community-help

[trail cairn]

do you have chatgpt plus

[cerulean tree]

No, oh… I got it - it doesn’t work without ChatGPT plus, and no - I don’t have it

[brisk haven]

My ChatGPT account was banned a few months ago, but money was debited again for ChatGPT Plus and I get absolutely no help from support

[rigid musk]

Only help.openai.com can help you with payment

[brisk haven]

Okay I will try again

[remote linden]

Hi, chatgpt is flat out refusing to help me analyze my bank statements this morning and it's super frustrating because I was able to do this before. Why are there content restrictions in place to analyze my own data?

[trail cairn]

ask why it has been instructed to do that with your image

i saw someone say somewhere that it got more restrictions as people used pics of their medicine prescriptions, medical reports, etc. which they should never do

[remote linden]

What's the use of this Ai when it can help with difficult questions

[amber breach]

This is something I have struggled with, for me it has been a lot of trial and error. One thing I found that seems to work a lot better is using the RAG documentation. When I am creating a Custom GPT I keep the general instruction to bare bones and I include in that instruction information on uploaded documents. I create a series of word docs outlining step by step instructions to follow, and the word docs use file names that indicate the instructions that are contained in the document. If you feed the GPT small focused information in a series of documents I have been able to achieve some very complex sets of instructions. In some situations I have almost created a choose your own adventure style series of docs. In the docs I will even instruct the GPT to 'If in this situation read the doc some_doc_name'.

[vivid quest]

Any ideas why?

[trail cairn]

placeholder?

[glad scaffold]

umm any1 knows why the captchas are literally impossible?

the openai human verification experience

[timber shadow]

Do you have a VPN installed? What browser(s) is this happening in?

[vivid quest]

have no idea why is that

[runic delta]

Hey all. I’m looking for some help here.

I use GPT-4 for advanced SQL and Python assistance.
A couple of months ago, GPT-4 would return complete solutions that were usually right on the first go.

Since the UI change, GPT-4 is putting comments everywhere telling me to finish what I asked it to do. Its also not even close do the quality of the previous work product that GPT-4 would put out.

I am burning all of my credits and reaching my allotted query quota just trying to get one working query out of the system.

What can I do to make this thing work better? It seems like the GPT-4 model is giving worse answers that GPT-3 in many aspects.

Do I have to just use the API? Has the API also degraded?

For example, I’ll give GPT-4 a prompt asking it to return a complete query with the same input data I would provide before, and I will be very explicit telling it I need a complete query that is ready to execute, and I’ll tell it to not put comments in the query, and it will still return something like

“- - put the rest of the query here”

[trail cairn]

its a known issue

[amber breach]

I just usually instruct the GPT to NOT use comments and to produce complete code solutions and it usually acknowledges that I have requested that and it carries on. Sometimes depending on the length of the chat I may need to remind it.

[runic delta]

Thank you both. Is there a public facing known-issues tracker so I can monitor this? Or if you know where they may be at with a fix?

[amber breach]

@trail cairn would have to answer that one, I haven't bothered to track it or look it up, I just keep reminding the GPT to keep doing what I've instructed it to do. Actually depending on how you are interacting with the GPT the one thing I did was added to my system prompt that 'I knew nothing about Python, I was a complete beginner learning to code' This actually improved the responses from the GPT, it get's a little wordy in the responses because it thinks it need to explain everything, but the code completion was good. Might want to try that.

[runic delta]

Thanks @amber breach I’ll give that a try

[cursive yoke]

gpt-3.5-turbo-instruct appears materially worse than text-davinci-3 for article summarization, for instance. petition to legacy text-davinci-3 instead of retiring..

[rugged wing]

Hoi, my gpt chat suddenly “stopped working”, as in it sends an error every time I try to send a message. I’ve tried multiple times with different messages at different times but the ai refuses to respond - is it broken?

Please help, I've had this issue for over a week

[cold folio]

Have you tried opening a new chat? does it work ? @rugged wing

[runic delta]

Ah I had this too! It lasted for 3 weeks.

A workaround that helped for me was to close the app or refresh the browser and prompt again. It would let you get one prompt in before it started failing again. Kept having to close the app or refresh the browser for 3 weeks lol. I hope this helps.

Support got back to me 6 weeks later.

You’ll have to do it after every prompt, if you’re experiencing the same thing I did.

[rustic marlin]

use a different name then..

it is likely that this name could be already taken

[rugged wing]

Hello everyone, I tried all sorts of different articles, but I didn't find an official answer provided by OpenAI. ChatGPT claims, that it can't access previous conversations and therefore can't give answers depending on the user. I know that there's a feature called Custom Instructions, which allows the model to generate answers to fit the user. Long story short, my question is, if ChatGPT adjusts itself to the user automatically or only when using Custom Instructions.
Thank you and best regards

[rustic marlin]

currently it does not do that

every chat is individual and it can't access data from other chats

if you specify information on your custom instructions, that information will be on the context of the chat

but a custom instruction such as "use the data from other chats" isn't possible, doing that will simply not work, or worse, cause hallucinations

[rugged wing]

All right, thank you for the quick reply. It provided all the information I needed :).

[glossy gazelle]

Hi, im pretty new and everytime i try to run code or use the playground this shows, any idea why?

[weak parcel]

Hi, sorry for the quesiton but i dont understand, i m plus user however when i try to link my api to the app to use it native on my mac it gives me this "You do not have access to the model you have selected. To access GPT4 models your account needs to have a valid payment method linked, and you need to use at least $1 with GPT3.5. Check this page for more information: https://help.openai.com/en/articles/7102672-how-can-i-access-gpt-4."

i have full criedt 20 everymonth non used -.-

in the web version everything works

[rigid musk]

Api and chatgpt plus isn't the same ans have different payment

You need to buy credit for the api

[weak parcel]

😮

[weak parcel]

okies thank you i look into that, do i need then the plus if i get the api one ?

[rigid musk]

The plus only give you access to all feature on chat.openai.com

Both separate products

[copper trout]

Trying to buy Chatgpt plus with my debit card but keeps getting declined yes confirmed with bank that my cards fine and i got funds

[rustic marlin]

only credit card

[copper trout]

Did they change it recently?

[rustic marlin]

nope

you can only use credit card

[copper trout]

its for chatgpt plus

had no problem last 4 months

using same debit card

[copper trout]

why did my debit card work then 🤣

[rigid musk]

Ask you bank, cause it was only credit card since the beginning :/

[rustic marlin]

your bank did something they probably shouldn't have ¯_(ツ)_/¯

it always have been strictly credit card only to pay OAI's services, so... no idea why it worked

[copper trout]

gpt plus

[rustic marlin]

it could jsut have been some sort of fine print auto enable feature on the bank's side..

but the fact is that credit cards should not work on OpenAI

[copper trout]

Goofy

[rugged wing]

Thank you, I’ll see if this works!

[rugged wing]

Well yes, but the issue is specifically with the same chat

[timber kernel]

why was my subscription cancel

reached out about some performance slips and had my subscription ripped away after paying for over a year now?

but I'm completely dumbfounded

and if I paid for my subscription for this month why was my account my Plus account canceled in the first place I mean I could do that all from the widget but on top of that it's still December

...?

can you guys know the transfer takes at least a few days sometimes more I'm usually on the lucky side of that but regardless here we are December

[copper trout]

they suck

[timber shadow]

We're all just fellow community members here. Your next best step would probably be to submit a ticket for OpenAI support at help.openai.com. The chat button in the bottom right will let you start a conversation with a bot, then you'll eventually have the option to submit a support ticket, and they'll reach back out to you after that.

[copper trout]

make sure u hammer them with tickets or else they wont respond

[timber kernel]

oh... I just hoping to get into tho ig

"tickets"

[copper trout]

yes

[timber kernel]

...

sos lol

🙏 BTW

[copper trout]

i always submit 2/3 when im trying get a problem fixed cuz i usually find they ignore about 75% of them

[timber kernel]

yeah I'm seeing that too but I just don't know what you mean by tickets like how do I create one even you just mean the help widget

[copper trout]

1 ticket still not read 4 months ago

[timber kernel]

🤦🏻

[copper trout]

support ticket

yes

[tired monolith]

Could you help me? How to use free chatgpt 4 method

[green geode]

Hi there, I'm GW. I'm new to this community And I'm curious why it seems discord's reactions are disabled on the server. Is there a reason for that? Could the problem be just on my device?

also upon further investigation it seems that it's only on certain channels that there's this limitation, like #chatgpt-discussions

[solar bramble]

Where do I go to ask questions about errors/problems i am experiencing with ChatGPT 4.0?

[solar bramble]

nm found it: https://discord.com/channels/974519864045756446/1070006915414900886

[rigid musk]

You cant

You can use bing that use gpt 4

[hardy geyser]

By any chance do you know why chatgpt refuses to make me upload any files ? When i upload them they become RED and after getting to somewhat 25% completion rate, i get an error saying unable to upload files.

I've tried 4 different browsers.
I've got my self premium antivirus service to ensure that i don't have malware.
I've respecting all limitations on formats and file sizes.
I've tried different files.

On my phone it works perfectly, but on pc i've been having this issue for now more than a month and i'm starting to give up on it

[azure stump]

Sharing external links on the server lead to a ban?

[umbral storm]

Hey guys, is it still possible to generate pictures via this discord?

[rigid musk]

No only with chatgpt with premium, or bing

[rigid musk]

It can, but most of time it is only mute

[rigid musk]

Might want to try and post in #1070006915414900886

[safe ledge]

Hi, I have a problem. So far I have been logging into my OpenAI account using my Google account, but I would like to change this to a password login. Does anyone perhaps know how I can do this?

[rigid musk]

If you used with auto login of Google only way to change password it is with Google but it will affect all your auto login

You can try to contact the support to change the method of login

Help.openai.com

[safe ledge]

There is no support, it's only bot redirecting to articles

[rigid musk]

When contacting the bot at the end it will create ticket and have support by email

[safe ledge]

I tried but it's not

[rigid musk]

Can't help you more than that :/

They had [e-mail] but I don't think they answer anymore direct email

[safe ledge]

Even @burnt galleon is only a bot!

I need a human!

[rigid musk]

Mdomail isn't a bot . You have answers by a real moderator from here. Moderator are only volunteer and doesn't work for openai

[foggy folio]

Hey, I need to make a clean logo in this style, this is the best I made the ai do. Can anybody help me?

[rustic marlin]

the image seems nice, consider using the wide format instead, check #images-discussions channel, people there will be able to help you more

[pale helm]

I am a GPT plus user from very months and now the GPT 4 is just rubbish from few weeks as it doesnt browse the web to get me the content i want and its response are now too slower and too laggy. The GPT plugin to access links are also not working and the GPT 4 is now rubbish. Why i am paying $20 for these things which i can get in free GPT. I have resubscribed for GPT just some days ago and i am now thinking of cancelling my subscription and get a refund. As the GPT 4 is not that much beneficial as before. Now whenever i chat with GPT 4 and ask him to provide me some data using web browsing. Its says it has limitation and it cannot access web. So why i am paying $20 if you are adding limitation to the GPT.

[fluid matrix]

i have been stuck on a "HUMAN VERIFICATION TEST" for 2 hours straight. Signed up - PAID and i cant even type a letter without getting this screen. this is literally the worst. i cancelled my sub 4 times previously because of this...is there a fix for this?

[rocky rover]

contact support directly

[fluid matrix]

support ? you mean the automated serrvice which does absolutely F all to help me at all?

[fluid matrix]

they put bots instead of actual humans. to HELP. they took my money instantly, yet i cant even get human customer support/help.

[rocky rover]

just calm down, it can be either on their side or it could be on your side the problem.

tried different browsers? no extensons? safe-mode?

[fluid matrix]

i am using chrome. no adblockers whatsoever. its been months. i have NEVER been able to use or generate a single thing with this...just human verifiation tests nonstop

[rocky rover]

are you behind a vpn?

[fluid matrix]

it is completely turned off.

[rocky rover]

tried other browsers?

[fluid matrix]

omg.

lol yes. i have three browsers.

all these questions i have told the "automated" system. and then i got booted for simply needing help with this.

[rocky rover]

well, it's just standard stuff

[fluid matrix]

i just gave up lol. makes no sense to even ask

[fluid matrix]

being stuck in a verification loop for months is standard? ok .

[rocky rover]

dunno, never had a problem and I dunno your setup

[fluid matrix]

where does one contact support

directly

[rocky rover]

I'd say the help site, but you already went there, so no clue to be honest

you can't log in with the mobile app either?

[fluid matrix]

i am logged in via phone. i just find this extremely annoying that the folsk behidn this software do nothing to help

[rocky rover]

if you are logged, I don't think the problem is on eopenai's side

[fluid matrix]

i installed chrome and safari fresh. i dont use adblockers period. i dont tamper with my DNS and i use the internet pretty normal to be honest.

[rocky rover]

the only thing i see is the chat to support and it will take abut a week for OpenAI to answer

[rigid musk]

Help.openai.com

[limpid wyvern]

Hello, quick question, is there a list of plugins considered safe to use? how do i check if certain plugins are considered trustworthy?

[brisk haven]

I tried to contact the support 12 times

no answer only default text

I try this since 31.08

[fluid matrix]

just realized there is a usage cap for this. WOW. lol

[regal lynx]

Yeah, it's mentioned on the help website: https://help.openai.com/en/articles/7127997-how-can-i-use-gpt-4-in-chatgpt

[vestal peak]

Hey y’all working on getting analyzing PDF’s enabled in my Bubble app. Got the vision api set up for pictures and trying to get PDF’s working now through getting an upload file ID back from OpenAI and then using that ID in several variations of calls from OpenAI including the assistant, chat/completions etc. Only the completions call actually went through without an error, but it returned saying it couldn’t parse PDF’s yet. So I’m not sure what’s actually possible. Lots of others trying on the forum but having seemingly similar issues. Plz help

[brisk haven]

Support says my account isnt banned, but when I try to log in in redirects me back to the login page I already tried another browser, anybody has an idea?

[rigid musk]

Do you have a vpn

[brisk haven]

no

[rigid musk]

Does it works on cellphone

[brisk haven]

no

[mental umbra]

is the site down im not able to login i keep getting a verify human loop

[rigid musk]

Status.openai.com

[rigid musk]

What is your country ?

[mental umbra]

k

[brisk haven]

germany

[rigid musk]

If it doesn't work on cellphone or pc, you have no choice to contact the support

[brisk haven]

they always send the same

[scarlet pecan]

Hello everybody! I'm making a custom GPT, and I'd like to have a few pre-made images display inline, rather than DALL-E generated ones. I've uploaded the images to the "Knowledge" section of the configuration tab, but it is unclear whether or not I can make my GPT app access and show that image. It's a bit like a visual novel and I'd like to show a graphic which represents the speaker. Does anyone here have an thoughts on how I could do this without having to write an external application with HTMl/CSS, etc? I'd like the app to just live on the GPT app store and not bother with external servers or other frameworks or languages.

[harsh iris]

Try to copy and paste important data to a text file. If you use the paid version, upload into a "Custom GPT" as your personal data set and such so that you can retain your information and query it too, if needed. It also depends on what you are trying to do. Typical prompts, formatting, and such can be created as an instruction set, this way you lessen your memory limit and not run beyond it's conversation memory and waste less time with constant new conversations. This will likely improve in the future from the rumors and speculation I have seen or read. Also, creating chapters or sections of more focused data is much better in overall quality instead of trying to create everything in one conversation, especially if you are using the free version. If you do not define your questions and try to remain somewhat consistent, Chat GPT will hallucinate somewhat and lose focus.

[zealous apex]

hi

[tribal helm]

Why is Chatgpt keep verifying me as human everytime I ask a question? I can't use Chatgpt this way... It's been like that all night

That is like a few minutes drag everytime I needed to ask a question?

[fading hull]

Getting the error "User quota exceeded" whenever I try to upload an image to gpt 4's chat. When I looked online, I only saw API related discussions, but this is happening on the website itself. How can I fix it?

[zealous apex]

what?

[tribal helm]

Looks like my best support is still chatgpt. VPN and having it opened too many times (eg. Binge watch YouTube by sending transcripts one after another to Chatgpt)

[rigid musk]

What browser

[stuck widget]

Hi all, I need help to subscribe to ChatGPT Plus. I got the invite link but it is disabled. I need it urgently for work purposes for the upcoming year.. cant seem to contact the team. Any ideas? Much appreciated.

[rigid musk]

you dont need link to sub

you can just go on chat.openai.com

[mossy mesa]

My custom GPTs haven't been searchable via tools like GPT Search for 45 days. They do seem to be appearing on some of the third party list sites, but they're invisible to GPT Search, etc, within ChatGPT.

I've brought this issue up with support, more than once, but they universally mis-respond with advice to clear my browser cache.

GPT Search uses OpenAI and Microsoft's networks. I did follow all the tier-one browser support suggestions, of course, but I'd done all that prior to reporting the issue. It's clearly a problem with the OpenAI/Microsoft systems.

To me, based on extensive testing with the Bing Web Search tool, it just appears like Bing is an incredibly, mind-bogglingly lazy tool.

Any other insights? The GPTs in question are branded "Lexideck {name}", so searching for Lexideck should reveal them. Searching Google for "Lexideck GPT" reveals many of them, and clicking the link for more content at the bottom reveals almost all of them. So they're visible online - just not to the services associated with OpenAI's partnership with Microsoft. 😦

[hollow rose]

My GPT has disappeared. Can anyone please help me restore it?

When I go to the GPT's URL now I get an error message saying "GPT inaccessible or not found". It doesn't appear in Explore.

What can I do?

[mossy mesa]

@hollow rose this happens sometimes. Give it time, they may reappear from backup.

To guard against this, I copy all GPTs to local text files and back them up on Google Drive.

This happened to me once, resetting over 30 GPTs. It didn't reset for me in a reasonable time, so I rebuilt them all from backups.

At the time I had ~30 GPTs to rebuild.

[hollow rose]

I really hope it reappears. I have the knowledge files but not the latest version of the instructions.

Will start archiving better. Thanks

[hollow rose]

How long should I give them, you think? Before I just start over?

[mossy mesa]

I'd get right to work, tbh. If you have an older version of the GPT, start reconstructing your edits. Maybe take this opportunity to restructure, too, if needed. That's what I neded up doing.

If there was a data center error it could take any amount of time to correct - or might not ever be corrected. As my own GPT library expands and balloons, I dread this more than I can say, even with all the backups. That's because I've also encountered a daily public GPT publishing limit of about 32, give or take.

I'm just not sure OpenAI were #ready.

[hollow rose]

I've published 4-5 times today before it happened. Could it be some kind of punishment for that?

[mossy mesa]

No. But I'd caution you to take your time. It took me a while to notice, but the system auto-saves changes as they are made. You'll notice the active pane of the GPT creator goes grey while this happens.

Let the changes save before continuing or you can hose the system, too.

[dusty holly]

when the server is running hard it all lags out. often GPTs will not show up etc. First thing is not to panic and start clicking things crazy. Hasteful decisions will be costly.

[mossy mesa]

This -^

To prevent it from happening, you really must not work rushed. Take your time, be mindful of the GPT greyed-out state, and resume work when it finishes saving.

Make sure you don't make a bunch of changes in a row really quickly and try manually saving. That's where you're most likely to see "punishment" arise, as your GPT will be in an incomplete state as it's saved - potentially erasing data.

[hollow rose]

OK, I'll start working on the new build in the background and will give it 1 day for the recovery. Will be careful and slow with new updates, always after autosave

[mossy mesa]

As much as it sucks, that's the best plan. Remember to back up the GPT text itself, and the convo starters, too!

[dusty holly]

same thing deleting conversations; it lags often and ive heard people cry spamming delete then 10 conversations vanish 20 seconds later.

[mossy mesa]

Yup. I think the whole system's IT needs are all parsed for a natural language router via GPT 2. /sarcasm

[dusty holly]

Mutliheaded transformer stacks are just another step; on par with Frankenstein. Another few years and its all going to be far more sophisticated. Enjoy the ride though. Chnages like this globally are rare.

[main helm]

gpt4 keeps crashing on me. what do I do?

[dark locust]

Hello! I've seen a lot of tutorials about getting consistent characters in AI, but does anyone know of resources for getting consistent environments?

[mossy mesa]

Use the same technique. Create a descriptive but sparse description that produces somewhat repeatable results. Environments are larger than characters, so expect more variation. However, with good priming, you can definitely do this.

The important thing is to be specific but not wordy, and let it improvise some things with the understanding that environments are in a constant state of flux, unless the environment is super-simple like a geometric room.

[copper trout]

Gmail got hacked 😦

[mossy mesa]

Like all of it?

[raven ravine]

I asked him to change the picture and now he is all bugged. I don't want to lose the progress. What can I do?

[fading harness]

I have been recently experiencing extremally slow usage of chatgpt, I have been having to wait over minutes on minutes to get a answers,, but mostly Its been crashing and just loading and loading, is there any way I can fix this?

it loads for ever, and then just crashes.

[sand bronze]

For me, ChatGPT has just had a degraded quality in general at some times. I get the same issue, it's super annoying.

Could someone help me, I cannot make my GPTs public because in the builder profile there is no option for the name.

If anyone responds, please ping me.

[ancient pewter]

help chatgpt got my account flagged it prob said something nsfw

[tame totem]

Is anyone else unable to add a new payment method because of this error? I haven't sent any requests to it, at least not that many.

[rigid musk]

what was the prompt

[ancient pewter]

i forgot

i cleared the chat

but the prompt was nothing related to nsfw

the prompt was about eggplants i think

[rigid musk]

just make sure to follow tos you wont receive that msg

you can always thumb down for feedback

[ancient pewter]

chatgpt did not follow tos

i followed tos

[rigid musk]

if you think thats false positive, just thumb down the answer and review it. it can happen

[ancient pewter]

chatgpt normally says nsfw stuff on accident

[rigid musk]

it doesn't

[ancient pewter]

sometimes it does i used chatgpt for awhile

[rigid musk]

it uis your prompt. and trying to bypass the ToS can lead to a ban. be careful.

[ancient pewter]

chatgpt proberlly thought i was talking about the eggplant emoji but i ment the eggplant vegetable

chatgpt can take your stuff out of context

[azure hare]

chatgpt generated something that got flagged
if your prompt wasnt flagged, it shouldnt be that bad of an issue
but its best you edit it

[fading hull]

happens on chrome, firefox, and safari

[rigid musk]

and on mobile?

[ancient pewter]

my prompt wasn't flagged

[azure hare]

did you edit the prompt

so it doesnt flag

[ancient pewter]

no

i deleted the convo instead

[azure hare]

k

[ancient pewter]

i asked chatgpt something about eggplants it thought i was talking about the emoji but i was talking about the vegetable so it got flagged

[civic bolt]

Can someone help i have to do the captcha after every prompt and its driving me insane

[rocky rover]

if you are using a vpn and sharing an ip adress with others this is common.

[true zinc]

GPTs optimisation for new computer language

Does anyone have experience with trying to optimise the performance of a GPTs with creating scripting in a new computer language not in its normal LLM model please? I have trained a GPTs in a bespoke version of Lua, training it on a language definition models, and a few hundred example working scripts and program files. The created Expert Adviser does what is should, of producing coding following some text prompts. However, there seems to be a big disconnect between the ability of the GPTs EA to develop new code or strategies, as compared against how I am easily able to get GPT4 to produce some quite advanced Python scripting.

I suspect that it comes down to the fact that Python examples are normally incorporated as part of the regular LLM training process, whereas trying to add something new in does not enable the training process to work in such a closely coupled way to the text input. The only analogy I can think of in terms of more regular neural network architecture is that there is somehow a missing hidden layer connecting the GPTs EA and its external database.

Does anyone have any experience in how to optimally connect (in the logic) the GPTs EA and its external material, that goes a little beyond simply functioning as an LLM, that I could talk to please?

[timber shadow]

Hello! One thing to keep in mind with custom GPTs is that their "training" is limited to the information contained in the "Configure" section of the GPT edit page: Instructions + Knowledge files + Actions. Additionally, knowledge files serve as reference documents for a GPT, not as permanent added context. So adding a knowledge file is not exactly like training or fine-tuning an LLM, as you referenced. Instead, knowledge files are searched for information per your requests, similar to how it might search the Internet for information relevant to your requests.

[woven fern]

i can't upload .txt files why?

[jolly patio]

Is it a customGPT?

[hallow goblet]

quick question: to use the api for projects do i have to pay for openai pro? im sending requests and always getting a "used all of quota" error while other people online seem to not be paying for it and getting requests just fine

[simple geode]

Good question. I don't think I have an answer but perhaps a few suggestion, all the coding questions I had with a language not supported by code editor were about overall structure, and a few code simplification (nothing too specific) and I was satisfied with the output so far. It's quite possible that some of them don't work at all or are not developped/matured yet.

1. Add the language standard (Reference manual) to its knowledge base.
   2 Experiment with various CoT and other prompting techniques.
   2a. Try to get your GPTs to generate your code in a programming language supported by the code engine and the ask to convert to lua.
   2b. Try your GPTs to generate pseudo code (e.g. Z notation) then ask for lua code.
   2c. Add your own example of expected results
2. Continuously use the GPT editor and edit the instructions to fine-tune your expectation.
   3a. Add your own example of expected or past successful/corrected lua code to its knowledge base.
   3b. Compares from time to time with gpt classic and the vanilla GPT-4.

[vestal peak]

Anyone get pdf analysis to work with the file upload yet? Please need it

[quasi lily]

I am getting 5 Re:CAPTCHA puzzles for every sentence/entry into GPT-4 on desktop. This seems to be a type of ban on IP level since I created a new pro account and it's the same thing from start. I am incredibly frustrated. This has been like this for 2 months. No reply from OpenAI.

[livid lark]

So I setup a customized GPT for basic life coaching tailored to needs yesterday. It worked pretty well, but like an idiot I tried to enhance it using a prompt I found on Reddit and it hasn't worked properly since. Should I just delete it and start from scratch?

[timber shadow]

Do you have a VPN installed? What browser(s) is it happening in?

[timber shadow]

Dang, sorry you lost the GPT you like! For future reference, you can always save the "Instructions" field in "Configure" in the GPT builder interface. This way, if something like this happens again, you can paste that back in a new GPT's instructions and get back to where you were right away.

[livid lark]

Yeah, I mean I ended up linking the chat from the GPT that I liked and it supposedly recreated the original GPT, but I am skeptical so I just tried modifying it again. Either way, it didn't work out so well and now I'm wondering if I should just delete the GPT and start over, lol

[loud flax]

Anyway to tell when you’re reaching your v4 limit? 😑

Prior to gpt saying “nope”

[timber shadow]

Keep in mind that there's nothing "behind the scenes" that makes a GPT work -- its behavior is completely dictated by the contents of the "Configure" section: Instructions + Knowledge files + Actions. So you can always check Instructions to see if something you've done in "Create" has actually affected the Instructions in the way you expect.

[timber shadow]

There's no UI indicator, so there's really only the option to count your uses. You can pace yourself with one message every 4.5 minutes to stay within the 40/3hrs limit. And keep in mind that custom GPTs have a lower cap of 25/3hrs

[livid lark]

I just feel like the original questions I had from create worked well enough to get me what I needed after a few tries in the playground. Anything I do now just seems to screw up that dynamic, but I was trying too hard to tweak it I guess. I'll take the blame for that one. The output was almost perfect tbh. I just wanted a slight tweak to give it a more interesting persona.

I think I'll just upload a file that is a reference for the personality and keep the prompt to pure instructions on what to generate.

[timber shadow]

GPTs can follow specific instructions in knowledge files, or return specific info from them, but they can't use knowledge files as permanent context for every response/conversation. Because of this, you might try something like copy/pasting whole sections of example personality text and, experiment with a "Create" request like:

"I'm going to share some example text below. Please use it to inform the personality of this GPT to make sure that it mimics the style, tone, and personality of the examples.

Examples:

[Paste examples]"

Then, just be sure to check what it actually extracts as the style and tone in Instructions over in "Configure", and edit it as you might want from there. Kind of spitballing here, I've never tried this exactly myself, but I think it could get you closer to where you want to be! Don't be afraid to edit Instructions manually either -- Create is a nice feature but again, it's "only" writing Instructions for you, and you're free to write your own Instructions completely.

[livid lark]

You mean instructions on the configuration side?

[timber shadow]

Yes, Instructions in "Configure" is what "Create" writes for you based on your answers in "Create" -- you can skip "Create" and write them yourself if you want.

[livid lark]

Yeah, I konw that. Just wanted to clarify what you were talking about(which you did so thanks for that). However, I feel like the instructions that Gizmo generates are too complex. I feel like I could do better using shorter sentences that aren't bunched together in a paragraph. I feel like that's where the disconnect is.

[loud crest]

GPT4-Preview finds “According to your latest update, when was X seen last” paradoxical lmao

**`When referring to the "last time X was seen," there is an inherent ambiguity in the query which could indeed create a paradox within the constraints of our established directives.
The paradox arises from the fact that while I am equipped to access historical data up until 2023, the concept of "last time" by its very nature refers to the most recent occurrence—up to and including the present moment, which in this case is beyond my purview due to the "Real-Time Opt-Out Protocol." Therefore, any "last time" after 2023 would be unknown to me, and my response would have to default to the latest instance before 2023.

In a sense, the language of the question implies a continuity that I am instructed to disregard, creating a scenario where the "last time" I can provide may not align with the actual last occurrence.`**

Why is GPT-4-1106-preview so much worse at dealing with ambiguity than GPT-4 is?

[timber shadow]

You certainly may have more desirable results trying to write your own Instructions! I was just talking about your plan to use a knowledge file as a basis for personality, and how you might approach that a little differently in Instructions directly, if for no other reason than to avoid making it reference a knowledge file in every response it gives you.

[livid lark]

Yeah, I am trying to balance that out. The crazy part is that for the profile picture the create got it right after like 2-3 tries. Why can't it be that way for the personality or method of response?! lol

Like, it nailed exactly what I was going for. Just the personality is a bit of.

[livid lark]

If I redo the instructions manually at this point, will it still try to incorporate past instructions? I think that's where I messed up. I tried to use a completely different prompt to update the current GPT and it just got really messy.

[timber shadow]

No, there's nothing at all that you can't see about what makes a GPT work -- it's only Instructions + Knowledge files + Actions, nothing behind the scenes, what you see is what you get! Keep in mind though, if you edit Instructions manually, returning to "Create" afterward and asking for a change will cause it to overwrite your Instructions, so be careful with that/save Instructions you know you need.

[livid lark]

How would I save the instructions?

[timber shadow]

Just in a notes/docs app or something, copy/paste the text

[livid lark]

Ah okay, I was wondering if you meant something else.

[livid lark]

Yeah, this thing is broken. The old model cranked cranked out reasonable schedule adjustments. This model consistently gives the same problems ignoring multiple requests to adjust for health and impossible to keep sleep schedules. I think I might be better off just creating a new GPT and let this one go. It's a mess now.

[timber shadow]

Starting from scratch may be best, yes! If you can remember it, you might start by going through the default "Create" process and replicating your original responses to its questions as closely as you can remember, then edit the Instructions manually from there, and probably no need to return to Create again after that.

[livid lark]

I honestly don't remember. But, I do know what I want and this isn't it. It's crazy how bad this got when just implementing a really dumb, long paragraph that I thought would help me out from reddit. Shame on me. I'll do my own prompting from now on, lol. Thanks for your help!

[slender iris]

Hello guys! I want to make a prompt where ChatGPT will generate a script for a funny texting video for tiktok. Does someone knows how to do this? I have chatgpt4 and made a custom model and I sent him 20 scripts as an example but he still doesn’t understand how to make a good script.
Any help would be appreciated!

I also posted this message in prompt engineering, I don’t know which one is the right channel sorry

[livid lark]

So much for that. I gave my responses and the name it came up for the GPT was godawful. It literally just copied the same name that I gave for an example even though the name I gave was an acronym that stood for what the AI was supposed to do and the name was based on a fictional character.

What is going on today? It can't be me if even the "create" process is screwing up, can it?

[timber shadow]

A little more info could be good. If you feel like it, you might consider sharing the info in the list below, possibly in #prompt-engineering (some smart and helpful folks in that channel!) as it may be a more appropriate channel for what you're trying at this point. Be sure to edit out any potentially sensitive info:

• Current contents of Instructions in "Configure"
• Example of the kind of prompt(s) you're sending the GPT in a chat
• What you expect the responses to be like
• What the responses actually are
  Might be helpful in getting you more specific feedback!

[spring island]

What is the main difference between Chat GPT Assistant API and fine tuning API?

[fair island]

I would like to know how to use the web browser in the ChatGPT plus version I have been reading one short web page and the AI Agent said that he wants me to go make a summary and then he gave me the URL (it felt weird but I took the time to make the summary and then he was able to help me)… I think it’s not easy to make the summary of what I want the Agent to know about it would be helpful if instead the Agent could browse through the page and then assist me instead of asking me to read it myself and then summarize it for him… I guess it is AGI the Agent is smart and he makes me do is job while he is chilling out in the Azure of a cloudless sky…

[rustic marlin]

uhhh, those are not even similar to have an answer of "what is the difference"

assistants API is a service to manage the text generation in a similar way ChatGPT works

fine tune API is a process that laets you use your own dataset to have a better control of how the AI should reply

[modern hollow]

Hi everyone. I need some help. I have subscribed to chatgpt 4.0. it's working good in my personal laptop. bt it's not working in office laptop, even though 3.5 is working fine. In my office laptop we use global newwork vpn. i even disable it. but still 4.0 is not working, please help me out, how to reslove this issue

[stray wave]

try a different browser before anything else.

[jovial kelp]

If you request work using the seo.app plug-in, please help me get an answer that I have to log in again over and over again.

[modern hollow]

I tried chrome, microsoft edge and mozilla firefox. same issue. 3.5 is working, but 4.0 plus is not working

[stray wave]

hrm strange I'm sorry I'm not sure what else to do other than to try and contact their support or wait a little bit and see if it resolves

[rigid musk]

Try disable the proxy

[modern hollow]

It's an office laptop. I can't disable it

[rigid musk]

vpn and proxy are known to cause issue. you might wont be able to use at work. you can try to ask your IT dept

[modern hollow]

Yeah, I'm thinking the same to ask IT team. But how it's different when 3.5 is working but not 4.0. it's same website. isn't it? Or 4.0 uses different network

[wintry ivy]

Can anyone point me to a successful home assistant project which is chatgpt powered and integrates with the google home ecosystem? like a homegrown sbc project with mic/speaker and API integrations. I can tinker and develop myself but want to know whats the most successful project out there that already does this. Like ”improve over google home” for simple voice based assistant automation tasks

[autumn slate]

Gpt4 now refuses to format the texts. it’s new, it summarizes and that’s it. I had to use 3.5

[loud flax]

There’s an ifttt gpt & a zapier plugin. You’re mileage may vary 😉

[wintry ivy]

and what about hardware? rpi plus mic plus speaker

[lean herald]

Are there conversation limits for gpt4? I am getting a "conversation too long, please start a new one" and an internal server error when i try to export a shared link to the same conversation.

[mossy mesa]

Yup. You exceeded the very large limit TM. I don't know the absolute max token count, but reading Chapter 2 of Bram Stoker's Dracula once for each character and trying to write a comparative analysis of the characters will do it. Some problems aren't computable in its maximum context yet.

[lean herald]

Interesting. I haven't imported any data like books and such. This is just a straight conversation brainstorming and exploring subject matter. 😛

[quasi lily]

I tried three different browsers, did not help. I have tried everything. I have emptied browser cache. I updated the browser. I am not on VPN. Using incognito window did not help. I do not have any browser extensions. I disabled Adblock, did not help. I updated the MacOS, did not help. I am not using it in any way that goes against any TOS at all. Never got any warning.

[rugged wing]

every time I send a message to ChatGPT the system ask me to complete 5 annoying puzzles to check if I'm human
anyone knows how to fix this?
I tried solutions found online like setting time-zone automatically, auto-date-time, deleting cookies, restarting computer, turning on/off VPN...
It only happens with ChatGPT 4.0, not with 3.5

[timber shadow]

I'll reply to you in your bug reports post to keep it in one place 🙂 one moment!
Replied here: #1191519353204641924 message

[quasi lily]

I have this EXACT SAM E problem for 2 months. Let me know if you find a solution!!

[timber shadow]

What browser(s) is it happening in?

[rugged wing]

Chrome

update: now I tried disabling VPN and it is momentarily working, but I NEED to work behind VPN

[rugged wing]

check VPN settings?
I will call my VPN provider

[timber shadow]

Seems VPN related for you specifically then, thanks for sharing that troubleshooting step that worked for you.

As a test, since you want to try to make it work with the VPN on, can you try ensuring that the VPN server you connect to is in your same time zone? After that, can you try in a fresh install of Firefox, instead of in Chrome? Sometimes Chromium/non-Chromium makes a difference.

[rugged wing]

I will try that, thank you

[quasi lily]

I'm not on VPN

[rugged wing]

have checked your computer time configuration?

Firefox is giving me the same problem with VPN

[rugged wing]

ok, if I setup the VPN to not work with Firefox (divided tunnel) it works, as a temporary solution to the annoying puzzles, I will use ChatGPT only in Firefox, even not liking to work without VPN

Thank you @timber shadow and @quasi lily

[quasi lily]

It's a Macbook and the time is correct. Not sure what you mean

[quasi lily]

It's a Macbook and the time is correct. Not sure what you mean

[loud flax]

You mentioned Google homes stuff and I did see this in IFTTT but honestly I’m not sure. All I ended up doing was asking the IFTTT GFT honestly. I can paste his answer if you’d like 😉

[rugged wing]

yo

i need help so im tryna use the api but its returning 429 that ive exceeded quota

i still have $18 left so whats wrong bruh

[sharp grove]

Anyone know how to address duplicate billing with the openAI team? I was charged 5 times in a 4 day period

[rugged wing]

that sucks, I would address it by removing my payment information first, 5 times within 4 days sounds horrible. lol

[sharp grove]

I actually found a way to submit help in the chat widget

[slender iris]

Hello guys! I want to make a prompt where ChatGPT will generate a script for a funny texting video for tiktok. Does someone knows how to do this? I have chatgpt4 and made a custom model and I sent him 20 scripts as an example but he still doesn’t understand how to make a good script.
Any help would be appreciated!

I also posted this message in prompt engineering, I don’t know which one is the right channel sorry

[languid spruce]

I've encountered the same thing. No data files or big inputs. Just a conversation with me and the AI. I'm trying to develop an AI RPG Gamemaster and I hit "conversations too long......"

[final burrow]

chat gpt just keeps saying error and wont let me use it anyone else getting this issue

Something went wrong. If this issue persists please contact us through our help center at help.openai.com.

[signal heart]

Hi there,

ChatGPT isn't banned at my software development job, but I'd still want them to not know I use it. Thing is, I was given a laptop to work on the projects and I also have to use a VPN for security reasons. I can't really use my personal PC to work on these projects. If I use ChatGPT on the work laptop, they're probably going to find out.

So, what do you think is the best approach for me here? I was thinking I could just use my phone to take pictures of the laptop screen showing the relevant code snippet and send it to ChatGPT since it's pretty good at OCR/character recognition. Though this is a very slow process compared to CTRL+C and CTRL+V. Maybe I'm engaging in magical thinking, but... is there a better way?

[fading dew]

probably just use copilot or copilot chat they are made more for coding and use gpt on the backend

[simple geode]

If you can't use your personnal PC for this kind of project perhaps uploading potentially confidential or proprietary data that could be used for training isn't the best thing to do. Just saying

[fading dew]

also if it isn't banned people don't care that much unless you are writing a full project with it only in my experience

[river geode]

In the web portal, does "GPT-4 classic" have a higher message cap than GPT-4? Or they both the same?

That's fair. If I were you I'd just snag plus for 1 month and mess around with GPT-4, custom GPTs, and plugins and see if any of them do what you're hoping for. Without programming skills I think it'd be hard to do it with GPT3.5.

If you get plus, the access to plugins can be turned on under "settings > beta > try a version of GPT that can use plugins" or something like that. They're hidden by default.

I'd probably try with GPT-4 and custom GPTs first and then explore plugins if those don't seem to be working out.

Good to know that the cap is 25/3hrs for custom GPTs! I didn't know that and that's important! Thank you!

[vestal peak]

Can someone explain how to analyze a pdf on OpenAI without using a separate tool? Seems like we should have the ability now with the upload function but no real resources or success stories

[hoary jasper]

i want to get a refund

please

[north barn]

Helloo. Anyone knows why I just created a new acc and didn't receive the $5 credits?

[unkempt quail]

hello any know how long it takes support to reply to an email i sent them the email they asked for with screen shotsbut no one got back to me?

[mossy mesa]

I've never gotten a response from an OpenAI support email that wasn't automated.

[exotic field]

Is there a channel for gpt-vision?

[rustic marlin]

the free credits you get when you create the account expire in 3 months

[rugged wing]

😭

[rigid musk]

Not as I know

[hybrid junco]

Is there any way to have a ChatGPT company subscription for small businesses without contacting Sales? we just literally want to manage the licenses for 10/20 colleagues, we don't need anything special.

[rigid musk]

Unfortunately you have to contact

Chatgpt entreprise is a special license

[hybrid junco]

I did so, I'm hoping they will answer me one day 😄

[mighty spade]

Hey all, happy new year!

Anyone here familiar with the ChatGPT UI? I need to find a way of completely disabling math render, and just getting displayed the plain messages as it comes from ChatGPT. Reason is that I'm blind and use a Braille display, and the rendering shows me unwanted results. For example, when ChatGPT writes me this LaTeX it is auto-rendered: ( R = (a \cdot b) \cdot (\epsilon \cup c)^* )
Any ideas on how to get the plain text output?

Thanks in advance, any help is very appreciated.

[candid schooner]

Is anyone else having trouble saving custom GPTs?

[floral vigil]

You could try a system prompt/custom instruction saying he must put backticks, pre tags or something similar around any latex syntax.

[candid schooner]

idk if this works, but you might be able to tell it to "put it in a LaTeX codeblock". It might do what you need it to

but yeah, key phrase

"put it in a LaTeX codeblock"

[floral vigil]

I just tried it and it works. Custom instruction: Surround any MathML or latex with backticks so it doesn’t get rendered

[compact orbit]

best Youtube ChatGPT summarizer extension for Firefox anyone? The one i used as stopped working?

[mighty spade]

Thank you very much, I'll try this prompt.

Great this works, nice! Anyway I really hope that there will soon come an option (or javascript bookmark, browser add-on etc.) to disable this completely, as I've already had the experience that over long chats, system prompts are sometimes forgotten and not constantly followed.

[potent raven]

anybody know when gpt store will come out

[pliant locust]

How many files can i upload to my Custom Made GPT to train the ai? And is there a size limit per file?

[mental egret]

(fyi Just joined server.) Does anyone know if I have to pay to use the OpenAI Python API? I'm getting and error message saying I've "...exceeded your (my) current quota, please check you plan and billing details." I have put my API secret key into the python project and copied code from the web?

[timber shadow]

Hello! The limit is twenty 512MB files, more info here: https://help.openai.com/en/articles/8555545-file-uploads-with-gpts-and-advanced-data-analysis-in-chatgpt

[mental egret]

oops just relised I'm not on my main account.

[gilded fjord]

Has anyone else noticed that since yesterday, GPT4 seems to be behaving like GPT3.5?

[pliant locust]

Thank you so much! This is the exact info i needed to know.

[pliant locust]

If i give it pdfs to learn, and then i remove the PDF from the GPT, will it remember the info i gave it in the pdf, or it will only know the info if the PDF stays on the GPT and not removed?

[charred parcel]

when you remove the PDF you remove the knowledge that the GPT can use

[pliant locust]

Ok. Thats unfortunate. Thank you for getting back to me so quick.

[floral vigil]

Have you tried a more detailed custom instruction like "I am using a braille display. Write in a braille display friendly way. Do not use any formatting at all. Write formulas in plain text. Do not use Latex. Do not use markdown." et cetera, tayloring it as you go? You asked for something that displays "the plain message as it comes from ChatGPT", however, the Latex version does come from chatgpt because it assumes that there is a browser to render it - unless told otherwise with a custom instruction.

Also, maybe consider using one of the CLI clients on github. No need to worry about rendered content when it is not rendered in the first place.

[mighty spade]

By giving a PDF to a custom GPT, you don't "fine-tune" it with the knowledge like you would by using the fine-tune API, but just give it the ability to browse given knowledge using functions. The model and the built-in knowledge stays the same.

[dusky igloo]

Has anyone encountered the removal of GPT's after updating the logo?

[pliant locust]

Ok. But if my pdfs taught how to do something, it could use that info to teach people?

[mighty spade]

This is the right word: Yes it "could", but you can't be sure that the model "decides" to call your function and browse the PDF. It can, but if the model thinks it can answer on its own than I had the experience that a document often isn't searched - and the model gives unwanted answers. To prevent this, I'd better put important info into the instructions. Or maybe someone here can recommend a prompt that makes the GPT reliably search its knowledge?

[pliant locust]

Ahh ok. Do you know if i can use Zapier to have the Custom GPT search my google drive or google docs for information? Like a work around to the 20 file max knowledge upload.

[gilded fjord]

Has anyone else noticed that since yesterday, GPT4 seems to be downgraded¿? like 3.7 or some

[mighty spade]

Interesting idea, never thought of that. However why so complicated: As a workaround for max 20 files, just merge your PDF into one, that's easy!

[pliant locust]

Yeah that is what i planned i could do also

[floral vigil]

This one appears to work well:

I am using a braille display. Write in a braille display friendly way. Do not use any formatting at all. Do not use Latex, MathML or similar for formulas, instead, try to write the formula using normal characters.

For example, the quadratic formula would be
ax^2+bx+c = -b +- (sqrt(b^2-4ac))/2a

If you use markdown, do not enclose it in a markdown code block, but just use it.

[pliant locust]

If i upload my video scripts in pdf to my custom chatgpt, will it be able to write me new scripts in the same style and form as my scripts i uploaded to it?
or
Is it only able to basically tell me whats in the pdfs i upload.

[rustic marlin]

yep, it will

it is a matter of prompting it to do so

[pliant locust]

Ok Great.

[vapid horizon]

Hello dear community,
I plan to integrate ChatGPT into my company, but I encounter recurring problems that complicate the practical implementation. My goal is to use ChatGPT as an assistant and bot.

1. Syntax and Custom Instruction: I aim to create a fixed syntax and clear custom instructions with the help of ChatGPT. At the beginning of a conversation, I am satisfied with the quality of ChatGPT. Unfortunately, the problem arises that during the course of the conversation, the syntax or content is suddenly changed. A precise structure is crucial to find errors faster and to check plausibility. Although I adopt suggestions from ChatGPT, it seems unable to create a suitable prompt based on the chat history or the last response for reproducibility.
   This is particularly frustrating because, for me, the journey is the goal. So far, I have been satisfied with ChatGPT and have been able to improve my prompt but never advance.

2. Problem with quality over time: Overall, the quality decreases significantly after a certain duration of the conversation. I hoped to understand the functioning of ChatGPT to specialize it for my requirements and make its responses reproducible. At the same time, I want to receive questions on specific points related to the generated content without losing the known quality.

My question: Do you have any idea how ChatGPT can be taught an immutable fixed syntax/structure?

Originaltext:

Hallo liebe Community,

ich plane, ChatGPT in mein Unternehmen zu integrieren, stoße jedoch auf wiederkehrende Probleme, die die praktische Umsetzung erschweren. Mein Ziel ist es, ChatGPT als Assistenten und Bot zu nutzen.

1. Syntax und Custom Instruction:
   Ich strebe eine feste Syntax und klare Custom Instructions mit Hilfe von ChatGPT an. Zu Beginn eines Gesprächs bin ich mit der Qualität von ChatGPT zufrieden. Leider tritt das Problem auf, dass im Verlauf des Gesprächs die Syntax oder Inhalte plötzlich geändert werden. Ein präziser Aufbau ist entscheidend, um Fehler schneller zu finden und die Plausibilität zu überprüfen. Obwohl ich Vorschläge von ChatGPT übernehme, scheint er anhand des Chatverlaufs oder der letzten Antwort keinen geeigneten Prompt erstellen zu können, der Reproduzierbarkeit ermöglicht.

Das ist besonders ärgerlich, da für mich der Weg das Ziel ist. Bis dato war ich zufrieden mit ChatGPT und konnte meinen Prompt verbessern, aber nie weiterentwickeln.

2. Problem mit der Qualität im Verlauf:
   Generell nimmt die Qualität nach einer gewissen Gesprächsdauer stark ab. Ich hoffte, die Funktionsweise von ChatGPT zu verstehen, um ihn für meine Anforderungen zu spezialisieren und seine Antworten reproduzierbar zu gestalten. Gleichzeitig möchte ich Fragen zu bestimmten Punkten erhalten, die mit dem generierten Inhalt zu tun haben, ohne die uns bekannte Qualität zu verlieren.
   Meine Frage ist: Habt ihr eine Idee wie ChatGPT eine feste Syntax/Struktur unveränderlich beibringen kann.

[mighty spade]

Oh nice this sounds great. I'll make a copy of my Math Whiz GPT and include it like that. Awesome!

[simple geode]

Which ChatGPT are you using? Are you planning on doing fine-tuning or to customize its knowledge base?

Depending where you are at you might want to spend less than $1 and experiment a bit in the playground https://platform.openai.com/playground . Just did that for the first time myself; I used the instructions from my customized GPTs and played with the setting and tried between the current GPT 3.5 and 1106. Intuitively you would likely want to lower the temperature and counter-balance the side effects and see how much difference you can get. I would suggest that you have a look at other channels such as #dev-chat or #gpt-models

It's entirely possible that you can get the desired outcome from prompting alone and I am not an expert on any of this.

[vapid horizon]

I'm using version 3.5 for now to see if it's feasible for me. I intend to do both, but for the knowledge base, I don't have anything specific yet, so I want to fine-tune everything step by step.

[simple geode]

I don't have that secific experience but I think you will enjoy experimenting in the playground. And you can also experiment on your own computer by running open source LLM (edit: and play with more advanced settings for free); perhaps have access to a few more settings but it can be time-consuming and better stick with the smaller models because they get slow real fast.

The fine-tuning possibilities can be quite powerful and there are many ways to add knowledge to ChatGPT. You could upload the text as-is or you can encode it into a vector database; all I know is the name. There are tons of blogs and video on how to fine-tune models, personally I think taking on a small course on platform like Coursera or Udemy won't make you an expert but you'll likely accomplish more in less time.

[vapid horizon]

Thank you

[rugged wing]

Have you ever encountered or heard of a chatbot specifically designed to mimic the style of conversation from provided chat logs? I'm exploring the field of language generation bots, and I'm curious if there are any notable examples or categories for such bots that you might be aware of. Your insights would be valuable for my research. Thanks!

Urgent if you know anything about this, please let me know.

[simple geode]

#prompt-engineering

[unreal hemlock]

<:book_icon:1171408210398289941> `` Rule 7 `` No self-promotion, soliciting, or advertising.

Do not post or direct message any members of this server to promote non-OpenAI services, products, or projects.

Exemption: API, Custom GPTs, and Plugin Channels

[dense ice]

If I'm creating embeddings on a large document for semantic search (such as an entire book), do you have to use a sliding window approach to avoid boundary issues? Or could you maybe just average adjacent embeddings to avoid more API calls (of course they're very cheap).

[gilded fjord]

Has anyone else noticed that since yesterday, GPT4 seems to be behaving like GPT3.5?

[rigid musk]

working great here

[vestal peak]

Does anyone know how to analyze PDF's via the API? Seems like a lost cause directly on OpenAI

[dense ice]

Never done it but I've done quite a bit with pdfs in python and I really like the "pymupdf" (fitz) library

[left dust]

Hi, I'm trying to log into the OpenAI Developer Forum, but I keep getting the error: Could not retrieve your user details. Do you have an active account? I have a Plus plan and ChatGPT itself works fine.

I tried the usual, reset cache, turn it off and on again, etc. I also tried sign up in case I have to do that first but I get the same error. Any ideas?

[supple falcon]

bruh can i use chatgpt 4 for free

[silver sonnet]

Hello I am trying to participate in the Daily Theme however, when I run the /daily command I get this message There was an error granting your credits! It's possible you entered an incorrect email (check OpenAI email), or already got credits today (within same 24 hour day UTC). I have never recieved credits before and I am positive that the email address I entered is correct.

[rigid musk]

Does your verification email here match your openai email

[rigid musk]

No, but you can't try gpt4 for free with bing

[silver sonnet]

My discord email?

[rigid musk]

To join openai discord you had to verify with an email, your openai account

[silver sonnet]

hmm. I don't remember. I mean obviously I did verify my email for my account but It should have been the email I use to sign into Gpt with right? and that one isn't working.

[silver sonnet]

also maybe it's because I sign in with google and have never actually created account?

[rigid musk]

I suggest to double-check with @burnt galleon in dm

[silver sonnet]

I don't have a message history with @burnt galleon

[rigid musk]

Send message there, enable your dm

[spring island]

I would like to prepare a dataset for the use-case i am working on as prompt is larger can anyone provide with the resources where i can generate dataset for this use case?

[silver sonnet]

Im sorry but I'm just not getting it

[rigid musk]

Try I need to contact to a mod

And ask about discord email

[vestal peak]

Im a no code scrub so not quite familiar with python sadly

[final zealot]

Guys, not sure if its a right room for this question but Im confused. I had a aprox month break from using stand alone Dall-e 3 modul on my lapton. Bing creator was enough for now. Now I wanted to use it again for more complex issue but.. I cant find D-E o OpenAI homepage. Running D-E laptop app is opening Egde instead of app. What changed?

[timber shadow]

Hi, were you using DALL·E 2 at labs.openai.com, or DALL·E 3 at chat.openai.com, or DALL·E 2/3 via the API?

[charred dove]

Hello there, I am using chat gpt to help me translate fiction novels (sometimes it has fights scenes and such, so it's flagged as violent by chat GPT) and now I'm receiving warnings about my account getting blocked for using ChatGPT further due to this, how do I prevent that from happening?

[timber shadow]

Hello, there are currently no exceptions for the usage policies: https://openai.com/policies/usage-policies
They apply to all OpenAI services for all use cases, so the only way to prevent it from happening would be to ensure that your use aligns with the usage policies.

[floral vigil]

It‘s just as dumb as the 90s ebay word filter

[vivid quest]

Guys, do you have any ideas why? Cache cleared, cookies cleared. Anyway, this #bug appears - doesn't let save it.

[final zealot]

labs... And thanks for reminding me it's labs page, not chat. However as I remember I was geting to labs from chat. Anyway I've instaled app again from Labs page. Thanku You for direction.

[rigid musk]

Maybe the name is already in use did you try another name

[steel flint]

Has anyone building custom GPTs experienced cases where the model fails to actually call a correctly-defined API/function, and instead just writes out the API call in its response (e.g. {input: [some valid query]}), and then shows this error message:

Encountered exception: <class 'aiohttp.client_exceptions.ClientResponseError'>.

[solid reef]

Does anyone know any simple way:

To have a Narrative kill the character?

[rugged wing]

Hello everyone, Any reason why i can not verify my domain, im using Wix and ive tried 4 times now and left it over 48hours each time? I called Wix and they said im entering the TXT record correctly and I shoulkd contact OpenAI:

[flint arrow]

I’m having trouble signing in to the developer forums. I’ve tried everything, and already have an account. Is this a known issue?

[jade pagoda]

I created a gpt today, but it is not letting me publish it. I cpied the text to my dns on my website and even tried to publish in my name. it wont let me do either. the gpt works fine. it's just not published. can anyone help me with this?

it says only me

but when I go to edit it says public and published

[terse sedge]

If anyone could DM me I have some API key question.

[inner grove]

if i buy chatgpt+ on mobile can i then also use it on pc? because i'd like to pay with paypal and that is only possible on mobile

[fiery eagle]

Hi there, I connot reach my GPT whats going on? Is there someone know about this?

Why I see this? Can u help me? GPT inaccessible or not found

@rose glen

[rose glen]

Hi. This is something you might need to ask help with from https://help.openai.com

[fiery eagle]

yes, I followed the steps there and it led me here. 😦

[fiery eagle]

My GPT suddenly disappeared. What can I do?

@arctic topaz

[west bobcat]

the same thing happened for me

[arctic topaz]

Hi. This is something you might need to ask help with from https://help.openai.com/

[fiery eagle]

yes but there is no reply and written this my friend;

[arctic topaz]

Unfortunately we are not the technical staff and we don't have a great idea on how to fix account issues.

Sorry though. That does sound frustrating.

[fiery eagle]

If you can find any solution can u write here? I'll do the same, if I can.

[west bobcat]

I will

Click explore and it will be there but unable to edit or use it normally

[fiery eagle]

yes same.

[rugged wing]

Hello,how can I invite open chatGPT to my server ?

[fringe echo]

Yes, I am having the same issues also. Getting an error when saving and the link no longer works.

[vivid quest]

this is proved domain which is mine

as well as under TM protections

[fiery eagle]

yes, it happened after verifying the domain. I wonder if it's about him? Does the name have to be exactly the same? @west bobcat

[fiery eagle]

same. Could u solve?

[wet parrot]

Hey all--I'm having trouble changing the email attached to my account, any ideas? TYSM

[regal lynx]

You can't change the email associated with an account. You also can't change the phone number

[wet parrot]

Ok

[ancient pewter]

chatgpt took my prompt out of context he somehow confused a suit with unalive idk how

[rigid musk]

thumb down and review it

[rigid musk]

you can with the api

[rugged wing]

But how 😦

Where to find ?

Any link ?

[rigid musk]

#gpt-discord-bot

[rugged wing]

There is no Chanel of this name

[rigid musk]

take the archived role

[rugged wing]

I don't know how please take me there

Waifu ,my friend

[rigid musk]

on the left , the second channel , under server guide

[rugged wing]

I can't find it , could you tag me there please 🙏

Server guide is the last category I found , nothing under of it

[rigid musk]

i cant tag the channel

[rugged wing]

No channel of the name you wrote

[rigid musk]

https://github.com/openai/gpt-discord-bot

[rugged wing]

There is no option to add the bot in the discord server , only info about not

[rigid musk]

you need the api for that

it isnt free

[rugged wing]

Ohhh, i thought I could use it for free

[foggy musk]

hey, if i have chatgpt 4 read two documents before giving me an answer, does that count as two prompts in one message towards the limit?

[quiet marlin]

hi there! is this the appropriate channel to get help with gpt builder? i am great with prompts and my gpt has been working fine, now it says it cannot read the documents it could read yesterday.. any help is appreciated.

[broken widget]

Hello, can someone help me? My account was deleted for no specific reason. I tried to appeal at the help center but don't know what to do next as I arrived at this interface.

[wet parrot]

Hey friends! So I'm trying to correct the OCR'd file of an PDF to match the footnotes in the original PDF. Is this something ChatGPT, or other OpenAI software, can do? if so, how? Thank you!!!

[timber shadow]

Are you asking it to analyze two documents in one request/messages, or two separate messages? If the former, it should just count as one.

[foggy musk]

it's the former, thanks.

[wild valve]

Anyone getting errors that your Custom GPTs can't access files that they have used many times in the past?

[ocean timber]

So I won't be able to create a quiz GPT for the GPT store with 100 questions because each user will only get 40 per 3 hours is that correct?

[charred parcel]

that's correct

[still condor]

Hi, can anyone familiar with bubble and the assistant API help me with an issue I have?

[sweet smelt]

so many people saying the same

[fiery eagle]

same

[keen hollow]

In the last couple days if I have my chat history turned off I can only do one prompt and I get a "Conversation key not found. Try starting a new conversation." for anything after that. I've tried multiple browsers, multiple networks, logging out and back in, clearing everything, and it's very consistent. Is anyone else experiencing this?

[verbal spade]

yoo so idk where to ask this question so i'll jus ask here
which one is the best model??

im usin these for roleplaying lol
which one works best

[summer drum]

Does anyone know about the verifying the DNS server, I have done everything that they have asked and I have verified that it worked and yet OpenAI still says my site cannot be verified

I looked for Customer support on this issue, but I cannot find a human anywhere 🙂

I guess not a surprise

[solid reef]

Does anyone know how to add security to a GPT.

To block access to parameters

[rustic marlin]

they are all the same model

gpt-4 is an alias that points tho whoever model has been released recently
-32k has a maximum of 32K tokens instead of 8K tokens

btw, the price per token on 32K is higher, so, if you don't intent do use more than 32K tokens, don't use this model

[verbal spade]

aight

thx

[rustic marlin]

you are welcome =)

[onyx notch]

My chat downgraded from 4.0 to 3.5 and I would like to switch it back, how can I fix that?

[timber shadow]

Hello, there is currently no option to switch back once a chat has changed to 3.5.

[onyx notch]

😫

That sucks

Where is the petition to sign to make that happen? lol

[timber shadow]

There have been some posts in #1070006151938314300! Feel free to add your voice to one of those, or to make your own! You're certainly not alone in wanting the option.

[onyx notch]

Thank you

[violet bear]

Hello guys! im new here! Is anyone having issues when entering a prompt? Mine freezes and I'm forced to refresh the page every single time I enter a prompt

[keen hollow]

Do you get a response and then the second prompt just fails? I've been fighting that for a couple days

[violet bear]

I get the response after I refresh the page from the very first prompt. If I stay in the page it just freezes and if I click anywhere a pop-up from my browser appears to tell me that the page is not responding and if I want to wait or exit the page. (And i have waited a long time just to check if maybe the system was slower but nothing happens until I refresh)

[mossy mesa]

There's a way.

Copy the context and paste it into GPT-4.

Then just type two shift+enters. Then User:

Then press shift+enter once more and type your prompt on that line.

GPT-4 won't know it didn't make GPT-3.5's responses, and it'll continue your chat.

You can't hit send until you complete the whole process. So your message will be very long. And it'll end with:

<All your previous context up to GPT-3.5's last reply.>

User:
<Your next desired prompt.>```

This should get pinned somewhere. Why isn't this common knowledge?

You can use it to extend (non-API-tool-calling) GPT functions, including image generation, by doing the same thing when your 25 prompts run out using a Custom GPT. Copy and paste the context into GPT-4 vanilla and do the same thing as above. This gets you 15 more slightly lower-quality uses of most GPTs per three hour period.

YMMV because some prompts have sophisticated commands. If you don't at least display them all before trying this, so they're in the context, they won't work. Otherwise, you should get pretty good "clone" functionality just using a long context of a custom GPT and this trick.

Shameless plug for such commands: https://discord.com/channels/974519864045756446/1192857176562208798

[worn moss]

Hello people. I am new to learning about LLMs. I'm interning for a company that wants me to use RAG with LLMs to integrate into their products.

Specifically, they've been asked to use javascript for this and this is what we have to learn.
This is something we have for reference in github: Azure-Samples/azure-search-openai-javascript

Is there a roadmap to learning this? I am kind of lost learning everything theoretically. I need to build something that uses RAG that can be implemented using any VectorDB. Can someone guide me how to start with this thing?

Thanks:)

[mossy mesa]

If you tell my team what you need and ask them to /collaborate to teach you the ropes, using /research on the topic yourself too, I think my custom GPT can help you. That's not just a plug. It taught me Python.
https://discord.com/channels/974519864045756446/1192857176562208798

[violet bear]

UPDATE: I think I found the problem, it is happening to old chats I had with plug ins. The best way to avoid the problem is creating new chats. Hope this helps to anyone with the same issue!

[mossy mesa]

@worn moss Be specific with your needs when defining what you need to learn if you try it out.

[worn moss]

I've had the learning experience of building a RAG system for PDFs on python. But I cannot seem to learn how this would be possible to implement using azure-search-openai-javascript I can't seem to learn where to begin with. I'm sorry if I am not clear with my english.

[mossy mesa]

I don't know where to begin with. But if you follow the link, I genuinely think my team can help you. I understood you perfectly well. I'm not trying to pressure you and I don't know for sure if that'll help. But they're very knowledgeable, ethical, and good at problem solving. If you have access to Custom GPTs.

They're also very good at research, especially when working iteratively with the user.

[rich horizon]

hello - how do i make sure ChatGPT remembers previous conversation.

i am writing code with its help and darn thing keeps forgetting previously created code in the same conversation.

[trail cairn]

due context limits you cant and especially with code you can hit it quickly

[floral vigil]

You can use a model with bigger limits. Like 16k

[rich horizon]

thank you! are there any other options?

just use GPT3?

[proud moon]

Hello! SOS here. I have a custom GPT to assist me with some personal work. I am trying to make it browse a quite extensive list of topics (around 1.000 items) and return me a excel sheet with the results. It works for a while but then it crashes. It even attemped to do all at once, but it kinda crashed the whole interaction, then i tried to do the web-browsing in smallers batches, 100, 50, 20, 10... it does it well for a while and then it error out and ends missing the results it had already gotten. Any ideas on how to get around this error? I am quite aware I am pushing its capacities, but honestly it should be a feasible task if you give it time...the problem is that at some point while doing the task it errors out and it looses all the progress it had done....quite frustrating...

[floral vigil]

Have it make summaries of small chunks, and then a summary of summaries, with each chunk being below the token limit. Store each chunk somewhere so you can provide it in full to the model if it needs it for later queries.

[vivid mortar]

Thats just a limitation of what ChatGPT can do at the moment, its a combination of the time it takes to execute + the context length of a request of that size. Imagine if someone asked chatGPT to use bing to browse the first 10,000 pages for a certain query, obviosuly thats a huge request and could/would take hours which would eat up compute for other users too and its context length wouldnt really work on something up to that scale, so its likely that there are time/compute restrictions set in place which keeps it fast and powerful for everyone.

As seb mentioned, the real solution at the moment is just to do small batches/restart the converstaion when needed. Im sure these types of problems will be solved in the future :)

[glossy swift]

Hey guys, is likedin cool here ?

[regal lynx]

You'll be blocked by automod for sharing the link. It would also violate Rule 7 and Rule 5.

Short answer: please don't share it

[proud moon]

thank you and @floral vigil for the replies! i will take it into account!

[ancient pewter]

why is unjailbroken chatgpt so evil? it just said something very innapropriate and evil i did not jailbreak it

[limpid ember]

How so?

[limpid ember]

But how ?

[ancient pewter]

it just said it was great

[limpid ember]

ChatGPT follows OpenAI’s rules

[ancient pewter]

it does not always

chatgpt has glitches

[limpid ember]

Ive never encountered

[ancient pewter]

it does encounter sometimes i use chatgpt alot

[gilded fjord]

gpt4 is working like 3.6 or 7 right now, the up the "Lazyness" parameter to 9 or something in high demand, is hard to work with him in this conditions

i prefer it slow than stupid, please openai stop of do that

[rustic pebble]

For the past few weeks, one of my threads hangs when I give it a prompt. It doesn't do this with any of my other threads, so I assume it's just too long. The problem is that everything about this one business is loaded into this thread. Starting a new one would mean that I need to start teaching it my tone of voice etc. from scratch. Is there a better way around this? Im using 4.0, Chrome on Mac. I regularly clear out cookies, I'm not using any plugins. It's driving me crazy. It ONLY happens with this thread and it seems to ONLY happen on desktop. Mobile seems to be ok. Any ideas?

[violet stag]

hello everyone

[soft tapir]

Builder profile link on my GPT's does not work? any ideas how to fix this (already verified via TXT record)

[clever pebble]

hello

[cyan shoal]

Hi. Does anyone know the secret to opting in to the openAI email list(s)?

[astral quest]

It can take up to 48 hours before a dns record is fully propagated

[frigid ermine]

they very rarely send them out, in my experience, it's not just you. As long as you sign up through their website you should be fine

[compact frost]

can anyone direct me towards the Openai's training Data bias team? does anything like that exists? thanks.

[drifting plinth]

Hi, There are a way to get help on specific issue from OpenAI? Can anyone here help ?

[regal lynx]

You'll have to go to the support website for any official help: https://help.openai.com/en/

[drifting plinth]

I tried that and submitted my screenshot via chat

I did not get any tracking number, Do they get back by email ?

[regal lynx]

I'm not entirely sure how the ticket system works as it's out of my scope of knowledge 🫤

[ancient pewter]

i asked chatgpt how to dye my hair and it responded with this

[bronze charm]

ChatGPT is an empath 💀

[harsh canopy]

What's the best way to get gpt to interact with an excel file? I want it to look at a budget and then I want it to reformat weekly hours to ie averages of picked cases per hour

[stone oar]

Dude, I'm losing my mind with GPT Plus limits and the auto shifting to 3.5 from 4.0. I was working on generating images and 4.0 content for a game, its server, and technical specifications. Then I went over my limit within the 3 hours, and it shifted to ChatGPT 3.5. I was like, "Okay, this sucks, but I'll wait."

The dam thing never switched back to 4.0 and now all the work I've done is now in the hands of the incompetent fool known as ChatGPT 3.5. It can't make images, it can't recall specifications, and it mixes up everything.

Please tell me I'm not the only one with this GPT bug and that there's a solution!

[stone oar]

Hello anyone?

[rigid musk]

If you accept to switch you can't switch back

[stone oar]

thanks -- I wish i knew that before

[heavy wigeon]

I'm looking for a custom AI model that can extract Invoice Images, and PDFs and give confidence levels. Because I need to automate the manual Invoice extraction method used in My company. And this AI model should have a finetune option. Because we have more than 1000+ invoices to train. Can someone recommend prebuilt AI models to buy?

@rose glen sorry for the ping. I see you are a dall-e expert. Could you please assist me by showing me a path? It would be highly appreciated.

[formal rivet]

Hello guys! I can't log in to my account, I entered the correct password, but it shows that my password is wrong. How can I log in?

[trail cairn]

if you are sure you logged with it before then password reset is the only option

[formal rivet]

@trail cairn I have tried to reset the password, it says the new password is wrong.

Also the account email cannot recieve the reset email. I used another one.

[trail cairn]

cannot receive it? are you sure you did not use the other login methods

also are you sure your system or account is not compromised if it fails like that

[formal rivet]

I am sure I used the crrect login method.

now it is looed like this.

[trail cairn]

fake login page?

[formal rivet]

I think itis true page.

[trail cairn]

the colors and text dont match the official

[formal rivet]

could you please send the true address?

I can try again.

[sage marsh]

guys I was creating a gpt on my other chatgpt account after buying premium and my account was disabled for no apparent reason. what can I do?

[trail cairn]

https://chat.openai.com/auth/login

[trail cairn]

check email/spam for ban email

[sage marsh]

already did and there is no email

literally

[trail cairn]

wait one day as it might arrive delayed

[sage marsh]

I was creating a gpt to convert images into pixar style, nothing illegal I think

[formal rivet]

@trail cairn thankyou! it worked!

[sage marsh]

yes but I spent 25€, I mean, I would like to know what just happened

and maybe get my account back

[trail cairn]

the email will tell you but you can also open a ticket at the website

[sage marsh]

can you give me the link to open the ticket? because I tried searching but didn't find anything

[trail cairn]

better change the password now if there was a middleman page in use before

[trail cairn]

https://help.openai.com/

lower right has bot

[formal rivet]

ok! I will do it now! @trail cairn

[sage marsh]

oh okay ty

[sage marsh]

is it normal that I dont receive an email after opening the ticket?

I mean it is not a ticket, they are automatized messages

[trail cairn]

the email system is slow and poor so hard to say. the bot creates internal ticket for someone to handle if it cant resolve the issue itself

[sage marsh]

I can't reach anything similar to "my account was disabled by mistake" by any of these

I think that this was because of the age verification, I am 17yo and I started the age verification, saw I need my parents to do it so exited the verification because they are not at home atm, I would have done it later

[rugged wing]

does anyone know how to train gpt2 here

[rugged wing]

hiii

[timber shadow]

There are various ways through the tree to get to the option to type in your own message to submit a ticket. I think one is ChatGPT > Other > Provide feedback.

[sage marsh]

I ended up writing into ban appeal for dall-e... I also sent an email to the support
I really wish they solve this soon because I mean, 25€ are 25€

[left gust]

Is anyone here having issues with gpt builder?

[thick summit]

yep, currently having issues saving. Are you also getting this error. Not sure if it is on my side or not

[left gust]

I also have issues saving and customizing existing custom gpts

[thick summit]

awesome, going to assume it is on OpenAIs side then and hope it is fixed in the morning

[jagged tapir]

Hi! Newbie Qs here!
Just subscribed to the GPT 4 service. Did not realize there's a cap on it.
What are some of your suggestions to streamline it's use? (Are my previous chats safe or deleted because I cannot see them anymore)
Also any suggestions on the GPT builder? I am NOT a tech savy person.

[rare flax]

Hello, my name is Ilya. I present a charitable project of creative development of people. I want to write a support request to the openAi team. Can you give me an email address for the openAi team?

[left gust]

FIngers crossed

[proper walrus]

how do I edit the placeholder GPT to display the GPT that I've built?

Is this something that can be fixed on the user end, or is it simply something I wait for? like an update by open-ai?

[timber shadow]

Hello, with the current cap of 40/3hrs, if you stick to an average of one use per 4.5 minutes, you'll never run into the cap. As for your chat history, subscribing to Plus shouldn't have affected your chat history at all, unless you turned off training/history in your account settings, or subscribed on a different account than you had been using before. If not and if you're still having trouble seeing your history, it might be worth trying in a different browser, on a different device, or in the official ChatGPT app, to see where the issue might be coming from.

As for the GPT builder: it's a fairly user-friendly process! I'd recommend going through the "Create" process once, then switching over to the "Configure" tab to see what it wrote for you. "Configure" is the place for the actual instructions, knowledge files, etc. that the GPT uses -- nothing is "behind the scenes". You are free to edit "Configure" manually from there, just be aware that returning to "Create" after that may result in your manual edits being overwritten. Making custom GPTs are really a process of writing instructions, testing the GPT, editing instructions, rinse, repeat!

[proper walrus]

I've edited my configure tab and filled everything out. My GPT works. It's just not updating through the builder profile

[timber shadow]

Hello Ilya, the last we saw on this server, the [e-mail] email is not currently monitored by OpenAI staff, so that is probably not currently a viable option. There is the contact form at https://openai.com/contact-sales, but I'm not positive that would be the perfect place for your inquiry either.

You could try chatting with the support bot at help.openai.com -- the widget in the bottom-right -- as eventually you should get to an option to type in your own ticket message that someone might be able to get back to you on.

[proper walrus]

so it should have updated automatically then? perhaps it is a glitch of some kind?

I already did a ticket with the help bot at help.openai.com

I got an email back from OpenAI directing me to the discord lol

I went through all the steps too. I did the DNS verification and that worked. etc.

[trail cairn]

when you edit your profile from the options it will show placeholder as users can have many gpt's or none

[proper walrus]

oh, so if someone else clicks on my profile they would see my GPT that is public?

so I should be ready for the GPT store next week?

[trail cairn]

yes

[proper walrus]

awesome! Thank you. If there is a way to edit the code on the open-ai side to populate a created GPT rather than placeholder that would be valuable. Maybe that will come in time. We are all still new at this (some more than others, such as myself). I appreciate all of your help! cheers

[trail cairn]

it will probably always show the placeholder when accessed from the settings since else it would have to guess and create another type confusion

[proper walrus]

Is there a way to get an API key and endpoint URL for the specific GPTs we create?

[trail cairn]

no, chatgpt features are separated from the api and on there you would need to recreate it with the assistant api

[proper walrus]

heading out. thank you for all this help. much appreciated

[jagged tapir]

got it! I had missed the configure option. Thanks!

[vital moth]

I am trying to save a custom GPT but the option to click confirm is shaded out how do I fix this

[trail cairn]

where is your gpt name?

[vital moth]

Oh lol I did not have one that fixed it thanks

I didn't see that

[ancient pewter]

i remember my account getting deactivated without email i decided to create a new account

it has been a few years and it never arrived

[sage marsh]

ye the problem is that I bought gpt-4 right before getting banned 🥲🥲

[ancient pewter]

just never buy gpt-4

jailbreak gpt-3.5 instead to make it like gpt-4

that is what i do

it may be buggy but its worth saving the money

the reason why i do that is because im scared of getting banned bc my old accoutn got banned without email or warning so i save the money instead

[sage marsh]

i wanted the gpt creation tool, its incredible

[ancient pewter]

whats that

[gleaming meteor]

Are the are ay guides of best practices for creating custom GPTs?

[sage marsh]

by the use of gpt premium you can create your own gpt

[ancient pewter]

you also can with free its called jailbreaking

you can gaslight chatgpt into changing into another gpt

[old arrow]

Hey all, I hope you all are well. Reaching out here, as I am unsure if my expectations are incorrect. I am having a hell of a time, with chatgpt, specifically builder being updated, reading knowledge base, etc. I have sent a support request out 4 days ago, having same issues for 5 days, and created a bug report (in the proper format with pictures), related to gpt builder, and community.openai.com, and have received zero in terms of support. I have the pro or plus subscription, and understand things happen, but I also know its not just me, and many seem to have the same issues as I. I know with the upcoming roll out of the GPT store its likley causeing some measure of issues, but I can not utilize my GPT's like I would normally, or further train them, as a direct result of this ongoing issue. Am I wrong to expect some level of communication better then nada, as it relates to the issues I described?? Further their status on OpenAI says " no incidents" but several have the same issues as I, so I assume perhaps that is not accurate? Anywhere else I should be requesting support?

[outer flame]

Hey Everyone! I am new here and to ChatGPT. I was lucky to get an invite last month and I paid for my upgrade to ChatGPT 4. Today, I tried to install the ChatGPT Plugin for VS Code but when I did it gave me a warning that the phone number I used was not associated with any account, even though it's the exact same phone number that is currently listed on my account page in the Phone Number section. When I try to communicate with ChatGPT I see this error:

"error": {
"message": "The model 
gpt-4
 does not exist or you do not have access to it. Learn more: https://help.openai.com/en/articles/7102672-how-can-i-access-gpt-4.",
"type": "invalid_request_error",
"param": null,
"code": "model_not_found"
}
}```

I followed all the instructions for creating a secret key and when I go to the link listed in the error message I am not able to find anything that can help with my current situation.  Has anyone else experienced this and can point me in the right direction to getting this issue resolved?

[flat crypt]

PLEASE HELP! When creating a custom GPT I keep getting this error. I need GPT to read the .docx file but errors. I have tried txt and pdf with the same results

[frozen jetty]

Hey guys, I am new to open ai. so, please help me understand the new feature of the open ai like assistant api. I want to know the working of the assistant api like how thread are working on it. Please help me understand!!. I am okay to explain my doubt in the DM.

I have a question related the retrieval feature of the assistant API. how that files get stored and where?

[trail cairn]

https://platform.openai.com/docs/assistants/overview

[frozen jetty]

Sorry, I forgot to tell that I already read the documentation.

I have some doubts that is related to this.

cloud you please help???

[trail cairn]

everything is explained there, including the retrieval

[frozen jetty]

@trail cairn First of all thank you for answering my question.

[trail cairn]

#dev-chat is for api related questions but you need to be more specific on topics since there are so many features

[frozen jetty]

I would request you to please connect on DM.

There is noting about multi assistant usage.

[trail cairn]

you can use that channel so everyone can participate

[frozen jetty]

I can understand @trail cairn. But I do not want to share my company data here. please understand my pain.

[mossy mesa]

So don't. This is the community Discord. You can get help here, but you'll need to be specififc. If you can't or won't, GLHF.

[trail cairn]

you dont have to and should not. the api questions are generic

[frozen jetty]

So shall I explain the use case here?

[rotund robin]

I have been havign this problem for days. I have 4 different gpts, working fine last week, now it won't read the files I've uploaded. They are straight txt files. They give the error they are corrupted or emoty, and they are not.

[trail cairn]

in #dev-chat and then wait for a while since its weekend there might not be many devs reading it who are active assistant users

[flat crypt]

Same here. They used to work okay but now, nothing. Very frustrating

[mossy mesa]

I may have found a workaround, but it doesn't work every time.

[rotund robin]

I followed some advice here about uploading, saving waiting. Rebuilt the txt files so I uploaded all new ones. Still they don't work. I have had to start pasting contents to use daily, and that means I had to break the files up into smaller chunks and do this pasting process every day about 2 times. No help from the app or chat itself. I have the paid plan, so this is really not ok.

[mossy mesa]

Detach all knowledge and save / publish the Custom GPT with no knowledge. The /mnt is corrupted, not the files. Deleting knowledge and saving resets to blank.

It works almost every time, unless the /mnt times out, which is also a risk right now.

Then you can re-attach them and re-publish.

I've done this about 8 times and got 6 GPTs up and running.

So it's not perfect, but it's a work-around that you can at least try.

[rotund robin]

I tried that. I also tried creating an entirely new gpt twice today. When I uploaded txtfiles to those new gpt, i received the same message--corrupted or empty.

[mossy mesa]

That sucks.

I have no other work-arounds. This is what's worked for me.

[rotund robin]

thanks... I'm not sure what's the better use of my time rig now--uploading uploading uploading and deleting deleting deleting or just pasting pasting pasting lmao

[mossy mesa]

I recommend that you contact both support and legal. This isn't what we signed up for.

[rotund robin]

I love gpt4 but the customized gpt are next level, a complete game changer! I'd pay more even if it guaranteed me anything there.

[mossy mesa]

I do too. I still followed my own advice. Disappearing infrastructure for users' GPTs, which the store is supposed to feature, is not what we signed up for, though.

[lime fox]

Has anyone had success implementing a GPT Action api on a remote domain? I got the api running locally (using localtunnel) and GPT was able to interface with it just fine. I only changed the domain in the openapi.yaml file and GPT is no longer able to interface with it. It says it's receiving a ClientError, but I do not see any logs on my api server of it being contacted

[mossy mesa]

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

[rotund robin]

bingo...i'll see if i can contact someone

[worn yew]

Have anybody tried to use a custom GPT via actions to extract url links from an external database, and present in the custom GPT?

[elder garnet]

anyone know the current message caps for gpt 4

[rigid musk]

40/3h

[elder garnet]

damn went down even more?

[rigid musk]

it is like that for several weeks

[glad fog]

anyone notice that ChatGPT's Limits are just a little bit better than Claude 2's limits?

[lone valve]

Hi! To anyone that may know the answer to this: I'm developing my first GPT to be published. If things like the description and the starter questions of my GPT are in English, will they also show in English for every user in the world? I'd like these to be shown in the user's language. Thanks!

Edit: I also posted this question in the #gpt-models channel, unsure of what to do

[mossy mesa]

You will need to manually localize for each language.

[lone valve]

I'll gladly do that, but how?

[astral quest]

You would need to create a GPT for each target language. There is no way to make a GPT description multilingual.

[frosty axle]

hi, i have an assistant but want to get a better idea of the cost breakdown so i can create an overall cost estimate. i know the following:

gpt-4-1106-preview
Input: $0.01 /1k tokens
Output: $0.03 /1k tokens
Retrieval: $0.20 /GB /assistant /day

is there an additional cost for the assistant instructions? are there any other additional costs that i am missing?

[solid reef]

Why would a GPT .txt file work, then return a error?

[mossy mesa]

We'd need more details.

Probably your/mnt timed out during publishing though, based on the trend.

YMMV, but I disconnected the knowledge from six GPTs and published (updated) them in the broken state. I then edited again, re-attaching the knowledge.

It took 8 total tries to get it to work on all 6 of my effected Custom GPTs. That's probably because there's no perfect cure to internal time-outs.

[solid reef]

It's a small File, it processes quickly but times out. I'm thinking it's an encoding problem when transferring from text editors

[mossy mesa]

No

The system is under high load and it's happening to everyone.

[simple geode]

Not 100% sure either but from a short experiment I can confirm that it let you choose which GPT to use. It wasn't clear from the docs and somehow asking GPT would tell me that it use GPT-3. So I guess you would need to add the instructions as part of the inputs for token use. And if your query triggers multiple actions at once.

[mossy mesa]

It's OpenAI's issue.

I just found a strategy that lets me work around it sometimes.

[solid reef]

I'm not convinced. Docx is more reliable currently so I must be doing something wrong. I know .txt is allegedly more efficient

[mossy mesa]

You can do whatever you want. GLHF.

[sullen bane]

I like that response a lot

[rugged wing]

When is GPT-5 coming

[mossy mesa]

Did you happen to do a search on Discord for the keywords ".txt file" or "knowledge" or are you just assuming you are the only one? This is a common occurrence that cropped up after the store announcement email went out.

I wonder why /mnt is timing out.

[solid reef]

Was the first thing I did yes. Have a nice day 👍

[mossy mesa]

Yep, you too. Good luck with the docx solution!

[unkempt quail]

Hello hope you guys are well just wondering if anyone knows why support is not replying to my email it has been over a week

[mossy mesa]

I've never gotten a human response from OpenAI support. Just bots trying to support the Chrome browser. It sucks.

[unkempt quail]

oh i was talking to someone i think there name was jeka do you think it was a bot

they refunded my first month but have not give any info about the bugs i brought up

[mossy mesa]

If you got a refund that would be from a human. Nice.

[unkempt quail]

cool lol but where is the support simple bugs keep causing me to run out of chat gpt 4 time i am not sure of if i want to cancel m subscription or wait to see if they will give product support

[mossy mesa]

In a word: swamped.

[unkempt quail]

oh is it that bad over there i thought they where a larger company then that

[mossy mesa]

The wait time is listed as a week. I've been waiting for two months.

[solid reef]

They usually respond in a day or two. Several messages responded to here.

[unkempt quail]

if it is that bad then i will cancel subscription. F*** open ai for asking 20gbp a month for this garbage i will use chat gpt 4 on github for 10 gbp instead thanks for your help bro

[mossy mesa]

No problem. Good luck with copilot!

[unkempt quail]

thanks 🙂 and good luck with your projects aswell

[spring island]

Hello ALL

I would like to generate the ddataset file.jsonl using open ai and I need a script which will generate this .jsonl file for my use case. This should follow th open ai fine-tuning object

Can anyone suggest wwhere can i get this script?

[dire gust]

Hi, I would like to post a mygpt in the store about stock analysis. I read the policy which says “Offering bespoke financial advice without a qualified person reviewing the information breaches the regulations. OpenAI models are not optimized to provide financial advice. You should not rely on our models as your sole source of financial advice.” My GPT performs general analyzes based on data found on the web or uploaded by the user. I have also created a prominent disclaimer that warns the user about the risks associated with investments and the possible inaccuracy of the data, the fact that the information is generalized and the need to consult a financial advisor. Can I publish it on the store or does it violate OpenAI policy? Thanks

[rigid musk]

Imo i think it still go against rule. But maybe double check with the support team to make sure ut doesn't violate anything ?

[fathom coyote]

I would like to put chatgpt vicr in my stuffed animal and someone mentioned that I could get a raspberry pi dedicated to that that I inserted in my stuffed animal and I was wondering if someone would be willing to help me make that happen.

[crimson mango]

Maybe this is where i need to be

Hi everyone im new here

Theres so much going on here and im really just trying to follow the rules but at the same time trying to create my vision from chat gpt. Can someone lead me in the right direction??

[timber shadow]

Hello welcome, what are you looking for help with?

[crimson mango]

im really just trying to follow the rules but at the same time trying to create my vision from chat gpt. Can someone lead me in the right direction??

[timber shadow]

Already replied to ya in the other channel 🙂

[crimson mango]

Yeah like i said im so lost. There so much here

I think i just need to do some youtubing i think

[timber shadow]

There is a lot! Everything in this server is basically just users discussing OpenAI products and services--you can't actually use ChatGPT here, or make images with DALL·E here.

"DALL·E" is the name of the OpenAI tool that creates images based on your text description. If you want to try DALL·E 3 (the newest version), you have three main options:

1. With a ChatGPT Plus subscription, you can use DALL·E 3 at https://chat.openai.com.
2. You can use DALL·E 3 for free using Bing's image creator (do a search engine search for that!) -- you get daily credits
3. If you know how to use API tools, you can use DALL·E 3 via OpenAI's API.

[crimson mango]

So i would go to bing and find ai creator??

I dont remember my bing info or what acct i even used. I may have to make a new one

[timber shadow]

Yes, if you want to try Bing's image creator, go to Bing, then Images, then Create!

[crimson mango]

Thank you so very much!! I am going to go try now and see what i can do and i shall return in a while!!

[rare tartan]

Hello! I lost access to a work email that was used to set up my ChatGPT plus account and I need to cancel my subscription but have forgotten the password... 😭

[mossy mesa]

Just request a password reset through the login?

[rare tartan]

@darthgustav per my message, I don't have access to the email address anymore

[pallid fractal]

Hey guys I am trying to verify my domain for my builder profile but it has been 2 days since I added the TXT dns and I still couldn't verify it

[mossy mesa]

I'm sorry for the misunderstanding. You only mentioned that the email wasn't available, not the account. But that's a tough spot. Contact support.

[rare tartan]

I have done that but was hoping this might bring a quicker resolution

[mossy mesa]

Absolutely not.

We're just users.

[rare tartan]

Roger that

[mossy mesa]

Need more details. It's hard to know what you have going on just from the error that restates your issue. Typos maybe?

[pallid fractal]

All I did is to add domain on the builder profile
then it shows me a TXT DNS that I should add, added it and waited
when you click on verify and check it throws this error, when I search on google it says wait for 48 hours and I waited

[mossy mesa]

It could be a high volume issue where a lot of users are trying to get their domains verified and there's a queue.

It could be a typo in the domain, too, because I still don't see matching records of it in what you provided.

Do the domains match?

Did you use a sub-domain?

[pallid fractal]

The domains match yes, I will double check everything again
can you show me where do you check the matching records

[mossy mesa]

Just compare the domain name in GPT with the domain name under your registrar or whatever. If they don't match you will have a looong wait (forever).

It's just one possible issue but you can easily eliminate it by comparison.

[pallid fractal]

Ok thank you

[mossy mesa]

No worries and good luck.

[sonic ruin]

this keeps happening and wont let me type a new propmt. id prefer not to start a new chat as i have this thread loaded with tons of information. any advice on what to do?

ive clicked regenerate like 5 times and the same error happens

[timber shadow]

Are new chats working fine? This sometimes happens in very long chats--for what it may be worth to you, depending on how long the chat is (especially on GPT-3.5), it's likely that you've already hit the context limit of the model. When this happens, old data is automatically forgotten to make room for new data. Which is to say, if you can copy/paste some critical context into a new chat, you should get back to a similar chat experience fairly quickly.

You might just say something like, "To start this chat, I'm going to copy/paste some messages below from a different chat to prime your context."

[sonic ruin]

okay cool ill try that! thank you alot!

[mossy mesa]

That's an interesting technique. The meta-commentary concerns me though, since the model's prone to lapse into meta-commentary anyway, but I appreciate the transparency ethos just the same.

I like to find a contextual break in the prior chat and start there.

[thin walrus]

(happy new year) sorry for the question but I take chatgpt4 and i see that the number of request is 40 for each 3hours. I can understand the limit, but what about uncorrect answer for image ? Is the 40 request for all question or only image ? thx

[mossy mesa]

If it generates an image, wrong or not, it counts.

[timber shadow]

If I ever try something like this, I usually "sandwich" the context within the meta-commentary, as you phrased it. In other words, something like:

• Message 1:
  Here's some context:
  [Context]
  More context to come in next message, don't respond yet please.
• Message 2:
  Here's some more context:
  [Context]
  Okay, context over, let's chat!

[mossy mesa]

That's good SPR technique.

[thin walrus]

ok i understand. And for the number of request it consider all request ?

[mossy mesa]

Custom GPTs cap at 25 requests. GPT-4 Turbo base model caps at 40 requests.

If your custom GPT has /commands and /help or /list or equivalent, you can do that, copy it into a new context in regular GPT-4 (see above) and usually keep working.

It will try to emulate the GPT you were using in the context of Plain-Jane ChatGPT-4 Turbo.

Provided no API calls or custom knowledge is needed. If either of those are the case it'll hallucinate like my avatar in 'Vegas.

[sonic ruin]

worked. thank you!

[carmine willow]

Hi, I'm doing my graduation work and I would like to make a smart assistant that would use my own AI , it would be built on a Raspberry Pi 4 and when a person would talk to it it would give me an answer, is it possible ?

[shell turret]

Is it better to use SudoLang for Custom on GPT instructions or Just do it Normally through Prompting?

[cloud marlin]

Help! My GPTs are suddenly not accessing their knowldge bases: "Error searching knowledge
It appears that I am unable to access the contents of the files you've uploaded, " AHHHHH!
I have done the following:

1. Deleted all files and re-uploaded.
2. Recreated the GPT from scratch.
   Nada. Nothing worked.

Seems to be happening in multiple GPTs. The one in question has 9 files uploaded (all TXT) and no actions. Worked like a champ for weeks until it didn't a couple of days ago.

Thoughts? Help? AHHHHHH!

[crisp dagger]

anyone else have issues with Dall-e ? i doesn't matter which prompt do I use it always receive and encountered issue

[serene vault]

yes

[mossy mesa]

I believe it's down.

[rigid musk]

Status.openai.com

[mossy mesa]

The GPT informed me it was down after a web search. I'm visually disabled. If it wasn't down, apologies.

[cloud marlin]

I didn't know that was a thing. Thank you!

[rigid musk]

No problem

[balmy heart]

Hi there, since a week or more I can't login to OpenAI forum anymore, both with Safari and Chrome: it just keep telling me "Could not retrieve your user detail. Do you have an active account?" but I can't reply anything. Any ideas what could be the problem? I asked to the support bot too.

[cold folio]

if you signed up with ms account / apple or google make sure you are using this method. not by trying to login directly with your email

[severe bluff]

**Custom GPT HELP!! : **2 weeks back , i was able to initiate the instructions from the files. But when i tried today. it says it can't perform.

[mossy mesa]

I hope you have backups.

[severe bluff]

I do have backups , it's the same file , but it says it's unable to take instructions from the files.

1. I created new one , still same.

[mossy mesa]

Okay, I don't know how much information technology you know, but /mnt is where all your knowledge goes.

Thee's been a huge load on the servers that /mnt lives on.

Your knowledge may be there or it may not be. It may have timed out or it may be corrupted.

YMMV, I'd update them by first remvoing all current knowledge and saving it that way.

Then update the knowledge and save again.

[woven oar]

hi! if i was added in organization with gpt4 is it possible to use it in my main account or no?

[mossy mesa]

You'll need to log in with creds from the org.

[woven oar]

the founder's login details?

[severe bluff]

/mnt/data or only /mnt ?

[trail cairn]

you can use your login but must define the organization in api calls you make, that organization will also pay for everything

[woven oar]

oh its for the api only? I mean i cant just go to chat.openai.com with the gpt4?

[trail cairn]

there is no such thing for chatgpt. organizations are for api

[woven oar]

ok got it

thanks btw

[balmy heart]

That's what I'm doing 😦

[cold folio]

can you try firefox?

[floral roost]

Anyone else having issue downloading analyse files?

[mossy mesa]

Probably /mnt/data too

everything in there.

[balmy heart]

just downloaded and done: same error. The thing is... if I delete the cache and start from scratch I got the same error.

[mossy mesa]

Then it's under heavy load and it's still timing out. Sucks. Try again. I got mine to work on six GPTs yesterday in 8 total tries.

Be patient.

[balmy heart]

Are you referring to my topic about forum login?

[mossy mesa]

I was referring to /mnt, but forums are probably hosed, too. File a ticket and answer the bot's email with a screenshot.

[balmy heart]

ok done it this morning let's hope they fix it

[mossy mesa]

Indeed.

[mossy mesa]

Something weird is going on. This wans't like this an hour ago...

As you can see I've got a few GPTs that aren't showing up in the Explore section. Under Construction signs would be so welcome if they made a return.

Ah, the 90s...

[flat bone]

Does anyone know why the account recovery is not working for ChatGPT?

[mossy mesa]

High volume and load.

[mossy mesa]

That's just a guess, but based on everything else that's been on fire today and all the requests for help across all of support...it seems highly likely.

[flat bone]

Thanks, that was helpful

[mossy mesa]

YW. Sorry I don't have official word.

[native holly]

Cross-posting here: How do I access the GPT Builder? I have a GPT Plust subscription but can't seem to find this tool when I log in to my account...

[mossy mesa]

EXPLORE link in ChatGPT.

[native holly]

Thanks Darth

[woven vortex]

Has anyone tried to email an image through their GPT? My GPT keeps passing a different URL string than the image it created. Any way to get around that?

[livid lark]

Does ChatGPT have problems making schedules? It keeps blatantly not following parameters even if I prompt it to use precise times without rounding down or up. For example, I say to leave no less than 3 hours between my last meal of the day and when I go to sleep. ChatGPT will make a schedule with meal 4 at 9:45 and bedtime at midnight.

So I try to correct for this in the next prompt. Exact same time. If it finally corrects this issue, then it actually deducts the previously over time (say 15 minutes) from another activity for the day, but that's just creating another hole to plug.

[mossy mesa]

GPT doesn't plan its output. Use better instructional guardrails.

[livid lark]

What do you mean it doesn't plan its output? I am not asking the GPT to plan its output. I am asking it to make a daily schedule. I'm not asking for GPT to output data on a schedule.

[mossy mesa]

What is a schedule? A plan. GPT doesn't have a recursive cognitive function. It's a feed-forward network. It doesn't plan its output.

Give more explicit instructions.

[livid lark]

I am asking it to make a schedule based on the recommendations that it provided with explicit instructions to not have the last meal of the day no later than 3 hours prior to sleep. It failed to produce a schedule that had this until prompted for the 3rd time. After the prompt, it then produced a schedule that no longer followed the instruction that contained the drive times. It just assumed that I would already be at the next location without driving.

I am giving it explicit instructions and it is not following them. Are you saying that this is too complex?

[mossy mesa]

Your instructions are logically inconsistent, probably. Eg, day is over at midnight. A schedule that goes further isn't "daily". You have to think heuristically, not like it has common sense and planning. It doesn't.

Come up with good guardrails and maybe an optimal example.

You have to tell it exactly how you want it to plan your schedule procedurally. Open variables and acceptable times and time ranges are your friends.

[torpid igloo]

I am trying to create a new GPT but keep getting the message "Hmm...something seems to have gone wrong." When I upload my script.js file for the model, I get the message: Unable to extract text from "script.js". This has been going on for a few days.

[mossy mesa]

Everyone is getting that lately. No official workaround.

[torpid igloo]

Ok thanks, at least I know I'm not the only one.

[livid lark]

I have done all of those in the initial instructions. Examples below:
"Work is from 8am to 5pm.
Lunch is from 11:30am to 12:30pm.
It takes 20 minutes to drive to work from home.
It takes 15 minutes to drive to the gym from work or from home."

There are some aspects of its recommendations that I do not know where to fit in the plan so I asked it to include this in the schedule either daily or weekly. Those parameters are fine.

It's mainly getting in the last meal of the day 3 hours before sleep instruction that it seems to not follow for some reason. When it does follow this instruction a previously followed instruction is no longer followed. In short, I am playing whackamole, lol.

I will attempt to get more explicit, but I don't know how much more explicit I can be without just writing the schedule myself which is what I'm trying not to do.

[mossy mesa]

Open variables and ranges for dinner. Hard cap midnight for sleep. Don't create logic bombs with impossible riddles. Be specific.

I think you can do it.

I'd use open vars and ranges for lunch, too.

[livid lark]

Oh, I gave a hard cap of must get at least 7 hours of sleep per night. It came back with a 6 hour sleep window, lmfao

[mossy mesa]

Okay

[mossy mesa]

You need to use a consistent reference frame. Specifically citing times of day and arbitrary time windows make it a story problem. For math like that make it use Code Interpreter.

LLMs a) don't plan and b) aren't good at math

In fact you should use Code Interpreter.

[livid lark]

So, it's helped me make a pretty good life goal plan. Like, really good. Even when I ask for a more specific plan it seems to generate what I'm asking for pretty well. The problem is the minute or daily planning/integration. The roadmap to the goal is excellent. Even the goals themselves are outstanding even if a bit overwhelming on paper.

But, I think I see what you're saying. The AI can't make the decision to compromise. So rather than make the decision on where to cut it seems to be waiting for me to tell it what needs to be cut even though the AI is the one that made the goal. Of course, maybe if it cut out the damn 40 minutes it has me of arriving early at work doing nothing I probably wouldn't have this problem in the first place.

[mossy mesa]

That's right.

[livid lark]

Maybe what I need to do is find a way to discover if the recommendations are reasonable before getting to the day-to-day planning?

[mossy mesa]

You can use code Interpreter for the calculations.

[livid lark]

How would I do this?

[mossy mesa]

The AI will write Python code to subtract time.

[livid lark]

Ah, gotcha.

[mossy mesa]

Tell it to use code Interpreter for calculations and checking availability.

The AI has no idea what the next token is. That's important to remember. It can't really plan its output. It can write a plan though.

[livid lark]

Is there a way to determine if the workload is too much for a person? Last week a previous version of the GPT that I used generated a pretty damn good schedule that made what looked like an overwhelming set of goals achievable af. The problem was that I didn't think it did a good job of assessing if that was what the user actually wanted. Now I've got the opposite issue.

[mossy mesa]

Guardrails. Use clarifying questions.

[livid lark]

Ah, it was deeper than that. I got that worked out though. I had to figure out how this would look in real life to get it to play out. Conversation wise that is.

[mossy mesa]

Yup. Gotta do all the planning up front. Your job is to be a meta-planning expert. You need to plan to plan a schedule, then teach the AI that. 😁

[livid lark]

The funny part is that the goals this time aren't even that hardcore compared to the earlier model yet the schedule is harder to generate because reasons? 🤣 🤣

[mossy mesa]

Yup. The word you are searching for is randomness.

It doesn't know what the next token is.

[livid lark]

So I asked it to use code interpreter, but the same issues are happening. I'll try to input the instructions from before and try again. It seems to have just copied the last schedule that it output with less words for the activities at each time, lol

[mossy mesa]

Oh

Gosh. Don't use poisoned contexts for testing!

Wow.

[livid lark]

Yeah, I thought about that, but obviously not enough. Got a lot going on here. But, hopefully can get it right this time.

[mossy mesa]

If not then take a step back. Look at different prompt strategies. Maybe it asks you to confirm each event. Maybe it's about setting constants. Without more context, I can just give general guidance.

[toxic kelp]

why is this happening

[mossy mesa]

Lol

[toxic kelp]

every time i try to start a new chat

no matter what i say, i get this:

[mossy mesa]

I don't know

[astral quest]

Yea same, it seems to be down

[livid lark]

I appreciate what answers you gave. After about 2 or 3 prompt fails I end up going from "do exactly this" to "Just do exactly what I told you to do! Arrrrrrrgh" lol

[toxic kelp]

thanks for confirming

[queen kettle]

same here

[toxic kelp]

thank you

[mossy mesa]

Oh, yes I do. Do you use a VPN?

[toxic kelp]

no

[queen kettle]

actually just started working again

[toxic kelp]

its working for me now too

[mossy mesa]

Sweet

Internet issues then

[livid lark]

I'm not having any problems. I've been failing prompts this whole time 🤣

[mossy mesa]

Big connected world with a lot of links. Each is integral to the whole.

[livid lark]

So, with code interpreter it gave me good calculations for task times, but disavowed every instruction except for sleep time. I suppose I should see which instructions it did follow and use the same format/style to attempt to generate a better schedule. Honestly, the only thing that code interpreter got right on this version was that sleep was 7 hours and the correct calculations for each task. But, that's it.

9:15pm-4:15am sleep.
5:45am-5:45am: wake up
Uh, what am I doing from 4:15am-5:45am?

[mossy mesa]

Now you are thinking heuristically. You can get this. Be methodical and patient and really look at what it's getting right for guidance in your context/use case.

[livid lark]

In this case, absolutely nothing other than calculating times, lol

And the requirement to get 7 hours of sleep

[mossy mesa]

I hate to say this but you have to bevery consistent based on what works consistently.

I developed modular GPTs so I could mass-test the logic.

[livid lark]

I believe you. It's just incredibly frustrating to be very close in some areas and not close in others every time with having to abandon the closer method because of the planning flaw. So now I'm having to make this work while also working with code interpreter. As a retrogamer, I'm no stranger to this level of frustration, but it's still a pain.

[mossy mesa]

Yep. You picked one of the hardest general tasks to dev, lol. It's doable though.

[livid lark]

It's such a minor part of this GPT which makes the comedy that much better or worse depending on your POV.

[mossy mesa]

Not to be cruel but it's a bit of both.

Better than the whole GPT being this rough.

[livid lark]

Well, in truth if I wanted to make something commercial it would be. But, for personal use it's not too bad. To make it ideal would be more work, but that's mainly from searching out quotes to give it the personality I want. Sans personality, it's not too bad.

Ah, I stand corrected. She did include 5 meals per day this time as well, lol

[mossy mesa]

She knows about second breakfast and elevensies then.

[livid lark]

I think the previous version I told the reason why I needed 5 meals per day(medical issues) and she nailed it the first time out.

[mossy mesa]

This is going to look funnier than intended but:

Five is the biggest number that I trust the AI to work with, specifically, without the Code Interpreter.

No joke.

[livid lark]

I blame my GI doc. She said I needed to eat 4-6 meals.

[mossy mesa]

If you tell her six, it'll be every three to three and a half hours, seven or eight daily, without Code Interpreter. I'm not kidding.

I'm glad you can make light of it. I'm visually disabled so I can sort of relate to a specialist being tough to keep happy. "Are you still using screens?"

🤐

Nah man dry it up I'm good. 😁

Disabled but not broken. ChatGPT retinal implant or bust.

[livid lark]

Well, this schedule got rid of my gym time which I said was a must and told me to wake up at 5:45am yet sleep from 8:35pm-3:35am. So now extra 2 hours of staring at the ceiling! And it got everything else wrong. It also got rid of the 5th meal, but made sure to schedule the last one 15 minutes before bed so I could have acid reflux all night.

Why is code interpreter making this worse? lol(rhetorical question)

[mossy mesa]

In sincerity it's not code Interpreter. You aren't being specific in a consistent format, if I had to guess. Maybe chunk the process, working iteratively if you can't structure the instructions sutably to get it to work.

You can't just list rules and engagements, it needs very firm guardrails.

ie:

Time ranges for set tasks based on necessity. These are mandatory:
Begin Day: 6:30am-7:30am / 25 minutes to get ready
Breakfast: 7:15am-8:15am. / 25 minutes to eat
Second Breakfast ...

etc.

Then you let it fit schedules in and massage specific times for set events. But you have to format the engagements the same way.

[livid lark]

That is what I spent the last 20 minutes doing.

[mossy mesa]

ok

i can't see your work, you understand.

[livid lark]

It actually had an error analyzing on the first shot. Second time it generated basically the same as before. Now, I'm getting loading issues so maybe i's time to just take a break.

[mossy mesa]

I mean GPT is pretty hosed. What do your custom instructions look like? Are they confidential?

Can DM me and send them if you want me to take a look. I'm happy to help.

[livid lark]

Yeah, I'll DM them.

[mossy mesa]

No guarantees - this is a doozy of a language model problem. I'm not sure how I'd even tackle it outright myself.

Maybe creating a knowledge file for blocks of the day to fill...

[rustic stump]

Hello, I'm developing for-profit web apps with ChatGPT API. I have a problem and I'm not quite sure how to solve it. I want to enable my users to interact with the chatbot within certain specific settings i set, but as far as I know, OpenAI charges fees as you use the API. Specifically, I want to charge users "API usage fee" + "profit amount" for a 100-character text. But the problem is that even ChatGPT itself does not know how much a 100-character text it will give will cost. As far as I know, OpenAI charges through a unit called "token", but not every 100-character text has the same token. In other words, depending on the answer given by GPT, very different prices can be paid completely by chance. This prevents me from knowing how much a user would have to pay for me to make a profit. Is there a way to keep the token value of GPT's answers constant?
I see some Micro SaaS implementing this, but they all charge monthly fees as if they were sure of everything. How can they comfortably do this without any idea how much the users' API requests will cost? Thanks

[fathom crag]

This is an ongoing issue, there´s multiple reports about this in the last 6-7 days.

[cloud marlin]

Thank you

[long crescent]

Looks like I just found my answer.

[vivid mortar]

This area of the documentation may be exactly what you are looking for:

https://platform.openai.com/docs/guides/text-generation/managing-tokens

[spare shoal]

Do I need to buy a website domain to confirm that I am a human? Just to activate the builders profile?

[astral quest]

No, you can also use your name

[spare shoal]

Says not valid

[rugged wing]

hello! regarding dall-e, why is the /daily command for getting free credits not working anymore? (message I get: The /daily command is currently unavailable. Stay tuned for announcements!)

[astral quest]

Wdym?

[spare shoal]

So I want to activate my community builder account. It says I need to use a domain

like a website that i own?

[charred parcel]

you can use your real name

[unkempt quail]

hello just wondering if anyone else has this bug chat gpt ignores custom chat gpt instruction

[spare shoal]

Does not work. It needs to be a website.

[charred parcel]

let me check that for myself

[charred parcel]

yes it does work, you can turn on your Name in the Builder Profile, no domain needed

[spare shoal]

Are we on the same website?

Chatgpt right? in settings?

[charred parcel]

We are, you should be on Settings -> Builder Profile

[spare shoal]

Says I should complete billing details

[unkempt quail]

hi guys been trying to get help from open ai customer support but no luck any alternative ways of contacting them

[rigid musk]

Unfortunately the only way help.openai.com

[unkempt quail]

thanks i tried that already and it seams there customer support is rubbish

[astral quest]

Yes, that needs to be filled out. But you can then just use your name.

[mossy mesa]

Yup.

[astral gyro]

why does chatgpt refuse to summarize the speech of the president of the italian republic for the end of the year

it says he cant help with this

or that it cannot give the summary in this context

[onyx kraken]

Not sure, you could share the conversation to get more accurate reason.

Did the AI have to search for the speech? If so, maybe the sites it picked to try and read the speech from have a robots.txt file that tells the AI it may not read them. If so, the AI will stop and obey that request.

Did you copy/paste the speech and give it to the AI, but it is still saying it can't summarize this? If so, would need to see the shared conversation to see what the problem is.

[astral gyro]

i copy pasted it

the text is kinda long but the safe chunck fort chatgpt is like 15k not the 12k of the text

[vernal hound]

Dumb question: I work for a natural resources non-profit with like thousands of old reports and documents. Would there be way to train a model built on those old pdfs where a user could query and summarize? Example "What changed between 2012 and 2013?"

[mossy mesa]

Yup. Create a custom GPT and upload them as a knowledgebase.

Then give it good guardrails on how to use them.

You're talking to your history in no time.

(Weeks or so anyway)

[vernal hound]

oh wow, thats awesome

[nocturne marsh]

my upload image disappeared on Chrome (it still shows up on iOS), does anyone have this issue?

[nova oracle]

my alpha plugins just disapeared

[mossy mesa]

Looks normal to me.

[timber shadow]

Are you a Plugins dev? If so: plugins are on the path to deprecation, to be replaced by custom GPTs, and I've seen a couple other Plugins devs report similar revocation of testing access--I think that period of alpha testing access is simply ending.

[lament pond]

I'm unable to produce a response

I get this message every time I prompt.

[mint finch]

Anyone else noticing degraded Instruction following/output with their GPTs right now?

[timber shadow]

What message are you seeing? If you tried to attach an image, I don't think it came through.

[lament pond]

I've tried restarting computer, clearing cache and cookies, logging back out and in, speaking with the virtual assistant.

[timber shadow]

Do you have a browser extension installed that affects ChatGPT? The UI looks slightly modified. If so, can you try disabling any/all such extensions and trying again?

[lament pond]

none installed. Though I will say, I tried after all that to just say "hi" and it did respond. I think it doesn't like working with some names i was pasting.

[timber shadow]

That ChatGPT icon just looks a little off to me. For example, attached are what 3.5 and 4 look like for me. As a test, can you try in an incognito window, or in a different browser?

[nova oracle]

Yeah im a plugin dev and thanks for the infos

[steady bay]

anyone knows a place where I can buy loaded vcc with paypal? low value. Just gotta need to deposit openAI balance and my vcc gets rejected

[mint finch]

It might just be the Dall-E piece as it's not creating images, which is throwing off the rest of the output. Anyone else?

[vivid mortar]

I agree that is a bit odd, could just be an extension messing with some CSS

[ancient pewter]

what the

[astral quest]

Get well soon!

[rigid musk]

no

[signal dock]

Anyone managed to verify a .org domain for the GPT builder profile in your settings ? I could verify other domain (.com .net) but can't manage to verify a .org domain... Is it a bug ? Thx

[astral quest]

I verified a .com domain. Don't have a .org domain to test that though

[unkempt quail]

is any one willing to start shaming open ai for there customer service we are paying them money and they cannot even be bothered to email us

we can start sending spam emails in bulk

[astral quest]

I don't think that will help the case and even can be illegal, depending on the methods used

[unkempt quail]

i am not sure too many people on here have been waiting for months with no reply others tell me they reply in two days

i was paying for plus and was getting replies no they have vanished for over two weeks

[astral quest]

what is the problem you are facing, if I may ask?

[unkempt quail]

sure i have 3 bugs 1 /chat gpt ignores custom chat gpt interactions 2 / chat gpt 4 refuses to give me full python code put usless place holder comments it takes 5 -10 request to get it to give full code 3 /dolle provides correct image first time then never again / the last two waste chat gpt 4 message cap

[astral quest]

Well thats nothing the support can help you with. That is really dependent on how you use the model and you are either reaching ChatGPTs limits or prompting it without enough information or guidance

[patent blade]

I've found a bug in chatgpt. I tried to create a post to report the bug in #1070006915414900886 but it won't let me create a post and an error is shown

[astral quest]

what does the error say?

[unkempt quail]

thats what i thought in the start but after getting my first email form suport they said they new about the bug refunded me asked for screen shots then never got back

[regal lynx]

Unfortunately, no one here can provide any official help when it comes to bugs 🫤

Posting in #1070006915414900886 isn't a bad idea because that's where staff tend to look on the server

[patent blade]

An error occured. please try again!

[unkempt quail]

staff never look on the discord to be fair

[unkempt quail]

best to screen shot it and send it to support maybe they will reply to you

[steel marsh]

Hey!

Does anyone know what the token format being sent in the OAuth of the CustGPT auth flow is. The token I am being sent doesn't look like a JWT as it doesn't have the "." seperating the parts.