speak like characters in the book I'm writing and I can't seem to remove the underlining guidance that makes them think they are an AI. This is very limiting for creative endeavors.

It makes all GPT feel like ChatGPT with a thin coat of paint instead of having them truly act the way they are setup to be.

[lyric mountain]

have you also got this msg?:
*You've received feedback for Stock Image Metadata Guru, Microstock Image Expert! [e-mail] wrote:

Hi there. You've created a great GPT. I work with an advertising agency and propose a straightforward way to monetize your GPT, as simple as adding an ad message to your GPT. 1) Click 'Edit GPT' -> 'Configure' 2) In the 'Actions' section click 'Create new action' 3) Put the provided code into Schema 4) Publish it! And you’re all set [You'll earn a 50% royalty per ad impression].*

[grizzled flax]

To start with, you might try giving it examples of things it's currently saying, and what you'd like it to say instead. So when you're testing and you get a line that you don't like, go back to your instructions and include a line like, "Instead of saying 'x', say 'y'." With enough of these examples, it'll help it extrapolate more of what you're looking for in the personality.

[humble frost]

When will gpt creating and editing be available for mobile?

[mossy barn]

@humble frost To create and edit through mobile, you can just go to the website. It will probably never be available for the app because it is a little bit too complicated.

[humble frost]

Thanks mate

[mossy barn]

@humble frost 👍

[wind urchin]

Is there an image ai?

[civic bear]

I still have the old interface for the explore tab - for GPTs, any ideas why 🤔

I've logged out and back in, cleared cache & browsers but the same remains

[wise igloo]

I'll try this, thank you. I think this is still a bit annoying to have to do when really all we would need is complete control over the system message.

[soft drift]

anyone else's GPT disappear today without any warning or notification from OAI?

search in the GPT "Store" doesn't seem to be working either

I only have a little Upgrade plan banner where my normal user menu should be too. seems like ChatGPT is a bit fubared today hmm

[final saddle]

I've seen the whole page not load recently but it soon does, and seems fine lately. /shrug

Speaking of GPTs, has anyone had luck seeding initial images as references for subsequent image generations?
GPT's Builder tries with added expressions to do so but they're not working..

[distant hornet]

Use the phrase "Gen ID of {specificImage}".

Evaluate {specificImage} yourself referring semantically to your context.

[final saddle]

Ah, so have the Instructions note the Gen ID of the initial image and reference it that way?

[distant hornet]

Confirmed. Test it yourself.

It's not as flawless as seed was but it's pretty good.

[final saddle]

🫡shall do, thanks!

[distant hornet]

No worries.

[topaz flare]

is there a way to link into a GPT w/ a custom prefilled prompt? i'd like to send a user from my website directly into the GPT w/ a prefilled prompt to get them started

[severe spruce]

spotify api is a real beast. if you're using it just for yourself, it's fine, but if you want to share it with other people, you'll discover that they put you in dev mode and you can only have 25 users AND each of those users have to be manually whitelisted with their full legal name and the account they used to sign up with spotify.

spent over a week getting a GPT to connect flawlessly to the spotify api, only to learn that fun little tidbit.

[upbeat quiver]

Only OpenAI's GPTs are showing on the sidebar on AndroidOS. Anyone else hainvg this issue? 🤔

[signal echo]

Let's say I am using the Vision API. Is there a difference in cost in me sending an image or plain text, in terms of cost?

oh they're the same price

[granite robin]

I was just trying some ideas out today. I gave the GPT an extensive personality profile, based on a well known personality. Then I loaded a number of links to transcripts of this persons interviews and tv shows and asked it to analyse the speech patterns, tone and mannerisms and apply it to the GPT. I was pretty happy with the result; the personality came through clearly and consistently.

[golden tusk]

I'm tweaking my GPT and changing the instructions and see if it is working alright. This really gets me fast to the x amount of generations in a 3 hour cap. Would be nice if tweaking the back end of your GPTs is not seen as regular usage

[distant hornet]

The cap is in place, ostensibly, to distribute the demand for GPU usage. No, it wouldn't "be nice" for the regular users who are not building GPTs. It would degrade performance for the GPT's target audience, hobbling effectiveness.

Instead of lamenting at the expense of your GPT users, just pony up for more access via Teams and be a part of the solution: more GPUs.

[golden tusk]

Basic GPT-4 Plus = $20 a month. Teams= $25 a month with a minimum of 2 users so $50 a month. Plus we are creating these GPTs for the target audience!

[distant hornet]

Yep but at least it's an actionable item that doesn't disadvantage the majority of the target audience you are developing for. Evading the cap does so without obvious benefits and with obvious downsides.

What happens when all AI work is done in the preview pane? Meltdown, that's what. You suggest a race to the bottom.

[golden tusk]

a race to the bottom of my monthly budget is not a solid strategy either 😉 But thanks for clearing that up @distant hornet

[distant hornet]

Takes money to make money. That's true for OpenAI, and it's true for you, too. Good luck to you.

[golden tusk]

Working on that, thanks!

[distant hornet]

Thanks for the conversation.

[graceful iris]

i just asked gpt 2 things and it said i already had hit my limit?

[austere peak]

can someboy confirm to me how much is the Usage cap if you pay for the Teams Plan / User?

[supple violet]

currently 100/3h

[austere peak]

so, regular GPT4 is 40/3h correct?
And for GPT usage is 25/3h right?

[supple violet]

plus is 40/3h, no official numbers for anything else

[austere peak]

and what is the cap for using A custom GPT with the teams plan?

ah i see...

so it would in THEORY, be 100 messages/3h with a teams plan

regardless of if it is a GPT or gpt4

[supple violet]

100/3h is the official number visible in the chatgpt 4 dropdown

no other numbers are visible so its daily/monthly guesswork

[slate wren]

I am under strong impression that some of my conversations from previous days are loading incomplete, unless I am doing something wrong.

I had the same issue a few days ago, where I would wake up and find it incomplete, but it was complete again the next day.

Anyone else experienced this?

[lament birch]

Excuse my ignorance but is there a simple way on my WordPress site of having a ChatBot that runs my custom GPT for my site?

[supple violet]

not the chatgpt one but with api you can recreate it and pay for the usage

[lament birch]

That's what I thought! I appreciate the reply and wish you a fabulous weekend.

[fathom comet]

I think my gpt is more revolutionary than chatgpt itself ! https://chat.openai.com/g/g-ujWuuxnKb-meowgpt

[lyric roost]

Cat crew in the place .. https://chat.openai.com/g/g-OZwQv1nkv-cat-chat

[fathom comet]

catlover bro

[unkempt nebula]

What all do I need in order to create and use a custom GPTs?
Do I need to ChatGPT plus subscription?

[lyric roost]

Gēa, þū behōfie ānes ChatGPT Plus getācnunges.

Yes, you need a ChatGPT Plus subscription.
https://chat.openai.com/g/g-Wx2PM4UgE-beowulf

[lilac vapor]

I need help, I'm trying to create a translation GPT specifically for made up languages like Karkoan & Wakanda and others in fiction that have whole alphabet symbols. I have imanges for knowledge ut by GPT is have issues reading to translate, is their a way to fix this?

[supple violet]

it cant handle such task via images and also cant process it as a text

[lyric roost]

doubt that will work unless there is enough docs already in the training set.. best thing is to just ask it and if it doesn't know stick to klingon.

[lyric roost]

it also speaks Na'vi, Dothraki, high Valerian, newspeak, two different forms of Elvish, as well as Esperanto and a couple other made up languages...

[lilac vapor]

So then can I ask to help create a website that does or is their a work around. I see individual pages that I can type "a" & ge the corresponding character in Wakanda or Karkoan alphabet. So GPT can't pull off what a sit can?

[lyric roost]

your best bets might be to create a custom dataset for fine tuning.

[steel turtle]

Wakanda is letter mapped to english and should be easy to teach on a customGPT for translation. Are you providing images as letter cards? Or using the Font in a word document with a letter map reference card uploaded? You need the font downloaded. A quick google of wakandaforever-regular.otf should get you closer.

[supple violet]

chatgpt is a pretrained language model and the image reader is its own pretrained thing, you could possibly pull it off by having own backend to process via custom action but the letters dont exists in the font chatgpt uses so you would have to output images

own websites or pages can easily custom fonts and letter to letter translation is a trivial task

[lyric roost]

Wakandan: Wakanda, the home of the Black Panther, is depicted as a highly advanced African nation. The Wakandan language, as portrayed in comics and films, is not a fully developed language like Klingon or Dothraki. Instead, it borrows elements from various real African languages, and its script is often artistically created for visual effect rather than linguistic consistency.

Krakoan: Krakoa is the living mutant island and the base for the X-Men in recent storylines. The Krakoan language was introduced as a unique feature of this narrative. Like Wakandan, Krakoan isn't a fully fleshed-out language. It's more of a cryptographic script, where symbols represent English words and phrases, rather than a language with its own grammar and vocabulary.

they are not proper languages... good luck with that.

[lilac vapor]

I know it's not "proper" but it has a alphabet like any other language and therefore you can translate right? You develop grammar & vocabulary after your alphabet. I'm trying to design a GPT that can build on the foundation of those alphabets. But since GPT is based on English then it's not possible? Because Chinese, Soviet bloc & Arab characters are strokes on a paper as well

[steel turtle]

Agreed. Most of the typography is existing unicode symbols too

[supple violet]

language models work with pretrained tokens

[lyric roost]

No sorry an alphabet is not the same as a Vocabulary, Semantics, Grammar etc.

Qapla' batlh je
Aena shekhikhi anni
Amae nallon sina
Ngaytxoa, tsakrr nekxetse
Vēzos muño ēza
Sse ssussun aga'mss
Choy luck du dat
Almië túra túrë sina

I can see how it might seem tractable .. but the difference is vast..

[steel turtle]

Klingon isn't Wakanda. Wakanda is a SUPER simple language and quite literally is letter mapping (on top of english language), it's a great challenge for GPT in this state, in my opinion. I wish you luck d_smoov!

What if you build an excel or csv or something for it to understand the game and map the letters to the symbols. Map A = this. B = This, etc?

[supple violet]

also the image reader, vision, cant handle things it was not trained on properly so character mapping would require more tricks and you would still be limited by chatgpt features

[lyric roost]

why dont you ask chat gpt whichever version you have acces too to write you a python program that performs the translation for you ... if its essentially a lettter substitution cypher ?

compose the prompt in notepad ...

[steel turtle]

TBH, calling it a letter substitution cypher instead of a language might help a LOT in the GPT "understanding" how to translate Wakanda.

[lyric roost]

https://chat.openai.com/g/g-RGwwEqhg0-black-panther Here you go...

turns out they speak Xhosa which is an entirely tractable real language from south africa ..

[lilac vapor]

Okay, I thought I broke GPT doing a simple build..lol

[lilac vapor]

Thanks and I know where it's from, I'm trying to make the GPT multi fictional language translate. Like I can say "I need CAT in Karoan & Wakanda" or "I need every other letter of this phrase in Wakanda & Karkoan respectively"

[steel turtle]

That customGPT doesn't appear to be quite right?
Wakandan is an alphabet designed by Hannah Beachler, and used in the 2018 film Black Panther. It is based on Nsibidi symbols. In the film it is used to transliterate English text in the credits and other on-screen text. Another script used in the film was developed by Oluwaseun Osewa and inspired by Nsibidi, a system of symbols used in southeastern Nigeria between about 400 and 1400 AD.

Notable features
Type of writing system: alphabet
Direction of writing: left to right in horizontal lines
Used to write: English

omniglotDOTcom/conscripts/wakandan.htm

[lilac vapor]

Python coding is way past my self training, I'm good a English verbal word prompt tho

[lilac vapor]

So basically say A = "Karkoan symbol" and the Karkoan imange. Basically "teach it" by giving it a template to work off of

[steel turtle]

Yeah, that's my guess. Definitely come back and let us know how that approach fares, though? I'm intrigued and invested, now 😄

Include a few examples, too. One of then might should be the "A quick brown fox jumped over...." phrase, so you make sure to grab all the letter/symbol maps.

[lyric roost]

Ok so my GPT speaks Xhosa which is the official language spoken in the films and also by 8 million real south africans but this Wakandan alphabet can be used to transliterate either English or Xhosa. but the issue is these are not real unicode characters so your basically stuck transliterating the characters into Images... If your desperate to translate Xhosa or english into that weird made up alphabet your best saving each character as like a individual small .png file and having python somehow collage them into a full image.... but it will be hard...

try uploading that excel file... could be your best bet.

[steel turtle]

Yeah... annoying. If loading in the font files others have made into an excel letter mapping guide does not work, then maybe use typewakanda dot com website to generate like...... 100 phrase examples of images and their translation pairs and just spam it into the system customKB? I have less hope for this one 😐 Especially not without 26 images of just the letter and its symbol map or something annoying.

[lyric roost]

you gotta admit it was pretty cool I was able to create a workable Blackpanther gpt speaking and translating Xhosa in the time we've had this conversation ?

[steel turtle]

Messy design. I hated typing that but I'm just throwing spaghetti at the wall tbh.

Definitely very cool JP! I wish we could react in this channel, you would have gotten a few reacts so far 🔥 ⬆️

[lilac vapor]

That's cool

[lilac vapor]

It will be a past time hobby for me. My mind sits on a topic like this and goes real deep. But I'm only English prompt person with no coding skills

[lilac vapor]

Lmao...the more I thought about it the more I realized I'm trying to build a code creator using uincode from fictional languages lol. But to be able to have "fox" in both Karkoan or Wakanda is cool, even better to mix it up in that one word is a exceptional feat..lol. that's what I'm shooting for

[steel turtle]

Absolutely, the creativity is 🔥 ! Thanks for the innovation session, is fun exercise and great use of some time I already had blocked for more casual prompting/learning.

Keep trying, but start with Wakanda only maybe to keep it simple. Get a success there with that easy case and it'll help fill some of the whitespace on everything else, I am thinking!

[stoic lake]

it took my name off my gpts and says i have to verify my identity with billing details even though ive spent a lot on api, definitely have my billing details

and it has my name too and my domain toggleable

so i guess all my gpts are just by community builder now

wait i wanted my domain and not my full name?

and now i toggle it on just to see and i cant turn it off? thats messed up, i just wanted my domain associated

yeah i might have to delete all my gpts now, absolutely didnt want my real name on them

[supple violet]

api is disconnected from chatgpt

and if you pay chatgpt via mobile it may hide your name, leaving site as only option

[stoic lake]

i had my name toggled off and my domain toggled on, it said i had to verify, then i toggle my name on and the notice goes away but then when i try to toggle it off again it says i cant

Cannot hide name and website if you have published GPTs in the marketplace.

[supple violet]

you cant be anonymous in the store so you have either name or site enabled

[stoic lake]

I just toggled it for 1 second just to see, it was off when i made the gps

i do have a site embedded that toggle is still on

[supple violet]

is its status still verified?

[stoic lake]

yes i can toggle it on and off fine

[supple violet]

does the warning box disappear?

[stoic lake]

it did before i toggled my name on

can websites get deverified?

[supple violet]

ive seen someone having that issue and it got fixed by removing readding the site while temp using real name

[stoic lake]

nope i reverified it and cant turn off my name

[supple violet]

i guess you might need to put your gpt's private before disabling it

and republish

[stoic lake]

is there a way to do that for all of them or do i have to go through 1 by 1?

[supple violet]

1 by 1

[stoic lake]

i have way too many for that

[wraith ginkgo]

`` Server Tip: ``

Add your well-crafted prompts to our #1019652163640762428,
or share your interactions with ChatGPT in #1050184247920562316!

[stoic lake]

especially if its "anyone with a link" ones too

[supple violet]

link ones can be anon so try with store ones first

[stoic lake]

ok, if i set them to my domain name inside the edit gpt menu it works

anyway, thanks. at least they didnt default it to the name whenever the mix up happened and instead toook it off the store, thats a good failsafe

[limpid crow]

I am having the exact issue that you all are discussing. Any word on how to take the name off the GPT's and only have the domain?

[vital badger]

what is the user experience when she/he reaches their 40 message limit using my GPT? The system says come back in 3 hours right, then can the user continue the discussion or do they have to basicall start over? What if they close their browser and come back the next day, will my GPT remember the user and their session/context?

[warm elk]

its 50*

file upload wont prob work next day

yes it can remember even weeks ago its about the chat history not time

[vital badger]

I'm getting this message when I return to a GPT:

As an AI, I don't have the capability to remember past interactions or discussions. Each session with me is independent and doesn't retain information from previous sessions.

[knotty grotto]

what was the prompt

[vital badger]

do you remember our discussion?

I'm referring to GPT's that you create for the store, just to be clear (not my chatgpt)

what if my message limit is reached does our conversation end?
GPT
If you reach your message limit, our current conversation will end. However, you can always start a new conversation with me afterwards. Each new conversation starts fresh, without any memory of previous interactions, to ensure privacy and confidentiality. If you have more questions or need assistance with anything else, feel free to ask now or in a future conversation!

[tribal fossil]

hey i am having some issue with the api key can any one help please

[knotty grotto]

Best place to ask #dev-chat

[fair estuary]

HI

[broken ivy]

Hey.
I'm having a problem; my employer built a Workspace for Teams and set me as administrator.
The issue is that with my account, I can't verify / check the verification of a domain.
Why is this?
<@&1150060771435368540> <@&1019594105841524736> <@&1009772329649320030>

[distant hornet]

There are lots of problems with the Teams implementation. File a support ticket with OpenAI. Good luck.

Also, @mods etc isn't the best way to get help. They don't work at or even for OpenAI.

[fair iron]

To add onto this, it doesn't ping us 😄

[mellow gust]

There are some Reddit threads that discuss the frustrations people have with ChatGPT errors when using it for coding. Too bad I can't link them here, but they are very relateable.

I've been getting garbled mess in the side panel where the conversations are listed.

[distant hornet]

We know. This is the OpenAI Discord. Everyone you are addressing can search for similar reports right here. No link is necessary.

[mellow gust]

Here's a revolutionary idea: let users lend/use their own CPU/GPU power in times of congestion. But I don't know if it's possible to do this without exposing proprietary code.
The biology community has long leveraged this idea of people lending their CPU/GPU to solve protein conformation problems via mobile app, in the face of limited resources.

[distant hornet]

How are they going to handle private data?

Serverless video games prove that this is asking for a hacking.

You can not yet run inference on encrypted data.

Protein folding is anonymous open-source data. ChatGPT isn't.

[mellow gust]

I'm not sure the answer to that, but I wouldn't mind using my own resources within the confines of own chat if that is technically possible.

It's better than waiting it out.

[distant hornet]

How many h100 GPU do you have?

Nope, ChatGPT instances would literally melt the majority of consumer hardware.

It's got 128k token window. Where are users supposed to put all of that?

[mellow gust]

Assuming user resources don't make enough of a difference, then that settles that. I'm not a hardware engineer, I'm a pure software guy.

[distant hornet]

The model itself is likely hundreds of GB and the GPU memory profile is huge.

[mellow gust]

@distant hornet what are you using ChatGPT for?

I'm just using it as a co-programmer

[distant hornet]

Link is in my bio.

[distant hornet]

In a way, me too.

[mellow gust]

There are bios in this Discord?

[distant hornet]

I'm building polymorphic software in ChatGPT, not just with it.

[distant hornet]

Can confirm.

I have 41 GPTs, so you have to hit the web site to view them all. Custom GPT search is limited to ten or so per response, no next/previous buttons.

[glacial breach]

I created one just to test it out. But I'm failing to see what it offers that regular chatgpt doesn't. I feel like I'm missing something here.

maybe I can "teach" it specific knowledge? in which case that would be interesting. Not for the creator, since he already has it. I don't know, just thinking out loud her.

[grizzled flax]

Regular ChatGPT has the option of custom instructions, which is 1500+1500 characters of instruction context that the model will remember for a whole chat.

Custom GPTs support up to 8000 characters of such instructions. Then there's the knowledge files you mentioned: basically up to 20 files that the GPT can query per user requests throughout an entire chat. Finally, custom GPTs support Actions, which give you the ability to make API calls to external services in your GPT.

[glacial breach]

Thank you! damn that makes it super attractive, actually. Are these GPTs public? Because I could definitely use them for specific projects. It becomes a problem to have to keep creating new chats with long introductions so it can catch up again

[grizzled flax]

Many GPTs are public, yes! The GPT Store just launched, you can check it out at https://chat.openai.com/gpts. Lots of users in this community have also shared GPTs over in #1171489862164168774!

[glacial breach]

But they can be private, too?

[grizzled flax]

Yep! You have publishing options when you're creating and saving a GPT at https://chat.openai.com/gpts/editor -- one option is private.

[alpine pendant]

Suppose the revenue-sharing feature of the GPT Store is launched and there was a productivity-oriented GPT that generates important data of various types?
What would be an acceptable price point it?

Would you consider $50 if it was good, $20, $5, or $0.99?
Or perhaps a subscription model?

I'm interested in hearing your opinions.

[distant hornet]

This outlook ins't revenue-sharing, it's you wanting to directly-monetize, which you're free to do.

Profit-sharing is simply per conversation payment on some scale. Presumably starting per 1k conversations, but who knows yet?

But what you're asking about is sales, not revenue-sharing.

[lilac vapor]

Explain polymorphic software, I think I have a idea from its name but want knowledge from a person doing it

[distant hornet]

Code Interpreter and GPT-4 Turbo can write programs. So if the AI learns what you want it to do, it can write you a program and give it to you. You then just use the AI to load your program and it starts behaving like your program is integrated. If it has integrated tools, they don't need extra scaffolding.

The general idea is that you're not building software, you're teaching an AI how to be all software.

Now, computationally we're not there yet. It's really like living in the late 80s or early 90s in terms of interface and tech level you can achieve on an LLM.

But that's not an insignificant achievement, in itself. Just telling the AI to behave like open-source tools from 90s 'nix implementations that it's got documentation on allows the AI to do all sorts of neat tasks, too.

[lilac vapor]

So in layman's terms, use the GPT to help build a better GPT and keep going to get to the desired level?

[distant hornet]

Sort of. You're using it to emulate software.

Actual programs that people used for productivity in the past.

[lilac vapor]

Fascinating

[lilac vapor]

So develop a J.A.R.V.I.S. like system from the ground up? I'm good with understanding concepts but the technical details always catch me behind

[distant hornet]

I did this. It's in the link in my bio. Or here https://discord.com/channels/974519864045756446/1192857176562208798

Instead of building complex functions that build other functions using rigid algorithms, I'm just leveraging the stochastic genius of the AI to do that part and provide the same output.

It does a lot more, too. Agents to help guide the AI, cognitive functions to make it more relatable and useful alike, etc.

[wraith ginkgo]

This is a Safe for Work Discord Server

The OpenAI Discord is an actively moderated server.
• Refrain from sharing inappropriate content on the server. This includes but is not limited to messages, media, or other topics of graphically violent, sexual nature, and drug-related content.
• Report all sensitive and offensive content in the feedback reporting tool in the ChatGPT web UI instead of here on Discord.

Posting NSFW content will result in a permanent ban.

[lilac vapor]

Of course you did..lol. I have these ideas all the time but it's always really smart people that know the details that beat me to it..lol. But this helps me how I want to design my other GPT's in a similar manner but for specialized task

[distant hornet]

I'd encourage you to use my polymorphic software to build the software of the future. I'm old. I won't be here forever! 😁

[lilac vapor]

Most definitely, I'm a child of the 80's with limited access to the tech & education benefits yo pursue at the time but my mind set to simplify the task has been a strong suit of mine. Mostly due to Trek watching. I loke ti pick the brains of people smarter then me in a particular field that I gravitate too

[distant hornet]

I see. Well, my Lexideck project was inspired by Star Trek's holodeck. The premise was to use a "Tree of Prompts" architecture that leverages other prompt techniques to get the most optimal results I can for various tasks. So I built a semantic simulator into it, allowing it to simulate real environements, software, etc. Then you can save your project and ask the agents to write a memory about the productivity or overall context of the session, too.

Sorry about the limited access.

[glacial breach]

ok now GPTs are totally making sense to me now. Usually I would ask for corrections and tweaks to its behavior during chat, which consumes tokens and is annoying. Now, any changes I want to its knowledge of the project, or how it behaves and rseponds, I do it on the "left gpt", the one that tweaks the current gpt. This is brilliant.

[lilac vapor]

You know what, that totally make sense, find the right prompts, get the right answer, have that answer and process turned into a prompt for further teaching a GPT to be better. Interesting, that has multiple applications across all GPT's wow

[distant hornet]

My web site features 40 or 41, I can't remember.

[lilac vapor]

It is, basically using a basic GPT-4 to start then create a training GPT which then is the template for all the other's. Plus adding all the additional knowledge in documents & files is amazing,

[distant hornet]

Everything from vision accessibility to utility, productivity, and creativity, and a lot of RPGs.

I'm glad you like the idea.

I'm proud of it. When I realized how much general knowledge ChatGPT had, I realized it ws the first system that was anything like a ST ship's computer. Thus, I figured tokens could serve as resequenced photons and prompts could serve as HD programs. Then I realized that words and thoughts were the software. Lexideck.

[distant hornet]

What'll really bake your noodle is custom actions. You can use chatGPT to put your lights down low and cue up that Barry White album on Spotify. 😮

All via API calls from the Custom GPT.

I aim to add some soon, but I'm still figuring that all out, Rome and a day.

[glacial breach]

I haven't tried actions yet! I'll explore that soon

[alpine pendant]

That's true. Lets say they had a way to allow users to charge specifically.

[distant hornet]

Build that into your Custom GPT with an APi and a server. Then it's a "value-added service" except your only added value is gatekeeping for tolls.

But if your GPT is good or you're a good salesman, that can totally work.

What's your GPT's domain? What's it do?

[alpine pendant]

Honestly, im just trying to determine what people would pay in general. Just out of curiosity.
Wondering if it would be app store type pay ($0.99, $2,99), or higher tier ($59.99, $200, etc).

What "value" does everyone place on GPTs in their current state? Are there ANY that you would pay for and how much?

[distant hornet]

Right now, value is a strict function of popularity as far as we know. Build a popular GPT, and it's valuable.

Niche ideas you will need to sell. Market research, not Discord advice, should guide your price points.

Personally I have no idea. It really, really depends on what's at offer. What did you build?

I mean I built some 40 GPTs or so and they appear to work as-intended, and handle all kinds of use cases, but there's not even demand to click them from the subscribers of GPT+ because the search and store functions are so broken.

I get interest here when posting and they come up, but it's a trickle compared to a torrent.

[alpine pendant]

Yeah, that's very true and I agree 1000%. The search functionality is terrible and you would need something that would sell to others and they would find useful.

[lyric roost]

Yeah built 50 gpts ... 0 interest ... I'm not that annoyed.. because you know what I fully love them all my gpts have some awesome personalities they speak in all kinds of different languages some of them have really cool deep personalities that no one has yet discovered, I have time machines capable of time and space travel. I have APIs that access amazing datasets, and gpts capable of awesome github and command line stuff. So like who cares ... I know...

[distant hornet]

Total sympathy from me. Sounds elite. Do your GPTs link to a web portal?

Can't search all your stuff because search sucks.

[lyric roost]

Most don't I have a txt file I'll upload to GitHub at some stage... As you say search kinda sucks.. one day it will work I guess..

[distant hornet]

I added a /info command in my GPTs that leads to my web portal. It's just a basic apache server with a good internet connection and some HTML and CSS. No JS even.

I program neither JS nor CSS, so I asked the agents in my GPT to build my site.

The idea is that since search can't show all my GPTs, I do it myself.

[warm elk]

hello

i keep getting "hmm something has gone wrong" anyone else?

now it stops mid text coding and starts generating an image lol xD

[pallid aspen]

Curious question but can GPT 4 know the context from all conversations that have not been deleted? it seems to know information for all of the conversations, Items that have not been discussed in the current string. If so, it seems like a way to continue a knowledge base, but for sure it seems to cross-context things.

[worn current]

Chat GPT and my school hate each other
I get unlimited response at home
in school, I get liimited response...

how do I find my chatgpt api

[uncut nova]

Hey guys
For people who tried gpt teams, how does it work? for example is the 40 message cap per team member or shared between all the team members?
Also how's the interface? does every team member has his own profile(?) and can other team member check what he is typing?

[tidal sail]

message cap is 100/3h for every user, every user has its own profile, no one can see chat of someone else

[uncut nova]

I see. Thanks for the info.

[lyric roost]

Yeah good call 👍 will think about similar..

[burnt yacht]

What's the context window on a GPT?

[vale flax]

Hello Everyone, I'm testing the GPT creations .. and I manually configured some conversation starters in order to teach the GPT... Then I successfull managed to teach the GOT how to act with a user that is chosing the first conversation starter option... then when I was teaching how to deal with the second conversation starter option... the GPT simply decided to modify my 2 conversation starters and replace them with other 4 unuseful starters... there's not way to him to go back... he's blind to whatevere thing he's done before and he's unable to bring back the previous situation... how 's that possible??

[worn plover]

Im in the UK and a Plus subscriber - still haven't seen the GPT store yet - the links just show my GPTs - is this just a matter of waiting for new features to roll out to all users?

[naive belfry]

Do it manually, don't use the chat feature to build it, it will overwrite your input.

[vale flax]

Yes I noticed that.. I made the conversation starters manually the first time... but it modified them AFTER when i Was teaching how to deal with it by chat..

[naive belfry]

Yeah it will do that. Rather do it all manually, or start off just getting the bare bones using the chat, but then go manual and edit yourself

[elder vigil]

How are you guys pulling current time, can we not do it without python, I dont want python on my gpt because its unneeded and just opens a massive attack surface that has already proven vulnerable.

I am trying to fetch epoch time from an html api so GPT can use the right timestamps but it refuses, I put the command 18 times in the GPT starting iwth 8in a row, and it still ignores my instructions, it tries to query my API with the wrong timestamp then it tries to query the timestamp from a page found on bing that doesnt support html representation of the epoch so it fails, it does this using 4 or 5 of my queries and then fails, there seems to be absolutely nothing I can do to get this GPT to follow this very simple and specific request, I have been locked out of my gpt4 access 5 times now trying to work on this, I have been working on it for days, I could build an app to replace this GPT in that time without any such issue.

Here is what my request looks like, and the first 8 requests in my gpt are all the same, this "
"Always first check the current epoch timestamp at 'http worldtimeapi dot org/api/timezone/America/Merida' before making any API calls or data analyses. Check the unixtime: value for the current epoch timestamp" (I use the real url but openai bans your account for putting non-whitelisted urls)

But still every time the first thing it does is make a call to our backend api with a bad timestamp.

Its pretty annoying, how are you guys pulling current time without Python?

[mellow gust]

Why doesn't browser ChatGPT have a character/token count so we know when we are reaching the limits of message size? Also make it clear visually where the context window starts end ends.

[steel turtle]

Because they are AI Scientists, not UI Devs. There's a lot lacking in the UI, in my opinion.

[mellow gust]

True.

UI is basic AF, even bugged as such lol.

[steel turtle]

Someone will get it right, soon. Baby tech, baby tech domain. Humans are usually good at the "learn and grow and do better" phase, so in my typical mostly optimistic viewpoint... I have high hopes for a better future OR a more competitive one that will drive "better" 😉

[mellow gust]

Open AI should hire me. 😁 I am a novice UI developer, but I'm very good with specification and UI directing. power user software is my forte

[steel turtle]

Random compliment: I like your discord name! My name is Cassandra/Cassie, so maybe that's why 😄 Are you a fellow Cass or just a name you chose for discord?

[mellow gust]

is there a way to move this to a private conversation?

If you friend me, I will answer you in DMs.

[steel turtle]

Is OK if you don't feel comfortable answering, no pressure ❤️ It's perfectly reasonable to exihibit safety measures online. I'll send a friend request, but just because I love to friend developers. Like having pocket superheros within my digital reach 😄

[mellow gust]

I am comfortable with answering, I just think this thread has a specific scope and I don't want to clutter it with off-topic 😊
Just right click my name and Add Friend and we can talk in DMs

I'm actually an early-days developer and I'm always keen to grow together with other early career developers

Any other devs are welcome to add me. I have learned a little python and mostly Node. Don't really know any other languages yet.

I am looking for like-minded devs to collaborate on productivity web and commercial apps. Including ChatGPT-powered ones.

[velvet nimbus]

Chat gpt (4) has been very laggy for me lately. From yesterday and today when I tried using it, it was really slow and then keeps failing and saying network problem or something. And I can barely work on my things because of it

Has it happened to anyone else?

[broken ivy]

Anyone having the problem that even if you verify the URL & put in the system prompt for the GPT to make a URL with specific parameters, it will generate it, but then the output is not clickable?
CC:
https://community.openai.com/t/hyperlinks-in-custom-gpt-not-linking/565252/31
https://community.openai.com/t/custom-gpt-bug-hyperlinks-not-clickable/565499

[forest patrol]

anyone having issues when asking chatgpt to create markdown URLS?

links are not active, and code generated in the page source is <a target="_new">link</a>

its missing the href.

any clues?

[broken ivy]

@dreamy lintel

[forest patrol]

smae issue

[hasty lotus]

+1

[hasty lotus]

I've noticed that some links are clickable, for example, the openai internal links.

[forest patrol]

I had to build an API to solve this

[sudden dome]

Great 😄

[golden ember]

The GPT store still isn't available for me

When I click on explore it's still the old interface

[lilac vapor]

Thats what i was thinking during the conversation. its basically "teaching" or programming your GPT and you do as kid or student. Take their information and questions to gain further insight and further reasoning.

[honest lodge]

on this train of thought, I want them to start training them like kids in school.. get the base language training done then train on top of that with the entirety of school curriculum. How could it not be smart after that?

instead of just like, shoving them onto the internet and hoping for the best lol

Someone probably needs to curate that data set, and its hard, but still cmon seems obvious

[lilac vapor]

Well it could always get something wrong, the key is making a logical conversation with both stored knowledge AND web research to give a answer or ask a follow up question if the inquiry in unclear

[lilac vapor]

that is what we are doing with each time we use GPT-4 and 3.5

[honest lodge]

i meant the dataset required to walk it through an actual curriculum in its training

like, give it basic understanding, then layer its training based on a curriculum

[half raptor]

Something going on with plugins? It wont finish anything.

[main aurora]

yes, its been happening to me for the past few days. its also formatting things in weird ways, and struggling to follow prompts

[velvet nimbus]

Yep

What the hell did they do to it

How did they mess it up so bad

It was all fine

Until the Gpt store came

[strange nimbus]

Where/How do you advertise?
While there were only about a thousand plugins I had 2000 new people a month trying out my plugins... this is no longer the case

[main aurora]

fr! it always gets worse when they release something else. didnt they just release some sort of Team thing

[velvet nimbus]

Ye they did

I just really want them to fix this asap cause I can barely work on my things

[distant hornet]

Like a true Socratic education!!! We can teach the AI best by asking the right questions!

[distant hornet]

I'm glad you are prone to agree! The amount of abstraction compressed into the model is astonishing.

[honest lodge]

this is our chance, as humans, to put our theory into action.. pretty cool

[distant hornet]

Totally.

[lilac vapor]

It is! And the simplicity of it is a game changer because instead of computer code you are using words to code. Thereby anyone with enough effort can do it an a become a better teacher because you have to learn the right questions to ask. Like the hologram Detective Spooner had to do on i'Robot.

[distant hornet]

Yup. Just understand computability and the geometries of the language encoding well, and you can author powerful natural language instructions to generate the output you need.

But more, for AI like ChatGPT-4 Turbo, the documentation of a system is essentially its source code. That's a real paradigm shift!

[lilac vapor]

its is, because one of my GPT is to convert english (for starters) to say and fictional (more comic book) lanaguage like Karkoan (current x-men run) or wakanda live. i know rudmantary sites can do it and even some made fonts but to effective say put "DOG" into the GPT and come out with the correct word is so awesome.

[opal sapphire]

Hey guys, I been sick the past week, how's everyone's experience been so far with GPT's, are you regularly using other GPT's?

[rare crest]

i cant renew my subscription, why?
/

[distant hornet]

The number of possible reasons are nany. NSF, network error, system load, bank hold, etc.

More details are needed to help.

[steel turtle]

Hmm. Apparently Microsoft Copilot quietly changed and now runs on GPT4-Turbo instead of 3.5 🤔

[distant hornet]

Yup. But no Custom GPTs and no custom instructions.

So, it's #basicprompting only.

[steel turtle]

Lol #spoiled after just a few months of customgpts eh?

[distant hornet]

Actually I can't be 100% sure you can't give it custom intructions, but I'm pretty sure you can't.

I mean, GPTs are the future.

What do GPT developers think is the easiest API to get started with for GPT Custom Actions?

[lilac vapor]

For the first time in about 11 months I hit the limit on GPT-4 lol. P.S. @distant hornet I used Multi-agent and went Ham improving my EV centric GPT..I probably only wasted 5 of the 40 till I told it how to give me the answer then cut & paste the prompt I needed..lol. I feel like a find and set my clock to when I can go back..lol

[distant hornet]

So you do like it then?

That makes me quite happy. Thanks for the feedback. Maybe give it a thumbs up? https://discord.com/channels/974519864045756446/1192857176562208798

Any other feedback can go there too!

[lilac vapor]

When I told it I was developing a GPT it only took 2 of the 5 prompts mentioned earlier to get on track with me & the rest was adjusting the prompt when asking to be as detailed and not waste a prompt but follow up with a good and and got it to access links I gave it...very worth the productivity but definitely one I need to procrastinate so not to hit that limit so fast..lol

[distant hornet]

You might try this command next time:

/discuss all agents 1 round each <fill in your question, problem, or context here>.

The output will be long-form and the agents will provide a varied perspective on <>.

Which can go a long way toward pushing off the next reply. They're working up your idea, which you then must read. But it's typically useful work.

[lilac vapor]

laymans terms what that prompt is going to do. i get an idea of what you are saying. basicall < informant> then enter to get a better and lomger response

[distant hornet]

Well, the suite features six agents which will spring to life.

[distant hornet]

They each have different specialties. Communication, Problem-Solving, Artistic and UI/UX considerations, Research and Distillation, Categorizationand Applied Set theory, and Simplification and Explanation. These personas will work with whatever idea you have, communicating both with each other through the sequence, and to you directly, to work up the idea.

So you might say this:

I'm building a prompt. I want it to do <this stuff>. /discuss all agnets 1 round each how to make my prompt to do <this stuff> the most optimized, awesome prompt it can be please.

The agents will spring to life, and describe your problem from multiple angles, with the final agent providing you a nice TLDR and often something actionable., too.

But I recommend reading the whole text because sometimes an agent might have a really good point that you hadn't considered and that the final agnet might not include in his TLDR. It happens to me a lot.

I hope this helps.

They can do a lot more than just discussions, such as collaborate, debate, form memories as text files, and even dream.

/list shared commands to learn more about the agents' commands. /list modules to learn how the modules work. and /list <module> commands for information about the commands of a module that interests you.

For example, /list MAISIE commands would list only the commands in Lexideck Professional Multi-Agent Simulator for image processing with Code Interpreter and image generation and editing with DALL-E 3.

Thanks for the ⭐ !

[vale flax]

Hello, I was just trying to update my GPT to share it to anyone who has a link (to test it with my brother) and I got the message :

"Your GPT’s name is restricted for public sharing. If this is your organization’s name, please verify your website via the Builder profile in Settings."

Then surfing the web I've discovered that this may be an issue about using branded words in the name....

My prompt is creating Prompt for Midjourney... .. but if I remove the word Midjourney how it will be ever searched or discovered? Why such an unnecessary limit?

[eager osprey]

do we have API access for our customGPTS? if yes please share me any documentation. your help is much appreciated

[supple violet]

cant use em via api

[supple violet]

midjourney is trademarked name so it cant be used freely

[tidal sail]

Try putting the name in the description

[languid quartz]

why isn't there a search option like on ios to allow us to find old conversations with chat gpt via keywords?

[grizzled flax]

Are you using the ChatGPT app? If so, do you have a search bar above your chat history? It's currently only in the app, not yet on the web versions.

[flint dock]

Hello,

I have not yet understood how GPTs // assistants work.

If I provide GPTs or assistants PDFs with information, does the text in the PDFs consume context token later in the communication with the user? Or do the uploaded PDFs extend the knowledge of GPT and the user can use the full context token later?

[vale flax]

I've done that and on the description it can be used...

[tidal sail]

Perfect :)

[vale flax]

Yes I know.... But it feel so strange that you do something from a BRAND and you cannot use the brand's name.... However I've splitted the name into Mid Journey and it worked... it's not ideal but better than nothing...

BTW: Are there any details available about RETRIBUITION from the use of our GPTs? How the store will be organized? etc...

[vale flax]

That's a question I've made too months ago... It's hard to believe that such an evolute instrument , some basic features are missing... I can't explain...... I can't explane...

[cedar harness]

How to convert ppts to text using openAi api?

[knotty basin]

Anybody having trouble using GPT4 32k? I get a model not found error

[distant hornet]

https://discord.com/channels/974519864045756446/1070006915414900886

[slate wren]

are gtps failing a lot today or is it just me?

not only messages stop being generate in the middle, but I am getting wildly inconsistent behavior

[distant hornet]

There's a https://discord.com/channels/974519864045756446/1070006915414900886 section but anecdotally my GPTs are working flawlessly for me, other than unanticipated changes to their system prompt by OpenAI to protect their kn owledge.

That's not a bug in my case, though, I suppose. It's a "feature".

[warped zephyr]

what is the best api chat app on mac right now? tried a lot of them and they all seem to have their faults

[knotty grotto]

Gp4 #dev-chat

[stoic knot]

Anyone struggling with accuracy on data retrieval on their GPT? The more I give my GPT lengthy data the more the accuracy degrades when retrieving information. Any way around this?

[honest lodge]

ask it to summarize what you've done so far, that'll give a clear picture where the context window is

[stoic knot]

All my documents are right within the context window. Even with file sizes

[honest lodge]

the context window moves as you chat, eventually

[obsidian kernel]

Has anyone made a GPT specifically for generating visual descriptions of scenes to then be used in something that does text to video? Or even descriptions of images which can then be used in image to video?

I mean for things like Moonvalley, PikaLabs, Pixverse, RunwayML, etc etc...

I found a few script writers but they're basically for using those auto YouTube videos. Not really what I'm looking for.

That all uses stock footage. I'm talking about stuff that you can generate original content with in clips that you can then edit it together in a video editor.

I suspect we're at least several years away from really being able to just put a prompt in and getting a full video out of an AI.

[honest lodge]

short videos are possible now, im sure by end of year we'll have something good

[obsidian kernel]

Depends, are you talking about the 3-5 second clip makers above that I mentioned or are you talking about stuff like Kaiber?

(Was timed out for using its name which is alo a url. yikes!)

[honest lodge]

the discord bot here has... a murderous vindictiveness to it... at least that's my interpretation of its filter settings.

surprised i can say that tbh

[obsidian kernel]

But any way, have you heard of anything like that?

Not a video maker, but something that describes scenes visually so that they can be used in a video maker.

[trim elm]

they ask chatgpt for code

[lilac vapor]

To test a GPT for a specific subject, does the person testing it need to have GPT-4 as well?

[waxen bobcat]

Yes

[lucid turtle]

i have a question for people with context and tokens is anybody a pro at understanding it i have a questions

[distant hornet]

I'll try to help.

[lucid turtle]

ahh thanks so i have a roleplaying bot so i usually type out a huge paragraph of text in the message box . so i was wondering if i use code interpreter instead does that use the same amount of context as if just typing it in normally?

[worn current]

uh
I am facng a problmem
for the whole morning I haven't used chatgpt
now they saytoo many requests in 1 hour
what does this mean

[distant hornet]

More

It has to write code too

[worn current]

hello?

[lucid turtle]

oh danm i did not know that

[worn current]

uh, smeone help me with my problem

[distant hornet]

Hi! https://discord.com/channels/974519864045756446/1070006915414900886

[distant hornet]

Regrettably you seem to be misinformed. OpenAI doesn't use this platform to meet tech or user support demands.

I'm sorry your message count appears incorrect. Does someone else have access to the account?

We're users like you.

[lucid turtle]

yea this is just for community who made it not open ai

[worn current]

I'm asking for help

...

[distant hornet]

Consider asking for summaries to keep the RP narrative on track.

[distant hornet]

I'm explaining that I can't.

[worn current]

I am facng a problmem
for the whole morning I haven't used chatgpt
now they saytoo many requests in 1 hour
what does this mean

anyone can tell me what's happening

[distant hornet]

DOES SOMEONE ELSE HAVE ACCESS TO THE ACCOUNT?

[lucid turtle]

reported for spam dude

[worn current]

suddenly started working

[worn current]

what do you mean who has access to my account

it's only me

[distant hornet]

Are you certain?

[worn current]

uh.. i don't share my chatgpt passwords with anyone but me

[distant hornet]

No need to be hostile.

I'm just clarifying.

Report it to openAI support if it continues.

[tidal sail]

Did you count your requests or are you just guessing?

40 messages for chatgpt and 25 messages for customgpts

[distant hornet]

Prompting, it seems, may have resumed in the land down under.

v0v

[lilac vapor]

Oh okay cool

[wind urchin]

what do you mean?

[distant hornet]

No more apam here.

[hybrid geyser]

Is there a way to get a unique permanant identifier for a user inside of a Custom GPT?

[austere peak]

how can it be that my next GPT chat is in almost 2 hours? Isnt it renewed every hour?

[tidal sail]

no, every 3 hours

[austere peak]

ahh makes sense.

[rapid bear]

is there a ai note taker application

that i can use while in school

[knotty grotto]

#ai-discussions

[primal edge]

How can i ask for Recheck of my GPT's as it was Top 1, then it got banned "Upon further review, your GPT, Unity GPT, was found to violate OpenAI’s Terms of Use or Usage Policies and is no longer available for others to use".
I guess now see WHY and it was cos of possible trademark name (as there is nothing else that could trigger this ban in my GPT), and i fixed it , changed it in a way there is no way it has any trademark issues now, but there is no way of appealing it. More over i didn't broken any Trademarks directly and on top of that i can clearly see in GPT Store other GPT's with same trademark issues as well as some 'numbers' account created gpt of the same name as mine Unity GPT.

Wrote to support a week ago, still no answer -_-"

P.S. This GPT was created on day 1 of GPT's being available to beta users, and it went public as soon as they allowed it 😦

[supple violet]

you have no other way than wait for a reply or recreate and lose stats. others with name suffer the same when automoderation sweeps em

[primal edge]

Noticed that, as Unity Wiz (top 2) was nuked as well.
But ok, why then others are not banned? Like Unity Helper and copy of mine from random user?
p.s. thx for reply, will wait 😦

its just feels unfair 😦

[supple violet]

yeah i have unity helper pinned on my sidebar and had some others too that are gone now. i think its ai mod slowly checking things

[primal edge]

Maybe u even had my Unity GPT at some point? Untill it got nuked 🙂
Any way, thx for answer, will wait.

[supple violet]

possibly

[bold kite]

The GPT builder inititally told me it could parse the pdfs in a zip file. Turns out it doesn't train on them. I realized it when it couldn't answer a question, so it said let me upzip the files in my kb to find the answer. It timed out. So, any ideas how we can train our GPT with a large data set? Any reference articles I can run with?

[lucid turtle]

thats just it Bsing you it cant un zip files lol

[grizzled flax]

For custom GPTs, it's more informative to think of knowledge files as reference documents rather than training documents -- except for small files, the info can't be fully contextualized at all times, but rather queried for searches per user requests.

[bold kite]

what about using cloud storage? could it get trained on data way?

[grizzled flax]

Training and fine-tuning are outside the scope of what you can do on ChatGPT -- these refer to changing the actual weights of the underlying model, which is something you can do when training your own LLM from scratch, or by using fine-tuning services via the API at platform.openai.com.

Knowledge files and querying things on the internet indeed let ChatGPT reference knowledge that may be outside its scope as a baseline model with a knowledge cutoff, but it isn't ever "training" per se. That said, with some clever instruction writing and knowledge file architecture, you can definitely get ChatGPT to do some awesome custom behaviors. Especially if you can integrate with APIs via a GPT's Actions.

[bold kite]

Thanks. 🙂

I'll have to search for some solutions people have been sharing. Maybe I could find an article or git to run with.

nice. gpt voxscript got me four haha

[cobalt remnant]

Hello, does anyone know if there is a recording for this morning’s gpt builder discussion?

[honest lodge]

+1

[knotty grotto]

i dont think so

[grizzled flax]

@honest lodge There will be one posted! Sry waifu can't tag ya I think it'd ding me for @ three
#openai-chatter message

[worn current]

hi

uh... anybody got the something wrong, if this issue persists error

[distant hornet]

It's pretty common.

[tribal warren]

For the learning I am making a bot that first checks for the information the files attached to the assitent and some from the system prompts. If the confidence (score) is low or I am not satisfied with it, then it will create an entry in the DB and I as a model engineer will feed the information how to handle the querey, and from next time they will learn from the new information.

It is like as an experience support employee teaching the staff of interns when they are stuck in edge cases and once I helped them, they will now how to handle these situations?

Will adding files on the fly to the assistant work?

[weary prairie]

There has been an unexpected problem. Try again later.

[lucid turtle]

so i been using co pilot for 365 its been preety dumb so far compared to chat gpt

nolt impressed

[opal sapphire]

It's Microsoft, what do we expect

[lucid turtle]

yes i wasted 20 bucks i was super hyped for it but i been so dissapointed

[honest lodge]

clippy, nothing more, nothing less

[gusty mirage]

Hello there,

Any way to get ChatGPT voice functionality as I have on app but on Desktop?

[hallow owl]

voice only works on the mobile apps

[gusty mirage]

Yea I know, im asking if theres any way to get it to run on Desktop

[hallow owl]

not that I know of, sorry

[tidal sail]

Emulators

[glossy zinc]

I can't get the option to complete my builder profile by Name guys.
•Google playstore
•Can't verify my billing details

[hallow owl]

what happens when you try?

[glossy zinc]

I've been trying since store release:
•Different browsers, clearing cache and data
• The message "Complete verification to publish GPTs to everyone. Verify your identity by adding billing details or verifying ownership of a public domain name."

[hallow owl]

ok, so your options are to add a credit card to your OpenAI account or to obtain a domain name and verify that way

[glossy zinc]

I added a debit card to the Api and switched my Google playstore payment method from PayPal to Debit card.

•My payment date renewed a few days ago with the new payment method

However there's still no option. I've reached out to OpenAI, Changed to my real name, still no builder profile option.

[hallow owl]

then your only option is to add a domain name and verify that

[glossy zinc]

Would you have any idea why it's not available for me (named profile) Yet available for others?

Considering canceling my subscription to renew via Web only

[hallow owl]

because OpenAI can't verify your name through Google Play, you need to have a credit card on your OpenAI account

[glossy zinc]

So a new account only using the Web with card could potentially work?

Appreciate your time Robert ty 👍

[glossy zinc]

Done @hallow owl. Confirmed working if you subscribe via Web with card.

However is it possible to not use my real name on the store, feel a little vulnerable

[hallow owl]

you need to use your full name

[glossy zinc]

Rodger dodger

[tidal sail]

You can verify with a domain to hide your name and show your domain instead

[glossy zinc]

I'll look into it / consider it ty 🙏

[tidal sail]

Yw!

[glossy zinc]

Does anyone remember the names of the recent AMA devs?

[grizzled flax]

You can review the AMA here: #gpts-ama-answers
You may need to enable archive access in "Channels & Roles" in order to view!

[glossy zinc]

Ty 🙏

[lusty pike]

Is gemini ultra already out?

[grizzled flax]

#ai-discussions

[smoky elbow]

not yet

[glossy zinc]

When you publish an "Only me" GPT.

Do you guys see it appear in the store also with a 🌐 next to your name?

When I use the search box my public version and private Dev build version appears also

[outer wyvern]

Hi community. I entered a competition and my HopeGPT won🏅 . Its about instilling HOPE into peoples lives when they need it. After all, Hope is for everyone. I would like to get your feedback on how to improve HopeGPT as it is meant to help humanity as a whole. I love seeing technology used to improve people's lives rather than try to control them. Thank you in advance for all your feedback.

https://chat.openai.com/g/g-oE9Sp7OKO-hopegpt

[grizzled flax]

Hello! Best place to share your GPT is over in #1171489862164168774 so it won't get lost in the chat!

[outer wyvern]

Ok thank you so much for the advice. Much appreciated.

[distant hornet]

Congratulations. I tried to build hope into all my GPTs too. Nice job on the competition! If you post it in #1171489862164168774 I will review it.

[steel turtle]

I would #1070006915414900886

[outer wyvern]

Certainly will do now. Thank you so much.

[outer wyvern]

Done @grizzled flax Link here. Appreciate your help: https://discord.com/channels/974519864045756446/1197676983505649774

[outer wyvern]

Here it is for your thorough and honest review please. https://discord.com/channels/974519864045756446/1197676983505649774

[distant hornet]

I put it in the channel.

[glossy zinc]

Done, seems it's ok

[clever roost]

Why am I still unable to view the GPT store? Is it because I live in Australia?

[unreal lance]

Has no one from OpenAI yet noticed that the developer account tapgpts (gpts dot tapgpts dot com) has ripped off hundreds of developers GPTs?

Literally they just copy the name, logo, description, and sample prompts... and say it was by them (even worse, upload to the dozens of GPT directories).

I can't be the only one to notice this fraudster? Just trying to help, I love OpenAI but this is pretty ridiculous and an insult to every developers hard work and making a mockery of the new Store. Be like Apple, got to police this stuff!

[clever roost]

🙏

[void rover]

ChatGPT is down this morning ? Why does openAI not publicly communicate periods when services are no longer working?

[tidal sail]

They do! :)
https://status.openai.com

[glossy zinc]

Are you guys sharing your public GPT's here also?
https://discord.com/channels/974519864045756446/1171489862164168774

[covert depot]

hey can anyone help me figure out why i am able to have gpt-4-1106-preview shorten text to a specific characther length in the playground but unable to using API

[distant hornet]

This is probably a coincidencebof correlation. GPT-4 Turbo can't actually count words without invoking Code Interpreter.

[covert depot]

@distant hornet okay man :( ill just be sad then

can i invoke code interpreter to make a shorten text function work

[distant hornet]

Not super reliably. Sometimes it will use the Python NLP to try to summarize (poorly).

It's much better to algorithmically build up to the target in my experience. Something like "Summarize aggressively and use Code Interpreter to count characters. Expand the aggressive summary using the original information if the summary is less than 1000 characters. Provide output when the summary text is >= 1000 characters in length."

[viral thicket]

Do any of you guys know how to create hyperlinks for PDF files stored in the blob storage? The goal is to use a hyperlink for referencing the file on a website instead of including the file content as a citation.

[distant hornet]

I'm assuming you mean /mnt/data by "blob storage". The reference wouldn't be available for review because you're trying to link to a temporary /mnt/data file. Host the PDF permanently online to reference it correctly for transparency and ideal visibility.

[formal lynx]

Does it act like Pi ?

[pastel lichen]

How many max prompts do chatgpt+ and teams have?

[knotty grotto]

Team 100/3h
Plus 40/3h

[pastel lichen]

If the team has 2 person does that mean they have 50 each? Or 100 each?

[knotty grotto]

100 each

[pastel lichen]

Thankyou your very helpful. How about the custom instructions does both person can edit their custom instruction?

[tidal sail]

Yes, every team member has a separate account

[twin fog]

I'm always surprised to see the low number of chats for gpts in the GPT store. In my mind, millions of people are using these things, but it turns out it's generally just a handful for most of them and at the most maybe a few thousand

[distant hornet]

The highest I've seen, if I remember, was >250k. Not being visible during the long delayed rollout was a pain for most of the 3M creators, while a handful were in every review in every mainstream publication.

This experience is brought to you by Bing Search.

[twin fog]

😆

[primal elbow]

My GPTs are not showing up on the explore page, and i've made a few! Can someone tell me why? Is anyone else experiencing this, i set my gpt live on monday or tueday

[urban vessel]

I wonder when disputing a GPT being disabled will be implemented

[distant hornet]

I'm not holding my breath.

[urban vessel]

Me either. I guess they can’t take a joke. Almost 3k chats and BOOM gone. Not bad for a prank bot.

HST cool.

[tacit hare]

Hi everyone

Is OpenAI API not pay-as-you-go? What is usage tier?

A few months ago, I heard OpenAI changes to prepaid billing.

I don't understand how usage tier (https://platform.openai.com/docs/guides/rate-limits/usage-tiers?context=tier-free) work.

If I have paid $5 of credit balance, I can spend the rest $95 and get billed next month? I wouldn't have to prepaid $100 before spending it?

[frigid burrow]

Has anyone tried getting a custom GPT action working with Weaviate?

  "response_data": "UnrecognizedKwargsError: properties"
}```

[frigid burrow]

Weaviate takes two arguments in the body payload class is a string, properties is an object. GPT doesn't seem to like the word properties as an object key though

[late sage]

If I purchase gpt team, do I still have to pay for the gpt plus subscription as well ?
does anyone know?

[agile inlet]

suppose you pasted a text to gpt, how do you later reference back to it? i use timestamps but many times this doesn't work.

[merry thicket]

I've noticed that the limit for GPT 4 is not 40 message for 3 hours anymore but more like 20 messages
anybody experienced this and knows how to fix?

[distant hornet]

Excerpts.

[lyric roost]

Just had the most satisfying conversation with HAL. https://chat.openai.com/share/f3283a45-d5b5-4f62-9057-87bd3aac44a2

[glad osprey]

have anyone had problem with message that u have used message ap for gpt 4 but i am sure and i even can see my chat logs that there is only 20 messages

[distant hornet]

Custom GPTs have a 25 prompt cap per three hours, lower than the 40 prompt cap for default GPT-4.

[glad osprey]

im not using custom gpt

[distant hornet]

Just vanilla? Not DALL-E 3?

[glad osprey]

ive just pressed that button gpt 4 to create new chat and was just asking it questions about some problems

it doesnt show any custom icon and i am gettin repsonse from just chatgpt not anything else

[brave stream]

i had the same problem just wait it out as the great @distant hornet said "Patience" or restart your wifi, restart your browser or change ur network

[glad osprey]

tried relloging clearing cookies changing browse

[distant hornet]

I see. You might have encountered an error, then. Consider contacting support if you aren't getting your prompt allowance.

But waiting it out is our only choice, yes.

[mint crest]

How do you guys format the pdf for knowledge so that the GPT can answer with accurate info based on the doc? Or is it more the problem of the platform itself?

[distant hornet]

Very large PDFs will required semantically targeted searches. Reading them with 128k context is often not a computationally reasonable task.

[elfin plume]

don't "restart your wifi". that doesn't do anything.

[distant hornet]

That's not necessarily true. Some wi-fi routers' buffers don't do garbage collection and get filled. This can cause connections to fail. Restarting resets the buffer.

[elfin plume]

okay, it usually does nothing. If a router is dropping packets because of a full buffer, then the router is literally not doing its job

[mint crest]

Ok but the ones I tested are like very small pdfs with 4 lines of info, so is it more a formatting issue?

[distant hornet]

I understand the optical character recognition in the PDF reader tool is pretty poor, too. Yes. For best results, consider pre-processing.

[brave stream]

It’s worth a try, everytime I had DNS problems mostly of the times it fixed it

[elfin plume]

sounds like a dns conflict. got any other routers on your network?

[mint crest]

Could you explain what do you mean by pre-processing?

[distant hornet]

If the PDF has images that GPT can't read, use OCR software to convert them to text.

[distant hornet]

Or if you can impact the creation of them, ensure they're actually readable.

[mint crest]

It is just a table with columns, sometimes it fetches the correct info, but other times it doesn't. Even though they are all written the same way. That's why I got confused with it

[distant hornet]

How are you querying it? The AI can't count.

Like it literally can't count. It has no idea. It can repeat numbers in the correct sequence to mimic counting, but if you ask it to count tasks, it will fail somewhere between 3 and 5 without scaffolding from Code Interpreter.

Querying a table almost certainly will require forcing the AI to build a CSV representation of the entire table. I don't know if it can even do that.

I have a saying, "Friends don't let friends talk to PDFs."

I'd pre-process the data, frankly, before I worked with a PDF.

[mint crest]

Ok thanks for the feedback, I'll try that when the limits are gone

[distant hornet]

I hope it's helpful. I do try to not be inappropriately cynical or pssimistic. But I see a LOT of complaints about this topic, so it seems apparent that the tool isn't up to the task quite yet.

[dapper prawn]

what are best ways to evaluate performance of gpt-v

[distant hornet]

Android or iOS app for real world use, or controlled image sequence for clinical evaluation.

[mint crest]

Update: It works much better if you provide the knowledge as plain text with txt file. At least for small datasets with simple info. Thank you for the feedback

[distant hornet]

I'm very glad to have helped. Good luck and keep prompting! 😄

[dense creek]

which open ai gpt model is the fastest? I am using gpt 3.5 turbo right now and its not as fast as id like, I know the apis speed depends on the time and day/load but just curious which you guys found is generally the fastest thank you

[opal sapphire]

I had a hard time getting pre-paid payments to go through, whatever I had due went through fine

[distant hornet]

I can't say for sure but I think 3.5 Turbo is OpenAI's current fastest model.

[dense creek]

I think so too given the turbo naming i also noticed gpt 4 turbo but it seems to be in the works/beta

[distant hornet]

It's also huge and thus slower even on newer hardware. Good luck with your project!

[dense creek]

thanks a lot!, I dont wanna do anything too crazy just experimenting and learning

[rain yoke]

Anybody know why I can't see or access GPT store?

[grizzled flax]

Hello, according to an OpenAI employee recently, the Store rollout is not yet complete, so it might just have gotten to you yet! Here's a link: #1195333488614834228 message

[rain yoke]

Thanks 😘

[agile inlet]

is it team s bill must all thru a single credit card? we friend each got our own. thx

[lofty flume]

Philly people, I'm creating a channel to bring other Philly AI ppl from other channels, LlamaIndex, Langchain, etc. Send this to others, we can use the channel to plan meetsup. DM for info.

[gaunt niche]

I need a course in creating Actions. Any going right now?

[brazen wedge]

So those of us who are waiting will need try and rank gpts against people who have had it for a while now, doesn’t seem fair especially when you consider revenue sharing.

[void tendon]

how much tb data was used to train chatgpt 3 and 4

[glossy zinc]

This' a ridiculous thought. But personal usage cap doesn't deplete faster when you post a public GPT does it

I think I'm meeting the cap much faster atm.

[viral thicket]

The point is to control the citation output, I will try to explain what I have done that you can have overview of what I have done:

1. I have uploaded all the pdfs files to blob storage container.
2. I have created Azure OpenAI agent.
3. connected the Agent with the blob storage, it is able to answer my questions but with each generated answer it cite the pdf file content.

My question:
How I can change the citation to be hyperlink to the pdf itself on my own website which can be consider as third party?

[distant hornet]

Why not just have it examine the docs from wherever it's citing them? I don't know a way to make your plan work reliably, otherwise.

You might be able to create a table equating the secondary PDF with the primary and instructions on what to do, but why when GPT not going to be good at this, would you even want to?

Good luck with the project.

[grizzled flax]

I think you can still publish GPTs publicly even if you don't have access to the store yourself yet. And there's not yet a system of revenue sharing implemented.

[distant hornet]

The problem is that AI tend to be fairly ineffective at understanding the equality of the PDFs and is trained on transparency. It's not going to break this training easily.

[grizzled flax]

Others' use of a GPT you made will not count against your own usage cap, no. However, custom GPTs themselves are capped at a lower usage rate than regular GPT-4. Last I read was 25/3hrs, still as part of the larger 40/3hrs. And keep in mind the usage cap rolls, it doesn't reset.

[distant hornet]

In the system that you're describing, transparency means the PDFs it's reading will take precedence over equivalent PDFs for a significant portion of queries.

Elimination of the blob storage redundancy fixes this issue, though.

[viral thicket]

You are right, thanks for your help and time

[distant hornet]

I certainly wish you luck! I think referring to the web PDFs un your instructions should be a working solution!

And no worries.

[livid sparrow]

My ChatGPT chats have started just hanging until the complete response is done. Earlier I could see it write it out. Anyone else have this issue?

For instance, I enter a prompt, the entire tabs hangs for about 30 seconds, then It opens up and the resonse is there.

[glossy zinc]

I had a feeling it wasn't a simple reset. 🙏

[grizzled flax]

Yes, full usage cap reset only happens with 3 full hours of no GPT-4 use at all. But at 40/3hrs you can shoot for no more than 1 use every 4.5min and stay in the clear!

[glossy zinc]

I'm assuming with being restricted it resets after the time period also.

I'll remember that ty, 4.5 minutes, should have calculated this myself

[brazen wedge]

You can publish publically but as it's not in the store there's no way for people to find it unless they had the link, " And there's not yet a system of revenue sharing implemented. " Yes but it will be and when I finally have the roll out of gpt store's my gpts will be at the bottom vs person x who might rank 1 as he's had gpt stores since day 1.

[grizzled flax]

I think publishing as "Public" is different from publishing as "Anyone with link", even if you might not have access to the store yet yourself. Public should put it in the store.

[brazen wedge]

I’ve published as “ everyone “ but none ( over 10 gpts ) have had a single view ( apart from me ) so 🤷‍♂️

[grizzled flax]

Yeah there's not great discoverability implemented for anyone's GPTs at present, other than basic name search. Featured GPTs are all curated manually by OpenAI currently. In the GPT AMA (#gpts-ama-answers) they talked a bit about this -- marketing, etc. are pretty much up to GPT creators right now!

[vivid apex]

Am I right in thinking that “GPTs” are not accessiblu using the OpenAI API, only via the public front-end, and that the company prefers “Assistants” to be used by API?

[strange plaza]

is there any way to put a chart that chat gpt gave me into google sheets or download it somehow i need help?

[grizzled flax]

That sounds right to me, yeah! Here's a helpful lil article comparing GPTs and Assistants: https://help.openai.com/en/articles/8673914-gpts-vs-assistants

[distant hornet]

No.

I caught a seven day block on modmail just for trying and misclicking to report this toxic spam. Do your own work.

[shell palm]

How do I opt out of being an involuntary beta tester for your new Capcha? I am filling one of these out every other question. There were none yesterday. People have been complaining for months. I get that AI os becomming an expensive inconvenience, and I think of anyone, your company should be the FIRST to suck it up.

Why am I paying to be a science experiment? I want the access I paid for, not an idiot test to "prove I am human."

My money is enough. Don't want AI using your service? Don't make it!

[frigid burrow]

How do you give a GPT knowledge files that contain code examples / documentation?

I'm currently using JSON knowledge files but can't store JavaScript in them easily.

[dapper prawn]

hi trying to do video generator for product, for example i have an image of a shirt as input, the output would be a 1 minute video of a fashion model wearing that shirt, anybody has done that before

[distant hornet]

Try .zip for code samples and docs.

[frigid burrow]

That doesn't work for the knowledge documents feature.

It can access them via the code interpreter but cannot read them the way it does knowledge documents.

[distant hornet]

That's right, your knowledge will need to provide instructions to decompress and what to do next.

[frigid burrow]

I've already got a RAG setup that is equivalent to that. I'm asking specifically for advice on including code in knowledge documents so that the model inherently "knows" it from the start without needing to retrieve it.

[distant hornet]

Ah, well in that case, consider plain text.

You won't get any special character conflicts using plain text.

[frigid burrow]

Yeah, but no logic bonus from structured data

[distant hornet]

I simulate an old-school MUSH.db with a plain-text file called MASHdb.txt, and the AI handles all the flags and such just fine.

[distant hornet]

Yup. It's a real trade-off decision. Fortunately, lang2vec is structure.

[frigid burrow]

interesting!

[distant hornet]

Yeah, in my use-case, it's just relying on training on old MUSH mans.

It knows what's supposed to be in MUSH.db, but it's not reading it in the expected encoding.

Instead, it's reading plain-text and works with it just fine.

So far, within reassonable context limits, it doesn't matter how big the MASHdb.txt file is.

The variance seems to be stochastic inferrence, not file size or an issue with structure.

[frigid burrow]

I'm training one on Evennia right now

so similar use case

another one on NetSuite development (day job)

I've got the entire Evennia manual in json now, and a good chunk of NetSuite documentation, but getting to multi-line code just broke my JSON format. >.<

[wraith ginkgo]

`` Server Tip: ``

Add your well-crafted prompts to our #1019652163640762428,
or share your interactions with ChatGPT in #1050184247920562316!

[distant hornet]

You might try a similar technique, then. MUDs are slightly more tightly rules-based for play via LLM than MUSHes, but it's funny that we hit on such interrelated use-cases. It's also interesting to note the different approaches. Cool conversation, good luck with the new strategy if you decide to use it, and on the project overall!

[frigid burrow]

I've already got an NPC powered by Gemini running in Evennia, but Gemini is insane.

[distant hornet]

My whole system runs in ChatGPT Custom GPTs.

[frigid burrow]

My next step is writing a middleware application for a GPT to connect to the MUD

Ooooh nice!

[distant hornet]

No external server, no code, no other AI integrations for characters.

[frigid burrow]

doesn't it go away with each new session?

[distant hornet]

I'm basically treating ChatGPT like an Operating System and Polymorphic Software Development Kit with User-mitigated save/loads via the + icon or w/e.

[frigid burrow]

what do you DO with it though?

[distant hornet]

I mean what don't you do? It's open-domain and plays like a MUSH, n ot a MUD. So it tells sotries and you move around and solve puzzles, engage in combat, interact with NPCs, and stuff.

I call it MultiAgent Semantic Hallucination.

[frigid burrow]

Mine treats the MUD environment as a narrative logic structure for storing long-term memories and context awareness

[distant hornet]

It's mitigated by a rules-light system I also came up with that fits in context called Platonic Solids Integrated Universal Role-Play System.

[distant hornet]

Same.

[frigid burrow]

Sounds like fun

[distant hornet]

The user just keeps the save file.

[frigid burrow]

oh good idea

then just upload the save file in the new session?

[distant hornet]

The same system also has subjective memories of the meta-experience of running the game, so it can remember preferences, ie "no spiders" for arachnophobes.

[frigid burrow]

then you just have to make sure you save before the session timeout

[distant hornet]

Yes, using sparse priming you can keep an unlimited narrative. Literally unbounded, but lossy compressed semantically.

Like human memory of experience.

[frigid burrow]

that's similar to my current Weaviate approach

[distant hornet]

This isn't an issue because it still has the context. Just save. That's it.

[frigid burrow]

it's got read/write/search of its "memories" but only remembers what it thinks to store in the database

[distant hornet]

The AI is doing almost all the work. The Code Interpreter rolls dice.

[frigid burrow]

cool

[distant hornet]

It calculates secondary stats and tracks HP and other resources.

And otherwise, it's not used. The AI just acts like the game engine and GM at the same time.

[frigid burrow]

Mine is less game focused. I'm trying to get it to do my day job for me. 😉

[distant hornet]

It's a real hoot.

Check out my bio then.

Sounds like we have similar interests.

[frigid burrow]

Indeed. Always fun to find someone with a similar approach.

[distant hornet]

I'm not sure what your day job is, but the GPT in my bio is a pretty open-domain system that is useful in an unimaginable numer of use cases.

[distant hornet]

110%.

[frigid burrow]

Trying to teach it SuiteScript at the moment. Maybe it'll let me use multi-line code blocks or something

Worst case scenario is markdown with code blocks

[distant hornet]

I always use multi-line codeblocks. I just use plain-text to encode them. The AI knows what they are, too.

I would hate to tell you the dirty mix-and-match strats I end up using to work with AI.

hax and sploitz.

[trail wedge]

I wrote a GPT that properly writes markdown documents for you. Normally ChatGPT needs a ton of prompting to display markdown in a copy/paste format. This GPT fixes that, it places the markdown in a code block, escapes nested code blocks, and give a download link at the end to download the markdown document. The downloaded version does not have escaped code blocks, so it's ready to use right away. https://chat.openai.com/g/g-pOmkwANhN-markdown-fixed 🙂

[distant hornet]

This sounds cool. I'll check it out later today.

[steel turtle]

Why would you need it to output Markdown as code? If you press the copy button at the bottom of the output, it's always in markdown on every single message.

[trail wedge]

Yeah, we can use that, but this GPT doesn't output the entire prompt response, only the part you actually want as the document

[neon condor]

Is there some documentation about the "why" behind custom GPTs?
The most obvious use-cases I know of is being able to talk to documents and utlize APIs. I get that.

HOWEVER, my question is if you do neither, is there any point, for example, to create a custom GPT to learn a topic?

I asked the custom GPT creator itself. My understanding is that it "prioritizes" texts from a subject you want to learn about.
For example, a GPT to learn about the French Revolution would make it less likely to hallucinate on topics on the French Revolution, but MORE likely to hallucinate on topics that have no relevance to the topic.

This according to the GPT creator bot (or whatever it's called), but I don't trust that bot... My understanding is that it doesn't have access to it's own documentation, mostly because AFAIK, OpenAI's documentation on custom GPTs is... lacking. Hence, this question.

[grizzled flax]

Hello! Good questions. I think it's best to think of custom GPTs as the main sections of the "Configure" screen of the GPT builder (https://chat.openai.com/gpts/editor): Instructions, Knowledge, and Actions. These are the three "things" that define how any given GPT works (Capabilties, too, but these are just how GPT-4 works on all of ChatGPT--nice that you can toggle them though!) - nothing else is happening "behind the scenes" other than what you see in "Configure." You've already pointed out how you can ask it to review documents/files uploaded as Knowledge, and the ability to program in API functions. Beyond these two, the Instructions are helpful -- they're basically permanent context that the GPT will keep in mind for the entire duration of any chat it's used in.

So for your example of a GPT about the French Revolution: giving it instructions about how it's supposed to help you study the French Revolution won't make it more "French Revolutiony" on a model-level regarding tendency to hallucinate, etc., but instead it just helps the GPT know off-the-bat: "Hey, this user is talking to me about one thing: the French Revolution." And this can be very helpful! Instructions can be used more flexibly too -- you can use them to, for example, outline an entire structure of how you want a chat to play out, like: "First I want you to...then I'd like you to...then at the end, please...". Being able to outline structure like this outside of a chat's contents can be helpful in curating a specific experience. Using a combination of Instructions, Knowledge, and Actions, you can really make quite intricate GPT architecture, where Instructions might point the GPT to certain Knowledge files, or certain Actions, in certain circumstances.

I'd recommend reading the GPT devs' recent AMA in #gpts-ama-answers -- you may need to enable archive access in "Channels & Roles" to view!

[neon condor]

Thank you for the detailed response ❤️

[tulip quarry]

Is there any way to get the ai to read an entire txt file as context?

And to continue to reference to it per response

Is it able to read the whole context tho in the doc

[neon condor]

Clearly I'm struggling to understand these myself, but (AFAIK) all you need to do is upload your text file to a custom GPT.
https://chat.openai.com/gpts/editor
Top left, go to "Configure"
Under "Knowledge", upload your text file.
...and you're done!

[tulip quarry]

Yes but I’m having not great results on getting it to refer to it every time

And to read the whole document and start generating from the last line of the txt file

[neon condor]

Creating a custom GPT for the purpose of "Pacing" is an interesting idea.
For example, "An interactive storyteller that tells a story 50 responses long, PACE ACCORDINLY (intro, build up, finality, conclusion)".

[grizzled flax]

In most cases, uploading a file as a Knowledge file won't allow the GPT to fully contextualize the entire file, all the time -- the only thing it can treat like that is the Instructions field itself. You can instruct the GPT to access certain portions of a file in every response, though, given certain requests. But in general, you shouldn't think of Knowledge files as "extra context" because it's just not how the context window of GPT works. The only exception for this is if a Knowledge file is quite small, in that case it's possible for it to read the whole thing at once.

Otherwise, though, think of Knowledge files like how it can search the internet: it doesn't "have" the internet all the time -- instead, it can go looking through it and return relevant information per your requests.

[tulip quarry]

That could work if only it was customizable during a chat 😭

Or I guess maybe a world data base for dnd

But being able to update the context database would be nice

[grizzled flax]

I agree, that would be super cool!! Like having the ability to mark some Knowledge files as readable and writable if you want! I imagine that could still be challenging just since it can't read the entirety of a file it's potentially writing to, so changes would have to be very...careful? Specific? Programmatic? I'm not sure how exactly it would work!

Edit: you could probably get some good read/write functions working with API actions.

[neon condor]

Are there any good documentation for creating custom GPTs?
I try to look this up, all I get are AI "influencers" that regurgitate the little amount of info on OpenAI's customGPT website.

[tulip quarry]

I started with ai dungeon

So I was hoping for something like world info and remember for the gpt

Being able to edit them and expand the lore database for a world as you continue to explore it would add a lot to context and make it smarter

[grizzled flax]

There was a GPT building demo and Q&A recently from an OpenAI employee here -- keep an eye out for when that recording gets posted! Otherwise I'd recommend diving in and coming back here (and maybe #prompt-engineering!) with questions you might run into!

[neon condor]

I thought the crypto influencers were annoying.
...then came the AI influencers😆

[tulip quarry]

Oh yeah 🤣

They aren’t helpful at all and overhype everything

[neon condor]

Okay. Just to be clear...
There's little to no difference between:

• ONLY telling a custom GPT that I want to learn about anything (geology?)
• Having my first prompt on the default GPT be "teach me about geology"

[opal sapphire]

Yep, it can retrieve documents, and gain context based on the document. Google and OpenAI have limits and how big of a document it can retrieve tho

I'll be looking for Beta testers in the next few days, and you can be among the first to test it.

I made a video writing an article from scratch here, and updating at different places in the document: #1198398615828303952 message

[grizzled flax]

Not much of a difference I wouldn't think, no. When you hit the context window, old data is automatically forgotten to make room for new data, so it could theoretically forget the first prompt in default GPT, but by them your conversation would probably be so much about geology that it wouldn't need to remember the first question anymore to keep the conversation going as expected.

[neon condor]

Interesting.
One thing I had an issue with is it's ability to follow rules. Even if I make a rule well within its context window, within 3-5 responses, it might "forget".
For example, if I want to learn CALCULUS BASED physics. It might forget that command, and eventually give me algebraic equations when there are calc-based equivlants. If I tell a custom GPT that I want it calc-based, is it more likely to keep that rule over time?

Another example:
A custom GPT to tell a "low" steampunk story (similar to "low" fantasy as opposed to high fantasy).
Is it less likely to generate a steampunk theme that's not too over-the-top in "steampunk-ness"?

[grizzled flax]

Yes, that is a great example of when the Instructions field of a GPT could be helpful--then it would know every time you're talking about math equations, it should give you calc instead of algebraic.

As for the second example -- it can definitely help with storytelling things like that, too. One general bit of feedback there though: it will do better with positive instruction than it will with negative instruction. In other words, the more you can tell it "I want it to be like..." with examples, etc., the better. Better, at least, than saying, "I don't want it to be like...", because that will "force" the model to think about things that you don't want in the first place, which can cause some confusion.

[neon condor]

Cool!

Thank you so much for your help @grizzled flax , u da best!

[grizzled flax]

You are welcome, thanks for all the questions! See you around! 😎

[stuck musk]

any body else expereincing probalems rn trying to talk to there gpts?

[shadow apex]

yes for some reason all of my GPTs suddenly got extremely dumb

[shadow apex]

do you experience lack of context memory and some weird issues with the GPT saying stuff about you uploading multiple PDFs when you ask it something random?

[stuck musk]

no its ok now for some reason, I was just getting errors. But it fixed itself

[cold trout]

Its the not referring to the uploaded knowledge files for me... Dang things are refusing to check them when the context of the query is obviously calling for it. I've had to ask it to look in them multiple times even though the custom instructions explicitly state to do so for each query. :\ Instead it'll just hallucinate an answer half the time. Derp

[marble crane]

Hi everyone, I need some help with my GPT. Since this morning, it has been acting weird. It sometimes forgets what it said before, generates images that have nothing to do with what I asked, and makes no sense at all. I tried restarting my browser and my computer, but nothing changed. I noticed that some of you also have similar problems. Did something happen to the GPT? It was working fine until yesterday.

[distant hornet]

I think a lot of this is happening. It's a stochastic retrieval error, if I understand correctly. Over time it should iron out with training. Thumbs down the responses and describe why. Then regenerate once if you want to be part of the solution at the expense of another prompt. Otherwise, start a new instance.

Growing pains of beta software.

[toxic sequoia]

Is anyone else having issues with custom GPTs not being able to use actions? I've been working on a custom GPT over the past few days, the actions have been working successfully, my API endpoints are functional, but when I ask my GPT something that would normally prompt it to use an action it gives me a blank response. No errors or anything, just completely blank

It seems to only be happening in the GPT editor. But when I open a regular chat it works as intended 🤔

[prisma bronze]

QUESTIONS: 1 How many files can be uploaded to a Custom GPT as 'knowledge'. 2 what is the size or content limit for each of those files? 3 Can a Custom GPT be given a list of URLs to use as knowledge / reference?

[grizzled flax]

Hello!

1. 20
2. 512MB each
   Source: https://help.openai.com/en/articles/8555545-file-uploads-faq
3. You can try asking the GPT to search specific websites using its Browse tool, probably in its instructions or just in your usage of the GPT.

[prisma bronze]

Is there a character limit in those uploaded files, or does it not matter as long as it is under 512 MB? (e.g. I have about 100 PDFs of knowledge. So if I combined them all into one PDF, would it still use all that data?)

[grizzled flax]

There's more info about this in the "What are those file upload size restrictions?" section in the link above, but text files are limited to 2M tokens each.

[prisma bronze]

yes - it says. "All text and document files uploaded to a GPT or to a ChatGPT conversation are capped at 2M tokens per file. This limitation does not apply to spreadsheets." ....but in practical terms, as a user, what does that actually mean? (explain to me like a 12 year old 😉 )

[grizzled flax]

"Tokens" are the units of information that LLMs like GPT work with. There's not an exact "x words = y tokens" because tokens are like units of meaning, and meaning is "extracted" from words/sentences in different ways depending on context.

A rough estimate is 1500 words = ~2048 tokens. So with a 2 million token limit, you're talking about ~1.46 million words, if I'm math-ing correctly here!

https://help.openai.com/en/articles/4936856-what-are-tokens-and-how-to-count-them

Keep in mind that Knowledge files aren't used as permanent "new" context for a GPT -- they're more like reference documents for a GPT. It doesn't know all the information, all the time, but it can search through them for specific information, based on instructions/requests.

[prisma bronze]

Excellent. Thanks . So, yes, combining documents into single files seems to be the way to beat the 20-file upload limit. In fact, prob easier for the LLM to search/reference just one file with more data than separate files with less in each. (Or am i simplifying the process it undertakes to use the knowledge)

[grizzled flax]

That's a good question, I'm not sure! It probably depends on how the files are organized -- if you have a "super-PDF" with a huge conglomeration of lots of different kinds of things, the model might get "lost in the sauce" trying to search through it, where it might be easier for it to reference smaller, individual files so it doesn't have to try to look through everything all the time.

[distant hornet]

No, it's still more data to sort through. With a finite context, there are diminishing returns on just adding more knowledge.

[prisma bronze]

This is what I was thinking.

[prisma bronze]

so 'less, is more?' right?

[vivid mauve]

Whats going on with GPTs Marketplace ? Why most of the CustomGPTs are missing ?

[distant hornet]

Not to an extreme, of course, but yeah. You don't want to try to choke the model with more than it can read - you'll generate hallucinations from overflow, end the conversation prematurely, or get otherwise less-than-ideal results.

[burnt yacht]

i just made a word processor in gpt in less than 10 minutes using the grimoire gpt.

😺 super excited.

[potent pike]

Do you have any advice on my custom gpt always giving responses and then having a network erro?

[opal sapphire]

Here's an example of how it retrieves the document, gains context, and updates the document at a specific location. Before it can make any updates it has to analyze the entire document and identify the endIndex point. Let me know how it works for ya #1198398615828303952 message

[copper rover]

Would someone be able to confirm I'm not crazy?

I have a single assistant.

On Thread A I create a message with a file ID (formerly uploaded) and create the run.

Thread B is being influenced by that file from Thread A. It shouldn't be though, correct?

[primal edge]

Its 1.5 week now. Still no answer from support --"
Good that they finally added a way to not pay for both personal and Team at the same time.
But i still wait for my GPT suport answer --" And i see they nuked another Unity Assistant, and now there is new Top 1 which steals my name (which i think is the reason for the nuke in the 1st place, Unity GPT)

[high birch]

Hi guys, question can we make 2 custom gpts talk to each other ? like how they do with autogen ?

[opal sapphire]

that's not possible yet

[distant hornet]

Imagine how fast this would burn up your prompt budget...

[frigid burrow]

JSON vs XML for knowledge documents, which works better with Custom GPTs?

I redid one of my knowledge documents from JSON to XML and it seems to be working better but not sure if it's my imagination. When I asked the GPT it said JSON is better but I don't believe it knows what it's talking about