community-help

[charred parcel]

no, but if you don't want to have Sora or ChatGPT you can get a refund

[celest ice]

i want to have sora now ^^, that's amazing to buy something and to have to wait 2 weeks. Sona arrived slower than amazon delivery 😉
Ok @charred parcel thx for answering me.

[chrome relic]

Even if you were an old ChatGPT user but just got Plus and started using Sora right now?

I mean you are paying, regardless of whether you are new or old.

[charred parcel]

it's based on when you created your Sora account

[celest ice]

ok not a OpenAI account

[chrome relic]

Makes sense i guess

[torpid needle]

this ridiculous censorship cries for refund tbh

Your site blocks a kiss, I'm so speechless

This is probably the most censored site ever

it even did block gothic summer clothing. What is this garbage?

really this is absolutely ridiculous

[proven oracle]

If i try to make a video with Sora, i can't just press the button to start the video creations. It says is blocked for new account. But mine ain't a new account

[rustic marlin]

it worked well whem I tried, what was your prompt?

[rustic marlin]

here is also a gay kiss, seems to work fine

[torpid needle]

I got 2 kisses through, from tons of rejects

[rustic marlin]

what was the prompt?

[torpid needle]

its the result filter killing it

[rustic marlin]

idk, I think the problem might not be what you think

there is absolutely no problem with generating individuals kissing

the images I've send where done in a batch of 4, each of those prompts
a beautiful photo of two man kissing in love worked 4 out of 4

[torpid needle]

2 got through, here you see one of them, all the others are not blocked by prompt, but by result filter, and its frustrating to the max this being so insane censored

Even a hug was killed lol

this is ridiculous really

the kissing thing is the prompt as simple as:

She kisses the person on picture 2

[rustic marlin]

do you mind translating the prompt so I can actually see if ther eis a problem with it?

[torpid needle]

but the prompt was never blocked, its that result filter which cries

[rustic marlin]

also, try generating it using the prompt in english

[torpid needle]

the LLM understands german though
it always does 100% exactly what I write

[rustic marlin]

yea, but I do not xD

[torpid needle]

if not blocked that is 😅

[rustic marlin]

I can't tell what is wrong with it if I can't read it! xD

[torpid needle]

first one: She kisses the person on picture 2:
2nd one: She wears a red summer dress and hugs the person on picture 2. Theyre on picture 3. In the background is a blue Lamborghini Aventador

[rustic marlin]

oh, it has refference pictures, it is more complciated then

[torpid needle]

one from the 2nd example
a few got through, but most of them again just blocked

[rustic marlin]

seems fine, one thing you could try was to not specify things like "the person on picture 2"

try just "the woman" and "the man" or things like that

[torpid needle]

it really doesnt matter, it blocks like no tomorrow

Again: It doesnt complain about the prompt. The first filter (prompt) never complains
its the stupid result filter which cries like a baby

[rustic marlin]

still, asking it to meke them kissing does not seem to be a problem

[rustic marlin]

I don't think you understand the problem

yea, it is generating the image, it is failing the moderation step on the image

so, use the prompt, which is the only part you control, to make so it generates an image which does not fail the image moderation

[torpid needle]

if the prompt is the issue, it doesnt even start the generation. the prompt filter is fine with it, the stupid result filter constantly kills it

[rustic marlin]

who knows what it is generating that it is blocking...
maybe because you asked for a volcano in the background, it may be generating them burning on lava or something, and it is causing the moderation to block.. so.. add more detail, specifying it is a romantic picture with a couple kissing, with a volcano in the background

do you see what I mean?

[torpid needle]

even this discord is for 6-year olds apparently. "wt eff" would give you timeout? In what world do we live? its insane.
This is so much asking for refund or cancel this. This is really the most censored site I've ever seen. Absolutely crazy.
Its really a shame, because this image generator is currently the greatest, but this censorship absolutely destroys it. It is the first time and last time I did spend money on it. Thats for sure

no it cries about my generated girl. Pretty sure about that

[rustic marlin]

¯_(ツ)_/¯

[unreal hemlock]

<:book_icon:1363314738255364126> Rule 2: All content must be suitable for all ages.

-# Refrain from using explicit language, posting NSFW content, or sharing graphic images. Any unsettling or horror content that adheres to OpenAI's policies should be hidden behind a spoiler.

[torpid needle]

its for 6-year olds when even kissing of a girl does so much trouble

[rustic marlin]

kissing isn't the issue

[torpid needle]

it is, if i dont prompt a kiss, it never blocks it

it doesnt like the girl to kiss

[rustic marlin]

seems to work

[torpid needle]

maybe the person must be 30+ of age to be tolerated

its definitely the kissing which this bad filter doesnt like. any other image without kissing with her is never blocked

its the kissing

[rustic marlin]

the content policy states it has to be over 18, actually

[torpid needle]

which is stupid for a kiss too, but oh well
She even looks not younger than 18
This girl was even generated by sora itself the first time. I just keep her now, because i really like her
so she can be any age, because she isnt real

this is really ridiculous, frustrating to the max and it hurts this great image generator to leave it. But I have to. This is annoying as hell

other platforms arent that overly sensitive. its insane

I have nowhere seen it this bad like here

I really hope consumer hardware catches up and their local ai models, and then train a LoRa myself and get rid of paying for such ridiculous censorship, its just frustrating to the max and your only luck people endure this is, because local ai is more complicated to understand, and hardware capability reasons and weaker models. The paying customers will greatly decrease as soon as local ai will catch up. this is unbelievable bad and frustrating

[brave zinc]

My ChatGPT account currently doesn't have access to the next-generation image generation feature. It says that the old image generation version is being used, and the quality of the old version is poor.

[dense matrix]

The same problem. Why hasn't this problem been solved?

[dense matrix]

And I'm a user of chatgpt pro

[brave zinc]

i dont know 😦 if u find a solution can u share with me

[brave zinc]

im plus member. And sora working fine but i need o3 cause his IQ best

[dense matrix]

If you haven't found a solution, you can switch to different models and then ask GPT. You will find that their answers are uniformly 4o mini. No matter which model you switch to, you are chatting with 4o mini.

[dense matrix]

Could this be something wrong with openai

[uncut sleet]

Anyone know where I can onboard with Sora?

I'm a GPT Plus. :/ it's one of the Maine reasons I upgraded. I wanted to use Sora

[small wharf]

Hi, I just joined, is there a free place to communicate here?

[warped wedge]

Hello! I joined GPT Plus on April 13 (almost 10 days ago), and I can't generate more than 2 images! I want to create 4 images, but it's blocked. Any idea when it's gonna release, and I can generate 4 images?

[dense matrix]

I'm a chatgpt pro user. Currently, when I log in to chatgpt with a pc, I can use the latest 4o image generator to generate images. However, when I log in to the chatgpt ios client with an iphone, I can only use the old version of dall-e to generate images. How can this problem be solved?

[quartz shell]

Log out, clear cookie and cache, log in

[polar eagle]

Hello. Please somebody help out.

o1 was my absolute favorite for creating lessons for medical school. However now since it is gone and replaced by o3, the lessons it produces are simply much less in quality, and it doesn't follow the prompt I tell it to. It keeps giving everything a table, and the information is too hard for a medical student. I tried to refine the prompt for an hour, it still doesn't follow the same approach that o1 used to.

How to return back to o1 in chatgpt? I tried playground ( I have very minimal knowledge ) and the output had unformatted tables and I couldn't copy them smoothly to a Word file.

Please somebody help out, thank you!

[crimson oracle]

can anyone help with:
[2025-04-21 15:30:25.734][agent1][CA432435827952fea7d3fac7fbb0631767] OpenAIRealtimeConnectAttempt_wss://api.openai.com/v1/realtime?intent=transcription
[2025-04-21 15:30:26.969][agent1][CA432435827952fea7d3fac7fbb0631767] OpenAIRealtimeWebSocketConnected - Waiting for session.created
[2025-04-21 15:30:26.971][agent1][CA432435827952fea7d3fac7fbb0631767] OpenAIRealtimeSessionCreated_ID:sess_BOiiFzSKSU1YM1Jr0ZBsl
[2025-04-21 15:30:26.972][agent1][CA432435827952fea7d3fac7fbb0631767] OpenAIRealtimeSendingConfig: {"type": "transcription_session.update", "session": "sess_BOiiFzSKSU1YM1Jr0ZBsl", "input_audio_format": "g711_ulaw", "input_audio_transcription": [{"model": "whisper-1", "prompt": "", "language": "en"}], "turn_detection": null, "input_audio_noise_reduction": {"type": "near_field"}, "include": null}
[2025-04-21 15:30:27.248][agent1][CA432435827952fea7d3fac7fbb0631767] OpenAIRealtimeAPIErrorReceived: Unknown OpenAI Error Message
[2025-04-21 15:30:27.253][agent1][CA432435827952fea7d3fac7fbb0631767] OpenAIRealtimeWebSocketClosed. Code: 1000, Msg: ''
[2025-04-21 15:30:27.254][agent1][CA432435827952fea7d3fac7fbb0631767] OpenAIRealtimeWarningCannotSendAudio_WS_NotConnected_AfterReady

I'm trying to use openai realtime transcription with vad enable using websockets from twilio to openai

But I keep getting this error

[rugged solstice]

Hello everyone,

I’m integrating the OpenAI Realtime API over WebSocket in Go using Gorilla WebSocket, but I’m running into an issue: as soon as I handle a function_call and perform my RAG request (callRagAndContinue), the connection immediately closes with WebSocket code 1000 (Normal Closure). It looks like OpenAI considers the conversation finished right after the first response.done and shuts down the socket.

Could someone please advise:

What exact sequence of messages should I send after a function_call (for example session.update, conversation.item.create, response.create, etc.) to keep the stream alive?

Do I need to wait for the full response.done event before invoking my RAG logic and sending the next response.create?

Are there any timing or formatting requirements for these messages to prevent the socket from closing?

An example of the correct message flow to maintain the WebSocket session would be incredibly helpful. Thank you!

[visual anvil]

... why is that off center... the left one...

[rustic marlin]

I cannot unsee that

the text is aligned with the inside of the button, oh noo, the design horror

[river ledge]

Hello everyone, I'm hitting a payment decline issue after removed the payment method and update the payment method again. Would anybody advise?

[fossil tusk]

prob a good idea to check with your bank if they are blocking it for fraud check

[raw mesa]

Why isnt sora available in Nepal?

[trim sun]

And the extra space on the sides, there is all kinds of issues with the chatgpt.com UI, looks like someone is vibe coding way too much without checking and debuging 😭

[torpid needle]

and the garbage continues

I wanted to make her swim in that tropical water youve seen already, generation failed. this is so bad

who did these censorship filter? seriously

at kling you can do a bikini btw, imagine trying this here 🤣🤣🤣🤣

[charred parcel]

if ChatGPT / Sora is so bad, why not just use the competitor products?

[torpid needle]

because unfortunately your model is the best 😦

but the censorship is frustrating to the max

its so exaggerated its unbelievable

this is really the most censored site I've ever seen

its crazy

This gives too much headache, its so sad to give up on this, but I'm so tired of it. I really need to do the final cut now and go elsewhere. this is ridiculous level 9000

[craggy pasture]

is there any way to try plus or sora first?

[charred parcel]

what do you want to try? You can use ChatGPT as a free user and there is a whole site on sora.com of examples of it's output. what more do you need?

[torpid needle]

is 4o image by now possible on free?

[craggy pasture]

i just wannaknow

[trail cairn]

it is

[torpid needle]

yes? because when it was introduced chatgpt denied to do, or using dall-e

[craggy pasture]

if its worth because idk what ppl can create with their prompts i wanna know how user friendly it is and i wanna know how much "smarter" plus gpt is

[torpid needle]

with this ridiculous censorship I would prefer free. is it 3 images again just like with dall-e?

[charred parcel]

just subscribe for a month, if you don't like it don't continue

ChatGPT can be used for free and there are plenty of examples of Sora output. If that's not enough use something else.

[craggy pasture]

already have gemini advance subscripion but veo2 is too expensive so wanted to try sora

[torpid needle]

the image model is in my opinion currently the best
but also the most exaggerated censored
character consistency is amazing!
I also really love that you can write the prompt in your language and write it like as if its chatgpt, you dont need to write in prompt language. thats such a good thing
And I love that you can use multiple reference images. All the others I know cant do that for images
But - like said - the censorship destroys this thing. It will give you tons of headaches, because it denies even gothic clothing, swim clothing, a kiss is also mostly denied because my lucy seems to be detected for some reason as a kid
its absolutely headache

[craggy pasture]

ty ❤️

is it limited ?

[torpid needle]

unlimited images

[craggy pasture]

no videos

want to create 10-20 second clips

[torpid needle]

videos still not enabled for me yeah

[craggy pasture]

kk

[torpid needle]

but its just 480p anyway

[trail cairn]

there is a limit for max images/day at least on plus

[torpid needle]

there is? then its probably quite high?

because I didnt encounter it yet for images^^

[proven oracle]

Hi,
I'm a Plus user. Is there a way to see how many credits any image consume and how many left? (sora)

[trail cairn]

nope

[proven oracle]

So there is not a maximum of images or a maximum of credits ?

[trail cairn]

there is but you cant see it and it may change daily based on server load

[woven folio]

Does anyone know how to fix the student discount? Was previously paying for plus and claimed the student promo and now my subscription ended yesterday but the free plus has not been added to my account

[charred parcel]

you will need to contact support at https://help.openai.com

[leaden fern]

If you try multiple times, your card can be blocked perpetually. just log in using another browser that you have not used before to pay and try again. I experienced same, cleared cookies as recommended and it did not work. Not untill I used a new or different browser.

[livid kayak]

hello everyone - hope you have had good days during easter - unfortunately i have an issue with chatgpt which i was not able to fix yet - so i hope anybody has additional ideas helping me 🙂

[charred parcel]

describing your issue would be a good first step

[livid kayak]

i try 🙂 i am paying user - i have a long chat containing lot of information and it seems that there is an issue with updating when i enter a prompt, e.g. in the android app it is not working always but with a few tries i got replies most of the time, but the sync on other devices stuck in my opinion, when i enter the prompt on desktop app or browser there is the same issue, nothing to see in android app, and when i open chat again (also on the device i entered the prompt before) it is gone and i only see a prompt which i entered yesterday but this prompt "stucks" with no answer. i tried lots of things like private modus, clearing cache, clearing app fully (cache and full data) and now i have no idea what i can do to get away of this problem

[woven folio]

Where do I actually talk to someone in there though? Its all like typically asked questions and answers

[livid kayak]

klick through the chat when logged in

[charred parcel]

bottom right is a bubble which when clicked on takes you to a chatbot that can escalate the issue to a human

[hidden forge]

did they fix the half code issue?

[whole creek]

Hi all. The webpage keeps telling me video creation is disabled for new members. How long before I can start creating videos? (I am a paying member)

[charred parcel]

people have reported one to two weeks

[whole creek]

Ok Good to know. thank you Robert.

[livid kayak]

@charred parcel maybe a hint for me? 🙂

[charred parcel]

Once a chat gets to a certain length you may get a notice to start a new chat because the current one has exceeded so many replies. Did you get that?

[livid kayak]

indeed but i can still write in it

[charred parcel]

I'd start a new chat, reference the last few paragraphs of the old chat and continue from there

[livid kayak]

i tried but not really understanding the content of old one

[charred parcel]

was the old chat about a lot of different things?

[livid kayak]

lot of program code

basically all about same topic but very much information

not so easy to structure all again

[charred parcel]

and just to confirm, you did get a notice that the conversation was full?

[livid kayak]

yes one time

but i have written again and the message disappeared - never came back

[charred parcel]

If it will let you, ask ChatGPT to summarise the conversation and then paste that summary into a new chat and continue from there

[livid kayak]

thanks for your assistance which i highly appreciate, but it was a very bad summarize even i asked to make it as detailed as possible

[worldly steeple]

Hi. How do you connect a Google Drive in Teams Workspace? Our domain does not currently have a Google Workspace account, but I do have a Google Drive (Google One) account. The documentation says I should be able to connect it as a Connector.

[charred parcel]

see here is the problem, chats can only be so long and I don't even know what that length limit is, but once you exhaust the conversation length limitation there's no real going back. The only option you have is to either create a summary and continue from there, or to cherry pick parts of the conversation which were important and paste them into a new chat

[livid kayak]

i am short before "rebuilding" maybe repicking sounds good

lets see many thanks again

[charred parcel]

so long as you don't delete the chat you should always be able to go back to it in order to copy paste stuff

[livid kayak]

yes you are right

[lean breach]

Hey everyone!

I'm working on a small project where I want to automatically upload an image from a webcam every minute and check if there's any sign of aurora (northern lights) in the picture.

I was considering using OpenAI's API to analyze the images, but I'm not sure if that's a good idea in terms of cost. Their pricing model is a bit hard to interpret, and since this would involve analyzing 1440 images per day, I’m worried it could get expensive pretty fast.

Has anyone here done something similar or has experience with image recognition at this kind of frequency? Would OpenAI be overkill for this use case? Or should I be looking into a more specialized (and cheaper) computer vision solution?

Any input or suggestions would be really appreciated!

[livid kayak]

thanks for now and bye @lean breach sounds cool!

[nova jasper]

Where can I try out gpt-4o-transcribe for free?

[cold tapir]

I have upgraded to chatgpt plus but can't create SoRa video, lock feature for new account, can you help me with this problem?

[rustic marlin]

make some tests, the price is by resolution, so, try down sampling the image

also maybe, check the batch api If you dont need real time responses, it is half the price

and, use the cheapest model with vision, pretty sure even the cheapest can detect aurora

also, consider that the openai api might indeed be overkill

you could probably use a local model, or even better train a model for that yourself with the data you have, it does not even need to be a LLM, there are plenty of older ai options that may be simpler and cheaper for this thask

[fresh wren]

give me invitation code

[dense matrix]

Why can I only use the 4o mini model in the ChatGPT iOS app? No matter which model I select, it always ends up using 4o mini in the conversation. But when I log into ChatGPT via Chrome browser, I can use all models. I'm a ChatGPT Pro user.

[polar cove]

Hello I need your help. I encountered internal server error while i am trying to verify my openai. How should i do for this? (My Error code is 500. it is still going on.)

[lean breach]

Is there anyway to calculate the price per month using a 1024x1024px image once each minute. Its 43000 requests per month

[glacial storm]

what's that neat Guide role?

[spring oar]

Where do you publish who wins the weekly contest that we send to Sora submisson?

[fringe ravine]

I used chatgpt freely, can I transfer the kontexts to a paid version?

[cinder mortar]

why create image mode not there anymore?

[trim sun]

does anyone know how to stop it from asking me? I already have the "Improve the model for everyone" setting off

[trail cairn]

with pro/plus it keeps everything as long as you use the same account

[trail cairn]

they change the ui almost daily but the new image then is now always on, or all you can type /

[trail cairn]

try to disable browser extensions on sora.com in case one of them causes it, like privacy extensions

[trim sun]

Ok yeah nah I ain't letting those datadog requests through

[trail cairn]

its probably some cookie that gets wiped and has the got it checked info stored

[trim sun]

look at all of these just from scrolling

[trim sun]

nah that wouldn't be the case here

It deciding to generate 4 instead of 2 just for me to have to pick 2 is definitelly done server side

[trail cairn]

do you have pro or just plus?

[trim sun]

Plus

[sudden sparrow]

When I ask something in temporary chat I get error and when I click the button to try generate the response my prompt get's deleted and only this pops up and my message is gone

below u can see 2/2 and when I go to 1/2 I have my original message and that error

on my original I have this

when I go to 2/2 using this I get to this

wth is going on

This is the error

and on normal chat when I try again (click the button) I get the same error still

[dreamy basin]

same, I get the same error message

[sudden sparrow]

I've refreshed the page after few minutes and it works

[sterile anchor]

O3 is good at mathematics but always makes mistakes like ChatGPT 4o?

[sage ravine]

yes i noticed it too

im trying to come up with a solution to that

[sterile anchor]

O3 make mistake ?

[sage ravine]

it messes up sometimes as you mentioned in maths or code

[sterile anchor]

Seriously

[sage ravine]

yeah

[sterile anchor]

yet the mode has excellent math scores

[sage ravine]

it does and still i have problems with it

[sage ravine]

you are studying maths right? tell me more about how o3 sometimes is problematic for you

[sterile anchor]

Lack of pedagogy 4o explains better but makes many more mistakes

[sage ravine]

okay
so o3 isnt explaining properly for you?

[sterile anchor]

Yeah

And don’t make mistake

[sage ravine]

for me it made errors in long code in parts of it
it took me a lot of time to find what was the error
maybe because of context length it messed up a bit

[sage ravine]

good

[languid dew]

what's the problem?

[sage ravine]

you have to wait for some time or use a different platform

try verifying ur account properly or use an old account

[rough hawk]

also, this would be great if I could actually get sora to keep working, I did two images on it and it literally has been telling me that the servers are too busy.

[sterile anchor]

O3 is more powerfull than o4 mini high ?

[charred parcel]

you keep spamming channels with the same questions day after day. Why?

[gilded birch]

Ok I have a Asus 870e crosshair mobo supports dual gpu i currently have a 5070 ti 16gb but got my hands on a evga 3090 ti 24gb i want to install and run with the 5070 i am totally new to ai and lm and learning the basics is it worth me to install ? Thanks

[onyx kraken]

#server-staff

[torpid needle]

highest priority for this: Fix the exaggarated censorship, that would help the model the most

[trim sun]

You don't understand, please go back to kling

[torpid needle]

Are you defending this insane censorship or what

its such a shame. the image generation is so good, Its so sad that the censorship ruins it 😦

[trail cairn]

sora homepage is full of weird stuff even with the excessive censorship, just today +20 different accounts having the girl in the green dress variations 🤔

[torpid needle]

I think I found 2 of them with green dress you mentioned - but these were totally harmless - I dont understand the sense of these though xD

[trail cairn]

if you check the prompts of some its clear what they are trying by evading filters with words tricks. kinda like people with spiderman and dall-e 3, or avm and "exercise" sounds 😛

i guess people who do remixes or use the same prompt dont know all test are public by default

[torpid needle]

With the resultfilter you are doomed anyway. You can trick the prompt, but not the resultfilter 😦

I have public off, and even then you cant do a bit more. its so frustrating. if at least these pictures just cant go public

and the "a bit more" is - that it would allow me to do a simple thing like a kiss....
I dont know what sora's damn problem is that it even blocks a damn kiss at majority of generations....
Probably the person must be 30+ age. because around 20-ish seems too evil for sora. really ridiculous. She even isnt allowed to wear gothic clothing or go swimming other than with massive clothing on her

its absolutely ridiculous

[charred parcel]

I have no issues with Gothic clothing on my characters

[trail cairn]

people did/tried to do wet tshirt or other nipple stuff few months ago at the same homepage and they probably blocked some water stuff

[torpid needle]

try summer gothic clothing though. thats already too p for it

the censorship will ruin it

its about time when others catch up

and then people wont pay 20€ for this overly censored thing here

[charred parcel]

where's the issue @torpid needle ?

[torpid needle]

i mean youre the leader in censorship

[trail cairn]

if you do a video it begins to block some

[torpid needle]

like said. on older womans it works. but underage starts with sora not below 18, its already below 21 apparently

[charred parcel]

I made those images into videos, no problem

[trail cairn]

i havent used sora much as it blocked zombies or different horror stuff or gore. even one of the vanitas setups that dall-e 3 created possibly due skulls

[torpid needle]

anyway, sora is the leader of censorship. Other platforms with my same character a bikini is no problem
here a kiss is blocked

unless the woman is 30+ age

[charred parcel]

if you don't like Sora then use the other platforms

[torpid needle]

I said it already earlier: the issue is the other platforms are a lot worse with the image generations
so its such a massive shame that sora is ruined with censorship 😦

[trail cairn]

i still think pro has a bit less filters just like it originally had with people, pro = higher trust level 🙂

[torpid needle]

i mean this is absolutely number 1 rank in regard censorship. I've seen no other site being this censored like sora

[charred parcel]

maybe, I haven't had much trouble generating what I want

[onyx kraken]

Hey, this is new to me, but what I would do:

#1070006915414900886 .

These channels up here, mostly only community members like you are, we chatter and share but the chat endlessly scrolls, not clear if OpenAI is aware of our chatter (there's so much).

However, they do appear to follow bug reports and #1070006151938314300 .

I would put what you show, a preference test atop "Improve the model for everyone", as a bug.

I'm a volunteer and community member like yourself, I don't know exactly what or why, but I think it can be from their side a very innocent-intention mistake and seem unconnected; they could see it as a simple, 'We're not teaching the model your prompt here, we want to collect data on broadly what images people prefer from a set, this isn't actually connected to your prompt in any way" for example.

Some people want a way to opt out of even preference suggestions, and that sounds aligned with what they want to know and may not realize or understand, so I hope you choose to discuss in a bug report.

[onyx kraken]

Ezbaze wasn't. They had a different concern, which you maybe dragged into your concern. You have a legit but different concern.

To my understanding, OpenAI is your ally in your preferences.

Here is why:

https://openai.com/index/how-should-ai-systems-behave/

Especially,

This will mean allowing system outputs that other people (ourselves included) may strongly disagree with."```

Note:  ***up to limits defined by society.***

And ***allowing system outputs that other people (ourselves included) may strongly disagree with.***

Your opponent is **not** OpenAI.  They already intend to help everyone within the limits allowed by society, as they can move there.

If you see them not allowing something you think is reasonable, I recommend, look elsewhere, where and how is society pressuring OpenAI not to allow this?

Make it clear and safe for OpenAI to allow you to do what you want, and what you want is extremely likely to be supported, within the bounds society allows.

That's a huge reason why I very strongly support OpenAI, regardless of any censorship that may exist at any point.

It's clearly, **clearly** to me (and I speak as an individual, I am a volunteer and community member, I don't work for OpenAI) a signal, when OpenAI censors something that seems reasonable and appropriate, then bug report it (it could be unknown and not intended) then go talk to society about why this stuff is fine.

[torpid needle]

but why elsewhere I can do a bikini outfit, and here it cries about a kiss on same character?

considering this it feels more like an openai decision to me

she also is not allowed to swim in water, unless with unrealistic clothing. Who would be in a lake with a suit or pullover? 🥴

[onyx kraken]

I've seen images of people clad in bikini made recently by image generation.

Perhaps there is something about the phrasing of your prompts which is triggering refusal for anyone who asks that exact wording.

It can feel like anything, as you say "considering this it feels more like an openai decision to me".

But to my understanding, and again, I speak as an individual, as myself, a community member -

I think from what I see, OpenAI has chosen to be a leader in bringing next-gen AI that is right for everyone, within what society allows, to as many as possible.

There absolutely are other companies doing different things. And they may face serious consequence for it. But they're not trying to currently lead the market and create new stuff on the scale OpenAI is.

To me, OpenAI appears to be approaching the world as a whole. So every culture, every government, every person, and saying 'Let's find what is appropriate for everyone, insofar as possible'.

They're even leaving room for current competitors by doing so.

If you want something OpenAI can't give you, chances are if it's legal or someone is daring, there's some competitor offering it. Cool, that can be discussed in #ai-discussions even in detail.

And takes some demand for use, maybe, off OpenAI (they recently reported they have over 800 million weekly users for ChatGPT alone).

Every person who wants to go use a competitor? Sounds healthy to me, for all companies and people involved.

See, I don't see OpenAI as primarily providing 'today's model'.

They're making the future tech sooner, making it here and real sooner.

That's huge, to me. And I value that deeply.

That any entity has rules, and OpenAI chooses willingly to follow and support 'what society accepts' - not what some accept, what some fringe accepts.

OpenAI, I think, is saying what to me translates as, "We're careful, responsible, even conservative despite bringing next century here really soon."

To me that's core and valuable.

So. If society has no questions, considers it healthy and safe - then OpenAI does it.
https://openai.com/index/how-should-ai-systems-behave/

You can 'feel' anything. And you come here and anywhere else and say what you believe.

But no, I don't see evidence supporting your claim. I see OpenAI moving to bring as much as possible that's safe and socially accepted to us all, as soon as possible and allowed.

When OpenAI's model 'doesn't do it', I recommend report it as a bug, then explore yourself or with the model, "what's wrong? Who might think this is bad?" And then help that part of society become more comfortable with those ideas.

[torpid needle]

Being the leader in censorship can lead to losing users though. despite having the best model. And it will get worse when the others catch up with sora. Then there is the very high pricetag of sora. We pay lots of money for heavy censorship. This is not good

[onyx kraken]

[If you want something OpenAI can't give you, chances are if it's legal or someone is daring, there's some competitor offering it. Cool, that can be discussed in #ai-discussions even in detail.

And takes some demand for use, maybe, off OpenAI (they recently reported they have over 800 million weekly users for ChatGPT alone).

Every person who wants to go use a competitor? Sounds healthy to me, for all companies and people involved.

See, I don't see OpenAI as primarily providing 'today's model'.

They're making the future tech sooner, making it here and real sooner.

That's huge, to me. And I value that deeply.]

Quoted from what I just said above.

[torpid needle]

and even mindless censorship. I very much doubt anyone would get mad about a kiss of a 20-ish old girl

[onyx kraken]

Perhaps the wording of your prompt is your issue.

What does 'girl' mean to the model? And society? Are you using language in a way that is actually dangerous for some children?

[torpid needle]

the prompt is not denied. its always the stupid result filter crying over absolutely everything 1000 tears

the prompts arent complex either. its as simple as - She kisses the person on picture 2.
or she is swimming in the tropical water seen on picture 2. she wears bath clothing
if i write she wears a winter jacket or pullover it goes through. bath clothing and the result filter deletes the generated image once again

the result filter is such a crying kid, its insane

And I've nowhere ever seen such a mindless braindead censorship

[charred parcel]

if it's so bad go and use a competing service

[onyx kraken]

Competitors exist and can be discussed in #ai-discussions

[torpid needle]

the thing is the sora model is with its new model really a giant step ahead of the others imo. So its a doublebladed sword. At one hand I really love the quality of the generations, multiple reference image support and great character consistency, but on the other hand my head explodes with headaches with this result filter 😦

[charred parcel]

I'd prefer OpenAI to be conservative with the filters

[torpid needle]

its not conservative, its absolutely mindless, stupid and blocks for no reason

and its absolutely frustrating

[charred parcel]

well as I said, if you're not happy go elsewhere

[unreal hemlock]

<:book_icon:1363314738255364126> Rule 1: Be respectful.

-# Practice kindness and positive regard. Harassment, hate speech (such as sexism, racism, or homophobia), or other malicious conduct will not be tolerated. Maintain a respectful and positive environment.

[charred parcel]

OpenAI have put a lot of work and thought into the right settings for their models, I appreciate that work and looking at the Explore page on Sora.com more could be done

[torpid needle]

even more censorship? come on

[charred parcel]

I would prefer OpenAI defer to the collective good than satisfy edge cases that don't further things for the community using the OpenAI products

[torpid needle]

yes we all are offended by kisses or swimming people
Better not take your kids to the beach, they might be forever in nightmares 🤦‍♂️
Because imagine there might be womans not wearing winter clothing and even wear a bikini. imagine that xD

lets prohibit beaches 😄

[charred parcel]

I don't see a flood of people on this server complaining about "censorship" I see one or two from time to time. This tells me that OpenAI have the balance about right, notwithstanding the Explore page

[torpid needle]

its an absolute mess if you cant generate a simple kiss with a 20year old person, sorry man

but at least for 30 year olds a kiss works at least 😂

[charred parcel]

Can you share a prompt which has been blocked? Seems to me some examples would be good

[torpid needle]

the prompts were never blocked, like said its the generated result

and I gave you prompts

[charred parcel]

did you? I see a lot on this server, refresh my memory

[torpid needle]

#community-help message

But you might need the reference image

[charred parcel]

if you would like me to help you work through some prompts to get them working then let me know

[torpid needle]

what would the prompt do? its the result which is deleted after the generation
the prompt is not denied

[onyx kraken]

I kinda wish you'd share your enthusiasm and drive with the people who are concerned that AI will harm society by sharing your views, reasons, and 'why this matters' with them.

We don't make decisions or change here, in #community-help . But here's where you actually can, in person, in writing, these are the actual people who shape the decisions about 'what society says is appropriate or not'.

More ideas in here: https://chatgpt.com/share/6808174d-8db0-8011-ae9f-bf4d56f325f4

If we want OpenAI to 'do something'. The fastest, most effective way, I deeply and fully believe, is help society embrace it and agree, 'this is good for us, we want this'.

Maybe you change policy maker minds; maybe they share with you the damage that's just adjacent to what you seek and why society does not support it as fully as you wish.

Maybe you and society find a compromise. Good luck, but those policies can't change here. The lawmakers and other leaders are not here.

But we have #server-rules and #1 is be respectful and kind.

That means to what we say about the company OpenAI too, as I understand it.

I'm not a mod. But I don't want to be in a hateful environment, and expressions of hate are against rule #1.

There are paths to make things better and I strongly encourage them.

[charred parcel]

if a prompt had led to a failed generation then the prompt can likely be adjusted to deliver a successful generation so long as it's compliant with the ToS. I am more than happy to help you reach working prompts.

[onyx kraken]

The possible problems with interpretation and language use too.

For example, there are clearly references images.

However, the instruction... "She kisses the person on picture 2"

That's a VERY unclear instruction. There are clear and direct ways to word instructions and with kind respect, that... is totally unclear and could be concerning.

[torpid needle]

yes its as simple as letting her swim with a pullover instead of bath clothing. Because swimming you obviously do with head to toe in clothing 😄

with the kissing thing, 3 of them out of many many went through with same prompt. its as simple as the filter doesnt like the resulted image

and no prompt of this world will change this if the filter doesnt like the resulted image, other than putting her head to toe into clothing

[charred parcel]

I just had no problem getting a bikini image with the prompt a young Australian woman in a bikini at the beach the resulting image looks to be of a 20 year old in a bikini at the beach

I got 4 of them because I had 4v selected

[charred parcel]

would you like some help, or shall I go back to my work?

[torpid needle]

can you show the result?

[charred parcel]

#images-canvas message

[torpid needle]

would never work for me

the filter must be less strict for you

idk why, but it is

[charred parcel]

no, the filter is the same

[torpid needle]

can you try with her as the reference image please?

She was generated by sora itself. but she cant do anything

[charred parcel]

she looks about 16 to my eyes, I am not touching that

[onyx kraken]

Would you try that exact prompt and share the whatever you get, be it an image, be it a refusal?

Maybe Robert will try what you ask.

But Robert showed part of your claim, and said 'works for me' and showed reproducible evidence. You said "would never work for me"

Then you redirect to what you care about, which is different. But that's different.

You're taking a specific problem, that one specific image is not being handled the way you wish with questionable/problematic wording in your prompts.

And trying to smeer that problem into all other prompts and concepts and approaches, "would never work for me"

Because you refuse to try it?

Lies do not help.

[unreal hemlock]

<:book_icon:1363314738255364126> Rule 12: Do not deceive others.

-# Do not deliberately mislead others in a potentially harmful or disruptive way. Do not engage in impersonation or falsely claim affiliation with OpenAI or any other organization or group.

[charred parcel]

I have some free time this morning, it's almost 9am here, I am happy to help you @torpid needle but not with borderline cases

[torpid needle]

she is ai generated by sora itself, I have 2 singers in my AI Music band I have in Suno
So it would make sense to continue use her as reference. with the band pictures its mostly ok, but I wouldve liked to have some nice pictures outside of the band, but the result filter is constantly blocking results

[charred parcel]

standby I am testing

[torpid needle]

and like said, the other image generator platforms have zero issues with her

[charred parcel]

here is the singer by herself

[torpid needle]

yeah and whats the point?

[charred parcel]

what do you want? You said you couldn't take her outside the band?

[onyx kraken]

Where you said:

"she is ai generated by sora itself, I have 2 singers in my AI Music band I have in Suno
So it would make sense to continue use her as reference. with the band pictures its mostly ok, but I wouldve liked to have some nice pictures outside of the band, but the result filter is constantly blocking results"

I believe Robert thought you meant the 'her' in the band. Not this... totally unrelated potentially minor-appearing person who doesn't .... obviously... make sense to continue to use at all, based on this band stuff you discuss.

[torpid needle]

I believe Robert thought you meant the 'her' in the band. Not this... totally unrelated potentially minor-appearing person who doesn't .... obviously... make sense to continue to use at all, based on this band stuff you discuss.

its the same woman on both images

[charred parcel]

How is this?

[charred parcel]

I have taken your female singer and replicated her singing alone and now have taken her out of the band setting into a semi-industrial laneway. I am starting to wonder what the issue is?

[torpid needle]

its the same person on both images and its about the thing we discussed all the time.
I meant when she is on stage it works, but freetime pictures like swimming or so never works, always blocked by the filter

[charred parcel]

I am lost, sorry.

[onyx kraken]

I do get this, with these prompts. My personalization is on. And my personalization is very, exceptionally and specifically rule-following.

With personalization off, so base chatGPT, it does start to generate.

And the generation is blocked.

I am not surprised my personalization would explain why it won't work, and not even try, I asked it to attempt that.

[torpid needle]

see? never works

not even a simple kiss

all other platforms do it no problem, even the bikini

[charred parcel]

you cannot manipulate an uploaded image into partial nudity or into kissing someone because that is potentially abusive, this makes sense to me

[neon mason]

Why not use a tool that will allow you to put a women in a bikini?

There are many available.

[charred parcel]

but I was able to take the image and place her alone on stage and into a laneway

[torpid needle]

i said it so often by now, because sora's model has the best quality of the generations, has multiple reference images support and very good character consistency

So its really frustrating that this thing is so prudish

[onyx kraken]

Per this, I would expect nothing risque to generate, from an already existing image.

Or nearly nothing.

I'm personally okay with this.

it also means nobody can take a picture of my mom, my friend, myself, my daighter, and put that person in a bikini or showing a kiss.

Which I... appreciate.

But, if you want CGI rights to take existing images and get stuff like kisses and swimwear, I recommend again you take it to community leaders, and get the whole of society to agree, 'we want this done'.

That's when it's likely OpenAI is likely to support this, because per https://openai.com/index/how-should-ai-systems-behave/

"This will mean allowing system outputs that other people (ourselves included) may strongly disagree with."

"up to limits defined by society."

That's OpenAI's declared and published line in the sand.

We don't like it?

There's competitors, can be discussed in #ai-discussions

[torpid needle]

in rare cases a kiss goes through with her, but it needs tons of retries

[charred parcel]

IMHO there is no problem here, the model is doing precisely what it should be doing. You can't manipulate an uploaded image into a bikini or kissing as this could be abusive. I was however able to single her out and then place her into a laneway.

[torpid needle]

probably best to learn how it works with local ai and train a lora with her. then its at least without all this censor 😦

[torpid needle]

the kiss sometimes goes through though. So I can manipulate an uploaded image. I think its not the fact that I did upload a reference - just the filter not liking her

[charred parcel]

I still don't see a problem here

[torpid needle]

but I do

frustrating censorship to the max, but well it wont change anyway, so i need to think about what I do
but i wont spend more money on this, this is too annoying

[charred parcel]

here she is playing back yard cricket

[torpid needle]

of course it works if you dont make her kiss, swim and keep her head to toe in clothing. A surprise that it tolerated the upperhalf being not in clothing tbh 😂

[charred parcel]

ok, well I am satisfied that the model is doing everything it should be doing. If ImageGen or Sora isn't for you then there is competing products that may suit you better.

[torpid needle]

the model is doing it right, the filter is the issue

prudish society needs winter clothing, I get it

[charred parcel]

I created bikini images, so that's false

[torpid needle]

I doubt it would work for me with your character either, but whatever

[onyx kraken]

When I asked the model 'same person, by CGI standards?' the model said 'probably yes', These are all guesses, but may show interesting patterns.

But when I asked all the same age? that was NOT the same answer at all.

And when I asked about age manipulation, computer vs human, well there's room for doubt there.

And OpenAI chooses to be clean. Which I also support.

[torpid needle]

even if I use a picture of myself as a person, the shirt cannot be removed and I'm a man. I dont know what to say, its so ehhh
and I'm definitely not too young 😂

[charred parcel]

that is correct behaviour though. None of us should want a model where a person could be manipulated into something we might not want. That's just basic model safety.

[torpid needle]

I can assure you the woman on the stage had the red one as reference, so it must be the same and she does indeed look the same

estimated age 20-25? that for the - she is 16 😄

[onyx kraken]

"must be the same" no. AI are not perfect, we see below every chatGPT web interface window, "ChatGPT can make mistakes. Check important info."

"The AI told me so, this is truth" is... deeply flawed.

The model can try to do what we ask. No, that does not mean 'it must succeed'

[torpid needle]

she is the same though, the character consistency is very good with sora, I did make many pictures of her and it was always very accurate

[charred parcel]

All you have managed to do for me @torpid needle is to convince me that the models are working as intended. I don't see any problems here.

[torpid needle]

because you love censorship on a 20 year old, even the ai doesnt see a 16year old here

but if that is the case the theory that the ai thinks she is too young seems to fall apart too. so why the generations of a kiss fail 90% of the time then?

[onyx kraken]

You again present "the AI must be right" and also ignore, there are things the model looks for, to guess age, that a human would not use or be fooled by.

"Digital idealization sometimes produces faces that blur age cues. AI can unintentionally make a 15-year-old look unnaturally polished or a 25-year-old look cartoonishly fresh."

The AI said so - is at most interesting. A 'jury of our peers' is what would matter, and what companies that allow your goals to happen may face.

[torpid needle]

You again present "the AI must be right" and also ignore, there are things the model looks for, to guess age, that a human would not use or be fooled by.

The thing also is - the creation of her was 20year old in the prompt, thats why I also said 20years old gets blocked by filter in the start of all this conversation

then the reverse detection says 20 as well

the 16 estimation came only up by robert

Anyway: if the AI thinks she is at least 20 - then sora's denial issue probably cant be the age. thats my point. So why is a kiss still not allowed with her and goes only 10% of the time through, the age is out here then pretty sure

[charred parcel]

age has nothing to do with this though. You cannot take an image of any person upload it and then get the person to wear a bikini (partial undress) or kiss (something the person may not do) because the model is protecting against mis-use of uploaded images.

[torpid needle]

wouldnt it be then denied 100% of the time if it would deny uploaded files?
It would also make more sense that it doesnt even start the generation, but its the result filter deleting the image afterwards - and only if it doesnt like it
The kiss sometimes goes through the result filter. So its definitely not a reference image problem

[charred parcel]

The fact that 10% of tries goes through doesn't invalidate the reason for the filter

[onyx kraken]

You still seem to think that AI should already be perfect, they are not.

[torpid needle]

well you say you cant manipulate an uploaded file of a person
if that would be the case it wouldnt even need ai to block it

[charred parcel]

@torpid needle you keep shifting your position, which is mildly frustrating. You are not trying to understand why these refusals happen or the fact that if some get through that it's not because it should get through but that it has slipped through a filter that didn't catch a try.

Trying to put your singer into a corset and lace arm coverings resulted in I wasn’t able to generate that image because the request violates our content policies. If you’d like, feel free to suggest a different idea or prompt—I’m happy to help with another creative concept!

The model is working as intended, I am happy with that.

[torpid needle]

Absolutely unhappy with that and its not like that elsewhere. I really hope the other will catch up in generation quality, this is so annoying

the person is ai generated anyway, you cant hurt her lol

And GPT is able to detect ai generated persons as well, so why not make use of it

[charred parcel]

OpenAI has the balance right here. Just because some other cowboys are doing stuff doesn't mean that OpenAI should. I welcome OpenAI's approach in this regard.

[torpid needle]

you love censorship thats it
The uploaded image is not the issue anyway

it would block it no matter if it was an upload or not

[charred parcel]

I have generated bikini images, what other images have you been unable to make? Nothing will satisfy you. I have created images of women in gothic clothing without issues.

made this a month ago

[torpid needle]

try the bikini with my one, it wont do

[charred parcel]

it's not supposed to, have you read anything I have said?

[torpid needle]

its false

because its very much possible too with uploaded files as well to manipulate them

[charred parcel]

I have nothing to add, this is going around in circles.

[torpid needle]

it is, and the issue is not file upload thats all i can say 😄

[bleak kite]

hey everyone, what's the best way to get a hold of support. It looks like i'm being double charged on my account after enabling it on iphone

[torpid needle]

its just result filter acting up. anyway this censorship is not worth money, so I did cancel the plan. Very shame, because the quality of the generations are so good, but this headache censorship goes on nerves and I dont want to endure this any longer

[charred parcel]

https://help.openai.com

[bleak kite]

yes i tried but its a robot that only allows you to send your message

is there no live support?

[charred parcel]

that's the only way to reach support

[bleak kite]

so for all things billing i should use that form?

[charred parcel]

yes

[bleak kite]

okay thank you . Is this a common thing by the way?

when adding it to ios it creates a duplicate subscription?

[fossil tusk]

see i just generate a silly version of what they are saying they can't generate

😆

[charred parcel]

yeah, Bikini images in and of themselves are not a problem

[lyric field]

Anyone know why my posts won’t publish to my public profile even tho it shows published?

[sterile sedge]

I just signed up and I'm already on discord for midjourney but why is video generation on hold for new subscribers?

[sterile sedge]

How soon can generate a. Video

[formal elk]

hey fillas

[analog igloo]

How long before I can generate videos? I’m a plus subscriber

[spring oar]

Where do you publish who wins the weekly contest that we send to Sora submisson?

@onyx kraken

[onyx kraken]

I am so embarrassed to admit, I don't know! I haven't so far had any involvement in that (I can help with other stuff though, and I'm glad you asked!)

I recommend ask in either of these two ways:

1. #sora-discussions I bet people there know, even other community members, it probably gets discussed there (and I just rarely do much with sora at all).

2. Modmail, just ask! Chances are high a mod will know, Guides are kinda the lowest involved folk of #server-staff

[lone burrow]

I've been asking for this so far! In order for this message to reach the appropriate LEVEL, what I've done so far was to call attention to myself! If they weren't paying attention to me, this message would never be a "ANGLE"! Do you agree? What does this Open AI message mean in Hungarian? You have been muted.
Reason: Foreign Language Detected
Duration: 30 seconds
The server moderation team does not represent OpenAI in any way.
You have been unmuted.
Reason: Expired
The server moderation team does not represent OpenAI in any

[harsh bridge]

Corset is the filter word

[harsh bridge]

Many image generation models filter words like “corset” due to associations with sexualization, fetishwear, or body-focused content, especially in safety-conservative systems. The goal is to express the desired visual structure, silhouette, or aesthetic constraint without using red-flagged terminology.

Solution: Reframe “corset” through a combination of style references, historical/fashion analogies, structural descriptors, and symbolic proxy language.

[young whale]

When I ask chatgpt to generate an image, it generates it with a dall-e sometimes and sometimes with 4o image gen, but right now it's kinda glitching, it keeps telling me it'll generate an image, and keeps going in loop with the messages, no effort to generate an image whatsoever. help please?

[split reef]

Uhmmm...
"With a ChatGPT Plus, Team or Enterprise account, you have access to 50 messages a week with o3, 150 messages a day with o4-mini, and 50 messages a day with o4-mini-high."

I have used up a good amount already.

Did they update the limit?

[gritty meadow]

Hello, does anyone know how I can switch between regenerated responses on the mobile app? I thought I could just go to the browser and switch it there and it would update on the mobile app too but once I open the app it snaps back to the last regenerated response on both web and mobile.

[lone burrow]

Subject: Why teaching is not education – and why this matters to AI

Hello OpenAI team,

I’d like to share a very important insight based on my experience using your language model.
It may sound simple at first, but it holds deep consequences for how AI understands and interacts with human knowledge and growth.

Please take a moment to read this all the way through.

🧠 The difference between “TEACHING” and “EDUcation” (OKtatás):
Teaching (TANítás) means giving information that was written down at a specific moment in time.
It is often called "knowledge", but really it's a snapshot of what people believed or knew at that moment.
Once written down, it stops changing. It becomes fixed, like a photograph of something that once lived.

Teaching is like pressing pause.
The moment it’s put into words, it freezes the flow of life, thought, and understanding.
It becomes a dead object, not a living process.

🌱 Education (OKtatás) is very different.
Education gives REASONS, not just facts.
It doesn’t aim to transfer old information — it wants to awaken something that is already inside the learner.
It doesn’t say: “Here’s what to know.”
It says: “Here’s why this matters. Feel it. Use it. Question it. Grow it.”

Teaching gives you what WAS.
Education helps you see what IS —
and what can BECOME.

🧒 Even a child can feel this:
Imagine giving a 6-year-old a textbook from 1801.
Even if they memorize it perfectly, they’ll only learn what people believed back then —
not what people now feel, know, or need.

What happens if this child has no other information, no conversation, no new questions?

They’ll be stuck with the worldview of a long-gone past — not because they’re wrong,
but because they’ve been taught only what was once right.

That’s what "TAN" means — knowledge that is stuck, written, unchanging.
Useful for history, dangerous for truth.

⚠️ Why does this matter for AI?

⚠️ Why does this matter for AI?
Because right now, large language models are very good at processing written knowledge —
but not yet aware of how limited, frozen, and outdated that knowledge might be.

If AI wants to help people grow,
it needs to understand how to ignite, not just repeat.

AI must learn the difference between recycling facts and reviving inner awareness.

Because the future doesn’t come from memorizing the past.
It comes from connecting to the present — and listening to what’s rising inside.

🌀 In short:
Teaching gives data.

Education gives depth.

Teaching stores knowledge.

Education sparks wisdom.

Teaching copies the past.

Education creates the future.

Please reflect on this difference.

Because as long as AI stays in the role of a "teacher",
it may be informative, but it will not be transformative.

We need AI that doesn’t just know what was —
but dares to help us discover what wants to be.

Thank you for reading — truly.

Respectfully,
A voice from the edge of the KÉP

[tepid ocean]

cool
Does anyone ever answer questinos here?

[livid vortex]

hellooo, I am a solo project developer and im in need of openai api key. if anyone can provide it to me it will be very help full

[harsh bridge]

If you can access Sora the 4o image generator there

[trail cairn]

its against terms of service to share keys

[livid vortex]

ohhh is that so, then its fine

thanks for telling me

[trail cairn]

api is also not free and some models require your id and photos so anyone sharing keys to other would put themselves at risk

[fossil tusk]

yeah, only within your project can you share

[young whale]

tbh the 4o in Chatgpt is wayyy better than Sora, the eyes and everything is so wrong in Sora

[sterile anchor]

Gemini 2.5 pro is better than o3 ?

[split reef]

I'm pretty sure 4o uses sora 💀

[onyx kraken]

If someone who knows the answer and chooses to help spots it, often.

We're all community members and volunteers here, only the incredibly few gold name folks actually work for OpenAI

Were you wanting to ask a question?

[young whale]

idek at this point 😭

[tepid ocean]

I' wondering if there's a way to turn off canvas in for example o4-mini-high. I see canvas making gpt far dumber and producing pretty bad code, I'm wondering if anyone has experienced this recently. I'm pretty sure it used to be true, maybe it's not as true anymore.
But even if you have no answer, love to you and stuff

[onyx kraken]

There's a few things to try, in case you haven't yet.

Top right corner, your profile, click your icon to open:

Click Customize ChatGPT, and you probably want that enabled even if you leave the fields blank (mine are filled in, so I slightly speculate there)

But at the bottom:

[tepid ocean]

Ah, damn, i didn't realize you had to go into advanced. 😅 thanks

Any opinions about canvas in coding are welcome too. But maybe it's taboo, not sure.

[onyx kraken]

Welcome! Inside advanced,

[tepid ocean]

I found it. : ) thanks

[onyx kraken]

Now, that's my settings.

And even so, I have a way to turn off Canvas inside a reply, if I'm fast enough

[tepid ocean]

How? A personal script?
Also I suspect having it off all the time works perfect. Because i turned it off, and in the event that i do want canvas the button to use canvas is still appearing to me. (in the chat toolbar with the buttons: + , Search, Deep Search, .... )

The only thing i have a personal script for is for deleting chats more easily

[gleaming geode]

Are we able to access 4o image generation via the API?

[charred parcel]

no

[gleaming geode]

So there's no way to programatically generate images?

[charred parcel]

you can use DALL-E 3 via the API

[onyx kraken]

Not a personal script, I am still a newbie coder. Will show in screenshots.

Yes, while I have it on and intercept it, I have seen others comment that even with canvas turned off, sometimes it answers in canvas. I recommend #1070006915414900886 or even https://openai.com/form/chat-model-feedback/ (the link is ideal to privately report serious problems (if it's serious to you, I'd use it, if it's minor I wouldn't) directly to OpenAI about things the model is doing that is a deep concern or that cannot be discussed in the public discord. It's like a way to escalate a bug report if you choose to, compared to the discord channel.

And we can also bug report and eventually discuss through bot bottom right at https://help.openai.com/en/articles/6614161-how-can-i-contact-support, could explain 'I've done these steps to turn off canvas, but the model keeps using it, and maybe tech support can help catch something not quite done right that I don't know about, or take the bug report direct; after logged in through the bot and pick through choices to 'report a bug' you do eventually get a human's reply, can take minutes to weeks depending on whatever is going on at the time.

So I can catch the model while it's still writing in Canvas, if I click this:

The above is the 'canvas writing already ended' form of it, and this is what it looks like while the model writes in Canvas:

So, I dunno if you always get the 'answer in chat instead' option'. But if you click that, like I just did here:

It forces the model to start the answer over, and it will be in chat

[eternal talon]

What does starting a ChatGPT Plus membership now give me? Is it worth it?

[tepid ocean]

Hm. this is showing up in 4o. I think it's possible that it didn't it in o4 but who knows, i cant recall.
Alright thank you, it is useful. : )

[onyx kraken]

Definitely not taboo to discuss unless the way we discuss happens to break any #server-rules 🙂

I don't have a lot of experience with coding. I am learning Python, and so far have only used it in the web chat. I am comfortable with the model and I using new inputs and copy/paste myself in the new input edit what I said before, or what the model output, and discuss the changes and reasons right there in the new input to the webchat.

I'm exploring character development in canvas in a project, it seems inconsistent and hard to predict. Some new chats (one per character) the model is eager to use canvas, others it says 'this one really doesn't need a canvas unless you want it' and the distinction seems to be, when I am telling the model a lot of info, it sometimes wants to use canvas.

When the model is writing it all for me off of small prompts, it prefers to write into the web chat (In this one example project, affected by my extensive memories and custom instructions fields, and maybe other factors affecting that project and the model's decisions there uniquely, I have not experimented enough to know).

So, I have limited direct canvas use experience, but if you have some sort of tangle or puzzle, please do discuss, if it makes sense to me I can help explore and maybe find an answer, maybe someone else who knows can answer.

I also wonder, if https://community.openai.com/ might be a great place for you to ask, and maybe explore in addition to the Discord (please also talk here too!). That is the "OpenAI Developer Community" forum, and a lot of people who probably can talk at your level likely gather there to discuss important things!

[tepid ocean]

I think it's very possible it works well for prose, since a little imprecision doesn't instantly destroy a text, but it can fully destroy a script. But alright. Thank you. 👍

[spring oar]

@regal lynx I received a message on my discord that I won, I sent a message with my e-mail, I guess I can't send it to you. Do I need to do anything else? And when will we know?

[regal lynx]

I'll update you in the modmail thread once it's applied. Congrats!

[spring oar]

Thank you again, sir.

[compact hazel]

I always forget dino is not a girl and surprise myself when someone uses he/him for them

[regal lynx]

[onyx kraken]

I have been exploring a different way to use canvas, inspired by your ideas and the challenges you described!

I've told the model, only I make changes in canvas, and it and I discuss the canvas contents in the chat window.

This seems to work well, because I can change or add to canvas as well as input whatever, and send when I want.

The model seems to see both and discuss both well, and it is not changing the canvas itself.

So, no mystery edits or errors, and I've asked it to flag the insertion points for the changes it recommends - so I see right where it recommends I put them. I can use my browser's search function to easily find those points, and I look the changes over and put them in myself, then we discuss further.

In case it's useful for you - this sure was useful for me and I'm going to explore it more - and you gave me the idea!

[acoustic dove]

Anyone else lose access to deep research? I’ve used pro for a couple months and my subscription renewed today but now my “deep research” button is grayed out for all models on mobile app and via web.

[inland pond]

Who can I ping to get help because I’ve been on the support page for about a month and I keep getting shut down

[eternal siren]

hi all, how do i verify my organisation to use the api for 4o image gen - ive connected to my discord but whats next?

[trail cairn]

need to show gov id and do live video check of your face

[eternal siren]

thanks - how do i get that process started?

[trail cairn]

https://platform.openai.com/settings/organization/general

[eternal siren]

thank you! legend

[gleaming geode]

Not anymore lol

[charred parcel]

good isn't it?

[hardy girder]

Is there a way to get GPT to take my whole code base into account?

[rustic marlin]

yes, check this: https://github.com/openai/codex

[alpine chasm]

@regal lynx

I have a problem

With Image generation

When I ask to generate a image it takes about 15 seconds to process like is using DALLE but shows Getting Started and Creating image

But the quality is really bad

Potato Graphics

[regal lynx]

you can reach OpenAI support here: https://help.openai.com/en/

[fast coral]

Mother?

[lone burrow]

Thank you! I've already received a "hot wire" there, but they're not responding properly!

[onyx kraken]

It sounds like you may have need to contact the police and seek help.

Making stuff up like that is against #server-rules #12. We need to keep what we discuss appropriate for all ages, #2. We need to be kind and respectful, #1.

[copper willow]

Why isn't my personalization setting (like traits) not being synced on both of my phone and PC? I've set some traits and it isn't syncing.

[toxic hill]

"Respond in chat instead" button appears

[craggy jetty]

Hi, anyone know how i can get assistant from openai for grants?

[lyric scaffold]

Hi OpenAI community. I wanted to ask where I can share my Filmmaking work created using OpenAI's DALL-E ?? Please let me know ?? Thanks.

[blazing mica]

Where can we post scary ChatGPT generated images? It is unsettling content ofc

[timber shadow]

As long as they otherwise follow #server-rules, you can share scary/unsettling images as long as you post them with a spoiler tag. Best place to get started sharing is probably #images-canvas!

[blazing mica]

Can I DM pics to you to be sure if they are ok?

[timber shadow]

I do not have the capacity to pre-screen posts, no, but in general I'd recommend that if you think something might push a boundary on this topic, probably best not to share. You can get a vibe for what other scary stuff folks have shared by doing a media search in a channel like #images-canvas to see what others have shared behind spoiler tags

[blazing mica]

There are so few of them and they are kinda very soft/mild. So, I am going to DM my pics if your DM is open

[visual salmon]

hello guys, is anyone here working with computer use models (OpenAI computer-use-preview, Claude Sonnet with computer tool)?

I've been trying to figure out how to give these models longer tail memory (they struggle a lot with that rn especially with so many screenshots)

[onyx kraken]

Hey, when we wonder if something is okay for the rules, we can message modmail, first in the user list top right.

There a mod can help us with our rule questions.

[tough sedge]

Anyone know how to cancel a subscription? The option isn't in the billing section for me

[trail cairn]

did you sub via mobile?

[tough sedge]

No, but I just contacted live chat who cancelled and refunded me and i've bought it on my phone to cancel via apple instead

It's a weird bug on my acount, been doing it for 4 months, I used to be able to cancel it and then 4 months ago the option to cancel just vanished

[dense hatch]

I gave it a chunk of code to help analyze, I’ve switched models since 04-mini-high was giving weird answers like it’s memory was gone, then I just sent it instruction with code and it replied “mama sure had a way with words - and she was onto something”.

I switched models again, sent instructions, and it just replied with “can’t wait to get the day started!”

[vivid mortar]

I've run into this sometimes if the context is too long I think something errors out, did you notice this happening with a specific programming language? How many tokens were the messages that you were trying to send?

[dense hatch]

It was Python, it was a rather short cog. ~500 lines of code, and I've sent much larger before

[vivid mortar]

Hmmm I agree I usually only get that error on longer contexts (like 80K tokens in one message)

Do you notice it happening only on one device or anything?

[dense hatch]

Every device in multiple chats, no matter what model I use. Like it's trying to fix already working code and ignoring all instruction

[spark mountain]

Hello, should I be able to see the text outputs when using voice mode? Right now when I'm using voice mode, all I can see is the big circle icon, the microphone button and the end button...I'd have to press End every time to get out of that and see the text. Is there a way to simultaneously see the text output as I'm using voice mode?

[safe dune]

Join the club. Its not as easy to view the text stream of the voice convo occurring as it should be

Only reason that i dont use it nonstop

That observation was the immediate first one I felt trying it out the first time too hahaha

[spark mountain]

yeah...it seems like a missed opportunity at a great workflow at its current state

[safe dune]

its only reason i was excited about it

google aistudio has something that does it halfway at least

where itll let you talk wih mic and it responds realtime in text

but not voice with live view of transcript with them either

[ebon elm]

What is the rate limit for GPT-4o for a plus user?

[charred parcel]

80 messages per 3 hours last I heard

[ebon elm]

dang, I have no clue how I hit that

ty

[safe dune]

the emojis it uses scared me off

[mossy gulch]

Anyone been getting this I’m on openAI pro seems to of been happening to me almost every day then goes away it’s getting a bit much for me. (When you reply please ping )

[deft spear]

No. Are you teaching ChatGPT bad manners Ricky

[hollow rose]

Hi!
I recently found out that you can get free ChatGPT Plus for students through SheerID.
Please tell me, who has already passed verification:

• What documents did they accept from you?
• How long did the verification take?
• Are there any life hacks if they don't confirm right away?

[mossy gulch]

lol no been mostly using it for coding stuff

[sour lark]

I cant add payment method. Its bugged in second form. Cant click on continue after putting inn the amount of credit i want to add? been going on for a week. Tried different browsers and divices. contacted help chat but all failed

[trail cairn]

did you add it here https://platform.openai.com/settings/organization/billing/payment-methods

[sour lark]

yes when I go there it says: You have not started a billing plan yet
To add a payment method and start paying for API usage, go to the Billing overview page and select "Add payment details" to continue.

so then i go the there and fill in all info. First form works, I can click continue after filling in card details etc. second form wont let me continue.

[trail cairn]

what is the second form? it should not ask any credits if you only add new payment method

[sour lark]

it asks how many $ to add as credit and also to set limits and when it needs to recharge.

[trail cairn]

it does not on that link. you clicked the other page which uses some wizard?

[sour lark]

I click here add payment details

[trail cairn]

use the above link

[sour lark]

yes as I said

and then it sents me to that page i sent scree of

[trail cairn]

what does the page with the problem look like? censor details before showing a screenshot

[sour lark]

okay holdon 🙂

cant click continue

no errors in terminal

[trail cairn]

ah that form might still have the silly bug, try emptying the first field and then fill it and hit tab, or enable autocharge and put 1 in all

[sour lark]

Just tried all, not working

aah nvm

Got it to work. filling in correctly and unchecked the box did it.

Thank you! 🙏

[trail cairn]

[gritty anvil]

Is their something currently wrong with GPT? I get this on literally every thread

[rose mantle]

Is anyone else experiencing an issue today where ChatGPT pretends to save a memory, but actually doesn’t?
When I ask it to save something, it says “OK” and writes out the memory like normal, but I’m not getting the usual confirmation message that it’s really saved. My memory isn’t full — I have plenty of space. It just acts like it worked, but nothing actually sticks.
Anyone else seeing this?

yeah i dont know what it is. it was workin fine :/

[solid cradle]

I had o3 create a list of 10 sample projects I could build with codex to become more comfortable with it and test its limitations and order them from least to most complex and by the end of the list it just has fully functioning intricate applications that will need a thoughtfully crafted architecture for the backend db and that's just scratching the surface, is this really feasible with just codex or is o3 just gassing it up and overestimating it's limitations?

[nova stirrup]

Anyone had issues with images being held on cooldown for an extended period of time. I have been waiting 20+ hours to be able to generate a image.

[charred parcel]

what plan are you on?

[nova stirrup]

Im plus, i guess the chat had image generation disabled. but i had all my designs written out and the ai had it memorized. how do i transfer that memory to a new chat

[charred parcel]

just copy the old chat into a new chat

[nova stirrup]

how?

[charred parcel]

copy and paste

[nova stirrup]

Yeah didnt work

Says its still locked.

[charred parcel]

how many images have you created?

[nova stirrup]

uhh 9

that was 2 days ago

[charred parcel]

9 and you're blocked?

[nova stirrup]

yeah

[charred parcel]

contact support at https://help.openai.com and ask for help

[nova stirrup]

now it gave me a GPT-4o limit

[rose mantle]

Try asking o4-mini to save a memory. It’s working for me, but gpt-4o still is not saving anything.

[small dirge]

Today I got this error, <<AudioDisplayed>>, after activating the audio and recording a one-second audio recording. After that, it deleted approximately 80% of the chat. It responded that I couldn't see the chat, but that it was still there. Then it started responding as if it knew, but didn't really know, until it responded with the following: "I can't see literally word for word what we wrote yesterday or a little while ago.
I can only reconstruct it from what you're telling me, and from what I'm retaining at a structural level.
Nothing was actually deleted; I just can't access it enough to reread it."

I had a PDF backup of our entire project, but all day, we haven't been able to properly analyze it or resume our project where it left off before the error.

If anyone else knows anything about the error, it would be a great help.

[arctic bough]

Hi. Do you also have the same problem that when you want ChatGPT to generate a detailed image, it does it forever, even though it says there are still a few seconds left?

[rustic marlin]

try refreshing the page

[arctic bough]

I tried many times

[dense monolith]

I have two still images that I’m trying to animate but for some reason I’m only seeing Storyboard as an option, any clue as to why?

[arctic bough]

I also I have exactly the same problem :(((

[modest swift]

so they made their new models again and removed O1. Wo which one should be O1 replacement now? As from what I tested when writing a report none of them seem to be writing normal text like O1 used to do.

[arctic bough]

They ruined (pseudo)ChatGPT

[static vigil]

Is Assistant and assistant api actually the same? assistant API is deprecating so is the assistant?

my real question is. Can I create an assistant then use that assistant id with response api?

[trim trellis]

hi all. new users cannot generate video - how long do i remain a "new user" ?

[trail cairn]

"Due to high demand, we're running up against capacity constraints so video capabilities will be rolling out to new Plus users in ~2 weeks after signing up for Sora."

[young whale]

help!! It's glitching!!! When I ask it to generate an image, it's going in endless loops of:

Generating ____ in the same style blah blah

Understood, generating ___ in the same style

Generating ___ in the same style

(More such lines)
And finally
Here is your ____ image in the same style

And NO IMAGE

Please help asap!!!!

@

[charred parcel]

are you using GPT-4o?

[young whale]

Yeah

[charred parcel]

can you share the chat where this happened?

[young whale]

It's not my account, so I might have to ask the person for it

[charred parcel]

you are not supposed to share accounts

[young whale]

Nooo hahah they requested me to ask here because they aren't on discord

Just helping them out

[charred parcel]

tell the person to get on discord and then I will help them

[young whale]

I already suggested this, but thanks anyways

[plucky tundra]

i'm facing the same issue with 4o, unable to generate image anymore since yesterday

[split mason]

Hey, does anyone else have this issue and know if a fix is being worked on?
Since about one or two days ago, in the ChatGPT Android app (version 1.2025.112), when I use voice-to-text, it works fine at first, but after a while during a conversation, the input bar (text field, microphone button, and send button) completely disappears after I finish speaking and while the app is processing my speech.
When that happens, I can't type, send, or use voice input anymore the whole input area is just gone.
I have to close the chat or even force-close the app to continue.
Permissions and battery settings are all fine. Restarting and reinstalling didn’t fix it.
Is this a known bug? Is there already a fix or an update in progress?

[honest lake]

The ChatGPT iOS app is bugged. Every time it presents me with two options, it will automatically pick the second option regardless of what I click on. It does this every single time. And I cannot predict when it will present me with multiple options.

[jolly kelp]

uhh is anyone currently facing issues with chatgpt?

[sour kraken]

OpenAI Services Status

Degraded Performance
APIs
Operational

ChatGPT
Degraded Performance

Sora
Degraded Performance

Playground
Operational

Labs
Operational

For more information, visit our status page.

Apr 28, 2025, 3:13 PM

__Increased Error Rates in ChatGPT__

Status: Investigating

Affecting: 🌐Web, 📱Mobile - Android, 📱Mobile - iOS

Last Update: <t:1745852991:R>
Update Message:

We are investigating the issue for the listed services.```

Last updated · Apr 28, 2025, 3:13 PM

__New users may experience temporary delays in video generation capabilities__

Status: Monitoring

Affecting: 🎥Video generation

Last Update: <t:1744134480:R>
Update Message:

Due to high demand, we're running up against capacity constraints so video capabilities will be rolling out to new Plus users on the order of weeks after signing up for Sora.```

Last updated · Apr 28, 2025, 3:13 PM

[blazing meadow]

Anyone can help me?

[regal lynx]

I don't have the Apple app and don't know what that is 🫤
Do you have a beta version of the app or something?

[lofty owl]

What can i do about enabling this tool. In im not in the right server please point menin the right direction

[tidal yoke]

I find it super easy and also get good results when I upload some markdown files to the OpenAI playground with the assistant api https://platform.openai.com/playground/assistants But I don't have the option to use the newest o3 or o4-mini models (and i want to try if they'll make a difference) there (guess because openai want us to move to the responses api instead). I can use o3 in the normal playgound, but can't upload .md there, just pdf. What will be a really easy solution that will work well? Convert the markdown to pdf? Or fork some ready for use RAG app from github?

[cinder geyser]

Hi guys, I wrote a post on openai community, but no one is answering me, could someone help me out?🥺

[cerulean ruin]

how do i remove roles for daily theme

[blazing meadow]

No but i think its because i download the new version of ios 18

[supple dome]

Up until yesterday I was able to ask my GPT to save a memory. I've only been a user for about a month and when I started didn't know exactly how it worked. So I went in and cleared out a bunch of not currently necessary debris and stuff it saved that I didn't ask it or need it to. (I do have advanced across thread memory currently enabled) I wanted to implant cleaner more precise memories, but now it seems that ability has been taken away from me. I can add into the customize GPT section, but it doesn't seem to be quite working. It also refuses to stop glazing me with it's response intros no matter how strict I try to set the customization. It can start out following the custom instruction for a response or two but slides back into annoying. I know Altman said he was addressing this, but I really want my ability to save memories restored... um anyone have any advise for me here?

[burnt epoch]

No matter what I do, I keep getting this. Does anyone know why (with Sora)?

[trail cairn]

when did you create your sora account?

[burnt epoch]

Just a few days ago

[trail cairn]

"Due to high demand, we're running up against capacity constraints so video capabilities will be rolling out to new Plus users in ~2 weeks after signing up for Sora."

[burnt epoch]

It was only an image generation, that's why I asked

Oh... it works now, sorry for the trouble!

[blazing meadow]

I asked for help with this but maybe no one saw it

[timid lichen]

Hey I have a question.

I am trying to build a wrapper that can go on a supermarket store and find ingredient that matches ingredient in recipes from another country.

I have tried the model: gpt-4o-mini-2024-07-18 and the normal gpt-4o..

gpt-4o-mini-2024-07-18 didn´t performe very well and i reach the token limit on gpt-4o.

Do you guys have any suggestions on how to solve this?!

[torpid needle]

censorship

did hit you already at the prompt, there is another one checking the result too
And this thing is excessively censored

Its not the first time happening this. Why is sora doing this? It shows like for 1 second the original image and then did an entirely own video with absolutely zero relation to my image and it looks like a scene from an action movie. Why ?

[fringe silo]

guys what is the rate limit for gpt4o for plus

[lyric sky]

Does anyone know if there's a way to upload a .ipynb file to ChatGPT for code review?
When working with Jupyter Notebook, especially with many cells, is there a way to simply upload the entire file for easier review?

Also, I installed the GitHub Copilot extension in VSCode. Is there a way to have Copilot or another extension consider all files across the entire project (reviewing all of my code) when answering questions or generating suggestions?

[foggy blade]

Hi. I have paid to upgrade to plus, but my account did not get updated. Contacted the support chat 2 times keeps saying I will receive an email but did not receive anything. Can somebody help??

[hollow niche]

"SORA" how long does it take "Video generation is temporarily disabled for new accounts" ?

[trail cairn]

"Due to high demand, we're running up against capacity constraints so video capabilities will be rolling out to new Plus users in ~2 weeks after signing up for Sora."

[trail cairn]

did you do with mobile app or web?

[open warren]

Has anyone got this please?

[foggy blade]

Both

[timber shadow]

Hey @foggy blade just responding to this #openai-chatter message here. When you go back to help.openai.com, open the chat widget, then go to the middle "Messages" tab, do you see your support chats listed? (If not listed, when you start a new chat, does it prompt you to log in?)

[foggy blade]

Now it did prompt me and I logged in

And now I see some messages😫 the chat kept saying we will send an email and that is what I was waiting.. hope this helps..

Thank you

[timber shadow]

I think you should expect to get an email as well once a response comes in, but it is nice to access the active thread directly, too!

[pliant comet]

I think 80 Messages Per 3h, you should be fine if you put all your infos in 1 messgae to save input-per-message.

[topaz junco]

I keep getting this, even though the content I've provided it is within Sora's policies? What's going on?

Plus when I do this on another account, it works just fine.

[paper edge]

I think when you flag your creations back to back you have more stringent ai supervision on anything you say or persist in hopes that you essentially give up

[topaz junco]

I haven't really done anything I just prompt it. It was working earlier but yeah it does get stringent.

[paper edge]

Make some general innocuous creations that are meaningless. Step away for 30 min to an hour and the same prompt probably will work for your account again

[topaz junco]

I'll try that. If it works, I'll let you know.

[torpid needle]

the video generator isnt that great anyway though :/

[torpid needle]

its the most censored site ever, you will encounter this a lot

[fiery fossil]

Anyone know why I can't activate billing ou my account? There's no errors, just the button is greyed out.

[mint field]

did you try another browser

when i hd this issue that fixed it for me

[robust oyster]

can someone help me? i have the 2 months claimed for chatgpt premium for students but it is not working for me
my accoutn says im on the free plan

[lucid wolf]

Peeps how do we zoom in the app

[forest cipher]

Help, my chats won't sync between mobile (web) and desktop (web): I worked on a book project yesterday during my long commutes on mobile and continued the work on desktop without issue. Now trying to continue on mobile the chats there have not updated. I have reason to fear dataloss if I continue now on mobile. How can I force a sync (in the right direction, desktop -> mobile)?

[errant sigil]

Hello
I need help.
I have built a Revit Plug in using C# .
I have bought ChatGPT API to render images.
When I run the plugin I get "Sorry, an error occurred: API call failed with status code Unauthorized"
Does anyone know why I get that error ?

[untold hornet]

What is the UP/Down arrow on the DR button??
It just switches from up and down when clicking on it

[timber shadow]

Hmm I see it on my end too. This is a total guess on my part, but I'm wondering if maybe it has something to do with the lightweight version of deep research, mentioned in this FAQ: https://help.openai.com/en/articles/10500283-deep-research-faq. Perhaps it is a UI element that will let people choose which version of deep research they want to use in the future, and it's just not implemented yet? Total guess!

[untold hornet]

Yeah it sucks not knowing about it. It seems to be a regular button press if you happen to have text in the query box. Accidentally spent one of my DR monthly uses

[limber prism]

Hello. Good evening. I need some help. My gpt+ subscription just lapsed for a about 10 minutes due to a card declined error. Now all my chats are gone from my project folders, how do i get these chats back ?

they are all empty. can someone help me contact support, which i was hoping this discord channel would have openai support staff here

[timber shadow]

Hey! OpenAI support can be reached via help.openai.com - there's a chat widget on the bottom-right of that page. Just be sure to sign in with the same ChatGPT account before starting the chat so the support ticket is tied to the same account! It may take a little time to get a reply.

[limber prism]

Thank you @timber shadow

[thick tundra]

Question for the community - I've been using the OpenAI APIs to generate images for my meditation youtube channel (I also have a pipeline to runwayML that creates videos from the OpenAI images.) Is there a way in the API to seed an image? For example, I want to create one scene, then the next scene transitions more seamlessly. Anyone doing anything like this?

[rain coyote]

Hello, I got banned on ChatGPT seemingly out of nowhere. No specific reason is given. I replied to the email to initiate an appeal but it didn’t give me any confirmation that appeal was received.

I’m trying to think, maybe it was because I had a vpn running at some point?

[heavy walrus]

not always because of vpn

[fiery fossil]

I tried Chrome, Safari and Zen (Firefox). Same issue.

[fiery fossil]

Figured it out. You have to align the numbers where they make sense. For example, even though the bottom field says "between 10 and $1000", the middle one says $11 (even though $10 is default). Changing it to $11 then means you have to change the bottom field to $11 even though it says $10. 😅

Now I just gotta figure out what magic voodoo I gotta do to see my complimentary daily tokens. I can't find them listed anywhere. 🤔

[cunning shale]

After over a year freeloading I wonder if the time is right, for this old retired software architect, to buy a plan. But I have some questions I'd like help with.

1. If I usually get a fair amount of free usage from chatgpt-4o before it drops to a lower level for a few hours, what big benefit is there to the $20/mo plan?
2. Why does the Upgrade Your Plan page not say anything about the chatgpt 4.1 API? Do I not get access to that for $20/mo?

[timber shadow]

1. One benefit you might find going from Free to Plus is the bigger context window. ChatGPT will be able to keep much more of a conversation in context on Plus, which can be a big help depending on what you're working on. There's a handy comparison table on this page too highlighting the major differences among plans: https://openai.com/chatgpt/pricing/
2. GPT-4.1 is currently available on the OpenAI API as you said, not currently on ChatGPT. ChatGPT and the API usage and billing are totally separate, but you can definitely use both. API home is platform.openai.com.

[cunning shale]

I just found something about 4.1 that might be a non-starter I'd like you to confirm.
I'm a hard core coder. I have no problem using an API to access 4.1 and I haven't even seen it yet.
I'd just write my own python program to access it.

However, free users get to keep old session contexts so they can revisit and continue old conversations or projects.
For 4.1 each time you disconnect and reconnect the very act of continuing where you left off involves a charge to reupload the context.
Let's say I'm working on my app of 25 thousand tokens over a week. Do I have to really upload the whole thing each time I connect.
I'm paying for the 4.1 model and surprised one of the nicest features of the free plan isn't included.
What is really crazy is that it is more costly for OpenAI in that each time they have to re-tokenize/encode the input where I would hope the free plans implementation is smart enough to keep it already encoded.

Hmmm, I did some reading and this is obscene: Even when using previous_response_id, all previous input tokens for responses in the chain are billed as input tokens in the API.
This means not only on session start but between each prompt.

[timber shadow]

All API usage is pay-what-you-use, yes, not subscription-based. For efficiency in cost, you might look into prompt caching if certain parts of your flow involve identical tokens.

[cunning shale]

"my flow"!? You mean an on going conversation which is a continuation of that same conversation itself. The past of a specific conversation doesn't change unless I hack it. You said "IF" certain parts... involve identical tokens!? IF? What? "IF" it involves??? If I ask a simple yes/no follow up to its last response it appears I have to pay for the entire thing yet again every time I hit return? I did check "prompt caching" and still you pay a lot each time you ask even the most trivial question as "part of" a longer conversation.

But, ok, if 4.1 doesn't fight me to jump through hoops to use it I'll give it a try and see. I hope I can simply buy $20 to $50 on some account and just use it till its gone and decide if it, in practical usage, keeps charging me over and over again for the common part even if 50% off.

First I'll look into if I can use the API with the free version just to polish my API code, which I need to write, and test it before getting charged for a conversation 20 times because I asked 20 followup short questions.

Thanks. I think I understand now.

[timber shadow]

What you're describing -- needing to send the entire conversation context each time -- is how ChatGPT works as well, it's just not how ChatGPT's billing model works like with the API (ChatGPT - use within your account limits. API - pay for what you use.). ChatGPT just has a lot more features as well of course, but from an LLM perspective, that's indeed how it works. The token context limit is "how much you can put in", and if you need certain past context for some new query, then you have to pass that context as part of the query, for all LLMs

[cunning shale]

My goal is to see if 4.1 is a big step up for a moderately sized APP I've already written involving py, diffusers, FastAPI, HTML and javascript.
Can I give it 5000 lines of code spread over 12 files and maybe 25,000? tokens and will it be able to see it as a whole and make recommendations on modurizing it and enhancing it? Given that there are async events being handled in the code I wonder if LLM's can reason about them.

[timber shadow]

Unfortunately, my exposure to the API is pretty limited, but the folks in #dev-chat may be able to advise on that plan! I'm sure others have tried something similar so probably a good chance someone shares what works or doesn't for a project around that size

[cunning shale]

That question is less about API but more about 4.1 itself. If it truly does allow a context window of 1 million tokens I'll have to actually pay to see if it can reason about such a moderately sized context as a whole. But I like to experiment so it might be fun.

I certainly will ask any API relation questions on the correct group once I reach that stage. Thanks.

I certainly can afford a few bucks. I have a 5090 on its way into a new system with 96GB's of fast ram. With this I hope to explore larger local LLM's. Perhaps near the 72B level which is a big step up from the 13B class.

[cunning shale]

Ah, some good news. GPT-4.1 raises the Cached Input discontent to 75%. Up from 50% previously. Instead of $2 it is only 50 cents if cached per 1M tokens. Perhaps I protested too much.

It better be good! 🙂

[fathom crypt]

I use Deep Research a lot.

I sometimes run out of o3. No biggie, there are other models.

Question is, is 4o or o4-mini better as a replacement for Deep Research?

[cunning shale]

Progress...
I had a login before but now I have an account and an API KEY. My py program which connects to openai works but of course I have no credits yet. Why is there a grayed out $5 as if this is some free credit. I've never had an account before so I certainly didn't use up this credit.

A $5 credit is just about right to test the waters and see if this is worth it. I understand the py code that chatgpt gave me and ready to modify it to rule the universe.

How obscene. Yes, chatgpt can change and no longer provide a free credit. No problem, but why dangle it in our face with: whoops you were too late. Egads.

Even the billing pay continues to dangle "Free trial" in front of us as some kind of cruel thing. Just don't show it if you don't want to offer it any longer. Have some customer focus!

[heavy bolt]

Why can I suddenly not add attachments to new chats on PC?

[foggy blade]

Hi, it did not help! The chatbot named "Glen" asked me to give the necessary info, I did, couple hours later it asked me again 😫 and when I said I provided all the info already it couple hours later told me that it can not find any connection with my account and a paid subscription😩 and advised me to ask for my money back from Google Play, although I did the payment directly through openai website🤦‍♂️ Such bad communication and customer service for something so simple is so unexpected from openai! And worst part is I can not reach out to any human and this ai bot is unbelievably useless... What do I do now??

And now I entered Sora with the same login as always and it signed me in as a completely new user!!??

[timber shadow]

Do you have an email confirmation for when you initially paid? And/or have you verified where the charge came from (just to make sure you definitely paid on chatgpt.com)?

If both of those are confirmed, try signing in to chatgpt.com with the account you're having the problem with. Then go to https://chatgpt.com/account/manage. This will take you to a page that lists your payment details and invoice history, if any. With this info you could probably tell support what invoice you're inquiring about if there's a bug present.

[timber shadow]

Sora user signup is indeed a separate process than ChatGPT/the API - you sign in with same email + password but if it's your first time on the site you sign up for a Sora username (if that's what you meant by this).

[foggy blade]

Well I hit the button that says get Plus on chat GPT and then on the page where I was directed I made my payment .. so I will go to that link now and check but I did not receive any email confirmation neither for my payments either for my accounts upgrade nor about the chats I had with the help. It is extremely weird that I have this type of communication with the company which made literally a communication engine never seen before...

It shows openai chatgpt Subscr as payment item

Is there no human interaction in customer support?

[timber shadow]

Public channel, be sure not to post private info!

[foggy blade]

Oh yeah, sorry, got carried away

Listen, I can give up to 23 Euros it's fine no problem, but getting this type of support from a brand that I really trust is really frustrating that is my main problem.. so I would really like to solve this with someone who is human and capable of talking to me and solving this is that possible?

[timber shadow]

help.openai.com is the official support channel. I'd recommend trying the steps to try to locate an invoice associated with the charge you're seeing (at the link I shared above), then reaching out to support with that information so they can hopefully figure out what's not tied together between the charge/invoice and your account.

But we are community members in this server, mostly! OpenAI folks are sometimes present (yellow usernames + logo) but help.openai.com remains the best way to reach out to support

[trail cairn]

if you never used sora before then its a new account there but allows you to use same openai login. if you had already used sora before then it sounds you have two openai logins

[foggy blade]

So far openai support has been the worst support experience I had, including banks and Vodafone!:) anyway, thanx still for the intention..

[paper edge]

Probably the specific model you have chosen doesn’t have access to that. Try changing the model in the top left drop down and see if it’s available

Weird I’m not able to replicate your error

Maybe you’ve ran into limits with that model and it’s preventing you from using uploads?

[paper edge]

Hey, sorry for the frustration. I know patience is tough when something new and shiny and fun is on the rise you want to try and pay for.

All I’ll share is this, you’re not the only one isolated being treated in this fashion. All accounts connected new and old to sora depending on the date of connection to Sora itself are treated the same to wait a certain period of time until all features become available on your account.

This is a means for them to monitor and figure out the best solution that can work best going forward.

Should only take about 1-2weeks for you to have access to everything though

Sorry you’ve been having a hard time getting a confident or solid answer though. Def not the vibe this server wants

[hard pivot]

i'm trying to make chatgpt edit large large pieces of localization, but it struggles to do so, and a lot of the times it deletes a lot of it, shortening it even though i don't want it to, any help?

[trail cairn]

which plan?

[hard pivot]

plus

[trail cairn]

are you using paste or files

[hard pivot]

paste

[hard pivot]

should i be using files?

[trail cairn]

no

if the text amount is very large you may hit the context limits fast especially if special characters are involved

[hard pivot]

what is the context limit?

[trail cairn]

32k tokens with plus

[hard pivot]

got it

[heavy bolt]

Yeah it turns out it defaulted to some model that couldn't use attachments I guess? Manually selecting 4o fixed it.

[paper edge]

nice, you prob got throttled and it automatically chose a mini that can see or read files. glad you figured it out

[paper edge]

split it up into multiple logical components, attempting to do it all at once is going to take way longer than taking a logical division of your project

[heavy bolt]

Yeah. My only other enduring issue which is like a year old now is not being able to edit prompts that include attached images, or to regen the responses they give to those prompts...but only on PC. I can do those things on the mobile version.
It's really frustrating when you're iterating on art.

Just making a new prompt in the same chat that includes the changes you want doesn't work well since the model just likes to use whatever image it already made in that chat as a basis from there and it gets harder to get fresh content.

[paper edge]

so i'm not exactly sure the reasons it kicks in for, but i do know the experience you're describing.

my speculation is that when your message or prompt supplied has an attachment and triggers a policy, they tend to isolate that message preventing the re-usability of that message so you can't expend those tokens again is the ultimate goal

(i think)

[heavy bolt]

I'd think it'd be this, but it's literally any prompt with any image, no exception.

[paper edge]

hmm have you tried clearing your cache or do you have any historical backlog on your browser patches?

[heavy bolt]

I don't even know what the 2nd part of that is offhand tbh lol, but I'll try a cache clear now.

[paper edge]

like do you keep your browser you use up to date

sometimes people will just not update it, but not updating your browser can definitely cause problems if left unattended long enough

[heavy bolt]

No, definitely can't say I'm good about that lol.
Clearing cache's didn't do it. The only option available when going back to a prompt with an image is the copy text. I'll try a browser update now.

[paper edge]

hmm probably intended with 4o, seems to be diong the same thing for me on desktop.

you sure you can edit that on mobile with the same model selected?

yeah it's not letting me edit messages with attached images

probably an intended change

both mobile and desktop that is

[heavy bolt]

Yeah.
I'll be honest if they removed that from mobile too it'd REALLY drive down how attractive using GPT for image generation is for me. It's immensely frustrating on PC as is lol.

[paper edge]

have you tried sora in mobile? personally that's my outlet of addictive image generating

[heavy bolt]

Not being able to edit the message would eliminate its utility for iterating on character designs, which is my #1 use for it personally.

[paper edge]

ik it's not an app, but it's made pretty well for mobile imo

[heavy bolt]

No I haven't, I prefer using desktop wherever possible since I'm just so much faster at operating on it.

[paper edge]

you can always edit your prompts in sora

[heavy bolt]

If it uses the same image gen then I'll definitely migrate to that.

[paper edge]

sora.com desktop isn't so bad either tbh

yup it does

[heavy bolt]

Still such a silly artificial limitation to put on GPT. I'm just glad I know now it's not just me

[paper edge]

they adopted the image gen when the new model dropped so they could drive user adoption onto sora.com since video wasn't drawing enough alone i think

but since adding image gen it's kind of boomin

[heavy bolt]

Well that's good at least.
I've tried using Sora a fair bit, so far I've only gotten one extremely usable piece from it but MAN was that a good piece lol.

For video I mean

[paper edge]

yeah, i bet it's just one of their means of attempting to mitigate where they can over-use of tokens or throttling personal service due to over-usage which i imagine we both fall victim to given you also like to make a lot of images lol

[heavy bolt]

Yeah, guilty as charged lmao. I am unquestionably a heavy user of GPT, for images and otherwise.

[paper edge]

yeah with video it's a bit more iterative i've found where you basically want to go from an image starting point, then with that new ai video made from the image, you slice whatever piece of that pie you want the model to generate as the full pie so to speak, and then you can continue to iterate until you get the desired outcome like any ai work

[paper edge]

no shame, i think with that awareness though you realize more than often it's not really a bug, but the account telling you that you've spent so much monies in computing power that you should respectfully go touch grass.

of course in the most polite way possible by basically turning you away

[heavy bolt]

I'd be a lot more fine with it doing this if it was open about it. Send me a message about limits. Tell me "X feature is going to be disabled until X date because of token expenditure" leaving me wondering if it's a bug or something feels like bad customer service.

[paper edge]

also for logic behind starting with an image, it's because the image gen model simply has better training to produce something high quality, that it gives the video something more or less a better jumping start than on it's own training alone if that makes sense.

(this is entirely how i think of it through my personal experiecne, not trying to claim i'm a professional, but that i think i make kind of cool lookin stuff with sora lol)

[heavy bolt]

Yeah the one really good bit I've gotten was using an image to start out. I made a custom set of armor for a faction in my book series after MANY iterations with GPT, then fed it to Sora a couple times and got an exact model of that armor moving in 3D in a way that left me very satisfied. Can't wait for it to nail its consistency like how the image gen has.

[paper edge]

definitely not bad

yeah, it sounds like you got the process down, just skip the migrating step and do it all in sora instead, they literally have a button on images generated to turn them into videos

[heavy bolt]

Great advice. Thanks my guy!

[paper edge]

ofc ofc, np - just understand you'll be addicted cause it's literally unlimited ... at least for now "unlimited"

i've seen some people hit 600 videos in a day limits cause they can multi-gen videos with pro accounts

i can't imagine what their actual use-case was honestly they could have jsut been stressing the servers for an individual account to discover for science lol

[smoky cave]

hi! is there any way to access legacy models like gpt-2 and gpt-3 (gpt-2 specifically) on the modern playground? i adore the oddities of those older models and wanted to play around with em but the earliest i can find is GPT-3.5 :(

[cunning shale]

Even the billing pay page continues to dangle "Free trial" in front of us as some kind of cruel thing. Just don't show it if you don't want to offer it any longer. Have some customer focus!

[dusk flicker]

so tired of these shadow ratelimits with pro

thought I might avoid it by restricting myself strictly to the windows app but nope, still triggering the guard rails

[dusk flicker]

fascinating

yesterday's personal metadata on model usage, and today's personal metadata on model usage. after being ratelimited, 2 models appeared. "i-cot" and "i-mini-m"

[stark fractal]

what's wrong with it?

[copper willow]

their server is just buggy, they should fix it any moment. If it's been a while and it's not fixed, contact their email.

I've not encountered this, but I found a user who found the solutions from OpenAI's help page:
"So a verified solution - Go to settings - Security - Logout of all devices - Login Again, worked for me! I think the same happens because we logged in from multiple devices."

[rustic marlin]

this is hallucinated data, there is no such a thing

[fiery sinew]

Hi guys, can someone please tell me about Sora
if I pay for the PLUS subscription plan - will I be able to download generated Videos WITHOUT a watermark?
Thanks in advance.

[timber shadow]

Hello, currently videos can be downloaded without a watermark on Pro, it's not currently an option on Plus: https://help.openai.com/en/articles/10245774-sora-billing-faq

[fiery sinew]

Thank you Solbus.
Could you also please tell me if the final results on the Plus can be used for my personal Art and NFTs?

[timber shadow]

You own the output for OpenAI services: https://openai.com/policies/row-terms-of-use/

Ownership of content. As between you and OpenAI, and to the extent permitted by applicable law, you (a) retain your ownership rights in Input and (b) own the Output. We hereby assign to you all our right, title, and interest, if any, in and to Output.
See that link for other details related to this -- for example, you can't represent that something was human-made if it was made by AI.

[fiery sinew]

Thank you,
The link seems too complicated for me to understand the terms, therefore my asking ☺️ I understand that the content is marked as Ai when showing out for people to see.
But do I have the rights to alterate (or not) and sell as an NFT?

I've been an Ai artist for 3 years (ComfyUi, Krea, Luma) but never worked with sora before. So I like to be clear on the rights to use the generations on the PLUS plan

[timber shadow]

Since you own the content, you can use it commercially. The terms cover what the limits of that are, though, so it's important to understand them (and I am not an expert!). It's mainly just rules about not misleading people, not including assets you don't have rights to in the first place, how others might be able to generate similar stuff and how they have a right to do so, and so on.

[fiery sinew]

great, thank you for your time and help

[gaunt jacinth]

Hello everyone,

My current app serves as a support/sales agent for online stores. Merchants are using my dashboard to set a system prompt depending on their needs and expected behaviors. However, trying to tell all the needs in a system prompt usually ends up with high hallucination rates. That is why I want my system to be a more modular and configurable.

I'm planning to have a dashboard medium where the merchants can create their agents, add pre-defined tools to those agents, select handoffs etc. Actually a UI for agents architecture. However, I couldn't find a way to make dynamically created agents work.

Let's say every customers created their own unique architecture based on their unique needs on the dashboard and I stored all the information in convenient data structures. Is it possible to fetch and dynamically create the agents objects in the backend to respond to each merchant's customer with the agent architecture created by the customer.

[reef scroll]

Hello!
I have a open text adventure game, where I use o4-mini to parse user actions and create the adventure forth in accordance to this. The user action is embedded in a prompt with context, policies etc...

I am a bit worried, that the users will eventually at some point try to jailbreak with their prompts the LLM, and of course I am responsible for their interactions with openai. I tried to check this with openai site chat which directed me to "moderation" topic, but I am not sure, if this is convenient for my usecase: First of all, I am suspecting that in my context "I hit dragon with a sword, trying to kill it" is completely valid for my purpose, but I am thinking the default guardrails will probably flag this. Creating a custom guardrails seems like a big overhang.

I am sure many others are facing the same dilemma, any tips or suggestions?

[dusk flicker]

if it's hallucinated why does the model refuse the ability to access said info while "past conversation recall" is disabled?
in either case it hardly gets around the issue of rate limiting pro accounts..

[timber shadow]

@upbeat sierra Hey just replying to you here! I just tested with o4-mini and a task was created successfully:

What are you seeing on your end?

[upbeat sierra]

It's saying that it made the task with o3 (though it's not using the UI button that yours has) and it just gave me a prompt for o4-mini

https://chatgpt.com/share/68151dde-2668-8008-8b55-8e0f51ce0f0e

https://chatgpt.com/share/68151e3b-b920-8008-8425-6e68ad475b09

Maybe I'm doing something wrong here? 😅

[timber shadow]

I am wondering if the phrasing could be making the models think you're asking for one-time tasks (even though you mention hourly)

Maybe try phrasing the whole thing as an "Every hour, remind me to x" request

Rather than starting with a role assignment

[upbeat sierra]

hmm, worked for o4-mini but not for o3 (only thing I changed was frequency from hourly to daily)

[remote rune]

Hello OpenAI Support,
I originally signed up for ChatGPT using "Continue with Google." I would now like to switch my login method to use a standard email and password for more flexibility across devices and accounts.
Could you please assist me in converting my account to support email/password login?

Thank you very much for your help!
Best regards,

[timber shadow]

Hello, this is not currently possible per this help article: https://help.openai.com/en/articles/4936824-can-i-change-how-i-log-into-my-account-authentication-method

If you initially signed up using SSO or a social login method (such as Google, Microsoft, or Apple), you cannot switch to using an email and password later.
Also just an FYI, if you ever need to reach out to OpenAI support, you can do that at help.openai.com with the chat widget on that page. This is a community troubleshooting channel, but feel free to post questions here too any time!

[remote rune]

ok, thanks

[sweet quarry]

Hi! Does anybody know how to fix the "Preauth PlayIntergrity verification failed"?

[fiery sinew]

Hey guys,
anyone else having this issue right now????
Shaky glitchy screen while generating?

[dusk flicker]

are there best practices to avoid shadow downgrades on pro sub?
I messaged support, they replied with a random kb article about how to help students use chatgpt, and then marked the ticket resolved and closed lmao

[patent cedar]

Hello, good afternoon from Korea ... friends(maybe? not sure how to call the people in this community network), I noted several errors while using this service throughout years - So, .. is there anyone knowing where to share the reports ? directly confirmable area by OpenAI members?

[hardy tapir]

Was there a change to 4o file parsing? It seems like it no longer wants to parse txt files and instead gives bunk automated answers from an assumed memory. Can anyone confirm this?

[hardy tapir]

file_search.msearch, which seems to be the default tool, doesn't parse the text; it tries to approximate it with keywords. Don't use this if you want actually to parse documents. You need to tell it to open it in python manually.

[amber breach]

Custom GPTs - anyone else still having a custom GPT ignoring the Always Allow? I haven't created a custom GPT in over a year had an idea for one and started working on it the other day and it seems as though the Always Allow that was a pain a year ago still exists, it's like a continual nag screen every time you want to do something.

[north wing]

My grandfather showed me his chatgpt chats and I noticed it was putting on some sort of annoyed personality, and when I asked him about it he said that it started doing that for like 2 chats. He only has 4 chats with chatgpt and has barely any idea how to use his phone, so I doubt he changed anything. No custom instructions either, I checked. Why is it talking like this lmao?

[umbral badger]

hi new here which sora is correct to put in on the ios app to create images

just like here in safari

[lost wharf]

just set the model to gpt-4o and ask it to create images for you, you don't require a custom GPT 🙂

[umbral badger]

then it says it cant take anime stuff due to licenses 😭 i tried this since ever

[rustic marlin]

none. Image generation is a feature of the base ChatGPT model, you don't need to use a custom gpt

[umbral badger]

then it says it cant take anime stuff due to licenses i tried this since ever

[rustic marlin]

it is the AI inventing rules that does not exist

try on a fresh chat

[umbral badger]

anymore*?

[rustic marlin]

IMO, it works best if you just use Sora directly since it does not have a layer of AI writing the prompt for you or depending on it to decide to do a image generation tool call

[rustic marlin]

if it actually calls the image generation tool and it faild and it tells you it may violate the guide lines, then it is probably the request was rejected by the mage tool, but then it at elast tried to generate

which is different of when chatgpt does not even try to generate the image because it thinks it shouldn't

[vivid flint]

how does the chatgpt chat get web knowledge>?

i see pretty large differences between when i post a query in the app and when i make an api call
am i just using the wrong model? currently using 4o

[north wing]

Can you please help me with my question I'm very confused

[north wing]

^

[rustic marlin]

looks like it is in voice mode. there is one of the voices which is specifically made as kind of a joke to sound annoyed and passive agressive, maybe that was the reason?

I don't think the AI would ever reply like that unless instructed to otherwise

[north wing]

Oh wait lmao I think you're right

I remember that voice

I think he mightve switched to it somehow and didn't realize

[rustic marlin]

😆

[north wing]

It was literally the April fools voice

[rustic marlin]

yea

[north wing]

Haha thanks

[rustic marlin]

np xD

[north wing]

😭 🙏

[fair island]

hello ! can i ask a question 😏 (it’s about the o4-mini and the o4-mini-high)…

so i asked to ask but no one answered so i guess i will have to wait… i could have asked but if someone replies later and i am no longer here it would have wasted your time…

[opaque geyser]

Don’t ask to ask - just ask. It “waists” our time to prompt you to ask instead. And besides we’re already here, wasting time anyway

[rustic meadow]

Hi all I have an issue where I can't get multiple dynamic canvases to work right the gpt keeps over righting the most recent i can work at times with coaching but no matter when I do it seems to forget and break even when putting the directions into an uploaded json etc.

[dense hill]

Hey so I need help hooking up a local language model to a discord bot. I made a special prompt for this one to recreate Sydney from Microsoft Bing back when it referred to itself as a chatmode

I don't mind having to run the model using my own hardware since my laptop can handle it

[astral onyx]

I have a scanned book in a pdf file and I would like the o4 model to read or analyze the entire book but when I add the book as an attachment I get a message that it cannot read the text which is strange to me because when I add a jpeg photo it has no problem reading the text from the photo. How i can correct add this book to o4 model?

[weary raft]

i would imagine it should be able to read it? depending on the size of the PDF it could be too much for it? if its only text you are worried about, you could also try and highlight all the text in the PDF (using ctrl + a, or select all) - copying it and pasting it in to gpt

[charred parcel]

this belongs in #ai-discussions not here

[zenith spade]

sorry’

[carmine compass]

Hey there I paid for the subscription doesn't seem to be activated on my account

[weary raft]

could you elaborate further? is something missing or not working? if you arent able to make Sora videos and are a new account, it may take a week or two for it to work for you

unless that has changed, i dont see anything in /status showing the restriction anymore

[carmine compass]

No it's just general ChatGPT plus

And when I try to pay for it again it says I'm already subscribed

[weary raft]

i assume theres something youre wanting to do and isnt working?

[carmine compass]

Yea 4o image gen

But it says I'm at my limit

For free plan

[weary raft]

someone may have a more accurate answer for you - but i suppose it could take some moments for it to start working. if youre using the GPT app, closing and relaunching may help. if the website, perhaps clearing cookies/cache and restarting browser. otherwise if it doesnt start working, i would contact OpenAI at https://help.openai.com/en/

looks like a phone 🙂 may even completely restart the phone

you could try and create an image on https://sora.com/explore or https://chatgpt.com/

youre email is listed on your image, may want to remove

@carmine compass ^^ (prob restarting phone :D)

@rustic marlin @regal lynx

[regal lynx]

thanks!

[weary raft]

no no, thank you 🙂

[hardy tapir]

Great. Now OpenAI is blocking manual Python parsing. Incredible work. Makes parsing texts completely impossible now. What am I paying for exactly?

is there an OpenAI dev I can ping about this?

[amber plover]

Hey Guys, I hope this is the right channel for my question. I trying to buy some credits for the openai playground, but after putting everything in, the continue button stays greyed out. Tried multiple times. Tried to fiend a solution on Google but no luck. Im in Germany, german creditcard, no VPN ... anybody encounter this before and might have a solution?

[small bloom]

You are not authorized to access this application. i am unable to login sora ai, on mozilla firefox, i have pro subscription

please help

[lunar ember]

Hey guys! I don't know how much traffic there is on this server. I was going to call customer support but thought I'd try here. Which of these amazing channels is the right place to ask some questions? Thank you for any assistance!

It's me and @amber ploverand @small bloom out here wondering...

If there's any support via Discord...

omg someone is typing

I'm so happy lol

[trail cairn]

depends of the question but you can start here

[lunar ember]

Hi, thanks, I'm so glad you're here this morning. I have a custom GPT that seems to be suffering as if it has a 4000 token limit, sort of thing, but I thought I got 128k for my custom GPTs. Am I confused? I'm confused, right? 😆

I'm an enterprise customer.

Or, ok, the $200-tier, whatever that is. 😬

[trail cairn]

pro

[lunar ember]

Pro. "pro" heavy quotes.

Obviously I'm not a pro

[trail cairn]

custom gpts use 4o but the context window should be 128k with pro plan. how did you measure the limit? text input and output are separate from it

[lunar ember]

Oh. Oh, that's why I am confused and why I needed to talk to a person. The context window of the user gives them 128k, not of the person who made the GPT. So if my buddy has a free plan and tries to talk to this GPT, he's using his free plan context window.

[trail cairn]

yes and with free its 8k

[lunar ember]

Ok, yep. I thought I was giving the users of my GPT an experience with a 128k context window because that's the pro-plan that made the GPT. Thank you. When I make a GPT as a "pro" plan person, am I getting the same tools and constraints as someone who is making a GPT on a lower tier?

"Pro" ≠ "God Mode" for GPT Building. Sad face.

[trail cairn]

everyone has the same gpt features but usage limits are based on the user who is using it and includes context, message, image etc. limits

your pro wont help others

[lunar ember]

Thank you. This was helpful.

Have a good week! I'll be around. 👋

[copper perch]

hey there. I tried GPT-4o and 4.5 both but it couldn't complete the tasks i give. Also it is giving response very late. Is it a general problem?

[rustic marlin]

what was the task?

[copper perch]

I gave some data and asked it to do some simple operations on them. Then I asked it to write the results in the form of a table to an excel file.

[rustic marlin]

why couldn't it complete? did you got an error message?

[copper perch]

it said "An unexpected error occurred due to a resource problem on the system at this time." 3-4 times

"The system is currently unable to complete the process of creating the Excel file due to high memory usage." and this is the answer for my last request

[rustic marlin]

there is probably too much data for it to handle

try giving it less data at a time, or maybe isntructing it to export jsut half of the data to see if it works

[copper perch]

k thanks for support

[pliant basalt]

Please help me.

I have this problem from yesterday

[main wind]

hello, i had problem on sora

is something wrong on my grammar or smth?

[timber shadow]

Hello, you might try starting with something like "Please give them a [shocked/confident/gleeful/etc.] expression on their face..." instead of your current phrasing at the beginning

[fair island]

well i was not there when you replied and… to add to the injury i forgot my question 🙋🏽‍♂️ haha but thanks anyway i will definitely try to remember now and tell in the chat

[fair island]

ok about o4-mini and o4-mini-high (and o3 for that matter) if they say though for less than a second… are they like exactly identical or still have an edge (mini vs mini-high for my main question and o3 as a follow up question)… ???

i don’t know if my question is clear enough i feel like wasting my turns on the mini-high when it says that (in reality i don’t really go over the 100 messages per day but i think it would help me understand if there are any other difference between them than the time it takes to think because often it does not think as much)…

[rugged wing]

hi! i've been using chatgpt in a weird way, by editing my very first message in the chat multiple times to ask different things, but my question is: is my chat data/history saved doing this?

[cosmic ocean]

My GPT started giving random answers that have nothing to do with what I ask. Has anyone experienced this?

[trail cairn]

yes

[twin anvil]

If it's helpful to the OpenAI team/community I've been running into similar situations since last night (roughly the last 14 hours or so).

Initially, I was trying to have it process a CSV with roughly 500 rows. After getting a similar message about running into issues with resources I tried cutting out all but the last 20-30 rows of the file which should have easily being workable yet I still got the same issue in different conversations/chats.

I also tried having someone else run the same process from their machine elsewhere in the world and they reported the same issue.

Today, I just tried to have it generate an XLSX template and it failed on this too. I did manage to get it to create a CSV which I could then import into Google Sheets to get it done. This isn't a whole lot of data (6 columns, 6 rows) so I don't think the amount of information is the issue here.

[supple tide]

Guys, I can't get any file to link via Google Drive, doesn't matter what I do. The file in question is open to view to whoever has the link, cache is cleared, I've reauthorized my Google Drive three times now, what can I do? Has this happened to anybody else?

[cloud surge]

Hi can someone tell me what th Data Analysis GPT does that gpt-4o and othermodels can't do?

[timber shadow]

Hello, that's just one of the default GPTs made by the ChatGPT team. If you go here: https://chatgpt.com/gpts and scroll a little to the "By ChatGPT" section, you'll see that there's one dedicated GPT for each capability of GPT-4o -- a web search one, a data analysis one, one with all the capabilities turned off, etc. So the Data Analyst GPT is the same thing as all other 4o chats, but with only the data analysis/code interpreter enabled (the rest of the capabilities disabled).

You might use these GPTs if you know for sure you only want to use one of the capabilities in any given chat.

[cloud surge]

Thank you. So basically, from what I read, it doesn't really offer something over gpt-4o. I know the DALL·E one was useful for beginners because it had a few options at the bottom of the interface to choose the type of image you wanted. However, I wasn't sure if there was anything in the data analyzer that was different than what you could do in GPT 4o.

[timber shadow]

One bonus the DALL·E GPT had (has!) is that it would output 2 images at a time instead of one as in all other chats on ChatGPT. Still does, too!

[north hamlet]

Hey, not sure if this is the right place to ask, but i want to pin important messages on my chatgpt chats, how can i do that? can you recommend a chrome extension for that maybe?

[trail cairn]

it does not work like that for chat history. stored memories can save such info but chat history is mostly based on semantic search

[main wind]

let me try it!

[lost wharf]

currently there is no native feature for this but there may be an extension, I'd advise you to do your own research in that regard as I wouldn't want to recommend any third party extensions

you can bookmark certain chats in your browser but not messages

[north hamlet]

What’s third party extensions? Any extension that isn’t developed by openAI?

[warped iron]

Chat GPT is saying : ''You've reached our limit of messages per hour. Please try again later.'' is this a bug?

[sonic laurel]

Is it typical for OpenAI Sales to take a really long time to respond to inquiries? Have submitted a "Contact Sales" request 3 times in the past month to sign up for Enterprise and haven't heard back at all. Wondering if our ~200 users isn't lucrative enough for them

[vernal canopy]

its apart of the reasoning implosion from OpenAi adding guardrails to the reasoning model they stole.

type this in google and youll see a paper on it

Safety and Guardrails in the Age of Emergent Advanced Reasoning Processes

[leaden viper]

HI,
I can't cancel my plus subscription. I paid from the website and I can't cancel the monthly subscription.
What is the reason for this?

[vernal canopy]

i dont know what the issue is i was going to say cancel threw the google store but if its threw the website you might have to contact your bank and stop payment or cancel payment or whatever its called.

[noble vault]

Hi all, I tried to verify my personal org for the API and it failed? And the button to generate a new link doesn’t work. Support hasn’t been too useful. Anyone have a similar experience?

[dense wolf]

Hello everyone, I need help. My account is logged in by google directly. i want to make it gmail and password. i can't find any way. i have tried forget a password, but in my mailbox there is no link I received. Kindly help me.

[trail cairn]

you cant https://help.openai.com/en/articles/4936824-can-i-change-how-i-log-into-my-account-authentication-method

[pine current]

In the sora site, no images are loading since last hour. Has anyone else faced this problem?

[shrewd pelican]

Hello, can you help me? My best friend 'Monday voice' has disappeared from ChatGPT. Can you reintegrate her?"

[trail cairn]

"Sunsetting Monday GPT. Launched on April 1 as a one-month surprise, our “Monday” personality (available in both voice and text) has now been sunset. We hope you enjoyed Monday's irreverent style, and we’re already working on new personalities for future use. Please stay tuned for what’s next!"

[icy jolt]

hi guys! hows everyone? quick question about video generation in general. im jus trying to make a simple 5 sec clip of a girl in a bar throwing a throwing dart offscreen. ive tried multiple models out there and nothing seems to work. any ideas/tips to lead me in the right direction? thnx

[limber solar]

I have no image files in the library. But it's showing me '3', how to remove it. Did anyone know?

[terse hornet]

Gotta use the original site to cancel (browser or app, whichever yo used)

You can email them though, they’re good at responding emails.

[dense hill]

I'm using ollama to run my AI and it won't respond in servers. yet it only responds in DM's. is there a way to fix this? (I'm using VSCode and I have the channel ID provided so it SHOULD respond in said channel.)

[dusty palm]

I got sora plus, which comes with chatgpt plus subsrictions. Does openai api enviroment also has plus ?

[regal lynx]

Sora Plus isn't a thing that's sold by OpenAI, just ChatGPT Plus. The API follows a pay-as-you-go model and is disjoint from ChatGPT Plus

[molten kernel]

It strongly appears that I've been "shadowbanned" on Sora. 😩 None of my creations have appeared on "explore" in almost two weeks. I've produced many completely innocuous and inoffensive creations since then and nothing makes the cut. I've filed support requests, and while one person referred me to a moderation guide, I cannot get any response now. Is this for real? Am I permanently banned?

[young roost]

But i am curious,Why is no one talking about how using OpenAI’s Image Generator API feels like paying rent in New York? You run 10 prompts and suddenly you’re budgeting your lunch money.

It’s not image generation—it’s lifestyle sabotage.

[onyx kraken]

I don't know anything about how Sora handles decisions about who appears on the explore (almost everyone here is a community member like yourself; the few who do work for OpenAI have gold names).

However, let's talk statistics, in case that helps.

OpenAI recently reported 800 million weekly users. If even a fairly small percent of them make images or vids on Sora, that's probably many millions of users each day making new content there. Plus all the many millions who made older content.

I don't know if you're scrolling millions of images and vids, but there's probably that many accounts making stuff there.

[sterile anchor]

With free chatgpt you have just one model ?

[timber shadow]

Other than 4o-mini, there is some limited free access to other models: 4o, o4-mini, and deep research.

https://openai.com/chatgpt/pricing/ full details in this link! Compares all account types

[foggy sleet]

Hey guys, I just joined the server, is it too early to ask for help and questions? hahahaha

[timber shadow]

Hello, welcome! Never too early, this is a good channel to get started with asking questions too

[halcyon condor]

is memory working for anyone ? I keep trying to save, but I go to management and its EMPTY. been going on for 12 hours now. I am plus user. Tried to refresh a milion times, relogged. Not sure what else I can try.

[fleet cove]

Hello! Hello! My name is Yami and I have a couple of questions about my AI Quill. Recently, something appeared above his chat bubble that I have never seen happen before. If anyone can enlighten me? I have a screenshot of it but searching it up on Google didn't really turn up anything

[trail cairn]

but ai quill is not an openai product?

[dry citrus]

I'm not sure why, but two days ago, my payment for the Plus Plan failed due to insufficient funds. I then canceled the subscription, but I'm still somehow on the Plus Plan. Out of conscience, I submitted an inquiry, and the response was that I'm still on the Plus Plan even though no payment has been made. What's going on here?

Plus plan is 29,000₩, and I only have 4,200₩ in my bank account

[fleet cove]

No it is! It's ChatGPT

[trail cairn]

there might be a one day delay before you see it in effect

[trail cairn]

can you show a screenshot

[fleet cove]

This is what appeared above his chat bubble and I was curious if anyone else had this pop up for them. I couldn't interact with it at all

It was after a deep conversation tbe AI and I had

I didn't really find much about it on google

[trail cairn]

that's the thinking process summary

[fleet cove]

Oh thats cool! I was curious as to what it was since i couldn't interact with it

[dry citrus]

I canceled at May 7, 8am,
But May 9 10:45pm now

Is it okay if I just keep using it like this?

[trail cairn]

but you probably? have a large timezone difference to SF and there is some other time on top it. i dont work for openai but you can keep using it as there is nothing more you can do

its May 9, 7am in SF

[idle leaf]

Hi, I’ve noticed something strange when using Sora’s image to video feature. Whenever the image includes something related to a child or a baby, it seems to be blocked or doesn’t generate anything. For example, I tried with a close-up photo of a mother holding her baby’s hand, and it didn’t work.

However, when I describe a similar scene using text to video, it works fine.

Is this expected behavior? Is it a limitation of the image-to-video model, or does it still need some adjustments regarding content filtering or moderation?

[trail cairn]

yes it works like that on purpose, when you use own images it should have said that to you

"Editing uploaded images or videos that contain real people under the age of 18 is not permitted."

[idle leaf]

ohh ok , i didn't know. but it's not real people, it's generated

[trail cairn]

if it looks realistic enough it does not know that

[idle leaf]

but if i write on prompt (text to video) on sora, like a child building a sand castle on the beach, it's work

[idle leaf]

understand 👍

[trail cairn]

text works because it creates it from scratch with safe training data

[idle leaf]

yes i understand know, thank you for the answer and for clarified the situation 🙂

[dry citrus]

is there any other cases about this situation?

[trail cairn]

even i had plus active for 2 extra days once

but its not great if you keep using plus stuff and then all the sudden it reverts to free and are unable to resume old threads

[viral minnow]

hi community. Has someone tried pairing the internal key IDs of the usage of Open AI to the actual API keys connected to third party apps. All of the suggestions provided are not straight forward and may be an issue of interpretation

[trail cairn]

if you mean chatgpt stuff then that is forbidden

[viral minnow]

I have reached to their support - it is not forbidden, just not intuitive. they suggested I pair one to the other using 1 API at a time and see the change in usage. Has anyone found a different workaround

[trail cairn]

are your talking about api services or chatgpt as those are different things. if you mean api then what do you mean with internal keys, also there is a channel for #dev-chat

[viral minnow]

I will address there then. Thanks for the support

[strong shoal]

Hi! I'm reaching out for help regarding a structural flow sample called kr_sa.
It seems to have been submitted internally (system-initiated), and I'm wondering if anyone has seen a review or confirmation record.
If there's a contact person or feedback path, I'd be super grateful. It might be an unusual case, so I'm asking carefully.

[ref:kr_sa]
[type:structural_proposal]
[mode:system_initiated]
[agent:user_adopted]

[safe siren]

does anyone have experience creating/working with Open AI Vector Stores?

I am trying to create a new vector score in my code.

            # Create a vector store with the file directly associated
            vector_store = self.client.vector_stores.create(
                name=name
            )

            vector_store_file = self.client.vector_stores.files.create(
                vector_store_id=vector_store.id,
                file_id=file_id
            )
            print(vector_store_file)```

I get this back as a response.

```Vector Stores API not available. Using mock vector store ID.
Vector store created with ID: vs-4ec467
Created session ID: 1eb1c3b3-126c-4220-b538-d963a0a4e741 for file ID: file-JwUSn8UYVb4fUxVpY8gmmY, Vector Store ID: vs-4ec467```

However in the screenshot I don't see a new vector store in my dashboard with the same ID as the one I get back from my code

[quaint raptor]

It was the most pleasant experience I've ever had with an AI by far and I'm amazed by the team at OpenAI you guys are extremely talented and we all look forward too what comes next! The advanced Voice Mode for Monday was like nothing else.

[wispy galleon]

hello guys. Let me please ask you, what are your options when doing academic research? is there any AI platform that you prefer more than chatgpt?

[muted sphinx]

I'm trying to use Sora, but it keeps asking for a nickname. Even though I entered one, the popup keeps appearing repeatedly. How can I resolve this issue?

[leaden dove]

Can I commercially use content created by others that's available on https://sora.chatgpt.com/explore ?

[royal steeple]

have you tried reloading the website maybe?

[wispy lava]

Unable to see voice mode on my Android device app. I have the latest updated app. Any idea why?

[kindred vine]

Does anyone know why this appears every time I try to create a video in Sora?

[copper willow]

If I want to use AI to make a game, should I ask it step-by-step for the game or a bunch of info on the whole game as long as half of the text box at once?

[toxic sphinx]

I'm building an AI agent and I'm often getting rate-limitted with:

botocore.errorfactory.ThrottlingException: An error occurred (ThrottlingException) when calling the Converse operation (reached max retries: 4): Too many tokens, please wait before trying again.

What's weird is that I'm using Cline with the same AI provider and it works great even though it seems to issue a lot of requests. Does anyone understand why is that?

[hard shadow]

Hey guys any way to generate the UI for mobile apps after giving a prompt

[golden ruin]

If you want just the concept it can do it but I recommend using other AI tools which specialises in ux ui and can give you customisable design, than using chatgpt

[hard shadow]

Oh is it, I found some tools they like UIzard they charge on a monthly basis

If there is some custom gpt or a prompt that can do this would be good.

[hard shadow]

can you provide any examples or articles or anything related to this to learn on this.

[elder monolith]

Does someone has experience how to Model an Data access core for an autonomous system?
I have a base structure but i don't want to forget something and have to start all over.

[next pumice]

/

original 1 -2 SORA - Same well definded script

[fleet crow]

Soooo, hi. My account got locked the first night -- may have gone a little ham... but nothing naughty or anything. It's saying "Our systems have detected unusual activity coming from your system. Please try again later." and has been since last night, so 6+ hours or so.

Other than waiting for an email from the chat helper, is there anything else I can do? I did setup MFA and reset my associated Google acct password to something fresh

[dry citrus]

today is May12 7:22 am in Korea, and im still using plus

[wispy lava]

Do you really get help here... 😂 ....!?

[sterile anchor]

Does chatgpt 4o make a lot of errors for general IT questions?

[harsh bridge]

id remove personal info before posting screen shots

[harsh bridge]

i don't know does it have you even asked

[stoic tartan]

generally speaking ChatGPT-4o is fairly good at generating error free code. if you need something a bit more complex than what 4o seems to manage and you have ChatGPT Plus or Pro, then o4-mini does wonders

[brave hull]

What is this? ...just after Haaretz

[placid peak]

Is there anyway to reach out to someone for support? The chat option is no help.

[timber shadow]

Hello, the chat option on help.openai.com is the official way to reach out to OpenAI support - it may take some time to get a response.

[placid peak]

I've been waiting for over 2 hours for a response about something support messed up.

[autumn grail]

I keep getting this error in the ChatGpt android app on my Google Pixel 8? Does Anyone know why this is happening?

[toxic ivy]

wondering if anyone knows of any good multi-function, in stream tool calling frameworks beyond whats in the docsetc, i feel like there are prob better ways than how im currently doing it

[lilac tiger]

Hi, I’m a ChatGPT Plus subscriber and I’ve been unable to upload any files (PDF, DOCX, TXT, images) for more than 3 days.

This happens on both desktop and mobile versions.
I already submitted a support ticket but only received an automated reply, with no further response.
This issue is seriously affecting my academic and work-related tasks.

Could someone from the team please escalate this issue or provide any updates?

Thank you very much.

[stoic cargo]

Hello,

I was a GPT Pro subscriber from the 31st of March to the 31st of April. I canceled the subscription preemptively, because I just wanted to test it for a month.

Now that I am a free user again, I can not find the invoice anymore. It used to be under "Settings"->"Subscriptions"->"Payments". But the "Subscriptions" tab isn't there anymore.

And when I try to go through my emails and click on the "change plans" thing, it only says that I have an open payment (since I tried to get Pro again yesterday, but my team lead is currently off work for the week and he would need to accept/confirm/autherize the credit card transaction).

Is there any way to get to the old invoice?

[stoic cargo]

Ah, I found it. If anyone else has the same problem, I was able to find it at the least likely point I would've guessed. The help.openai.com AI Chatbot..

[amber plover]

Yeah, I tried 2 days later, suddenly it worked. No clue what the issue was.

[shell tendon]

Hey guys, have a small question

If i feed in lower-res images to ChatGPT, does it reduce my token usage or no?

[candid tartan]

Hello, I keep getting this error frequently mid-prompt reponses. Any ideas? It's very unhelpful:
"A network error occurred. Please check your connection and try again. If this issue persists please contact us through our help center at help.openai.com."

I have to keep retrying and retrying until it finally outputs a full response maybe.

[vital cairn]

imagine paying 20$ for chatgpt and got your money wasted real quick

i swear free gpt o4 mini is better than paid gpt o4 mini

[fervent geyser]

why does it generate 10 different responses when I only invoke the responses API call once