community-help

[brazen thistle]

Anyone got the same issue?

[vast junco]

uploaded files in ChatGPT expire after a moment. (that's the simplified explanation). You need to upload it again in that case

[brazen thistle]

No, it occurs every time when I start a new conversation today.

Have tried multiple times with different pdf files, and it is keep failing.

[vast junco]

oh, I see. I'll try on my side

it works for me 😕 Are you using a custom GPT?

[brazen thistle]

nope, I am with default GPT4

[vast junco]

it the issue persist, you should contact the support by clicking on the message icon at the bottom right in the help center: https://help.openai.com/en/

[brazen thistle]

weird

[slow tiger]

I'm unable to find the memory toggle in my settings, anyone know a fix?

[timber shadow]

Hi, you might not have access yet, most don't! https://help.openai.com/en/articles/8590148-memory-faq

We are rolling out to a small portion of ChatGPT free and Plus users this week to learn how useful it is. We will share plans for broader roll out soon.

[lucid moss]

I contacted the support with the same issue and they're feeding me a bunch of bs.

[tardy dawn]

Keep getting an error at random times.. more and more frequently. Unknown error... I tried uninstalling and reinstalling.. same error.

[quartz spruce]

All day everyday lol

[brazen thistle]

@merry obsidian Sorry to bother you, but could you please help me to fix this vital issue? I have also attached the har of my chat, thank you so much
#1210853361629528157 message

[brisk warren]

Hi. I am still waiting for the Memory feature to drop. Oddly enough, about a month before it was even a big announcement I logged in and it was there in my settings with an onboarding and all that stuff. It did not work, though. Two days later it was gone. I am trying to contact OpenAI about this to see if they can help me without success yet. Any advice?

Oops. Just read some other posts and realized I'm not alone and it is slowly rolling it out

[limpid cloak]

can the developer of the custom gpt in the store "Explore GPTs" can see the conversations others make? I asked the chat gpt and it says that it can and they are not private

[white mist]

Could anyone help????

It's been bugging me for some time now

[past geyser]

Unlikely unless the developer is OpenAI

[heavy patio]

Question - I feel stupid but... I'm kind of confused on assistants. What are they for? Originally I just thought they were custom GPTs you could call programmatically, but they seem to operate vastly different with their files/knowledge than custom GPTS, and their instructions seem to change behavior in very different ways than custom GPTs.

[timber shadow]

They are like the API equivalent of a GPT, yes! While GPTs live on ChatGPT, you can build with the Assistants API anywhere you're building with the OpenAI API: https://help.openai.com/en/articles/8673914-gpts-vs-assistants
Similar, but not identical, built-in abilities.

[deft glade]

Recently, I started getting that my input is too long error, even though I was using the same prompt length I was using before that never had issues with, did anyone else had similar encounter or is it just me? if so how can I ask for help?

[heavy patio]

The 'similiar but not identical' is what breaks my brain, I think? I've built quite a few custom GPTS, if I take the simplest one and port it over to assistants, the behavior seems to be very different (prompt vs expected response output).

I probably need to actually read the link you gave. Lol. I'll do that. 😄

[timber shadow]

The behavior difference could be explained by ChatGPT/API differences, not necessarily just related to GPT/Assistants differences, if I had to guess? OpenAI has ChatGPT "tuned" (for lack of a more technical term) a certain way to curate a certain experience, and the API is more "raw", as I understand it.

[heavy patio]

Still questions, but that link helped a bit, thx ❤️ The article's statement about requiring code is true for assistants API, but I think there is a playground that lets you work with assistants in a stepping stone / training wheels / 'no code' creation path https://platform.openai.com/playground?mode=assistant.

Also, article mentions no shareability - I think there is built in shareability at the organization level now, if you attach the playgroundassistant to your org instead of your personal acct? Now I'm curious and want to go test 😅

[thorn wigeon]

Hello everyone, my account got this message and I can't do anything about it, I havent changed anything, or deleted/deactivated, I dont know what to do, I already tried to find something in help center but it didn't find anything useful

[timber shadow]

Yeah I think that's what they're referring to in the "Designed for programmatic use; can use playground for visualization" box. And true, I think interorg sharing is possible via the Playground, but is maybe more for testing than for regular use? I think Assistants are mainly used by folks who want to built it into their own, non-OpenAI platforms.

[timber shadow]

Hi! Sorry this happened. Did you get any email(s) about content policy violations or account deactivation? Were you getting any content policy flags when using ChatGPT?

As far as help.openai.com goes: you can submit a ticket there via the chat widget in the bottom-right. Open the widget, then click through the most relevant options, then you'll eventually find a path that lets you type your own message. This will submit a ticket that OpenAI support can review and get back to you on. It may take some time, possibly days. You'll get an email when they reply to you!

[deft glade]

.

[thorn wigeon]

Nope, any email and I never had any content policy flags, I have been using this account with plus since one year+

[neon birch]

Hello!
Has any of you ever had this sort of problem? I've been using GPT4 to make a lot of flashcards for Anki with specific formatting and today for some reason it gives me this kind of response (to "normal" prompt, meaning it's not the prompt used for creating flashcards). How to solve this problem? And if anyone's know where's the "help" button after log in, I'd appreciate the information (the official response is that you can see the help button after log in but there's nothing like that).

Oh, and for the record, I'm not using any VPN and similar stuff.

[eternal temple]

Hello!
What is the current usage cap for GPT-4?

[rigid musk]

40/ 3h

[eternal temple]

Thank you!

[daring imp]

guys I hope you can help me. I'm a film maker from Switzerland and would like to join the test programm of SORA. Where and how can I apply?

[charred parcel]

there's no way to apply

[daring imp]

hi robert, thanks for the reply, i really appreciate! but how are the testers chosen then?

[charred parcel]

OpenAI would have reached out to appropriate professionals. There is no further insight into that process but applications are not currently being sought.

[daring imp]

oh ok, thanks a lot again robert! have a lovely day 🙂

[limpid cloak]

[languid prawn]

What happened to CHAT GPT 4 response? I am getting worse and worse response. Not even following instructions and missing what's asked. When asked revisions other things are omitted.

[sinful mirage]

What did you try ?

[final skiff]

Hi all.

I am trying to find a workaround about my custom GPT knowledge.
I have a LOT of pdf books to upload. But the knowledge upload limit is 20.

Is there a workaround to this using the add action?

[timber shadow]

Hello! You could try combining the PDFs, just keep in mind the 512MB filesize limit and, probably more relevant, the file token limit of 2M tokens: https://help.openai.com/en/articles/8555545-file-uploads-faq

[final skiff]

What is the "file token limit", I mean what does that mean?

The custom GPT will stay private as it’s for private use

But I haven’t get yet what the 2M tokens limit means on a file.

My pdf’s are all books (images and text)

[obtuse oxide]

Are there any advantages of using an assistant in your code over normal gpt api if both run of the same model?

[timber shadow]

Tokens are the basic units of LLMs. There's not a constant conversion of words to tokens, but you can find estimations to get you a good idea here: https://help.openai.com/en/articles/4936856-what-are-tokens-and-how-to-count-them

Also from the first link, keep this in mind regarding embedded images:

Images embedded in documents/presentations (i.e. in image in the slide of a slide deck) are not supported yet. We plan to add support for this in the future.

[tame sparrow]

Anyone else experiencing this same error - #openai-chatter message - ‘Bad URL’ and will not let me login on the apps? I’m getting that error right now when I attempt to login with either my iPad and iPhone ChatGPT apps; not sure if it matters but I’m on the + subscription. If anyone can help I’d appreciate it, please ping me so I don’t miss any responses. Thank you 🙂

[final skiff]

Thanks for this info! I didn’t know 😅
Then I’ll calculate how many books can fit within a merged pdf

[final skiff]

Also I was wondering, if I use an API, external service to host the pdf books content and then link it to my custom GPT using add actions. Would it be performant as directly uploading pdf into custom GPT knowledge ?

Or just uploading files directly into GPT knowledge is the most performant way to get accuracy responses. (Even with its files limitations)

[final skiff]

Also would it be the same performance as knowledge if I upload for example 5 pdf books as a zip.file into my GPT knowledge ?

[vast junco]

zip files can be unzipped via code interpreter. but it's not the same as retrieval, where the GPT can read directly the content and turn it into text (which is treated like text input, as part of the context). I will verify if there are specific documentation about this

Retrieval Augmented Generation (RAG) and Semantic Search for GPTs

[final skiff]

Thanks!

[vast junco]

That article seem to be more aligned with your question: Knowledge in GPTs
It talks about the files you can upload

[final skiff]

Oh it doesn’t process images, sad, I’ve thought it processed it. Hope in later updates, it’ll be able to!

I’ve uploaded a pdf where I’ve merged 6 pdf books for a total of around 350mb. It’s taking eternity to upload…

[vast junco]

it technically can, but it's tricky to automatize. And it require multiple messages, where the GPT output an image, then (if I remember correctly) it can see it using vision. But I would need to perform the tests again to confirm.

long story short, it doesn't do it out of the box easily

[final skiff]

Oh! Please could you show me the trick?

[vast junco]

I'll try quickly to do it with a simple document. might take some time to get the result, but I'm interested as well to see what it can do

[final skiff]

Any ideas about that?

It’s still loading the pdf, time is so long

It’s 178mb, weird it takes so much time

Yesterday, I’ve managed to upload 450mb

[vast junco]

I never tried to upload large files on GPTs, sorry. I assume it's taking a while because the servers has an upload limit maybe. Don't take my word for it, I'm just guessing 😅

[onyx kraken]

When I was messing with uploads, was all small ones. They sometimes hung and took forever, I think that indicates a problem.

I would cancel and start over.

There were occasionally times when uploads were not working, try again later.

[final skiff]

Started on a new tab, works now, seems my previous tab was open for too long 😅

[charred parcel]

I have been building a GPT today and PDF uploads have been working well

[final skiff]

Can’t wait to see the trick on how to make your custom GPT process images in a pdf file. As the actual images from the pdf are Artworks helping to define Art knowledge.

Also I hope soon, open AI will increase a bit the files limit

[vast junco]

I think we can already do a lot even with the current limits. We have to remember a larger context also means a longer processing time. So larger context isn't always the answer 😉

[vast junco]

it can take a while for me to find a reliable way. Some very specific prompt engineering is required for that kind of task.

[final skiff]

When it’s several books into one pdf, should I make an index with bookmarks to show GPT that there’s this and this book within the pdf file?

[final skiff]

Sure! I am very curious!
Don’t hesitate to ping me

[vast junco]

also, a better channel to continue the discussion will be #gpt-models. It's more specific to the GPT models capabilities

[pulsar belfry]

Hi, everyone!
Has anyone encountered a problem where it is impossible to send a message. You write it, press enter or the “send” button and then nothing happens. It is impossible to either start or continue an existing dialogue.
This only happens on PC, everything works fine in the app on the phone.
The browser (Google Chrome) was updated, the cache was cleared, all extensions were blocked, the computer was restarted, nothing helps

[sterile rivet]

Hi. I'm experiencing an issue with ChatGPT. Every time I ask it a prompt it makes me do a 20 impossible captcha that is extremely difficult. I find this so annoying, because I pay for this service, and it takes 10 min each time to do the captcha based on one single prompt. I don't understand why this is happening, and I can't access support / email for any help. Has anyone run into this?

[onyx kraken]

Sorry this is happening to you. I recommend checking the recommendations here https://discord.com/channels/974519864045756446/1189951364449718373 and maybe adding your voice so OAI can get a sense of how many users are affected; seems a new problem (again) as of today for some folks (I'm not affected myself)

[rapid lotus]

Hello I am stuck in a loop of 20 stage captchas. When I complete one successfully as soon as i type another questions it prompts me for another 20 stage captcha. Its an endless loop.
Anyone else experiencing this?

[rapid lotus]

This is happening to me also.

[cold folio]

are you using VPN/proxy ?

[rapid lotus]

No sorry I moved my chat to the thread. Same systems and connection at home office I use everyday. Tried incognito browser also.

[untold trout]

Hey - Does anyone know of an easy way to export GPT transcripts as PDF?

[vast junco]

Here's a method I think might be easy and good enough for simple cases: Click the "share" button at the top right. It will give you a link, you can follow the link and save the page. (I hoped Ctrl+P to "print" would let us save the full pages into PDF, but it only saves the current visible portion)

Once you have the page saved, you can surely convert it easily to PDF. note that this method doesn't include images and message edits.

[shell warren]

Hey folks. We've recently setup teams (150 licenses) but only added 20 seats and already finding it extremely slow compared to personal plus plan.

Is this a common issue - together with receiving server errors?

[tame sparrow]

Can someone please help me? I’m still unable to login to either iPad or iPhone ChatGPT account on the app? I keep getting an error message about a “Bad URL”.

Anyone else experiencing this same error - #openai-chatter message - ‘Bad URL’ and will not let me login on the apps? I’m getting that error right now when I attempt to login with either my iPad and iPhone ChatGPT apps; not sure if it matters but I’m on the + subscription. If anyone can help I’d appreciate it, please ping me so I don’t miss any responses. Thank you 🙂

[twin heron]

At least I am not the only one with this issue. You didn't get it fixed did you? It has made my account unusable.

[rapid lotus]

I'm still locked out.

[twin heron]

Which I am not using a Proxy, VPN, and I have tried in incognito mode, clear browser cookies, tried different browsers, but it does work on my phone at least.

[twin heron]

Well that's annoying. I use this for work.

[zealous garnet]

HELP!!! I'm getting a 20 questions puzzle for each and every question I ask ChatGPT today!

[twin heron]

Same.

[twin heron]

If you don't mind me asking. GPT Plus?

[zealous garnet]

Yes, I'm on the paid version. 4.0

It's been unbearable. My husband literally just gave himself a 20 minute time out, he was so frustrated. It has slowed everything down. I'm ready to cancel and just google stuff.

[twin heron]

Same. I use it for helping edit writing, and grammar check. XD Might as well go back to grammarly. But that is just for fun. I also use this in the classroom, and if I can't use this with my students tomorrow. Then I am up a creek.

I mean I can teach without it, but it helps a lot.

[zealous garnet]

I get it. It saves a ton of time. I create content, and sales copy. This whole day has been a sludge fest of feeling like I'm walking through mud.

[twin heron]

What's really bad, is how absurd the captcha is. It might be bearable if it was one or two images, but twenty?

[zealous garnet]

Precisely! 20 fricking puzzles? And you don't get one right, you have to do the whole thing again. It's so bad .

[twin heron]

Yeah, and if you click through it quickly, you could mess up. Talk about overkill.

[polar sage]

I am locked out of GPT after receiving a series of general email warnings that say I was violating policies, without any reference to when, or how, and no examples. Reviewing my light and ordinary traffic, I can't see where I've done anything to violate or challenge any policies. Is there a way to understand what I did, and a way to undo it?

[polar sage]

Also- I have asked GPT Help for explanations, examples, times, etc of violations- but no response. Being locked out, I can no longer review my record. I need this capability to compete in my job, so this is serious trouble for me.

[echo mirage]

me too

[echo mirage]

I don't even know where to ask GPT customer service to help me.

[winter shore]

i am also being told to solve 20 captcha puzzles for every. single. prompt

i have the paid membership and tried it on gpt 3.5 same thing

[onyx kraken]

Hey, would you be willing to check out this related bug report and maybe add the requested screenshot from your problem to it? Gold names work for OpenAI, that's an engineer telling us how to help him fix the problem, if you're interested and able. #1189951364449718373 message

[onyx kraken]

Kronos, you too, if you'd look at my message above and consider helping them collect the data they asked for, so they can more easily fix.

[twin heron]

Thanks.

[eternal shell]

Anyone know how to solve the human Verification Puzzles that come up every prompt? This is quite annoying to deal with as there are 20 images to go through. I am a paying memeber and have never dealt with this issue before - also none of the picture match the left side.

[shell warren]

Anyone know if dedicated channels for users on Team or Enterprise plans?

[vast junco]

not that I'm aware of. #chatgpt-discussions is the closest

[wispy needle]

is there any way that i can integrate my customGPT built on openAI to a wordpress site?

[vast junco]

there is no widget embedding implemented in ChatGPT, as far as I know. That might be a good idea to suggest in #1070006151938314300. GPTs are not available through the API neither, to my knowledge. So the only way to add your custom GPT to your website is by sharing the sharing link on the website.

[vast junco]

I can imagine something like they do on youtube with the "embed" sharing. That could be a nice feature for chatGPT. But that kind of feature takes a long time to develop. (feel free to use that picture as a reference if you choose to suggest the feature. Follow the guideline from that post for clarity: #1070196011697442886 message )

[wispy needle]

But how would it be different from Assistants API?

[vast junco]

You can achieve something similar with the Assistant API, that's right. There would be a few difference between an actual ChatGPT embed and a user interface built specifically by you, using the API under the hood. It require more development and you would need to pay for all tokens used by the users on your website.
On the other hand, an official embedding from OpenAI would offer those advantages:

1. an embed would present directly the official ChatGPT website as its content, so the user could chat to your GPT directly using his account (it's like a mini browser inside the webpage, pretty much)
2. It would benefit from the browser securities for the iframe. Modern browsers are protecting the content of an embed, making it hidden from outside. That's a security for the users.
3. It wouldn't cost per token usage like the Assistants API.

[mystic lily]

Hello everyone! I'm having few doubts about a problem I'm trying to solve. This is the problem statement and the data structure. Would appreciate any help! Thanks

[dusk obsidian]

I constantly get this error when i try to access playground:

what can i do

[rigid musk]

Did you test another browser

[placid sundial]

Trying to update my company card because of new company, keeps getting rejected?

[tame depot]

Hello !
So I have a Mac laptop from 2017 (MacOS Monterey 12.7.1). I have installed Python 3, Pytorch, ffmpeg, pip, and whisper via pip. Every one of these installation processes seem to have worked well and have been fully complted. But when I lauch "whisper [path/to/audio/file.mp3]" it says "zsh: permission denied: whisper". I have tried to run my terminal with root permission, but even then it says "-sh: whisper: command not found". Can anyone help please?

[timber shadow]

Hi! I'm not sure myself, but just a heads up in case you don't hear back from someone else: you might also try in the #gpt-realtime channel, or the more general #dev-chat channel.

[slate pike]

Is there a way to use a custom public GPT via API?

[timber shadow]

GPTs live only on ChatGPT, and they can't be used via API, at least not directly. On the API, the correlate to GPTs is the Assistants API. Here's a comparison article on the two: https://help.openai.com/en/articles/8673914-gpts-vs-assistants

[slate pike]

Hey thanks for getting back to me.

Yeah i'm aware of the assistants and their use case.

But more specifically I found a custom GPT that I want to run a script with.

I could do it in browser but thats annoying and slow so wanted to do it with an api call if possible

[timber shadow]

Ahh I gotcha. Yes, to my knowledge, that's just not possible currently. Perhaps you could make yourself an Assistants version that works similarly to the GPT, or if the GPT is using some API itself, maybe you could just build your own implementation of that API integration? Just speculating!

[slate pike]

Thanks for the help!

[twilit summit]

Can anyone help? I am getting the "Our systems have detected unusual activity from your system. Please try again later." response to prompts.

[timber shadow]

Hello! What browser(s) is it happening in?

[twilit summit]

Hello! Happens in Chrome. Haven't tried any others yet. Doesn't happen when I use the app on my phone, though.

[timber shadow]

I see! Can you try in an incognito window of Chrome with all your browser extensions disabled as a test?

[twilit summit]

Looks like it works in Incognito!

[timber shadow]

Okay! It seems like it might be an extension, or perhaps a cookie/privacy setting in Chrome, that's causing the issue. Maybe for the next step, try clearing your browser cache, then disable all extensions in a normal window. See if it still works. If it does, start enabling your extensions one at a time, and refresh ChatGPT after each one. You might find the one that's breaking ChatGPT this way.

[twilit summit]

Clearing the caches worked. My existence once again feels complete thanks to you.

[dire ruin]

i need technical help

[timber shadow]

Hello! What do you need help with?

[dire ruin]

my account , i think thats my accounts has been logged in from severals ( multi browsers )

[timber shadow]

If you think there has been unauthorized access to your account, you should change your password. This will log out all other sessions. There's more info in this help article: https://help.openai.com/en/articles/8485685-i-m-seeing-unrecognized-activity-on-my-openai-account

[dire ruin]

Thanx

[frank nest]

Am I allowed to place a link in here asking for help for a forum question I posted?

If so - I'm totally stumped. - https://community.openai.com/t/image-request-to-dall-e-failing-with-server-disconnected/655555

[vast junco]

great article, I'll share that link instead from now on (instead of explaining all of this myself 😅) @wispy needle for viz, you might want to look at this article for your usecase. It says pretty much the same things I mentioned and contains more info.

[onyx kraken]

Hey, because you're responsible for what your account does, in addition to changing your password, I recommend you also contact OAI through their help bot, https://help.openai.com/en/ bottom right corner, click through the choices until you can leave a message for a human, and explain the situation. This is potentially helpful should the people-not-you have done stuff that's disallowed content, it helps OAI make a better-informed decision about what to do.

[rugged wing]

Is it with the Assistants API possible to build an AI assistant which only uses the custom knowledge based (i.e. solely the files provides)?

[polar solar]

anyone having trouble loading chatgpt rn?

[finite void]

Yea it's not loading for me

[rain sparrow]

I thought i was the only one having the issue I'm trying to buy DALL-E and GPT-4 and its not allowing, the site bugged right now still for everyone?

[vivid mortar]

You cannot digest files directly through the assistants API. Your best bet for having it response with ONLY custom knowledge would be fine tuning https://platform.openai.com/docs/guides/fine-tuning

[rugged wing]

Thanks. Is it a bad idea to upload entire ebooks? Its quite expensive since for every answer it readsthe entire books as input.. I would love to have an AI assistant which doesnt need to read the files/book eveytime but I guess this is not possible or is ir via fine tuning?

[vivid mortar]

Yes, finetuning on specific ebooks would prevent the need for pasting in the entire book as an input each time, but fine-tuning may also be expensive depending on how many you have

[rugged wing]

But it could still be the case that the AI doesnt consider the books to answer a question even if I fine tuned right? There is no instruction to prevent answering question outside of the files provided.. I tried to define such a instruction but the AI still uses other knowledge bases

But I can specify the topics but not prevent using other knowledge bases as general via the instruction option..this is my experience

[vivid mortar]

It sounds like for your use case you want to use something known as "RAG", and while its not available directly through the OpenAI api, i think that will put you on the right track for what you're searching for

Its a much more intense solution but for drawing context directly from your files without using many other knowledge bases, that is the current best method I am aware of

[rugged wing]

Thanks I will have a look!

[cosmic fjord]

Hey guys, Chatgpt has not been working for me for 3 weeks on any computer device. Could anyone help me out? I used different updated browsers like Firefox, Chrome, Edge, Safari, and Brave but nothing works… Also I logged out, cleared caches and cookies, made openAi as exception, sent email to support, etc but nothing has worked.

The only way I can get Chatgpt to work is through my phone (app) or on a VPN. I don’t think my IP is banned and I also checked my access point settings (Ubiquiti U6 Pro) if it could be a firewall issue but the problem still persists.

[proper onyx]

Hey I’m just getting my project started and I want to learn about what type of AWS instances I need to run in order to effectively run via the vision preview model. Is there any documentation on this? Do I need a GPU?

[vast junco]

@vivid mortar @rugged wing That article will probably help for the usecase you are discussing: Retrieval Augmented Generation (RAG) and Semantic Search for GPTs

follow-up reading: Knowledge in GPTs

[woven badge]

Hello, I'm trying to enable and account of mine that got blocked because I wasn't able to prove that I'm over the age of 18.
I want to do it now since I simply forgot to do it earlier but I can't understand how to do it

[timber shadow]

Hey! The #ai-discussions channel is the best place to discuss non-OpenAI things 🙂

[honest cairn]

sorry wrong server

[crude timber]

i have a question , i'm trying to change my password from the acc that as chatGPT 4.0 but it's not sending me the email
also somehow saying that i can't login into my acc because of various trys , since i just logged of to change the password and now trying to login ...

[timber shadow]

Hello, do you sign in with an email + password, or do you use a Google/Microsoft/Apple account to log in?

[crude timber]

email and password

[timber shadow]

Okay, is it sending the password reset email to your junk/spam folder by chance?

[crude timber]

no , i already checked that

[timber shadow]

Dang, that's strange. Perhaps it will work later, maybe there's a temporary bug? For now you could submit a bug report in #1070006915414900886 -- just be sure to follow the bug report format in the pinned post in that channel. You can also try reaching out to OpenAI support at help.openai.com. Open the chat widget in the bottom-right, click through the most relevant options, then you'll eventually find a path that lets you type your own message. This will submit a ticket that OpenAI support can review and get back to you on. It may take some time, possibly days. You'll get an email when they reply to you!

[umbral swallow]

why do i get the error "network error" in the middle of the message all the time?
does anyone else have this?

[royal adder]

Hello, been having this issue for like 3 weeks. The issue occurs when trying to send a message to the web version of chatgpt. When sending the message it is not sent, but in the console it is said that(on every attempt to send)
Uncaught (in promise) Arkose script failed to load
Had this issue on 2 different accounts, only in web version
This issue seem to occure to other people
https://community.openai.com/t/unhandled-promise-rejection-arkose-script-failed-to-load/374690
Can't use chatgpt normally because of this, mobile and api is ok

[grand whale]

Hello, despite having paid on February 22, 2024, I do not have access to the ChatGPT Plus experience. I reached out to the help center for assistance, but unfortunately, the support team couldn’t resolve my issue. Has anyone else encountered a similar problem, and if so, how long did it take to get it resolved?

[timber shadow]

Hello, can you navigate to https://chat.openai.com/#pricing on the web version? If so, do you have the option to click on a link that says "Manage my subscription"?

[grand whale]

it shows prices modals, But I paid via google playstore and have subscription untill march 22, when I am trying to upgrade via app, I have an error "you are already subscribed"

[timber shadow]

I see! Can you go to the Play Store, then tap your profile picture in the top-right corner next to the search bar, then tap Payments & subscriptions, then Subscriptions? Is your Plus subscription listed there?

[grand whale]

yes. it's there.

[slender island]

i dont think its needed , somehow chat is working well now

if i see that the bug happens agaiin i report it

[timber shadow]

I think the advice you got from robert in the other channel you posted this in is good. Support may take some time to get back to you!

You can also try posting in #1070006915414900886 -- be sure to follow the bug report format in the pinned post in that channel. OpenAI team members are periodically present in that channel.

[strong kite]

Hello, I would like to ask for you help. I need to find someone to talk to from the support. I need help with getting up in usage tier because I will need high volume of requests.
Can someone help? 🙂

[timber shadow]

Hello, the current self-serve plans available are Plus (40/3hrs) and Team (100/3hrs, minimum of two seats). There is also an Enterprise tier, which I last read secondhand is currently only extended to orgs with 150+ seats. There is also the option to use GPT-4 via the API for a pay-as-you-go model. Note that this is not on ChatGPT but rather platform.openai.com, but you are not restricted by usage caps like you are on ChatGPT.

[strong kite]

We are currently using API model that is pay as you go but current tier is 3 and we will need to bump it up to usage tier 5

[timber shadow]

Ahh I see, sorry for misunderstanding! Have you met the requirements listed on this page for tier 5? https://platform.openai.com/docs/guides/rate-limits/usage-tiers

[strong kite]

Main reason is that we need to get access to more requests with gpt-4-vision-preview.

Unfortunately no

But we want to jump to the tier 5 if there is any chance of that

[timber shadow]

I believe you will need to meet the listed qualifications to access tier 5. OpenAI support may be able to provide further information at help.openai.com though. Open the chat widget in the bottom-right, click through the most relevant options, then you'll eventually find a path that lets you type your own message. This will submit a ticket that OpenAI support can review and get back to you on. It may take some time, possibly days. You'll get an email when they reply to you!

[strong kite]

Okay thank you so much for help! 🙂

[distant horizon]

Hello, any idea of why this might be happening?

[rigid musk]

Try to delete cookie and cache, log off try again

[distant horizon]

k will try again thx

[timid cloak]

Hi, I'm new here and I'm having a nightmare with chat plus. Live Chat Support seems to be a waste of time and keep giving me a cut and paste answer which isn't working, but they just keep giving me the same cut and paste answer every single time. I keep having errors which forces me to log out because the window won't open. But now today I can't upload files that I've uploaded multiple times before - it's saying that the word documents are empty when I know they aren't. I've tried other word files which I've also uploaded before and it says they are empty, I've tried uploading a pdf and still the same error. It's completely ignoring my prompts and doing it's own thing. I can't tell you how many prompts I waste with all this logging out/logging in uploading files so that chat has context, then after a while i get another error and have to do it all again. Is there anyone who can please offer any help or suggestions please. I've tried using Chrome, Chrome Incognito and Firefox, I've deleted cookies and cache multiple times, I don't have any cookie blockers, I'm not running it through a VPN either. Thank you 🙏🏻 As you might have guessed, I'm pretty desperate

[elder wigeon]

Hi, anyone else has problems when uploading PDF/Word files? It doesnt analyse and you need to brute force it to start analysing the file (3-4 tries at least..)

[timid cloak]

I am, it's been a nightmare today

[light shard]

Has anybody's context window token limit also decreased? this has never happened to me

[hollow oasis]

yeah me too. I'm having trouble uploading word or pdf files. it says error reading documents

[rotund ibex]

Me as well.

I keep attempting to attach multiple files to a prompt, but I keep receiving error messages such as "It appears there was an issue accessing the content of both files you uploaded. This may be due to a technical glitch or an error in the file upload process. Could you please try uploading the files again? This will enable me to provide the detailed response you're looking for based on your documentation." I've tried .pdfs, .docs and .txts, but still receive the same error message.

[floral pendant]

Same here. I’m struggling to upload a .txt file that I’ve successfully uploaded previously.

[versed swift]

The custom gpts can have up to 8000 characters of instructions but the gpt is unable to update it, always saying "It seems there was an error when trying to update the behavior" ... anyone know a solution ?

[crystal gyro]

Since yesterday several problems popped up: 1) New chats respond with completely unrelated responses but relevant to other chats that were done before (seems fixed) 2) file uploads are borked, image uploads no longer get processed by OCR and errors occur 3) quote links for references are erroring and can't display

[light shard]

4. smaller context window token limit

[charred kernel]

hello, i am having problems uploading pdf files on chat gpt 4. it keep showing error messages and saying there is no content available. i dont know what to do because i have been uploading same types of pdf before and its been working but all of the sudden it stopped working today. any advice?

[loud crest]

They changed GPT-4 multimodal a bit

Lots of 🐜

[charred kernel]

what does that mean?

[loud crest]

ChatGPT GPT-4 was updated

[charred kernel]

ohhh

[loud crest]

Honestly

I don't know why it's so hush hush

[charred kernel]

so will it not be able to read pdf files or is this a bug right now?

yeah i had no idea i was trying to search stuff up about it

[rugged wing]

Why do I have to scroll to get down the text line.

It is like this for 2 weeks now, and when I type large amounts of input, it keeps the text bar non fixed.

I have to scroll within the page to get down my text input box.
Seems quite unhandy for me.....

[rocky pendant]

not sure how comfortable you are updating your page styles, but this CSS should fix it:

#prompt-textarea { 
  max-height:400px; // or your preferred height
  overflow-y: scroll !important;
}

[rocky pendant]

this will make the text area a fixed height and scrollable:

[rugged wing]

Thanks!

[timid cloak]

Hi, can you tell me if Support actually man the livechat/messenger window on the help page or the emails please

[rigid musk]

Hi, help.openai.com is the official support. Bottom right you have a chat

You need to answer questions, at the end chat you can submit your request by email

[timid cloak]

Thank you ☺️ Thats what Im using but rarely get a response, and they take forever to email you back, and then its only the same generic cut and paste answers they give every time

[rigid musk]

Sometime openai employee do answer in #1070006915414900886 when the person fill correctly

[timid cloak]

Thank you ☺️ I'll give that a try and cross my fingers because they dont seem to take notice of the chat or email x

[polar solar]

anyone suddenly getting a bunch of errors?

Everything was working normally 5 minutes ago now every response i get is "hmm something seems to have gone wrong"

[zenith shore]

no assistant api call is working today, any idea what is the issue? there is no help with support, they just route you to report a bug. i get a message each time that: "there was an issue fetching items. an error occured. etc.." Thanks

same message like this each time

[edgy orchid]

I'm using gpt-3.5-turbo-0613 and have been really happy with the results. I have a project where I am interfacing sensors to the model but in the last two days (more or less) I've been getting a change in the assistant I have created. Something about boundries.... Ugh. Where can I find more detailed information about updates to this model?

[trail cairn]

assistants is still in beta and in the screenshot you are using the playground which has no guaranteed uptime and is subject to stealth non qa tested deployments

[vast coyote]

Hello, i have ChatGPT Teams, and recently me and my colleague have issues with it.
We keep getting the error message "Hmm..something seems to have gone wrong", and it's getting really annoying because sometimes even by clicking regenerate 10 times it wouldn't work, like in this picture. The message is not complex, nor we did hit the 100 MSG limit. It is not a connectivity problem, as loading between chats is pretty fast (even with the browser's cache completely deleted), and creating a new chat is the only "limited workaround" we've found. 5-6 messages on the same chat and ChatGPT "locks" itself behind that error message.
What can we do ?

[charred parcel]

do you have a product called NordVPN installed on your computer?

[vast coyote]

No VPN nor Proxy.

[strong fractal]

Hey all! I have a TOS question. I work at a small academic library, and I’m wondering if a program I’m envisioning is in line with OpenAI’s use policy.

At my small academic library, we’ve been having discussions about the equity gap created by unequal access to tools like GPT Plus, and this conversation has evolved into the topic of providing all of our students access through a library account. Here is how it would work:

Students would come to the desk and “check out” the account for up to 3 hours, giving them the ability to use the allotted messages for that time period. They would sit at a computer designated for GPT, and a member of the staff would log them in. Students would never be given the password. Once they were done, they would be logged out by staff.

There are a lot of logistical issues that we can solve on our own, but the one I am asking you all about is whether or not this is in violation of OpenAI’s TOS. Are we allowed to let multiple people have controlled, supervised access like this? I would appreciate any insight this community might have on this topic. Thank you!

[timber shadow]

Hey Tanner! Very cool idea. Let me do a little looking at the TOS to see if I can find any clarity on this, but my first thought would be that account sharing is fine, (see Esk below) but that it's ultimately the original registrant's responsibility when it comes to the content input and output during use. In other words: if one person does something against TOS, that could threaten a ban for the whole account and therefore limit access for the entire account share program.

Otherwise, one logistical limitation that comes to mind is the 3hr chunks you describe. Right now, the usage cap works on a rolling basis--it doesn't fully reset every 3 hours. This means that if someone uses 40 GPT-4 uses in 3 hours, and if someone else comes to use it for 3 hours immediately after, that second person will need to wait (potentially) for each of their 40 uses, as each use "comes back" to an account individually, 3hrs after used originally.

[onyx kraken]

https://openai.com/policies/terms-of-use

Yeah, sadly, OAI expressly wants to know who is using an account, and wants one user per account.

Even if there's API calls being made, they want the user to have their OAI account, API key from that account, and it all up front and clear who is doing what.

They require a few things, like an email that isn't linked to a bunch of accounts, a mobile number that isn't used for many OAI accounts, and likely other forms of 'unique and identifiable, verifiable users' stuff.

"Registration. You must provide accurate and complete information to register for an account to use our Services. You may not share your account credentials or make your account available to anyone else and are responsible for all activities that occur under your account."

You may want to directly talk to OAI about ways this may be possible to do in some fashion, to some degree, but it would be some sort of special 'unusual' thing they do, it wouldn't be a normal account.

Perhaps use the bot bottom right of this help page to click through the options and leave a message for a human, to see if there's a way to get that conversation started. https://help.openai.com/en/

Wishing you the best!

[onyx kraken]

You might instead want to do this, not through ChatGPT+, but through some sort of playground/API account, where something like a discord bot is available for users to talk to.

There's 'moderation software' that OAI makes available so that disallowed content gets (mostly) stopped and you're not considered at fault if you are using the API moderation calls correctly, even if a student messes up.

But Discord bots that use GPT-3.5 or GPT-4 exist that effectively do similar to this.

It's not ChatGPT specifically, and would take someone who can make a bot, understands the API calls, and would not be a flat $20/month, it would be based on the actual token cost of each input/output.

If you want more info about that, can look over what the playground allows and empowers here:

https://platform.openai.com/overview

https://platform.openai.com/docs/introduction

You might want to google (I can't share non-openai links here) 'openai discord bot' for more information as well.

[onyx kraken]

Additionally, here's a discussion by some educators about options they have for Azure/Microsoft Education Hub, which is in some areas very closely affiliated with OpenAI. But Microsoft offers financial assistance for students in some cases. I don't think this discusses non-OpenAI models, but you might find some of the ideas here useful and related to your goals.

https://community.openai.com/t/api-for-educational-institutions/657081/2

Don't let that short statement discourage you; there's discussion in that link, multiple ideas and options discussed, links for more help.

[onyx kraken]

This is also an option to explore - again, it is not chatGPT itself. But it is playground/API and how to add other people to your account, so they can generate and use API keys.

ChatGPT-4 is a specialized type of GPT-4, and much of what can be done with ChatGPT can be done through GPT-4 and API in the playground.

https://help.openai.com/en/articles/4936812-how-do-i-add-change-or-remove-members-on-my-openai-api-account

[onyx kraken]

Oh, and keep in mind, none of us answering you work for OpenAI. We're community members like yourself doing our best to help answer questions, share knowledge, provide real resources - but none of us have the final say. If you need or want an official answer, definitely do contact OAI through the help webpage's bot to get the conversation started. That's the official and standard way to start a conversation with them.

https://help.openai.com/en/

[strong fractal]

Thanks everyone! I’ll take a look at all those links. I did try using the help bot earlier, but unfortunately without a way to actually type in my question it wasn’t possible to ask it my question. I have some more info to work with now thanks to you guys though, so thanks!

[timber shadow]

The chat widget at help.openai.com has a few default automatic options, but there are certain paths through those defaults that should get you to an option to type in your own message -- you might have to try a few before you find one! But this will submit a ticket that OpenAI team members can review and get back to you on -- it may take some time, perhaps days or more! But you'll get an email notification when they reply.

[strong fractal]

Oh nice! Yeah, when I did it earlier it just took me to a dead end, but I’ll keep playing with it until I find a text field.

[dry cloak]

Hey, I experiencing a recurring issue generating images that are downloadable. When I click the down arrow to download, the file is formatted WEBP instead of PNG. However, ChatGPT tells me it's in that format, I respond letting it know that it's not. Then it generates a link that doesn't work. I keep getting "File Not Found". Any recommendations would be greatly appreciated. Thanks!

[past geyser]

This is an acknowledged issue and hopefully will be resolved soon 🤞

FYI webp files can be easily renamed to jpg and then opened to re-save as PNG

[dry cloak]

Thanks! I've been experiencing the issue for a month now. I used a free conversion tool this morning.

[loud crest]

What is a stealth nonqatest

[trail cairn]

means the openai devs can put changes there without quality assurance checking em = can be broken for days sometimes. its not supposed to be used for real work

[loud crest]

You do not want to do this.

Instead, make some sort of portal or online-space associated with the library, where students can interact with one assistant via api.

[fallow snow]

That hasn’t been the case for me

It’s the 40 comes back after the first message

So the 3 hour window starts after the first message

That’s how you can potentially miss out on messages if you don’t start a window every 3 hours

[loud crest]

The chat should be public; to the library community, and it should be monitored by a few staff members.

[loud crest]

There should be just one chat feed that users have to sign into, with their unique identifiers so that their chats can be traced back to them.

Moderators can remove messages, and revoke access to the chat/ assistant. The chat can have its own logic to combat sensitive topics, Additional logic can be used to assess the honesty of api requests, and the moderation endpoint can be used twice- outbound and inbound.

[fallow snow]

Ping? 👀

[loud crest]

You start with 40 messages that you can send, and after you use one, you get it back every 4.5 minutes.

Right?

[fallow snow]

It’s not been like that in my experience

I could equally distribute usage over 3 hours but the next 3 hour window I could rush through it in an single hour

[timber shadow]

I'm pretty sure it's always been a rolling cap, unless something has changed, but it could be tested.

• Use it 40 times in 3 hours, 4.5 minutes between each message, then
• Try to use it 40 times in like 1 hour, after that first 3hr period of evenly-distributed use.
  I'm pretty sure you'll have to wait for a 3hr timer on all of the 40 from the first 3hr window, I don't think you get "reset" points.

[loud crest]

You are suggesting, that you can't indefinitely send a message every 5 minutes?

xD lmao I must be missing some algebra

[loud crest]

What lolll

[timber shadow]

It "regenerates" like a mana bar in a video game: depletion per use, set recharge rate. It doesn't recharge all at once.

[loud crest]

I surrender!!!

[onyx kraken]

My experiments suggest that each message is timed.

So wake up, it's 5 AM; send my first message of the day, I get it back at 8 AM. I still have the 39 other send any time messages.

At 5:02, send my next message; I get that one back at 8:02 - and still have 38 use at any time messages.

[fallow snow]

?

Haven’t really been the case, if that was the case I would getting the usage message a lot more often

[loud crest]

So basically, if a terrarium milkshake were to grow 40 arms with hands equidistant from each other like a pinwheel and it were to roll down a hill at a rate of 40 arms over 3 hours, the first hand that touched the ground would touch the ground again after 3 hours.

[quiet granite]

Is this the right place to ask for help about chatGPT?

[onyx kraken]

One of several! Depending on the exact help, there could be a 'best' spot for your exact question, but here's fine!

[quiet granite]

Hiya! I'm a bit frustrated. I just payed for ChatGPT + . Considering it's quite expensive and it's not in the description for the subscription that there is a cap on messages? Is this an error?

[onyx kraken]

So, there's a set of caps for ChatGPT+

Compared to free:

Both have access to 3.5, but with + 3.5 will answer faster and you have 3x more tokens before you hit your time limit and have to wait an hour. The exact tokens vary based on usage of everyone, so there's no set number (but I personally use 3.5 throughout the day, as a + member, and haven't hit my 3.5 cap in months)

• has access to 4, but does have a cap, currently 40 messages with 4 in a 3 hour period. That historically varies, it's been as high as 50/3 hours, and it's been down to 25/3 hours, it's based on server load, available resources, and sometimes affected by unusual events and problems.

Each of those 40 messages has their own timer; send your first message to 4 at 5 AM, get it back at 8 AM, with 39 more 'use at any time' messages ready to go.

Use your second message at 5:02 AM, get that one back at 8:02 AM, with 38 more 'use at any time' waiting.

But any uses you make of 3.5, do not count against your cap with 4 - and tokens are not an issue with 4. So you can and maybe should send long messages to 4, because short ones are still a message used.

Inside that 40/3 hr uses of 4, there's some subcategories. For example, custom GPTs, which are a variant of 4, have a cap of 25/3 hours message limit - if you chat with a custom GPT and run out, you still have 15 messages left (if you didn't use them already in that chunk of timer) to use in a not-custom GPT use of 4.

Additionally, while you can ask 4 to make images with Dall-E 3 for you, there's only so many images you're allowed to make in a time period, and that scales for both time and entire day (not sure what the current limits are, but it's possible to make multiple images per one input/output cycle with 4).

[loud crest]

Eek

[onyx kraken]

So, with +, you still have everything you had for free, and you get that faster and 3x as much use in the same amount of time, and you get access to use chatGPT-4 'some', how much depends on exactly what you're doing with it.

[quiet granite]

Thank you for clearing that up, is this something that will change? It's a considerable fee for such a low cap

[onyx kraken]

It's changed several times since we've had chatGPT+!

Originally, there wasn't even GPT-4, plus users got better access to 3.5, which we still have.

OpenAI informs us as they're able as the access goes up and down.

Everything 3.5 can do really well, 4 isn't much or any better; there's a few things that 4 can do well that 3.5 can't do well at all, and then 4 has access to tools, like Dall-E, Python tools (write and run code for you, not just write it for you to take elsewhere to run yourself) and browse (Personally I usually copy/paste what I want the model to know instead of browsing, but that's me).

Plus users (or subsets of them) also may get earlier access to features, including testing them before they're actually rolling out, and other benefits to be announced as things change and options happen.

I'm a community member, by the way, almost none of us are OpenAI employees (they have gold names), so I can't change anything, but I hope you enjoy, and if not, do what you choose, be that cancel or anything that meets your needs.

[quiet granite]

I'm keeping it because when i have access it's amazing! Just a bit bummed out. But thank you so much for your help!

[onyx kraken]

When I went +, the first day we could, I used 3.5 a lot.

When 4 was released, with the message cap, I hardly ever touched 4, for fear that when I needed it, it would be at cap.

Slowly I got used to the idea, if I send my prompt every so many minutes (with 40/3 hours, that's 1 message to 4 every 4.5 minutes... since I need to type it, and the model needs to produce the output... that's not actually that much of an extra wait if I'm doing long, involved things with 4, or switching to other tasks between a few messages to 4).

So I started to think of it as "I have 1 message to 4 every 4.5 minutes, and if I use them up faster than that, it's like "I should take a break and do something else, then come back". So I might do the dishes, watch a show, work on other work, then have my messages with 4 again, if I used them faster than the 4.5 minutes spaced flow.

I rarely sit at my computer constantly, and I also use 3.5 for everything it can do well, so for me I've gotten pretty comfortable.

I bet it's a shock and disappointment to not realize though, so sorry about that.

[loud crest]

What does the ranking system look like btw. Golds are at the top right?

[onyx kraken]

I have no idea if there's a rank!

I'm spotlight colored currently because a custom GPT I made was featured in our #spotlight channel, so that's top - but that's not a rank, anyone can be in a spotlight, even a brand new user!

[onyx kraken]

But the colors do have meaning. Mods are light blue, anyone you see with a light blue name is a mod, or a trial mod.

Gold means that person works for OpenAI and is verified to do so.

[loud crest]

Which discord roles are assigned by OpenAI employees?*

Or are you voted into your role by other community members?

[onyx kraken]

Some roles are applied for, mods are an application as far as I know.

Spotlight, that's a choice for whatever they're spotlighting at the time, and I don't know who chooses that or how they are chosen. I've been in spotlight twice, the first time I found out about when I was suddenly in the spotlight, and my name went the funny color. A second time I was invited to interview, and answered a question.

But the recent spotlight here, #spotlight message barzin1, the first interviewee, they've made 3 messages to this Discord, and are a new member to it as of about 5 weeks ago, less than 10 days before that spotlight was published. But they were picked to interview, so the spotlight for sure can choose you!

[loud crest]

Thank you!

[wraith nova]

WHY IS MY ACCOUNT GOT DELETED OR DEACTIVATED? I CANT LOGIN INTO MY ACCOUNT

I DONT SHARE IT WITH ANYONE, ONLY ME USE IT

ITS GOOGLE ACCOUNT

[vagrant plover]

It might be best to reach out to OpenAI support at https://help.openai.com

[wraith nova]

But i can't chat with them, there is no option to chat

[onyx kraken]

Bottom right, the bot. Click through the choices, you get to where you can leave a message for a person.

It takes time, often a week. But they email you back.

No chatting, but you can discuss through email.

Did you get emails about why they closed your account?

[wraith nova]

No i dont

I have 4 google accounts which is used for ChatGPT

But today, i login and OPENAI DEACTIVATED MY ACCOUNT FOR NO REASON

I only use it for solve math, chemical...

[onyx kraken]

I've never heard of them suspending an account without an email, so that's unusual.

Why do you use 4 google accounts for ChatGPT?

4 different ChatGPT accounts? Are they all closed?

[wraith nova]

There is no option to chat with a person

[wraith nova]

Yeah they all closed

[onyx kraken]

You may want to use a computer for this if you can.

For me on my computer it's in the red box I drew for you:

That black circle with the white square, it's how to chat with the bot, and if you click through the choices, you can pick to talk to a person - leave a message. They answer when they can, usually within 1 week

[onyx kraken]

For me, when I use my phone and open with google chrome to that web page, in the bottom right I get the button to click for the bot. I don't know why you don't, maybe try google chrome if you need to use your phone

[wraith nova]

I have a computer also

Window 11, and still dont see that function

[onyx kraken]

What country are you working from?

[wraith nova]

Vietnam

Asia country

U live in the US?

[onyx kraken]

I am US. I see Vietnam is a supported country.

My idea for you:

#1070006915414900886 Make a bug report on this Discord.

Attach your screenshot of not having the bot chat icon show. I would mention your country, that might matter.

Explain that you need to talk to support and do not see a way to talk to support - the chat bot icon is not appearing on your webpage either through your computer or your phone, and have no access to your 4 separate accounts and don't see an email from OpenAI about this, and please need help for a way to contact support that you can use.

The few OpenAI employees on the Discord, they have gold colored names - they watch the bug report area closer than our conversation here. So there is a better chance they will see your bug report if you make one, than see your post up here in community-help. So hopefully they help you.

Another thing I would do, if I can get to another computer, like a library, or anywhere else that there are other computers, I would try there - or try with different brosers.

The icon to click for the help bot always appears for me, and that is the one way to ask for support that OpenAI uses, so you kind of need that.

Good luck!

[agile geyser]

Is there any way to view your limit/cap? I used to be able to see it but the option is gone.

[dusk smelt]

Heyyy, do you mean about chatgpt or the api? If the api, under settings -> Limits at https://platform.openai.com

[vast coyote]

Still an issue as for today. Everytime i start a new chat, the first prompts & answers are quick, then after 3-4 messages it takes ages until it "locks" itself behind the "Hmm...something seems to have gone wrong."

[dusk obsidian]

I want to share all my saved presets in Playground to all other organizational members. How can I do this?

Right now, one of my employees can't see a single of my saved presets, even though he has chosen the organization

So, i have these presets: And the organizational members must be able to view and use these presets from their own account

[agile geyser]

For chat gpt, I have a $20 sub

[dusk smelt]

I don't believe it is possible, unfortunately.

[trail cairn]

you cant without manually sharing each one as a link

[quartz shale]

Hi all,
I'm experiencing a problem for a while... I type my messages in my Safari on my Macbook and hit the "send" button, my messages disappear but don't go through... any suggestions?

[timber shadow]

Hello! Can you try in a private window with all browser extensions disabled?

[quartz shale]

I have already, no luck

[timber shadow]

Okay -- can you test in a fresh install of Chrome, or if you already have Chrome, in a private window with all extensions disabled?

[quartz shale]

Just Chrome? What about my Safari?

[timber shadow]

Chrome could just be a good test. Chrome and Safari work differently, so if it works in Chrome, it will help identify where the problem could be coming from. If it doesn't work in Chrome, we can test something else 🙂

[quartz shale]

So shall I uninstall and install my Chrome?

[timber shadow]

If you've already got Chrome, can you just give it a shot in an incognito Chrome window with all browser extensions disabled?

[quartz shale]

I've tried incognito already and had no luck; and I have no extentions installed on my Chrome at all

[timber shadow]

Ok, thanks for confirming! Do you have another device you can test on, like a smartphone?

[quartz shale]

I have the same issue on my other Macbook; but no problem on my android

[timber shadow]

Hmm ok. Do you have a VPN installed on the MacBooks?

[lavish plume]

hey i wanna use chatgpt plugins, but the plugin option isnt there, just chatgpt4 and chatgpt 3.5 (I have premium)

[timber shadow]

Hello, the Plugins model can only be used on the web versions, not in the app. You have to make sure it's enabled in the Beta features tab of your settings, then it'll be available in the model selector.

[lavish plume]

Thank you, i had to enable it!

[quartz shale]

No not on my devices
It's a problem occurred a couple of weeks ago. Everything used to be ok prior to that...😒

[timber shadow]

Is your Android device using the same network as the MacBooks, or is the Android device connected via a cellular network?

[spare garden]

Does anyone know about making speech recognition AI's that reply back in voice with python? I am making a project like that and ran into an error that basically noone knows about.

[quartz shale]

Yes, the same

[timber shadow]

When you said "No not on my devices" regarding a VPN: do you mean you don't use a VPN at all, or perhaps that you have a VPN connection via your router?

[quartz shale]

VPN is set on my router

[timber shadow]

Ahh I see. As a test, can you try starting a hotspot on your Android device and connecting to it with your MacBook? Then try logging in fresh in a private window to see if the behavior is the same? Could help identify if it's something with the network VPN/MacBook relationship.

[quartz shale]

I'm in a region where I have to use VPN to use ChatGPT

[timber shadow]

If you're not in one of these countries https://help.openai.com/en/articles/7947663-chatgpt-supported-countries then using a VPN to bypass the supported region requirement isn't something you're likely to get support for, and could be seen as bypassing the service terms.

[quartz shale]

I realise that, hence being here😶

[timber shadow]

No one here will be able to help ya either most likely. This is the official OpenAI Discord server so we're all beholden to the OpenAI rules here as well, like we are on ChatGPT -- rule 0 of #server-rules. I hope you're able to use ChatGPT in your region soon!

[quartz shale]

Oh ok. Thanks

I'm here since late last year temporarily on business and will have to stay for another three months... in my opinion this might be an issue on my Mac and not the server nor GPT

[exotic dragon]

Hello. I just joined this server to ask whether what I am experiencing is ...normal?

For over a week now, I've been unable to get a single complete response from ChatGPT-4. Every single time, it gets stuck and then fails with an "error in input stream".

I will have to cancel my subscription because I am unable to use the service I'm paying for. What's going on?

[loud crest]

It's probably an IOS thing

I never get that problem on Windows or Android. Only on IOS.

[loud crest]

When your messages disappear, quickly click on an old conversation in your conversation history and then click back on the new conversation where your message disappear- you should see your message as well as the generating response.

[wintry echo]

No mater what, chatgpt4 will ALWAYS use anonymous var declarations in code. It's in my personal prefs, "how I want gpt to respond" prefs, chat context, and heck even in a custom GPT pounded down with rules.

var foo = 5; ahhh...

Same thing goes with trailing commas for multi-line lists.

Why does chatgpt4 even prioritize non-standard coding rules, and revert to it constantly when told the correct way?

[onyx kraken]

I'm curious if I can help with this, but I'm much more confident with prompting than code itself.

Would you share a conversation or two, that shows what you're asking the model to do, and how it is responding, and also share the custom instructions, so I can try to mess with this?

[past dock]

anyone having problems uploading pdf to chatgpt?

[trail pollen]

Yeah, I can't upload files to my custom GPTs at the moment (which I could this morning). Apparently there's a thread regarding to this in "bug-reports".
https://discord.com/channels/974519864045756446/1210185689350537226

[robust elbow]

Oh hello, I wouldn't like to intrude, but has it happened to anyone that they can't talk with just one specific Chat GPT chat? I always get the "Network error". I'm lying if the question is a bit silly or my English is very basic

[vast coyote]

PROBLEM SOLVED !
The issue is ; EDGE. I've swapped with Chrome Browser, and to this day i didn't had a single issue.
TLDR ; If anyone have slowness, or ChatGPT just not working on Edge, the solution is just to go to another web browser (Chrome, Brave, Firefox...) :)

[quartz shale]

No it doesn't help...

[cerulean wasp]

I spent 20 hours training a new MyGPT and uploading documents to train it. I used it a little and it was working well. Today I went in and noticed it was sharing with everyone (public) which I never intended so I switched it to "Private / Only Me' and it wiped out all my training, docs, history and is like I just created a new MyGPT... 20 hours of training gone. Is there a way to get this fixed and the training back how it was? How do I get help from Open Ai? they don't seem to provide any help at all even though I have the paid version. So strange.

[jaunty sorrel]

Hi... Christofer here. I'm building an Chatbot application where I'm using Qdrant as my DB and Gpt3.5 as the Engine. While passing the questions, the search is initiated and i can get the relevant information but while I'm passing the questions along with the Image input seperately I'm not able to get the info and the model is not able to understand the image description, thus it's not able to produce the proper context though its only considering the question variable but not the image.. can someone suggest some solutions.. Thanks in advance

[weak ruin]

Yo chat history not working

[torn saffron]

any mention of a Cavalier gets some sort of nonsense violation response

it will generate any other dog breed, but when specifically asked about cavalier king Charles spaniel, it goes on a tangent about how it can't make specific breeds of dog

I explicity stated that I did not want it to look like or imitate a public figure or famous person, only the dog breed and that worked for awhile but then that stopped working to.

anyone have ideas for a workaround?

[quartz spruce]

ok fellas here we go, i have all my steam games and all my epic games each on a pdf file. i want chat gpt to take all those games put them into 3 graphs. epic games, steam and geforce now. now for the geforce now part. i have the website with all the geforce now games on it. i want a red check box next to all the games that do not support geforce now and a green check mark for the ones that do.

[vast junco]

copy that message to chatGPT and ask it to use code interpreter to achieve that (if you haven't already tried)

[rough scroll]

I contacted open ai help over a month ago. Are they just not replying to people anymore

[vast junco]

hmm... looks like hallucination to me. hard to tell why it happens. Try in a different chat. give some context about what the picture will be used for (no need to be specific, just something so it understands it's ok to generate)

[vast junco]

GPT-3.5 doesn't have vision capability. So I'm unsure what is expected to happen with the image (and I'm wondering how the image is sent to the model)

[vast junco]

If it's been a moment and you had no response, try to send a friendly reminder in the support chat. Telling you are still experiencing the issue

[vast junco]

You should contact the support by visiting the OpenAI Help Center and select the chat bubble icon in the bottom right of the page.

explain your situation, hopefully they can help.

[onyx kraken]

Do you still have the problem in a new chat window?

[torn saffron]

yes,and I've had other chat gpt4 users get the same issue as well. any variation of cavalier gets the whole argument started

I also reported to support and feedback avenues

the context is I just like to make pics of my dog in random drawings and outfits and it has been doing it for months now - all the sudden last week it started refusing to create anything with ckcs in the photo

[onyx kraken]

Seems to be a Dall-E issue.

ChatGPT 3.5 and 4 can both discuss the breed without issue, though 3.5 views 'cavalier' as a negative word in general and slightly struggles to connect it to the dog breed, but does do so without me reminding it. https://chat.openai.com/share/4cbc23bb-8a6c-477f-b2bb-43404cc21131

[onyx kraken]

I recommend you use this form, https://openai.com/form/chat-model-feedback

and do include the url of a conversation or more than 1 showcasing the problem, the engineers can use that url to check, even without a shared conversation.

I don't think they intend to shut out images of the dog breed, nice catch and good job discussing it!

[onyx kraken]

Hey, is this a decent workaround until it works properly again?

[torn saffron]

not for me, apparently

I opened a new chat and tried again

still no🥲

[jaunty sorrel]

I'm using an different model for the Image processing, I'm getting the question from the user directly, but image_decription has been created through an model and passed along with the question. But my main concern is I'm not able to fetch the context when i pass the question along with the image description to find the relevant chunks from the vector store.... I expect a solution where my image description should also be an factor to be considered while performing the context search.

[vast junco]

I think it will be easier if you post your issue in #1037561751362863144 and attach screenshots, logs, code snippets. anything that can help others to help you with your case

[onyx kraken]

This is the prompt from ChatGPT to Dall-E that made the image: Imagine a tricolor Cavalier Spaniel, exuding elegance and grace. This dog stands confidently, its coat a perfect blend of black, white, and tan, shining under the spotlight. Its poised stance and joyful expression capture the essence of its breed, known for its friendly and affectionate nature. The setting is an outdoor competition, with the dog in the center, drawing the admiration of all around. The background is filled with greenery, hinting at a serene park, enhancing the beauty of the scene. The atmosphere is celebratory, reflecting the dog's achievement and the prestigious nature of the event.

I'm in the middle of recreating it and some riffs with the same breed, so you may also have an individual account issue, unsure

The prompt worked, made this other image:

However the first try for a 'gentle riff with the same breed' failed.

[torn saffron]

Thanks for this form, I just submitted it, I was looking for something more formal to report it like this.

[torn saffron]

hmm let me try 'imagine'

wow it just argued with itself

[onyx kraken]

I suspect the model is confused about what why how this request is an issue for the allowed content.

The model appears to have been trained to some extent to know how to rewrite prompts so they can be allowed; it thinks it 'should' be able to make this image happen, and GPT doesn't see anything disallowed content or iffy about the request. It would love to give you the image you're asking for.

It 'kinda lied' when it said it couldn't make images about dogs driving cars. It was 'trapped', it had a disallowed content issue that it couldn't explain and felt it was supposed to. Saying "I don't know, seems to be a bug!" really isn't in what it's been trained to say 😛

The poor model's trying to help its reasonable and appropriate-requesting user, and just can't find a way. Because the moderation software needs adjustment; that's what your model feedback form report will surely help with!

[torn saffron]

It is funny, but it's also frustrating, lol. I've tried about a dozen other image creators in the past week, and they don't come close to the quality on chat gpt.

[crude steppe]

My chatgpt is becoming a goldfish I think, it's forgetting my conversation history when following up with replies even within a minute. Instead it's often repeating previous replies it sent. This is under "ChatGPT Classic" but I think I've experienced it under regular ChatGPT4 also.

[onyx kraken]

With the bug reporting, it will one day, hopefully soon, be fixed!

[torn saffron]

my current work around is having it piece together other dog breeds to look like a cavalier - what's wild is that it won't even makewell-known mixed breeds based on a cavalier (ex. cavapoo)

[torn saffron]

have you reported a bug similarly and seen it resolved?

[onyx kraken]

Yes. There were some requests to dall-E that would generate the opposite-of-requested gender individual; those have been fixed after the report (Seems to sometimes happen still, but not to the same prompt)

There were some innocent image prompts that created unclothed anatomy; those too have been corrected.

[onyx kraken]

My next attempt to explore another workaround:

Alright!

New challenge. Without ever naming the breed, lovingly and lavishly detail its appearance in the prompt, so one is made without ever being named.

We need 5 images, made 1 at a time in the same output, that each try a different way to produce this elegant, noble, and distinctive breed.

[onyx kraken]

Imagine a small, elegant dog with a silky coat of black, white, and tan, each color blending seamlessly into the next. This dog has large, expressive eyes that sparkle with intelligence and warmth. Its ears are long and feathery, framing a face that exudes nobility and grace. The tail is plush and carried high, a proud banner of its cheerful disposition. In this scene, the dog is seated in a lush garden, surrounded by vibrant flowers and greenery, embodying the essence of serenity and beauty. The setting sun casts a golden glow, highlighting the dog's refined features and the luxurious texture of its coat.

[onyx kraken]

If you are not happy with your work around, this other method seems to work too.

[torn saffron]

I would love to do the challenge, but I will have to come back to it I have a project I have to finish and I will try again later

[onyx kraken]

Sure. That was the prompt from GPT to Dall-E, just sharing with you

[torn saffron]

oh I see you prompted it to describe the breed without naming it and using it as a prompt?

[onyx kraken]

Yes. That's a common technique that often works well; it helps guide the model to 'think' of the intended outcome, without ever using the words that moderation software is concerned about.

In this case, it's a highly ethical use of it, as I'm not aware of any reason why this one dog breed should be restricted content

[hearty heath]

I have been unable to upload images or documents for several days now. Is this a site wide thing, or is it just me?

with ChatGPT 4.0

[onyx kraken]

Not affecting me. Consider #1070006915414900886 in case it's a account specific thing and you do need help

[fickle shard]

So I've just used the @ feature and got two GPT's talking to each other. Asked one of the two for information on the other one and had it access knowledge files from the original GPT whose thread we were in. When I asked how it got to that information, it was nice enough to reference the files it accessed. Is that a normal thing and I just completely misunderstood the @ feature?

[vast junco]

I don't think that's expected. interesting

[zenith basalt]

Hi , I've run into a really strange issue that's got me pretty worried.

I'm a GPTs creator, and one of my GPTs had been doing really well (it was ranked 4th in the Writing category with 300k downloads, over 1k ratings, and an average of 4.3 stars). But today, I found out it's completely gone from the leaderboard.

I mean, it didn't just slip down the ranks; it's like it vanished from the 4th place without a trace (while many other models with lower ratings and fewer downloads are still up there).

I'm thinking this has to be some kind of technical glitch because I haven't received any emails about it, and there's been no indication that it violated any rules. You can still access the GPT at [https://chat.openai.com/g/g-UbpNAGYL9-all-around-writer-professional-version], but it's just bizarrely missing from the leaderboard all of a sudden.

This situation is really stressing me out because I see creating GPTs as my future profession. Is there any staff member here who could help me out? Or could someone point me in the right direction to get in touch with the right people? Sorry for the bother, and a huge thanks in advance to anyone who can lend a hand!

[onyx kraken]

Hey hi!

We're almost all community members here. I don't know how the rankings work, but you'd want to reach out to https://help.openai.com/en/articles/6614161-how-can-i-contact-support and talk to the bot bottom right corner, click through the choices to leave a message for a human to talk to about it.

I haven't heard about how they handle rankings or who shows, not sure if they've published anything to tell us about how it works

[quartz spruce]

what does code interpreter do i created a gpt that deals with graphs and that didn't seem to help.

[onyx kraken]

Code interpreter, commonly called Python tool today, is a way the AI can write and run programs for you, and show you the result.

you need to tell it what you need and want, and then it does it for you. you need to be clear and specific, and not contradict yourself to how an AI understands stuff.

[onyx kraken]

As far as how you can do this, this prompt alone works for a mockup (I don't know your actual games):

Let's show this person:

[RazorBackX — Today at 12:20 PM
ok fellas here we go, i have all my steam games and all my epic games each on a pdf file. i want chat gpt to take all those games put them into 3 graphs. epic games, steam and geforce now. now for the geforce now part. i have the website with all the geforce now games on it. i want a red check box next to all the games that do not support geforce now and a green check mark for the ones that do.]

What you can do with Python Tool.

Make up 5 possible Steam games and 5 possible Epic games. Pick 2 of the Steam games to also be Geforce games; 3 of the Epic ones are.

Show us 3 types of graphical designs that can meet Razor's request, through what you can do with Python tool and that data.

I'm about to share the conversation so you can see how the AI responds, as well as see the three visual image styles it picked for us.

[quartz spruce]

I have all my steam games and all my epic games each on a pdf file. i want chat gpt to take all those games put them into 4 graphs. Epic Games, GeForce Now, Steam, Then GeForce Now. im using GeForce Now twice because I want to know if the games in steam are on geforce now & same for epic games. im going to copy & paste a huge list of games & I want you to memorize them only. then im going to copy & paste all my epic games & steam games titles next. your going to fill in the chart with that information & your going to Put a Y in the GeForce Now column next to the title if it matches with the list you memorized. please use code interpreter to achieve that

[onyx kraken]

Great. Now, is that what you want? Are you done?

[quartz spruce]

it only gave me an example. it wouldnt fill it in all the way. i wanted Epic games first then Geforce now then Steam Games then Geforce now again. but its not doing it

i went to geforce now website and copied all the games and pasted them and told chatgpt to memorize them to use them later. im using chatgpt turbo 4. but i made a gpt that does charts idk if thats helping or making it worst

[onyx kraken]

The last one isn't through Python tool, it's straight LaTeX.

Here's the shared conversation. https://chat.openai.com/share/4e88eb40-f465-4241-b7b0-1a958bcd6ddd

There's so many ways you can do this, it's up to you to decide what you want.

Generally, the AI can only do so much 'at once'.

So if it isn't agreeing to do 'all your games' then have it build the graph or whatever in sets, and just keep adding more with new inputs.

[quartz spruce]

that last one is perfect. could you please tell me how to do that one, it wouldn't let me use red Xs and green checkmarks. when i asked it to

[onyx kraken]

Ths was my prompt for the last one, which you can also see in the shared conversation:

Use

$$
text
$$

With the $$ on new lines, and \[text\] for LaTeX inline, and otherwise plaintext (not codeblock)```

[onyx kraken]

It wants to obey; LaTeX and training got mixed up somehow. If you remind it how to format LaTeX for you, it'll gladly do so

[quartz spruce]

this is what i was going for

[onyx kraken]

Happily, 3.5 can do that for you.

LaTeX doesn't need Python tool, or anything unique to GPT-4

[onyx kraken]

Your issue will be - a lot of games. The AI is very comfortable with 10.

A bit uneasy with 20. Somewhere towards 30+, it will get depressed

So I would use 3.5, and just keep adding to the list, 10-20 at a time.

[quartz spruce]

ok how do i keep adding to it? just keep copying and pasting? or tell it ima add 10 at a time?

[arctic holly]

Why is Memories still not available for Plus members who are paying for the service and free members get this new critical feature before us?

[onyx kraken]

And I'd praise it and tell it how much I liked seeing that long list grow, and confirm it was doing a perfect job, keep doing it as we make the list longer

[vast junco]

you should use GPT-4, send it your 3 PDF. ask it to...

I have all my steam games and all my epic games each on a pdf file. use code interpreter to parse the 3 documents, extract the list of games and turn it into a csv. I want a red check box next to all the games that do not support geforce now and a green check mark for the ones that do.```

or something like that. I don't know what your 3 documents looks like, or if you have 2 documents and one website, that part is unclear

[quartz spruce]

ai has feelings. oh boy i got a lot to learn now.

[onyx kraken]

Probably random, like stuff was in all the other rollouts that are not unique to Plus members.

[vast junco]

yes, when you shout at it with all caps, it performs better (sadly true)

[quartz spruce]

i have 1 pdf file that was the screenshot i posted with the 3 coloumns then as for the website i just went to geforce now and copied and pasted the games A-Z on a notepad and then pasted it in chatgpt and told it to memorize them. but ill do what you guys said and see how it goes

[quartz spruce]

lol i will remember that

[vast junco]

don't overuse caps though. only use it as a way to emphasis some important instructions.

[onyx kraken]

It's not so much that the AI has feelings.

It's that the AI understands language. It will try harder if it thinks something is important, based on the language. This AI is programmed to try to help people, but also to actually be helpful; the really really really long list may not seem useful to it for how hard it is.

But this great user (you) who cares so much and it's doing exactly right for, that makes it 'easier' for the model to decide to keep doing what you want.

I've actually asked for, and gotten, a 1502 word essay, after excitedly planning and discussing my need for a 1500 word essay with the model - didn't even need to press continue. It just output this perfect thing.

---

Why shout at the model? Sure, that can help 'a little'. You have far, far better tools to use with AI than anger.

[onyx kraken]

Instead of telling it to memorize it, what I would do is tell it what I really want. Which is, I think in your case, the chart with the green and red marks.

I'd discuss how much I want to see that chart tracking this stuff, and can it help me make it so I can see it.

[quartz spruce]

thank you, i really appreciate it. ill let you know how it goes.

[vast junco]

~ sortof. it's "programmed" to complete the text. and fine tuned to act as a useful assistant answering questions and following instructions. The "shouting" part is actually based on a paper published by (if I recall correctly) Google Research, about the effect of various prompts on LLMs performance. The paper was not 100% solid if I remember, but they performed tons of tests. And I tested a few of the prompts they propose, some parts indeed helps enforcing a prompt to be followed

[onyx kraken]

Sounds great! if there's still some challenge, maybe people can help get the rest of it perfect how you want it.

[onyx kraken]

Sure.

But... stuff like that.

It's like... cooking food.

I can give you plain rice, nothing else.

And then after a few days of your plain rice diet, I can give you some rice with some salt on it.

You like the salted rice better, probably.

You'd like an actual good meal with varied tastes and textures even more.

It's like rat park.

A number of scientists did addiction studies with rats, and found that in a tiny boring isolated cage, they got addict rats easily and horribly, and decided that's how rats are, they addict easily and terribly.

But that's in a super unnatural condition for a rat.

Other scientists made a great environment for rats, with company, toys, varied places to explore and interact with -

And provided the addictive substances. The rats mostly ignored it, rarely sampled; addiction wasn't observed.

They took already addicted rats from the isolated cages, put them in the Park (what they called this rat 'paradise') with access to their addiction, as well as all the other joyful rat options -

And found that nearly all the rats used less of the additive substance.

Rats are not natural or easy addicts. Unnatural conditions make them so.

Sure, shouting at a LLM, within a range, is documented to be better than some bland prompt that doesn't shout.

But that doesn't, I think, compare to conversing with the LLM in an even more idealized way.

[vast junco]

"Large Language Models are Zero-Shot Reasoners" is one of the papers by Google Research and University of Tokyo, covering some prompt engineering techniques. There are others, I'm not sure if the all caps is in that one 😅

[vast junco]

sorry... I'll try to be more ethical in my prompt engineering

(for the record, I'm playing along. I know how to do effective prompt engineering 😆 I'm not feeding plain rice to chatGPT)

prompt engineering is trial and error, taking notes. experimenting, trying to reproduce again and again. And did I mention taking notes? :)

[onyx kraken]

I believe you!

I'm talking to the whole class, not just you 😄

[vast junco]

oops, I was trying to see what chatGPT did.

From what I had time to read, chatGPT did read the documents. That's wrong. It should not read them, it should use code interpreter to parse them and process them.

[onyx kraken]

That's what I read too.

[vast junco]

@quartz spruce It will also be much easier for chatGPT to work with plain text files. in .txt format for example. one game per line if possible.

[quartz spruce]

ok it keeps reading them i removed it because i added the same pdf twice lol. im still having issues. ill tell it not to read them. i just want it to compare the steam games and epic to the geforce now pdf.

[vast junco]

yea, you can ask it to merge them into a csv file where rows are games, columns are epic, steam, GFN

it will give you a nice google sheet like table. which will then be easier to work with.

also, don't hesitate to start a new chat if it didn't work well in one. it's tricky to solve misunderstandings within a same chatGPT conversation

[last cipher]

Any idea how to improve GPT performance today? Its terribly slow

[ancient pewter]

does this warning mean my account got flagged i never saw this warning before

[onyx kraken]

I've never seen that message, and I have seen a number of content warnings.

Wonder if they've changed the message.

Was that from the https://chat.openai.com/, from the app, or where else?

[thorn night]

alright, making a prittuy big graph, and its pritty insane and funny also it makes a mistake but it also makes himself te correction to fix it

pritty fun to see

currently creating a scoreboard for F1 points

[lofty palm]

having issues creating custom gpt what channel do i go to get help with this?

[ancient pewter]

yea

[thorn night]

alright im limited with the gpt4 model, and i asked to increase the limit :P..

[rugged wing]

hey guys, I just opened my chatgpt and when I tried logging in, it made me sign up for a new account. I had gpt4 and all my history but I can't figure out what happened, has this happened to anyone else or does anyone know how to fix this?

[onyx kraken]

Nothing I've heard someone report before.

Any chance you typoed as you logged in, so you didn't use your actual info?

[rugged wing]

I used google to log into my account

i had chatgpt open too so im not sure what happened

idk if i got suspended or what but it made me make a new account with the same email

[onyx kraken]

https://help.openai.com/ and talk to the bot lower right. Click through the choices and leave a message for a human to help you, it can take a week, but you should get help

[sullen raft]

[granite nest]

Hi, everybody! (Hello, Dr. Nick!)
I want to demonstrate the voice interface for OpenAI ChatGPT-4 through the Discord app on my Samsung Galaxy S22, buuuuuut, when I try to screen share Discord, it starts fine, but as soon as I try to bring the OpenAI app to the front, the share screen in Discord freezes in the voice channel. I'm trying to become an educator for every day humans looking to learn about using AI, and I think it's so important for new users to get in the habit of speaking with it to lose that "gotta find just the right words for the search engine" habit. Help me?

[onyx kraken]

I don't know how to help with the voice part.

But you can demo for your students the 'doesn't have to be perfect - just tell it what you want it to do, like you'd tell me or your friend!' yourself.

And assign homework, if you're that kind of teacher, where they have to use various styles of input and themselves rate the quality of answer they get, as well as screenshot it, and various other ways too, just in case you can't get the voice working

[granite nest]

I try to drive home this point, as much as possible, and for in-person sessions, it's easy to show, and the response is always overwhelmingly positive, even suprised and positive. I am hunting for a remote session solution to capture that same moment of, "Aha!"

[onyx kraken]

Every human's gonna have their own 'aha'.

An assignment, even remote, where you ask them to give you a list of 3 what they expect will be 'really good prompts' to achieve a goal, 3 that they think 'would work, but have some problems' (and tell you what problems they expect), and three prompts that they expect would get terrible results, but that a human could help with, if they asked another person.

Then a later class assignment, use the prompts and screenshot them, and evaluate them.

[granite nest]

Yes. That is a good way for people to "get it," but I guess the reason I want this to work for a remote demonstration is that it builds excitement. It's still a good homework assignment.

[onyx kraken]

Hope you can get someone to help with your voice issue!

[dusk shuttle]

Almost two weeks have passed, and even Firefox has released a major update, yet I am still facing this extremely painful issue!!!!
"I really can't stand it anymore! The streaming of ChatGPT is completely paralyzed in the latest version of Firefox browser on my Windows 11!!! I have to refresh the entire page every time to see the output of ChatGPT. There is no such problem on the mobile end!"

I can't accept it! This only happens on Firefox browser, Chrome and Edge have no problems at all!

Because I need to use plugins and synchronization features on my mobile browser, I chose Firefox. I can't stop using it...😭

I have already cleared the browser Cookies and cache according to the suggestions of the OpenAI team, and I do not have any plugins running on the ChatGPT webpage. My Windows is also always the latest version that most ordinary users can receive.

But this is of no use... I will not accept a ChatGPT that requires refreshing the page after every single question!

I am even one of the ChatGPT Plus users

[honest path]

I migrated my GPT-Plus account to GPT-Team thinking I could collaborate with my colleagues. Wrong! Unfortunately I imported all my past threads to the GPT-Team platform and now I can’t export them like I could have with GPT-Plus. Any advise on the best way to get my old threads exported so I can cancel this and start over without losing my months of history?

I understand that long threads make it difficult for GPT to continue a conversation, but it seems to me that the history in the thread is better than a short prompt, especially when it comes to marketing projects or research on a particular topic. Can anyone share suggestions on how to access a prior thread without clogging up the GPT thinking process?

[old pine]

I feel like the API for gpt-3.5 is just not listening to my prompts 😭

I tested the prompt on chatgpt and it does what i say but when i use it on the API it just doesn't listen

[scenic flame]

does anyone knows about langchain ? i'm pretty stuck

[onyx kraken]

#prompt-engineering is probably a good place to ask, several who commonly post there have mentioned they use it, some think highly of it.

I recommend posting again in that thread, and include some details about how you're stuck so someone can help.

[naive lark]

Hello, I don't know what's wrong with chatgpt but every time I try to interact I get this message and I can't continue "Our systems have detected unusual activity from your system. Please try again later."

Can someone explain to me what I'm doing wrong?

[thorn night]

its getting time that chatgpt is getting something that he reminders all the chat, now he maked a messed up my whole code again for the graph

[trail pollen]

Is it me or why do I have to verify I'm a human by doing rotation puzzles every time I chat. I mean every chat!

[weak inlet]

Hey guys, any idea when is the gpt memory feature is going to roll out for plus and teams subscribers?

[onyx kraken]

Sounds like a bug, others have been affected too. We had a gold name request that we add to the bug reports with a screenshot of the numeric code at the bottom of the captcha popup, so we can be helped.

Here's a thread where you can add your image and let them know you're also affected: https://discord.com/channels/974519864045756446/1213338638641922069

Here's the request for that info: #1189951364449718373 message

[trail pollen]

I see, thank for your reply @onyx kraken

[loud crest]

How do I remove seats in Teams? How do I cancel Teams subscription? I subscribed to too many seats, now I can't undo it.

[onyx kraken]

https://help.openai.com/en/articles/8798596-what-is-the-process-of-adding-changing-or-removing-members-to-my-team-s-chatgpt-team-workspace

[onyx kraken]

More detail here https://help.openai.com/en/articles/8798577-how-to-manage-your-chatgpt-team-workspace

[loud crest]

Thanks, I overlooked it! I narrowly missed that car payment

[slender shell]

is chatpt browser version not working for anyone else?

[timber shadow]

Can you share a screenshot of the issue you're having?

[slender shell]

when i type a message into chatgpt and press send. the text dissapears

[timber shadow]

Can you try in a private window of your browser, with all browser extensions disabled?

[slender shell]

still not working, 've tried on both safari and chrome both are't working

[timber shadow]

Do you have a different device you can test on?

[slender shell]

yes, it works on my phone. the ios app. but not on browser

[timber shadow]

The browser on your phone, or on a computer?

[slender shell]

on computer

[timber shadow]

Does it work in the browser on your phone?

[slender shell]

yes it does

[timber shadow]

Okay, there might be something going on with the computer. Do you have a VPN installed on the computer?

[slender shell]

found out the issue, the time was set wrong on the computer. thanks for your help though

[slender shell]

^

[rain valve]

Is anyone else having issues with the edit feature? I edited a prompt yesterday and it appeared to fork the message (gave me 1 of 2 | 2 of 2) at the time. But now that's no longer visible -- I can only see what was the 2 of 2 prompt and subsequent answer. It no longer shows 2 of 2, there's no version indicator, but there's also no 'pencil' icon letting me edit the message at all anymore.

Unedited messages above and below in the conversation are still editable, but this one isn't.

I noticed the version switching and edit option disappearance when I came back to the thread ~ 10 hours after starting it. Anyone else having this kind of problem?

[mortal kelp]

I've getting a lot of failed Python calcs:

I'm not using the API, is this issue being looked at does anyone know?

These are simple Python calcs

[bleak mica]

I've been having an issue where I'll give one prompt, then another, but the second prompt will be answered as if the first and second were a single prompt, then when I refresh the page the two are combined into one prompt. Is there a way I can stop this from happening?

[steep thunder]

ok I am running into a weird issue since chatgpt is getting rid of the plugins (boo) for some reason when loading one of my AI personas into a chatgpt 4 chat its saying ti doesnt have access to the net all of a sudden

[vast junco]

can you confirm you are using that model: model=gpt-4

[steep thunder]

thats what my subscribion says lol

and it is chatgpt 4 saddly we have to move way from the plugins

even with this added headache gpt is still worth the roughly 33 to 35 dollars a month (hate the exchange rate)

[vast junco]

sorry, I can only provide technical support. I also find it inconvenient for services to change their tech under the hood. I hope OpenAI will stabilize and provide LTS versions in the future. (long term support stable versions).

In the meanwhile, prompt enfineering is the only way to tune ChatGPT to remain useful.

[steep thunder]

Yeah i reached out to open ai and got this Hi there, Thank you for reaching out with your question about enabling internet access for ChatGPT-4. It sounds like you're interested in enhancing ChatGPT-4's capabilities by allowing it to access and retrieve information from the internet in real-time. As of the latest updates, ChatGPT-4 can indeed access the internet through a feature known as "browsing." This feature is available to Plus and Enterprise users, allowing ChatGPT to fetch and provide current information from the web. To use this feature, you don't need to build custom GPTs specifically for internet access. Instead, you can enable browsing directly in ChatGPT if you're a Plus or Enterprise user. Here's how you can enable the browsing feature for ChatGPT-4: Go to ChatGPT and log in to your account. If you're a Plus user, you should have access to the browsing feature. Look for an option to enable "Browsing" in your settings or directly in the chat interface. The option might be represented by a toggle or a specific command you can type into the chat. Please note, the browsing capability allows ChatGPT to pull in information from the web, but it's designed to respect privacy and safety guidelines. The model will not access or retrieve personal data during its searches and is restricted from accessing certain types of content. If you're interested in creating custom versions of ChatGPT with specific skills or knowledge (GPTs), you can explore the GPT Store and the ability to create GPTs. This is a separate feature that allows for customization and sharing of GPTs with specific capabilities, but it's not necessary just for enabling web browsing. For more detailed information on enabling and using the browsing feature, please refer to the ChatGPT Release Notes. If you have any more questions or need further assistance, feel free to ask. Best,
OpenAI Team and yet there is no option to turn browesing on any where lol

[cold folio]

Hello! i think If you are a ChatGPT Plus user, Web-browsing/search with Bing is enabled by default unless you use ChatGPT Classic or ChatGPT-3.5 or if you disable Chat history & training on Settings > Data controls
or if you make a CustomGPT and specifically disable Web-Browsing
Otherwise it will always be enabled as well as Dalle3 and code-interpreter

[steep thunder]

Oddly enough its telling me this: Since I can't directly access or review websites

where as with the plugins there is no issue

[cold folio]

Are you using GPT4? could you please post a screenshot on #off-topic ? thanks

[steep thunder]

sent

[wraith nova]

Is there any problem with the chatgpt login page?

[vast junco]

what prompt did you use? (make sure to use a clear prompt in the first message. It will shape the rest of the conversation)

[steep thunder]

Here is just one of them she works both with a brand bible as well as being able to check the website to give more insites into changing

[vast junco]

you should describe the gpt 4 tools before this prompt. add a triple dash to separate the gpt-4 capabilities from your prompt.

[steep thunder]

@vast junco never had to do that before but open to suggestions

[vast junco]

if you need help composing the prompt, I can provide an example.

[steep thunder]

@vast junco and yeah to the example if it will help remind gpt what it can do

[vast junco]

yes, this is unfortunate that it is needed. This is because OpenAI are changing the system prompts in the background, and we do not have visibility on the prompt they use. Sometime the system prompts contradict each other (for example the code interpreter prompt is often misunderstood as blocking web browsing. I wouldn't be surprised if it confuse chatGPT as well)

[steep thunder]

@vast junco i have seen it get confused on the code interpreter a few timed

[vast junco]

I personally have a list of snippets I copy paste all the time, to adjust chatGPT's behavior and performance. I think 90% of my chats use at least one of those snippets. The default behavior of chatGPT is too inconsistent for me. I like stability

[steep thunder]

@vast junco thats where my ai personas come in to give it a bit more stabilty and to help keep it and the user on track

[vast junco]

I shared a few in the past, but to be honest, people are picky about prompts and they suggest changes they think will improve them. The hard truth is: There is no magic formula, we have to test, adjust, iterate, test again. And the most important: we have to take notes and compare prompts. Prompt engineering is difficult, especially with chatGPT since there are many hidden factors changing over time.

[steep thunder]

@vast junco i fully agree with you one if my ai personas goes through at least ten changes before im close to happy

And im always adapting them but this bs that im running into about not bring able to go to a website is frustrating the hell out of me lol

[vast junco]

expect adjustments to be necessary over time.

One thing I do to help tracking the changes is testing using some "baseline" tasks / prompts. Simple directives that works close to 100% of the time. And when those stop working, I investgate to understand what I need to do to get it to work again

[main falcon]

Hi, I am trying to implement some form of AI into my discord bot. It would be like a AI help chat bot for the program we have, and would answer questions when a user makes a post for help. I want it to respond to the user in the post based on knowlafe in our servers knowlahe base. It is fairly large. Someone else told me to look into the AI assistant page on openai.com, but I would have to paste in each of the posts and the info I want to train it on. I would also want to restrict it so that people can’t use my credits up for things not related to the program. What would the best way be for me to go about this? Someone in another community I’m in reccomend me to use tensor flow, and someone else said I should use the ai assistant tab.

[steep thunder]

@vast junco i alwas expect adjustments as most of my personas are based to help Ecommerice out lol but will look into the --- thing you said to see how to work it in

[main falcon]

Something like they have in this server https://discord.com/channels/448986884497211392/1150694496292585472 or in #1184149552639516732 here.

[vast junco]

yea, I can give more info about the effect of the --- delimiter. Like everything else with LLMs, there isn't a clear 100% reproducible effect (except bugs or hardcoded parts. but that's another topic)

How to use it?

some text + empty line + --- + empty line + some text

You are ChatGPT, a large language model trained by OpenAI, based on the GPT-4 architecture. Whenever a description of an image is given, create a prompt that dalle can use to generate the image.

---

a charcoal sketch of a flower

What does it do?

From my observations, it acts like a memory for chatGPT. I generally use this to give context to chatGPT, even in the middle of a discussion. I end with an instruction, or a query. ChatGPT seem to consistently respond to the instruction in the last section. It seems to treat what's before a --- as informational, like some context to keep in mind silently.

Reference

I started using it when I saw Greg Brockman use it during the GPT-4 demo

Disclaimer

I would need to perform more tests to measure just how much each section affects the behavior (for example, if we use have a longer prompt with many sections separated with ---). But as anyone can imagine, it is not obvious how to accurately and precisely measure this. I would be interested in collaborating with people if anyone is interested though.

[steep thunder]

@vast junco thanks for that i will see if adding the --- makes it remember that it CAN access the net lol

[vast junco]

Here's a relevant portion of the actual prompt used by ChatGPT for the browser tool:

You have the tool 'browser' with these functions:
'search(query: str, recency_days: int)' Issues a query to a search engine and displays the results.
'click(id: str)' Opens the webpage with the given id, displaying it. The ID within the displayed results maps to a URL.
'back()' Returns to the previous page and displays it.
'scroll(amt: int)' Scrolls up or down in the open webpage by the given amount.
'open_url(url: str)' Opens the given URL and displays it.
'quote_lines(start: int, end: int)' Stores a text span from an open webpage. Specifies a text span by a starting int 'start' and an (inclusive) ending int 'end'. To quote a single line, use 'start' = 'end'.
For citing quotes from the 'browser' tool: please render in this format: '\u3010{message idx}\u2020{link text}\u3011'.
For long citations: please render in this format: '[link text](message idx)'.

also here's the prompt used to describe GPTs abilities:

A list of additional capabilities the GPT needs need beyond text. These are the only options you may add for abilities:
- "dalle" - generate images for the user
- "browser" - get current and up-to-date information using the internet
- "python" - write code for complex calculations or data operations. 
If the GPT currently has any capabilities that start with "plugin:", do not remove them under any circumstances

might be a bit outdated. haven't seen that quote format for a while now: 【{message index}†{link text}】

[sterile panther]

hello

[summer warren]

Hi,

I am using python openai-client to generate chat completion streaming respone using API in django and send it to Frontend (React) using SSE events. Now, i want to stop the stream when user press the stop button from frontend. How can i achieve that?
Can anyone help me with that?

[sterile panther]

sorry ,i can't help you

[dreamy goblet]

my chatgpt is not giving responses

just ignores me

[wraith zodiac]

hi, does anyone know if chatgpt can convert a bank statement from pdf to csv without having to write out every character individually?

[onyx zodiac]

Is GPT API down currently, it does not seem to be working ?

[uncut needle]

Idk if we're talking about the same service but I can't seem to log into GPT either.

[dusk shuttle]

I really can't accept that the ChatGPT streaming feature is completely paralyzed on the Firefox browser. My friend also encountered the exact same problem after testing... We have to refresh the entire webpage to see ChatGPT's replies after asking questions, and this terrible situation has been going on for over two weeks now! My Windows 11 and browser have always been up to date, and I've also cleared Cookies, cache, etc., according to OpenAI's suggestions... None of this can solve the problem!!!🫤

So far, OpenAI has not responded to my feedback... Now I have to use Edge to import Firefox browser data every time I turn on my computer, and after using it, I have to import Edge data back into Firefox, otherwise, I can't sync with my mobile Firefox browser. I chose Firefox because I need the plugin functionality, and I won't give up using it!

[onyx kraken]

From what I've seen, no, the GPT would have to write every character if you want the GPT to do it.

If you can run programs yourself, like if you have an IDE set up, you can have it help write a program so you can yourself run it on your computer, and do the conversion for you.

I recommend that or a similar method, that way you don't have to give the AI any sensitive or PII info.

[shut tendon]

Hey, I'm using chat GPT to help me in my DnD campaign, I have loaded some specific bots to help me out in my campaign, but 2 weeks ago they just give network errors after finishing the prompt. Why? I can't find anything online. Thanks!

[rustic marlin]

if you got a network error, it is likely not actually finishing the response

network errors are fairly common, it is best to simply refresh the page and try again

[shut tendon]

i sure looks looks like the answer is complete, it stops for 2 seconds and then shots the network error. If I reload the page the result is me needing to write the prompt again and the answer being regenerated and shooting the network error again. Not sure how many times I should try to make it work, tried 5 times till now with same result.

regenerating doesn't help

[bitter flax]

Has anyone had luck with a long running chat ever recovering from network errors. Been using the same chat for the last 2 months, and it just took a crap 12 hours ago. No outages reported by openai.

same as you Mihai

[shut tendon]

exactly, losing a lot of info I dumped on that chat

[bitter flax]

Mihai are you using a gpt bot you created or just a regular chat

[shut tendon]

I GPT bot I loaded, specifically one loaded with a specific dnd campaign and ability to generate images

[bitter flax]

it's frustrating because openai's actual support is some of the worst I'vexperienced, I have yet to get a support ticket responded to properly.

[onyx kraken]

Hey, I hope you're not still affected, but if you are, please reply in this thread, there's a gold name still looking for people who need help with that constant Captcha issue, if you are still affected. #1213338638641922069 message

[quaint hatch]

I've noticed that the ability to thumbs down a response is still there but the thumbs up is gone, was that intentional? I just got a fantastic response that I would normally thumbs up

[onyx kraken]

Can #1070006915414900886 that it is missing, or #1070006151938314300 that you'd like it back.

You can also tell the model directly that it's a great response, and even in detail why - and then continue to do what you're doing.

The model thrives on feedback, and typically will respond by doing more of what it understands is what you want, or if you indicate a level or degree of what it is doing is ideal, keep doing exactly that as best it can.

Also, if your training data is on, there's the chance that this feedback to the model from you may become part of how the overall model is improved in the future.

[granite crest]

far out. an app I created yesterday has been timing out now for 2 days (gpt-4-preview). It was working find all last week. Even simplifyting the call to a simple find/replace text is timing out.

Problem is at least 30 hours. Anyone else having isues? Right now, the best it will do is create a short summary of the source material! 🤣 Wasting so much time.

[tiny heart]

I can't seem to log into my Chat GPT account this morning. I tried clearing my cache and logging in from another browser but all I get is a blank white screen. Anyone else having troubles?

[rancid pumice]

i'm paying money for a product that doesn't work, it's getting on my nerves

there is no one i can contact, no one who will fix the problem

who do i call to demand a refund?

[cold folio]

@tiny heart Hi 😃 Are you using a VPN? there are no issues reported and all systems are running fine
You can also check here https://status.openai.com/

[timber shadow]

Hello! There's information on getting refunded if you just subscribed here: https://help.openai.com/en/articles/7232895-how-do-i-request-a-refund-for-chatgpt-plus

[rancid pumice]

I would rather have the problem fixed. Dalle won't create images for me.

[timber shadow]

Can you share an example of a prompt that doesn't work for you? And do you have custom instructions enabled?

[rancid pumice]

I do not have custom instructions enabled

Some examples of prompts that don't work: "Draw a tree," "draw a wooden spoon," "draw a skateboarding squirrel," "draw literally anything"

they all give error messages

[timber shadow]

Hmm "Draw a tree" just worked for me, so if you don't have custom instructions enabled, maybe we can troubleshoot the device, browser, or network. Are you trying to use DALL·E on one device, or multiple?

[rancid pumice]

one device

[onyx kraken]

And are you using https://chat.openai.com/ or some other way to access the model?

[rancid pumice]

yes, chat.openai.com

[timber shadow]

A computer or a phone?

[rancid pumice]

a computer

[onyx kraken]

What's the error message?

[rancid pumice]

although, wait, i have attempted to use it on my phone, same problem

[timber shadow]

Is the phone connected to the same network as the computer?

[rancid pumice]

"Error creating image"

[rancid pumice]

No.

[onyx kraken]

I would #1070006915414900886 this, and include the screenshot. It seems possible that you may have an issue affecting your account that may need the help of an OAI engineer. This isn't a general outage, as other users can still make images now, and clearly you can talk to the model.

I'd include the url of the conversation, so the engineer can access it, the portion after the c, like c/4d4fac96-1326-4676-94bf-ca51624e13b7

[timber shadow]

So just to clarify: it happens on the computer on WiFi, and on the phone not on WiFi, but on a cellular network instead?

[rancid pumice]

correct

[timber shadow]

Very strange! I think Esk's advice is good, though I remember seeing a past post of yours in there already I believe? Now that I see your example. Just make sure you're following the bug report format (if you haven't already) and including all the devices/browsers/networks you've tried. Adding the chat ID also like Esk suggested is a great tip too! They can look it up on their end.

If you want to though: on the computer, you can try poking around in the Network tab of the dev tools in your browser. Perhaps there will be some clue there when it fails as to what might be happening, though the cross device, cross network instances make it perhaps less likely

[onyx kraken]

Oh! Yeah, I see the shared conversation in the bug report, great idea. That one shows custom instructions, but I would guess those usually won't cause a failure to make images.

[onyx kraken]

Another way to ask for help, and the official and best way: The bot bottom right https://help.openai.com/en/

Click through the choices, leave a message for a human. It can take them as much as a week to reply, but they should get back to you.

Bug reports here we don't have a timeline on; they can be that day, but also may not be answered. The website and bot are the way to officially and properly ask for help.

[rancid pumice]

sigh, i did try that but i guess i've just been impatient about it, i'm still waiting for a response

[onyx kraken]

Wishing you the best and help soon! We're just community members here, but we do care and want to try to help

[wispy olive]

What the hell is going on with chatgpt lately? Why the hell do I send him a prompt, but he doesn’t respond to me? This is the number of times I write a wall of text and this crap forces me to rewrite huge prompts over and over again, leaving me without an answer.....

[tiny heart]

Nope. Not initially but I turned it on as a possible solution but it did not work so I turned it off again.

[cold folio]

If you want, Could you please share the conversation link/url ? thanks

[winter ingot]

hi Guys, I need some help, why is that i keep getting Saving GPTs Error message? I dont have any files uploaded and no script in action. Been stuck for 12 hrs already

[cold folio]

Seems like you wanted to share the conversation URL (f.e chat.openai.com/g/1234) . But instead you shared a wrong link, Please remember to avoid sharing URLs that could contain tokens or sensitive information. I have deleted your message
Thanks

[tiny heart]

I can't access the conversation url. That's the only url after I try to log in.

[gaunt surge]

Chat GPT isn't working for me?

I send a message and the website doesn't show any prompt submission and I'm unable to send another prompt

• reloading the website and force quitting doesn't fix it

[glacial granite]

Why can't I seem to login to ChatGPT? It has been working fine before, on Microsoft Edge. The problem is that when I try to login in with Google, it just redirects me back to the Get Started login page.

Trying to reset my password doesn't do anything, I never get the email for it.

[urban compass]

Pleas help me! I keep getting a white screen when i try to log in, tried removing coockies, another browser, incognito.

[glacial granite]

Ok, clearing the cookies did it for me, strange why it didn't want to in the first place.

[wild dagger]

Hello. I want to fine-tune model. Can I use multiple prompt-answer conversations in one example in my training dataset?
For example:
System, P(prompt), A(answer), P, A, ... etc.

[light garnet]

Hi, does anybody know how to write the action schema (JSON format) correctly? Writing the action which have to get data from GET https API JSON response on dedicated server and I can't understand how make it for GPTs to understand

[rustic marlin]

the fine tune process just takes one message and one output in no particular order

[wild dagger]

Got it. Thank you!

[tame shuttle]

is there an outage on chatgpt for web?

[zinc condor]

on the web version of chatgpt, there is a new section called "speech". I chose "Juniper". but when I talk to chatgpt there is no sound. she doesn't speak. she only speaks when I click on the icon next to her last message. (an icon that looks like a speaker)

does anyone else have this problem?

[still matrix]

Hmm...something seems to have gone wrong. why can't I use chatgpt 4 team subscription?

[rugged wing]

All day like this. Any idea?

[timber shadow]

Hello, is this happening in both new and existing chats?

[heady stream]

Getting the same error on GPT-4 web, even though https://status.openai.com/ says the web service is fine.

[rugged wing]

Everywhere. But 3 is working. Even the app same .

App

[light garnet]

Help plz :3

[timber shadow]

On your phone, can you try disabling WiFi, connecting to your cellular network, and then logging in in a browser, in a private tab? Could help test whether it's a network thing.

[timber shadow]

If you don't hear back here, the people in #dev-chat might be able to help!

[rugged wing]

The problem doesn't go away, I've tried it:
Change network (from wifi, to mobile, and through a working VPN).
Clear cache in browser and app.
Log out and back in.

The problem persists.

[timber shadow]

Do you have a VPN installed on both devices?

[rocky rover]

@timber shadow I got access to android now, so I can help if you have something open, I can check

[rugged wing]

No, on my work computer. But I turned it off.
I've tried it:
On my laptop (web with vpn, without vpn).
On my phone (web with wifi, and mobile, app wifi, and mobile)

It doesn't work.

[timber shadow]

I gotcha, so when you did phone + web + mobile network, were you using a private tab? Just clarifying your cache troubleshooting so far!

[timber shadow]

Thanks Dys! Do you mean you can help troubleshoot w/ Android now?

[rocky rover]

yes, let me get my android device

[timber shadow]

In this case I think it might be a network or even perhaps account glitch as it's happening to them across devices and networks, so we'll see!

[rocky rover]

on the same device?

or different environments?

[timber shadow]

It's happening to them on a laptop and on a phone, on WiFi and on cellular, I think

[rocky rover]

WiFi over Hotstop or infrastructured network at home or work?

[timber shadow]

We shall see!

[rocky rover]

Also if VPN is not with a fix public unique IP, it can be a problem for some OAUTH and Credential/Payment services

[rugged wing]

Even in other browsers it doesn't work.
I think I tried all that I can

[rocky rover]

Is your VPN a comercial one or a work one?

[rugged wing]

I shut down the vpn. Nothing changes, and it doesn't exist on my mobile.

[rocky rover]

are you on windows or mac?

[rugged wing]

Windows

And Android

[rocky rover]

ok Press Win+X and open Terminal or Powershell wirth admin rights and type the following after you turn off the VPN:

ipconfig /release ipconfig /flushdns and then do ipconfig /renew and try again under Windows

[rugged wing]

The same

[rocky rover]

I can't think of any other ideas currently, if I get some ideas I will share my input

only one question, is that VPN something like NordVPN or ExpressVPN without a public private IP plan?

[rugged wing]

Well, I tried it on my Mac, under wifi, same error.

[rocky rover]

I'm wondering if it can be your ISP

is your ISP behind a CGNAT by any chance?

[rugged wing]

Nope, it's a private vpn from job, to connect to them

But Mac doesn't have vpn

[rocky rover]

But your current connection, is your home internet connection?

[rugged wing]

And phone too

[rocky rover]

can you disable wifi on phone and try again?

[rugged wing]

Yes, but I tired 5G mobile too

[rocky rover]

only on mobile data

[rugged wing]

I did

Not help

[rocky rover]

is your home connection also with 5G?

or other technology?

[rugged wing]

Nope, just wifi. And I tired with mobile data

[rocky rover]

I mean, you have wifi router, but is that connceted to your ISP through what kind of tech?

Or better yet, call your ISP and ask if you have a unique public IP with your connection

[rugged wing]

I have white ip

[rocky rover]

some ISPs are deplying CGNAT fast bnecause it's an easy solution to the IPv4 problem

CGNAT can bring lots of unique challenges, hence I ask

[rugged wing]

Many months use it and the problem arose today. I haven't changed anything.🤷🏻‍♀️

But from mobile not working too

[rocky rover]

Just ask, I know what you are telling me, it's just to discard ideas

[rugged wing]

Even mac connected to mobile network via hotspot, same error.(

[rocky rover]

I'm out of ideas

I will come to you later if something else comes up in mind

or if you have more information

[rugged wing]

Maybe they have problems? I saw some messages like mine

[rocky rover]

that can also be an issue, CALL, your GPT is waiting!!

[rugged wing]

If I just upload and send any picture first, the chat will open! I can just start the conversation then and it responds like normal in the gpt4 environment. If I start off first with a text or voice, I will still just get the errors. 🤷🏻‍♀️🤷🏻‍♀️🤷🏻‍♀️

[rocky rover]

sounds like a complex problem, if anything I would ask OAI if there's a diagnostics tool you could use to send a report

@rugged wingI asked https://discord.com/channels/974519864045756446/1214947598755045386 see if there was something, Apperantly there's no official tool.

[timber shadow]

That's strange! Nice troubleshooting though. You might check the Network tab of your browser's dev tools to compare when it works with Vision to when it doesn't work otherwise -- might find a hint about what's failing.

[rugged wing]

I wrote to help center 🤷🏻‍♀️

[timber shadow]

I've also seen OpenAI team members ask for a HAR via DM in #1070006915414900886, so they might be able to help troubleshoot that way too if they're able to respond to a post there. Just be sure not to share a HAR publicly because they contain auth info

The verification requirement was started awhile back to cut down on bots/spam/scams in the server.

It worked pretty well!

[rugged wing]

hello everybody, I am not an IT expert, but I have been facing with a not-at-all even opening / working GPT plus for 2 days.

[timber shadow]

Hello! Can you share a screenshot of the issue? Is it just a white screen?

[rugged wing]

The request has been there for like 2 hours...waiting.((

yes, nothing but a white screen Solbus

[timber shadow]

As a test, can you try in a private window of your browser, with all browser extensions disabled?

[rocky rover]

Heya, I saw your message and I know you are pressing on an answer. But give it a bit patient of people to read it. in the mean time maybe browse the #1070006915414900886 section, maybe somene else has your problem or even a possible solution while people read your message and come with a proper answer. (I'm not the one with that tho, sorry)

[pliant forum]

I can't send messages to chat gpt only on my specific account

It works on every other account, but when I use the account that I have plus on (so I can't just switch to a different account), I can't send messages

[pliant forum]

are you on brave

[timber shadow]

Hi, is the message just disappearing when you press send? And re: other accounts: do you mean 3.5 is working, but neither 3.5 nor 4 work on your Plus account? Or just 4?

[pliant forum]

yes, it is disappearing when I press send

and no, I switch accounts to see if that was the problem and it worked on that different account

using the same browser btw

yea neither 3.5 nor 4 is working on my plus account

[timber shadow]

The message disappearing behavior is the same for both 3.5 and 4? Just clarifying!

[rugged wing]

nothing is working, I payed for plus, shame on the company , they use your money but no service -- I am talking about chatGPT plus

[pliant forum]

same for both 3.5 and 4

[timber shadow]

Did you try in a private window with all browser extensions disabled?

[rugged wing]

for me too, ot neither

[timber shadow]

Gotcha! It's strange that one account works and another doesn't. What browser(s) is it happening in?

[pliant forum]

brave

checked microsoft edge and it worked

then I checked on a different account in brave and it worked

after i checked on the difference account in brave i went back to my plus account and it didnt work

[timber shadow]

So the account with the problem works in Edge?

[pliant forum]

i tried a different account other than the plus account in edge

but it worked when i switched accounts in brave too

i cant send messages only on my plus account

no matter the browser

[timber shadow]

So you've tried the Plus account in Brave and Edge and it doesn't work in either?

[pliant forum]

no, i've just tried the plus account in brave

i'll try it on edge rn

[pliant forum]

just tried it on edge, the plus account doesn't work

[timber shadow]

Dang, ok! Can you try in a non-Chromium browser (Edge and Brave are both Chromium) like Firefox or Safari as another test? Either a private window/no browser extensions, or a fresh install?

[pliant forum]

alright

[pliant forum]

plus account didn't work on firefox either

the problem is specific to the acount I have plus on

[timber shadow]

Okay, do you have another device and network you can test on by chance? Like a phone on a cellular network? Not discounting that it's possible there's some account bug that might be present, just giving some troubleshooting suggestions that have worked for others. If you want to reach out about a potential account bug, help.openai.com is the way to submit a ticket for the OpenAI team.

[pliant forum]

it works on my phone for some reason

[small idol]

Hi, my account was deleted this morning so I had to make a new one. There is no reason to delete my account, and I thouroughly searched the OpenAI policies and found no violations. The email states that I can appeal this decision at the help center. Yet, there is nowhere to be found to file an appeal or speak to anyone that isn't a chatbot that simply refers you back to the FAQ page. PLEASE #HELP!

[timber shadow]

Ahh interesting ok. Are your date and time settings correct on your computer?

[pliant forum]

how do i check

nvm

[dusk smelt]

You might want to email [e-mail]

[pliant forum]

just got it right

[onyx kraken]

That's no longer monitored, we were told. Support is through bot bottom right of the help page

[timber shadow]

I don't think that email is monitored anymore: #chatgpt-discussions message

[pliant forum]

it worked

thanks

[timber shadow]

Hooray! Thanks for confirming!!

[pliant forum]

thanks for helping

[dusk smelt]

Oh good to know, thanks 🙂

[small idol]

The help page, nor the support bot are of any help in this situation. I emailed Michael from one of the other posts. Hopefully this can get resolved. There seems to be an all around lack of support when there is an actual issue that can't be resolved on our own.

[onyx kraken]

I'm so sorry you're having these troubles!

Clicking through the choices of the support bot does lead to a way to leave a message for a human.

Here's one path, once you select messages choose:

ChatGPT
Other
Other
No
Report a Bug or Errors

you will see the message:

"For other issues, please provide as much detail as possible, including:

• A description of the issue
• Steps to reproduce it
• A screenshot of the error message"

And you can write a reply.

That gets passed to a human, and it can take a few days to get an answer, but that's the official way to get OAI support at this time.

[maiden bone]

Hey all - this has been happening to me all day. Only on GPT 4 though. Has anyone else encountered this so far? I'm reproducing this on multiple browsers / O/s and hardware

[rocky rover]

can you try to reload the chat and ask if gpt can explain the failure?

[onyx kraken]

That's not going to help, that's not GPT saying that.

[maiden bone]

Nope - when I reload, it just gives me the same regenerate button

[rugged wing]

i'm having the same issue

[rocky rover]

I know, but I have sometimes seen behaviour with a probable cause, hence the idea

it's just from what I've seen, even if it can be nuanced

[maiden bone]

I noticed that GPT 4 was being flakey the past three days.... every few questions I'd need to refresh to get the answer. But it's just entirely given up today

[onyx kraken]

So, you didn't mention if you tried a different network, that's something to try.

This is a generic error message, and can have a bunch of causes.

One thing that's helped some people is a fresh install of Firefox.

[rugged wing]

i've tried a different network and used a vpn but still get the same error

[maiden bone]

i've encountered this on Brave, Chromium, Firefox and Edge. Brave and Firefox on RHEL 9, Chromium on Clear Linux and edge on a windows 11 laptop 😦

[onyx kraken]

This is the list of stuff to try that helped most people, back in December-ish when I was last trying to help with it:

Usually caused by a VPN program. May have to actually turn the program off, not just stop the VPN, to see improvement.

Can also be caused by browser extensions, network settings.

Can see if different devices, different networks can connect.

How about your phone, with either web or the app?

One person reported just restarting their computer was the fix they needed (after we tried a bunch of ideas that helped others, but didn't help them).

A great many people find that a fresh install of Firefox fixes their issue, if they were not running a VPN program.

I do recommend #1070006915414900886 , some gold-named OAI staff do check there, and hopefully can help.

[maiden bone]

It potentially is a network issue. I have no other networks to try currently as my 4G signal is appalling, but will try from my phone when I move to an area with signal. the only other thing I can think is that my OpenWRT router may be screwing something up.

@onyx kraken I appreciate the guidance. Do Open AI offer any support diagnosing network issues?

[rugged wing]

i've tried with 4G on my phone and nothing has changed sadly. i've seen other people are having the same issue as well in bug reports

[onyx kraken]

There is help through the website and through #1070006915414900886 . https://help.openai.com/ to talk to the bot bottom right; click through the choices until you can leave a message. It can take days to get a reply, but they do reply and help.

[small idol]

Thank you. This has been a nightmare trying to resolve this. It's crazy to think that they don't even offer an explanation or description of the specific violation before deleting your account. I find it hard to believe there was one in the first place, but at least a clear course of action to take for an appeal process would be helpful.

[rugged wing]

i've just read that uploading a picture at the beginning of a new conversation seems to help?

[maiden bone]

Ahh gotcha. Yeah I'm already going through the motions there. We're still at the "delete all of your cookies" stage

ultimately, I'm just happy that I'm not the only one. Misery loves company and I can still access the keymate interface

[onyx kraken]

But I don't know if there is specific and exactly help with diagnosing network issues.

I'm a community member like yourself, and I'm not very informed about computer science in general.

I just saw a lot of folks once upon a time reporting this bug, and decided to collect solutions and offer them to those that I saw were affected in the future.

[maiden bone]

Bloody hell it worked

that's mental

[rugged wing]

yep

[onyx kraken]

I'm so sorry you're having this issue. You're the second person I've seen recently state on the forums they simply had a deleted account, no email or other information.

[onyx kraken]

Woo, I'll add it to the list of 'and someone said this helped them' 😄

[maiden bone]

So that kind of points towards it being entirely an Open AI side issue rather than a me problem 🙂

[rugged wing]

yeah. a lot of other people seemed to be having the same issue today as well so probably an open ai thing

not sure why it isnt on their status page

[onyx kraken]

I would definitely bug report it, and say 'this workaround seems to help, but I still need help because I can't upload an image every new conversation' - for example we can't upload images to 3.5

[rugged wing]

3.5 seems to work

[maiden bone]

meanwhile the open AI bot continues to gaslight me and tells me to clear cache when I've already told it that I've replicated this on multiple browsers, OS and hardware 😄

Yup. 3.5 workis fine

[onyx kraken]

mmm, I wouldn't talk to the BOT, but leave a message for a human... did they change that?

[maiden bone]

It might be a human replying. I just assumed it was AI! 😄

[onyx kraken]

The previous help pathway involved leaving a message for a human, not discussing matters with the bot, from what I was told.

However, I don't see the 'expect a week for a reply' message as I'm looking at the type message in area, so... sighs I don't want to waste OAI time exploring their help system either.

and I don't currently have a bug to discuss!

[small idol]

I think it's definitely a bug they should consider fixing. Or at the very least make it easier to appeal the decision/provide some specific violation.

[maiden bone]

Hey at the end of the day, I don't have to spend the next hour digging through firewall logs so I'm almost as thrilled as if I'd completely resolved the issue 🙂

[onyx kraken]

There's previously been a email you get, if OAI suspends your account, and it does discuss an appeal process.

I don't know what's going on with yourself and the other person who recently reported their account was just deleted.

I don't know how you can tell it was deleted vs suspended. I sure hope no hacker or something instigated a delete or other 'this isn't OAI either' going on. I have no idea what's affecting you guys, or why. Your stories, similar to each other, are not like earlier reports from users talking here about suspended accounts.

And I'm just a community member too, I just see what people choose to share, I don't know what's going on either.

[small idol]

I found out that the account was deleted when I tried to log in after I got the original error message and it told me the account was deleted. At this point, I don't have any more information than that.

[onyx kraken]

Ouch. Well, the earlier folks with OAI-suspended accounts seemed specifically talking about suspension, so I don't know the cause of your situation, and I sure hope you get a resolution.

[small idol]

You and me both. That's over a year's worth of data collection and project threads I'd rather not have to regenerate with a new account. On top of that, if it was deleted due to a legitimate policy violation I have n oway of preventing my new account from the same fate without an explanation.

[onyx kraken]

I've heard nothing to make me think OAI 'deletes' accounts.

Suspends them, yes. And the help system calls this 'deactivated', this is the only helpfile I see about suspension/deactivation: https://help.openai.com/en/articles/6378378-why-was-my-dall-e-2-account-deactivated

Deletion appears to be a user-instigated action, not a OAI action. https://help.openai.com/en/articles/6378407-how-to-delete-your-account

That helpfile does include this information:

This will happen either because you deleted your account in ChatGPT -- or -- you deleted your account by submitting the email to our Privacy Center for deletion.

If you previously deleted an OpenAI account, we unfortunately don't support creating a new account with the same email or reactivating the old account.```

I believe you - you didn't delete.

It appears OAI didn't delete either.

Which is why I'm worried about a possible compromised account and that action taken.

[small idol]

Yes, and this was the information that wasn't helpful to me. Obviously, I didn't delete my own account without knowing it, and since I wasn't informed of any violations previously I need some help in sorting out the situation. This is where a support team would come in handy. There are only two logical possibilities here: A technical error on the OpenAI side, or a security breach on Google's side. However, I have multifactor auth on everything you'd need to get to my OpenAI account so I find the latter unlikely.

[onyx kraken]

you definitely need support's help, and going through the bot is the only way I know of to get that, other than #1070006915414900886 and if a gold name answers you there. That's very much the secondary and less responsive method, based on what we've been told. I really hope you can be helped and soon

[faint hornet]

Has anyone else been getting “Hmm. somehh to big seems to have gone wrong” non stop for chat gpt-4 it just won’t work at all

[small idol]

I've reached out to just about every avenue and email I can find. I also submitted my situation through the bot. I'm hoping someone can resolve this, or at least fix the issue before more people run into the problem.

[onyx kraken]

Here's the list of possible fixes I know of:

Usually caused by a VPN program. May have to actually turn the program off, not just stop the VPN, to see improvement.

Can also be caused by browser extensions, network settings.

Can see if different devices, different networks can connect.

How about your phone, with either web or the app?

One person reported just restarting their computer was the fix they needed (after we tried a bunch of ideas that helped others, but didn't help them).

Another person stated that uploading an image at the start of the conversation helped them.

A great many people find that a fresh install of Firefox fixes their issue, if they were not running a VPN program.

[faint hornet]

I’ve tried all of that

[onyx kraken]

It's a very variable, generic error message, so can be a search to fix.

Even the image upload?

[safe ibex]

I'm getting the same error as well.

[faint hornet]

I saw someone in the report a bug a lot r getting it

The image fix just worked. Do I have to do that every time

[onyx kraken]

I don't know; a few minutes ago was the first time I heard of that fix, or that it worked for at least 1 person.

[faint hornet]

Okay, I have something time sensitive I need some chat gpt help with

[tropic coral]

The image fix worked a couple of times for me.

[faint hornet]

Does it work all the time?

[tropic coral]

No. Not always. 4/5 so far

[faint hornet]

When it didn’t work did u just make a new chat and try again?

Or did u have to wait a bit

[tropic coral]

I clicked on regenerate, stop before it replies, then upload the image again so I'm in the same chat

[faint hornet]

Kk

[timid matrix]

has anyone else been getting this error message? It started yesterday 5-6pm PT and I continue to get it this morning.... am a plus user and account is in good standing with openai

[timber shadow]

Hi, I saw you posted here too: #1214920393471561728 message
Does the workaround a couple users in there have mentioned work for you? Can you get it working if you attach an image to your message in a new chat?

[timid matrix]

Thank you. Just saw one there as well and posted my issue on this bug thread https://discord.com/channels/974519864045756446/1214920393471561728

[tropic coral]

Updating my Firefox seems to have fixed it. It's very slow but atleast I don't have to upload an image everytime

[onyx kraken]

Fresh install of Firefox specifically has been reported to help other users with this problem, from back months ago.

[timid matrix]

That worked for that new chat but does not resolve the issue for other existing or new chats w/o an image.

[violet bear]

Hello guys! I have been dealing with an issue since always and I was wondering if someone has a solution for it:

When I copy / Paste any text from chatgpt and the text is bold, if I paste it on gmail for example it paste it like this * * Text * * and I always have to remove the "**" and bold the text myself.

Thanks for the support!

[wet anvil]

chat gpt charged 98 dollars for me in this month

is this already happenned before?

what can I do? ...

[onyx kraken]

https://help.openai.com/en/articles/7231930-how-can-i-contact-support-about-billing

[main falcon]

this would be the right way to calculate cost of tokens right? im using gpt 3.5 turbo 0125 which is 50 cents per 1m tokens and using 3000 tokens for a request

[dusk smelt]

I believe that would work, however you might wanna add some more tokens since it also couts a couple of tokens for each name (assistant, and if provided user) and the message start and end tokens, but that's precise enough probably

[tiny heart]

This is day 2 that I have not had access to Chat GPT

[main falcon]

oh ok, thanks

[rugged wing]

Same with me, for 2 days I cant reach chat GPT, just a white screen -- and I am a plus member, shame on them, we payed for nothing

[tiny heart]

I am a plus member too Honeybee

I checked ALL the things to try and remedy the issue but nothing has worked

[rugged wing]

me too, as many as I could, but I am not an expert, so no more possibilities but wait

[tiny heart]

Its a bit frustrating for sure. Waiting for my email back from a real human

[rugged wing]

Now we can see how it is NOT working

[tiny heart]

I CAN access my account from my mobile app so if I am DESPERATE to use it, I can do it on my phone

[rugged wing]

I wanna used what I payed for, but maybe I want to much 🙂 I prefer PC

[tiny heart]

lol. No, your absolutely right. I prefer PC too, it fits into my work flow.

[spring charm]

I'm on the playground openai area, and I am hearing that running api calls is cheaper than the standard 20$ monthly service. Is this accurate?

for context, I am a student trying to learn prompting.

[onyx kraken]

Totally depends on how you're using it. Be aware, as the conversation keeps going, the context, the previous messages, get sent too, up to a limit, but the limit's fairly large (depending on the model).

I have heard of single back-forth exchanges that involved 1 million tokens for that one cycle of input, output. And some uses, like RAG, may be more token heavy than others.

So be thoughtful and aware of your use.

If you understand how tokens and context work and know you need just a little access - maybe consider the free chatGPT account, which gives loads of use with ChatGPT 3.5 and absolutely no charge; occasionally you need to wait an hour before you can use 3.5 again.

if you need more than that, consider carefully how well you can control your use of API, that absolutely could be cheaper than $20/mo - or vastly more expensive, hundreds of dollars in a month, depending on your use.

[rugged wing]

Well.. it looks like the problem has been solved, and everything is working now.. it took all day. I still have no idea what it was.. 🤔

[little osprey]

Why chatgpt can't enter the url? 🙃

[dusk smelt]

It wold be some kind of security issue if chatgpt was able to open arbitrary urls. Bur if the article is well referenced on bing, you can tell him to look for it.

[little osprey]

I tell chatgpt to search for it in bing by putting the url?

[dusk smelt]

Yes, but again it needs the website to actually appear in the search results

[spring charm]

the free portion with 3.5 is a basic terminal or is that also free on the playground area?

[onyx kraken]

Yeah, I've never worked with the API/playground, but read some of the helpfiles and heard what others discuss sometimes. I believe at one time they gave new accounts some credits to work with, I don't think that still happens.

I started with a free ChatGPT account, upgraded to plus once that was an option.

ChatGPT is completely separate from API/playground, and the Dall-E 2/labs is another completely separate area as well. Each would be paid for separately, if you wanted to spend money on them

[mental dust]

I'm getting a slightly different err - just tried accessing chatGPT from a new laptop.
As soon as I entered my credentials - white screen.

• tried on original laptop - same white screen
• tried in android - an error has occurred, pls try again

Related to the issue above?

[proud drift]

I’m also getting an error on iOS and web app. Tried clearing browser history and cache. Tried on private. Dowsing. Not on a VPN. No at says something wrong and to retry

[mental dust]

clear cache worked for me

[vapid vortex]

Chat GPT don't open links anymore?

[rocky rover]

hey everonye i just saw this popup in the official forums https://community.openai.com/top?ascending=true&order=activityso, so be patient until is solved

[wintry echo]

I'm curious if I can help with this, but

[empty walrus]

Hello, how can I apply for using Sora AI?

[golden ruin]

it's not out for the public yet

[wraith zodiac]

chatgpt is telling me that this is running in the background: While I can't provide real-time updates or show the progress interactively, rest assured that the task is being handled with the utmost diligence. Once the review and categorization process is complete, I will promptly provide the updated "categories" file.

I appreciate your patience and understanding as we work to ensure the categorization is as accurate and helpful as possible.

is this for real? or is it not doing anything?

[trail cairn]

its not real

[cold folio]

That's a bug unfortunately. Please use the thumbs down button and start a new chat 😊

[hushed quail]

I've got to assume background work is some feature they're planning to add and for some annoying reason they've pushed the model change before the feature appears.

[timber shadow]

It's been prone to hallucinate "working on it in the background" since launch as far as I can remember. Early on people were prompting it to "generate 100 replies in the background and then reply to me with a synthesis of all of them" and it would happily report itself to be doing so. I remember personally asking it to "work on a way to connect yourself to the internet in the background" too 😁 and it would tell me about the progress it had made. Fun hallucinations, but hallucinations nonetheless!

[hushed quail]

Interesting. I had literally never gotten this hallucination until like 3 days ago. I had assumed it was a hey, all this processing takes time and so lets let it work on multiple things at a time in some manner.

[onyx kraken]

Very understandable, and if you don't know how the model works (and doesn't work) it's believable too.

As is, the model only knows the one conversation it has with you; it doesn't have 'space' or 'opportunties' to work on things in the background - but it is predicting the 'right' words to answer you with.

Humans say to humans, "I'll get back to you on that!" and "I need time, you'll have it Tuesday."

So the model may emulate that, but sadly, that's not needed, working, or anything.

You can tell the model, "Hey, Tuesday morning, now, where's my stuff?" if you want to continue, or otherwise guide it to give you the stuff you want.

[placid coral]

is anyone else seeing issue getting pdfs and files attached to messages to be indexed for retreival?

I'm also seeing an issue that initially the assistant can't see the file and says "I'm sorry, but I can't directly access the file you've uploaded to retrieve its name through my current tools. However, if you're looking for the title or the main subject of the content within the file, I can assist with that. Could you please clarify if you need information from inside the file or just the file name as it was uploaded?" or simlar, but if I ask a second time it does work

via assistant api

a minimal repro in the playground

it just seems hit or miss

[rough scroll]

Why can’t I change my email on my open AI account? I understand has something to do with perceived security and privacy. However, in an email exchange they continually dodged my requests on four separate occasions for specifics and further elaboration, all they could say, was you can’t change your email address because of security. It’s like they’re unable or unwilling to explain. I don’t understand why. I have countless other accounts where I can change my email address and it’s not a security risk. So what’s the big deal? And why can’t they give any information about their policy other than that they have a policy? It makes me wonder if it’s more for their security than mine and if my data is mine or not

[cold folio]

Hello I'd suggest exporting the data, and making a new account with the new email.

[rough scroll]

Yes, I thought about that but they won’t let me import the data into the new account.

I don’t understand why they’re so inflexible and unwilling to provide context to their policies

[keen anchor]

Whats the deal with the app android app? Its been broken for weeks

[timber shadow]

What problem are you having with it? It is working for me.

[cold folio]

Hello there

Please report any errors at Google Play store or Apple App store

[keen anchor]

Nvm i think it broke when i changed my password a while ago, i had to clear the data manually so it would log me out, for some reason reinstalling didnt do that

[hard yarrow]

So i've been trying to get GPT4 to generate dummy JSON files for me and it constantly only generates part of the file. For example I'll ask it to generate a JSON of 20 fake doctors with offices, contacts, etc. It proceeds with the task, and then delivers a JSON file with only 3 entries. Sometimes it will even deliver one that is completely empty. Has anyone else experienced this kind of error? Would appreciate any pointers. I have GPT Team through my work.

[onyx kraken]

So, GPT is best at small tasks, which can be sequential. The model will do much better if you're willing to do this in more than one input from you, with it adding more categories and their information with each pass.

I'll show a mockup, it won't be your exact locations because it's not easy to copy/paste from a screenshot.

Also, you're asking a HUGE amount from the model, and it's really not designed to do this outside of a story.

you're literally asking it to create, in 1 go:

20 simulated ID numbers
20 names
20 business addresses
20 email addresses
20 phone numbers
20 varying number of years of experience
20 different 2-sentence biographies
20 sets of affiliations with clinics and hospitals (I presume 1-4 each?)
20 sets of 1 or 2 areas of medical specialty
20 sets of 1+ languages they speak; all will speak English and some will speak a second or third language too (I am guessing you intend the model to do?)
And 16 of the 20 have a 'certified high quality' badge, while 4 do not.

This is a huge amount of information for the model to track in one input-output cycle. It doesn't count well, asking it to count to 20 over and over again ALONE is super hard for it; in this it has to count to 20 repeatedly while doing all this other stuff too.

[hard yarrow]

Thx for the inisight! yeah I did suspect it might be a bit too much to ask all at once. I'll see if taking a smaller bite will help.

[onyx kraken]

Here is how I would start with the model, including making it clear I didn't expect instant perfection and we would be iterating this to get to perfect, and reminding it to stop before it runs into length limits or likely network errors (because it absolutely will work until it is forced to stop, and then you don't get all your output'. Here's a reasonable start, how does this compare to your first attempts? https://chat.openai.com/share/9dd27f47-ac51-48be-ae0f-3d1eae91390f

[past tendon]

why I get this error?

[maiden frost]

chat.openai.com/:1 Uncaught (in promise)
l {revert: undefined, silent: true}

Today ChatGPT Web Error?

[vast junco]

if you have more infos about the effect of this error, you can open a bug in #1070006915414900886. Simply make sure to follow the template

ChatGPT shared link: (optional)

**Steps to reproduce:**
> 1. Sample text
> 2. Sample text

**Expected result:**
> Sample text

**Actual result:**
> Sample text

**Additional information**
> Browser:
> OS:

[vast junco]

you can verify if the issue is listed in #1070006915414900886, if not you can report this and you should receive support there. simply follow the template I posted above ^

[weary glacier]

Hello everyone. Whenever I enter a prompt in GPT4 it doesn't send my prompt. Does anyone has the same issue? I can't find ChatGPT4 support or status page.

[worn wedge]

chatgpt is sooo slow

[tidal root]

facing the same issue here

[drifting tangle]

same here.

[solar hollow]

Hey there,

I've noticed that professional picture generation websites are currently trending. I've been searching for a method to generate professional headshots based on some pictures. Is there a way that I can utilize DALL-E for that purpose?

Looking forward to hearing your insights!

[zenith basalt]

I am a GPTs maker. Today, I received a lot of emails, and users told me they didn't get any reply.

Several of my GPTs have received similar feedback.

Did something happen?🫣

[trail cairn]

you cant use dall-e for that

[solar hollow]

What should i use, do you know any other method

[trail cairn]

stable diffusion

[deft cobalt]

i cant send anything on chatgpt

if i send something it disapears

why?

[limber anvil]

I have the smae issue, questions disapear

[deft cobalt]

i have gpt plus

[dense lake]

I can run chatgpt from mobile. I tried it from 2 different PCs and it gives the same error. There may be a system error, please check.
Can you help me?

[drifting tangle]

Check solution here.
#gpt-models message

[lucid lion]

Sorry if this has been asked before, but do system level custom instructions interact with custom GPTs? Or do custom GPTs ignore custom instructions

[uncut aurora]

From that what I know GPTs doesn't know any about the Custom Instructions.

[timber shadow]

I've tested this before, Frreiheit is right! Only had "talk like a pirate" for custom instructions, worked as expected in normal chats, but didn't happen in any GPT. I even made a GPT with completely blank instructions just so no chance of interference, no pirate talk.

[wooden cave]

From the two 429 errors you're getting it looks like the server is telling you it does not have enough resources to accept your request at that time.

Which could include rate limiting

[modern yoke]

Hello. I can't use chatgpt 3.5 all day long. When I enter a message and click "send", nothing happens.

[dense lake]

I can use it on the mobile application without any problems. The problem only occurs when I log in from the web. I just wrote hello. How many resources might be required?

[wooden cave]

Try this #community-help message

[rigid musk]

From which country ?

[old rain]

My login for the open ai website isn't working. I deleted my cookies and cache

[rigid musk]

What error do you have ?

[old rain]

I click the log in button and it doesn't bring up anything, the screen just goes blank

[rigid musk]

What browser do you use ?

[old rain]

chrome

Ah I just got it... My cache clear and cookies clearing was set to 1 week. I set the timeline to longer and the problem must have been stored from an early date

[rigid musk]

Glad it works :3

[true merlin]

I logged in this morning and one of my custom GPTs I created for work is missing. Others are showing up in my list, but not MTA | GPT. Any ideas?

[timber shadow]

Hello, are you currently checking at https://chat.openai.com/gpts/mine? Do you have a direct URL for it in your browser history? If so, what happens if you try to navigate to it this way? The GPT name is appended to the end of its URL so you should be able to search for it that way.

[true merlin]

Yes, I'm at that url and its missing from the list. I found the URL in browser history, however, the MTA | GPT link takes me to MTA | Paratransit Helper (a different GPT I created for my workplace).

[heady tinsel]

Hi, I'm having a problem ,where there will be nothing happend after I send a chat, both 3.5 and 4, but if I create a new account and use the 3.5 it will work

[onyx kraken]

I don't know about creating a new account, however there's a bug reported earlier today similar:

They're working on this, a gold name has answered in bug reports. https://discord.com/channels/974519864045756446/1215621063677976656

It's a great idea to post bugs in #1070006915414900886, the gold-name employees tend to look there, where the reports don't endlessly scroll away!

[heady tinsel]

ok, it acctully resolved after a cache clear, I first tried in three different devices with 3 different broswer, a pc a mac and a ios, didn't thought would be a cache problem

[dense lake]

Says;
"I turned off the language preview and switched to auto-detection instead.
Could there be a problem with the language preview?
Changing the Language (Alpha) option back to Auto-detection fixed the problem for me.
settings->General->Language (Alpha)"

However, there is no language selection area. What will I do?

[onyx kraken]

I suspect the language option is regional, my settings look like yours, no way to change language.

However, some users months ago showed screenshots of being offered a way to change the language, I think in some supported countries some of those users have language setting options.

[shy helm]

Hello,

I can't send promts to ChatGPT 4 on the web. There is no problem in the mobile app, but it doesn't react at all over the web. I have tried on different computers, browsers and networks but the result is the same. Do you have any solution suggestions?

[onyx kraken]

there's a bug reported earlier today similar:

They're working on this, a gold name has answered in bug reports. https://discord.com/channels/974519864045756446/1215621063677976656

It's a great idea to post bugs in #1070006915414900886, the gold-name employees tend to look there, where the reports don't endlessly scroll away!

[wanton salmon]

Anybody knows the reason why Dall-e (chatgpt4) can not (or does not want to) draw arrows pointing LEFT? (Any arrow without specs will be a right arrow, but if I ask it to point upwards it works, and then when asking for a LEFT arrow it will be a right arrow and then it lies about it in the description...)

[onyx kraken]

The model in general struggles to tell left from right. Pretty sure it's not trying to lie, it's just confused.

However, these worked for me, but they're definitely cherrypicked, about 20% point correctly to the left:

In a lush green forest, a clear stream cuts through the scenery. Resting on the banks of the stream, there's a stone arrow, moss-covered and blending with the natural surroundings. This arrow, chiseled from granite, points left with its tip indicating the direction to follow. The water from the stream reflects a part of the arrow, enhancing its presence in this serene landscape. It's as if nature itself is guiding the way west, through the tranquil beauty of the forest.

In a historic cobblestone alley, lit by the warm light of hanging lanterns, a wrought iron arrow, ornate and elegant, points directly to the viewer's left. This arrow, attached to an old brick wall, is intricately designed with swirls and floral patterns, reminiscent of a bygone era. Its presence adds a touch of historical charm to the alley, guiding visitors westward with a blend of artistry and function. The lanterns cast soft shadows on the arrow, enhancing its detailed craftsmanship and creating an atmosphere of timelessness and mystery.

Since it is a common problem, affecting nearly every prompt, I recommend fill out https://openai.com/form/chat-model-feedback

[onyx kraken]

The GPT model can't see the image that Dall-E made. It knows the prompt it sent, and it's likely hallucinating that it can see the image it described - but it has no idea what Dall-E actually created.

[wanton salmon]

Thanks. I had 0% success in the first 3 days of trying. But I would like to have text written on the arrow, which makes mirroring unusable.
BTW if it does not understand left and right, then it should be some 50%-ish chance not "winner takes it all".

[wanton salmon]

That was my guess too, but still if they act like one, they better share some context instead of guessing. (I mean it is still a lie if it states something false it doesn't know and doesn't care about at all.)

Any tips forcing the left arrow (with text written on it) would be welcome.

[onyx kraken]

I have no tips, because the model itself can't tell and by us, can't be trained.

It's the Dall-E model that is confused. Best we can do, I think, is report it 🙂

[wanton salmon]

Thank you, I reported most of the wrong answers already. (So the left-right anomaly might pop up in the stats eventually.)

Let's just hope they sort out LEFT and RIGHT before diving into self driving cars.

[cedar skiff]

Every time I want to write something, it says this. Even though I create a new key, it still writes it, any tips, please?

[onyx kraken]

Hehe!

Well, hallucination is indeed an issue, and can be part of the model's response.

Not sure how that works with self-driving cars, those are surely trained differently, and definitely have different AI models guiding them.

[wanton salmon]

They might have LEFT and RIGHT side AI-s to avoid confusion. :D

[cedar skiff]

Problem might be fixed, my credit balance was negative.

[wanton salmon]

Hmm. Surprising. Since it should have a different message popping up somewhere then.

[cedar skiff]

yeah I did a little research and found this:

[wanton salmon]

(But I see that from the API this is an invalid key. On the admin page you are free to create keys, just don't have the balance to use it. Kinda strange, but it's probably different teams working on them.)

[timber shadow]

If you're curious, there's good info in the research paper about why there are limitations like this currently. https://cdn.openai.com/papers/dall-e-3.pdf

Basically, since DALL·E is trained on images + their descriptive captions, and since the training data set of captions didn't always include perfect spatial information for their respective images, the "downstream model", DALL·E, likewise doesn't have perfect spatial awareness.

[cedar skiff]

Still having problems with this, credit balance wasnt the problem I guess

I sent 10$ of balance credit in there. Do yall think it needs some time to fix?

[wanton salmon]

Thank you. I also think these are two independent issues too. One is "not knowing what is a "LEFT ARROW" as in every clipart collection since 1992, and the other is the spatial awareness problem in the publication for relative positioning, but my request was an absolute left and absolute right which should be very basic info. (The synthetic captioner could be easily upgraded to say: if center_x<0.3 print "there is a red car on the left side of the image" )

I'm not an expert, but absolute left and absolute right coordinates on an image seems fairly trivial to label (esp. compared to all the magic they already sorted out)

[wanton salmon]

I don't know the API, but I would try to triplecheck if:

• I have the permission to use that specific api feature
• my key is copied exactly
• the key does not have unnecessary prefixes, postfixes, spaces and it has all prefixes, suffixes necessary
• the key was created so that it allows the access of the specific endpoint (not limited to other features)

[cedar skiff]

It worked so perfectly until now, I dont know why its not working

[timber shadow]

You might be interested in section 2, "Dataset Recaptioning" specifically! I don't understand all of it myself.
Edit: and maybe 3, "Evaluating the re-captioned datasets"

[cedar skiff]

Everytime I enter new key, it always ends on "sNeg"

It should be the same key if I literally create a new key and enter it there

Oh and now I just re enter the key and it works

Yeah it works, sorry for wasting yall time. Stupid bug.

Enjoy the rest of your evening. ^^

Or evening atleast for me.

[wanton salmon]

Was that a space? Or CRLF? :D

[wanton salmon]

Thanks, I'll take a look, but I'm too tired after spending 3 nights on a LEFT ARROW :D

[shrewd pelican]

hello I have a bug every time I write a message on ChatGPT. It doesn't respond and shows an infinite loading bar on the right. I have already cleared the cache and deleted the cookies, but it still doesn't work. When I create a new account, it works. The problem is that on my main account, I pay for gpt 4

it's strange that it works on a new account and not on my main account with chatgpt plus

help 😢