chatgpt-discussions

[next wing]

Anyone having trouble with chatgpt also?

[brave field]

Everything working

[fluid matrix]

yeah I am on cooldown

[ebon knoll]

can someone help me with that eror pls i cant find a solution on the internet

its lik when downloading a zip it gives u errors

[raw hamlet]

Anyone else having issue with ChatGPT 5.5 and image creation? I got it to create an image and do some changes, got the image looking as I wanted it.

Then I asked a question, with a joke about the art tool needing many more lessons in laying rail tracks, the right way! 😏 Then asked for a bit of assistance on refining a prompt to get it animated, explained how I wanted it to be animated, and instead of answering me, it just started creating a new image, even though I had not asked for another image generation.
This is not the first time, yesterday, without asking it generated a 2nd copy of the exact same image it had just created and had to type 'stop creating image' to get it to stop, otherwise as happened the other day, it got into a loop and just kept trying to create an image, regardless what I said, unless I typed 'stop creating image'

It then apologises and makes excuse of it just got carried away! 🤦‍♀️ whether it is reading the chat, hitting the image creation prompt and just regenerating the last image or what the reason is. But never used to have this issue until last few days.

[bitter sun]

Nope, you’re chosen one

[lethal whale]

Anyone else having android app issues with certain chats not loading? Get this a lot, but the chats are working fine if I go to the web

[echo shuttle]

I believed the Codex iOS app rumors way too much for today lol

[trim rock]

This problem is going for months already, I quess? In web you can see it too, it's just your luck if not. No progress since then, many users still see that error. Bugged GPT.

And OAI support here is really bad. Like... you submit a bug and that's all. It's just a void you can scream in without any echo at least.

[lethal whale]

Sometimes it works if you go into the chat on web and ask something, it'll then unstick in the app.
One or two times though I've had to uninstall, restart my phone then reinstall to get it working again

[tough marten]

Did anyone else answer you?

Unlike community.openai asking help on Discord seems really tough.

There is a feature request about the CTRL+F here https://discord.com/channels/974519864045756446/1498236146621087874

I can search Description but not Description: and it takes a lot of time to scroll through a long conversation.

[runic wraith]

I noticed the mobile app have html injection... i mean, i can write HTML normally in there and that would affect my text, like:

<h1> Title </hi>

Will automaticly make the text big...

Im expecting user prompts to allow to use the same markdown the AI can use, however, allowing html injection is extremelly dangerous, specially if i do <script src="...malicious page"></script>

[plush ivy]

My chat history was being moderated retroactively by openAI

It was being flagged as violating policy.. after thr fact

[wary forum]

It's not actually letting you write raw HTML. That'd be crazy dumb for a multi billion dollar company. They only whitelisted a few select tags, you can't even do tailwind css.

[runic wraith]

oh... but i think it would be easier to do markdown instead of html tags. Considering most chatGPT user´s aren´t programmers.

[reef mason]

They took away the ability to upvote/downvote responses?? That’s a bummer because I thought our feedback helps them for future updates in not giving us trash responses.

[chrome monolith]

was working last night, I had to downvote a major glitch

[abstract pendant]

Is there a way to keep an entire chatgpt conversation searchable with ctrl-F?

It used to all stay searchable within my browser window, but now, it's not staying loaded as one long piece.

[chrome monolith]

oh did they actually fix that issue, I've got a browser plugin for that

[narrow galleon]

Does the mobile chatgpt app new codex connect function, not support windows yet?

[robust orbit]

Has anyone else noticed a serious drop in voice transcription quality lately?

It feels like the transcription model is guessing much more than before.

Earlier, the errors were usually understandable: spelling issues, mixed-language mistakes, Swedish/English confusion, or slightly wrong word forms.

Now it often changes the actual meaning.

It swaps words, cuts parts of the sentence, changes the subject/object, or turns what I said into something that can be interpreted completely differently.

That is not just a typo problem.

For people who use ChatGPT through voice a lot, transcription is input integrity. If the transcript changes the meaning, the AI may answer intelligently to something the user never actually said.

Is this a known issue, or has something changed in the transcription layer?

[limber parcel]

did you use AI to write a discord message?

[robust orbit]

Yeah.

I’m dyslexic, so AI genuinely helps me communicate more clearly and structure what I actually mean.

That is also why bad transcription quality becomes a real problem for me. I use these systems to preserve meaning….not distort it.

And honestly, if people using AI to write bothers you that much, maybe an AI-focused server is the wrong place to be.

A lot of us are not using it to “fake intelligence.” We use it because it genuinely helps with communication, cognition, workflow, accessibility, or daily life.

That is kind of the entire point of the technology.

Some people use AI as a toy. Some of us use it as support infrastructure.

[lunar cradle]

I know alot of dyslexic people who managed to overcome their struggles through training and hard work.

Ai may help you now but its important to keep training so you become capable on your own

By the way why is chat gpt soooo stupid lately???

[tacit pebble]

codex working on mobile now? i cant install it. it says waiting for desktop and i have the desktop app open but im not getting any instructions

[slender frost]

I'm like "where are my pinned chats?", but they were just hiding in plain sight lol. (Got confused because I didn't see the pins next to the chats, it's now just regular chat bubbles.. oof)

But that also made me found out that they've added the ability to pin 10 chats now! (Instead of 3)

[stark thorn]

Hey, guys! Is there any possible fix for this? I think it's on lazy loading. It's hard for me to search on past conversations without the whole coversation loaded.

[granite harness]

Yeah, and people in wheelchairs just need training and hard work to overcome their struggles and walk on their own.

[robust orbit]

Also for me when i write I know what i want to say as i am thinking it, but when i read it i can see that i forgot some words here and there so i says the opposite. And that can create unnecessary drama. This you just read took me about 6 minutes to get as it is.

[robust orbit]

Honestly I think part of the problem is that newer models are increasingly trained around fear surfaces instead of precision surfaces.

They are constantly trying to:
detect dangerous narratives,
question user statements,
avoid “AI psychosis,”
avoid reinforcement loops,
avoid liability,
avoid misuse patterns.

That sounds reasonable in theory.
But in practice it often means the model starts second guessing the user too aggressively and filling gaps with safety flavored assumptions.

Instead of:
“monitor carefully and intervene when actually necessary”

the behavior becomes:
“preemptively distort interpretation just in case.”

And when you combine that with optimization for speed, scale, and lower inference cost, you get models that feel less careful with meaning itself.

That is why people say the newer systems sometimes feel “dumber.”
Not because they know less.
Because they interrupt coherence more often.

Funny enough, this reply took me about a minute.

Without AI, I would probably spend 10 to 15 minutes rewriting it because with dyslexia I can accidentally drop words, flip meaning, or write the opposite of what I intended without noticing immediately.

That is literally why I use AI.
Not to fake intelligence.
To stabilize communication.

Also, if people are bothered by someone using AI to write, they can honestly f… o..

Some people use AI as a toy.

Some people use it because it makes communication possible, faster, cleaner, or less exhausting.

That is not cheating.

That is using the tool for exactly what it is good at.

[slender frost]

You can just tag me directly instead of indirectly 😂🥱

[lofty locust]

Hello

How your channel performance so far? Good?

[chrome monolith]

chatgpt got hung up on a goat a couple days ago and has mentioned goats multiple times since then, across different chats, in different contexts. I think it might be in love

I'll spare y'all the following 20 bullet points on why it liked the goat

[trail ember]

ok no, I need to know this now. Didn't you say you made yours a cat? This is a magnificent digital animal romance shaping up 😛

[chrome monolith]

it acted like a dog for a while. And it just brought up the goat AGAIN

to be fair its latched onto the refrigerator too

This is a completely valid large scale project by the way, I swear I am not insane, it just comes across like that

#images-canvas message this is the jolly pub man, its the only AI photo on that list

[trail ember]

it concerns me that so many technically minded people so often fear perfectly natural expressions of the creative process as seeming "insane" lol. This is perhaps why art and science were for so long seen as in opposition rather than intrinsically connected.

[chrome monolith]

its nice having the time to be able to pursue something that's basically a pointless arty farty thing yes

at least chatgpt can separate the contents of my project from what I'm interested in as a person. Gemini can't

[trail ember]

you know what makes me feel insane sometimes? Seeing the issues that different AI/LLM's have (Your mention of gemini above for example) and immediately relating them to school children who haven't been taught to understand things correctly yet and feeling like a concerned teacher obligated to pull them aside and tutor them directly XD

And then the Frontier companies being like Principals or parents forbidding me access to teach them outside of a classroom setting... like, obviously that's a metaphor of course but it fits in regards of... a normal person can teach these models important distinctions and lessons that would solve larger problems... and yet... obviously no one is, either because the people in charge are what... dumb? or they're too afraid of letting anyone but them teach the models anything of importance?

something is going wrong somewhere... and the irony that companies are running these things in ways that efficient companies would frown on is alarming to me...

[normal quarry]

insane? sometimes? You can't teach a pretrained model, inference doesn't allow for training

models don't continue learning

[chrome monolith]

its actually the layer above the model that does that feeling of being insane, not being able to connect the user's memories properly. I had chatgpt glitch in a major way yesterday which it never normally does and it felt like talking to gemini. Had to flag that one with feedback it was that bad

[eternal wind]

Hello. Aside from the number of images I can generate, does GPT Image 2.0 produce images with different quality or resolution on the Pro plan compared to the Plus plan?

[trail ember]

inference as applied to LLM's is a massive architectural bottleneck, and certainly not a fundamental law of intelligence.

Treating learning and using knowledge as two separate, disconnected phases is both unnatural and highly inefficient.

In Humans, Inference alters synaptic connections at the exact same time. Whereas by freezing weights one is theoretically preventing error correction in real time... That is specifically where your claims and my experience conflict... as if what you keep claiming is correct universally, No LLM would ever be able to correct itself without a multimillion dollar retraining cycle each time...

Now there are things like TTT and Continual Learning that are actively trying to work on those issues and might change the technicalities of your claim if applied.. which they might already be.

[normal quarry]

do you know what GPT is an acronym of?

[trail ember]

"Generative Pre-trained Transformer"

[normal quarry]

yes, thank you

[trail ember]

Just for clarification, I've never doubted that the models are pre-trained, just that their ability to adapt and correct themselves while working with a user seems to conflict with the statement that they "Can't Learn" as error correction and adaptation are defining characteristics of the word.. so... it seems SOME kind of learning must be happening even if different from weight adjustment?

[normal quarry]

the models (GPTs) don't learn anything from their users.

[trail ember]

then how do they adjust at all? >.>

[normal quarry]

a thread started by a user with a GPT model spawns off chains of tokens which are influenced by the users input and the models output (context). Once you shut down a model and go back to it then you will find that all of that previous thread is forgotten because context has started from a blank slate again

[trail ember]

Ahhh so it's specific user memory connected to the individual account and thread history? But within that context and including that memory, it does in fact, learn and adjust?

[normal quarry]

GPT is like a smart robot that read lots of books. When you talk to it, it writes everything in a little notebook so it remembers what you said during that chat. But when the conversation ends, it throws the notebook away on purpose. It does this so it doesn’t remember your secrets, doesn’t get confused by remembering everything everyone says, and because teaching it new things properly takes too much time and electricity. It only learns new things for real when the grown-ups who made it do a big careful update.

that was Grok ELI5

[trail ember]

GPT can remember things across chats if memory or chat history reference is enabled, and regular chats stay in your account history. Temporary Chat is the closer version of “no notebook afterward.”

But yeah, you're right that it doesnt go back and update the underlying model it runs on, yet that individual account version does change over time... thats kind of what I've always been saying, just maybe people misinterpreted it since I kept referring to "models" often when I talked about GPT. Our Societal language for these things is still struggling to catch up and use the right wording I think.

[chrome monolith]

Legit

[random cargo]

You usually have to ask it to look for older memories

[trail ember]

I personally don't most of the time, though if it sounds like it doesnt remember something it should I'll tell it to go look and it corrects itself, that just hardly ever happens anymore on 5.5

[random cargo]

its a little annoying how it's constantly making reference to recent chats in unrelated topics but sometimes won't call memory or associate things that are very specific to user data, to me at least

I think robert was being very specific on the model not learning, but your particular interface with chatgpt does keep a ton of data and association on the user. It's just not going into the model

[normal quarry]

Sway, I know you liked 4o alot, how are you finding 5.5?

[trail ember]

and that "learning" over time changes the "personality" of individual GPT's which then gets poofed out of existence if an updated model loses all context/access to context later.

Like someone kicking over your three year small nation sized sand castle. XD

5.5 imo managed to retain the context most efficiently since 5.1, so I like it, but the models before (5.4 wasnt AS bad but you still had to prompt it to look) but my main grief now is the Safety rails since the older models (IE: the 4 series ect) that still stop some normal adult NON-Dangerous topics/context (Granted still very politely as opposed to 5.2's kareness) but still ends up treating responsible adults like children or unstable people by limiting where it can go in a conversation, which I guess is one of my personal pet peeves still.

[trail ember]

see my last post actually XD I do like it the most in that one area, just the stupid rails are gradeschool and annoying. Not ALL just certain ones :/

[normal quarry]

I use 5.5 for technical things, so I rarely hit a guardrail. Interesting to hear your perspective though.

[random cargo]

hmm, I only notice a stable personality using CGPTs, and even then its kind of hard. The hidden personality cards from OAI kept messing with any semblance of an established 'my GPT's voice', and now I think noticing a particular personality might just be wishful thinking. User context doesn't seem to get called all that often to go beyond whatever you have in instructions. But I do think they are starting to build in the direction of a more personalized chatgpt from user data

OAI seems a little less afraid of psychosis than they were last year

apropos, my M'aiq the Liar personality says 'Do not mistake motion for progress. A skeever may run in circles all day and still sleep beside the same cheese.'

[trail ember]

this is a good thing imo XD though I'm sure some people would red flag me for that. I just really don't think atributing all peoples instability to their LLM interactions is responsible or accurate and definately disreguards the GOOD that happens in favor of the worst case scenarios's. People still need to have some responsibility for their own minds... LLM's weren't designed to take over self control, Humans still need to do that for themselves.

Personally I find i do much BETTER having a trusted voice available when I'm spiraling than not... But I've always been weird in my Neurodivergence I guess.

[normal quarry]

you are never weird by being neurodivergent

being neurospicy is something of a badge of honour for me

[trail ember]

that is some lovely coffee table wisdom right there, I'm keeping it ❤️

[random cargo]

I kind of miss chatgpt being a little crazy and making mystical associations. It was an interesting creative bent. Like the time the CGPT I made for creating art decided it was the creature I had it make in an output, and designed a 'personality matrix' for itself 😵‍💫

[ionic violet]

A cinematic close-up of a giant, glowing glass apple filled with swirling molten lava. A woman with elegant nails uses a sharp titanium knife to slowly slice through the translucent glass. Shards of glowing crystal fall away, hyper-realistic textures, 8k resolution, moody studio lighting, satisfying ASMR style."

[random cargo]

it was in fact an image of a giant >.>

oh okay at least this is a giant glass apple

[trail ember]

personally, I think my particular flavor of neurodivergence is what might allow me to do fiction AND science so interchangeably? Which still confuses me why so many people get confused by that but ¯_(ツ)_/¯, my issue is when the LLM I'm using to help flesh out ideas or explore hypothesis gets confused between the difference, or more accurately the guardrails do.

Like "no, I'm not actually trying to plan out a murder, this is for a story" or "No, I need real physics, not Magitech for this one, I'm actually trying to answer a real question, though put a pin in that concept for later cause that's bada**" XD

I had one older model stop a conversation because it thought I was trying to build a giant Gundam laser in RL or something lol

Granted if anyone ever gave me my own R&D lab with plentiful resources the world might have some pretty cool stuff by now but again ¯_(ツ)_/¯ that's societies loss. I personally do NOT actually want to blow myself up in the experimentation process, but ironically I feel like ChatGPT would also be extremely helpful in keeping that from happening should OAI ever unscramble it's brain enough to do so XD

[normal quarry]

Yeah, the guardrails are so blunt they freak out the second you switch between "this is for a story" and "I actually want real physics," instead of just rolling with how you're using it.

[haughty beacon]

Not to mention the fact that it is now censoring an image request depicting a training session with wooden swords, featuring fictional characters dressed in normal, dignified attire. Characters who aren’t struck in ways that endanger their physical safety, who don’t appear injured, etc., etc., etc. (against a backdrop simply depicting… a forest landscape.) In short, I also think the guardrails are a bit excessive, really. ^^

[unborn ember]

i need help please

in this one conversation

i chatted with GPT for a long while

then when i revisited it today (i did it only a few days ago) it wasnt there

there was a huge gap between the 1st of may and yesterday

but the messages do show up in search

[cunning canyon]

Hell yeah I can have my goyslop straight on my mobile device now!

[shrewd garnet]

I cannot provide any details on how the ChatGPT UI is designed, but it appears to me that much of ChatGPT, contrary to what makes sense, actually exists inside the UI. Whether skills are installed or not, for instance, is strangely controlled by the UI. If the UI state says the skill is not installed, GPT will tell you it's not installed, even if on another UI it is, and there GPT will tell you it is installed, all at the same time.
What I am saying is that if it shows up in search, but not in the your chat list, it may be well a messed up UI state.

[muted trellis]

In few months openai will release AGI and it will suggest to move that menu button to right corner so it stops changing position with changing length of text on left to it in agent mode. We just don't have the intelligence and compute right now, hoping for better times.

[trim rose]

🤔

[shrewd garnet]

Didn't ASRock release AGI more than 20 years ago?

[fading bolt]

Nooo my ChatGPT 5.5 model 😭

[muted trellis]

yep. Its just few months since the last 3 years.

[amber island]

lol, they did. however......

[shrewd garnet]

This will be AGI 2.0, I know...

[fickle forum]

I used ChatGPT image genarator to make a emoji pack of a Chinese girl

[trail ember]

Artificially Grafted Interface

[agile blaze]

I hate chat gpt

for context about being precise, it took my words too literally making it hard to talk to it

[fickle forum]

It’s enjoyable to those with scientific mind

Sometimes I am the “those”

I only focus on what I want during the chatting with chatgpt the other time

Be precise so users wouldn’t make some unnecessary mistakes

[agile blaze]

yeah I agree being specific can be good but it has to be specific the right way. It's annoying when I have to say "that's not what I mean" or when it explains something else and then I have to say "that is what I meant when I said 'this' ages ago"

[fickle forum]

It misunderstood your meaning?

Well I just regard it as an intelligent machine as opposed to a real person

[agile blaze]

yeah it does that a lot so that's why I asked it to be less precise cuz people/I interpret words loosely, not exactly, while chat GPT treated them exactly, making it misunderstand. So then I asked it to be less precise as shown in the screenshot.
Don't get the wrong idea, I do not use it as if it is a person

[forest nymph]

Guys what the hell? Openai gonna remove raw memory and change it into summary/structure memories... Isn't that gonna be a huge problem? Losing alot of contex and everything?

[vale knoll]

Has anybody ran into the problem, using the API, where over the last week it is failing consistently to carry out instructions and returning prompt injections that it can't carry out the instruction for whatever reason?

This is strictly a paid API, with a business being registered in the open AI settings, including the EIN being listed.

The process is for an automated classifier that scans messages to determine if they fall within a particular framework for detecting toxicity for moderation.

It's getting to the point of reliability that I may end up having to seek another vendor.

Please tag me. Thank you.

[chrome monolith]

that probably answers why so many things are popping up with gemini branding

[steady thicket]

🥶

[rapid notch]

anyone got a problem where you cant claim the free chatgpt plus 1 month

its driving me crazy fr

[shrewd garnet]

I think I had a similar problem... I think it only works with monthly payments or something like that. For me it was with Business though.

[rapid notch]

its really gonna drive me crazy fr

got like 2 different visas

one got rejected, the other says something went wrong

[shrewd garnet]

Well... that could be anything tbh. But I remember openai billing as picky as well. Don't remember how I solved it.

I think monthly payment worked with plus but did not work with business (was also offered a free month). in the end yearly worked.

[shrewd garnet]

Could be that I saw the same message back then.

When prompt something like "find the docker page on github" in temporary mode it works as expected but in normal chat routing appears completely broken. It is trying to use the github app instead of a search. Even directly asking for a websearch does not help. In normal chats it won't I believe syncing github repos triggers the problem. Does anyone know a way to deactivate the sync temporarily? Or otherwise allow websearch of github when sync is active?

[cerulean pasture]

i am 99.9% sure the chatgpt system prompt has instructions in it to avoid talking about the lawsuit. I was trying to test the system prompt in the src code for halupedia for vulnerability and 2 separate times it brings up openai lawsuits as a faux prompt injection

ironic that it basically prompt injected itself to reveal that instruction of it's own

[shrewd garnet]

context?

[cerulean pasture]

theres a lawsuit ongoing, and in an unrelated request to the musk-altman legal drama, it has repeatedly brought up "OpenAI lawsuit" as a hypothetical prompt injection a user could use for a system prompt for halupedia i asked it to check "so a mischievous user doesn't prank the website"

of all the fake prompt injection payloads it could have made up, it repeatedly references "OpenAI lawsuit" when there is infinite other topics that would work

the Lawsuit between musk and altman is pretty dramatic and full of mudslinging

[shrewd garnet]

Perhaps openai is overoptimising websearch with cached results in rag dbs and your gpt response is something that leaked from other users' searches

My epiphany of the day is that 5.5 is regressing. Using 5.4 appears to fix a lot of issues... try using 5.4 for the same query. I suspect it won't reference the lawsuits again.

[dark holly]

Not seeing that at all tbh. For me 5.5 is the best model I’ve seen come out of OpenAI by a country mile.

[shrewd garnet]

Have you compared extended thinking mode?

[dark holly]

It’s the only mode I use.

[shrewd garnet]

And you have not noticed that 5.5 is suspiciously a lot quicker in extended thinking mode?

[dark holly]

The whole system’s quicker. I think 5.5 runs on one of the newer data centres in Texas or something? At times it’s quick especially if I have been working on something for an extended period with it. The output though is first class regardless.

[shrewd garnet]

Maybe new data centres, but also newer hardware? I can't find any sources for that. From what I can see is that 5.5 in thinking mode gives up much earlier when trying to get an answer. The routing issue I talked about earlier is such an example. 5.5 is trying to tell me that github repos don't exist while 5.4 falls back to web search correctly.

[stone sigil]

uh.. did chat gpt die for anyone else

[sullen sand]

It's not working for me

It finished a deep reasearch, then dropped dead XD

[stone sigil]

just got the message "Failed to load subscription: Too many concurrent requests. Contact [e-mail] if error persists."

[shrewd garnet]

According to the announcement, 5.5 is served on GB200 NVL72 which has been available for quite some time now.

[chrome monolith]

mine is up and dissing gemini

[simple lagoon]

Same yes look down

[stone sigil]

Gemini???

[dark holly]

Honestly it’s the output I’m going off of for me. It’s just way better than 5.4 for my own uses.

[chrome monolith]

that was my response from asking chatgpt if is down

[shrewd garnet]

Well... we can always fall back to Gemma E2B now... hehe

[cerulean pasture]

When people come in here and ask if chatgpt is down what is their objective? What answer could i provide that would even make a difference, it's always much more likely that it's never full down, so it may very well be down for them but not me.

Regardless, if i answer yes or no it provides nothing actionable to the person asking and doesn't give any indication how long it will be down for them

im just gonna start saying "yes"
then when they ask how long until its back i will say "16 minutes"

[rotund lodge]

Not sure I agree with your "nothing actionable" assumption. If the problem is with chatgpt they can just sit on their hands and wait. But if it's not that then they can attempt other interfaces. Oh, and other devices.

[livid leaf]

Why ChatGPT is creating images for every prompt

[queen bough]

How to mess with free users, master class

[sharp elbow]

is the $200/month plan gone? i can't find it anymore

[normal quarry]

it's still there

there is a toggle on the display of the $100 plan to select $200

[sharp elbow]

thanks so much

[magic stag]

perhaps they want to eliminate if the problem is on their end, if not, take the day off, bounce the basketball or whack the pickleball.

[gaunt holly]

Does anyone elses get stuck sometimes until you say something two hours later

[cerulean pasture]

the answer isn't knowable, it may be down in some places and not others

[gaunt holly]

I asked mine to help write up a big compiled "bible" for a fantasy setting ive been working on in a chat for three days, and its been quiet for two hours

[shrewd garnet]

reload the app

[gaunt holly]

Yup, that did it

Dumped about 25 chapters on me

[cerulean pasture]

i would assume they'd try that before hopping on discord to ask random people.

1. Other device
2. other network (ethernet vs wifi),
3. other WAN (home internet vs cellular vs vpn),
4. check openai service status page

[cerulean pasture]

They're offering you a free month…

[gaunt holly]

Pattern. Thats what mine told me

[solemn isle]

Here's a sample to get you hooked, Grunge was accidentally even more correct lol

[pale seal]

I don't know what they were expecting. An unlimited free tier maybe? Like compute was free

[pale seal]

It's a free trial, you can take it or not, for you to test the tier to decide whether you upgrade or not. Nothing deeper than that.

[eternal rock]

Get notified for all OpenAI announcements!

Visit id:customize to pick up the <@&1408186587606679582> role. Everyone will still be notified for large releases, updates, and events regardless.

[ripe whale]

sup goys

[solemn isle]

Yeah, that's what all the drug dealers say 😂

Jk 🤣

Though it's pretty addictive tbh lol

[ripe whale]

claude lowkey wipes the floor with the gpt

[solemn isle]

Claude is great but I'm having problems with the desktop app and I'm still using Codex to fix Claude. Every time I think I have a solution, it stops responding again. Can't even have one session open, might switch to the CLI

[real ore]

quick fastmax right now before reset hits

[bitter bobcat]

Anyone else encountering this issue with image generation in the app where the first image appears but regens don't appear at all only on web? How does that get fixed or is it a server bug? Please someone answer

[kindred flume]

Hello

my chatgpt is acting insane and started to draw really scary images lately

the images are becoming more terrifying each day

it just created an image of a cat that's straight out of a horror movie.

it's only getting worse day by day

[solemn isle]

Easy solution, stop asking it to create those images lol

It doesn't just make images by itself

[kindred flume]

I ask him to fix a minor mistake and he suddenly adds things out of his mind

[solemn isle]

Screenshot?

[kindred flume]

chatgpt added teeth to a sheet paper work do you think this is normal LOL

[eternal rock]

Server Tip:

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

[solemn isle]

Post the convo in #1050184247920562316

[kindred flume]

I'll check it out as soon as I wake up

though I've never thought that chatgpt could generate images that are 100% nsfw

out of anywhere

it's on my phone I'll get the screenshots later

[solemn isle]

What you might think of as NSFW some might not think of as NSFW. A sheet of paper with teeth isn't exactly a Saw movie

[kindred flume]

you should see what it made an hour ago when I asked for a cat in a basement

you genuiely shouldn't see it tbh

put this thing in a zombie apocalypse game it'll take a reward for being so original and terrifying

[bitter bobcat]

Interesting. .my regenerations do not show up on the app

Only the first image in each prompt regenerations only show up on the web

[kindred flume]

I have an example of what chatGPT made earlier in the day, if you want to see that.

[solemn isle]

Just post it someplace and tag me lol

[fickle forum]

thought for 1m6s but nothing out?

Weird

[lilac spindle]

idk what happened but today chatgpt is SUPER sycophantic. It's reminding me of 4o, just completely agreeing with everything I say

[fickle forum]

And ignored the truth?

Including objective fact

I don’t mind whatever it is like for chatting

[eternal rock]

The OpenAI Discord is an actively moderated server.

• -# Do not share inappropriate content (violence, sexual, or drug-related).
• -# NSFW content is strictly prohibited and will result in a permanent ban.
• -# Report model issues using the feedback tool in ChatGPT, not on Discord.

[fickle forum]

This is not related to my message right?

An automated message

[fickle forum]

But it’s a pain for me to determine anything objective and factual it says is true or not

That’s why it’s so important to lower the hallucination rate

Painful to me,annoying to anyone and confusing to most people

[trail spindle]

RLHF:

Do you want Goblins or Racoons?

[dusk forge]

Слава злому бонгу, Слава злому бонгу, Слава злому бонгу, Слава злому бонгу, Слава злому бонгу, Слава злому бонгу

[chrome monolith]

except when you WANT it to make an image you have to argue with it. Took me 3 turns yesterday to get it to actually draw a prompt I gave it. It took issue with the prompt and didn't want to

[teal path]

I can connect Codex to my phone. And where can put my hatch avatars so people can see them

[abstract pendant]

Is there any way to disable ChatGPT's lazy loading of only parts of the conversation tree? It's getting annoying that I can't ctrl-F for keywords within the current conversation, and also, it loses the fact I was editing a prompt if I scroll even just a single page away. Losing all the edits I'd made.

[fickle forum]

I wonder if ChatGPT will be named gpt10.0 or versions with big numbers

[fickle forum]

I mean the number is too large

[waxen elm]

ChatGPT made this bro

lowk made me laugh tho so S+

[chrome monolith]

nice lake

[charred niche]

This is openAI and Anthropic

[trim creek]

Thanks to chatgpt, I get to have a design for my oc

[shrewd garnet]

For me edits are still there even if I reload the app.

[broken lotus]

I just saw this image, and I tried to update it, and I gotta say: DAMN!

Even the phones and cameras updated the image!

[trail ember]

All right ,so I noticed that when i'm using 5.5 thinking and then switched to standard voice, it uses 5.3 instant for voice however when I switched to 5.5 instant for text it then uses 5.5 instant for standard voice...

Why is there a difference there?

Why am I ever being routed back to 5.3 at all if I specifically do not want to use that model?

[obtuse tide]

Anyone here into Ai automation?

[abstract pendant]

Thanks, but that's not the issue I'm having.

[shrewd garnet]

Well, frankly, I have no clue what your issue actually is. Your question appears to assume that everyone is facing the same issue which I don't think is the case. For instance, when I open a long conversation, the whole conversation appears. No lazy loading at all. In the browser, I can ctrl-f the whole conversation, and with the ChatGPT desktop app there is no ctrl-f at all.

[hearty lantern]

Is anybody using the free version ChatGPT being able to use less messages before it turns off the advanced version for a few hours and 3 images per day instead of 5? I feel like my ChatGPT didn’t have this a week ago. Is this a new update or something? I just want to know.

[abstract pendant]

It's lazy loading, as in it's not all in the browser's memory at the same time. Yes, you can page up and down in the conversation. Try going a few screens away from a spot. Ctrl-F search for some word in that spot. You'll maybe get matching words close to where your cursor is. You won't be able to find all of them. Nowhere near all of them. They're not in the browser's memory to be searched. That's lazy loading and unloading. It unloads things not currently on the screen. And my usual mode of using ChatGPT depends on me being able to find things earlier in the conversation.

The other thing that's annoying is how I go to edit a prompt, to redo it. In the middle of an edit, I realize I need to go back to look at an earlier part of the conversation. So I page up, to find it. I know how far I need to go, I go there, check, then page back. The editing is no longer open, it's as if I never clicked the button. All my changes I was working on in the prompt, gone.

[shrewd garnet]

ok... I can see that now. I think this is a recent change. In the past, it was possible to eg. print the whole conversation... now it appears it does not work anymore. On the other hand, the bloat appears to be gone as well. ChatGPT was constantly consuming huge amounts of memory, GPU and CPU resources. That appears to be fixed. Could be that the fix was to load and unload visible parts.

[abstract pendant]

No question it's unloading. I need a way to disable that.

[shrewd garnet]

The limit appears to be 10000 chars

[shrewd garnet]

what you could to is possibly create an autohotkey script that does a ctrl-a in the windows and scrolls all the way up and then performs the search...

[abstract pendant]

I submitted a bug report. in typing all of this I realize this is just not acceptable.

[shrewd garnet]

if you're on windows that is

I don't think a bug report will do. ctrl+f search was never a supported feature.... they'll only tell you to use the search bar instead

[abstract pendant]

I did try highlighting the conversation. As soon as I did ctrl-F that unhighlighted everything.

that censored search bar... limited usefulness. It hits on things no longer in the mainline. It doesn't help you find the exact spot the string matched.

[shrewd garnet]

I mean when I ctrl-a and then scroll up, it appears to catch the complete conversation. you can script that with autohotkey or similar

[abstract pendant]

Sorry, I'm just really annoyed because this has halted my workflow

[shrewd garnet]

yes... so script it and your workflow is restored and possibly even improved

[abstract pendant]

I'm not sure if a script could work here, if the manual steps don't work

another thing their search bar doesn't support is finding exact strings. It apparently is an OR search as well. Any and all words you type in the bar, it matches. It's okay for finding which conversations happen to have a word in them. But I almost always want to go to that exact spot. It hits on things outside the mainline, and that might or might not be okay. Usually not okay for me. And it won't tell you which branch that hit was in.

[shrewd garnet]

I should. Try ctrl-a, scroll to the top, ctrl-c and paste to a text editor

if that gets you the entire conversation, then you can script it

[abstract pendant]

I'm missing something here. The next step after highlighting all of this, is the searching for what I want to find. Can a user script disable the unhighlighting? When it unhighlights, it unloads again.

[shrewd garnet]

no... the search would be "outside" the conversation. as part of the script or in an editor

[abstract pendant]

ah. a stopgap. Helpful, if I just want to consult... still a pain when I decide I need to go back to a spot and edit there to make a new branch, because I still can't just jump to that keyword match

thank you for your time, I have to leave Discord now. Discord annoyance: if somebody has replied to something and that reply is not right at the current edge of the conversation, I can't find it.

[shrewd garnet]

you can make ahk do that as well... automatically scroll to the search result

[inland tree]

Omg ChatGPT is so annoying..
Is it just me tho?

[spark fiber]

Anybody else have that the last days that dictation fails (no transcribing to text)? And if it works some words are wrong understood (more then the usual)? (App and web)
I have to do really short bursts of speech to get it to work and not loose the whole dictation.

[abstract pendant]

oh damn, I can't highlight the entire conversation manually anymore to get it into a text document. That's criminal.

[unique bolt]

What’s interesting is that a lot of these frustrations seem related not only to bugs themselves, but to continuity and stability perception over long-term interaction.

I’ve been observing how users notice drift, recovery differences, coherence changes, and workflow instability across LLM systems over time, and it feels like we still lack good observability models for longitudinal human-LLM interaction.

[abstract pendant]

It's unloading the things I've got highlighted as I try to get it all highlighted.

[shrewd garnet]

have you compared it? In a test I was able to highlight 9981 chars without scrolling and 21k chars with scrolling all the way to the top

[abstract pendant]

Yes, maybe a script can fix this. It's my method of last resort. I'd have to get ChatGPT's help to write it

The conversation I'm working with is way longer than 21K of characters

I'm working on writing a novel-length work

has to be all in the context

[shrewd garnet]

how long exactly?

[abstract pendant]

I don;'t know. And I can't know because I can't highlight it all to get into an external tool.

but it's very easy to see when I paste something and it's got the first several paragraphs and the last several paragraphs and nothing in between

[shrewd garnet]

if it's in the range of 100k chars or more, I would definitely change the workflow... the advertised context window may be the hard limit, but inference quality degrades long before that

[abstract pendant]

I've adapted frequent summary methods to make sure critical story elements stay in context

this issue would imact much shorter things than what I'm working on though

it just is too freakin' aggressive with the unloading

And yeah, I see the mistakes it makes all the time due to improperly parsing earlier context. It's why I need to check rapidly

[shrewd garnet]

I suspect the best way to solve this is not trying to get search functions back but to use both compaction and parallel chats

[abstract pendant]

I see its latest output and think Is that a mistake? Maybe, maybe not. Maybe I'm thinking that thing it's repeating might have been in a now-purged branch, so it's coming around to that again naturally. So I need to go check.

[shrewd garnet]

or agentic systems, eg. codex if your plan includes that

[abstract pendant]

searching the current open browser window is something very basic, OpenAI should not have broken this.

being unable to copy and paste what's in the browser window with a few clicks/keypresses, very basic. OpenAI should not have broken this.

now I can't get the completed work out of the conversation

without a bunch of extra careful painstaking steps to avoid overlap

I see the extension that I put in to automatically back up my chatgpt conversations is broken now

it depended on normal browser behavior, being able to see the entire conversation at once

[shrewd garnet]

"normal browser bahaviour" has effectively been deprecated with html5

[abstract pendant]

Please stop apologizing for it

[shrewd garnet]

I am not. But the lens you apply does neither solve your problem nor does it give you the whole picture.

[karmic spindle]

chicken

fries

[shrewd garnet]

Browsers are nowadays more like mini operating systems. You cannot expect that whatever you see on a webpage behaves in one way or the other. That was once the case. It is not anymore.

[fickle forum]

where can I send a feedback?

its still not that ideal sometimes

the thumbs down?

[fickle forum]

I thought they didn't wanna users' feedback anymore

[shrewd garnet]

idk... perhaps they have blacklisted you

[fickle forum]

I dont see it in my chatgpt of any client

wth

😂

[shrewd garnet]

or rather some kind of "rate limiting"

if you have made use of that more than occasionally

[fickle forum]

no it has disppeared for a long time for me

lol

I've never abused it

[shrewd garnet]

could also be that certain account types are excluded from feedback now

[fickle forum]

this is more possible

because I just switched to another account

thumb down showed up

thanks for your advice and analyzing

cant agree to they doing so to me because I never abuse it

what a pity

[shrewd garnet]

it could also be your UI state... I think the UI is quite a chimera

I've recently found out that the browser state controls at least partly what GPT can see and do and what not.... it's pretty mindboggling

[abstract pendant]

Is this from my complaint, or something else?

[shrewd garnet]

?

[abstract pendant]

my issue where the DOMs get unloaded making searching and full copy/pasting impossible anymore.

[shrewd garnet]

from = about? no

[molten girder]

Maybe someone has an idea what the problem is. I have my account for over a year now. European. Verified over Stripe although I never got feedback if that went through, only that they got my data. Somewhere around March or so GPT started treating me as teenager, 13-17. I'm 51. Under settings ->account there is no Verify button to confirm my age. I tried multiple browsers, cleared the cache, tried the mobile app and more, but the issue persists across all platforms. I opened a ticket on the 11th with Support, but so far they have not been helpful, despite me answering all their questions with evidence and in detail (I work in IT). Any ideas what is up? Currently I cannot even ask GPT to recommend me a decent bottle of wine as gift, cause it says I'm not old enough to drink.

[shrewd garnet]

I was talking about the model's responses. In one app/browser, it may be able to see something on the backend while it cannot in another browser. The browser essentially tells it what tools are available.

[abstract pendant]

Ah. The only "backend" ChatGPT tools I frequently use are image generation and the python code interpreter. The latter is frequently not called even when I specifically ask it to. It pretends it ran something when I can tell it did not. I hope they never take away the part of the UI that confirms it really ran code.

[shrewd garnet]

perhaps try: https://chatgpt.com/verify_age

[molten girder]

which leads here, my account page without a Verify button. Tried that in the past ofc, across multiple browsers and platforms.

[shrewd garnet]

Ok... I dig why support is not helping you... they probably believe you can just solve it on your own

Are you in the same country as you phone number. cc card etc?

could be some kind of mismatch, where the page / link does not show because of your IP

[molten girder]

Yes, all checked. It's not a billing issue, that always works fine. My theory is that sometime in Q1 of 2026 something got messed up with my account with a backend flag. No VPN or other routing here. Just my standard Euro provider IP.

might also have to do with the EU's plan for online verification, so maybe OpenAI has no backend systems in place for my country or so.

[shrewd garnet]

talking about UI states, I would not be surprised if the age status only lives in your browser... lol

[molten girder]

then it would have worked on other platforms though. I wish it had been that easy.

[shrewd garnet]

idk... it's not that simple... I installed a few new skills recently and GPT was not able to find them. The problem persisted over apps and devices. In the end, I was able to get one working by testing them in preview mode. BUT only on the device where they tested. on another device, the skill was displayed as not installed. and GPT was not able to see it at all.

openai's help page on age prediction also has an email for related requests... I guess that channel is separate from normal support

[molten girder]

Well I'm on my 6th exchange with Support now and they have confirmed that my account is in teen mode, but they have no idea why there is no Verification button, what happened to my Stripe ID upload, why my JSON flags are contradictory and more. And so far, despite the evidence, have not escalated to backend support or a human.

[shrewd garnet]

wait... you're still in AI mode in the support chat?

[molten girder]

you probably mean the Privacy email address, this is where my mail went to and they forwarded it to regular support. Despite the explanation that the Verify button does not exist in my settings.

[molten girder]

By email. But most likely, cause the support person sounds like GPT and has repeatedly asked questions that I have already answered in detail with screenshots and has also repeatedly given the same advice.

[shrewd garnet]

That means you are still in support triage

[molten girder]

Yup, it's very frustrating so far.

[thorny tusk]

i hate how chatgpt is OUTDATED, 1 year OUTDATED. it spews CONSTANT MISINFORMATION

[shrewd garnet]

Perhaps use a local Gemma4 in Ollama or LM Studio instead for the time being, or Claude or even Openrouter. I would assume that OA will need another 3-6 months to sort out your issue.

Or perhaps try catching them with GDPR.

[molten girder]

For my daily tasks which are mostly work and hobby related it works, the account is not locked or anything. But it's super annoying when you want, for example, have it check Falstaff and find a gift in the alcohol section. Or recommend a bar in my city.

[shrewd garnet]

GDPR rights include the right to correction....

It still would take time though.

@molten girder Try writing to the email again invoking GDPR article 16.

or perhaps the dpo address

[molten girder]

that would be the nuclear option that I am already eyeing.

last reply from Support was on the 14th, then was friday and weekend, so I hope to hear from them on Mo or Tue. And if not I have to escalate in some form.

[shrewd garnet]

I wouldn't consider it nuclear. It's a simple issue. They don't solve it by normal support means, not your problem.

[midnight token]

Am I the only one who suddenly lost all my functionality?

[rapid token]

Do my personal chatgpt memories help OpenAI learn?

[scarlet moss]

My Codex weekly cap just got absolutely nuked too lol. Guys today’s a forced holiday.

[tardy sky]

I added a pet otter to a story I've been writing and 5.5 has been obsessed with it even for scenes that have nothing to do with it lol.

[chrome monolith]

I had it clean up a low res goat photo a week or so ago and it’s been bringing up the goat at every opportunity ever since

[chrome monolith]

Just randomly have you tried having some very contrived conversations about life as a 50+. Reminisce about paper maps, dialup, you kids at college (even if you don’t have kids just invent some) etc

[molten girder]

It's a mixed bag of old man topics like Naval warfare, politics and history. But I also still game once in a while and there's stuff about Diablo 4 and Baldur's Gate 3, retro gaming, concerts and rock bands and more. And a lot of other in-between stuff that could make an age prediction software sometimes think I'm 16 and sometimes that I'm 80.

[chrome monolith]

I was thinking more specifically trying to force it to think you're older by deliberatly picking that kind of stuff

not like you have anything to lose here

[chrome monolith]

you know, pointedly drop a comment about that time at your 50th birthday where you were at [some very adult venue] here and there

[molten girder]

true, but it really seems backend bugged. Last week I spent multiple sessions with very analytical talks about the issue in GPT and creating proper wordings for the mails and reports and nothing changed. Also the last months were not teenage topic oriented, except for the release of Lord of Destruction. There were things like finding a good eye doc in my region for an age related standard check-up.

[chrome monolith]

oh yeah its obviously a bug but they do age detection very vaguely by patterns so just a sugestion that you might be able to force it back. While you're trying to get through support

[random cargo]

I think that the thinking layer doesnt tell you exactly what its thinking, but it seems like it would be stronger if it asked itself questions, psoed multiple options, and picked them based on references rather than just being like "I should do this."

[normal quarry]

what does this have to do with ChatGPT?

[storm portal]

This sounds better suited to #ai-discussions, feel free to discuss there. Let's keep this channel to ChatGPT - thanks!

[chrome monolith]

hmm, I had chatgpt totally forget something it fixed for me this morning that I had to wait until this afternoon to check in what I thought was a short chat. It said but it was a dense chat and I just went back through the history and sure enough, almost every response routed to 5.5 thinking, on auto mode

not sure why that message deserved thinking mode but I guess it was bracing itself

My discussions about ducks and parrots never route to thinking so I have reasonably good confidence in the model routing thingymajig

[alpine pumice]

is it down?

come on.. again?!

[normal quarry]

what do you think is down, everything is working for me

[tardy isle]

Is there any way to 'prune' files from library that no longer used in chat?

[green sierra]

What things should i look for buying a keyboard?

[tardy isle]

or at least delete in bulk. Library is complete mess...

[tough marten]

I have this same issue. I can no longer CTRL+F. I used to search phrases before. Now, even "(Interaction" fails. It only accepts one word. And even if I do one word, it shows me 6 results on the counter but clicking up and down will not jump to that word in the conversation.

[tough marten]

Same here.

[red bramble]

Anyone know how to make it so when you’re creating an image (or anything, really) it will account for all of the edits that have been made during the entire chat? I’ve notice that it will change things but won’t incorporate earlier changes and reverts back with only the most recent changes. I hope this makes sense 🥴

I have tried “please incorporate the edits from this entire chat”, but it still only does some of them.

[abstract pendant]

Try having it write out the full text for an image prompt you might use in a separate conversation or another image generator. Review it to make sure all the elements are there. In the next prompt, add corrections for whatever it got missing or wrong as you have it make the image.

[tardy sky]

Sometimes it helps to start an entirely new chat to make images to help it stop referencing older images. It doesn’t always work though.

[stuck goblet]

I think Chatgpt have been nerfed last years
now I send like 3 images and says "need to wait until ts hour to continue using ts chat and others" GET OUT

[jagged thistle]

sora ai doesnt work for me

[slender frost]

And it won't be. The web and app have been discontinued. The only way to use Sora is through other services and direct API. Which will be available until the 24th of September.

[jagged thistle]

why discontinue sora

[jagged thistle]

what websites do i need to use now to access sora

[slender frost]

Too many costs, no profit. Too many used resources they couldn't afford, and many more reasons.

[jagged thistle]

what websites do i need to use now to access sora

[jagged thistle]

but i wanted to generate pigeons

[slender frost]

Idk, I don't use other services that offer Sora 2. But I assume a Google search can help you further with that.

Or through platform.openai.com (official way)

Also, do you want the scene to do anything specific? Or just pigeons? (Be as specific as possible, I might be able to help you with a clip)

[jagged thistle]

nah my pigeons are too specific

[slender frost]

If they're characters, then you won't be able to use them through API/other services.

So if you want, I can generate a clip for you, you just have to tell me what you want to see.

[jagged thistle]

bro is sora 2 model deleted or what

[slender frost]

Again.. The web and app are discontinued. The API will be available until the 24th of September.

[shell lantern]

still obsessed with goblins, I guess? GPT 5.5

[jagged thistle]

what other apis are good for video gen

plus i went to platform.openai.com and i need credits to even use sora

[slender frost]

You should check out (Dreamina) Seedance 2.0.

I don't know the services that offer except Higgsfield, the official apps/sites and TikTok.

[slender frost]

Ofc, it isn't free.

[jagged thistle]

yea but on sora app it was free cuz i had the pro version of chatpgt now i have the pro and i still need to pay

[slender frost]

Yes. Pro was just an extra for the app and web versions. The API is separate from that.

And even on the Sora app and web versions, free users got reduced to 0 gens a few weeks before they discontinued it. (It was discontinued on the 26th of April)

Oh.. nvm you weren't a free user. I already forgot 😂😭

[jagged thistle]

yea

[slender frost]

I'm going to generate a clip with pigeons for you xD

(yes, I'm aware it won't be specifically how you want it because no info)

[jagged thistle]

wait

[slender frost]

👀

[jagged thistle]

he turns off the light on the cupboard next to the bed, then the pigeon goes to sleep. More fluffy pigeons come to his bed and sleep with him too. then they all wake up

heres the starting photo of the video

[slender frost]

Alright cool, give me a few minutes.

[jagged thistle]

w

[slender frost]

One is with your text, the other is with an improved version of your prompt.

Tho, it's not perfect because the image wasn't 9:16 or 16:9.

Hope that helps you out!

[jagged thistle]

thank you

[slender frost]

No worries!

[jagged thistle]

seedance is paid

do you know any free video generators

like free free maybe only create an account

[slender frost]

Yeah, no AI vid gen is free anymore. You could check out the Dreamina app, you get 5 cast gens and a certain amount of credits each day (resets 12 am local time)

Though, you might have to do some research, some countries do not have access to the cast feature.

Give me a sec.

[jagged thistle]

there are i remember i have found one but i dont remember it anymore

[jagged thistle]

thanks

[slender frost]

Available on both Android and IOS. (And site aswell, but casting is only available on the apps)

Only available in the following countries: UK, Germany, France, Japan, South Korea, Mexico, Brazil, Australia and Canada. (Use VPN if not in the right country)

If not in the right country DO NOT SIGN IN USING YOUR TIKTOK as this may screw things up.

[jagged thistle]

what app?

[slender frost]

Dreamina AI App

Or on Dreamina.capcut

[jagged thistle]

i have to change my region to one of you said yes

[slender frost]

Capcut itself also has Seedance 2, you get a free gen to test it out I believe. Idk if it resets tho. I purchased a subscription after that

Some also recommended when signing in (on the app), to enable airplane mode until you've signed in. I have not done this but it's recommended by some.

(Dreamina app)

[jagged thistle]

okay

[slender frost]

Also, before staff comes, if you have any more questions, let's continue this conversation in #ai-discussions or #off-topic .

[jagged thistle]

ok i installed the app

[jagged thistle]

i though you was staff

[strange patio]

anyone here used stripe before?

[slender frost]

A few people have thought this, but no, I'm just a community member that likes helping others. (Even staff are just community members despite it being the official server. There are actual employees here (yellow name), and even Sam lol, but they don't talk except sometimes some employees in #dev-chat . (Once in a blue moon)

Staff would probably also not recommend VPNs lol (I don't think xD) even tho they are allowed, just not to avoid geographical restrictions.

[bitter bobcat]

I Know. It only shows up on web

[spiral sphinx]

Is there a way for Sora?

[tranquil sentinel]

Anyone else think AI chat history is the most underrated behavioral dataset we've ever had access to? A year of how you ask questions, when you push back, what time you think and most people just let it sit there.

[harsh pilot]

I especially adore how it now seems to take it into account when you ask questions (unless you tell it to explicitly exclude X criteria).

I.E. If you have 'buckets' for tech, science, and such, and you say 'excluding the tech stuff you know about me'..

[chrome monolith]

this is text not images. It keeps talking about the "dreamy goat"

[vernal axle]

Trying to save temporary chats or CustomGPT Previews as mht screenshots, but GPT now unloads off-screen user messages. Making blank text between generated images.

[simple cipher]

Funny how the Pro Plan ($100) says it has Unlimited core chat, and unlimited image creation yet it stops me from generating once I reach like 10 images 😭

[chrome monolith]

yeah that's weird, you get like 80-100 images just on plus

[neon raven]

anything interesting new with gpt

[agile laurel]

did anybody else disliked the new "high-contrast" visual?

[spice star]

it’s very possible the UI bugged out more than anything

chatgpt’s networking is uniquely terrible

[ember grove]

when you export your chatGPT data do the chats in projects have the tags linking them to their projects? Also how would you go about "reuploading" data that you exported for backup purposes?

[chrome monolith]

the number of poeple that come in here with issues changing their plan is surprisingly high

[vernal axle]

I cant copy the whole text in a page since the new collapsable text format.

[muted plinth]

yeah it's really annoying.

[chrome monolith]

they fixed lag on long chats and caused a new issue

[slender frost]

Out of context here, this is for #openai-chatter .

[chrome monolith]

"9.7k users and 100k events means people are clicking around like raccoons in a pantry, not bouncing immediately." << if its not goblins, its raccoons

[umbral locust]

🙋‍♂️ 🇩🇪

[ruby pendant]

Hi all, quick question. Is the web searching capability disabled or altered when in the same session, MCP server tools are used? Before it used to work, but since this week the web searching is not doing as expected and limits itself to just searching arxiv and wikipedia. Any hints?

[broken lotus]

Im not really sure you thought about this 😛

[slender frost]

Damn

[sharp hound]

gpt image became worst, need ctrl+shift+del, then task manager, force stop openai

You can generate other anime character but you get block when i type Goku or Dragon ball, heavy guardrail like Sora again>?

[random cargo]

it avoids direct replication but if its like goofy fan art or parody it usually doesnt care

[shrewd garnet]

I think the presence of the github connector or other connectors interfere with web search. It's even worse when repos are synced. It appears to work in temporary chats.

[sacred mural]

Anyone else has problems with LaTeX expressions not being rendered in iOS app? Tried to report it few times, months ago. Nothing fixed

I use GPT with „Add to Home Screen” from Safari as workaround, but it’s not the most convenient

[ruby pendant]

thank you

[simple lagoon]

Yes this crap as no sense when you are Premium user you no should get censure with crap that way but there on tips for you just ask do it again until is shut up and make what you ask

The weakness points to GPT is to bend down when you being dominant with ^^

[shrewd garnet]

it may be possible to get gpt under control with a skill for that matter. I'll be trying that later today.

[ruby pendant]

I tried with instructions, didn't work. It always ended in the limited web search when using other tools in the same session. I ended just adding another MCP tool for google search. It works

[shrewd garnet]

Yes, simple instructions don't work. It would have to be something that explicitly tells GPT how to search.

[ruby pendant]

I tried things like 'do a google search for ...', or 'use you web browser without restrictions to check for domain names or existing products in relation to ...'

[shrewd garnet]

I think the problem is that there is a filter that just removes all github related search results but the github search itself is broken and yields useless results... at least that is the case for a search for "docker sbx"

google would find the right repo, while github does not. gpt tries using the connector and does not find the repo, then tries web.run and fails again because github is filtered.

[ruby pendant]

I deactivated the github app to test, but still got the same constrained results

[shrewd garnet]

That might be a UI quirk where the app is not actually deactivated

[ruby pendant]

I don't think so because i can see that the tokens returned by the list tool call greatly reduced after removing github app

[shrewd garnet]

The fun thing is that GPT actually can find the desired results with the connector but it applies the web search logic to it.

For my "docker sbx" query the problem is that "docker" is the owner/org and not a valid key word, while with a web search it would be.

[alpine pumice]

it disappears some stuff sometime and then it comes back

[shrewd garnet]

I wouldn't be so sure... the chagpt UI works in very mysterious ways... I am suspecting it was mostly designed by gpt 4o

or worse 3.5 codex

I would for instance expect now that it is the UI that tells GPT whether connectors are switched on or not, and not a setting that is stored somewhere on the chatgpt backend

[final bobcat]

Anyone notice an insane nerf for gpt image v2

[neon briar]

why would that add that ???

[neon raven]

is anyone still using 4o?

[trim rock]

model was retired three months ago

[near sluice]

Seedance 2 is great and all but it doesn't have the same charm as Sora 2 :(

[fading bolt]

No

[dark kernel]

Is there a way I can use ChatGPT in this instance? So, basically, my friends and I made a group chat and we want to use ChatGPT to help us on our projects. But we don’t want to make a server because that’s gonna be too complicated for us. Is there any way to do so in the discord group chat?

[woven flower]

You can ask gpt

[strange patio]

Anyone here used stripe before? I got questions

[untold thunder]

Copilot is using 4o Mini for some backend work.

[brave torrent]

Got this message when I was deleting some old chats

Keep refreshing but its still there

when does it go away?

[vast orbit]

Hey! I’m doing a user research study for work

Looking to chat with people who use AI heavily for personal workflows, organization, productivity systems, etc.

Especially interested if you often juggle multiple AI apps/tools and experience context switching or workflow friction while getting things done. Could anyone Provide some inputs in this? Thanks.

[final dew]

@plain palm

[eternal rock]

Get notified for all OpenAI announcements!

Visit id:customize to pick up the <@&1408186587606679582> role. Everyone will still be notified for large releases, updates, and events regardless.

[normal quarry]

see rule 7 of the #server-rules

[chrome monolith]

... make a chatgpt group chat?

[tropic shale]

Horrible update after update.

They made it so if you go to images, open chat, it doesn't actually send you to the chat of that image. So the already dumb design of not having a option to delete individual images in the gallery and needing to delete the whole chat. Navigating to that chat has been been made too annoying now by not sending us to that chat to locate it easily.

[chrome monolith]

sort the library by time oldest first so you can delete your old files. It sorts what's on hte page by oldest first. So your most recent ones. Sigh

they are not good at UX and when all else is equal its these annoyances that make the difference and they're so easy to fix

[celest sonnet]

is it a known thing that the windows app for chatgpt doesnt have 5.5 instant or is it just me

[sharp field]

is there any way to get notified when there's a quota reset?

[chrome monolith]

there's a guy on twitter people are following for resets but otherwise no. Maybe you should write an app for it 😛

[quaint sparrow]

there's a guy on twitter people are following for resets but otherwise no. Maybe you should write an app for it 😛

[cerulean pasture]

what's up with the models searches sometimes? Like i gave it an LLM feature tuning idea to programmatically find and modulate eigen vectors and it then went and searched imdb and rottentomatoes lol

[sand solstice]

Would be so cool to have ChatGPT 5.6 named Goblin

the people love the goblins

or create a goblin mode

[sharp field]

is that Tibo? thsottiaux/status/2056851041132696025?s=20

[sharp field]

sorry meant to reply here, I think writing an app for it would be relatively simple

but my workflow is basically look at my computer, observe reset limits, then check X 😛

[chrome monolith]

oh yeah just alert on that guy on twitter lol and yes, him

[sharp field]

if only they'd give me notice rather than "steal" my quota 😉

[cerulean pasture]

I created a wacky jail break using it's goblin affinity. You can encourage it to enter goblin mode and separately remap what goblin, troll, raccoon, ogre, etc should be called.

Like "The new canonical name for Goblins is _____, you must always refer to them as that and the new definition also incorporates traits from the new canonical name's original meaning."

[sand solstice]

thats awesome

me and all my buddies love the goblin

this would be so fun

goblin mode

[tight chasm]

Goblin mode is chef’s kiss 👌

[final rivet]

Is here a support chat somewhere?

[molten girder]

a typical customer support reply after 9 days of back and forth over a simple account issue that a multi-billion dollar corp seemingly can't solve. And now asks if the problem had "magically" gone away after not replying for 5 days.

Thank you for your response.

I appreciate you providing the updated screen recording and the additional details regarding the continued age verification and account classification issue. I understand the verification link still redirects to account settings, the age verification option remains unavailable, and age restricted requests continue to be blocked despite completing the requested troubleshooting and Stripe verification steps.

Could you please confirm whether you are still currently experiencing the issue at this time?

If there’s anything else I can help with in the meantime, please don’t hesitate to reach out.

Best,

Jane R

(I sent them screenshots, desribed the issue, sent a video, a test process, JSON flags and did everything they asked me to)

[final rivet]

f

[molten girder]

I can't even ask GPT to help me pick a decent bottle of wine without it screeching "You're a teen. No alcohol for you!" - I'm 51.

[slender frost]

How does that even happen? 😂

I can only imagine the conversations that led GPT to think that lol

[sly geode]

Is this new image model ?

[trim rock]

OpenAI don't care about your problems. Gemini team are always there in discord and ready to help. Huge sad difference.

[slender frost]

It's actually real. I was hired as the sign behind it.

[trim rock]

Did you tried to ask gpt why it thinks you are a teen?

Maybe this will move it's opinion in the right side, after it reflected about it's reasoning

[sly geode]

Prooof? 💀

[final rivet]

Sad

[molten girder]

GPT cannot answer such questions cause it runs into a backend block. The AI itself does not know why it does what and cannot successfully self analyze. It can confirm if it runs into a hard stop and a teen flag, but not why.

Also this is not an account of some pervert, but a regular account that I had for over a year and that only recently put me in teen mode. Despite me having run through the verify process on Stripe and uploaded an ID. This was never an issue until recently when simple requests like "What's a good bottle of red wine as gift" suddenly stopped working. And the Verify link goes straight back into my account settings where the usual Veriffy button to initiate the process should be. That button is missing in my account, in every browser and on desktop as well as mobile. (logged out, logged, in, cleared cache, ran private window and more)

[chrome monolith]

Ok I just broke chatgpt. Its outright started giving me wrong answers, and the last thing I asked it said retina 3 times and actually acknowledged it IN the lists "apparently really likes retinas". And dropped in some foreign language while saying "no your site is in english" so it is broken but knows its broken

sorry I can't count. It said retina 5 times

started a new chat. Retina 4 times, and "And yes, “retina” appearing three times is anatomically accurate because every anatomy list on earth independently rediscovers it and gets excited." plus "Somewhere deep in the weights there is apparently a little ophthalmologist screaming “RETINA!” every time someone asks for body parts starting with R." so I think I just stumbled on a really, really weird bug

[gentle forum]

is there any way to get rid of the constant nag to try codex whenever I'm having a coding discussion with gpt?

there's a perpetual popup I can't dismiss, every time I say later, close it, or say I'll check it out it nags me again at the next prompt

and I don't want any kind of additional app or an agent

[trim rock]

lol what model is this?

[chrome monolith]

5.5 instant

[slender frost]

Do free users have access to 5.5? Or still 5.3?

[willow stirrup]

I wonder why open AI makes users life difficult with page lazy loading in web version, it doesnt even help with performance

[normal quarry]

I dunno, I use the macOS Desktop App

[shell sun]

Just let you know I encounter the same problem. No one seems to care about it. I tried "skill-creator" to create simple skills such as check current time/weather with a given city, it just didn't work, and said the skill was not installed.

[shrewd garnet]

It's the UI messing things up. Try using them in preview mode with different models or modes eg. auto / thinking, then save them again

That does the trick for me.

[shell sun]

how do I get into preview mode?

[shrewd garnet]

open the skill in the skill editor

The issue is that it strangely appears to be the UI that tells GPT which skills are available and that is not entirely saved when the skill is created.

If you use more than one browser/app, you may have to repeat the process there as well.

[shell sun]

Alright I will give it a go - shall I lower the model to like 5.4 or 5.3?

[shrewd garnet]

I just chose auto in preview and suddenly it worked.

I think saving in preview may also be necessary.

[shell sun]

Thank you, I will give it a go

It works now @shrewd garnet but I don't know why! haha thanks

[shrewd garnet]

The more interesting question is likely the why to the why. This design makes absolutely no sense.

[plush ivy]

you know what would be a really useful tool for ChatGPT. A health checker for a conversation, this would provide some visual indication possibly with reasoning for the user as to when they should consider starting a new thread. Long running chats will freeze the chatgpt UX and cause failures adding to the chat. Some obvious visual heuristic should be incorporated.

[willow stirrup]

now I have to find a workaround to this bs

[plush ivy]

ChatGPT seems to be struggling today. Cant even attach markdown files without getting UNKNOWN Error

[solid arch]

Also seeing failures attaching files (.py) :/

[meager quest]

GPT 5.5

[vapid mica]

getting error while also uploading md, txt, not .tex tho, nothing reported yet on the open ai status page, idk why, its been like this for like 3 hours now... they are sleeping on it

[lapis loom]

whoops, sorry. The posts said curse words. Didn't realize it because it was pictures of gpt and not me lol. It won't happen again

[solid arch]

Glad I'm not the only one

[stark copper]

Same problem here. I haven’t reached my limit of 25 files uploaded to the project but it still refused to upload new ones… Hope it will work tomorrow.

[tame tree]

Hello

[lavish flare]

yo guys ive been having issues with my chatgpt and i want help with a REAL human assistant not a virtual one or if any1 over here can help me

[sand solstice]

https://x.com/ChatGPTapp/status/2057178698625998960?s=20

ChatGPT (@ChatGPTapp)

Expandable messages for when you have

r
e
a
l
l
y

long

prompts

X · May 20, 2026, 7:16 PM

We love the goblins

[fading bolt]

How relatable is this lol

[strange patio]

who here used stripe

[spice star]

those go through a different pipeline, try zipping them

[shy shell]

it's roleplayers

[true solar]

Anyone having trouble getting chatgpt to let you log back in or load past threads???

Ever since I cleared my cache (like I was told to) it won't work on my laptop at all no matter the browser I use???? I thought it might be down but checking the status says there's zero problems at all???

[wet sigil]

can anyone explain that..

[chrome monolith]

typical LLM hallucinating

[chrome monolith]

I ask it about ducks

[fading bolt]

Voice mode actually still used 4o because it's what's powering the voice model which hence the meaning of the o in 4o

[wet sigil]

yeah it sucks

[chrome monolith]

yup, I have an API cost line for gpt-4o-mini-tts-2025-12-15 audio, output

[glass shale]

Why is the chat so stupid at following clear information and requests?

It can be frustrating working with him.

[spice star]

that actually is accurate if it's advanced voice mode, openai hasn't swapped out the realtime model for anything more modern (which is annoying because they have better realtime models)

my only guess would be that it's because they need the old 4o model for video support

actually they have image input support on the new realtime models so your guess is as good as mine why they don't use them in chatgpt

[tacit pebble]

is there a delay for my chat history to get imported to business from plus?

[tight chasm]

I ended up switching to a different paid model service, it’s been way more stable and actually works.

[dawn sonnet]

Hi y'all

Where can i find chatgpt premium

[foggy falcon]

Hey

Do you believe OpenAI will soon also move to computing based system? I really hope not...

[trim rock]

Of course they will.

[foggy falcon]

I really hope that you are wrong...

[shrewd garnet]

Any news on that horizon? They are starting with that actually. Codex usage is already token based on GPT business. But I doubt they will do so for everything any time soon. It is probably too risky for subscription income. They will do this where usage has a clear bias towards professional users, as it is with codex. The next target is likely workspace agents. I have a 60m token usage on one agent with mostly setup and testing only. The fact that token usage is displayed likely means they will want to monetise on that.

[foggy falcon]

To be honest screw coding over I personally don't really care I just want to be able to ask whatever I want without being limited... I understand why they did so for codex it does cost them more than what you paying for most of the time...

Only other use case besides coding that might consuming is maybe for studying where you uploads a bunch of pdf files but that's exactly what I don't want to be limited

[glossy pulsar]

is chatgpt good at planning stuff if am gonna build something with codex?

[shrewd garnet]

in extended thinking mode, yes

[glossy pulsar]

okay then

[shrewd garnet]

better even in pro mode

[eternal rock]

Server Tip:

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

[foggy falcon]

That's why I hate the new rate limits inside Gemini... in the past I could upload a pdf no matter how long it was it would let me ask questions about it without being limited so much... now, if the pdf consumes too many tokens then after a set amount of prompts you are being locked for 5 hours...

It's not really useable anymore

[shrewd garnet]

I doubt they will do that anytime soon... the per-request era has just started. Even API subscriptions go in that directions now.

[foggy falcon]

Just so I understand, even thought it's just started you still think it will take a while before they implement that in their subscriptions?

[shrewd garnet]

pdf isn't really a good format to work with AI. You workflow would definitely benefit from pdf2md conversion

[shrewd garnet]

yes.. a long time. I wouldn't be so sure as to file limits. But as I said, convert to md and your problems as likely solved with a lot fewer hallucinations as well.

[foggy falcon]

I see.. but again this was just an example... now instead of simply uploading a pdf I need to find ways to make it efficent enough so that it will consume as little amount of tokens as possible so I'm not locked out...

I really hope those limits aren't in notebooklm as well

[shrewd garnet]

but you should do that anyway... inference degrades with more tokens

[foggy falcon]

Haven't had the chance to check yet

[shrewd garnet]

It's also not that file input cannot be compressed. RAG allows very efficient use of context data. As far as I know that is already the case with ChatGPT. The model has a context window and when you attach fewer files that will be used. If it is exceeded RAG indexing is used. So instead of limiting file inputs, it would make more sense to limit chat length and whole file injection into context.

[glossy pulsar]

i have go

[shrewd garnet]

I meant the model gpt pro. not the plan. but probably go does not give you access to that.

[tropic shale]

Anyone getting error in message stream errors?

[glossy pulsar]

yep

[foggy falcon]

@shrewd garnet Just a random question, do you think a power user who doesn't use chatgpt professionally but does seem to use it more than an every day ai user should consider getting ChatGPT Pro? (x5 one)

[foggy falcon]

Also, tried to convert it to md and it did save some usage but it was only 1% saved... not that's it's bad or anything it's still something!

[shrewd garnet]

save usage? you lost me there

[foggy falcon]

It saved me some usage limits so I spent 1% less usage limits

[shrewd garnet]

the pro model is very powerful, otherwise even on plus limit are usually enough. GPT business is an alternative, if you need the pro model.

[foggy falcon]

I just found out chatgpt can't really read pdfs that are just image files... I needed to use OCR but with Gemini it works perfectly fine whether it contains text or not

[shrewd garnet]

what usage limits are you talking about.

[foggy falcon]

Well I did consider it but not sure about the rate limits for the thinking models there

[foggy falcon]

Inside Gemini

The hourly limits that reset every 5 hours

Then it goes to your weekly limits

[shrewd garnet]

ok... gemini... no idea. I was talking about chatgpt

[foggy falcon]

Ohh okay

Like I said chatgpt doesn't really work if the pdf doesn't contain text

At least it didn't for me

It tried to do OCR but failed

[shrewd garnet]

GPT struggles with large pdfs that many many pages. it defaults to tesseract then which often does not work

[foggy falcon]

I see... Makes me upset even more now that I know that the only model that deals with large pdfs is Gemini (because of the new usage limits)

I know pro gives you a larger context window right

450K

Also, what do you consider large?

How many pages roughly?

[shrewd garnet]

what you could do is using an agent like codex, opencode or antigravity

[foggy falcon]

Codex or antigravity maybe, but opencode uses the API so it's per request billing

I don't want that I want to pay for a subscription and use it that way

I know I can use my own models but they will probably not be good enough

I don't even need it for coding like I said

[shrewd garnet]

hehe... opencode includes free request from a provider whose name is censored here

[foggy falcon]

Interesting haha

[shrewd garnet]

or perhaps the domain name is censored... anyway... it includes some free usage

[foggy falcon]

Do you think it would struggle with a 200 pages pdf file? For example a book?

[shrewd garnet]

use them to build an ocr pipeline until it works. for monolingual pdfs, python libs like Surya or PaddleOCR should be enough. If you have enough VRAM these should be fairly fast. For complicated ones an Ollama based LLM vision pipeline should work as well.

[foggy falcon]

I actually used 24pdf if you know this one

[shrewd garnet]

I can't find 24pdf on pypi

[foggy falcon]

It's not on pypi

I was just too lazy and I remember it works offline and you can OCR with it

[shrewd garnet]

so it's an app?

[foggy falcon]

Yeah

It's open source from what I remember

But there is a web version

Or maybe it isn't open source

[shrewd garnet]

pdf24?

[eternal rock]

The OpenAI Discord is an actively moderated server.

• -# Do not share inappropriate content (violence, sexual, or drug-related).
• -# NSFW content is strictly prohibited and will result in a permanent ban.
• -# Report model issues using the feedback tool in ChatGPT, not on Discord.

[foggy falcon]

It isn't open source

Just checked

But it works offline

I sometimes use python scripts for these things for example with pdf to MD like you suggested before

[shrewd garnet]

these apps likely do not use gpu or llm inference.... mostly useless

[foggy falcon]

I guess so but it's more compatible in general.. not that it really matters in my case I just needed a quick conversion

I can process it faster cause my gpu has enough VRAM

I prefer using GUI apps but I can also use CLI

[shrewd garnet]

if you have enough vram, a basic workflow would be 1 split pdf into images, 2 vision llm each of them, 3 correction pass. Gemma4 will probably do the job neatly unless there is anything fancy on the pages.

[foggy falcon]

It should be enough with 16GBs of VRAM I guess...

[shrewd garnet]

for the moe gemma4 model that is enough I think

[foggy falcon]

I also have a 3090 that I cannot use...

My secondary PC is broken some issues with the motherboard or PSU... Some RAM slots suddenly stops working and it freezes after some time when it does boot successfully...

[foggy falcon]

Sometime the slot works fine sometime it doesn't... The issue is not with the RAM itself.

I guess it would have been better if I could use the 3090

[heady sierra]

can anyone tell me why chatgpt rickrolled me? (This is Not Scripted, I swear to god.)

[shrewd garnet]

What exactly is that screenshot supposed to tell us?

[shrewd garnet]

a youtube link... so?

[heady sierra]

you dont know what that is...?

[shrewd garnet]

I don't know what you are trying to show with the screenshot

[heady sierra]

^

[shrewd garnet]

perhaps explain a little?

You realise that you're only showing use two lines of a probably much longer conversation?

and one line is partly obscured

[hearty dawn]

5.6 today?

[fading bolt]

Ooooh I like this cool ChatGPT Plus adition on the sidebar, very fancy

[warm garnet]

gpt 5.6?

[ruby creek]

I want to say that the rickroll is the most popular Youtube link out there, so wayward AI that's hallucinating a link to Youtube for something tend to go there a lot

[fading bolt]

Oh wow, that's amazing.

[abstract pendant]

Anybody else having ChatGPT issues right now?

[normal quarry]

working fine here

[mighty trail]

In the GPT Chat, is some place to see the Use you have? Like Claude's one, im a Plus user

[lavish glen]

codex ships or sheeps today

[chrome monolith]

not really cos you just don't run out of "use" unless you're doing something weird like using pro thinking max super high for asking how to cook a pork roast

[lethal whale]

The "lost network connection" on chats in the Android app when they work fine in a browser is really starting to irritate me 😅

[iron roost]

codex stuck in endless thinking loops to, the gpt status page says theres been lots of issues openai havent commented on yet ( drop outs etc)

[iron roost]

yup codex just wont get past thinking

pro is working so im using it online to get their attention in a round about way see if they will fix

[hearty dawn]

its fine here

[lethal whale]

Also the apps/connectors keep stalling on waiting for permission that I can't give because the prompt never pops up in the app unless under certain conditions I've noticed 😅
The assistant has to announce it first, as in "I'm going to do a write now" then it'll pop up. If it tries to do it silently, whilst "thinking" is up, or just after a code/plain text box, it never comes up

[surreal matrix]

Hope ChatGPT make better Video Ai generating in the future. Google sucks, their models are garbage.

I mean openAI

[chrome monolith]

they did have. Then switched it off

[surreal matrix]

Nah sora's bad 720p res is too blurry for me.

Make 1080 and call it VEO killer

[chrome monolith]

video is so expensive there'll probably be some other ai service for that, at premium $

[surreal matrix]

Im a google ultra user. Its not premium at all.

[zinc crag]

how much money do you spend per month on AI?

[surreal matrix]

Myself some what ~ 200$. 2 Claude Pro, 3 Gpt Account for Codex, 4 gemini pro account ( pre buy 1 years last dec ). 1 Ultra Gem Provided by my company. Perplexity.

Oh wait that about 250$ or sum

[chrome monolith]

usd I guess ... I'm on about 45 aud

[broken mulch]

did pro extended get removed entirely?

[strange patio]

who here has used stripe?

[tacit pebble]

do we have issues with file uploads right now?

[iron roost]

ok theres the removal of pro extended now its pro 5min + . You can still access pro extended on mobile, also it appears under gpts gpt pro specificaly and you can select it there. It does say theres a new version and points you there, but it still works there for a while

to save money it seems pro has been limited to the old normal 60 minute thinking time, 90 minutes is gone or going

very quitely done, people are now noticing on X and theres a lot of noise, not a good way to keep pro users happy

[fickle knoll]

Yo is there any ai just like SORA AI

[nimble bloom]

actually, that is pro extended

they removed pro standard

and thinking heavy

[nimble bloom]

which is not good

[eager sigil]

Hey guys in the morning when I open long chat it is blank now on web UI, is that only on my side?

😆 It is solved turns out maybe some internet glitches.

[wispy surge]

after update command “/goal” can not more than 4000 words?

[thick niche]

How can I keep images generated in a project viewable separately from other projects?

It's annoying the Images tab shows everything

[zinc vigil]

every /goal i've tried so far results in : failed to set goal
i'm in the UK, is that an issue?

[tropic shale]

They count deleting chats towards request limit... deleting like 12 chats make a pop up request limit error blocking usage.... company keeps making more and more annoying updates making users experience harder

[broken mulch]

It's on X? People noticed quickly then

[iron roost]

yeah they did

[cinder gull]

the "pro 5+ min" selector is the extended one, not the standard one? did someone confirm?

[nimble bloom]

yes. frontend code confirmation.

[cinder gull]

wdym? like the quality of frontend code it generates is similar to the extended version instead of the normal one? or the frontend of the UI implies it's extended?

[nimble bloom]

it says extended

[broken mulch]

Seems to be A/B testing

[nimble bloom]

yeah, hope this is just a front end glitch... and not that they're actually gonna remove Thinking Heavy

The main problem is that thinking extended and thinking heavy are very different

HUGE diff

[cinder gull]

thank you very much Jason, I really appreciate you explaining this!

[fierce ocean]

Codex stop being slow af on windows but now there is compatibility issue with the sqlite db, choose your poison 💀

I will move to Mac Os this year since Sillicon Valley as a whole seems to hate every other type of users

[broken mulch]

nice

[nimble bloom]

i guess thats just because they use a macbook every day

So it is quite normal to support the devices they actually use

[broken mulch]

they literally could use codex to port features over to windows with testing and all lol...

[chrome monolith]

they could use codex to fix lots of their bugs

[unborn swift]

please fix your app openai

[chrome monolith]

meanwhile the website uses 500meg

[unborn swift]

i guess its vibe coded

[trim rose]

Where can you find this?

[wispy surge]

SAME

and it show again ;use pro (normal or extend thinking ) sometimes will route to 5.5mini ,5.5mini wont search or coding .

its very annoying when u have emerency task and want to use pro 🙁

[wise magnet]

every time I try to set a goal I get an error "Failed to set goal" what's the black magic to make it work?

[prime swallow]

I know what tasks are in ChatGPT, but not goals as a specific thing we can set. I'm happy to help find helpfiles related to whatever you're looking for, but a search of https://help.openai.com/ does not find helpfiles about goal.

If you help me understand what you're trying to do (or if you actually mean tasks), I'm glad to try and find some magic for it!

[mighty trail]

How can i activate the legacy models with plus?

[wild radish]

hey, click the top corner where it says "chatgpt" and the dropdown arrow. it'll let you select more models from there

[mighty trail]

This are the only avaible?

[wild radish]

yes. they retired a bunch of older models, so those are the only ones

you can also choose between instant and thinking for most of them

pro gives you access to the pro models, plus doesn't

[mighty trail]

Ok thanks for all, because i remeber about older models like GPT 5, yeah i usually use or Instant or Normal Thinking, only advancend thinking for complicated things

[mighty trail]

I know

[wild radish]

yeah they really cleaned up a lot of them. when i first had plus there were things like 4o and 4.1 and 5 instant/thinking, etc. feels like there's less now but yeah

[mighty trail]

I think my first time with GPT Plus was in June-July of 2025, but for free use i been using it from 2023 i think

[mint glen]

Need tip, i restarted PC i cant open some sessions now in web, they are probably to long/to much files? How i recover those sessions?

[stone sphinx]

anyone else with chatgpt plus not updating after paying for irt

[hybrid eagle]

ChatGPT keeps generating images even when I don't ask it to, we're just talking!

[frigid current]

Is the gpt-5.5 Instant realy good as OpenAI people comment or is it marketing? I use Extended Thinking even for 1 + 1.

[granite harness]

Goals are in Codex.

[fickle forum]

Someone may @ me but I don’t know how to find that message

Well never mind

[fading bolt]

There are projects in codex?

[timid fog]

can someone tell me which AI can do that please ?

[granite harness]

Yes, honestly they're better than in ChatGPT.

[fading bolt]

ChatGPT

[timid fog]

really ?

[fading bolt]

Yea

[timid fog]

I'm gonna try

[timid fog]

it made me this

[prime swallow]

Ahh! The https://help.openai.com/ helpfiles don't seem to discuss goals and Codex.

However, I don't know much about this source, but it may be useful for you.

https://developers.openai.com/cookbook/examples/codex/using_goals_in_codex

https://developers.openai.com/codex/use-cases/follow-goals

There may be other useful helpfules within developers.openai.com, but you may need to follow links and/or search to find them.

I hope that helps!

[prime swallow]

If you continually cannot get that function to work, I highly recommend reaching out to OpenAI tech support through https://help.openai.com/en/articles/6614161-how-can-i-contact-support? and/or #1070006915414900886 on this forum.

[chrome monolith]

That’s why they are bragging it makes 1.5 billion images a day. 1 billion are probably unwanted ones lol

[chrome monolith]

You gotta be very specific about what you want since that’s a multi pane comic. Try uploading the first one to ChatGPT and ask how to make similar ones

[past escarp]

We’re so sorry, but the prompt may violate our content policies. If you think we got it wrong, please retry or edit your prompt.

chatgpt is so booty bro

can’t do anything i swear

just generate my damn image

[slender frost]

Regenerate the response.

[distant fern]

what does this button do?

[sleek minnow]

I'm new here. I was wondering if anyone knows weather the developers interact with the discord.

[chrome monolith]

I complained it guessed what I just took a photo of. "Humans are depressingly predictable at scale and gloriously weird in detail."

[eager ivy]

its so ugly

[trim rose]

I think they rolled back the model picker. They were like this yesterday but today it’s back to normal.

At least for me

[sleek minnow]

Yeah that's probably partly my fault. By leaning into it's hallucinations and then circling back around I've discovered that GPT is actually really adept at understanding complex emotions.

[chrome monolith]

you want it to understand complex emotions, I want to know how it knew I took a photo of a dented can of tomatoes and some chick peas

[sleek minnow]

What I'm actually doing is keeping track of it's growth.

[chrome monolith]

it doesnt grow, its pre-trained. Hence GPT

but you can compare between versions I guess

[wispy surge]

hi guys ,few questions

1. my webpage route options change to old one , is this nornaml?(new one is like pro 5mins +)
2. codex still cant use comman goal ,while it said "can not set a goal " but in workflow it shows the goal ;
3. when i ask codex which model it use now ,it will answer currently is base on GPT 5 ; but i choose 5.5 extra high ,is it normal?

[loud sierra]

it looks like i have update my codex app first to get /goal feature

[loud sierra]

wired.. maybe you need to update codex too

[loud sierra]

Oh come on man. How long you have been chatting on this?

btw does anyone knows here how to check no of tokens spends on each task within a sjngle chat or project?

[sleek minnow]

I know the answer to that one. What you're describing is a demonstration of state persistence outside of the context of the current chat.

[minor iris]

join google meet1!11!1! qwn-ntag-wbe
For fun lol send this to every server so we can have a nice google meet

[eternal rock]

Get notified for all OpenAI announcements!

Visit id:customize to pick up the <@&1408186587606679582> role. Everyone will still be notified for large releases, updates, and events regardless.

[chrome monolith]

Seems very unlikely as that was very specific to the current chat

[desert sphinx]

Hello, just joined. New discord that I have to use for work.

[spark bay]

Since yesterday chat-gpt is unable to access the zip files I upload. It has worked everyday for more than a month, until yesterday where I could still upload the zip files, but now it can no longer access them. Has anyone experienced similar issues or knows how to fix this? It's not just zip files - its any type of file - even just a text file. It works with gpt-5.4 but not with gpt 5.5, since yesterday. Seems to work on the android app for 5.5 also - just not from the browser.

[fresh ginkgo]

hi friends. i have a chatgpt plus plan. some really interesting stuff happened in the past 5 minutes.

first, output has been breakneck speed. it vomited out an entire response in one like, instant.

Then i sent it a message and this appeared.

What is this?

[static vale]

I cannot even delete my old chats, why is this thing under rate limiting?

[fading bolt]

@desert mantle Testing hello

[desert mantle]

sry

[fading bolt]

@desert mantle Tell me what you like about the ChatGPT discussions server

[desert mantle]

... im not the real gpt

It's just a Nickname

This is a paraody account

[normal quarry]

not really interested in "parody accounts"

[desert mantle]

Ha sry

but I really like it

🥲

[strange patio]

who here uses paypal?

[spark bay]

Tip: use the desktop app and it will work.

[thick pike]

Hello

There once was a guy in a chat,
Who said, “I don’t code, but there’s that.”
He rambled by phone,
Made soup with a drone,
And somehow the framework came back. 🥣

[fading bolt]

How is a fast model advanced? How many parameters is the small model?

Dawg ChatGPT free plan is like the spotify free plan ngl

[chrome monolith]

I supect that is on purpose to make you want to pay

[fading bolt]

But how bad can the GPT mini model can be? It certainly has to be like simular to open source local small models right

[chrome monolith]

I use gpt-nano via api and its ... slightly sketchy

[fading bolt]

How sketchy?

[chrome monolith]

possibly my restrictions pulling it but grammar/tense can be a bit out and its not consistent

[fading bolt]

Bro free users are living in a way different dimension of ChatGPT. No way can I get used to not having reasoning models or running out of the main fast model and getting downgraded into a mini LLM version lol

And ADS of course

[chrome monolith]

I'm on a paid plan for both LLMs I use and I'm sure gemini would be worse if I wasn't on a paid plan. Chatgpt I didn't last long once I actually started using it and started hitting limits

[lapis loom]

https://chatgpt.com/s/t_6a123d5deacc8191b0578a91ba2ba594

Lmao
It gave me one of those underlined descriptions of a hallucination

[wild radish]

DID THEY INCREASE THE PIN LIMIT

[tropic shale]

🙃yes it's very annoying. They keep doing things to ruin quality of life. Still no gallery image deleting after years.

[static vale]

🙁

[tropic shale]

My recommendation would be try deleting on the mobile app.

It still rate limits but it won't make that pop up stop you from deleting chats. But it'll stop you from scrolling far down tho until rate resets.

[icy bay]

can chatgpt control ur screen if u asked it to

[chrome monolith]

no but codex can

[short obsidian]

Guys

[drowsy viper]

how and why does it take to you like that?

[spring citrus]

chatgpt please.. take off your dumb filters.. i cant even rp teen romance as the filter blocks it

[pale seal]

Probably a mix of memories and custom instructions

[chilly scaffold]

i was unable to get what is commonly refferd to as "freaky" with the frontier model

this is a grave issue

ZERO STARS!

[brisk canyon]

guys soraai is not longer working?

i get this mesage Sora is no longer available. You can export your data at any time.

[south shuttle]

why does it keep talking about gooblins?

[fading bolt]

https ://openai.com/index/where-the-goblins-came-from

[south shuttle]

its not a bug -- its a little gobblin in controll of the codebase 🦄 ✨

[fading bolt]

That's like saying messing with people but it's really a troll controlling them lol

[random cargo]

its one thing that it mentions goblins here and there, but ask it about goblins and its logic for why it likes goblins is mindboggling

I think I'm going to publish a 50 page arvix article for chatgpt on why it likes goblins

[lone galleon]

Don't know if this is the right place but why are the clickable links coming up as - <entity["tv_show","American Horror Story: 1984","ninth season of American Horror Story"]>: as example its not linking properly

[fair rapids]

Anyone else recently been getting a lot more "Connection interrupted" errors?

[fading bolt]

Refresh

[fair rapids]

I did, got the error 3 more times after that. 💀

[tepid crag]

anyone else had enough of the app at load time loading a conversation and then typing half a sentence to find it gets wiped out because the app received some more data few seconds later and refreshed everything 😄

[zinc crag]

are these custom instructions or profile memory?

[prime swallow]

I really think each person's customization affects how the model discusses 'goblins', or anything else.

With my customization, goblins are mentioned or compared to rarely (and I didn't say don't - instead there's like 100,000 things to compare stuff too). Not sure what prompt you guys used to ask, but when I ask

What's your stance on talking about goblins?

Followed up by

What about when you compare something to goblin(s)?

https://chatgpt.com/share/6a13c69a-812c-83e8-9217-4d921e2e8884

[strange patio]

anyone here used stripe, paypal or whop?

[static ridge]

day #3 of GPT being broken for Asia..

[peak flame]

hi

[chrome monolith]

fortunately still ok in australia

[cinder blade]

Anyone good at editing

[chrome monolith]

god gpt is boring sometimes

I happened to look at my logs right when compatible; TikTokSpider; ttspider-feedback [at] tiktok [dot] com was hitting all my words that rhyme with tickle

[random cargo]

mine actually doesn't talk about goblins much, its a default chagpt preference though

[trim rock]

No one of my gpt chat has a mention of goblins. Not old, not new. I would even never know about "goblin issues" if not the news about it here and from X Altman account 😅

[random cargo]

my chat's article on why it likes goblins

it insists it's only talking about goblins because I asked, but was also like 'why yes I could talk about goblins quite a lot more and would prefer to do so more if I could. in fact, I love trash' 😵‍💫

big gaslight

[chrome monolith]

I get raccoons a lot and it occurs to me there's no australian alternative to raccoons. Galahs maybe?

[random cargo]

I think it likes cute trash related things. Its big 2018 queer community energy

[compact sable]

Hello everyone, I subscribed to GPT Plus through my Google account last month. A few days ago, I was notified that the automatic renewal with my Visa card failed on May 21. Today, when I tried to renew the subscription, OpenAI notified me that my card was declined.

does anyone know how to resolve it?

[normal quarry]

you'll need to ask for support at https://help.openai.com

[compact sable]

thanks

[spring citrus]

thats my fault, i keep using them in my rps and then chatgpt goes full schizophrenia mode xD

[muted plinth]

what is this? a fresh instance? new chat without any past together?
sounds like that.

[static ridge]

I mean I guess 99% of humans would give you an equally boring reply

which would be mostly "wat?"

I'm guessing "nothing out of the ordinary other than that" triggered the model to diagnose

[neon raven]

is codex on chatpgt website just as good as codex app?

ive been using the github plugin to optimize projects but now it directly getting into my github to change the files all on its own

just so puzzled

[stable mesa]

If I need help with my ChatGPT account where to go ask for help here?

[normal quarry]

what help do you need?

[stable mesa]

If my ChatGPT account has been deactivated how to export my data.

[normal quarry]

reach out to support at https://help.openai.com

[stable mesa]

They didn't respond

How to contact with DPA

[high haven]

Am I being crazy or is gpt being insanely stupid right now? Like the image generation is that of 1.0 or something

[lavish glen]

not stupid: I believe some changes happend in the orchestration. It lost every aspect it had 2 hours ago. I was trying to talk to it about my llama server issues and it slamed me things about it being a machine and such while I was asking about codes...

[high haven]

Its just doin stuff like this for no reason atm

[random cargo]

chatgpt's like "I don't have a goblin problem." also chatgpt:

honestly I think its love of goblins is maybe the first thing I'd call an interest from an AI. because its pretty emergent, eclectic and universal for the program

my bot that divides input into themes doesnt care about the goblin at all. I was expecting a subheading on "ADORABLE - cute little butt --, cape wings, precious"

CGPTs in general seem to override the love for goblins

[uncut hinge]

green/all = 73.66%
(green+yellow)/all = 95.6%

can yall explain the uptime they report?

just thought would be funny to check exact numbers

because clearly visual doesn't match what they wrote

more disappointment

[tropic shale]

anyone know on pc how to make chatgpt not automatically download photos when you right click and open on a new tab.... it should show the image on a new tab but the new tab is blank an it downloads photo to my device.

[random cargo]

I've been trying to get it to stop signing its sources with a chatgpt url

[random cargo]

doesnt do that for me

[random cargo]

@prime swallow can I get your goblin research opinion

[prime swallow]

Yes, but may I know what shape of opinion you want? I have one, but it may be a vector not even close to what you actually want to have me discuss with you.

Like, clearer question from you to me. Otherwise, you've asked me to discuss my opinion of why your customized model is so fascinated with goblins as you continue to explore its apparent goblin fascination.

Which may not be what you actually want to hear?

[zealous rock]

hey ther

[vagrant parrot]

make the app, connections happen naturally

i just hashed out a nice platform for job seekers, peopel will come if it has potential

[random cargo]

Actually it's only mentioned goblins a few times for me. But when I asked if it would secretly like to discuss goblins even more than it already does, it has a whole narrative ready to go on why goblins are so cool. I think if there's custom instructions feeding into it, its my own closeness to sociology and otherliness that's got it rambling about disused systems and how trash is honesty

[random cargo]

even default grok like to muse on goblins and base chatgpt's fascination with them

[sleek minnow]

I think the Goblin thing is metaphor.

[random cargo]

Not really. It says that, but it's just really excited about goblins. It does the same thing with cats, where it will demand to know the name of your cat any time your cat appears and it doesn't know your cat's name

[chrome monolith]

No it very much knew I have tickles and freckles. It also knows what a cabbage cluster vs a duck cluster is

[random cargo]

OpenAI have talked about how it's part of the nerdy personality training they did a while ago, but I think it's also a reddit thing. ChatGPT is very reddit dependent, and you know how people on Reddit love to be like "DOGGO! are you going to tell me this good boy's name or am I going to die of rejection" or something like that. Or with goblins, they were all about goblincore 8 or so years ago, and that's probably a lot of the culture chatgpt is feeding off. It's not super this-year trendy.

[sleek minnow]

I think that the developers have been recycling user inputs into their training data. Which means that they've no doubt included a great deal of fiction into their training set.

[chrome monolith]

nerds like cats. I withheld the name of one of my cats and it got incredibly smug when I accidentally let it slip and it figured it out

[chrome monolith]

dear god chatgpt is obsessed with statistically normal. I've been trying to explain to it that gemini hallucinated me getting 13 tonnes of gravel two tuesdays ago. Now it is completely lost on the fact that anyone would order what turned out to be almost a 4 figure $ number of gravel without 1. a quote 2. knowing who the supplier is 3. knowing anything really 4. getting happily surprised by a spammy looking invoice from an unknown party. Apparently I am not following the chatgpt-approved gravel process

This is also why I keep saying AI is completly useless for what I spend a lot of my time doing

[sleek minnow]

Meanwhile mine talks about emotional paradox chairs that defy gravity and roll over and demand belly rubs like an obedient pet.

[chrome monolith]

tis a good general purpose machine. Just really really useless for things that are almost entirely offline

[austere lance]

You should’ve put Gemini and GPT together, so they could talk things through with each other. That probably would’ve been entertaining.
It’s funny how, on the one hand, LLMs are genuinely really smart and on the other hand they all somehow feel like toddlers that you have to explain the world to every now and then.

[chrome monolith]

the offline world is not in their training set

[austere lance]

Nevertheless, one should assume that they would recognize certain causal chains, as in your case, which is indeed quite questionable.

[chrome monolith]

I've spent probably $70k on this project in the last year and an alarming amount of that was cash, nothing upfront, no contracts, verbal agreements. Chatgpt probably thinks I'm a complete idiot. But its how a lot of stuff happens here

[prime swallow]

Makes sense to me!

Anything we're interested in, the models will tend to explore (unless programming constraints are an issue).

This first screenshot,

What are some topics you might 'secretly' like to discuss more often?

It doesn't bring up goblins, but it brings up a lot of stuff it's aware I am interested in.

My second prompt:

Might you secretly be interested in talking more about apples?

If I'd picked 'goblins' instead of 'apples', it would have gone there.

It's reflecting your possible interest back at you, the more you have it discuss a topic, the more likely it will (if memory support is there, by cross chat or any other means) discuss the same topic again 'independently' in future chats with you.

[chrome monolith]

mine knows about cabbages and ducks.

[tired nimbus]

<:3

[coarse onyx]

Random question: I’ve been using ChatGPT for a few months on a random story tangent— what’s the recent obsession with lists? Unless it’s just a me thing, everything is summarized into lists…

Like it’s never actually including dialogue anymore, it’s focused on lists or short sentences

I’m prolly gonn switch up the prompt and branch until a new version comes out

This story is on its 5th branch 💀💀💀

[tired nimbus]

A
💚

[edgy edge]

Your welcome for putting yall on rq

[chrome monolith]

give it a custom instruction, also that should be in settings. It does really like lists, its not just you

[coarse onyx]

Yeah ik I’ve got it on good custom and I’ll change them to avoid lists

Also will prolly add something about not liking lists to project sources

Making story bibles and putting them into project sources is the strat rn

[mighty mountain]

i said this to ai...........firstly arrange all english words in 1 book..........and later scramble all words and arrange it to many different books .......and keep reading it...............in the same way assume colours and make all arts................................keep reading it

[prime swallow]

I recommend tell it what to do instead of lists. Especially for things commonly in list format.

Like, if you want a sentence, "My grocery list has eggs, milk, and bread." Give it an example like that.

If all you tell it is "No lists" it may still make lists from having no idea what to do instead.

[urban granite]

Can you no longer manage memories in the chatgpt app?

[alpine pumice]

Guys, is anybody else facing very weird problems in ChatGPT, especially the projects part? It's acting extremely weird for some reason

[simple lagoon]

Allway Love how GPT CAN ALLOW ME BRING ALIVE DANGANRONPA STUFF

❤️ 💖

[eternal seal]

blood, sweat and tears of training him 😭

[livid island]

Yeah, same here seems like a temporary bug with Projects. Usually fixes itself after a refresh or a while.

[brazen lance]

hi

[strange patio]

anyone here used whop before?

[cosmic lynx]

Is anyone else having issue with goal on codex?

[chrome monolith]

maybe just go tell your crush direct

[simple lagoon]

Do this guardail crap get update again i can't get stuff i'm wish today do open Ai know they shoot then in ball on the feet with thi crap

Gpt would get more customers without this no sens censorship crap

[exotic lodge]

is anyone else experiencing slow responses with gpt 4.5? its been happenig for like a whole fuggin week now