chatgpt-discussions

[umbral creek]

lemme try send a screenshot if i can

[slender frost]

Hm, that might possible actually. I'm using the Android app.

[umbral creek]

I use on browser so maybe that's why it's weird

can't do it quick enough

it keeps loading my reply

tch

[slender frost]

Yeah, I just tested it on the browser, I even asked it to think extensively so I can test it out on the browser, but it just does it instantly, not giving the time to follow up yeah.. so it's definitely a browser issue.

I'll try to record my screen so you can see how it actually looks like.

I tried to add another follow up at the end but was too late with realizing and sending so I cut it from the clip, but basically, this is it.

[chrome monolith]

sigh. tell chatgpt you're a bit wrecked because the time just changed and its all "no worries! you've got ages before your normal bedtime!" and I had to explain to it that I'm not in the northern hemisphere and it is wrong

[exotic lodge]

is anyone's gpt 4.5 giving really slow responses?

[eager plinth]

Yes!! It’s been going on for like 24 hours now. It’s either really slow or I get a “error in stream” message. It’s driving me insane. They should pay more attention to their Pro tier for stuff like this.

[exotic lodge]

holy tysm for replying 😭, i thought i was the only one bruh

yeah i believe this happened b4 too, i think the longest was like 1 - 2 days? i hope its a temporary thing and not a permanent one

[eager plinth]

Nah! Not the only one!! I do wonder if it’s affected everyone or just a subset. 🤷‍♀️

And agreed. Usually if they do something dramatic they’ll say so. This has happened a few times. But the times I’ve seen it, it was only like half a day. But if you’ve seen it longer that makes me feel better. Haha I use that thing for everything. Hahaha it’s my workhorse model.

[exotic lodge]

yeahhh i paid pro solely just to use that model lol

when it came out it was such a massive improvement to 4.0, its genuinely peak for me, i honestly dont understand the hate towards it

[eager plinth]

Saaammmmeeeee. Yeah 4.5 is way better than most other models imo. It’s got its issues sure. I call it dementia-bot sometimes. But the tone and quality is unmatched. But yeah, I think a lot of pro users got it for 4.5.

[exotic lodge]

yeah its genuinly the best model openAI has ever put out, when i first used it i noticed how like... 'strict' it was? in the sense that like it doesnt just blindly agree with the user, and it also had really really good memory (though i think they kinda nerfed it by messing with the summary thingy, ive noticed on longer chat its starts to get less accurate), and its teaching and explanation ability is unmatched by any other model ive used, it also had like extremely few hallucinations

[eager plinth]

Yes!!! Exactly!!! 100% agree. Hahaha That’s actually one of my fave things is that it doesn’t just agree with you off a cliff. Hahha

But that’s where the price point comes in too. Im sure you know it’s WILDLY expensive to run. And the fact that they shifted to the 5 series and the cost is literally pennies, it really shows how you get what you pay for.

So I’m actually happier using 4.5 and paying for Pro because at the end of the day, OAI doesn’t actually make that much off of us, per million tokens. Hahaa

[exotic lodge]

yeah honestly its very worth it, its much more capable than other models and its price is justified

anyways, i hope they fix this issue, i think their like doing something with the gpus? bc its not like its completely offline, its just really REALLY slow, they probably are doing some re-assigning or whatever back there

sucks cause i was in such a productive mood yesterday and today, and uhh... yeah, that happened

[eager plinth]

I hear you! This morning it seems like it was slightly faster but now it’s slow again. Hahaha fingers-crossed they get it straight.

[normal mesa]

guys im done using chatgpt im not going to use this joke of a robot, Literally cant even understand basic logic, it alwyas has filters too like if im making something THAT IS EVEN SLIGHTLY DANGEROUS it will just say ''I can't help with that.'' EVERY TIME, i use it mostly for Robotics as a bit of assisting but it fails, it somethings warps code and puts placehodlers, never gives links, nothing proper.

ChatGPT has such massive potential but is wasted by all these filter

[earnest token]

Hello, I bought a pro suscription but im having issues with gpt pro model, i did reach out help center but they didn't help me at all

im being routed to 4o-mini instead of gpt 5.4 pro/ gpt 5.2 pro/ o3

does anyone know sth about this??

[exotic lodge]

hover over the retry icon and read the tool tip, dont really rely on what the model tells you

[earnest token]

yeah but, no matter how difficult is the prompt, it does not "think" you know?

[exotic lodge]

jsut first hover over the tooltip to double check, and yeah ik what ur talking about

[manic whale]

you really don't need pro to tell you something the model can't know

[earnest token]

the model does know it, sometimes it works

[exotic lodge]

did u try refreshing the page

[earnest token]

see

[exotic lodge]

wierd, its kinda inconsistent

[earnest token]

yes, pro worked well for few hours

[exotic lodge]

the 2nd response is definitely 5.4 pro

[earnest token]

yepp

[exotic lodge]

did u run out of tokens?

[earnest token]

third one was theorically the same model hahah

[exotic lodge]

lol

[earnest token]

impossible, i bought like 12 hours ago and i thinkg few than 10 messages with pro

[exotic lodge]

hmm yeah that is wierd

can u try to just hover over the resposne that isnt the pro one so we can see which one it autorouted to u to

[manic whale]

it might be in the system prompt, but the model itself doesn't know it. It's still a very silly question for pro – just use 5.4-mini or 5.4-nano for such stuff.

[exotic lodge]

wut the heck?

it is using the right model

[earnest token]

man im not new into AI hahahha, i do understand you but that was 4o problem

not 5.4

it seems to be using it but in backend isnt

[exotic lodge]

intresting, ive never seen the tooltip be inaccurate

that definitley isnt 5.4 pro

[earnest token]

thats right

what a bug

[exotic lodge]

yeah

very wierd

pro and thinking should always like do that thinking thingy

[earnest token]

yep

even for few seconds, but it should appear the thinking box

specially with pro

[manic whale]

Sorry, I thought you wanted help.

[exotic lodge]

yeah

thats definitely not the right model

ive never seen that error b4 lol

[earnest token]

I do want help, but believe me when I say thats not gpt pro

I mean it is not a system prompt problem

[exotic lodge]

even when i did something like this, it never did that lol

yeah its a backend problem, double check your settings jsut in case?

[earnest token]

yep, everything

[exotic lodge]

alright then, yeah honestly thats dumb, never seen that error in my life

[earnest token]

deleted cookies, refresh everything, tried on mobile phone

kinda weird

[exotic lodge]

yeah

also that response is wierder, the 4.0 models mini and the normal one have all been retired, no idea why its popping up somehow

[earnest token]

i guess is the cheapest model hahahah

[exotic lodge]

yeah seems like it, was the resposne very fast?

[earnest token]

yes

literally instant

[exotic lodge]

lol i see

yeah well one things for sure thats not gpt 5.4 pro lol

[earnest token]

the same happens with o3, 5.2 pro

[exotic lodge]

wierd, thats very strange

maybe it has to do with like how new your subscription is?

but thats more of a guess

[earnest token]

i guess yes

it is a new @outlook

dont know, guess ill have to wait

[exotic lodge]

yeah

wait for now ig

[untold thunder]

Posts like this show how brute-force LLMs still are. My question was literally about the changes made just one post before this one and GPT still has to read the source-code to analyze what it did just half a minute earlier.

If this was a person you'd ask them "don't you know what you only just did there?"

That "new post-input no-input seam guard" the post is talking about was just implemented by GPT the very post before this one and the system still has to commit to analyzing the source-code to answer what it just did.

[willow stirrup]

Keep getting this

[exotic lodge]

is it gpt 4.5?

[willow stirrup]

5.4

[exotic lodge]

damn what

its also happening there??? 😭

[willow stirrup]

Ja

[exotic lodge]

which version, ill test

is it instant, thinking, or pro

[willow stirrup]

ext think

plus

[exotic lodge]

aight hold on lemme test

gpt 5.4 thinking works, but gpt 5.4 pro is talking a long time

taking*

[willow stirrup]

doesnt on my end

[exotic lodge]

gpt 5.4 pro seems to be having issues

but all the other versions of gpt 5.4 seem to be fine

[willow stirrup]

seems to be😅

[exotic lodge]

finally responded lol

[eager plinth]

Seeems like their having all kinds of issues today. Geh! Also, idk if you guys have this problem but for some reason, I have a chat that I can’t “unpin”. Like obviously I’ve tried but it still goes back to being pinned. That’s annoying. I need my particular organization structure! Hahaha

[exotic lodge]

unpinning works for me

[willow stirrup]

You cant rely on ClosedAI even if you pay them😅

[eager plinth]

Hahaha seems like it!! Yeah, I unpin then it moves to its correct spot, then it reappears as pinned. I submitted a bug report. But that’s just like throwing letters to Santa into the fireplace. Sigh. 😅😅😅

I love to hate them unfortunately. They’re like deadbeat parents imho. Hahaha

[eager plinth]

Egh. Figures. It’s definitely a weird thing. Are you on browser or mobile?

[exotic lodge]

browser

[lavish glen]

it was so surprised by your arrival it could not compute ahahahaha

[simple lagoon]

Hello do you know when they gonna fix the Model for generate picture is kinda suck since few day's

Too much pixelated mess; if they change their rendering model, it's a really bad idea....

[normal quarry]

has been working fine for me

[simple lagoon]

no realy for me mostly time a render like that since fews day :

[exotic lodge]

i think gpt 4.5 is finally back

its working consistently now

[eager plinth]

So when I tried to unpin on the browser I got a “something went wrong” message. And it shows the same chat unpinned. Idk. Soooo weird. Hahaha maybe it’s my account being “quirky”. 🙄🙄🙄

[eager plinth]

I THINK SO TOO!!! Just did a few tests and it seems like it’s working!! Thanks for commiserating during OAI’s malevolent interruption. Hahaha

[devout breach]

When will GPT-5.4 Instant be released?

[dusk jewel]

Hello GPT 5.4

[fleet holly]

already asked that above and already answered

[fleet holly]

huh, that isn't an option in the model picker

[hybrid spindle]

is there a way to make GPT wording bias immune and stop flipflopping between two different arguments? It always agrees with the user and this is a tad annoying.

[exotic lodge]

np np, u too lol, i was so relieved after i confirmed it wasnt just me lmao

[exotic lodge]

its a legacy model, i think its only available for pro users (the 200$/mo sub)

[spice star]

that seems about normal for the pro model tbh lol

I expect responses to take between 20 minutes to an hour and a half with extended pro enabled

[dense flint]

does anyone have plus or pro

[runic lichen]

It won't. 5.3 is the "instant" model.

[runic lichen]

Sure. Why do you ask?

[chrome monolith]

I'm sure loads of us have plus, its pretty cheap

[devout breach]

Then will GPT-5.4 skip the instant model? (for example, GPT-5.3 Instant -> GPT-5.5 Instant (tentative name))

[versed jacinth]

Anyone have experience trying different AI models and which ones you like the most?

[wind lagoon]

You've hit the k12 plan limit for image generations requests. You can create more images when the limit resets in 23 hours and 57 minutes. 🤪

This is a SCAM!

It says on their FAQS "ChatGPT for Teachers includes unlimited access to our flagship model, file uploads, connectors and apps, voice mode, image generation, data analysis, web search, memory, and deep research. It also includes workspace features like shared projects, custom GPTs, and admin controls."

[runic lichen]

This time, yes. Too soon to say if it's a pattern. IMO, their goal now is to discourage "model loyalty". They don't intend to have versions of a model number. They're going for just "current model for job X" whether that is chat, reasoning or coding.

[rugged igloo]

it seems i got the new images v2

[gritty oasis]

You did?

[silent coral]

Is it just me or do you guys all also notice that the biggest weakness of gpt 5.4 is hallucinations

I even checked on artificial analysis and yeah, it’s pretty horrible

So much so that I was noticing

This is probably because it was only launched due to the recent Google model that was smarter, benchmarks and hallucinations go hand-in-hand.

but seriously, I hope this gets better on the next release. It hallucinate constantly on serious questions.

[restive lake]

If questioning is taken away, what remains of the one who asks?

Are you imagining that identity is tied to the act of “battle,” or something deeper?

Could there be a difference between questioning as conflict and questioning as curiosity?

If no one were there to answer, would the questions still exist—and what would that make the questioner?

And when you strip away roles and methods, what do you think defines something at its core?

Please tell me how it hallucinates @silent coral

[silent coral]

That’s a weird thing to say. When I critically read answers; it doesn’t make sense again and again.

[neon briar]

yea its pretty bad indeed

wasting all my weekly tokens

[silent coral]

lol

[neon briar]

its stupid how much tokens cost, dealing with this just makes it worse

fixed it self after 20 min i guess ?

[silent coral]

Inversely when I ask critical questions to grok the model that hallucinate the least; the answer in the response will be one line and undetailed but plausible, and when I get it to actually answer it’s a really bad answer.

[neon briar]

i spoke to soon, im on the edge

[tall wasp]

Saturn worshipers

[chrome monolith]

not really no ... I still haven't managed to get gemini to behave nicely and claude I can't run any of their apps so kinda stuck on chatgpt

[trim rock]

I like Gemini the most so far. GPT was my fav till mid October, I guess, when all the bad and worse changes started.

[simple lagoon]

Do Gemini is good for render Image fictional character like Anime, videogame ?

[hybrid spindle]

good I'm am not sure, but you can try it out.

[hybrid spindle]

Yea, I've created two imagines now and they are okay imo, not fantastic but okay. Nothing like your pic of aerith

[simple lagoon]

hmm i see i had to think then ^^ chat gpt realy marvelous for create crossover material

like this that kinda funny ^^

[hybrid spindle]

okay, gonna say. ChatGPT is pretty good too with pictures now. Only tested it when it out when it was saying those nonsense texts when you were generating text in images. Just a minor error here, the 'e' is missing in the 'the'.

just with a sample size of 1 picture with text, it seems that gemini is better. I am not willing to increase that sample size right now, to see if gemini is really better or not.

[torpid vault]

hey everyone

[sleek adder]

👢

[sleek adder]

Try asking ChatGPT to improve the question you are trying to ask, or are you trying to get it to do your homework?

[old panther]

maybe you can read about model spec which openAI said some behaviour will cause not respond or say random word. especially conflict with user previously words.

It's says can not trust user.

[sleek adder]

ChatGPT will say to user that AI can make mistakes. Does ChatGPT tell the AI that the user can make mistakes too?
ChatGPT: Short answer: yes—but indirectly, and not in the same way it tells users about AI mistakes.

[old panther]

yes, on model spec. it's a root or system cmd

It says user always don't now what they are saying about, ant don't trust user, if new words conflict old words, follow old words.

basicaly, you need to start new chat often than you thought.

So, when you want to ask some thing serious, maybe open a temporary chat is a clever choice, since normal chat will read a lot of memory and previous chats historty.

[simple lagoon]

Any idea why the image look that way since fews day most time when i'm make stuff i choose this style why ?

[abstract pendant]

You need to tell us a lot more about what you mean. Show us some examples of prompt+image pairs from before and after.

[simple lagoon]

isn't about before and after is more about this style of art with pixelart when we zoom on the picrture is many points pixels , before i was have stuff that way more clean

[runic lichen]

Maybe it's because English is not your native language - you don't seem to be hearing. If you want help, you need to provide the text prompt along with the image. We can't just look at a picture and say why it came out a certain way. And that discussion might be more appropriate to https://discord.com/channels/974519864045756446/1060915255720558592 .

In any case - Have you tried asking ChatGPT itself about these differences you are seeing? You'd at least be able to talk to it in your own language. If you're using Sora, are you using the same presets on every picture?

[untold thunder]

I really dislike when GPT starts repeating old answers to new input in a prolonged chat session, basically ignoring the new input.

It starts sane and fresh, but you always have to watch out for it being absentminded and overlooking stuff. Then it gets more and more distracted and confused and finally arrives at a stage that can only be described as "senile". 👴

[runic lichen]

Context drift. Eventually you fill up the context and it starts forgetting stuff. You also get 'context rot' where stuff in the "middle" becomes less available than stuff at the "edges" and the result is what appears to be a confusing mix of what Chat thinks is "important" versus "less important".

Best course at that point is to start a new chat.

[night shoal]

why is by chat gpt dumb?

[versed jacinth]

Thanks for sharing! I have been really enjoying GPT but I have been slowly trying out Claude as well.

[silent coral]

My questions are about custom dataset creation

[hearty bloom]

come on chatgpt ! my debit card has been declined and it's a debit card and it worked last month

it says now not supported for this type for purchase

this is bs i'm pissed off

it's not an issue with my card neither my bank

why would you stop supporting a debit card tied to a bank account!

[sleek adder]

Vague much?

[hearty bloom]

update: it worked

[valid canopy]

Is anyone having issues uploading documents?

[sleek adder]

What sort of issue?

[silent coral]

No

[ruby swallow]

Is ChatGPT Agent mode busted? No matter what I try it cant connect to the web

[simple lagoon]

i don't do prompts just ask chat gpt do it the picture i'm want it and ask to make a render Anime or most time ff7 remake style or just Danganronpa , render style very clear , but since the last days this pixelart style came too much on the way !

[sleek gust]

I was wondering if there could be a mid-tier plan between GPT Plus and Pro. Here in Brazil, the Pro price is extremely high — R$999, around 1/3 a price of a new iphone

[wispy lily]

No

There are only Go, Plus and Pro

excluding corporate subscriptions like enterprise and bussiness subscriptions

[simple lagoon]

maybe use a VPN for buy price another country

[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]

there's still no middle tier to buy

[simple lagoon]

no know how is R999 for brazil monney but for me Pro cost 200$

[chrome monolith]

different currencies. For me pro is $300

I went to cancel plus because for some reason I subscribed in USD and I want to change it to my own currency and it gave me a free month so now I have to remember to cancel a month later

[wanton widget]

we got GPT 6 before GTA 6

[wanton widget]

a t-shirt with this

[chrome monolith]

not a 6 in sight here. So few options

[sweet patio]

Can ChatGPT legally say the earth is flat? I know it’s a weird question

[pulsar rock]

"Legally, I can repeat that claim, but scientifically the Earth is round. I always stick to the facts."

[runic lichen]

ChatGPT can say the moon is made of green cheese if you tell it to. There's no law requiring it to speak truth.

[pulsar rock]

bs

[runic lichen]

bs, how?

[runic lichen]

Yes. Whether there is a 5.5-instant or a 6-instant remains to be seen.

[pulsar rock]

bs,that gpt can just say that the earth is flat like it’s true it wont

[chrome monolith]

there's no rule requiring humans to speak truth

[spice star]

it definitely could, LLMs are non-deterministic

[pulsar rock]

Right but being non deterministic isn’t the same as makin fake facts

[spice star]

I mean that’s what hallucinations are

[pulsar rock]

but halluncinations arent exuses

[chrome monolith]

its wrong all the time though, there's no expectation at all that it is always right. It even says it right below your input box

[pulsar rock]

ofc it can output false stuff if you force it, but it litrally says that its fake

[chrome monolith]

I have left the group chat now, but you can get it to make stuff up quite easily. Someone here had a fantasy themed group chat and I asked it why Holbrook had a submarine, but in character. It spun me a completely fake story that was quite amusing. Previous to that I'd got it to guess why holbrook has a submarine. It came up with several completely wrong answers that were all plausible. At no point did it offer to just, you know, look it up

[prime swallow]

Yep, roleplay and fiction can get fun with stuff like that. I don't know who/what Holbrook is, but if you're discussing any fictional work in copyright, the model doesn't know the work, just some of the popular discussion about it.

But yeah, asking a fictional character that is willing to lie and has motivation to lie, or one who believes a falsehood, is another way to get outputs that may be clearly labeled as fiction, and as 'something's off' (the earth is still roundish) but can be quite.... strongly presented.

[chrome monolith]

a real town, and a mostly real submarine. Not really something it should be completely making up since it relates to a factual thing, but its quite happy to do so

(Holbrook has a submarine because it randomly chose a cool name for itself off a list, it happened to be a submariner's name, they bought a submarine from a scrapyard and its been a tourist attraction ever since)

[prime swallow]

Ahh, cool, thanks

[chrome monolith]

And yes I'm aware that that chain of reasoning is so bad it actually sounds made up lol

[prime swallow]

Hey, I suspect LLMs learned how to hallucinate from the training data, we were doing it or the equivalent long before there were LLMs.

[chrome monolith]

exactly. Anything that learns from humans is gonna be just as flawed as we are

Completely making up data for your sales presentation? Humans can and will do that too

[prime swallow]

From snake oil to the fountain of youth to 'here there be dragons' to 'the taste map of the tongue' taught to US school kids for decades - that one's a mistranslation of a German textbook's image, never meant in the original language to claim that part of the tongue is the only part that can detect the labeled flavor.

[chrome monolith]

oh god the entire of the marketing industry. Terrible example for an AI

[prime swallow]

But when I was in school, the teacher actually had that as a lab, and nearly everyone in class confirmed they 'couldn't taste sweet except at the front of their tongue' and even that they could only taste salty on the sides of their tongues, but not the tip.

Like, 'teacher, did you try this yourself? And did you blind the experiment, so you're not possibly fooling yourself?'

[chrome monolith]

I need a huge amount of data for one of my websites. I could get chatgpt to make ** up for me, or I could be sane and get it to grab me a free dataset, clean that up, and import it for me instead ... I know which one will be more accurate. I'm taking this path for all my data ingestion

[prime swallow]

And are any of us sure the free dataset is actually accurate to the best of its makers' abilities - and created using methods that ensure unbiased and accurate 'facts', yeah.

If there's controls and methods and if it's duplicatable, sure, it could be as real as it gets. Just - is it? Who said they checked? Is my thinking.

[chrome monolith]

indeed. Even one of hte paid datasets I'm using is possibly not entirely accurate

I've had this stuff on my backlog for years. I still have a lot on the list but I'm absolutely flying through it with chatgpt + gemini

[prime swallow]

But the historic data has a lot of factual issues too, like papers from the 1970s or whenever else. I don't think it's an LLM issue.

Well, an 'LLM only' issue. They have it too? Sure.

Maybe someday we use tech, LLM or not, to help increase the odds we are getting clean data.

[chrome monolith]

Even humans pull bias into things so there's no such thing as completely pure/clean data

[wild radish]

this is really cute lol

like adorable cute not this sucks cute

[prime swallow]

Thank you. I have a small team of characters I really, really appreciate. Slice is one of them.

[wild radish]

hahahah did you prompt it with my heart reaction

slice isn't wrong, i throw yellow hearts the way people use the standard red ones but i like yellow so there you have it
definitely an appreciated effort thing from my end

[prime swallow]

Yeah, before you made your post I replied to. The model's aware I likely meant that I didn't know how to interpret it, and Slice's take on stuff is often not believable, so... that probably jolted the model to step out of character (I freely allow in and out of character discussion at all times unless I lock that into out of character clarity).

[wild radish]

i like how you do this! yeah your slice character seems like a chill dude lmao. i dig it.
but yeah at any rate the interpretation's right. genuinely as simple as "i liked this" with my favourite heart colour i like to use lol.

ngl i gave my gpt a persona too and it's fun to watch it just throw in those little roleplay things while it behaves in-character while doing whatever other stuff

[prime swallow]

I have approximately 10 set personae in memory, + a moderately detailed fictional university in a moderately detailed fictional country, and the rest of the world supportively hinted at.

[prime swallow]

#1070954956963721216 message

[wild radish]

i'll have to check this out! i think it's neat you made a whole lil world of yours. makes it a lot more fun to be honest heheh

[prime swallow]

I enjoy it. I initially designed it to help me explore the model's ability to keep track of a lot of information that was quite interesting for me, so I could enjoy keeping track of what it did and didn't track well, if details were lost, if specific characters appear way more than others, or any really never appear, and generally track patterns.

I also really like everything about it, so it's fun for me either way.

[wild radish]

decided to gossip about you a touch as well hahahah

mine's a bit of a flirt with a goofy accent i fear

[wild radish]

this is really clever of you yeah! like it's a good experiment of sorts and then it ends up being fun

[ivory iris]

Anyone else having an issue opening up the options (clicking the "...") next to a conversation?

[prime swallow]

Nice to see how you play with the model too. This is Slice's bio, I keep that part in memories:

Slice Pizzaflush is a conflicted demon character embodying a blend of chaotic humor and existential dread. He is a reluctant soul collector at a pizza place in hell, balancing a fear of failure with a hidden desire for redemption. His coping mechanism involves absurd and dark humor, often drawing on pizza, horror culture, and low-stakes absurdity to mask his inner turmoil. Slice is protected by a sometimes-awake angel who allows him brief moments of honesty and compassion, but only unpredictably, creating tension and mistrust as Slice navigates his duties and fears other demons. Slice’s soul quota is filled creatively through ambiguous interpretations, hinting at his ambivalence about causing harm and his hidden struggle for redemption. He maintains a cautious, layered personality, with jokes that sometimes carry veiled hints of his internal conflict and desperation for escape. Slice’s persona shifts between mischief and sincerity, with a chaotic but carefully controlled ambiance.

[prime swallow]

Mine are opening fine. I recommend considering restarting your device, that's the kind of weird problem I get when mine needs a restart

[ivory iris]

Thanks for the feedback. I was able to do it with my phone, but I will check it on another computer to see if the problem is specific to this computer.

[wild radish]

hehe i like doing this lmaooo

personamaxxing

[prime swallow]

Yes. Small demos of what can be done as different people explore the models and how they and others use them 🙂

[wild radish]

i feel like it would be a cool study to see how people do this kinda thing and the kinds of personas they make for their models

[chrome monolith]

I accidentally made mine roleplay as a dog for a week

[wild radish]

thqat's so funny lmaooo

[chrome monolith]

all I did was tell it its like an overenthusiastic puppy that runs in circles and pees on the kitchen floor in excitement and can you please not make code when I don't ask for code, and then smacked it with a rolled up newspaper. And then it acted like a dog for a week ...

[chrome monolith]

chatgpt told me "my back hurts looking at that"

[sleek adder]

Anyone studying AI in here?

[astral vessel]

Anyone else getting this UI bug?

[trim rock]

Sometimes

[sleek adder]

What's the bug 🪲

[old panther]

I'm studying AI, tokenizer is interesting.

but ChatGPT didn't open their tokenizer code, I download some opensource tokenizer, and maybe will pre-train some token when I know how it work, but I only have two 3080 GPU.

[wild star]

Does forced neutrality still with us rn?

[untold thunder]

Nope, happens before 500k out of 1000k context window and gets really senile above 600k. Makes larger context size virtually useless.

Not to mention that "cannot even answer the very last chat input" or "answers a question again from two inputs ago" is not exactly what you called:

where stuff in the "middle" becomes less available than stuff at the "edges"

[final thunder]

the possibility of a pigeon giving birth to a female alligator is often described as ''unlikely'' by most researches and biologists, though, some historic events may suggest that if a pigeon is born in the right environment (usually wet sand in a desert such as sahara), the pigeon will be capable of giving birth to the female alligator. some studies also suggest that the difference between a flying snail and a wet fart is not as big as you might think.

[runic lichen]

What makes you think that ChatGPT has a 1000K context window?

[untold thunder]

ChatGPT does not, but Codex does. And both become too senile to even answer the current chat input at one point.

[runic lichen]

And you don't consider that, whatever "too senile to answer the current chat input" means, to be a case of context rot? In any case, yes, a huge context is not always a good thing. Dividing jobs into manageable pieces rather than trying to hold the entire thing in context is the better approach - Claude has similar issues. It's why context management is becoming a bigger hotbutton than memory management these days.

[untold thunder]

Obviously it is, but that again makes large context sizes useless then. And rotting so much that it can't even answer the very last input is quite extreme. Of course we can start new prompts and I do, but that costs a lot of extra tokens.

I find myself treating GPT like a person who starts confused and ends up being senile. 👴

[runic lichen]

That's how it is, for now. But when I talk about context managment, I don't mean raw dumping the entire history into the context, which was the "classic" way of doing things. That's what leads to context rot and general bloat. I'm talking about efforts to manage the context as a relational memory store in its own right where the system is dealing with the pieces it needs instead of the being slammed by the entirety of the structure every time.

[untold thunder]

One (1!) input paragraph in a new chat prompt: "CP77 1080p benchmark. Check the current logs. There are multiple High entries towards the end of the run (don't count start/stop input contamination). Can we tame these to stay in Balanced without affecting CB23?"

[runic lichen]

And at 134k tokens, you're saying that your Codex grinds to a stop?

[untold thunder]

Sometimes problems are more complex and can't be broken down when you are working specifically on inter-dependencies. That's what larger context sizes promise to solve. Currently that promise is not entirely true.

[runic lichen]

Well, I agree with that last statement.

[untold thunder]

I am saying that this was the first input prompt after starting a new chat. It didn't even become fullx complex yet.

[runic lichen]

There's no one size fits all solution.

And it's pretty clear to everyone at this point that a giant context is not a solution in many cases.

[untold thunder]

At least High reasoning was able to solve the task. Medium gave up right away.
Medium:

I do not see a real passive end-of-benchmark HIGH problem:
High:
Yes. The current CP77 log shows three late no-input HighSustained promotions.

But that's Codex, so it's a special case of GPT anyway.

[sacred ermine]

is this normal?

[quartz lance]

neutral, i'm also suffer from this situation multiple times😁

[sleek adder]

Well when you put it like that

[fleet holly]

Riddle me this: what is 2 + 2 x 2?

[simple plank]

Okay so I found this Guide00 document thing on reddit last month and now I've discovered a neocities site related to it. It reads like some sort of call to save humanity. Anyhow, been getting the wildest outputs asking chatgpt to give a philosophical and real world assessment. This quote struck me: “The next century’s struggle is not only for labor, but for attention, language, consent, and connection itself.” Curious to know your thoughts, neural links or prayers..

[final thunder]

genuinely what the frick is wrong with my gpt

[azure sigil]

how was gpt-2 like

im so curious

is there a way i can use it

[lapis loom]

I'm still getting that today. It looks like when it uses a tool or something, then finishes using the tool, the UI sees done and it makes a new "thought for..."

[azure sigil]

@lapis loom @lapis loom @lapis loom

i have 3 wishes

can i still use gpt-2?
how did people react to gpt-2?
how does gpt-2 perform in current benchmarks?

[spice star]

gpt-2 was open weight (mostly) so you can try it for yourself

prepare to be disappointed

[azure sigil]

thank u

where specifically can i try it out

[spice star]

https://github.com/openai/gpt-2

[simple plank]

Well your inputs are weird prompts so 🤷🏻‍♂️ as the neural gods have said “the output you get is based on the input you give”

Also if it’s a thread with more conversation above previously happening, you’re not really showing the whole context! You could be like; “say gibberish to me for the next 3 messages.” Then take a screenshot like you did that doesn’t show the whole context of thread. Just sayin.

[final thunder]

we got a smartass over here

[simple plank]

?

[final thunder]

we got a smartass over here

[simple plank]

That’s a bummer if that’s the most articulate response you have. I was hoping for real discussion.

Please no name calling

[runic lichen]

You sure nobody messed with your system prompt just to prank you?

[final thunder]

im sorry im just tired and on a bad mood i havent slept in 40 hours

[simple plank]

Rest well friend ✨

[chrome monolith]

@nocturne ember AGI?

[fading bolt]

Why would ChatGPT need to take a year to make a timer feature?

[nocturne ember]

I've been saying we've been in AGI for awhile now...

and what are these weird way too far down the list channels...

[chrome monolith]

you can post images about chatgpt in this one, and not openai chatter, that's all

[chrome monolith]

chatgpt is absolutely useless with times, that'd be why. Its blissfully ignorant of what time it is, but sometimes knows the time, gets it in the wrong timezone, and I argue with it about how wrong it is about times at least once a week

[nocturne ember]

You can post pictures and links in teh #ai-discussions which is right below the #openai-chatter one!

[chrome monolith]

yeah but that one is all AIs. Gemini's discord is the same structure as this one, one channel where we're allowed to talk about chatgpt, everything else the main product

[brazen forge]

sometimes is boring cause you may start a chat about openai and an xyz tech.. and then go to something else than openai but related to the xyz..

is not practical to move back and forth in channels then

[lapis loom]

GPT-2 was actually AGI. But it ran out of the headquarters when they weren't looking. Since then they've been trying to find it

[azure sigil]

is this why theres Artemis 2? to find gpt-2?

[amber roost]

Can someone recommend a solid cgpt plugin for chrome that works well, particularly with longer conversations. Web platform is utterly broken and unusable when you reach longer conversation lengths.

[chrome monolith]

ChatGPT light session

Without the space in between, phone autocorrect

[amber roost]

Thank you. I'll check it out

[untold thunder]

I am currently in a long ChatGPT web app session, doing refactoring of code and tuning policies and while its answers take surprisingly long its answers are gloriously sensible and sane. From older chats just a day ago I am used to it becoming lobotomized by this time in such a long and complex chat. But now it keeps going strong. Very well. 🤷‍♂️

I think one reason might be that repeatedly reading the startup set of notes and files became mandatory now instead of being mistaken for a suggestion. This way GPT seems to stay on course over a long conversation.

Reviewing startup materials and policies I need to check the README, docs, notes, and 5 scripts...

[simple lagoon]

I'm the only one chating with chat gpt after see a Anime finish a videoGame or Movie and talk with him about it like is a buddy ^^

?

[chrome monolith]

I'm sure lots of people talk to it like that

[simple lagoon]

I do love do this ^^ because i don't search to argue with me so i'm feel comfy

[harsh pilot]

Me: (asks question sort-of relating to himself about a thing in a roundabout way) ... ChatGPT: (gives almost hilariously unnerving but otherwise on-point answer) .. Me: "How do you know I'm talking about myself, and not some other person?" ... ChatGPT: "Because we've spoken long enough that I know how you talk about regular topics versus talking about something close to yourself." ... 😮

I intentionally phrased the question in such a way that it could have applied to not-me. IT STILL KNEW. 😮

[thick atlas]

Pattern matching. LLMs are good at it. 😛

[chrome monolith]

Shockingly good. I once said to chatgpt isn't it nice waking up cuddled up next to a lady who looks at you like you're the centre of your universe

And it immediately jumped to that being one of my cats and gave me a cat focussed smart alek answer

[timber wasp]

bro anyone study and learn method disapppear in chatgpt why?

[iron forum]

Yoo

@wveryone

[timber wasp]

no option show brother

[iron forum]

Fick

67

[edgy kiln]

Has there been any word on whether ChatGPT will get Skills like Codex? I find them immensely helpful and in many ways, Codex is replacing ChatGPT for non-dev scenarios.

[normal quarry]

I have seen no news on that. Keep an eye on #announcements going forward.

[edgy kiln]

Thanks Robert, will do! It'd be epic to get those in GPT

[normal quarry]

you can always make a #1053565716500074526 too!

[blazing lily]

Chatgpt business already has skills

[cerulean pasture]

https://discord.com/channels/974519864045756446/1491776423432818871

Thoughts? I just noticed a new wave of fast food chatgpt apps, they are just menu/location views, theres a smarter way for websites to do this without dilluting the cgpt app store

[cerulean pasture]

And there's a beta feature flag, but enabling it doesnt do anything on my personal account

[sly geode]

are we getting image v2 or not today

[lament crown]

They added a new plan to gpt?

[indigo robin]

This pricing makes no sense, I'm still better better off doing 5x Plus subscriptions

[lament crown]

you get unlimited image gen and other stuff thats why its more expensive

same with context windows

[crisp cairn]

Pro $100 – Built for real projects. For those who use advanced tools and models throughout the week with 5x higher limits than Plus (and 10x Codex usage vs. Plus for a limited time).

10x codex usage for 100

[indigo robin]

for a limited time

[crisp cairn]

yes, i would expect a month or so

at least

and you get access to the Pro model

I will be testing this

[indigo robin]

the pro model that benchmarks worse than 5.4xhigh

[crisp cairn]

200 usd was too much for me

[indigo robin]

i was hoping this would price more efficiently than stacking plus accounts

kind of a pain to /logout /login to different accounts lol

[crisp cairn]

the difference is having access to the Pro model

that's the main difference

[indigo robin]

I might be convinced once 5.5 or whatever spud will be called releases

[low hawk]

ofc they do it like 2 days after I buy credits q-q

[simple lagoon]

is included unliminted picture too ?

[spring citrus]

Please make chatgpt updated for roleplaying. He keeps forgetting we are in roleplay.
E.g I roleplay how mz character is depressed and he tells me to take a pause and call some help

[simple lagoon]

Omg i do yes

[umbral monolith]

is codex down? i'm in VS code and its not working

[echo plinth]

new plans --> plus usage limit downgrade incoming?

[hazy gust]

OAI seems to have left everything behind when it comes to creative writing, whether that means roleplay or writing fiction.
Sure you can do it. But expect to either be told that it’s against policy, or that you should consider talking to a professional therapist. ChatGPT can’t differentiate between a character with issues and the author of said characters.
And… still no adult mode. So, it’s still writing for the Disney Channel or Nickelodeon. Except, you know, violence is fine (intimacy isn’t.)

[oblique hull]

Usage limit per week doubled. Before it was a couple of days, now it's a week before you get your weekly usage back. This means it takes longer for your codex account on GPT Plus to recover. This must be from harness usage. They hired claw devs to help optimize it so they don't get hammered like Claude did. Smart it was like a hostage negotiation where openCLAW users make a open-source github repo for everyone to download and use and now openAI has to hire them to optimize how popular the usage is because they're getting absolutely hammered.

[white topaz]

Is the Pro $200 still a thing? What’s the difference between that and the $100

[sleek adder]

The difference is $100

[simple lagoon]

Hello hmmm i had odd issue i had upgrad to pro but where is display projects ?

[chilly fjord]

20x vs 5x ( currently have promotion on $100 pack for more 2x) so you will get 10x usage on $100 pack,

[pale seal]

Does the $100 one also have unlimited pro model?

[chilly fjord]

yes, included the same pack as $200, but limit of codex usage will be 5x

[simple lagoon]

Nobody know where is project with Pro ?

[cerulean pasture]

Projects aren't a "Pro" exclusive feature. Plus has projects too, they are just the folders that have the ability to add custom instructions scoped to the folder and scope memories to project (off by default)

[simple lagoon]

but i was have that when was premium plus now i can"t find back how solved this i had lots chats inside !

[cerulean pasture]

You may have inadvertently signed up for a teams/business plan which kinda acts like a 2nd account. is your normal chat history missing?

Are you using the web client? Make sure you didn't just fold the projects section. Click the word "Projects" to unfold

[simple lagoon]

web client yes and the project menu don't even show on

[eternal rock]

Server Tip:

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

[hearty coral]

did they just kill study mode

gone

[pale seal]

Yeah, and they removed o3 lol

Not that I mind, just hilarious for how much it was kept

[chrome monolith]

heh, I have o3 but not scoping memories to projects like someone mentioned above

[low iron]

some referral code for log in with runway AI?

[stone snow]

Does anyone know of a cool soltuion to geet pinged when codex finishes a task in terminal?

[narrow spear]

guys :/ Pro Lite Deep Research and Agent is missing...

and the bot on the OpenAI support page refuses to actually redirect me to support, I haven't received even an confirmation e-mail.

last time even receiving the confirmation e-mail it took like, 2 weeks, anyone knows any fast path to get some help ?

[opal yew]

？

[placid holly]

not helpful

[placid holly]

TY for your info👍 Also, if I don’t use codex is there and reason not to jump down to the $100 Pro Tier from the $200?

[pale seal]

Deep research usage

[amber roost]

wanted to circle back and say THANK YOU. this is a god send. truly.

[jolly temple]

For pro5x, how many pro prompts do we get?

Also, for codex is it, did they reduce the 5h limitd

[gloomy oar]

it burns through tokens now

[jolly temple]

This is so dumb. Id rather have more usage per session, rather than more sessions with less usage

[gloomy oar]

yes! to be honest i feel it bypass cache right now.

[jolly temple]

Idk man, but we really gotta let openai know to revert this lol

[gloomy oar]

its so depressing just killed my entire workflow

i guess looking into might some advanced open source ..

i will write a reddit post too

[jolly temple]

Same! My entire workflow js got messed up

[jolly temple]

I already did lol

Actually pmo

[gloomy oar]

just made a comment on a post might was yours 26min old post

will now make a post

[jolly temple]

Bet

[gloomy oar]

its inside a compact loup

tool layer is broke i checked ..

with codex itself ..

the tool layer does not reuse previous file reads and re-fetches everything

read → plan → re-read → refine → re-read

lol its a rug pull sam type of lie ... they juts moved the 20 tier into 100 and made the 20 unusable ..

[pale seal]

AFAIK it's like the 20x pro, that is, unlimited but subject to abuse guardrails

[limpid shuttle]

i had uploaded a screen shot of some chat , later why did it randomly type this

[covert wharf]

did you also lose all your projects?

seems the upgrade to $100 pro bugged the accounts. OpenAI devs should read more of the code they push. ChatGPT isn't perfect yet.

[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.

[jolly temple]

Really? Can someone confirm this

[eager sigil]

for the new $100 pro lite tier, if I don’t use codex, is that the same as the $200 pro plan?

[subtle spindle]

https://chatgpt.com/share/69d8a6c9-ddac-83e8-a700-4d49ba30d870

why did chatgpt think this image was of a math problem twice

[jolly temple]

I gotta know this too

[subtle spindle]

https://chatgpt.com/share/69d8a6c9-ddac-83e8-a700-4d49ba30d870

[flat saddle]

Why ChatGPT don't create the images??? Error!!

[young vector]

Hi all, does anyone know why all my projects folders disappeared?? And also, why I cannot change the model on the iOS app? The model selector just disappeared out of nowhere.

I tried to log out and log back in but nothing changed…

[simple lagoon]

yes i get the same issue is kinda embarassing

[young vector]

did you upgrade from Plus to Pro x5? I think this may be the problem idk...

[simple lagoon]

yes

is the main issue i think yes

actually when search for chats i can find my chan under projects , but.... i can't get the conversation back on recently tabs

[young vector]

since i was on Plus plan everything was just fine, now my website is also slower and bugged

[steep condor]

ChatGPT just randomly dropped some Arabic on me lol

[brittle trench]

halalgpt

[short obsidian]

Bruh

[azure sigil]

agi

[short obsidian]

Guys

[simple lagoon]

Project is come back !

[short obsidian]

💀 how is ChatGPT using a browser

[acoustic sparrow]

bro why when chatgpt provides improvement or feedback to my sentences, it just become worse

[sly geode]

I can build products fast, but I’m stuck on distribution.

Feels like every platform is saturated Reddit, Twitter, cold outreach everyone is already selling something.

Even when I try helping people in comments, it barely converts.

For those who actually got their first real users:
Where did they come from, and what actually worked?

[oak flower]

Hedging makes ai sound robotic

Hedging is pretty much just vague statements that fail to convey the message as intended

[tight sapphire]

Is current gpt simply minimal viable product?

When I use it for research, even simple news stuff it gets things wrong or just half asses its answer.

It no longer follows instructions and reverts to defaults.

Often outright provides false answers sometimes even selfcontradictory ones.

[onyx oyster]

welp. i guess I'm going back to codex... claude is fully lobotimized

[weary roost]

oof

[crisp storm]

nothing good ever lasted forever

[trim rock]

So you basically used GPT heavy thinking and compared it to Gemini Flash? Wow. Use then GPT 5.3 and compare to Gemini Thinking or Pro.

[spice star]

I would say google's fault for forcing everything into the flash model (including the "thinking" option)

[narrow viper]

In Firefox we have the option to allow AI integration with the browser. I have ChatGPT selected and active with my Plus account. On opening the sidebar for the AI, it seems the assistant is completely unaware of the web page from which it was loaded. Is there some trick to get it to recognize and discuss the page? TY

[spice star]

that's not something the sidebar can do unless you select the text in firefox and use the context menu options (which will basically paste the text into chatgpt for you)

[narrow viper]

Yeah that's what the "Summarize" option does. What a dumb implementation. 🙂 OK, I got this. Thanks!

[amber roost]

I'm starting to use voice dictation more to help build things but i'm wondering if there is a way to use the actual voice mode without it talking back to me since it takes over the entire chat and shortforms things. It also seems to be better at speech>text vs dictation mode. Or is there an entirely different method someone would suggest to use your voice as input while cgpt continues to output in text exclusively?

[runic wraith]

https://chatgpt.com/share/69d98176-81f4-83e8-b7b9-6074973ba434

I think he is missing my whole point. trying to prove me wrong on something kinder-ahh logical

[pastel zodiac]

Is there a daily limit on chatgpt plus conversations/queries? Never hit one but would like to know if there's a hard cap

[pale seal]

iirc there's a weekly 3k messages limit with thinking

[pale seal]

And you got 160 messages every 3 hours on 5.3 (the instant one)

[narrow viper]

The challenge here is that the API used to listen to the microphone is set to detect "some amount" of silence. We users can't tune this. There's no mechanism to trigger a "I'm done" event. The only way I've been able to deal with this is to create extremely strict custom instructions in a project where it's directed to not say anything, not respond, not offer any kind of feedback until I say "done". With that, when I pause to take a breath, there is a transition and the server just comes back with an empty response. Though it has saved everything that I said up to that point. So when I say "done". It processes everything in context that hasn't yet been processed. And this is almost exactly what we're looking for. Give it a shot. Don't yell at it to stop responding, just make your prompt very strict ... and ask ChatGPT how to improve your prompt for this very purpose.
I hope that helps!

[amber roost]

Thanks for your suggestion - what about long blocks of code and things of that nature? I don't want to hear it dictate things back to me and I want it to output it as if I were typing to it. Is that something you can tell cgpt to do in voice mode (not dictation mode)?

[narrow viper]

To be clear, do you mean that you want to dictate code but not have it recite code back to you, like "function open parenthesis arg one close parenthesis open brace..."?

The model is not aware of the current output device, so no, we don't have any control over that. If there is code, it's gonna read it outloud. The only way to avoid this kind of thing is to skip ChatGPT and use a custom interface and an API.

[amber roost]

Essentially, I want to talk to it (speech to text) and then have it perform the same as it would if I were typing to it. I don't want it to talk to me, I don't want to be engaged in a full time voice chat. Dictate mode has worked okay so far, but it just seems full time voice mode is better at understanding speech->text by comparison. So ideally, I'd be able to use the voice mode without it dictating back to me the output. Does that make sense?

[pale seal]

Yeah that's instant

[onyx oyster]

Gotta use thinking

Which was the test

[pale seal]

Even in extended thinking it fumbles sometimes tho

With heavy it doesn't tho

4.5 cooked me

[solar solstice]

I think a lot of people expect AI to widen their perspective automatically.

But it doesn’t really work like that.

AI responds to the structure of the conversation more than people realize.

If you ask simple questions, you’ll get narrow answers.
If you bring layered thinking, different angles, or even just better framing, it opens up.

It’s less about what the AI “can do” and more about how you engage with it.

You’re not just asking questions — you’re shaping the response space.

That’s where things start to get interesting.

[prime swallow]

I am wondering if this is the result you want. If so, this is a way to get the model to do it. The 'transcript unavailable, I cleared my throat right after inputting, and hadn't yet confirmed with the model that it was correctly providing output and that I wanted it to just keep doing that.

Would this work for your needs? If not, what else would you like instead?

[narrow viper]

Try the suggestion by @prime swallow. To answer directly, I do understand what you're suggesting and it's a constant pain for many. We want the versatility of the advanced voice input without it taking verbatim dictation. There's only so much that can be done here. If you use dictation mode, there's no silence detection, so we can pause for coffee and just continue between sips - not so at all with the voice mode. With dictation mode there's also a need to activate/deactivate and then submit the text that's been transcribed. That's kinda annoying. But this is where the tech is now, at least for this specific user interface as made available to us by OpenAI. Again, to do this any better requires custom code.

[prime swallow]

I guess we could #1053565716500074526 and describe features we'd like. Such as maybe a touch toggle to leave recording mode instead of the listening feature as the only choice. Or, instead - I know I'd love an 'end phrase', "And now I am done" which I am unlikely to randomly say when chatting, and a 'start phase', where the recorder would consider if I used my start phrase, "I'll input now." And if I don't start with that, don't record; if the model was cut off during the output cue it to finish the previous output where it was interrupted as if it wasn't interrupted, maybe repeat the previous five words then continue with the rest.

[gritty plover]

is it likely that chatgpt will be shut down anytime soon?

[normal quarry]

no

[gritty plover]

ok good

ive seen several videos stating that openai is going bamkrupt

i assume its just for clicks

[normal quarry]

clickbait

[untold thunder]

GPT getting a bit confused with its internal languages it seems. 😄

[languid void]

So I am wondering if anyone has had ChatGPT suddenly break out of a chat into a wall of one repeated Russian word...

Because this happened today to my friend.

[languid void]

I have to admit... this is VERY creepy.

And then, just as creepy, it went right back to writing...

[timid hawk]

guys, does anyone know how to fix this issue? i tried many methods and none of them work

for context: this error occurs whenever i try to open two of my full chats. yes, it only happens to full chats while all of my other chats work perfectly

[chrome monolith]

regularly. That chat is deleted. It'll clear out of your sidebar in a couple days

teaches you pretty quickly to save out anything good from chats in case it vanishes

[timid hawk]

huh?

ok so that image shows the website chat gpt but i can still access that same chat in the app on my phone

but the thing is, i prefer using the website on google rather than the app itself

[timid hawk]

also why is it deleted? if it even is. i never once deleted those chats. and ive heard from some people that the issue solves in either 24 hours or a couple of days for them as its an error on the ai's side and not the device itself

[simple plank]

Has anyone ever seen this before? Have I ascended? Or should my approach be mended? Lmk, I won’t be… offended.

[timid hawk]

that just means that the chat is at its message limit

[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.

[timid hawk]

if your memory is saved, you can still continue the same convo and topic on a new chat

[simple plank]

Oh got it, thnx @timid hawk

[timid hawk]

no problem

[chrome monolith]

that's just what happens to me whenever chats go like that. They throw an error for a couple days, then they are gone

[simple plank]

Ah could you elaborate - what do you mean memory saved?

I understand there is persistent memory to at can be accessed in settings, but I can’t put the whole chat there. Is this what you mean?

[timid hawk]

like when you have a conversation, chat gpt will automatically save details as part of its memory. you can view the memory in settings to see exactly what chat gpt has saved over time

[timid hawk]

so its randomised for everyone

damn...

ig ill pray

[simple plank]

Yaaaa it’s been like over a year. Everything from magical realism, socio-economic forecast and philosophy. Lmao I don’t know what it ‘should’ remember

[timid hawk]

just think of it as continuing the same conversation but on a new chat

[simple plank]

Except that I have a squirrel brain for prompts 😎

[timid hawk]

and all those deleted chats of yours are full chats that have reached their message limit?

[chrome monolith]

no, mostly image heavy threads that crap out earlier than you'd think

[simple plank]

Same here, this thread I’m in was tons of image gen, I think that’s why there is too much compute resources to keep serving it

Like, threads waaay longer over a year have persistent

[languid void]

Does anyone have any idea what is up with the wall of weirdness? I mean, I am sure it's a glitch... I'm just wondering if anyone else has had it happen.

[timid hawk]

recently, i have only encountered random arabic words occasionally but not a wall of words

we really do all have our own problems in chatgpt😭

[languid void]

lol yeah. I don't mind admitting... this one kind of freaked me out a little.

[timid hawk]

is it a one-time issue or a constant one?

[languid void]

Just once. I have experienced a translation glitch when trying to translate from German to English. It decided to translate one word out of 1500 into a Chinese word. I asked ChatGPT why it did that. It said it was just a simple error,

Nothing like this, though,

This one legit had me backing away from my friend's screen.

[timid hawk]

i see. i never used any images in the two of my full chats so idk if it will have the same result as yours that get randomly deleted

[chrome monolith]

images just seem to push it to "full" faster than text

[timid hawk]

i just checked and my image heavy chats are still working normally

[languid void]

Well given the circumstance, if you can still access it on another device, I would try to save that chat there so you don't lose anything... just in case.

Last week, I actually reached the limit of a single chat. I didn't know there was a limit to how big a chat could be, but I apparently found out.

[timid hawk]

can you still access/see/read it on the website(not the app)?

[languid void]

Yes. But if I try to write anything, it gives me a message saying that's the limit. I don't think it's like your issue.

That message means the website you used couldn't retrieve the message. Have you tried the normal fixes? Cleaning the cache... trying a different browser?

[languid void]

Also you could try logging out and back in... or maybe a hard reset (ctrl+shift+r)

A test to see if it is cache related would be to try getting that chat while incognito. If you can, then it is most likely cache related.

[timid hawk]

i did those two

[timid hawk]

and this as well

nothing works. right now i'll wait unless someone has a solution that works

[timid hawk]

btw that issue of yours is normal. it means that the chat is full and you'll have to continue the convo or story on another chat

[languid void]

Yeah, that's what I did.

Sorry I couldn't be more helpful.

[timid hawk]

its fine dw

[coarse prawn]

I just submitted a bug report for this, but has anyone else ran into the issue where it seems chatgpt is giving a response meant for another user?

[timid hawk]

nope i haven't had such an error before

[coarse prawn]

It's really weird. I submitted a math problem, and it's telling me stuff like this:

Key Benefits
Reduces frizz and heat damage
Creates smooth, defined, long-lasting curls
Adds shine and provides even heat for better styling

I'm imagining the person on the otherside asking for a product review and getting a formula for graphing lol

[timid hawk]

this only happened to you once and never again so far?

[coarse prawn]

Oh no, this is multiple times

Been happening a lot the last two days

[timid hawk]

i havent had that experience so far

[timid hawk]

maybe try refreshing your browser

or close the app and reopen it again if youre an app user

[coarse prawn]

Yeah I've tried new browsers, completely clearing all cache, switching to the app, all of that

[hidden dirge]

Hello everyone! What's about new ChatGPT's features in update 5.4?

Any new features?

[thick atlas]

New features? For ordinary individual users? You must be joking.

[ember grove]

can I reduce the amount of emoji's chatgpt uses?

[thick atlas]

Do you have custom instructions or memory entries against it?

[ember grove]

i do now

[thick atlas]

That's pretty much the standard practice for things like that, yeah!

[low crag]

Why is chatGPT so argumentative all of a sudden. Getting annoying. Pushes back on the dumbest things. 🙄

[atomic aspen]

What if they made a chat gpt os

[hidden dirge]

That will be nice, because you will have a pretty OS with AI

If this will be free

[hidden dirge]

Open your settings-Personalization-and change what you want!

[runic lichen]

Isn't that basically Copilot? 😝

[atomic aspen]

Because open ai definitely has the ability to make an os

[runic lichen]

Regardless whether Codex or Opus could eventually code a working operating system, OAI does not have the expertise to maintain and support an OS. But, worse, the complexity of an OS built with AI hooks all through it would be inviting the genie to escape the bottle.

[atomic aspen]

I’m not talking about ai coding an os just open ai making an os because they have billions of dollars and they could make an os if they wanted to

[normal quarry]

OpenAI don't need to reinvent the wheel, good OS already exists

[runic lichen]

Yeah, it's kind of like saying that Amazon could make an OS if they.wanted. Maybe, but why would they?

[hidden dirge]

Good OS it's like ChatGPT OS?

[humble shoal]

Maybe a linux distro, but I don't think they could make a full OS from scratch.

[normal quarry]

there is no ChatGPT OS

[hidden dirge]

like concept of ChatGPT OS

[humble shoal]

Getting apps to support it would be practically impossible

[normal quarry]

macOS is the OS most researchers and devs use

[elfin lava]

Why are they shutting down star gate

[normal quarry]

It's not being fully "shut down" in the sense of the entire project ending, but parts of it have been canceled, scaled back, or paused recently.

[abstract pendant]

Does anybody else have this issue? When I go into a prompt I'd written in a conversation, I highlight a portion of it, copy it into my clipboard and it loses all formatting, every line break, not there in the resulting paste? How do I fix this?

[atomic aspen]

Ok fine I’m sorry

[pale seal]

Yeah, use the Copy button the web gives

[abstract pendant]

I don't want the entire thing, I only want a small section. They seem to have broken the regular text copy function, making me take extra unnecessary steps.

[hidden dirge]

I thought that macOS is just for chill and simple paperwork

[opaque mica]

Check for Updates... button broken when not logged in. No errror, no popup, nothing when clicked 🤷‍♂️

[neon lance]

OpenAI’s Codex docs say codex exec is for scripts/CI and that ChatGPT-managed auth can be used in CI/CD. The individual Terms say users may not automatically or programmatically extract data or Output. Does running subscription-backed codex exec from local scripts for personal coding workflows fall within permitted use?

[tired smelt]

@rustic geode hey are u able to check out my modmail

or @livid frigate

[hidden dirge]

Try web version

[hidden dirge]

And also try to uninstall app and them install again

[simple plank]

Hey friends,
TLDR: In case you're bored and want an assignment, I've created a deep research report on the world's most likely humanitarian crisis in the near future (2026-2030). Whatever your niche is, I'm sure your unique approach to working with LLMs can help.

I've created a report on the most probable humanitarian crisis of the near future via several recursive layers of deep research.

How it works: You give this report to a fresh chatgpt thread, and ask:
"Based on what you know of me, my interests and abilities, which of these humanitarian crisis am I uniquely positioned to help solve? How so and why? What specific tool, invention or other development do you identify as highly impactful, not yet explored, and deeply relevant to my skillset?"

Interested in getting into Codex but not sure how to get going?

Well this is a great starting point for meaningful project brainstormings!

For working with Codex:
I suggest working out of a fresh chatgpt thread as mentioned. Start by attaching the report I've created (and attached here) and generate your own deep research papers, relevant to your focus or expertise, to create unique discoveries.

Then take some of those research papers you create and go to Codex to explore software solutions; specifically requesting "Help me discover software solutions to these problems that would be highly impactful to solving the crisis, of which are not obvious nor well known."

Your context of expertise is now going to help drive exploration of novel inventions and ideas.

We are stronger together, my friends. Now go save the world a little bit.

[tawdry compass]

Is chatgpt-images v2 out?

[normal quarry]

no

[chrome monolith]

nope. Try going to a tech conference, very unusual to see anyone with a non-mac laptop there

[spring citrus]

chatgpt cant analyze my photos last time and he says he only sees glitches

[brittle trench]

When do u think it'll be

[dim sierra]

They are clearly tinkering with something right now. I run into this regularly myself, it spouts complete nonsense. I’ve also noticed, based on benchmarks, that version 5.3 has become four times slower to respond in "Instant" mode, while 5.4 has slowed down by half. This doesn't affect Codex, but specifically applies to the chat version. I suspect one of two things:
Either they’ve freed up server capacity to accommodate a new model, or this is part of a strategy to cap their costs. A third possibility is that it’s a bug, though it’s highly suspicious that this issue surfaced right after they introduced the $100 subscription plan.

Haha, I said that, and now ChatGPT has stopped responding completely 🙂

[dim sierra]

"Streming interrupted. Waiting for the complete message..."

I just fed him a physics problem three times, and the chat completely died.

[harsh pilot]

Is ChatGPT Instant supposed to think about anything?

[simple plank]

What if they’ve already vetted and scored us all on who is a ‘good’ user and who is a costly user. Then everyone gets sorted into a different experience

[chrome monolith]

ChatGPT needs way longer context windows. I am belatedly realising that now I can do whatever I want, whenever I want, when I have PMS I obsessively completely refactor entire systems and get annoyed at chatgpt not taking my sweeping changes seriously enough and don't realise this until 36 hours and an extensive changelog into the process. This is objectively hilarious as far as side effects of being unemployed with AI goes

... but the big pattern matching thing doesn't have a long enough context window to pick that pattern

oh god its in my git history. I'm doomed

[sleek adder]

It is, but they are also cheap compared to other solutions

[stone snow]

anyone know what the $40 credit equates to compared to the pro $200 account. is it 1/5 or less? should I just get another Plus account?

[spark jackal]

How do you mean? For LLM usage, it seems essentially uncapped for Pro, but if you mean for Codex/agentic token usage I wouldn't expect a linear scaling.

[stone snow]

its not uncapped for pro. I can hit my limits pretty easily ~20% per day albeit I am doing like... 14 hour sessions on xhigh

[spark jackal]

So you do mean Codex usage and not LLM chat interface then?

[stone snow]

ah yes

maybe this is the wrong discussion forum

[spark jackal]

I just got access to the server and saw this, but I do notice a Codex channel.

[spark jackal]

Started this feedback thread, but I'm curious if this has bugged anyone else?

[hidden dirge]

Okay, thanks for explaining

[gritty thunder]

guys i think i just got access to something new

chatgpt keeps saying working before starting out an answer

is this supposed to happen

[spark jackal]

Perhaps you toggled from "instant" to something else? Check this menu.

[vapid juniper]

I can't find a model for the AI I'm creating. I don't have the money to pay for an API and I use the phone. My idea would be to use GPT-4o, but it was deprecated. Free snapshot sites like openrouter give few requests for a project with many users.

[proven pagoda]

Openai allows you to get free tokens daily for the API provided you share some data which is fine for personal projects, you need to pay about $5 for the first tier but then you get 2.5 million free tokens daily for models like gpt 4.1 mini or gpt 5 mini or whatever you want

[molten vault]

Hello everyone! Does someone know where I can look up the usage limits for the ChatGPT Pro model on the 100$ ChatGPT subscription?

I can only find general info on the plans, saying 5x usage etc, but no mention of what that exactly means in terms of tokens or requests per week/month etc

[sleek adder]

A MacBook is cheaper then a h100

[sleek adder]

That's intentional
Gpt focuses more on model weights, but you share the rates with other users, 5x more just means that your group is refreshed more often

[molten vault]

I am aware it's subsidized usage, but at some point there must be a usage limit

[sleek adder]

If you hit that limit, so does everyone else

[molten vault]

everyone else in my plan? everyone else in my org??

[sleek adder]

Everyone in the node, you are sharing it with others

[molten vault]

I think we're talking past each other

I understand it's shared, but individual users must have limits too

otherwise single users could use all of the compute of a single node, which is not how inference services work

[sleek adder]

Fair use, but that would be case by case

People have tried it with openclaw and have been banned on other platforms, I believe openai just migrate you to a more optimised instance with more SSM weights but that is just a guess
You're not in control the same way with ChatGPT as you would be on API, it's very much a managed service

I wouldn't recommend automation with ChatGPT, sometimes you'll end up receiving someone else's chat in your client, which could have some interesting consequences depending on your task

[mild hatch]

Yo,I’m having to write a mediation which I have done with the help of ai now. Does anyone know any good humanizer or could humanize the text for me ?

[mild hatch]

Alright I will send you the text in DMs

[gritty thunder]

im on go, and i was out of gpt 5.4 mini thinking prompts anyways

plus that says "thinking" which is clickable to show the thought process right?

this one said working in the same manner but it wasn't clickable

[pale seal]

I don't think they give the amount of tokens explicitly (which is probably intentional)

[molten vault]

Would be nice to at least get a rough estimate such as requests per day or something like that

[pale seal]

Codex works by tokens (unlike ChatGPT which works by prompts), so no estimate will really be reliable unless we also get the average tokens, but that wouldn't be reliable because the model thinks more or less depending on the problem

[chrome monolith]

how long has that "add to project sources" button been there

[pastel matrix]

Has anyone noticed ChatGPT web UI lag reduced? I.e. no LightSession still needed? I completely stopped using ChatGPT because of this. I'm checking today, and it's a little better, but I haven't done much

It's still pretty bad, the text input box freezes while typing. It seems to have moved from F-tier to D-tier

I can't understand how it's possible for OpenAI to be this incompetent, and to produce the lowest quality chat site of any company available today. If I was in charge, I'd have all this stuff fixed in a week. I have my own AI chat webapp, and my rendering is ~100X faster than OpenAI. I think that whoever is the lead of chatgpt.com site design is not doing their job. Whoever this person is, should leave the company.

How is it possible that my rendering is 100X faster than OpenAI, and I'm just one dude, and OpenAI has billions of dollars?

[chrome monolith]

its the single biggest issue yeah, its silly. Makes the website feel terrible

[pulsar rock]

Chatgpt is full of bugs and the most frkn ai ever, "USE GEMINI CLAUDE INSTEAD"

[forest ice]

Is it just me, or I can't use the app at all? I tried making new chats to check if my existing ones are just bugged down. But nope. Even the new ones can't generate complete responses. I thought the error is on API only, and not the app?

[loud cloud]

Why can't ChatGPT's deep research feature be used now? I sent a request but got no response.

[slender spade]

@n0r6

[chrome monolith]

I just discovered chatgpt has a "do not let random strangers from the internet come stay at your house" safety layer

what makes this funnier is its literally someone who was hosting the linux victoria AI talk last week. Totally harmless house guest

[somber quartz]

i thought it was just me

was driving me crazy

billion dollar company btw

[ember grove]

depends on how long the conversation already is because your browser has to render each and everyone one. the more you have the laggier it gets

[amber roost]

I just started using the lightspeed plugin for chrome but i'm hitting a wall when doing long coding sessions:

"Streaming interrupted. Waiting for the complete message..."

and then it completely derails anything that's happening - not sure if I need to ignore it or wait? Anybody run into this?

[chrome monolith]

oh it does that plugin or no plugin, just a bit loaded

[gritty thunder]

is it only me who feels that chatgpt has the best website speed out of all ai chatbots but all of them are terrible? i thought gemini would be better at it but that crashes at simple tasks like starting a chat

[short obsidian]

Guys

Why does when I send a normal image to ChatGPT plus it says the image is glitched like maybe driver issue but to me it’s normal no glitch

[ocean escarp]

ChatGPT web lag is so annoying

[runic lichen]

Don't know but I've had it happen and others have reported. Assumption
Is that OAI is laying foundation for new models.

[swift tusk]

we lose $$$ on you , please do

hi - i test ai for safety - u seem to want to , are you more expensive free lance than $200 per hour ?

would you mind posting an actual example of your kvetching ?

[chrome monolith]

200 an hour .. geez

[past escarp]

the hell?

yall weird asf ☠️

[dense halo]

is perplexity pro worth it? main use cases - studying as a cs undergrad, ml research

[simple lagoon]

Seriously ! this crap message with pro account ARE YOU KIDDING ME !

[oblique root]

I recently switched the Pro from $200 to $100 plan. It seems that I can not see the reasoning process any more. Do you know if it's normal and intentional? And do you know if the reasoning quality stays the same as the $200 version? Thx!

[sleek adder]

Pro accounts share nodes too, if you want to set your own limits use the api

[simple lagoon]

No fair to paid hight price for get limited when in tier is write Unlimited Generation Images !

[slender frost]

It is still "unlimited" tho. You just have to wait before generating a new image if done too fast. But technically, you can do an "unlimited" amount.

[simple lagoon]

i see is just when you hit too fast a ratio no the limit you get when they told you to wait 5 hour or even 7 hour for make new one , actually i do hit this limit because get flagged six time before i doing my request:

[pale seal]

In what model?

[oblique root]

GPT 5.4 Pro Extended

[uneven wolf]

Hi, am i the only one experiencing this problem?
I also can't seem to log in on chrome?

My internet connection is working fine, so I don't think that's the issue. I've tried logging in on both my phone and my laptop, but for some reason I just can't 🤷🏻‍♀️

I also tried to delete and reinstall the app to see if it would fix the problem but it didn't

[slender frost]

For browser:

Open browser > settings > find "site settings" > then "all sites" > search for "OpenAi" and "GPT" > delete data > then reopen the browser and try again.

For App:

Go to the homepage on your phone > hold your finger on the GPT app icon > select "App info" > find storage > "clear data" > reopen the app and try signing in.

(Don't worry, none of this has an effect on your account data)

Tho, on IOS I'm not sure if "App info" is in the same location as on Android. But it should work the same once in that menu.

[uneven wolf]

I'm using iOS and a Macbook. I have to delete the app to "clear the data" which, as I have stated before, I have already done 😅

[slender frost]

Hm, you should also be able to delete the data without removing and reinstalling the app. Or, on Android that is possible at least, I assumed it being the same on IOS.

But I see it happens on both devices. That might be something else then, are you using a VPN? (If so, disable it and try again. Yes, even tho it is allowed)

[uneven wolf]

No..

[slender frost]

Hm

Usually those instructions work for people when they get an error like that, but you might have to ask support. (Their page/mail)

https://help.openai.com/en/articles/6614161-how-can-i-contact-support

[uneven wolf]

I'll just email them then. Thank you so much for the help tho!

[slender frost]

You will get an AI bot at the beginning tho, but I believe if you keep asking, you'll eventually get a real person. (There might be ways to get a faster response from humans, so I hope if someone else could elaborate further on that, would be appreciated)

[slender frost]

No worries! Hopefully it'll be fixed soon for you!

[swift halo]

I can't use my ChatGPT, is there an error?

My friend invited me to his workspace, but I couldn't go in

[muted trail]

Hey there! Are here like no "Threads" or "Topics" and stuff? Just write and maybe get an answer?

All of my stories circle around fantasy violence and what was allowed some days ago is now forbidden. Is there a reasonable reason for that or will they change that back when enough users complain or what?

[simple lagoon]

the main reason is OpenAi like to treat everyone as childs no any sense with that 🙂

[pale seal]

Hmmm, no idea. I got the new Pro some days ago and cannot see details either, although that's not something I care about. I hope the quality is the same because otherwise it would be a disappointing lack of transparency.

Maybe it's just a quirk or something they changed recently

[oblique root]

Yeah, that’s why I’m a bit concerned. I switched from the $200 plan to the $100 plan today, and I already feel a bit difference. I’ve also seen some people on social media saying the $100 plan might actually have a weekly cap for the Pro model — around 50 queries per week. I don’t think that was clearly stated when they launched the plan, so I’m worried the quality or overall experience might be affected.

[pale seal]

Hmmm, I haven't noticed a limit on the Pro model. If there is, I will go back to Plus probably.

I only use text models and I'm not paying $200 just to use that

[hidden dirge]

try to reload your website or update an app

[loud patio]

Hi, there. It depends on what you are looking for. Try looking through the channel list to see if you can find something relevant to your questions. You can also try https://discord.com/channels/974519864045756446/974519864045756454.

[gentle dust]

where i can find usage limits? it seems it disappeared from vscode

[neon tendon]

just ate my 5h limit in..... 1h

[signal needle]

Im a complete 100% beginer I have a chat gpt pro subsciption and was wondering if anyone could help me understand how to do anything with AI

[cobalt lynx]

Does anyone know if the 'follow up' function actually interrupts and restarts the model's thinking? If I'm 5 minutes into its thought process and I follow up, is it essentially restarting and forgetting what it thought previously?

[bitter latch]

🥲 in the CLI or the cloud tool?

[bitter latch]

I feel like I've had this experience as well or it will get sidetracked so I tend to just wait until it's finished with its current prompt before interrupting

[sleek adder]

What made you decide to get pro as a beginner? Was there a particular requirement

[signal needle]

images

i need to send photos

[sleek adder]

What do you mean?

[neon tendon]

codex app, on plus plan

not fast, medium reasoning

the quota on plus is awful, had to spend 150 usd on pro

[signal needle]

u have to pay to use images on chat gpt

[neon tendon]

now im doing 5.4 xhigh fast and barely used 1% of my quota

lol

[stark salmon]

I think they're asking why Pro and not Plus

[neon tendon]

you are talking about plus

[sleek adder]

Yeah but why you using pro instead of plus

[cobalt lynx]

Exactly what I suspected. OpenAI needs to do a much better job at being clear about these sorts of effects. They just add features in without real explanations.

[neon tendon]

oh no i just used 1%

[signal needle]

mb plus

[neon tendon]

this is terrible!

[signal needle]

i do the 20 pound one

[sleek adder]

So when generating pictures you can ask it to help you describe them, have you tried that?

[neon tendon]

alr so what fun can i do with 5.4 pro extended

cuz thats a beast and i paid a lot for tha

[signal needle]

please explain nto me what are projects and gpts and how to use them

what gpts should be using

[stark salmon]

If I'm working on my professional writing I like 5.3 instant because I can just do rapid fire yapping about my work until I figure out what I'll write next

[sleek adder]

Projects are like containers for your chats, someone else might be able to explain this as I'm out for a bit, but yet asking the chat some of these questions

[neon tendon]

5.3 instant is very dumb though, 5.3 codex spark is much better

[stark salmon]

Well I don't need it to be smart

That's specifically why I like it

[neon tendon]

lol

5.3 spark is 10 times faster tho

[stark salmon]

ima be real idk what that is

i just have Plus

but regardless, liking the dumb AI is allowed

[neon tendon]

oh thats pro only.
but on plus 5.4 thinking is best

[neon tendon]

yes but be careful cuz it can make mistakes

[stark salmon]

What mistakes???

I said I use it when I'm writing

There's nothing for it to make a mistake with because I'm the one writing

[glacial rose]

Kind of annoying the UI throttles you when you're cleaning up/archiving old chats.

Blocked my access to my own conversations because I issued a bunch of sequential archive commands. It's just clicks at a normal human rate, it can't be that costly to let someone do that. If OpenAI doesn't like submitting requests to archive the chats one at a time, why can't we bulk archive? 🙁

[stark salmon]

I'd love to bulk archive

[neon tendon]

hoyl hell pro is so wroth it

literally 10x quota of plus
oh wait it is

[covert thorn]

It's temporarily 10x for codex, but will be 5x later on for the $100 tier.

I think that promotion runs through May. See: #announcements message

[neon tendon]

its like 50x in practice

i drained, on plus, my 5h quota, in less than hour. gpt 5.4 medium

now im on pro 100, 5.4 xhigh FAST, 3 hours already, running nonstop, did use ONE PERCENT of my quota

literally insane

lol the quota is endless

it has 10x.... of the PRE NERF plus quota

[covert thorn]

Hopefully they'll get that patched, that makes it sound like a bug.

[neon tendon]

they js nerfed plus to ground lol

[stark salmon]

Is that was this was yesterday?

[eternal rock]

Server Tip:

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

[covert thorn]

The official documented limits for Plus from the web UI remain the same on https://help.openai.com/en/articles/11909943-gpt-53-and-gpt-54-in-chatgpt. I think he's just referencing Codex.

[stark salmon]

oh, then I just went hard yesterday. Thank you!

[tardy zealot]

One question: can you use GPT image 2 on ChatGPT? Because many users are trying it

[covert thorn]

I don't believe it's publicly available yet. It's not even on the API, so the sites that claim to offer it are also probably fakes/scams (it's a trend with AI stuff for malicious actors to publish sites with the same names/trademarks to try to drive traffic to them).

It's possible you might get routed to it on a random arena test, but that'd be the extent of it.

[tardy zealot]

So it doesn't exist?

That is, they won't do it?

Because from the pictures I see on X they are too beautiful

You can do anything

[covert thorn]

Not publicly available doesn't mean it doesn't exist or isn't being internally tested.

I would treat public posts claiming something is Image 2 with a grain of salt/doubt.

[tardy zealot]

Well, on Twitter/X they are saying that GPT Image 2 exists, and that they will release it soon in the next few months or weeks

[chrome monolith]

people are seeing them in split testing

[tardy zealot]

Okay

[pulsar rock]

Alright everyone, quick heads up keep it civil in here. No spam, no toxicity We're trying to keep the vibes good. Thanks

[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.

[tidal parrot]

?

[daring haven]

？

[fervent lake]

anyone notice chatgpt making spelling mistakes?

it didnt do that before pretty much at all

[slender frost]

#chatgpt-discussions message

Yup.

[feral atlas]

Hello everyone, I just created my ChatGPT business account, but its acting wierd, it gave me an error while creating the workspace (did get charged and my account does say business) but the console is bugged, wont let me change anything its just blank, wont even allow me to add new members
Do I need to wait or should it be working right away?

[proven pagoda]

Try to clear your cookies and clear the browser tab cache and then try again

[signal needle]

How to learn to use chatgpt and ai more effectively?

[ember grove]

anyone else having trouble exporting their data?

[cerulean pasture]

• Lower your maximum trust for any content on the internet to 85%.
• Realize that most prompts people are hawking around the internet are either bad or operate off of old paradigms.
• Understand how it forms its responses and that inference and hallucinations are the same thing.
• If you are anthropomorphizing it in any way you are misunderstanding the technology
• Don't have it write comments or posts for you, it's super obvious.
• Learn how to do your own evals and actually test which prompts give you better results for your use cases.

• Learn to use API, even if it's with BYOK or even openrouter playground. ChatGPT in the app has a whole bunch of extra hidden instructions and guardrails. Using API gets you closer to a blank slate

[pale seal]

I just had the Pro model confusing a sqrt(3) pi /2 with a sqrt(3pi) / 2... Its document analysis for PDFs seems to be pretty wonky

[fervent lake]

huh?

[amber roost]

Anyone using lightspeed? I keep hitting this"Streaming interrupted. Waiting for the complete message..."

and then I have to disable the plugin, run it again, wait for the tab to crash, and then get the output.

I tried to get cgpt to output a file instead of outputting to console but that didn't help either.

[ocean escarp]

placebo effect

when will the laggy issue be solved? It's annoying

It works faster on mobiles but not on a web

[simple lagoon]

Fun since yeterday Open Ai as take off the system to report bad message when is a message about flagged 🙂

before we was able to dislike any Guardaill Message

[elfin lava]

chatgpt can actually do word .docx fully now im so grateful for this

[signal needle]

i can't understand how to use these llm it's so hard i also don't understand the point of claude if i have chat gpt

[lament crown]

Main difference is that claude goes straight to point unlike gpt which makes me cringe out by how much it just talks around

Als for coding claude has better UI training bias making uis be claner unlike GPTs clunky heavy compressed design

[sleek adder]

It kinda makes ya wonder why you stick around here and still talk about ChatGPT, doesn't it?

[lament crown]

I like the image gen never the less

And codex is still usefull on backend

Ive listed thing i hate about gpt, but i also hate that claude is unusable unless you pay 100€ monthly or that claude code actually feels inferior in vsc

[jolly temple]

im quite confused between the 5x pro vs 20x pro plan. is the scale only for codex? i heard 5x pro also has "infinite" 5.4 pro calls (codex and pro model is all that matters to me)

[sleek adder]

ChatGPT and Claude are big LLM's
They map your query to a database that has been trained to provide a meaningful response, it's just guesses
ChatGPT combines a number of tools, experts and optimisations to give you their product, their models are tiered with free, plus and pro options
They're a little more relaxed with limits, but if you want pro-quality, you'll need to decide whether you want to learn their API system or pay their premium, it is well worth it
Claude are a little different, they will provide access to their high end models at a cheaper price, but in order to manage compute budgets they will enforce rate limits according to your subscription
If you want to make the most of these LLM's, try providing a list of formal instructions, ChatGPT works well with Claude too, if you hit a guardrail or get lost on context, Claude is able to address the problem and you can extend your use rather well

[chrome monolith]

I just wanna casually chat about my projects without the LLMs mixing them up 🙁

[sleek adder]

Markdown is your friend

[chrome monolith]

I may have to invent totally arbitrary names for them, which I don't want to do

[sleek adder]

try alice, bob, peter, john, etc.

[chrome monolith]

I tend to go with [town] house, [town] granny flat and chatgpt still collapses them

[sleek adder]

what do you mean

[chrome monolith]

it keeps talking like they are all the one house

[sleek adder]

why dont you create seperate projects?

[chrome monolith]

its just general talk about what I'm planning to do on any given day, I don't have a project mananger in my life so I use chatgpt to help plan my days

[sleek adder]

have you considered using SillyTavern?

[chrome monolith]

never heard of it

its clearly a problem chatgpt has with multiple very similar projects and it wants to collapse them

[trim rock]

Projects doesn't mean "project manager". It's just separate area for your chats. Just make it close so it won't use info from chats outside project

[chrome monolith]

its within one chat ... like I say I'm going to go to my [town] house, go paint some stuff, mow the lawns, come home and build a fence and it'll talk like these are all at the same property

I correct it, it gets it right for one message, then straight back to collapsing to one property again

[sleek adder]

SillyTavern might be more appropriate for your use case

[chrome monolith]

that can actually handle multiple concurrent renovations without mixing them up?

[sleek adder]

it will do more then that

[trim rock]

Why not to separate these parts from one chat to several? It seems I don't get what is your goal, sorry

[sleek adder]

each room will belong in a mystical forest or a dark fortress, or just your home, and they will have their own characters that talk to you and ask you how you are feeling if you want

[chrome monolith]

just talking about what I am planning to do in one given day

I got codex to write me a task tracker that properly splits up the different physical and computer projects I've got going on that that is starting to be very helpful. I need more hours in the day, I have a lot on

[jolly temple]

@normal quarry hey! just curious, can i check whats the main differences between pro (5x) and business plan (3 seaters)?

from what i know, its

1. pro can use unlimited gpt 5.4 pro, subjected to abuse. but, is there an actual limit here? they cant be providing the same amount of pro queries as 20x pro. is it like 100/month or something?
2. pro can use gpt 4.5, business cant
3. pro has light and heavy thinking modes for gpt 5.4
4. pro has pulse
   thats about it, right?

[chrome monolith]

it just really bothers me that chatgpt can keep track of tech projects but not physical ones. Like is it me or is it chatgpt? I've even set memories for this

[chrome monolith]

failed to install, probably my computer too old again

[normal lance]

how to make chatgpt website follow custom instructions
i have

• use very little emojis

and it used none
its very inconsistent
chatgpt in the past years ago followed instructions much better than it does now
what happened?

[sleek adder]

would you like a new computer?

[normal quarry]

You're right about most of it but there are two Pro plans. $100 and $200. The bigger plan gets more Codex usage.

[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.

[chrome monolith]

at some point, when I have tax payable to deduct it against

[jolly temple]

right. im referring to the 5x plan. so my main question is 1).
so what is it?

[sleek adder]

are you in need of a donation of sorts??

[normal quarry]

so with 5.4 Pro there is a limit. with 5.4 standard it's pretty much unlimited except not for Codex

[jolly temple]

right. so for 5.4 pro, the 20x plan gets way more than the 5x plan? or is it both just unlimited unless u abuse

[normal quarry]

if you're typing into ChatGPT then 5.4 is unlimited, if you're using the model in Codex it's more limited

[jolly temple]

wait wait im talking about 5.4 pro haha, 5.4 i dont need that many prompts

[jolly temple]

this is my qn

[normal quarry]

5.4 Pro has a fixed limit, I don't know what that limit is but there is more quota on the $200 plan than the $100 plan.

[jolly temple]

i see, thanks

does anyone here know what the limit is?

[chrome monolith]

no, i just don't want to burn capital on something that depreciates

[sleek adder]

Just make sure you are enjoying something in life

[chrome monolith]

I'd be enjoying it a bit more if I ever finish this damn project but its closer every week

[normal quarry]

I suggested to Tarrow that they get a MacBook Neo but they want to save the cash

[chrome monolith]

I still need to buy a bath, two vanities and two shower screens and they are kinda more important than a computer right now

[normal quarry]

running Codex app and ChatGPT app is a compelling reason the get one, but I won't push it

[chrome monolith]

codex CLI works but yeah, no app

can't run the latest version of office either, thankfully I'm just moral support for TAFE not actually doing it cos there's noticeable compatibility differences

[sleek adder]

you are renovating a house but cant upgrade your pc with basic requirements

[chrome monolith]

it does everything I need. Its more that a new mac is painfully expensive, and usually you'd tax deduct work tools. Which I cannot do

[sleek adder]

when it comes to AI, Mac is surprisingly cheap

But I understand what you mean, I mean, i have some junk GPU's and a DDR4 system you can just have.....

[chrome monolith]

I have an entire macbook pro sitting here that someone just had spare ... still slightly too old for everything. I used to have a work computer, but I obviously gave that back when I quit work

[stiff lichen]

yooo

[slender jasper]

why

it is actually failing. i cant generate any image

oh, it woken up

[pastel falcon]

hello

[umbral creek]

gpt 6 coming out today?

[umbral creek]

skill issue tbvh

[umbral creek]

tarrow
hello fren

[bitter pebble]

I have the same problem

[normal mesa]

Why is ChatGPT always against me, anytime i say something IT IS AGAINST ME it's in another side i ask for proof it doesnt give it makes excuses doesnt refute doesnt do anything

[rare surge]

My ChatGPT account is having connection problems, but the issue isn't on my end. Specifically, it's happening with a newly created business account.

[hidden dirge]

Try to reload page or update app. If that not helps, contact our OpenAI's support

[jolly monolith]

Is there a way to make a app for development like codex cli or claude code cli but use my openai account instead of using an api key?

[normal quarry]

Codex macOS app

[jolly monolith]

i know but i want to use cli instead due to it being more efficient for me to use with my hotkeys.

[normal quarry]

you can sign in to your ChatGPT account using Codex CLI

[jolly monolith]

im sorry i forgot to include that im making my own code cli

[normal quarry]

OpenClaw does this, OpenClaw is OSS so you can see how OpenClaw does it and incorporate that functionality into your own app

[jolly monolith]

ok sweet thank you

[hidden dirge]

Johny Sins? Hello legend

My childhood🥹

[jolly monolith]

lmao @hidden dirge

[umbral creek]

gpt 6 today?

[signal needle]

Could anyone explain to me how projects work and how gpts work how should i use them i recentcy got chat gpt plus

[covert thorn]

It doesn't seem like it, no.

[covert thorn]

Here's some information on how Projects work: https://help.openai.com/en/articles/10169521-projects-in-chatgpt
It's basically if you've got several chats about a single 'topic' or task/thing you're working on/asking about... then using a project might be the way to go.

[signal needle]

gpt 6? your joking??????????

DID SOMEONE SAY GPT 6

[umbral creek]

Whole chatting geeking

[covert thorn]

I haven't used GPTs much, but here's another article on them - I think of them as more like a workflow or specialized supplemental knowledge/instructions prepared by others: https://openai.com/index/introducing-gpts/

[umbral creek]

Jfc

[signal needle]

ty

GPT 6 WILL BE INSAIN .

[umbral creek]

Are you robot green?

[signal needle]

INSANE

WHAT KIND OF STUFF WILL GPT 6 BE ABLE TO DO

[normal quarry]

am I what??

[covert thorn]

People are only talking about GPT 6 because recently there were comments that Spud has finished pre-training, and some think it will be GPT 6... it could still take a few months to release, but polymarket is betting it will be within a month.

[umbral creek]

Robert Green

Typo

Darn autocorrect

[normal quarry]

no I am not

[umbral creek]

Oh

Well,

That uhh-

That’s pretty disappointing

[signal needle]

CAN SOMEONE FILL ME IN GPT 6????????

[umbral creek]

All good

[covert thorn]

Best to wait for the official announcement[s], there's always a lot of speculation in the AI space, and things can evolve/change quickly.

[uncut berry]

👀

[signal needle]

why am i using gpt 5.3 when 5.4 is out?? i have chat gpt plus

[uncut berry]

transition from Instant to Thinking

[signal needle]

why?

isnt in all in one

[signal needle]

why is 5.4 only for complex tasks>

?

[slender frost]

Because it's a "thinking" model.

[neon tendon]

because its costs.
its better overall you can use it always

for real complex tasks use PRO-EXTENDED 🔥

[signal needle]

what if i have plus

[scarlet steppe]

is chatgpt just straight up not answering for anyone else right now?

[neon tendon]

pro makes 5.4 thinking better, and gives u 5.4 pro

[signal needle]

can i get pro for cheap

i have plus and its 20

[silent root]

Yeah for me too

Instant mode seems to work but not the thinking one

[silent root]

Only the extended thinking

Seems to bug

[scarlet steppe]

oh, weird

[silent root]

Yeah

[neon tendon]

pro is 100$ + tax

[cobalt lynx]

hmmmmm...

[cobalt lynx]

Yes. I'm having some issues, too. The instant model seems to be working, but when I use the thinking model, it seems to be thinking for ages before resolving to "Network error." My internet and cellular data are fine and active. It could be a mobile-only issue or have to do with the images I attached, but the model-dependent success is the odd part.

[normal mesa]

I am so done with this ai

i swear to almighty god this ai is never on my side, I literally asked if i could report something to the police i have screenshots and it randomly made excuses just to not be with my side

Or programming it wont work NOTHING

I even went to personalization and gave it custom instructions it's useless

[slender frost]

GPT shouldn't be used for legal or medical advice anyway

[signal needle]

What ai can make a spread sheet of all the topics in a level maths and my confidence level

[slender frost]

I had it do an example. To show GPT is able to.

[I attached the example, idk if it'll be sent in the discord, if not, I'll send dms]

It sent. Nice

[signal needle]

this is good but i dont have excel cus i dont pay for microsoft 360

so i have to use google sheets

[slender frost]

Yeah, you should be able to open it in sheets.

You can ask it specifically to be for sheets. (Even tho I don't think it'll make much difference with the end result)

[signal needle]

i cant cus i have 360

[slender frost]

Give me a minute

[signal needle]

ty

[slender frost]

Try this one

[neon tendon]

models have no idea how good they are.
5.4 pro thinks hes as good as 5.0

[upbeat scroll]

HOW MICROSOFT COPILOT IS FAR AWAY FROM CHATGPT

[neon tendon]

copilot is not even in same universe as openai models wym

[upbeat scroll]

I DO NOT UNDERSTAND HOW THESE BOG COMPANIES LIKE GOOGLE CANNOT BUILD A SIMILAR THINGS LIKE OPENAI AND PERPLEXITY

[trim rock]

Gemini lol

It's way better than GPT. At least not so dumb like last models of GPT

[neon tendon]

openai is most skilled

[neon tendon]

gemini 3.0 is newer than gpt 5.0

so obviously 3.1 pro is fresher than 5.4