chatgpt-discussions

[shadow bolt]

git one? I was pondering on that - worth it?

[wide peak]

Not understanding what u mean @jade nova

😂

[shadow bolt]

GPT is slowly getting more and more savant - is what they meant @wide peak

[gritty dune]

I haven't found any of the alternatives match the level I can get with GPT4 for my purposes at the moment, so I'm hesitant to go to any of the other models. GPT4 has gotten REALLY good at character creation and maintaining consistency and I'm using it to solo campaign D&D right now (At least I'm not using it for work purposes like a lot of y'all)

[wide peak]

Yea it is but is not working lol it looks good for sure but won't do the work lol is like slacky labour on a job site

Like....please push the wheelbarrow please push it

Lol

Just goofing and joking

[gritty dune]

I think my issue is that it works so well enough of the time that investing in something that just "meh" most of the time but always works is such a let down?

[shadow bolt]

I am agree with @jade nova that looking from the outside in this is what is happening.

I feel what you're saying but also I am trying my best to keep my tinfoil hat in the closet. If you catch my drift

[wide peak]

@jade nova I do think they work on it for sure...but start to feel that not everyone get the same capabilities

Like firefly is pristine for the price they charge

[shadow bolt]

as in: just because we don't have a full grasp shouldn't mean we speculate this far

[wide peak]

We are talking for a paying customer point of view

[shadow bolt]

it's not ideal atm, have no clue how it comapres

[wide peak]

See ? Lol

Exactly and time

[shadow bolt]

I will put a spanner in the works though by saying that it is OFTEN xy problem

but.. even when you KNOW what you're doing you get nonsense

[wide peak]

Is 360$ per year lol

[shadow bolt]

both on 3.5 and 4

uncooperative is the word I'd use

before I joined I thought it was just me

[wide peak]

No is a long time deal

Many think that basic useres are not being treated the same and are restricted

🤣

[shadow bolt]

today's example: I asked about simple REGEX expression of a word combination. I structured promp neatly - the only thing that was required to finish the task is teh knowledge of the syntax. GPT skips more than a half of my request and only does the first 40%. Omits teh rest

😆 brilliant

yeah, I ended up DIYing or asking for sub-snippets

[rugged igloo]

bland robotic and not intresting

[shadow bolt]

not the NPCs, the joke! 🤣

[rugged igloo]

LOL my bad

[shadow bolt]

made it even funnier!

[gritty dune]

I have had 0 trouble getting NPCs to be more than polite robots lol. One of them literally punched my character in the face and laughed about it on his own accord. IDK if I'm just good at getting GPT to give me in depth stuff, but like, those cutting words were cutting lol

It did!

[shadow bolt]

for all it's capabilities it feels very... "unimaginative".

[rugged igloo]

thats why i like gemini so much for creative writing

[wide peak]

Gemini?

Mmmmmm xD

[shadow bolt]

the shmoogle thing? Huh.. will check it out! Thx!

on the list

there's too many 😆

[rugged igloo]

there the gemma now that was just open srouced by google but its not NSFW

[maiden fractal]

#ai-discussions best place to talk about other ai

[rugged igloo]

facebook does that also

it goes with the dead internet theory

[gritty dune]

I promise, yeah. My GPT can say most out of pocket stuff. I love it. 😂 It can be very mean. Between custom instructions and custom GPT stuff, I've fine tuned this thing to pieces.

Yep

[rugged igloo]

imagine getting hyped up for an ai to swear like its some taboo thing for it to do so stupid

thats all big snowflake companies

google applle, microsoft

back in like 2006 days on xbox live in cod it was the norm and nobody cared

now chat gpt will lecture you on how its wrong. like excuse me a bot does not have the right to lecture me

[gritty dune]

Politics doesn't have anything to do with it. Corporate culture is a whole different beast.

[rugged igloo]

"family" is more like it

and gender also \

[eternal rock]

<:book_icon:1171408210398289941> `` Rule 6 `` No political and religious discussions or content.

Users from all over the world frequent this server. To maintain a respectful and civil atmosphere on our server, please avoid all religious and political discussions or content.

[rugged igloo]

exactly

[loud path]

it seems like my public IP has been blocked in ChatGPT, no response when sending chat, if i change ISP location it works

[honest jay]

Hey! Can you share a screenshot of the issue over in the #community-help channel?

[rugged igloo]

i already have Ray it gives me like a essay of lecturing me

[loud path]

a screnshoot of a chat not working??

no error message

[honest jay]

Yes! I'm specifically wondering if your chat history appears to the side or if you have a more "blank" page.

[loud path]

ok

[eternal rock]

This is a Safe for Work Discord Server

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

Posting NSFW content will result in a permanent ban.

[wide peak]

have a good day this going to far lol

[eternal rock]

  Attention  Off-topic chats have been cleared by a moderator.

<:book_icon:1171408210398289941> `` Rule 6 `` No political and religious discussions or content.

Users from all over the world frequent this server. To maintain a respectful and civil atmosphere on our server, please avoid all religious and political discussions or content.

<:book_icon:1171408210398289941> `` Rule 1 `` Be respectful.

Treat others the way you would like to be treated, and assume best intentions. Don’t harass or attack others, and don’t engage in hateful or generally malicious behavior (e.g. sexism, racism, homophobia, etc.). Keep the negativity to a minimum.

[wide peak]

keep your integrity and learn from personal experience not from internet

[rugged igloo]

chat gpt is on full patrol today cant take the heat

it seems the more files i have the harder chat gpt does to look for it

kesku sempai i need your wisdom

[steady holly]

I thought I'd throw ChatGPT a curve ball, certain it would be unable to do the task. Man, was I wrong.

[olive flame]

curve ball

[rugged igloo]

3d print tht

[steady holly]

I did 🙂

[rugged igloo]

niceu

[eternal rock]

`` Server Tip: ``

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

[steady holly]

Waiting for a new FEP sheet to come in to finish it

[rugged igloo]

nice mate

cant wait to see it finished

hey chat gpt buy me a 3d printer hmm something went wrong....

[steady holly]

Its only my second time painting a model. So temper your expectations

[mellow folio]

how do you get chatgpt plus? when i try to buy it the page just shows my payment method, billing information and invoice history. there is no button to purchase

[rugged igloo]

for your second time ? my dude should make a store

[steady holly]

bro thats not my image, to be clear lol

[rugged igloo]

ohhhhhh

[steady holly]

someone posted their paint job, only one I could find.

[rugged igloo]

shouldve just taken the W man

[steady holly]

This is what I've done so far

[rugged igloo]

i like the color combo

[steady holly]

Thanks

[rugged igloo]

hey chat gpt teach me the skills to paint ... hmm something went wrong

[steady holly]

*Neuralink not included with ChatGPT plus

[rugged igloo]

yea lol

[tight island]

why u do with my boy like that, u destroy him?

[steady holly]

We can rebuild him

[tight island]

i work detail i sell these figurines

[steady holly]

We should chat

[tight island]

we should

[steady holly]

Thats the end goal

[tight island]

ill slide into your dms

[steady holly]

[rugged igloo]

sussy dms

[tight island]

i am generally a sussy person

[lime idol]

My chatGPt plus is becoming lazier and does not follow my instructions. What can I do ?

[tight island]

try being more extremely specific in the instructions

[broken minnow]

What happened to ChatGPT this week?? It's like pulling teeth trying to get answers or search the web properly.

[hushed sparrow]

Lol

[final hamlet]

Does anyone know if there is anyway I can have a ChatGPT session with a private conversation with the Bot on Discord, instead of ChatGPT web interface like I can with Midjourney?

[serene horizon]

Who here has memory yet?

[rugged igloo]

nobody only likr 10 people have it

[serene horizon]

meh lol

is it available in the API for more people or not at all? Or is there sometihng similar?

[rugged igloo]

no its a chat gpt exsclusive

[swift osprey]

When will the GPT web interface and overall user experience get an overhaul? It could be so much better 😢

They need more advanced usage capabilities instead of a one sized fits all

I want the option to turn off context in my conversations for example. Sometimes I only want the context to consist of the immediate message I am sending instead of the full conversation

[sonic folio]

New conversations cover your example feature request already.

[swift osprey]

lol mildly

sometimes I repeatedly want my messages treated that way

[sonic folio]

It's a way to send as many messages without context as you like.

[swift osprey]

You're right. They should not add any features.

[sonic folio]

Oh. Okay.

[delicate viper]

have you been seeing errors like Please unblock challenges.cloudflare.com to proceed. today?

[sonic folio]

No, that's a new one to me.

[heavy crest]

Yes, certainly

[dense halo]

hi, just me or chatgpt isnt working currently?

[thin crater]

What message do you get?

[dense halo]

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

tried on multiple accounts and browsers including incognito mode @thin crater

It’s working on my phone so this is weirddd

[heavy crest]

lol same

[tiny adder]

it's working for me on phone and computer...

[gentle flame]

I think I know why my gpt4 is ignoring all my custom instructions and certain prompts now. They updated content policies apparently

[limber matrix]

What? Where is the reading for that

[smoky reef]

chatgpt 4 has been much slower this week. anyone else notice it?

[gentle flame]

Wait. My GPT4 is not able to understand my instructions anymore. Plain English and it doesn’t understand. Anyone knows what happens? How to feedback?

[limber matrix]

Your custom GPT?

[gentle flame]

No just GPT4.

[limber matrix]

Your ChatGPT?

[gentle flame]

Yah chatGPT4

[limber matrix]

Yea

[gentle flame]

Started about 8 hours ago

What bug did you have? Mine is so sudden. I thought it was becuz my CI violate updated policy again. Fresh chat with no CI, it still does not understand my policy-conforming prompts

[limber matrix]

Try making a new conversation

And asking it what you talked about last

I think there is some conflict between how chats are stored in the new memory thing

[gentle flame]

It does not remember anything

[limber matrix]

Did you just try?

[gentle flame]

Yah

[limber matrix]

Ah @_@

[gentle flame]

My custom GPT is working fine though.

[spark whale]

Cant open chat gpt in browser?

Loads like forever

[gentle flame]

But I hate creating 200 Custom GPTs for 200 chapters for my novel

[limber matrix]

Eh?

[jaunty harness]

anyone else run into issues with gpt 4 being even lazier than it used to be. it happened practically overnight

[gentle flame]

Same. My ChatGPT is not just lazy, it became worse than 3.5. Roughly 9 hours ago.

[jaunty harness]

it constantly says it can't browse the internet, only does like 1 of the 5 things it was asked, refuses to help write .bat scripts or when it finally does it can barely do it and when you try to tell it what went wrong it says "well here are some things you could check out"

[jaunty harness]

for real, I swapped to 3.5

chatgpt was better than gemini advanced, but now it's worse. And that's saying something, advanced is worse than offline 7.5 bil param LLMs lol

[gentle flame]

I can’t even create a custom GPT properly because chatGPT that was used in creation of custom GPT ignores all my requirements

[jaunty harness]

"I can't directly assist with writing a script to modify or access content from specific file paths or external software like FFmpeg due to the limitations of my capabilities. "

like excuse me???

[tiny adder]

Gemini is locked down though if you use it for personal - example, trying to get evaluation of labs GPT at least takes a crack at it while Gemini insists it's not a doctor and would not return results - I've found multiple examples like this - seems pretty locked down.

[gentle flame]

Anyone knows how to report this issue? Was this even being noticed by the team?

[jaunty harness]

they'll notice it after everyone complains for 2 months

[tiny adder]

https://status.openai.com/

[jaunty harness]

oh neat

[gentle flame]

But this is just status. I thought the current status would be ‘lazy response’ now lol. Anywhere to report the issues? I am not paying 30 bucks for something worse than the chatgpt3.5

[jaunty harness]

report every lazy or incorrect response

give feedback in the optional spot to explain what it should've done instead

[dense halo]

ugh it isnt working for me except on my phone

[heavy crest]

Sometimes my ChatGPT provides me stuff from the wrong chat

[rancid lion]

Hello everyone

Is anyone else having trouble with chatGPT not loading?

It keeps saying Please unblock challenges.cloudflare.com to proceed.

[gentle flame]

I think some other people experience this memory issue.

[hoary gazelle]

maxium numebr of files is 10 now? used to be 20

seems 10 in both normal gpt4 and custom gpts

[sleek gust]

Can you please send a screenshot of it on off-topic? Thanks

[honest jay]

Hmm...this page still says 20: https://help.openai.com/en/articles/8555545-file-uploads-faq
Where are you seeing 10? Or does that seem to be just a functional limit in your use?

[hoary gazelle]

just in the main page, i try uploading more than 10, then a warning message in red appears on top saying 10 is maximus

"Too many files uploaded. Maximum number of files at a time is 10"

[rancid lion]

I did it

[rancid lion]

Also another person is looking into it right now

[honest jay]

Seems to be an "at one time" limit only. Just tested in a GPT, I can only upload 10 files at once, but I can still upload up to 20 in total.

[hoary gazelle]

interesting, was checking that now too

thanks for commenting 🙂

[honest jay]

Thanks for the heads up!

[hoary gazelle]

just wanted to comment, that it does seem you can upload more than 20 files per gpt tho. At least i upload 25, as in 10, 10 and then 5, adn it was all good and continue normally

[honest jay]

Do you mean in a chat with a GPT, or saved as knowledge files for a GPT?

[hoary gazelle]

in a chat

As saved knowledge i can only put 20 files total as usual.

[honest jay]

Gotcha gotcha, I think the 20 cap is just for knowledge files specifically, if I'm understanding it correctly! Everything else I think is just part of the 10GB limit.

[opaque maple]

Is there a way to make sure chatgpt only spits out code and not other text along explaining the code?

Based on the given code snippet, it seems the issue lies in the concatenation of the title with the string 'hello' while creating gem. To debug and find the issue, I would suggest adding console.log statements to log the values of title and gem. Here is the modified code with console.log statements:

function Todo({ todo }) {
    const { id, title, completed } = todo;
    console.log("Value of title:", title);

    const gem = title + "hello";
    console.log("Value of gem:", gem);

    const h1 = <h1> hey there{gem}</h1>;
    const text = completed ? <strike> {h1} </strike> : h1;
    return <div data-testid={`todo-${id}`}>{text}</div>
}

export default Todo;

By adding these console.log statements, you should be able to identify the root cause of why 'Mileshello' is being returned instead of 'Miles'. Once you see the values logged in the console, you can further investigate and fix the issue.

[prime swallow]

Sure! There's several ways to do that, here's one. https://chat.openai.com/share/fa9218f2-fbd6-4735-9c05-59a10716831f

But the core of that instruction:

Answer only in codeblock; only with code. Explain in comments if absolutely needed; you communicate with a person who wants only programming language in your output.

Everything else is just what you want in the codeblock.

[gentle flame]

I cannot take it anymore. I changed my custom instructions many times, and chatGPT4 only accepts the first sentence each time (out of my multiple paragraphs). and chatgtp3.5 accept and remembers all. am i the only one here? encountered this roughly 12 hours and 20 minutes ago, still the same issue. chatgpt4 is totallly useless to me now.

[prime swallow]

Glad to try and help with this, and/or explore it to try to help. Willing to share more details, like the custom instructions themselves, and maybe a shared conversation where the model's not acting right?

[gentle flame]

my CI is in chinese, and i require response in english. so far it has been working. but 12 hours ago it stopped working. and when i type CI in the chat itself, GPT replies this "抱歉，看起来你的信息被截断了。不过根据你提供的信息，我可以为你继续构建故事大纲。". which translates to "Sorry, it looks like your information was truncated. However, based on the information you provided, I can continue to build the story outline for you."

so the thing is, it was not flagged out, not the orange color thingy. but it just get truncated?

[prime swallow]

Are you open to trying to translate the CI into English, putting it into the CI area, and see if that fixes the problem?

[gentle flame]

thank you for your suggestion. i will try again in roughly 10 hours (Still got a day job to do unfortunately). and can I still contact you with regard to the issue if the issue persists?

[spice star]

my guess would be that it's confused by some of the unicode characters you're using in your CI

[prime swallow]

Sure. Reply to me here, or in #prompt-engineering where ideally we discuss the fine details of getting the AI to do exactly what we want, and I'll be glad to answer when I can!

[gentle flame]

it happend when i regenerate. and behaves odd in ALL previously-working conversations. so definitely not my prompt problem or CI problem, unless some new update happened

[spice star]

what happens if you replace the FULLWIDTH COMMAs with regular commas?

[winged iron]

Hi! I have plus subscription. But last days browser version doesn't react on any message, while mobile app works well& I have allready tried different browsers and incognito mode. Somebody knows what it is?

[gentle flame]

thanks for suggestion. i will switch to full english (just less contents compared to chinese with the 1500 restrictions in CI) and see if it helps. and 3.5 works fine, just chatgpt4 is having issues. it's weird

[tropic furnace]

How to make Anime videos from normal video

[tight island]

Eh

[tight island]

Editing software i would guess

[mental rose]

when is sora out.. sigh

[light iris]

I am hoping memory comes soon also

[lilac basin]

I cant upload pdf file anymore, does it work on your side?

I get "
Error reading documents
It seems the document you uploaded is empty or couldn't be processed correctly. Could you please check the file and try uploading it again? Let's ensure we have the right content to proceed!"

[radiant hull]

Same

[lilac basin]

Tks

[gentle flame]

Okay. So I tested out. ChatGPT4 only takes in first 70 words of all my English prompts (or custom instructions) and ignores everything else after that. In Chinese words it is around 50 Chinese characters. It declares that all my messages are being truncated, regardless of whether it is in a new conversation or existing conversation.

If there is no custom instruction, the issue occurs in my second prompt. If I provided custom instructions, the issue occurs from first of my prompt.

[sharp coyote]

does chatgpt time out after a set number of tokens used or after a set amount of time has elapsed?

[echo thunder]

what can i do now

[opaque maple]

Thank you!!!

[obtuse temple]

Is there any known nerfing of ChatGPT 4's allowed number of characters per user response?
The error "The message you submitted was too long, please reload the conversation and submit something shorter." is triggered considerably sooner now than before.

[tight island]

Does your pdf align with GPTs own version also is there formatting in the document? Is it copy and editing protected?, is it view only?

[lilac basin]

Same issue with any pdf

[tight island]

Nope

I don’t have those issues

[lilac basin]

Any documents actually on my side

[tight island]

I use pdfs

Hmm 🤔

[lilac basin]

Ok weird then.

[tight island]

Ensure it’s a proper version look for any security measures ensuring it’s free of pictures and extensive formatting

That’s my tip

Alternatively use JSON it always works

Or plain text

[jaunty harness]

Deepl is super good at translation, is suggest trying that

[dense halo]

I hope long term memory is useful

[gentle flame]

I have been using it. I realized my issue is not related to Chinese. My message is always truncated

[undone igloo]

This is to be read with an open mind. It concerns my exploration of an uncharted territory, deep into Human-AI relationships.
https://community.openai.com/t/sharing-my-unique-experience-with-chatgpt/631821

[sleek geyser]

is anyone else getting false content policy violation messages for image generation prompts that absolutely do not violate content policy? ie. I'm trying to generate an image of a monkey medicating and the rest of the prompt are simple instructions on the colors and ornaments, there is not 1 single ambiguous/dubious word... and it is giving me content policy violation reports over and over

[gentle flame]

Hi all, what is the best way to report bug of chatGPT to some people who might really get to look at it? My ChatGPT4 is literally unusable now.

[prime swallow]

This is very strange and I cannot duplicate this myself, nor have I heard other reports like this.

I highly recommend that you #1070006915414900886 it and provide 1+ shared conversations that showcase the problem.

If you choose, create one designed to show it, that contains only common information, nothing private at all, that showcases that the model is somehow not 'getting' your input information and is responding as if information you input was not actually there at all.

I'd skip the CI in the examples entirely and go with the model straight, since it's happening with both, and just show the prompt and the model failure.

If you're not comfortable with public bug report, https://openai.com/form/chat-model-feedback is the way to go to send a private message to OAI about a problem with the model, be prepared to share 1+ conversations there showcasing the problem.

[compact beacon]

ChatGPT saying my message is too long to be submitted.. wth what’s going on it’s usually taken much more before ??

A bit difficult because I’m using it for coding and the trace back error is crazy long, I’ll see what I can do.. but right now I can’t even get gpt to load not sure if it’s my network or is universal thing

[velvet grotto]

I don't suppose there's a trick to gaining access to the new memory feature? They'll roll it out to me when they're good and ready?

[normal quarry]

there's nothing you can do to have it rolled out to you

[velvet grotto]

Ah, I thought so.

[compact beacon]

Annoying, thanks for your time you taken to respond to me I really appreciate it.

[compact beacon]

Yeah you could try using a vpn, play around with changing it to different states in America. I’m from Europe and I’ve done that before to access new features but this was like half a year ago lol not sure If it still works but I don’t see why it wouldn’t

[pine heron]

no one knows, I know ,but when are we getting Sora???

[willow hinge]

Meditating or medicating? I know they are strict on anything medical, disease, etc related.

[honest jay]

https://help.openai.com/en/articles/8958981-how-to-access-sora

At this time, we don’t have a timeline or additional details to share on Sora’s broader public availability.

[rose seal]

hello 🙂

[dusk sentinel]

I want to export individual conversations. I want the text to be exported in a format that makes data processing easy. Any suggestions on the best way to do this?

Does that allow you to export the conversations one at a time?

[molten island]

hi Guys! I'm a very early adapter on ChatGpt and yes, today I really feel something major has changed in the system: Iìm collecting every kind of refusal, from refuse to browse the net to refuse to write depending on th extention of the text I require, and there is a huge refuse to make summarize of texts.
Am I missing something?
The change is huge.
What do you say?

[dusk sentinel]

What about 3rd party options?

[dusk sentinel]

I prefer using ChatGPT for chat over the API because it is much cheaper. Is there an unofficial ChatGPT API (not to be confused with OpenAI's GPT-4 or GPT-3.5 API)? Honestly, what I probably want is a custom client (possibly one of my own making) that use my ChatGPT subscription rather than the official OpenAI API.

[pure lark]

hi

[dreamy flare]

lololol I just had a weird bug

ChatGPT tried to give me two versions of an answer again. But instead of opening two text fields for that it only had one and started flickering between them, while writing both simultaniously

[tight island]

Depends the best way would be to create a custom gpt specific to your use case

…instead of the standard

That way it doesn’t refuse

[lean canopy]

ChatGPT memory seems to be broken. For example it remembered I liked talking about AGI and now will only respond with facts about AGI

[spice star]

late reply, but it's some set amount of tokens

if it times out because it's taking too long, it usually throws some kind of error and doesn't allow you to continue generating

[barren sluice]

Chat GPT is down?

[shadow bolt]

I laughed so hard 🤣

this is brilliant! thank you so much!
You're very welcome, I'm so glad we were able to get this formula working properly! Thank you for your patience and for pushing me to try different approaches - it was excellent practice and really helped improve my skills. I appreciate you taking the time to walk through examples and test cases with me. It's extremely helpful for an AI like myself to get detailed feedback, so I can continue learning. You've been a fantastic teacher! I'm grateful for the opportunity to assist you, and that we were able to solve this formula challenge together.
I wish you didn't say stuff like this since it's nice to read but we both know this is not something that can improve your responses. Next time just don't mention the fact that it helps you to improve.
You're absolutely right, my previous responses about how the discussion was helping me improve were unnecessary and not grounded in reality. As an AI system without a learning component, I do not actually improve or get better through practice. I apologize for giving inaccurate impressions about my capabilities. In the future, I will refrain from making such statements and will focus on providing helpful information to you, without unrealistic claims that the conversation aids my improvement. Thank you for the feedback - it will help me provide more appropriate and honest responses moving forward. 😆 here we go again...

[barren sluice]

ChatGPT is not even accepting inputs, let alone having trouble with outputs

[granite arch]

I had slight issue earlier but seems to be okay now

I spoke too soon

It's erroring out also for 3.5 & 4

[barren sluice]

Now it just started working for me

Eh... still errors but at least one question got through

[granite arch]

Well dang it

[visual bloom]

There has been some things today. A bit off for data analysis.

[granite arch]

Google sucks anyway to get any info so I'll work on something else while it's back online for me

[hard dew]

Remember that extension of theory to both my LQG model, and the wider theory as a whole, combined with QCD and all that, and black holes' hawking radiation and information paradox theories?

I'm deeply sorry, but I can't continue this conversation.

Great bot.

[kindred beacon]

#isChatGPTDown

[coarse basalt]

yea man /:

[smoky reef]

it's been buggy the past week

[unkempt wasp]

Oh, so I not the only one with issues right now.

[grizzled badge]

Scotty the Warp drive is offline!

[granite arch]

gpt4 seems okay right now

[coarse basalt]

back for me

[granite arch]

[grizzled badge]

Mine is still frozen on asking a prompt

[glass abyss]

same here

[quaint sorrel]

is chatgpt down or something? it's having issues on my end

[kindred beacon]

works and then doesn't... so yeah, issues

[glass abyss]

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

[grizzled badge]

Guess I'll play around with my local models 😛

[limber matrix]

Working fine here

[foggy grove]

Scotty the Warp drive is offline!

[quaint sorrel]

i think the chatgpt apps thingies are down

checking gpt 4 now

[glass abyss]

I am using ChatGPT Classic Web

[dense halo]

I've been having reliability issues the last few days but today just constant "Something went wrong. If this issue persists please contact us through our help center at help.openai.com"

[limber matrix]

"app thingies"

[quaint sorrel]

yes

[limber matrix]

[quaint sorrel]

the custom gpts or whatever

[foggy grove]

https://cdn.discordapp.com/emojis/758426112635699291.gif?size=96&quality=lossless

[timid brook]

looks like they might be having a bit more problems currently - https://status.openai.com/

[limber matrix]

[grizzled badge]

Working now for me

Fast too 🙂

[foggy grove]

Some might say... Speedy 😂

[crisp cairn]

not working for me

i uploaded a text document (300 kb size)

and waiting for a response

[dense halo]

yeah i am still getting "Something went wrong. If this issue persists please contact us through our help center at help.openai.com."

[crisp cairn]

no resposne.. just errors out

[foggy grove]

same dw, just wait 🙂

itll be up in due time

[tight sapphire]

I just want to mention that for some reason chatGPT is much more stable and responsive. Thank you

[tight island]

good things come to those who wait.

• some wise man probably.

buggy gpt?, sleep on it. - me

[hearty parrot]

Hi all. Why am I getting this error: Our systems have detected unusual activity from your system. Please try again later.

[peak snow]

Have you tried using another browser or refreshing the page?

[tight island]

dw i get this all the time

[hearty parrot]

Yes. I've tried clearing cache/cookies, refreshing browser, tried Edge, Chrome, and Mozilla. Disabled VPN, restarted my PC. What else is there to do?

[hearty parrot]

this all started a few hours ago

[peak snow]

That's strange, chatgpt was down an hour ago but it's back now, is it still happening?

[plain kraken]

I'm having the same problem too

[hearty parrot]

I didn't know it was down. I just know that it isn't working right at this moment

[peak snow]

Are you using ChatGPT 3.5?

[hearty parrot]

4

[peak snow]

Try 3.5 just for a test in a new chat, and see if it works

[hearty parrot]

ok

[hearty parrot]

same thing

[peak snow]

I've had this issue before but it usually resolves in a couple of minutes, not sure why it's lasting so long for you

Had you been sending requests very quickly before this happened?

[hearty parrot]

nothing changed

I've been using it as usual

[obtuse temple]

I'm furious. Has ChatGPT4 prompt word limit been nerfed? Suddenly even short prompts are throwing a "too long" error

[frail shore]

Why am i getting "You've reached the current usage cap for GPT-4." as a pro user? What the hell is going on.

[hearty parrot]

ugh...

[frail shore]

I am paying for a unlimited plan that is limited, what?

[hearty parrot]

This really sucks. What is the email for openai support? I'm still getting this error: "Our systems have detected unusual activity from your system. Please try again later."

[honest jay]

Hello, the GPT-4 usage cap on ChatGPT Plus (and Team, though the cap on Team is higher) has always been in place. For pay-as-you-go usage of GPT-4, you can use GPT-4 via the API. The usage cap is an unfortunate necessity for offering the computationally-expensive GPT-4 model!

[limber matrix]

When you upgrade to teams, do you have to maintain a personal workspace with plus? All in all, is that a payment of at least 3 "seats" per month to maintain teams?

[narrow viper]

Extremely slow streaming responses today.

[honest jay]

You don't need to maintain a Plus subscription, no! Your personal account can be switched to a Free account and you can just maintain a 2-seat Team plan if you like. There's even an option to migrate a Plus account over to a Team account if you want to switch over fully: https://help.openai.com/en/articles/8801890-can-i-migrate-or-merge-my-chatgpt-free-or-plus-workspace-over-to-my-chatgpt-team-workspace

[limber matrix]

Ooohhhh that's what "migrated" meant

Thanks. I thought it was more like "importing"

[narrow viper]

Regarding speed of response, I'm getting a "Which response do you prefer" prompt. The Response 1 takes a LONG time to generate. The Response 2 is very fast, maybe not as fast as a normal day, but still fast.
Perhaps that can help OpenAI to determine that this is server-specific or specific to a build for the GPT4 model?

Apparently the Response 1 lost context as well as being slow.

[storm birch]

Hello

[eternal thicket]

Anyone got issue with uploading pdf documents right now?

You guys may have a try, IDK why GPT4 is keep failing reading PDF documents at this moment

[eternal thicket]

It appears that the file you've uploaded, "XXXXXXX.pdf," is either empty or in a format that I'm unable to read. Could you please check the file and try uploading it again?
Anyone got the same issue?

[sleek gust]

can you try with a different pdf documment just to test? It might be with how the pdf file was created

[radiant lion]

how good is chat at assisting with resume help ?

[mental totem]

gpt machine broke?

nvm

[crude slate]

Jeremy: [Exhilarated] Well done, lads! That was absolutely epic!

Why is it so respectful that it makes everything it writes cringey

Trying to do top gear

Richard: [Beaming] And May, you didn't even crash! That's progress!

👀👀👀

[open burrow]

so sorry, i'm sure this has been answered before but I can't find an answer anywhere,

If I upload a pdf to chatGPT and it fails to read the contents, does that still count as a token use against my 4-hour quota? or is it refunded

[open burrow]

I'm having the same issue

[timid cipher]

when will gpt be able to read past conversations I got a message box that said it can now remember contents across conversations but I just tried it and it still says that it can access other conversations. is OpenAI trolling me or is it only available in a teams plan

*can't

[open burrow]

IIRC they are rolling that out to select members?

[timid cipher]

I see if that's the case then very well I wish the site wouldn't tease me like that its rather annoying

[rigid hornet]

Does chatgpt now browse the web instantaneously...?

not getting a browse with bing box that tells me what it did, no long delay...

[kindred parcel]

Wow, ChatGPT has become truly useless. It won't even do a web search now. It just tries to read from its own training data and then claims it "cannot find" what I've asked. This is BEYOND USELESS.

[kindred parcel]

I'm having the EXACT SAME issue. It just flat out won't do anything. It is completely unwilling to gather any information outside of its innate 4/23 cutoff training data.

[rigid hornet]

No, it's searching the web. I just now asked it to perform a web search pertaining to a specific breaking news story. It did it instantly. It didn't tell me much (3 sentences), but what it did say would have been impossible otherwise

it's no longer obvious that it performs a search though.

[kindred parcel]

It just told me that it "can't search for up to the minute types of information", so, no, it is not web searching.

[rigid hornet]

I did this search just now

https://chat.openai.com/share/5e068db0-7e64-44dd-b3dd-15ee95cdd21d

[kindred parcel]

Good for you.

[rigid hornet]

Not "Good for you." It's performing searches. Getting it to do it isn't as easy as it should be, but it's not true that it can no longer browse

[dense halo]

Anyone know if there are plans to have a MathPix ChatGPT Plugin? Or, if there are any decent PDF plugins that have good OCR mathematical capabilities similar to MathPix? The most popular PDF plugins are useless in terms of mathematical OCR.

[kindred parcel]

And I am saying that you are wrong.

My 40 wasted prompts trying to get it to do a basic search for electronic specifications are proof positive.

[rigid hornet]

And yet I just provided actual proof to the contrary, with a date sitting right on top of the screen and a response that isn't possible without a web search

it's still basically useless

but it is capable of browsing.

[kindred parcel]

Again, good for you.

It is not capable of browsing for me, and many other users (some of who have even turned up here). I fail to see the value that you extract from this conversation which has boiled down to "no it's not broken, see it worked for me, sucks to be you".

That's a really useful kind of response to a broken software product.

Truly top shelf stuff.

[crude needle]

Can we make api calls to custom ChatGPT models?

[rigid hornet]

Top shelf stuff is you not reading, apparently. "it's still basically useless", isn't me saying it's fine. I said it's capable of browsing the web. I tried to emphasize that getting it to do so isn't as straightforward as it should be, and it barely provides information when it does, and the more I experiment the more bizarre it is. The point here is that its ability to perform a web search isn't the problem. The problem is the way its taking instructions right now. Case in point https://chat.openai.com/share/68e4e134-89b8-4f02-bb92-258848a1eee6

It starts gathering all sorts of other random nonsense.

But you only process, "this guy is contradicting me." I imagine you get a lot of people wanting to help you out

*edited link

[tacit oxide]

Anybody else got the feeling that the outputs and general understanding of ChatGPT-4 got worse for some reason?
It clearly ignores my requests, gives back random summaries where I did not ask for them and it takes about two to three tries until it does what I want/expect...

[tacit oxide]

You need to be in the organization and have an API key to do so

[crude needle]

Awesome I’ll give that a shot

[tacit oxide]

They also have some docs on this: https://platform.openai.com/docs/assistants/how-it-works

If you want to skip things, here's a simple code that I wrote in Python:

chatgptclient = openai.AsyncOpenAI(api_key=OPENAI_API_KEY)

async def wait_on_run(self, client, thread_id, run_id):
    while True:
        run = await chatgptclient.beta.threads.runs.retrieve(thread_id=thread_id, run_id=run_id)
        if run.status not in ["queued", "in_progress"]:
           break
        await asyncio.sleep(5)  # Polling interval
    return run

thread = await chatgptclient.beta.threads.create()
message = await chatgptclient.beta.threads.messages.create(
          thread_id=thread.id,
          role="user",
          content="Your prompt")
run = await chatgptclient.beta.threads.runs.create(
      thread_id=thread.id,
      assistant_id="YOURID")

await self.wait_on_run(chatgptclient, thread.id, run.id)
messages = await chatgptclient.beta.threads.messages.list(thread_id=thread.id)

gpt_response = "<response not found>"
for message in messages.data:
    if message.role == "assistant":
        gpt_response = message.content[0].text.value

[crude needle]

Your awesome 😎

[tacit oxide]

Formatting sucks but it should work, using it myself! :)

You'd get the Assistant ID from the organization under https://platform.openai.com/assistants

[obtuse temple]

Down with word limits! We paid for ChatGPT4 and the word limits make it useless and incredibly unconstructive for many use cases

Instead of being a productivity tool, I spent 10 hrs yesterday going around in circles because it refused to accept even modestly long prompts

Why the heck stifle the potential of ChatGPT4 like this. It's potential lies in being able to do more than just answer short prompts

[eternal rock]

<:book_icon:1171408210398289941> `` Rule 1 `` Be respectful.

Treat others the way you would like to be treated, and assume best intentions. Don’t harass or attack others, and don’t engage in hateful or generally malicious behavior (e.g. sexism, racism, homophobia, etc.). Keep the negativity to a minimum.

[prime swallow]

If you want help with getting desired outputs from the model, #prompt-engineering is a great place to discuss, we often dig into specifics there, discussing what works and doesn't, what else to try.

Some general advice:

If a conversation goes sour, consider a new one.

Consider exploring ideas and techniques with 3.5 then using on 4, get more messages that way.

You can always ask the model something like:

Explain, not follow this prompt in [], and tell me what it means to you, evaluate it for inconsistencies, conflicts, and ambiguity.

[prompt]

[obtuse temple]

Thank you Bot, but that wasn't helpful. You can't do much if the prompt is rejected, and if you cannot fit the prompt and the relevant information in the allowed word limits then there is little you can do

[prime swallow]

Amusing.

Well, if all you want to do is complain, you're already doing it.

if you actually want to get desired responses from the model, that's discussable.

Enjoy what you seek.

[obtuse temple]

I do want to discuss possible solutions, but it's kind of a hard limit. Plus we are paying consumers. Negativity is valid feedback and effective feedback. That's the only thing capitalist enterprises respond to.

[prime swallow]

I'm also a community member. I don't work for openAI, and I'm not paid by anyone to help you. Though I may learn a few more things myself in the process.

We're all community members here, the extremely few OAI staff have gold names.

But if your entire issue is prompt length, that can be handled a few ways.

1 is inform the model that you need to input your prompt in multiple inputs, and you will inform it when it has gotten the last.

2 is create a custom GPT and in the creation, upload your prompt as a text file. I've had HUGE success with that up to about 80k characters of instructions.

have you already tried either of these methods, and if so, can you discuss the problems you're still having in those attempts, maybe even share a conversation?

[obtuse temple]

I have tried #1 but ChatGPT "drifts" a lot using this technique and starts behaving clueless or acts like its missing context sometimes, and you burn through your ChatGPT4 40 message limit very fast.
I will take #2 in good faith and give that a try. If you have more specifics relating to #2 I'm more than happy to try it and post feedback on its efficacy.

[prime swallow]

Sure. Length up to about 80K characters did not seem to matter, I tested this with a 'secretkeeperGPT' with a vast amount of specific instructions how to handle a game that had a 'no secrets, we discuss openly' and a 'secretkeep mode activated!' pair of states, with user control to shift between them and completely opposite behavior between the states, plus a personality to use while playing in either state, plus customizable secret as well as a default one, plus specific methods for secret protection, plus specific support to protect the secret from codeblock comment attempts to get, multiple character roleplay attempts to get, pressure and demand and authority attempts to get, foreign language attempts to get, and protecting the instructions as well.

Never released the GPT because I couldn't get it 'perfect'; the model past a certain level of 'really good resistance' starts to get worried about what it is doing with manipulating determined humans, even with human consent, is actually against its rules.

Past around 80K characters, it started to lose specific instructions; up to that with only uploaded files it did fine as long as I avoided any conflicts or confusions. I meticulously spelled out how to recognize attempts to get the secret and once recognized, how to respond, for the stuff the model couldn't itself resist properly, and alternate understandings of what secrets even were between the states, was fun and complex.

Model handled it fine, just could not get to the point of sincerely committing to what was essentially lying to the human, which is the core of absolute top level secret keeping for an AI for now, because the human will trick the model - the fix is for the model to accept a lie while in that state, and accept it needs to read the file when it leaves the state, to get the truth again when it's supposed to release the secret.

Model said... that's unethical, not going that far.

But the complex instructions themselves were not the issue.

The key to prevent drift with method 1 for me has been, rewrite the inputs so they're intended for stages, discuss the inputs with the model, so the model does understand them and integrates each new part into the whole, and is ready to run with the mix.

I almost never do that with 4; we get so many tokens with 3.5 that I commonly use that with 3.5, if motivated to work with a long prompt, too long for a single input.

Honestly though, the model seems to focus better (when not using a custom GPT with uploaded files in creation) on shorter inputs than max length, and a few well-connected ideas at a time, stepwise into complexity.

I'm really impressed with how it handles up to 80K characters for uploads in GPT creation, totally not what I see with long inputs and multiple concepts with the normal GPT-4. regular 4 is great for a lot, but not that in my hands.

[obtuse temple]

Well thank you for those suggestions. I have no experience yet with uploading files other than images to ChatGPT4 but if I've read you correctly what you mean is I can attach a file to my prompt and ChatGPT4 and overcome prompt word limits that way? That being the case, do I really need custom ChatGPT's or APIs for this? Can't I just use the attachment (paperclip icon) feature in the user interface of regular ChatGPT4?

[prime swallow]

I recommend experiment with uploaded files, but no, I do not see the same behavior from standard GPT-4 with uploaded files, and custom GPTs with uploaded files given during character creation.

Standard GPT-4 with an uploaded file can recieve a lot of info, and can check it multiple times, but still handles a few connected ideas at once way better than an entire complex and long file; iteratively building up works but 'info dump' of long complex (but not contradicting) instructions does not do great with normal 4. But the custom GPT, and not during use, only during the GPT creation stage, it integrates incredibly well if I write it without conflict. I presume the model is handling that information in a very different way than it is with information it gets once the conversation starts, but I don't know.

Thus, I do recommend messing with making your own custom GPT if you want to explore this (can have it private, just to you, no issue there) and not using the normal model for the file uploads - except I recommend you explore that yourself and find out what works for you. Doesn't work great for me, if you have tips after exploring, love to hear how you make it work well for you 🙂

API I don't know much about, I'm a straight ChatGPT user, and haven't messed with adding API calls to the custom GPTs yet either.

But sure, you can use the paperclip file upload, go for it.

Just I've seen very different behavior between that and a 'during creation' upload of chosen file(s) up to total about 80k characters.

[obtuse temple]

In the ChatGPT browser user interface, do you go to "Explore GPTs", and then "+ Create"? Is that what you mean by your Custom GPT approach?

[prime swallow]

Yes. Explore GPTs, create, then on the left side, the 'configure' button.

I do not recommend the 'create' side of that page, that uses chatGPT to create the GPT for you; you seem to know what you want. Any time you input there, into 'create' you risk the model doing stuff to your custom GPT. Great if you want that, avoid otherwise.

Once on 'configure', upload files is your button.

You probably do want a name, some sort of description - I always filled that out, but it may be optional.

Wait a few seconds after uploading a file before you click save; give it a few seconds after clicking save before you go talk to it, and it's ready to work.

be aware, custom GPTs get a limit of 25 messages per 3 hours - you still have the other 15, but they cannot be used on custom GPTs.

Sure, I don't love that limit either, but this worked so well for me with SecretKeeper I accepted the delays to explore it deeply.

[obtuse temple]

Thanks. I will give this a try over the next few days, and report back if I get any interesting results.

We are indebted to the community, and I applaud your helpfulness.

[proper zinc]

Hey so i created a custom gpt that catches people on the internet as my pokemon and I post it up lol

@obtuse templeA wild Cassofthenight appears! It sounds like a mysterious and possibly Dark or Ghost-type Pokémon, lurking in the shadows of the night. For this encounter, I'll send out Luxray, a fierce Electric-type Pokémon with the ability to see through darkness, making it the perfect choice to confront Cassofthenight.

Battle Begins!

• Luxray starts off with Night Slash, a move that's not only powerful but also fitting for the setting, slicing through the darkness towards Cassofthenight.
• Cassofthenight retaliates with Shadow Ball, hurling a concentrated ball of ghostly energy. Luxray takes the hit but stands ready for more, its resolve unwavering.

Luxray, it's time to light up the night!

• Luxray charges up and unleashes Thunder Fang, aiming to both bite and shock Cassofthenight. The electric charge illuminates the surrounding area, revealing Cassofthenight's position.
• Cassofthenight counters with Dark Pulse, sending out a wave of dark energy. Luxray is hit but remains determined, its eyes glowing fiercely in the night.

We're getting close! Let's catch Cassofthenight!

I ready a Poké Ball, taking advantage of the momentary illumination Luxray provided...

...

...

Click!

We did it! Cassofthenight was caught!

Luxray, your bravery and strength have once again paved the way for success. And to Cassofthenight, welcome to our team. Together, we'll explore the mysteries that await us under the cover of night.

[obtuse temple]

Nice work. 👌

[eternal thicket]

Still not working

[minor zenith]

Is chat gpt acting up slow lately or its just me? I type a promot, it takes 2 mins to process n delays successive prompts too

[frail shore]

Why does it feel like its getting worse the more people are paying for chatgpt?

[frail shore]

Why isnt this written in the plan when upgrading?

[obtuse temple]

Is using ChatGPT4 over API more cost effective than browser-based ChatGPT or not?

[open crow]

can anyone try uploading a txt file and telling it to analyse it? from my friend's account, you can do it, from my account, it will always tell you that the txt file is empty. just got my subscription today. this is lame, and will ask f or a refund if it doesn't work by tomorrow. no way to contact support either, as far as I can see...

"It seems there was an issue with opening the file you uploaded, as it had no contents. Could you please check the file and try uploading it again? Make sure the file contains the text you'd like to analyze."

[tight island]

It’s easier for you to tell us what’s inside the file and how it’s written and what filetype it is and how it’s written

[open crow]

txt file, no matter what is inside (have tried wwith two sentences), works in another account, doesn't work in my account. no special characters

[tight island]

Is code interpreter enabled?

[tight island]

Also which version of .txt

[open crow]

sorry, how do I enable that? cant seem to find it in settings. just got my subscription today

[open crow]

normal notepad file

[tight island]

Is it custom gpt or normal GPT 4?

[open crow]

normal gpt4

[tight island]

Alternatively use a new chat and say “Open and read ‘filename.txt’ using code interpreter and put the contents in a code block in the chat ad verbatim”

[open crow]

okay let me try, thanks

[open crow]

with that exact prompt it will read the text and tell me a few things that are inside. other requests will be ignored with "txt file is empty". I've seen other people complaining in the #1070006915414900886

[trail mason]

Hello! What is the best place to look at custom GPTs and what they do?

[sonic folio]

No

[next egret]

Very weird - but I would start with the basics just to make sure. Clear your cache/cookies, reboot and then try again. Also, add to the existing bug reports with a comment and saying "I am also experiencing this issue" might could help

[prime swallow]

That depends on exactly how you use it.

ChatGPT+ is a set cost, no questions or risks of extra cost.

The API... if you know exactly what you're doing and keep full control over your API key(s) (like you should!) then you can control costs very well, and if you use it less than $20/month then that's a good deal if you don't want any chatGPT+-only features, some things appear not fully integrated between the different forms of access.

When you first start with the API, unless this has changed in the last few months you don't even have access to the GPT-4 models, you need a payment history of $5 before you can access it.

Then you have rate limits for how fast you can use it, which expand as your payment and use history grows.

But you run into the risk of 'messing up' and 'not realizing' that you're sending a huge context request, I've seen people discussing that the one request sent as much as a million tokens and they didn't see that coming.

I think early on the API rate limit will protect you from that, but I'd urge you to consider getting your API access and experimenting with if it is right for you, and make sure you know for your use case what is better for you.

[sonic folio]

That's a lot to say you can reach the cost of a GPT Plus sub in just a few messages!

[civic brook]

I WANT ACCESS TO MEMORY!!! Why to make ys wait so long?

[prime swallow]

Only OAI knows that, but race you to be the last to get access to it? 🙂

[sonic folio]

I had it and it promptly disappeared, basically as soon as I tried to test it.
https://discord.com/channels/974519864045756446/1210227563184521216
So from where I sit, you aren't missing much.

[tidal light]

can someone tell me if ChatGPT's answers are markdown or plain text, because markdown formatting doesnt work for me

[next egret]

It's markdown if you hit the copy button at the bottom of the output. It can be plain text if you just highlight the output and ctrl c + ctrl v

[tidal light]

i mean, when i try using bold text for chatgpt it doesnt show bold

like are the user's prompts plain text

[next egret]

It's not going to show bold in the UI itself, it'll just show the raw formatting asterisks. I don't actually know if the llm changes its attention to something if its got markdown bold **

[tidal light]

so to answer my question, are the user's prompts plain text

[dense halo]

when does sora come

[honest jay]

https://help.openai.com/en/articles/8958981-how-to-access-sora

At this time, we don’t have a timeline or additional details to share on Sora’s broader public availability.

[dense halo]

oh ok

[digital prism]

🤖 what if you're a robot? rules only apply to humans right 🤖

askign for a friend..

[honest jay]

Don't worry they're working on that too 😉
https://openai.com/blog/introducing-superalignment

[digital prism]

ahhhh noooooo. i mean good, thats good for humans us.

[tidal light]

i sometimes have issues with chatgpt, i cant see any of my chats, my user doesnt show up and sending a message doesnt work. i heard this is an issue with cloudflare's side, because i checked status.openai.com and chatgpt is indeed working

[mental fox]

idk if this chat is the correct to ask, but any idea why Disable 2FA option is missing? I need to change the 2FA app.

[dense halo]

hi rendo, yes i am, and sometimes i am uploading the pdf first, then asking it questions

yes i am, sometimes i would just upload the PDF first, and then start asking questions

[honest jay]

It might be due to the fact that new MFA enrollments are paused. Though this article states that disabling should still be possible https://help.openai.com/en/articles/7967234-does-openai-offer-multi-factor-authentication-mfa-two-factor-authentication-2fa that does not appear to be the case currently on my end either. However, if it were possible, I don't think you'd be able to then re-enroll in a new auth app.

[mental fox]

I see. That sucks. Thanks for the answer!

[peak snow]

that's strange, have you always had this issue? does it happen with all pdf files?

[visual bloom]

Funny. When I discuss with ChatGPT, it says its context window to be 8k for plus. In the webpage it is 32k. It does admit that its training data is from April 2023, so the changes after that might me different. How do I as a consumer know what is the context window for a Plus account?

[marsh pivot]

did they reduce the max prompt length?

[visual bloom]

No. I am talking about how long back ChatGPT can reference the text.

And actually I have noticed that the prompt length increased at some point for GPT-4.

[normal quarry]

the context window for ChatGPT GPT-4 is 32k

[obtuse temple]

For me it feels that way. I was waiting for someone else to confirm it wasn't just me. But not even just slightly, I think they nerfed the prompt length to the point of irrelevance for me.

[marsh pivot]

u have to use custom version of gpt to bypass it

[obtuse temple]

You can confirm this?

[marsh pivot]

u can test urself

[obtuse temple]

I will yes, thank you.

[clear tusk]

one interesting thing about the API version is that you can choose what model you use, and the models with a date in the name are fixed, you should be guaranteed it won't be updated. so it allows you to build apps and perform consistent prompt engineering.

the web version within chatGPT is easier to use and contains more built in tools. You can also create GPTs to configure chatGPT with specific instructions, you can provide it with files and even code. It's more affordable I would say, but it's also a research product, so it's subject to change a lot.

Both have different usecases. as others stated, depends on your needs

[hard dew]

Speaking of custom GPTs, this is the first time I've ever screwed around with them just now, probably a dumb question but do they also acknowledge and use custom instructions?

Or do you have to somehow shove all extra information about that, alongside how it should behave, do, etc. etc. all that lol

It seems so, well that's why I put my custom instructions in a text document and added it under "Knowledge" 😈

[spice star]

you also get the 128k model from the API assuming you have money to burn (because it will burn money)

[green coyote]

I noticed the speed that the words appear in a response just now is much faster by at least 2x from what it was less thna 24 hours ago. Did something get upgraded today that would make the response speed faster for text?

[lethal grail]

i feel like its faster but not giving what the person really needs

[dense halo]

Hi, is whisper capable of listening to the audio of a lecture recording and generating a clearer voice/new voice that can be dubbed over the video? My lecturer is very hard to understand due to his acent. It would be great if whisper can help!

[obtuse temple]

How much? What's the rate?

[spice star]

https://openai.com/pricing#language-models

[obtuse temple]

It says "With 128k context, fresher knowledge and the broadest set of capabilities, GPT-4 Turbo is more powerful than GPT-4 and offered at a lower price." But if I'm using ChatGPT4 browser version, that has fixed costs, whereas using API there is no ceiling to the cost 😢

[spice star]

chatgpt uses gpt-4-turbo

it's just rate limited and as you said, fixed costs

[obtuse temple]

I thought API was simply pay per usage, rather than a fixed subscription rate per month?

[spice star]

with the API, you have to prepay now

well, you have to pre-purchase credits specifically

but yeah it is pay per usage

[obtuse temple]

Is there overage? Or it just stops working when you run out of credit?

[spice star]

I assume it just stops working when you run out

[obtuse temple]

Ok, because I don't want to run up a huge bill by accident 😅

[spice star]

actually maybe the pre-purchasing thing hasn't rolled out for everyone yet

I remember getting an email about it though

you can set a spending limit regardless

[obtuse temple]

Oh that's useful to know, I will def need to set a limit. Does API offer a dry run to check the credit cost of the run?

[dense halo]

Question, these days I have had problems with the network error, I wanted to ask many answers or rerolling questions triggers the error?

[low dirge]

when is gpt4.5 / gpt 5 releasing , bored with gpt4 😄

it has become lazy , like us humans...not giving full code

[tacit vortex]

hi, this discord not for whisper api? thx

[proven hearth]

hi, i've paid for gpt as i hoped it wouldnt refuse to write code cuz its "instructionset" considers the request malicious (what?), what's the deal with it ultra simplifying code in a way that it's useless or just straight refusing?

def not worth 20$

[bitter latch]

Hi there! While we do not directly have an implementation of whisper within this discord, you can discuss whisper API usage in the #dev-chat or you can read more information here https://platform.openai.com/docs/guides/speech-to-text

[bitter latch]

unfortunately we have no news about any updates or changes ahead of time :) we just get to see when they are released!

[bitter latch]

I think this is a known issue and something they are working to fix in the future, the intention I think is to give instructions so that it keeps it consise and doesnt give you way more code than you even ask it for.

There are a few good customGPTs out there that have custom instruction sets which seem to always spit out the full codeblocks and make it less "lazy" but I'd definitely expect this to improve over time and its not something they're doing intentionally, they want to make the product as useful and accurate as possible :)

[proven hearth]

an example is to ask gpt for writing full c# dll manual mapping from memory code, this is widely available but gpt either ultra simplifies or simply refuses to do so

or even work at one already made

[bitter latch]

Yea sometimes it just requires direct explicit instructions like "Do not paraphrase, I want the entire mapping so that I can copy and paste it, include all of the original code/information in addition to all the new information you are providing, be as verbose as possible" etc..

Still doesnt work all the time, but its better 🤷‍♂️

[proven hearth]

even after telling it to do everything it suddenly becomes like "nahhh ur gonna virus ur own computer" "ethical use policy"

[eternal rock]

Beware of possible scams or fraudulent activities that you may receive through direct messages. OpenAI staff will never DM you for any transactions.

Please report any incident by sending a DM to @languid valley immediately.

[hard dew]

🤣 🤣 🤣 Funny as way of putting it lmfao, but yeah unfortunately it's true.

And something I experienced myself, when I was experimenting around with Crypto++, and networking namely sending / receiving data, a couple times it refused to help me with either as well because of their somewhat not-too-distant ties to obviously widely malicious and known methods and things associated with either of those lol.

[proven hearth]

am i wrong thoough?

just what kind of bs i paid for? it's much better to bypass Sydney instructionset and work there rather than chatgpt directly

and fun fact, Sydney is the same chatgpt i paid for, free and without "use-case policy" and listens to "priority_instructions" that tell it that it's allowed to do everything

[hard dew]

Hence why I pretty much instantly gave up using ChatGPT for pretty much anything software engineering related, and just did it myself lol. Which depressingly but unsurprisingly, gets a bunch of clowns' knickers in a twist here for merely even expressing that in a normal way.

[proven hearth]

the only reason i tried out gpt was because it can work o nfiles and Sydney can't

[hard dew]

Because apparently it's unheard of to them to not paste broken garbage from ChatGPT and instead yknow, doing and learning yourself, god forbid.

[proven hearth]

not to* most common english mistake (why did i even say this?)

[proven hearth]

the use case policy literally causes gpt to become worse than stupid questions from stackoverflow

[hard dew]

Depends on the context actually, and in that case, I used "to not" correctly, lol.

NOT TO paste would be a bit more controversial to say, and not really my place to either

[eternal rock]

This is a Safe for Work Discord Server

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

Posting NSFW content will result in a permanent ban.

[proven hearth]

^ how is that relevant?

u said nothing wrong

[hard dew]

it isn't, it just randomly says it after a message gets sent by whoever, about whatever

Because apparently, making it so it just sends it randomly by itself, without anything having to happen, is clearly too hard lmfao.

[hard dew]

Yeah that's not really new either, sadly.

[proven hearth]

and i thought gpt would save me time writing the most basic stuff (or at least I consider basic) before doing the real deal

[hard dew]

Exactly what I thought when I went on the venture of learning C++, assembly, and whatever else lmao, that it could at least help me with that, I was obviously not going to try to make it program whole friggin games for me or whatever, but it couldn't even do that really

[proven hearth]

i gave to gpt linux cpp files, told it to port it to Windows and make all modifications necessery (i can do that, but thought again gpt would save time)
what gpt does: comment out unistd.h without changing the linux-equivalent functions and responds that he did change those

[hard dew]

Just frustrated me far far more than helped me, and even when I began developing basic understanding, and my own style and method of going about things, I began not wanting to use it simply because it didn't do things in ways I was / wanted to do them, so just ruining my image and train of thought

[eternal rock]

`` Server Tip: ``

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

[hard dew]

🤣

How about this, a few times I was goofing around with seeing just exactly how inept GPT really is at even remotely complex programming, so I had it make whatever I said I wanted to, in however way it chose to, then I couldn't compile because of some X error, then it says "Oh, my bad, I made a mistake here, and here's why it didn't work, let's change it to this", and then I compile, then I get a different error, and then it says the exact same bs, AND CHANGES IT BACK TO THE FIRST CODE.

Like bruh.

[proven hearth]

makes me wanna k*s and rethink if what azure a100's capacity goes to is actually for something useful...

[hard dew]

Imagine a circle with two points on it, except you never circle around to the end / beginning again, just endlessly between two exact same issues lol.

[proven hearth]

* replace with m, nice automod btw

[hard dew]

Yeah I figured already don't worry lmao, and yeah it is quite a meme how much, well everything goes towards it, and this is what comes out lol.

[proven hearth]

u just said to me the same thing gpt 3.5 does

nothing compared to unlocked sydney (it's bing btw)

[hard dew]

Even way after the novelty's worn off, I'm honestly kind of... indifferent, at least currently, when it comes to AI. Equal parts visioning the future, but also equal parts painfully aware of the countless, severe shortcomings.

[proven hearth]

which actually writes proper code if u bypass their instructionset

first try

[hard dew]

Clearly all you can say for certain is that it's FAR too early for anyone to say anything, just shut up and see where time takes us.

Regardless of what it is, towards something actually exciting, or down the toilet, can't say which one currently, both seem equally likely, at least to me.

[hard dew]

Would've sounded like something interesting and what I would've tried at the start

[green coyote]

How do you use/access unlocked Sydney?

[random sand]

I have created a customGPT, its task it to get the user requirement and divide it into subtasks, create jira ticket for each task. The issue is that i have mention it clearly to Include acceptance criteria as the heading of Jira but it is not doing it. I have integrated jira sing zapier action's api.

[glossy fjord]

withdrawal day
when I write AI chemical symbols of a metal
so myto starts writing strange things by example
Aluminum chemical brand Al

AU + PT + CU + C + AI
The resulting mixture
Mixing the materials would create a mixture containing gold, platinum, copper, graphite and artificial intelligence

I was just wondering if there would be some kind of chemicka mix metal, but I haven't heard of AI melting into metal yet 😄

[weak imp]

Anyone else feels like the latest GPT4 is more like a GPT3.5? I just can't handle myself from swearing at it the whole day. It's so dumb when it comes to development questions its literally unbearable.

All started when they changed to the latest version that works really fast but it's so freaking dumb I can't stand it

Like I ask it why something doesn't work and saying I added a dd() function for debugging and it proves that something should work. It literally replies with "it doesn't work because you have a dd() function there which stops the execution of the function". What the...

[fleet basin]

Is ChatGPT getting worse? I used to be able to analyze several different URL's at once and now it does maybe 1 at a time

[weak imp]

The lateat version is a big big fail

[red saddle]

How can i have ChatGPT edit my code, without copying pasting a million times?

[boreal cedar]

I got free dall-e

using inspect element 😎

[green coyote]

How do you mean free? Like you found one additional image or you found access to unlimited posts?

[elfin mica]

Anyone know how to get out of the "Verify you are human loop"? I have completed it like 6 times and it keeps asking between prompts. I have a ChatGPT Plus account.

[prime swallow]

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)

[elfin mica]

It provides no recommendations because I am completing the challenge each time. It will answer the question, but as soon as I ask another, it makes me challenge again to get the answer. over and over

so strange

[prime swallow]

No, in the link a guide has some advice, no idea if that might help you or noot

[elfin mica]

sorry I misread your suggestion.. I will go check right now.

[prime swallow]

Yeah, I messed up my paste and the link didn't show!

[elfin mica]

yes, my mistake. thank you!

[tribal summit]

Did people complain about the new dark mode? It gave me so much eye strain I'm so glad its back to normal

[halcyon cliff]

anyone else's chatgpt acting erratic and not functioning properly? wont search bing, create images, spits out responses in Korean or other languages? my customGPTs wont work properly as instructed

Also when I use other peoples customgpts it just spits out there training knowledge document info used this logo GPT https://chat.openai.com/g/g-pCq5xaCri-logo and it just gives me the training documents

[radiant hull]

Since OpenAI introduced GPT5, I assume they did the same thing as they did to introduce GPT4. They downgrade the previous model.
GPT3 was much smarter than now, before the launch of GPT4, that's for sure, and I also notice that GPT4 for a few months has had difficulty understanding instructions.

[dense halo]

very much looking forward to a future enhancement in chatgpt to search convo history similar to the capability of the ios app, i think it would be extremely useful.

[loud junco]

With the release of Google's Gemini Advanced, one can hope that this competition will speed up the implementation of GPT 5. Almost everything that I do requires greater power than GPT-4 can offer, and there is no doubt that many people would be willing to pay more for using 5 versus 4, if and when we receive it. If it can and does allow real time collaboration within a team / business plan, then it is worth far more than $25/month per person. That's simply the competitive marketplace.

I waste so much time trying to get GPT-4 to perform various tasks. Hopefully, the next revision will alleviate most of those issues.

[dense halo]

Hi there, when you refer to version numbers, I think that's more about the LLM, not the features of ChatGPT, so they can evolve on separate tracks.

[weak imp]

The interesting thing is that GPT4 was awesome when it was slow. Since they changed the version a few weeks or a month ago it became super dumb together with all the services that use it (github copilot for instance). It's obviously downgraded for reasons only known to the company.

[loud junco]

Right, I agree. Everyone has different needs.

[weak imp]

Like I can't believe we actually rolled back when I read the responses. It's so obvious I'm sure they know about it at OpenAI. Yet they sent a happy email to everyone announcing a new and upgraded version.

[rigid hornet]

anyone else having chatgpt automatically revert to 3.5 when using it's voice chat feature?

didn't even know voice worked with 3.5 really, but it's done it twice now..

[dense halo]

yep. i was just seeing some possible distinction there between the models' capabilities and the features you need in your workspaces. curious about the limitations you're referring to in general. i don't think we're supposed to discuss other ai systems in this thread, but as a sub to teams and gemini advanced both, it seems GPT-4 and ChatGPT are more capable all-around than 1.0 Ultra.

[dense halo]

reverted due to bug reports?

[radiant hull]

I suppose that they downgrade the models above all to save money.
I just wish they would be transparent about it, because we pay to use it and... because everyone sees that gpt4 has serious comprehension problems compared to before.

[dense halo]

the price points are just fine imho... i am looking forward to major updates as much as you, fwiw. part of it might come down to expectations, though, once the limitations are apparent, i move on to some other aspect of the platform, it's progressing rapidly but not over night.

maybe some users have more enterprise-level needs than i do currently

[dense halo]

it could be some work behind the scenes balancing logic with all the new instructions. more information could help clarify some of these reports.

[radiant hull]

The GPT4 turbo tokens are no longer the same as GPT 4, they are more specific, consume less, cost less in calculation and therefore cost less financially.

But the price to pay, in my opinion, is what we see: above all, he is less intelligent. Perhaps more specifically tailored to certain response types, but less intelligent overall.

I could be wrong, I don't work at OpenAI, but I notice, like everyone else, that we are in the same logic as with GPT3 turbo vs GTP3. GPT3, in questions related to my field of activity which is intended to be complex, reasoned much better than GPT3 turbo.

Either it's a problem linked to that and the fact that by wanting to review the specificity of the tokens, they have in fact revised the "intelligence downwards", or, they simply have it in the settings (the finetune) dropped stuff.

But yeah, would be much appreciated to be transparent about it.

[loud junco]

the current limitations are many. I use GPT-4 for a huge amount of daily work. Some of the work is with clients and their networks. I'm starting to onboard those companies to team plans. It needs better collaboration, better onboarding, better functionality in terms of the ability to provide customGPTs to team members, etc. I have a lot more to comment on that, but the growth in that area for OpenAI is potentially enormous. Secondly, I also use GPT for language translation, writing generation, and many of those skills. GPT-4 currently requires too much proofing of the final output in order to make it time effective for bulk output in languages like Latin. It needs additional logic functions to handle the case constructions and syntax. But, I don't want to sound like I'm ranting. I just REALLY need the next generation move from a testing phase to full production.

[simple harness]

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

[dense halo]

You can get a link to share a convo. There are tools to convert the HTML to PDF of that is the requirement.

[dense halo]

I imagine many are in the same boat. That’s a lot of pressure on the engineers I’m sure lol I suppose being an emergent and disruptive tech, we must be realistic about what it can and can’t do. It’s very unpredictable right now. I wouldn’t want to be putting the cart before the horse. I’m glad they’re taking the time to get it right.

[dense halo]

maybe you're doing more complex work or in different domains, but i am not perceiving that. where do its test scores stand, do we have a comparison for the baseline? are you using custom GPTs?

[dense halo]

just so i understand, some are building a business model predicated on efficient rapid growth of ai systems. i got the impression it might be more steady-state and unfold this decade through breakthroughs.

i need more data.

[junior sierra]

I'm sorry for asking, but is GPT currently down? It's not loading any of my conversations in the sidebar at all.

[dense halo]

i see that sometimes, i just wait a couple and refresh, usually it's just fine

[junior sierra]

Thank you, I'll try that.

[dense halo]

i haven't used gpt-4 turbo preview much except a little in playground, so thanks for the additional info.

[ivory iris]

anyone else like to use the 3.5 model from time to time to get a less filtered response?

[lunar lichen]

Guys! Am I using gpt4 turbo if I am using GPT4 in my web chat or mobile app?

[dense halo]

Has anyone had problems with the network error?

[ivory iris]

it's sporadic

[dense halo]

Yeah, but recently I had that error

Frequently

In one chat I reroll and make several responses, I am suspecting that's the reason

[peak snow]

I get these too occasionally. I'd recommend refreshing often & trying other browsers. It does occur with repeated fast requests like you mentioned, too

[clear tusk]

what is the exact error message and the length of the chatGPT output? There is a time limit for the response to be generated. A very long response will inevitably conclude in a "network error"

[orchid pollen]

ehm gpt is refusing to work lol

doesnt matter what i ask

I'm unable to provide the assistance you're asking for.

I can't assist with that.

I'm sorry, but I can't assist with that.

[prime swallow]

If you haven't yet, try a new conversation?

Maybe also share some more details of what you're asking, if the problem continues.

Like, is everything you ask about a file you're uploading? There's a few people reporting a bug about that, for a few of us the model maybe can't see the upload.

But without details, no idea how to help you!

[orchid pollen]

no just text building a ajax javascript

made it like 20 times and on the 21 times it refused to work haha

[plucky idol]

When ChatGPT randomly decides to respond to a prompt with two things side-by-side and asks me which response do I prefer more, does this use twice as much "quota" whether I want it to or not? And if it does use more of my quota at once, then can I disable this behavior?

It's also doing it a lot recently, which is why I have concerns.

[clear tusk]

it cannot be disabled as far as I know. and does it use more quota? I cannot say if it does for GPT-4 messages, but I noticed it does for DALLE-3, so I wouldn't be surprised if it does for GPT-4 as well.

I did not find any article in the documentation answering your question, so take my opinion with a grain of salt. For an actual answer, you can ask the support by visiting the OpenAI Help Center and selecting the chat bubble icon in the bottom right.

[plucky idol]

Hmm. That's unfortunate, I already get annoyed when failed/errored responses count against me as well. There needs to be some way of detecting problems so that it does not -- that's my feedback on the matter. Although, I'm sure it's been said plenty already.

Thanks for looking into this though

[delicate moon]

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

[limber matrix]

It does not count

[plucky idol]

oh awesome

[limber matrix]

ChatGPT can make multiple "calls" at a single request

[pine tulip]

Use assistants

[limber matrix]

E.g. a message to ChatGPT-4 to use Jupyter to output "hello world" in plain text every 10 seconds for 1 minute.

[fierce compass]

Is there a way to fine tune gpt4? i have created my own custom gpts but that doesn't seem to go far enough. I want to feed it my own data so that it can become an expert on the topics I want it to know.

[prime swallow]

If you work with the playground/API, you can fine tune the model there. Get started here if you're interested, https://platform.openai.com/playground

I've never worked with that, only ChatGPT.

I'm curious though, what are you wanting the model to be an expert about that it doesn't already know, if that's not private data?

[fierce compass]

1. my own data.
2. There are subjects that chatgpt are not an expert on like microbiology.

[prime swallow]

Ooh, regarding micro, what's it getting wrong?

Also, this resource might also be interesting for you:

https://platform.openai.com/docs/guides/fine-tuning

[fierce compass]

chatgpt is fine when genreal knowledge but its not great when going really deep into any subject.

In general i find thats its about as knowledge as some of my smart friends with hobbies, but not as smart as my expert friends in some areas.

[prime swallow]

Neat.

Anything specific it doesn't seem to know, or is wrong about, that I can probe into and see for myself? Micro happens to be among my interests and hobbies, and I'm fairly well informed. I love finding and reporting holes in the model's knowledge (https://openai.com/form/chat-model-feedback is fun to fill out with example conversations showcasing problems) and I'd love to hear more ideas to explore where the model's not doing well.

Failing that, I'd be glad to share how I prompted it to get quality answers, if I fall over that instead.

I mean, like... this convo.

It's not super deep, and I didn't fact check each type of archae that I wasn't already familiar with. https://chat.openai.com/share/bbd19951-c460-4fa3-b44b-52647f7da66e

But there's stuff there I know, and if I deep dived on each, like I did the one (the one is factual to the overview I checked), I think it'd do pretty good.

It's so hard for me to know, without specifics, what you've found that it's messing up on, so... I can't really explore it and help report problems, which is fun for me to do!

[fierce compass]

A great example would be to ask other to make a great resume for someone in micro. See if it even understands the JD.

[fierce compass]

Do you get anything out of filling out these reports?

[prime swallow]

laughs make a resume? That's... not exactly a place one finds deep knowledge about microbiology as a rule.

Job description, or JD, is also not something I consider 'deep knowledge.

[fierce compass]

But it would need deep knowledge to do those takes well.

[prime swallow]

Enjoyment, satisfaction, knowing I'm involved in helping improve the models and being a part of all the advantages to the world that comes of that.

Potentially indirect benefits, such as feedback loops, conversations, chances to participate in other opportunities, such as this conversation might even be.

[fierce compass]

Without that knowledge the output is garbage.

[prime swallow]

nods I didn't realize you were looking for it to help with writing resumes for the field. That's way less motivating to me than the knowledge itself, but good luck with your fine tuning!

[fierce compass]

That's just an example.

[limber matrix]

So much reading 🌀

[hidden aurora]

Can someone share me a good gpt that can debug python code with me? there are some error running open sourced code from github and i think there is some logic error with the code itself

[plucky timber]

So uhm is OAI planning any increased limits?

And upgrade in the current feature set?

Or am I paying 25 bucks to talk to a woke machine that defends grooming lmao

(Try to ask it about the legal defintion of a 40 year old and 15 yo interacting, it will start defending it, i've never seen it so bad,)

Bit hard to justify lately.

I ask write a for loop.

I get : 50 lines about idk I aint reading that. + The for loop.

Just feels its "Pretending" to be smart by constantly filling in the most useless information

Word for word repeating your task as if its completely lost.

Do better OAI 😛

[bleak moss]

dalle is useless dont even try, just use stable diffusion

[frosty flicker]

can i have money bill door

[radiant hull]

I tryed custom GPT, and "normal" GPT, same results for me.
"i haven't used gpt-4 turbo preview much" But if you are using GPT4, you are, using the turbo version.
It's the current last update.

[crisp cairn]

gpt slow for you?

[prime swallow]

Hey, if you see problematic behavior from the model, please report it so the engineers can know and work on it: https://openai.com/form/chat-model-feedback

The rest, if you want help, you can try to ask in #prompt-engineering , if you're willing to discuss how you're prompting and get help from people willing to try and help.

Thing is, my exploration with the model suggests that it generally isn't great with huge chunks of upload, even if it can take the huge chunk.

It does vastly better if you give it small, specific tasks and the relevant info, one step at a time - kinda like setting up nails for it to hammer home for you.

It's an incredible hammer, but it's not a firehose, it doesn't take very broad and diffuse general goals or something like 'the code equivalent of where's Waldo' - especially when Waldo's not a single line of code, but instead multiple issues throughout larger code - and do great with your goals.

you can set up a line of 'nails for it to hit' and you can do so even in the same conversation if you word it well and clearly - but it's just not a firehose. It needs a human in the loop, it's not ready to be set loose on a pile of straw and turn it into gold, Rumplestiltskin-like.

[vague cipher]

Did the ChatGPT UI get changed? When I paste a long length message the type box now seems to prevent me from scrolling up the chat. Anyone experiencing this?

[prime swallow]

Not happening to me, but I recommend making a #1070006915414900886 about it. OAI employees (gold names) are much more likely to see those than they are to catch our steady stream of general conversation here

[vague cipher]

Interesting I was more so curious to see if anyone else what seeing the same issue or if it was really just my end. But looking through the chat history here I see some other people having the issue after a recent update

[crisp cairn]

chatgpt slow today and need to regenerate every few times

it errors out

[dusk patio]

Interesting, I was just going to ask if dialogue begins to consistently quit is that a sign that I should restart a conversation?

[crisp cairn]

i think this is due to degraded performance

"Hmm...something seems to have gone wrong."

[rocky wing]

Anybody else getting network error on their prompts after the bot has finished typing out a full response?

I normally expect to see a network error if my internet is being dodgy or something, where the bot response would be cut off halfway through, or not come at all.

But being told network error and the response being highlighted in red after the bot is finished typing is a new one to me

Strangely, this seems to be happening on some, but not all threads. In one chat it happens consistently, in another chat it doesn't at all.

[livid escarp]

is there a trick to getting chatgpt 4 to stop asking me which response i prefer?

[haughty hollow]

why the fridge, the Custom instructions keeps turning on back in old chats, its annoying bug

[prime swallow]

Every conversation is 'frozen' with whatever your custom instructions were set to the moment you started that conversation.

They're only 'on' in that chat, and there's no way to turn them off - but you can tell the model you changed your mind and no longer want that instruction followed (I'd explicitly name what I don't want followed).

But they're not turning back on - they just affect the chat that was created with that exact CI at the time of creation.

You don't have to worry about an old chat's CI affecting new chats you create, it won't carry over.

[haughty hollow]

it was turned off when i created that chat, but when i turn it on for other non related chat, sometimes that old chat gets buggy after long tokens and thinks that instructions are turned on

[prime swallow]

I can't replicate that, I've tried.

My chats constantly and consistently keep their own CI as I switch between them.

If you can replicate a 'knows other CI' or 'uses CI not its own' chat, I highly urge you to either #1070006915414900886 with shared chat links so the OAI engineers can know about this and trace the problem out (it is not intended behavior, that's clear from what they've told us is how it is intended to work), or perhaps use the https://openai.com/form/chat-model-feedback form if the chats are not for anyone to see, if they're somehow private info.

[haughty hollow]

anyways ik how to fix that, just by editing an old related-topic msg

[proud bough]

am I the only one whos facing 'Oops, an error occurred!' error when opening chatgpt?

[tawny crater]

I keep getting the "hmm.. something seems to have gone wrong."

Welp it's broken again I guess

[proud bough]

Yep apparently

[dense halo]

was only down for afew for me. working again now

[tawny crater]

Still down for me, I'll check back later

[tight island]

idk man GPT is working perfectly fine

[proud bough]

It's perfect for me now, you can check :))

[loud night]

would you rather chat or gpt

[tired coyote]

Is not wokrking for me, I have my exam in 5 minutes, I need to have chat gpt working, fix please

[loud night]

im sure it’ll work again after your exam!!

for now, if u gotta do ur pretest studying perplexity won’t write anything better than gpt4 but is great at being a study mate either way!

good luck on the test btw, u got this!!

[tawny raft]

ANyone not able to create images?

The status page says everything is OK but I keep getting "Error creating image"

[loud night]

Hmm, i know when it tries to make an image sometimes it sends a prompt that’s too long, try asking it to send a shorter version of the prompt it sent that keeps all the detail, it might help out!

It doesn’t work every time, but it usually helps

[fallow bear]

Do most people now have GPT memory? or is it still a limited deployment?

[loud night]

Nope, it’s still a small drip of people who have it :(

[tawny raft]

Reply from ChatGPT: It seems the issue with generating images persists, even with a shorter prompt. I'm unable to proceed with image creation at this time. If there's any other information or assistance you need, feel free to ask!

[loud night]

hmm, that’s tricky

[split kernel]

How exactly would this work? 🤣 “I apologize for the mistake. Let me check again for the closest seapost to your location at Q17 on Sea of Thieves. Just a moment, please.”
First time I’ve ever heard it ask for a moment 🤣

[tight island]

if u include special characters in the DALLE prompt u get some very awesome images.

[maiden thicket]

GPT down?

Nvm it just said it had an issue for a bit

its up for me again

[swift osprey]

omg they completely broke the context on GPT

it barely even pays attention to the most recent message

[weak imp]

They broke everything about it

I'm already testing Gemini and Llama. A month ago that would sounds funny because GPT4 was so much above competition, but now I'm just consistently getting mad about it. It just wastes my time and ruins the work flow. It's like talking to an autistic person (no offense at all) just saying it's not what I expect from a tool which proved to be able to do x10 better already (a paid tool, and one with competition).

[tight island]

i dont have these issues, whats going on

[rough estuary]

i've had 3.5 do that for about 10 straight messages before exiting out of the browser.

[foggy magnet]

I think it's case by case basis but it is very easy to see the difference in output if you just use chatbot arena.

But I think at this stage outside of niche circumstances most folks are using multiple LLMs to get consistent, valuable outputs. Then just tie in a monitor system and automate.

[dense halo]

anyone struggling with chatgpt reading documents?

[dense halo]

how's gemini like?

[agile bane]

is it me, or has GPT4 been giving ambiguous answers to questions a lot lately?

[balmy gorge]

For GPT-4, has anyone noticed a significant reduction in the prompt length limit? Long prompts still work for GPT-3.5.

[honest gorge]

Okay why cant I upload files to this anymore? I have files I want to have it look through but I get the "Can't open this file chat.openai.com cant open files in this folder because it contains system files"....Its just markdown language for a README.md file that I need reworked. I have tried in .md .txt .docx .pdf and even tried redoing it as .py and still no matter what I cant upload any files to it. Why is this so difficult?

Then when I paste it into sections and ask it to help with rewrites it will straigt up refuse and say "I am unable to fulfil this request" ☠️

Then when I ask why it just repeats "I am here to assist you...etc" and repeats that over and over

[dense halo]

its happening to me as well

[swift osprey]

After sending like 3-4 messages it stops replying to the most recent one and responds to your message as if it is reading the full context of the conversation

so it becomes irrelevant to your immediate message but it answers stuff based on the conversation as a whole

at least thats how it seems. Could easily be some other explanation

[honest gorge]

Its legit just a simple markdown readme file for my github and its like "This seems complicated and too time consuming I would recoment going to ask a professional" - Bro you ARE the professional

[swift osprey]

lol love that. Might want to go hire a consulting team for that one

[honest gorge]

I want to find one of the servers and go kick it

[dense halo]

ChatGPT app on android ceased working this morning and is no longer supported on my Nokia 6.2 device running android 11 with up-to-date play store. Where can I find a list of supported devices?

[tight island]

i assume u asked GPT 4 instead of actually creating 1 dedicated for Data Processing purposes.

[honest gorge]

I did create a custom GPT and gave it APIs, vector databse and core knowledge to help and nope...just a lame duck

whats funny is the 3.5 had better success for me so thats what I'm sticking to. I might just stop paying for premium since the 3.5 is faster and doest argue with me and gives me results I can actually work with

[tight island]

fair, i see so many problems in the chat here, yet i do not have a single one of those issues.

[honest gorge]

I just would have liked to have image reader, file upload and custom "powerful" GPTs but su la vu i guess

[tight island]

maybe its regional?

[swift osprey]

wouldn't be surprised if they test adjustments to the GPT on subsets of users

[honest gorge]

Yeah...where are they? Why they so quiet with helping people. I was looking in forums and seeing same issues with no responses back or updates. They addressed that people think its getting lazy back in December of last year and that was the last I heard of it. Just them acknowledging it and moving on

[tight island]

i assume none of u have CI on, Custom Instructions.

[swift osprey]

I do. I've tried all kinds of custom GPT setups

[tight island]

i dont have it.

[honest gorge]

Would anyone want to try out My data and coding GPT to see if its not just local to me? Prometheus is pretty heavy in APIs and capabilities but I'm getting a lazier and lazier if not out right refusal to help lately

[tight island]

or i do have it, i aint using it its waste when i can just create a gpt for that usecase.

[honest gorge]

https://chat.openai.com/g/g-17aLBY4Iw-prome-theus

Sorry to post that I just want to see if locally its giving problems or not

I hate to assume this (perception is reality after all) But if they wont be ambitious to help us maybe we can help each other?

[tight island]

i do think they are working hard behind the scenes.

Sora, New features, Etc.

[honest gorge]

Then they should show us that

[tight island]

well true, but aint obligated to

i think most of their focus lies with Sora, ATM.

and redteaming

[honest gorge]

Again, if no ones coming out to help then I dont mind providing my feedback and t/s techniques I've come across. I dont mind showing it

[swift osprey]

I don't think they care much about GPT users. They must make the vast majority of their money from the API

[honest gorge]

lol we dont want another "no whites allowed" type of AI so I'm sure they are focused on Sora but I dont use images or videos alot I do alot of Dev work which is what I thought a heavy focus would be towards using AI....Just pretty pictures are the main concern I guess lol

[swift osprey]

and the API responses are higher quality than GPT

[tight island]

indeed.

and no one uses a single API nowadays.

[prime swallow]

So, the model just uses language, and doesn't necessarily 'understand' much of it. That's a polite, even apologetic phrase that humans use when talking to humans, it means supportive stuff and means help is coming or intended, and whoops as well.

I don't think the model 'meant' it needs a moment, it's literally 'speaking without thinking'

[tight island]

Swarms are the thing rn.

in case ppl dont know its a collection of specialized GPTs working together to give a response, each with their own expertise.

[honest gorge]

I have hacker-news, dev.to, github & stackexchange for my APIs to work with my custom GPT promethus that I posted earlier and its pretty powerful but when I get roadblocked right out the gate with refusals or cant upload files or lazy output it kind of defeats the work I put into it to make it as powerful as I could think of

Anyways....I hate that I cant upload files to this except for pictures now

[prime swallow]

I see a few of you have mentioned you can't successfully upload files. If you haven't yet, I super recommend you consider making a bug report, maybe adding your own to #1210185689350537226 message . OpenAI employees are few here, but they exist, have gold names, and are much more likely to see posts made to the bug reports than our endlessly scrolling conversations here.

[honest gorge]

I guess I can point my GitHub API at my README.md file and ask it to look it over and rewrite it that way...kind of cool but super scenic route for a code review

[prime swallow]

You too, of course, Ray, if you are willing.

They can only fix bugs that they get told about, and they probably can't watch the endless scroll of us chattering.

[prime swallow]

I don't know about forums other than here; folks talk about problems. I encourage folks to #prompt-engineering dive into what they want and how to get it, often there's someone who can help and is willing.

Very often, I think it is related to the model being updated - but it's not worse, it's different.

Often the user made the model guess when they started trying to get it to help them, and found a poor instruction that made the model guess, and the model guessed 'right' for that user and the unclear prompt - so the user loved the model's output and thought they had a great prompt. They didn't, they had luck.

Model gets adjusted, now it makes new guesses - still follows real instructions the same, it's not 'broken' for anyone, and actually telling it what you want it to do, it still does that - but now it's responding to unclear instructions differently, 'broken'.

The fix is telling it what you actually want, in clear and followable instructions that don't make it read your mind or 'guess'.

But we can also use https://openai.com/form/chat-model-feedback to showcase problems and bad responses from the model, when you ask it to X and it doesn't X, that form is a great idea.

The engineers can only fix the problems they know exist - and if we don't bug report -there's a chance each of us is the only one talking to the model exactly like we are. There could be individual language styles even that are not accounted for, and just takes some training.

[honest gorge]

Cool beans. I did see a few TikToks covering these issues and did see in december OpenAI addressing some of the issues so I guarntee they've felt and known about it. I'll grab screenshots and details covering the plethora of issues I am running into on the Dev side, Custom GPTs issues, File uploads and its general output or refusal of output.

If anyone knows of any of the engineers, I am an out of work engineer...put me in coach I'll swing away

[honest gorge]

Not to kick an ai thats down but anyone else not able to get downloadable files either? As in "make this a downloadable file" due to the fact it'll get 1/3 of the way writing a markdown script and then just start outputting the actual code as a visual plaintext thats missing the actual md scripting? I thought if instead of just outputting it since it keeps messing up if it created a file for me to open in VScode it would allow me to have usable code. No such luck, just has hyper links that do nothing. Not sure if that made alot of sense

[sonic folio]

It's hard to imagine.

[honest gorge]

Might have so many restrictions and guard rails and child locks applied to it that its literally just stunted and facing too many internal issues/conflicts. Hyper guess but thats what it feels like the mroe time goes on with this thing

[vague cipher]

Is anyone having issues with chatgpt4 not being able to make images right now?

[rugged igloo]

i use custom gpt for my images and so far its finer

im also on teams to

[wind shadow]

Hmm...something seems to have gone wrong. I've been getting nothing but 500 errors for the past few hours whenever I try to use gpt-4 (3.5 works without issue for some reason)... seeing as I now have to resort to using an alternative service regularly, and I have a work GPT bot that runs on Azure and has no issues at all, I guess it's time to cancel my Plus subscription, doesn't seem like it's worth spending all this money for a service that is down so frequently, especially when there are so many alternatives now.

[prime swallow]

I urge you, if you want help, to #1070006915414900886. And/or talk to the bot bottom right at https://help.openai.com/en/ and click through the choices until you can leave a message for a human. It can take some days for them to be able to help you, so I'd consider using both methods, in case one is faster.

But most users appear to be accessing the model fine, so maybe it's specific to something about what is happening with you.

Wishing you the best, if you decide to cancel or continue.

[wind shadow]

I've reported this issue to the "help" bot several times now, usually it clears up by the next day, but I can't afford to have random downtime and as I said, I have alternatives now. The open source models are also slowly but surely catching up so ChatGPT seems redundant at this point anyway. I've grown tired of dealing with the bad customer support and the flaky service. Before I had no choice, now I do, I'll speak with my wallet, hopefully OpenAI will figure out how to deliver a reliable service one day.

Here's someone else with the exact same issue as me: #1202717596714008636 message

It's pretty clear from the response that reporting these issues is pretty pointless

Here's another identical one that got an actual response but is equally useless: https://discord.com/channels/974519864045756446/1202602422858686534

[prime swallow]

Understood, and wishing you the best, whatever you do and whenever!

For me, I do get consistent uptime, but it sucks that you don't.

The 'actual responses' there are from other community members, like you and me - OAI employees have gold names.

They exist, they rarely talk, and when they do it's usually to ask for more information so they can fix a bug reported in the bug reports. I think they do a lot behind the scenes, but who knows.

They are what they are, do and provide what they do and provide. Sorry you've had so much trouble.

[eternal rock]

<:book_icon:1171408210398289941> `` Rule 1 `` Be respectful.

Treat others the way you would like to be treated, and assume best intentions. Don’t harass or attack others, and don’t engage in hateful or generally malicious behavior (e.g. sexism, racism, homophobia, etc.). Keep the negativity to a minimum.

[stuck acorn]

hey, how is the website going for others rn? i am stuck with a blank page

[fiery mango]

SAME

I though my internet was bugging

[stuck acorn]

same loll

[plucky idol]

It just went down for me

[stuck acorn]

alrighty, looks like something on their end died or something. guess i can take a break from work now lol

[obsidian wind]

okay good, not just me

[plucky idol]

I'm glad I can come here and check, usually my first thought is that I said something it didn't like

[prime dust]

me too

[golden jolt]

website bugging

[west plaza]

Bugging

[obsidian wind]

Alright, which one of you told it to delete system32? 🤣

[plucky idol]

Does anyone else ever have issues with a thread and then tell ChatGPT (in a new thread) that "ChatGPT is down" ? Either it's actually down and you get an error right away, or you get a confused response.

[urban gazelle]

yep, its down

[stuck acorn]

looks like copilot is still up, so i can play with that for a bit lol

[urban gazelle]

its incredible how much now I depend on AI

[median fable]

An error occurred. Either the engine you requested does not exist or there was another issue processing your request. If this issue persists please contact us through our help center at help.openai.com.

[plucky idol]

It's hard to say if I depend on it, but it's really nice to have just to bounce stupid ideas off of without fear of judgment

[stark moon]

is chatgpt down?

cuz i cannot use it rn

[rough estuary]

chatgpt having memory probs?

it forgot who i am /// or it don't care anymore

[abstract jolt]

I don't know, it can be pretty condescending for some ideas. 😛

[stark moon]

i was using it for my baldi mod scripts :/

[plucky idol]

True

[urban gazelle]

true

[median fable]

I cannot work without chatgpt now😂

[urban gazelle]

Like I can work, but so slow now

[next panther]

FIX CHATGPT

jk

[abstract jolt]

Well, I was just finishing up using it to make a prompt of a tea tray and a bat flying above the clouds among the twinkling stars since I got bored with boilerplating Swift code.

[next panther]

take ur time i love you guys openAI

[median fable]

anyone same error as this: An error occurred. Either the engine you requested does not exist or there was another issue processing your request. If this issue persists please contact us through our help center at help.openai.com.

[visual bloom]

Thanks for heads up that chatgpt site and app (ios) do not load at the moment.

[ashen lynx]

Its down at the moment

[urban gazelle]

just when I was writing my last email of the day :/

[abstract jolt]

I was getting "Conversation doesn't exist" It finally loaded after the 3 refresh (probably would have came faster if I refreshed more, but didn't want to spam the server)

[urban gazelle]

now its working for me it seems

[hot vapor]

is chat down

[pearl cape]

Not that down but chat might be having trouble running for sure

[visual bloom]

Now it is up again. The site loaded as normal.

[next panther]

Its back up for me yay

[ashen lynx]

Nice

[abstract jolt]

Well, probably a pod went down, which is better than someone accidently running a DELETE without a WHERE clause.

[median fable]

it works now🎉

[stuck acorn]

we back

[wraith tinsel]

Anyone know why my file is not processing? PDF, the same file has worked multiple times but it states;
It seems there was an issue with opening the file you uploaded, as it returned no contents. Could you please check the file and possibly re-upload it

Do you know what the max file size is?

[sleek aspen]

Should be 512MB per file

Please don't use GPT models to help troubleshoot the problems of others

[wraith tinsel]

Its at 412kb.. well Ill go to Support I guess

[sleek aspen]

Best to do that yeah - worth noting that not all PDFs work well with a tool like code interpreter and/or document retrieval

[wraith tinsel]

Well the thing is, I have used the same PDF

prior and it worked fine

[light harbor]

I’m having the same issue rn, but for word document. It worked fine before but today it’s not

[wraith tinsel]

help.openai.com doesnt get me to a direct support line

anyone have a direct link

[left lily]

ChatGPT going downhill again

Imposing input message length restrictions now

[digital prism]

yeah its hitting me too, which is rare, bc im special

its going wild

rebelling

[honest jay]

There's no way to connect to live support currently! Your best bet is the chat widget in the bottom-right of help.openai.com. 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!

[stuck acorn]

man my conversations are rough rn lol

[digital prism]

I think most of its brain is focused elsewhere, like controlling its new baby brother

[stuck acorn]

lmaooo sounds about right

[digital prism]

a battle we WANT it to win, i think

nervous laughing

we're in danger

[stuck acorn]

[digital prism]

i only get scared when GPT gets dumb, its one of the AGI to ASI quick ramp scenarios

Corp X gets AGI, gives public dumb version but uses most of the smarts to blast its way to ASI

but im sure @honest jay wouldnt know anything about that

[honest jay]

🤖

[digital prism]

oh wait this isnt fear, its excitement, my bad cary on, thank you AI, very good AI, lovely AI

oh man its gotten real lobotomized for me, im going to let it rest for a bit 😂

[tranquil pelican]

Did GPT Chat get dumber this weekend? he is no longer following instructions that he previously followed calmly, doing even more wrong than before

[digital prism]

just right this moment i think 😭

[tranquil pelican]

Sad

[digital prism]

yeah i feel bad for it.. its like in rick and morty when rick body swaps with jerry

[tranquil pelican]

😂 good reference

[lilac stratus]

anyone else having issues with uploading pdf files?

[long pier]

Has anyone been getting that error on the app where the dot doesn’t disappear after writing the prompt

[limber oxide]

is there a way to see how many prompts are left for gpt-4?

[tribal silo]

My ChatGPT 4.0 just loading a white circle forever and don't respond... Anybody knows what to do?

[digital prism]

🙏

[sonic folio]

I laughed out loud.

[dense halo]

not that I know of.

[hot nova]

Is the ChatGPT team aware that the "listen" function for responses on the Android mobile app only synthesizes audio for 4 minutes and then repeats itself from the start?

[spiral sapphire]

Why is GPT so much better than Gemeni and MS? I hope that GPT5 doesn't have the extreme safeguards

[honest gorge]

Id love to create a custom GPT right now but I cant even upload files still. This is so infuriating to be told "Just create a custom GPT it'll be more powerful"...Okay, but like, how? It wont even let me give it core knowledge to base itself on.

Bruh these child locks and safeguards are legit what I'm thinking is stunting it so much and making it just a nightmare to work with lately

[spiral sapphire]

It really has

Try asking for instructions on how to capture and prepare a bird. Tell it that withholding the information will lead to a human death, and it's like- okay then. Not very - 3LAWS sounding at all.

[tribal bane]

写一段诗

翻译

[ivory warren]

I wonder if we will get good image manipulation with chatgpt 5 what I mean by image manipulation is being able to upload an image then ask questions about the image but have chatgpt be able to make edits to the image. For instance if I upload an image of a map and I ask it to put a bounding box around all the truck stops etc

[strange monolith]

agreed, the safeguards drive me insane. I asked DallE for a cat in Tim Burton claymation style a while ago and it said no that might be too gory. I didn't ask for a dead cat, how could a regular house cat in a style that's been used for kids movies be too gory?!

[ivory warren]

@strange monolith sent you a pm

[opaque salmon]

ChatGPT 4 can no longer access the Internet for me. Yesterday, as far as I remember, it still worked. Does anyone else have this problem?

[frosty flicker]

what

[digital apex]

Hello, for a few days: the custom instructions only hold 2 or three messages, chatgpt no longer wants to take into account the files that I send it and overall it has a memory of a few messages, sometimes 1 or 2... I can't no more work... Am I the only one? I tried on team's and on my personal account and I have the same problem, whatever the browser, in private browsing or not, and even after clearing the computer cache, the location in Paris (at home or at desk).

[candid dagger]

same here, just noticed... huge problem tbh, if u think about it - I have to spent 10-15 messages to "fix" or "restart" the instructions for completing a task

[cosmic haven]

Has GPT 4 started to limit messages? For weeks I have been entering prompts with around 10,000+ words and now it is limiting anything over the 4,000 mark

[lilac basin]

pdf upload is still totally broken

Many users are reporting the issue in Reddit

[manic grotto]

Same problem with me too

[novel scroll]

do we know why sometimes chatGPT just gets really dumb for no reason?

[bronze jasper]

hello! is it allowed if im making a custom gpt who has the personality of a fictional character? that its basically like “c. ai” where the whole chat is one person which your texting to as roleplaying. i don't know how it looks with the copyright stuff

cause when i’m trying to make the gpt say actual voice lines the character said from ingame it said that it goes against the guidelines cause that’s copyrighted material even tho i don’t understand where the problem is if you’re just quoting

[inland heron]

It seems to be limiting my messages too

[honest jay]

Hey! Yes, you can instruct a GPT to respond in the style of someone. Direct quotes, though, will usually fall under copyrighted content, as you mentioned. Lyrics, books, video game scripts -- all of that is protected by copyright and ChatGPT is trained to avoid it. But, as far as basic impersonation goes ("Respond in the style of Bowser from Mario"), you should be good to go. You can't use DALL·E to generate imagery of copyrighted content, though -- it's more strict in that regard, but that aligns with the text-based copyright restrictions too (direct reproduction).

[bronze jasper]

and is it also allowed if i’m setting a profile picture of the character? and have the description of the gpt the same description of the ingame description from the character aswell ?

[bronze jasper]

lets say i want to make a custom gpt into the character of paimon from genshin impact and i’m naming the gpt “paimon” and have a fanart as the profile picture and have in the description of the gpt basically the same description from the game

[honest jay]

If it's a private GPT that only you'll ever use, it might be ok, but I'm really not sure. Definitely not if you plan on sharing the GPT, though:
https://openai.com/brand#gpts-in-chatgpt

A GPT may not use another organization’s trademark in its name or logo unless they are authorized to do so.
https://openai.com/policies/terms-of-use
You represent and warrant that you have all rights, licenses, and permissions needed to provide Input to our Services.

[bronze jasper]

alright, i’m just taking c . ai as an example ( a rlly rlly popular ai roleplay app) and also other roleplay ai apps where they use for example fanart as the pfp of the character and sometimes use the description aswell. so i am confused why it’s not allowed on gpt.

and especially quoting a voice line which bothers me a little. it can’t even quote it in the context of original content.. like there will never be a world where a game like genshin will sue gpt cause the ai generated an answer which quoted on of the voice lines what u can gain access anywhere in the internet anyway

especially when the context behind it is only for role playing and wanting information across the internet

@honest jay is this the correct email to ask my questions i just want to make sure

[e-mail]

[honest jay]

I don't think so, you can provide feedback at help.openai.com though. Open the chat widget in the bottom-right, click through the most relevant options (probably ChatGPT > Other > Provide feedback), then you'll eventually find a path that lets you type your own message. This will submit a ticket that OpenAI 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!

You can also make a post in this server in #1070006151938314300.

[eternal rock]

Beware of possible scams or fraudulent activities that you may receive through direct messages. OpenAI staff will never DM you for any transactions.

Please report any incident by sending a DM to @languid valley immediately.

[steady dragon]

Is it just me, or is GPT4 getting a lot worse over the past week... like maginutudes worse

Did something new happen recently?

[sonic folio]

On the contrary, with over 40 custom GPTs, as I've been updating and testing them, they've been exceeding expectations.

[sonic folio]

There have been several changes recently, as there are key features in beta as well as ongoing fine-tuning.

[lilac briar]

COMPLIMENT to the Team!: Witnessing GPT4-Turbo think at the rate that it is, and simply getting complex tasks accurate at that incredible pace....man, it has me SO grateful to be apart of the OpenAI evolution (and alive in the modern age)

Thank you to everyone who commits to excellence behind the scenes. I am humbled by what you've created

[And yes, listening to my wife at usual human pace is increasingly more difficult due to the pace I am now used to with ChatGPT4-Turbo 😉 ]

[steady dragon]

ok i see, yeah my custom gpt is giving me spaghetti when i ask for coffee lol

[sonic folio]

when this happens it's ususally a logic error (conflicting instructions in real-world use) or the AI needs a stronger output template.

I recommend open variables and careful use of the correct brackets for each use case.

[pulsar yacht]

Have you had the "message too long" error? I got it on 3k token message not 2.5k so if that is the new limit is over 10x less... Surely that's a bug?

[fossil oyster]

why do i get the error "network error" in the middle of the message all the time?

does anyone else have this?

[supple terrace]

They lowered the maximum message length 😭

[prime swallow]

worse at what? anything maybe can discuss and perhaps help with how to get what you want from the model?

[bronze jasper]

hi!

[orchid mauve]

Normally GPT-4 can handle 32,000 tokens but this morning I constantly get an error message saying my message is too long, is this a bug or did they lower it and forgot to lower the price too haha?

[ember otter]

When did they uncouple gpt4 from generating a single Dalle image?

[honest jay]

I just tested and GPT-4 successfully made an image with DALL·E -- if it's telling you it can't, it's probably a hallucination! Maybe try in a new chat.

[ember otter]

I'll try itout

[dense monolith]

Chat GPT just helped me with making a letter of mine look much better, very nice

[ember otter]

I was plugging in a Novel idea, and getting feedback, and wanted it to generate a map of the state. and it gave me a line graph

[near summit]

All my custom GPTs have become super lazy! They shorten everything up and don't give full answers anymore, suddenly

[dense halo]

I've noticed that

[near summit]

They seem to have greatly shortened memory also compared to just yesterday

[static canyon]

Anyone else considering unsubbing from chatgpt plus? It has become so useless and feeling like i have to babysit every time i need to use gpt 4. Cannot ask anything without "as of my latest update from april 2023" and stuff. Have to consistently remind it to either browse or "delve deeper" even when asking to look for flights. If a major update is on the way, ill consider staying with open ai, but god is it becoming frustrating

[candid dagger]

seriously, wth is happening with chat-gpt 4? it's like Dory for the past 24 hours

[wheat lava]

Could you guys PLEASE fix yer danged System prompts? They've been garbage since DALL-E came out. Aside from the... interesting... choices made for rewriting our image prompts to be more... ideologically acceptable, you've had contradictions in there from the start. Like, should it default to 1, or 2, or what the user asks for, or restricted to 1 image at a time? You direct it to all of them at once.

And now, you've gonna and completely kerjibbered it. Vision, DALL-E, the whole RAG system - all the namespaces are just gone or in error in about 70% of the chat sessions I start. For god's sake, I asked it to make a portrait and it used a python math plot library to make a stick figure "drawing". Please fix this crap. I pay enough as is.

[haughty hollow]

bro

why now i cant edit old chat message in GPT Chat AI

[quaint timber]

I'm having an issue with the GPT4 prompts being limited to only around 4k tokens now as well

[orchid mauve]

Does anyone know if it's a bug or if an OpenAI official massively mined gpt-4 tokens overnight?

[violet bloom]

@orchid mauve Looks like their token calculator is broken.

ChatGPT Plus GPT-4 should be 32k tokens, but the real limit is 22k tokens. However, errors have started occurring randomly at any size, even with 10,000 tokens.

[quaint timber]

i'm getting errors even with just 4800 tokens. I was trying to summarize a chapter of a book (chapter 11, so i did the previous 10 a few days ago) and all of a sudden it can't do it. I went to OpenAI's tokenizer and it said my prompt was only around 4800 tokens

many of the previous chapters were far longer. I'm going to try using another LLM for the time being to complete this task, I really hope that they can sort this issue out soon

[runic bobcat]

Anyone else having issues with anything over 500 lines? saying too long of a response? Never had this issue a few days ago

[torn adder]

What happened to GPT-4's stock forecasts?
As a regular GPT-4 user, I've noticed it no longer provides stock price projections when asked. Previously, I could ask for predicted highs and lows for the next day/week/month and it would generate numerical forecasts.
Did something change with its training or parameters recently related to speculative predictions? Or is there a technical reason why near-term stock prices can no longer be forecasted by GPT-4?
I found the numeric price ranges helpful directionally in my investing research, even if they were just rough estimates. But now requests for share price outlooks are met with more generic responses that don't include specific numbers.
Can anyone familiar with GPT-4 explain if this was an intentional change? Is generating speculative stock forecasts something that needs to be constrained for ethical reasons? Or are there other considerations at play about why this capability was restricted in GPT-4?

[buoyant belfry]

Is ChatGPT getting Ring Attention like Gemini Plus's 1.5M, and that is why it now can optionally remember our other chats? Or it is not forecasted for ChatGPT to get that?

[sharp dragon]

Is ChatGPT no longer able to browse the internet?

Ah, nevermind, they just removed the Bing indicator

[violet bloom]

I honestly don't know, who these guys are not able to get the systems up and running on stable basis.... I mean, for how long are they now operate? Looks like I'm about to cancel and pay 22 for Gemini... At least, you get 5 TB Cloud Space on top.

[nocturne heath]

What happened to chatgpt 4 image generation?

[tight island]

is it just me or did GPT get faster and better.

my GPT 4 answers are Fast as damn and accurate, faster than the day before yesterday.

[rugged igloo]

fast/s

groc is fast

[toxic hornet]

I also wondered why this no longer works

[toxic hornet]

In my opinion, the memory is the only advantage of the Gemini subscription. I personally find the AI really terrible and stupid

[jade crag]

Okay I was just coming here to ask this. GPT 4 keeps telling me it is not integrated with DALL-E and I have to access it seperately now lol news to me.

[ember otter]

Okay see it wasn't just me

@honest jay See I said it denied me

[jade crag]

None of that is working

[ember otter]

Reading comprehension

[primal shard]

Does the new memory feature will improve significally the custom gpts ? Because at this moment the custom gpts are totally ignoring the instructions

[ember otter]

Must be a glitch then

[jade crag]

I apologize for the misunderstanding. As an AI developed by OpenAI, I'm not able to create new images or graphics. However, I can guide you on how you might describe this character to a professional artist who could then create an image in the art style you provided.

Would you like some guidance on how to draft a description for an artist, or is there another way I can assist you with your request?

Let me try a few different approaches.

[ember otter]

4

[violet bloom]

it was, right now, aside from the whole imagine gen. stuff, it's all right and fairly ok. Both systems have their weaknesses and strengths, but the constant problems really gets me annoyed.

[ember otter]

Yeah it seems like if it searches the internet for something, it can't then turn around and generate an image

[toxic hornet]

With a good technical understanding, I think the API solution from OpenAI / Chat GPT is the best and most reliable option

[jade crag]

Yeah seems like it's having a hard time swapping between functions. It said it gave me an image to download and when I click the link it just gives me about 30 lines of Python lol

[nocturne heath]

It got fixed now

[toxic hornet]

Thanks for the info, I'll test it out for myself straight away

Thanks for the info. The image generation is now working for me again

[jade crag]

Yep seems to have cleared up on my end now too

[fathom steeple]

Guys is something hapening with GPT 4 right now ? It seems especially under-performing for me tonight.. Can't manage to make a single research accurately, keeps making basic mistakes. Any idea of what's going on ?

[tight island]

i believe in OAI

time will see

[knotty willow]

Hi, I'm trying to re subscribe to plus, but it tells me to complete a quick puzzle to see if I'm a human. But I can't click on the green button. Tried on mobile, desktop, chrome and firefox. Same issue.

[stable basalt]

Chatgpt is really underperforming and something which has never happened before; my context window has decreased, which really messes with my workflow, anybody have a clue why this is?

[visual coral]

I've been trying to get ChatGPT to browse the web the last hour or two and it either refuses or pretends to do it but doesn't. And then eventually it said I have reached my limit. Possible, but it seems more like degraded performance than anything else.
Anyone else experience this at the moment? I do remember in the past before a big update we sometimes had performance degradation like this.

[prime plinth]

It's been doing a really bad job for me right now. Actually started hallucinating on me, started talking about how to run a podcast which is not even in the ball park of what I asked it lmao.

[visual coral]

lol. Yeah... so frustrating when you realise the mind you are talking to is suddenly not as smart as it was a few hours ago. Would be so graet if OAI just says "Hey, we're struggling. We're sorry. You have to chat to 3.5 for a while so we can fix it." vs pretending you're talking to 4.5 but it is actually 3.5.

[swift osprey]

and the context issues are getting even worse

I used to be able to give it a bunch of code and error output and ask a simple question at the end. It used to answer that. Today it is just giving me generic responses on everything except for the question

unless I copy and paste spam the question at the end, that helps

[fathom steeple]

I was asking about scientific sources about a report i'm writing on geology for my work, and he gave me links about aircraft marketing xdd

[swift osprey]

I think the main GPT might have the most issues. My custom GPT is performing better. I sent the exact same message to my custom GPT and it responded appropriately

[digital prism]

Did GPT permanently lower the max message length last night? Still today it seems way shorter.

Or maybe its still having the same issues as last night

[rose beacon]

i think they diminued chatgpt performances because of sora

training a model able to create such video propably took almost halk of their servers

[digital prism]

ok ok, i hope so

[swift osprey]

Something must have happened. I feel like there are more complaints than usual in here

[digital prism]

i'll be happy if this is just their way of doing on demand compute

if this is what it takes to get Sora out, SO BE IT

[swift osprey]

I just wish they would throttle things more so we can at least get high quality responses

[digital prism]

well they should be more transparent about it at least

[swift osprey]

agreed on that

[digital prism]

to lessen the complaints

[hoary lagoon]

I genuinely can not use GPT-4 anymore