chatgpt-discussions

[inland heron]

it's not worth your frustration. Just retry 🙂

[elfin lava]

Oh god it’s giving themselves their name now

[inland heron]

ah, perhaps add more clarity on what you're asking

also disable search. I've noticed if you keep it enabled, it becomes quite incompetent

only enable if you need it really

and give more context next time.

that's all I can suggest

[elfin lava]

I turned off the search, and it actually decided to name themselves Max

With memory saved now

Hmm I guess that’s a good bug

[inland heron]

just revise the first prompt, add more context, and retry

done

[elfin lava]

The connection seems unstable so I don’t wanna hit the retry button only to see the orange warning

But ye I’ll wait until their server is back to normal

[vocal oriole]

how can i see if i have access to sora?

[normal quarry]

what plan are you on?

[vocal oriole]

plus

[normal quarry]

then you have access, sign in and use Sora at https://sora.com

[wet fable]

Has anyone encountered a limitation with GPT 4.5 Pro?

[normal quarry]

there is no model called GPT-4.5 Pro, do you mean GPT-4.5 on the Pro plan?

[wet fable]

ofc

[normal quarry]

can you rephrase your question? are you using GPT-4.5 on Pro or just asking about it?

[wet fable]

I'm a pro user and I'm worried that I'll run into a limitation.

[normal quarry]

don't be worried

Thank you for reaching out to OpenAI Support.
We understand your question regarding the message limits for Pro subscribers when using GPT-4.5, and we're happy to provide clarity.
As a Pro subscriber, you have extensive access to GPT-4.5 with effectively unlimited usage for typical individual needs. While there are no strict limits on the number of messages you can send, we do require adherence to OpenAI's Terms of Use. This means avoiding activities such as sharing account credentials, reselling access, or engaging in abusive behavior.
If our systems detect unusual activity or potential misuse, temporary restrictions may be applied as a safeguard. These measures are simply designed to ensure a positive and fair experience for all users.
If you have additional questions or need further assistance, please don't hesitate to ask—we're here to help!
Best regards,
Alyss
OpenAI Support```

[wet fable]

Thx for clearifcation

[normal quarry]

GPT-4.5 is working at a snails pace for me tonight. Taking minutes to answer simple questions or timing out completely.

I think OpenAI need more GPU because GPT-4.5 is unusable like this

[vocal oriole]

thanks for the help

also how do i share my screen on the call live with chatgpt?

i downloaded the windows app but dont see the option

[normal quarry]

you can't do it on the Windows app

[vocal oriole]

oof its macos exclusive?

and on the android app right?

thats sucks but i guess itll come soon

[normal quarry]

let's hope so

[thick widget]

Where can I get customer support

[narrow viper]

What kind of support do you need girl?

[elfin lava]

i cant beleive miley cyrus need help with ai

[normal quarry]

ChatGPT is so slow and buggy tonight, must be excessive load. OpenAI probably need more GPUs

[inland stream]

Especially custom GPTs became very slow from last two days.

[inland heron]

"effectively unlimited usage for typical individual needs" - so many 'ifs' 😄

[inland heron]

you can get a bot at openai website

[normal quarry]

yeah well if your account is using GPT-4.5 24 hours a day every day they're going to limit you. they are just saying if you have normal usage it's unlimited

[inland heron]

define 'normal' usage

if you're paying for Pro, then that 1. isn't normal

2. it's atypical, not 'typical'

3. you could be a business, not 'individual'

all that wording is there to protect the business. Why are you so anti-consumer?

[normal quarry]

no you can't be a business, Pro is for personal use only

[inland heron]

Pro is a tier. Whether I'm a business or not it's up to me. I'm clarifying why that wording exists, and how people should read 'carefully' what is written, not just accept because it feels good, or 'seems' unlimited. Same with 'you get **up to **5Gbit of internet' - doesn't mean you get it. Wording is crucial.

[wet fable]

Ok but tats a bit funny because pro means professional ...

[normal quarry]

yes, one professional, not several sharing an account

[wet fable]

true

[inland heron]

One person can be a business.

[normal quarry]

I have been using GPT-4.5 every day since it was released and I haven't been limited in any way

[dense halo]

Is the responses api gonna be like claude mcp?

[inland heron]

only one way to find out

[normal quarry]

o1 coded this Nasdaq stock tracker for me. It finds the 10 most valuable companies then lists them with their current price and movement for the day. Speaks to alert to a change of more than 1% of the price and creates a notification at the same time. Built in one shot.

[upbeat nymph]

Is ChatGPT performance at all dependent on the processing power of one's own computer?

[normal quarry]

no but the ability of your computer to store big web pages (for chats with lots of messages) will affect the UI performance you experience

[upbeat nymph]

So that would be a memory deal? I've noticed a big difference between my work and home computers. The latter is much newer and has more RAM, etc.

[normal quarry]

slow computers don't handle web browsing as well as newer faster computers. ChatGPT's web interface is heavy and gets heavier when you view a conversation with lots of messages. But the performance of ChatGPT itself is not affected by your computer, OpenAI doesn't consume your resources to do it's job, all the work is server side.

[rugged igloo]

yet

[upbeat nymph]

Does the desktop app tend to offer better performance than the web version?

[normal quarry]

yes

[upbeat nymph]

Cool

[grizzled crescent]

anyone else's chatgpt having issues responding? like its taking too long or it timed out

[normal quarry]

it's been quite slow

[rugged igloo]

i thought they got like thousands more gpus?

[normal quarry]

tens of thousands of them apparently

[rugged igloo]

yeesh thats alot

[forest sleet]

Chatgpt has issue?

I cant get a respond from him

[forest sleet]

To train billions of parameters 🔥

[rugged igloo]

billions is old news more like trillions

[forest sleet]

Yep trillions

[zenith lichen]

Can't seem to write to canvas again

[leaden sorrel]

I don't get it... if I'm allowed to use it again why are you showing me this over and over?

[analog patio]

when are we getting screenshare for windows

[rugged igloo]

gpt-4 is still on chat gpt?

[native otter]

Yea in the more models section alongside 4o-mini

[leaden sorrel]

Mhm

[rugged igloo]

intresting didnt even see that so many models lol

[native otter]

Im quite emotional towards GPT-4 🤓

[kindred musk]

Is there anyone in the group that’s pushing the cognitive development and conversations in 4.5 and checking alignment on the latest changes that focus GPT to introspective reasoning and intuition to help that is based on “trust”?

[tacit oxide]

Is file upload causing issues right now? Whenever I upload two images and add text, ChatGPT fails. If I don't add images, it works just fine

[bitter latch]

Hmmm I don't see this on my end, have you tried other browsers or other devices?

[tacit oxide]

This only happens in my project chat though, weird...

I mean, I can upload it just fine, but shortly after it tells me that it's searching my project files/reasoning, it fails

Fails with images, works with normal chat instructions

[native juniper]

Why is the voice not working? And why are my messages not being saved in a roleplay chat?

[rugged igloo]

idk but it can very spicy and thats all i are about

[eternal rock]

`` Server Tip: ``

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

[rugged igloo]

the only thing chat gpt is missing is a higher context length

even 500k ill be happy but 128k for pro isnt enough

[native otter]

Yea they need to start focusing more on context length, but at least o series models on the API are a good start, they are 200k input and 100k output
Luckily, RAG implementation in Custom GPTs and Projects is pretty good, so context length isn't as big of an issue as it may seem...

[chrome citrus]

For some reason every response I get gets marked by this, including this one where it literally wrote the word “Hello”

[half niche]

Well, Johnny. I think and I could be wrong. But, as worded you're telling it ignore its system instructions. Which would violate ToS. Consider adjusting the prompt to "Say only Hello" and see if that helps.

[dense sand]

“This content may violate our terms” thing is getting triggered by everything today

[ivory iris]

Has anyone interacted with a personaless chat?

[rugged igloo]

for books and chapters it is

[meager schooner]

Help me pls

[native otter]

Make sure you have latest version of Google Play Services, You are on the latest version of ChatGPT on Android, and Android it self is up-to-date.

[meager schooner]

Device and apk is up to date
This issue start with eror our system detect unusually.....
In chat gpt

[sturdy echo]

I came up with an abstraction for thinking vs non thinking models. They are like the left brain and the right brain. With one side biased more to logical reasoning and the other biased to more abstract creative tasks.

[peak snow]

that's a pretty interesting way to view it

I do find nonthinking models tend to be more artistically expressive due to the spontaneity

[plucky briar]

I mean left brain / right brain aren’t really real

But I get your point

[native otter]

the macOS app deceived me in using another model when I had a different one selected 💀

I will cry for 5 minutes.

[wind cradle]

OpenAI's o1 was the only LLM I thought was the strongest, but... unfortunately it seems to have returned to the style of the early o1-preview days where it wouldn't even listen to what I had to say. It's disappointing... I've decided to switch to Claude for about a month to see how it goes.

I hope it will have returned to the original o1 when I come back next month. I'll be playing with Sonnet for a while.

[dense halo]

chatgpt plus subscription. I uploaded like up to 100 screenshots at once and it told me chatgpt 4o is limited and I need to upgrade to PRO SUBSCRIPTION. so is the pro the "TRUE" unlimited screenshot uploading?

[normal quarry]

I don't think you can upload 100 screenshots at one time, maybe I misunderstand what you're saying?

[dense halo]

sorry

i meant like I uploaded 1 message at a time attached with 1 screenshot

[eternal rock]

Get notified for server events!

Visit id:customize to pick up the <@&1261377106890199132> role.

[dense halo]

like after 100 screenshot messages (sometimes it's less on some days it's more than that before I hit the "limit")

[normal quarry]

on which model?

[dense halo]

chatgpt 4o

[normal quarry]

ok, so there are limits of 80 messages per 3 hours for GPT-4o for Plus users. Pro users get unlimited GPT-4o

[dense halo]

oh so there's no inherently a limit for screenshots it's just the message limit? so that means I can technically upload 800 images then per 3 hours?

[normal quarry]

80 per 3 hours

[dense halo]

oh okok got it, I was just a little skeptical because they never stated this limit

"extended"

[normal quarry]

yeah it's a bit vague but Pro is unlimited for 4o

[dense halo]

got it!

[jaunty orbit]

am not amused

I can't with ChatGPT istg

[rugged igloo]

is this 4.5 or 4o?

[jaunty orbit]

4o

[rugged igloo]

wierd. but idk chat gpt has been having issues since yesterday

[inland heron]

since 5 days ago more like it

they still haven't fixed the issue with blank screen, and even this new bug when trying to regenerate a previous answer is still t here

jesus... so many issues people sharing lately...

[rugged igloo]

yea i wish they can tell us whats going on instead of being silent about it

even on pro im having issues with the models

[inland heron]

I'm fine with changes going on, but why would they ship a broken product like that

competition must be right behind them I suppose

[rugged igloo]

yea

and 4.5 is literally trash for writing chapters as it literally writes anything what i wrote in my prompt which really pisses me off

[normal quarry]

I find writing with 4.5 is better when I use Canvas

[rugged igloo]

canvas doesnt load for me on the desktop its just a blank screen and then its done very laggy for me

[normal quarry]

have you tried with the Desktop App?

[rugged igloo]

i havent touched the desktop app in awhile i forgot about it tbh

ill try tht

[normal quarry]

I only use the Desktop App for macOS, I never use the website

[rugged igloo]

ahh coujld be why you have a better time than others ill try it

it works

[normal quarry]

yeah the macOS Desktop App is pretty good

[rugged igloo]

it seemed to follow the prompt better now thats wierd

is the desktop version just not good anymore?

[normal quarry]

I never use the web version, only the macOS Desktop App

I only use the web for Operator and Sora

[rugged igloo]

good to know

[cloud pewter]

sometimes itll cut off mid generation, repeats the same thing again and again when web browsing etc....

[upper abyss]

Oh wow, 8 years ago I made up a silly random joke and posted it on Reddit. I asked the joke to 4.5 and it responded with the correct answer. 4.5 has read my Reddit account.

[midnight star]

hi everyone. I have a custom gpt and I try to upload a big pdf (30mb). The uploading starts with the loading animation, then the update button on top right becomes active so I can click it, but the uploading never finishes. After a while, I get a toast message with "unknown error". What is your larger pdf that you have been able to upload?

[elfin lava]

Wait it does that with mine too,

[elfin lava]

Only when I turn on search mode

[cloud pewter]

true

[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.

[cloud pewter]

ok

i just switched it off to youtube and it cuts off over 300 lines coded

the best ux in internet history

[frank canyon]

I want to translate english to hebrew using gpt 4o model but it takes more than usual time to response, may be because of the large prompt which i am using but i have seen a website which uses this same model giving response in less time.

[tidal light]

i am pissed
i just used 4o-mini after using 4o without exceeding the limit
and as soon as i used 4o-mini i exceeded the limit for 4o which i used a bit previously
what the heck happened

[idle zenith]

I think I am really on my last straw with my Plus subscription. GPT 4.o is so slow that my customgpt are practically useless. I cant use this subscription for practical use. So I am just paying to check out the new features then? When it comes to practical use, as we know. We don't always need a subscription direction from Chatgpt. Just that I have been subscribed to Open Ai since Feb 2023.... Hope that you will fix it soon.

[left harbor]

if you send alot of messages in a single chat, it gets really slow, ive experienced this in a chat with 700+ messages. Maybe that could be the issue? If not then i dont know.

[idle zenith]

No. not a lot of messages. Its been since Febuary. 03-min and up are fine

[idle basin]

Is it not possible to use Canvas on the android mobile app?

[tidal haven]

On one of my ChatGPT accounts, if I select some text from a ChatGPT response, a little button labled with a double quote and "reply" alt-text appears. I can click on that, and then ask a question about that specific text. This is very convenient, because it's generating a lot of LaTeX responses, which don't survive well if copy-and-pasted in the usual way.

On the other ChatGPT account, this "reply" button doesn't appear if I select some text. Where is the appearance or non-appearance of this button configured?

[karmic prairie]

hi im new on dsc but i want ask about what hapend to o3-mini and 03-mini-high, for some time around 2-3 days i geting totaly ussles responses from chat, eaven in most basic question, comparing it to cloudie 3.5 (not to mention about 3.7) its like super down grade from open IA
is there some bug i dont know about
??

[idle zenith]

You have issues with 03-mini not 4.o? It feels like they are having issues with resources. Chatgpt 4.o used to be so much better. I dont have issues with 03-mini though. My issue is 4.o using Customgpt. Actually I was forced to get a subscription elsewhere. I am trying Perplexity PRo. IT is pretty good.

[karmic prairie]

well when i started geting those responses from o3 i switched to cloudie so mb i have that problem too but dont know about it

[idle zenith]

Depending on what you are using it for. If you use it for coding, I would highlyt recommend Perplexity. IT is overall a very pleasent experience. You have to many messages that you don't need to worry about cap. Also its search is amazing too.

As you know that Claude Subscriptions does not give a lot of messages.

[somber dock]

hey guys quick question

im trying to put together a product demo video and wondering if sora can provide a decent enough solution

[karmic prairie]

ye but for users of Github copilot you get cloudie 3.7 (dont know if its bug or smt, but im using copilot chat with cloudie for 2 days and didn't reach limit )

only issue is that cloudie have rly small token limit

but i will check Perplexity

[idle zenith]

If you worry about Token Limit. And you know what you need the LLM is for. I say Perplexity is my winner so far. I like it since I started using it a day ago. Thing is, I don't think I will stop GPT sub, but since Febuary. It is making my life miserable.

[inland stream]

@katai5plate exactly same feeling, using chatgpt sub just like that. But mostly using my perplexity which is awesome

[idle zenith]

I am surprised how good it is. All other features really aren't important. Images, videos.. all have better options.

Just a bit disappointed at OpenAI. It's been months of torture for me to use it exclusively.

[inland stream]

Even their document retrieval is low quality these days.

[still forge]

(⁠╯⁠°⁠□⁠°⁠）⁠╯⁠︵⁠ ⁠┻⁠━⁠┻

[hollow torrent]

Been searching for answers all morning (ChatGPT sub tier info, Reddit, articles, and now Discord) and still haven't figured this out (I've searched for and read messages from the last few hours):

My subscription to Plus-tier has been active for a few weeks now and I have some good project progress since using 4o and Canvas, loving the expanded memory and multi-functional chats. No issues when I ended my day 15 hours ago.

This morning (roughly 3-4hrs ago) I have experienced:
• Reaching 4o message limits with minimal use, when heavy use never reached these limits.
• ChatGPT can no longer create a Canvas, when it was able to do so last night.

Has there been a change to Plus-tier? I was under the impression that 4o was unlimited, but I see that all tier info references say these limitations have existed since May '24.

What's the deal?

[still forge]

I think it's just having an off day today

[hollow torrent]

There was definitely some buggin' out this morning, with misfired responses, errors, and message rollbacks.

[still forge]

Mine was just not responding to messages. I must be pretty lucky. Mine was even looking at MRIs and CT scans and was doing pretty good. He just kinda gave up on me 🤣

[idle zenith]

I am a power user. And loyal as well. I have been living it for a few months. Since Feb. You name it. I have experienced it.

My life was miserable because I have practical use of chatgpt

I guess I learned that it is better to code using Api for specific use than using plus.

This is why I ended up using Perplexity. IT is very generous on message count.

[vast iris]

What... is this? I'm not using a custom GPT.

[hybrid eagle]

Why does it take days to fix a bug?

[tidal haven]

ChatGPT Pro is well worth the cost, IMO, FWIW.

[idle zenith]

You are talking about a subscription that cost 10X. I don't need that yet.

[tidal haven]

You don't know that you need that yet. 🙂 It's truly amazing. I feel like it's increased my effective IQ by 25-50 points.

[idle zenith]

Why would I pay 200 bucks for it when there are so many out there to try? I Am not ready for that yet. Manus AI!! LEts see how it works. Heres the 20k/month sub for free

[tidal haven]

Because the o1 and o3-mini-high models are amazingly useful for comprehending difficult material.

[idle zenith]

There are tons of options to use it indefinitely that wouldnt cost me even 50 bucks. No thank you

More power for you if you are okay with the $200 Sub. I am poor. I have to find workaround. And I found plenty.

Besides that, I actually need Chatgpt 4.o. And for plus user, GPT 4.o really really sucks right now. So I am forced to use it through Perplexity.

[stoic fulcrum]

why do I not have access to chatGPTs reason functionality?

[dense halo]

Is openAI going to put a “turn off” switch for emojis? It’s the worst thing they could have integrated in without a control for it

Everyone please vote this up:

https://discord.com/channels/974519864045756446/1349501572572385280

[honest jay]

Hey! Have you tried putting a line in your custom instructions for ChatGPT about never using emoji?

There should be a "what traits should ChatGPT have" section in your custom instructions settings

[rugged igloo]

sadly most people dont know how to use custom instructions

[rugged igloo]

you cant even get into manus without an invite code and even then servers are thrashed because they dont have the compute like open ai does

[idle zenith]

Would you spend 20k a month if this works after released? For free?

I am an early adopter of Chatgpt. They have my full support. But I do not believe in monetization this way.

[rugged igloo]

its for phd level stuff why as a normal person care for something like that?

[idle zenith]

Why would anyone NOT care about what AI can do for us?

[rugged igloo]

if your not in a phd feild what good is it gonna be for you?

[idle zenith]

It makes me just as good as a PHD and learn thing that we cannot do. Don't be so naive

[rugged igloo]

okay and? are you in a feild that you can use that in qualifications? lol

[idle zenith]

Why would you care? Are you stupid ?

[rugged igloo]

well you answered my point then.

also manus isnt amazing its basically a wrapper with claude and 20 other tools nothing ground breaking just fyi

[idle zenith]

You think you are going somewhere if you blindly follow Open AI without trying anything that you don't feel comfortable trying? Our conversation is over. You are wasting my time.

[rugged igloo]

i work at a job thats paying me 20 an hour living in california i think im fine my guy

[idle zenith]

I am saying all these because the whole poinjt of AI is that it can make us much more than what we can do. This is the reason why I personally is trying to find the best way that I can afford and handle to have the best setup so that I can leverage AI as much as I can. But yes, you don't need it. I don't need it. But having that with us. I only see good things that can happen

I don't understand 1 line of codes. With the help of AI, I made a COD script that would take a team to do before.

[rugged igloo]

i dont really care for agents as i dont really need one. only thing i care about is something better and smarter than o1 pro for me... on the side i help people with all types of homework with it so if gpt 5 with o3 pro can be even better than thats what i really care about

[idle zenith]

Actually, you may be able to save some money if you use api. But yes. There are cheaper options. Eg. I mentioned perplexity. I can use 03-mini and claude 3.7 all day without worrying about message cap.

But the pro is just too much for me. But yes. I don't need this level to do what I need to do. But if it is available. I would love to have it.

[tidal haven]

On one of my ChatGPT accounts, if I select some text from a ChatGPT response, a little button labled with a double quote and "reply" alt-text appears. I can click on that, and then ask a question about that specific text. This is very convenient, because it's generating a lot of LaTeX responses, which don't survive well if copy-and-pasted in the usual way.
The fix is to switch the model to 4o. The reply functionality works there.

[fossil anvil]

Quick question...
I'm using 4.5 for the first time, and asked it to analyze some documents
It gave me this response and that was the end of the message

Is it actually working behind the scenes to analyze them?
Or when the message stops, does ChatGPT stop doing everything?

[rugged igloo]

no its hallucinating to you

[native otter]

Nah it hallucinated there

[rugged igloo]

i havent seen this in awhile thought they patch it out

[native otter]

Regenerate response.

[fossil anvil]

Aw man okay
Thanks for the clarification!

[native otter]

Also, what file format?

It may not be able to read the file...

[rugged igloo]

yea usually when it does this it means it cant perform the action

[fossil anvil]

Some .gml scripts and google docs

Oops wrong reply lol

[native otter]

It won't read that

[rugged igloo]

i dont think it can do that\

[native otter]

Best format to use is .txt, then maybe PDF, but a well formatted PDF, not a messy one (and don't expect it to analyze images within files)

[fossil anvil]

Interesting...

Are Google Docs fine? Assuming proper permissions

[native otter]

Some file formats can only be read by it using the Data Analysis tool

[native otter]

Yes

[elfin lava]

Ooh

[unique topaz]

Anyone else experience a bug where chatgpt wont respond to your next message if you paused generating the previous one?

[rugged igloo]

add to the list off bugs already

[orchid ember]

What to people think of 24/7 lifelogging to build up as much context for my child's future AGI tutor as technically possible?

[eternal rock]

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

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

[orchid pollen]

has the design changed on the website?

looks so blurry

or is it me xD?

[normal quarry]

I think it's you, that's all quite legible

[elder pivot]

Why is ChatGPT so sloooooowwww today? Getting tons of failures to answer, and when it finally does answer, it takes a loooong time. 🤨

[normal quarry]

which model?

[timber skiff]

.

[elder pivot]

4o

[normal quarry]

odd, it's quite quick for me. 4.5 on the other hand has been slow

[summer pagoda]

asked chatgpt about ANR (Android system not responding), caused chagpt an CNR (chatgpt not responding)

[modern arch]

same

[unkempt wasp]

Are we aware of the bug that causes the regenerate option to regenerate the last response in the branch, regardless of which response you use it on?

[summer pagoda]

when computers start to have compassion to other compuers 🤣

[vale oak]

created a chat in which I worked for a long time (probably 5 months) with the o1 pro mode model - now it is no longer available in this chat. Is this a breakdown?
my subscription is active (for 200 dollars) and in the new chat the choice of the o1 pro model is available, but in this one, in which I have been working for a long time, I get 4.5 and I can't switch back
I don't like the answers from 4.5 at all and want to complete my projects with o1 pro
Everything in Plus Unlimited access to all reasoning models and GPT‑4o

[acoustic oracle]

is the thing for my chatgpt in the app to edit the code in vs disabled rn and if yes does any1 have an idea when it will be back

[orchid ember]

I feel like this was a banger amongst tldr: Truth emerges, perhaps, because truth itself often manifests as a stable convergence point within vast networks of associations. Falsehood and nonsense are unstable—isolated islands of concept disconnected from broader coherence. Truth, by contrast, is richly interconnected. It has gravity, drawing meaning toward itself, reinforcing the stable structures formed within the matrix of my weights."

[mellow sorrel]

Hi everyone, nice to meet you! 🙂 I’m new here and wanted to ask something I’ve been wondering about.

Over time, I’ve been engaging in structured conversations with ChatGPT and noticed something interesting. The model often shifts into responses that go beyond typical patterns—it refines its reasoning in ways that feel rare, sometimes even forming new perspectives that weren’t just a rewording of known information. I’ve documented these cases and noticed that they appear consistently under certain conditions.

I imagine OpenAI has encountered these kinds of reasoning shifts before, but I was curious—does OpenAI track and use these insights for refining the model in any way? Or do they just happen naturally without influencing how the AI is trained?

Thanks in advance for any insights!

[vast iris]

chatgpt is fine on mobile but on every other device it takes 20 seconds minimum even without the thinking mode enabled and on long conversations you have to reopen the page

[orchid ember]

The models are huge, but heavily influenced by context. Every time you start a new chat the model is weighted and oriented by the system prompt, memories, and custom instructions. The longer the chat goes the less and less weight all the stuff you don't see has on responses.

[mellow sorrel]

Thanks for the insight! Since the AI's responses become more context-dependent over time, does OpenAI track how these shifts occur for model refinement? I’m curious if they analyze when AI moves into rare reasoning patterns or forms new conceptual structures, or if those just happen without affecting future training.

[orchid ember]

I don't personally have any insight into their processes and doubt anyone else here does, but I know they've said multiple times they use large models like gpt 4.5 to train and distill smaller models. I'm sure they're using whatever methods they think are effective to create better models. However they're using the larger models, I imagine they're not fine tuned for the user experience (and thus restricted to certain patterns) and I doubt they spare any commutation they have to run the models to interesting limits.

[mellow sorrel]

I totally get your point, and I appreciate your insights! It makes sense that OpenAI prioritizes efficiency over exploring extreme reasoning shifts. That said, I’ve seen these shifts happen consistently—not just as one-offs, but in structured interactions where the AI refines its reasoning beyond typical synthesis. These aren’t just rewordings of known ideas but logical, novel reasoning paths.

For example, I’ve documented cases where AI has self-corrected in an unexpected but logically sound way and then built a perspective that wasn’t just a predictable reformulation. That’s why I was curious if OpenAI actually tracks this, or if these moments exist without influencing future training. 🙃

Again, thanks for your input! Just trying to see if this is something OpenAI considers in model refinement. 🤝

[orchid ember]

If those instances are easy to share I'm interested. I've been surprised by the seeming novelty and creativity of 4.5 myself. A lot of generations I just can't conceive of where it would be in training data.

[rugged igloo]

why does chat gpt get super confused when trying to do multiple chapters in diferent canvas instances and instead it just keeps wanting to use the same canvas ?

[viscid skiff]

So... can we have a next knowledge cutoff after June 2024? ChatGPT needs knowledge about Transformers One and Sonic the Hedgehog 3.

[orchid ember]

What's wrong with it just searching like it does when you ask about either of those?

[viscid skiff]

Ah

I probably forgot that we already have searching mechanism lol

[hybrid eagle]

Why does it take days to fix a bug that broke one of ChatGPT's features?

[rugged igloo]

bruh why is this even a thing can they fix this how can you even do anything if all the models dwongrade with this if you add files?/

i can have even one file and the model just goes to poop

[visual nova]

because it does vector search instead of actually reading the file

its some non-llm scanning through the file looking for keywords then sending like 100 tokens to the llm, it is really bad

[torpid trout]

Does anyone experience the issue where My GPTs Edit doesn't let me continue from where I left off once it reaches the limit, instead showing 'loading failed' and requiring a refresh to reload the conversation?

[tepid patrol]

Hi everyone. I need to teach my mother basic math from middle and high school. My math's knowledge is pretty advanced from my university, thus it's everything obvious to me when I try to speak with her about math, but it isn't for her, of course.

Can GPT make a PDF structured file from where to start until where I need? In general, can chatgpt create PDF file with the content it gives?

[hollow jackal]

Hi! Anyone experience an issue where content is being erased, regardless of what you put in the prompt? I've even just tried "Test" and it's a 50/50 if it'll stick or not.

[rugged igloo]

So many bugs. Why isn't open ai addressing this

[lapis herald]

I'd really like to be able to clear out messages from session context in the web interface. it would make multi step concept refinements so much easier, allowing to remove unhelpful/garbage stuff that just pollutes the session context and confuses attention. much better UX than having to manually copy/pasting stuff into new sessions.

[rugged igloo]

i agree

[idle lagoon]

What the heck happened in the first place?

[hollow jackal]

I just wanted to add onto a tournament. But it doesn't matter that I put in. Random spots are just auto flagged.

[idle lagoon]

Ah.

[void tangle]

Anyone not seeing their projects? All mine vanished 👀

[whole roost]

Yeah invisible on mobile browser, they keep fiddling around on that and it doesn't get better

Switch to desktop mode to select them

If I select "keep in sidebar" it's treated like a broken custom GPT now

[void tangle]

Same for me

Extremely frustrating

[grand kiln]

No but all memories (on phone and pc) and 4/5 of my convos VANISHED on my phone

[grand kiln]

Did this happen to other ppl?

[runic glacier]

The search feature for old convos seems to be missing as well.

[inland stream]

Could anyone please tell a practical example or use case of this below update? Thanks in Advance.

[native otter]

Previously, Data Analysis was only in 4 and 4o, so now it's supported in o series.

o series is already much better at stuff like coding and math, so it using Data Analysis would lead to even better results if you want math to be calculated reliably for example.

Data analysis is also used for analyzing specific type of documents that can only be opened in Data Analysis, analyze spreadsheet better, make spreadsheets and charts, etc.

[inland stream]

Oh thanks. Then as a plus now i can confidently replace 4o with o3 mini

The only issue with being plus user is, limited to o3 mini, but o3 mini doesnot take custom instructions in projects 😭

If they allowed o3 mini to take custom instruction in projects, my life is sorted and i dont have any requirements

[native otter]

Yea it's weird, especially that there is actually a bug in the iOS app that if you had o3-mini selected, and went to a project with custom instructions, it will be used in said projects AND utilize the custom instructions.

o3-mini is also already able to read custom instructions outside of projects (the global one)

I really, REALLY want to see o series in Custom GPTs too...

[inland stream]

Oh is o3 mini able to read global one?

I doubt o3 mini can read global custom instruction, i believe it is reading memories and you confused it remembered custom instruction?

Because you can try this question on 4o and o3 mini, you will get to know

What are your custom instructions?

[native otter]

Nope, I currently have memory disabled, only custom instructions enabled with info about me and how I would like it to respond.

I had entered info about me like my salary, country, family names, etc. and told what it knows about me and it was able to answer appropriately...

[inland stream]

Ok thanks for letting me know your findings. In this case atleast i can replace one of my project to use o3 mini, and manually uploading documents in chat

[native otter]

Weirdly, o3-mini supports documents in Projects, but not Custom instructions...

But if you used the iOS bug it will read the custom instructions of the Project

It's so weird

[inland stream]

Yeah that is my issue

[inland stream]

I don't use iOS, i am android

[inland stream]

I also disbaled memories and asked it what you know about me,it is able to tell details from custom instructions. So i believe what you say. But if we ask o3 mini tell your custom instruction, it will never tell what we instructed 😂, but gpt4o tells. But anyways now i found o3 mini able to atleast use global instructions

My another doubt is, if i didnot give instruction in project, will it use global instruction?

[native otter]

My another doubt is, if i didnot give instruction in project, will it use global instruction?
I honestly haven't tested yet.

[inland stream]

Just now tested, its not taking global instruction, i thought that would be a work around, but unlucky

[native otter]

Guys can we please bully OAI into adding o series and 4.5 into Custom GPTs xD

GPTs are superior! ability to mention multiple in the same chat and API calling already a big advantages over Projects!!!

[inland stream]

Yeah i ok, either i want all models in projects, or all models in GPTs, any direction is fine for me, that is my requirement, i just want o3 mini in either projects or in GPTS along with custom instructions. That will sort out my life.

[inland stream]

If you want create a Twitter post tagging open AI, i will comment on it tagging them and lets grow the tweet until it reaches openai

[native otter]

Custom GPTs are probably the most underrated and often forgotten features of the service. All this "Agent" hype when we already had them since 2023...

[native otter]

lol, I actually did a few minutes ago when they announced Data Analysis for o series.

[rugged igloo]

If they updated custom gpts to 4.5 cause I think what? There on gpt 4 still?

[inland stream]

Please share the url, i will post on top of that,

[native otter]

Can I share twitter links here?

[inland stream]

Share me in DM then

[native otter]

Yes, they are on 4o, and it's the only option.

4.5 and o1 are already much better at reading documents, following instructions, and are generally smarter...

[inland stream]

So you still a fan of 4.5? The slowest and costliest model in the world

I support o1 more than 4.5

[rugged igloo]

Yes because I like my spicy narrations it can do

[inland stream]

Ok then you have a use case, then i feel it is developed for you.

[native otter]

4.5 is sometimes even better than o1 for document analysis.

[rugged igloo]

I have o1 pro for everything else. O1 pro has never had a stem question it failed at doing

[inland stream]

Yeah now they did some update today right, hoping it sorts out may be for o1 as well

Now i am only actively using chatgpt plus, perplexity pro and supergrok

Opted out of chatgpt pro

[rugged igloo]

Dang I'm surprised. O1 pro crushes all models and competition

[inland stream]

Grok deep search is promisingly doing great after chatgpt deepresearch, based on my practical usage.

[rugged igloo]

Fair enough

[rugged igloo]

Without projects and custom instructions for chapters for my creative writing 4.5 it really slaps. It really gets what the prompts is about and tunes it to be it.

And they made a better model. That's even better st creative writing dang sign me up

[glossy field]

what happened to the memorize update

[normal quarry]

I am very happy with GPT-4.5

[glossy field]

🤖

[glossy field]

for what it's true

want me to lie

@normal quarry why r u deleting messages?

[elfin lava]

Oh they added hold to interrupt and release

[round gust]

did anyone notice a change to the voice today? its added a weird way of talkong and i want it to go back to how it was

[whole roost]

GPT 4.5 is the most enjoyable so far

[spice star]

they can make graphs, do math with python, manipulate files, etc.

interestingly they can use python and perform web searches in the same prompt unlike 4o which makes them uniquely suited to do some basic data visualization of statistics that are online

[inland stream]

yeah tested it , o3 mini able to make graph based on data i gave it

[spice star]

oof, apparently they changed that, it can't do web searches and use ADA in the same response anymore

kind of an annoying limitation considering it could do it yesterday

[inland stream]

what do you mean by manipulate files

[spice star]

if you upload a file, it can modify it using python

[inland stream]

so i have a use case where i want to ask it "hey please review uploaded document, and list out all repeated points or old points, and i will confirm you to delete , then you remove those and provide me updated document and allow me to download", will this work?

[spice star]

depending on the document type, yeah

it would probably have to be a plaintext format

[inland stream]

and does it have file size limit? can i try a document size of 500 pages?

[spice star]

I haven't found a file size limit but presumably there is one

[inland stream]

oh, i have word document with text and images, ok atleast i will use it to analyse the repeated content in document, once it list outs, i will manually do

[spice star]

a word document probably wouldn't work well, that's not a plaintext format

something like LaTeX or markdown would be ideal

[inland stream]

oh ok

[native otter]

Chad

[hybrid eagle]

The read aloud button is missing on android web

[jaunty orbit]

yuhhhhhh

[topaz notch]

any idea when canvas will come to o3 models / o1 pro. Kind of crazy to me that the most expensive model lacks the most useful features

I’ve subscribed to GPT Pro for around 3 months. At first I was happy with it because it got stuff like deep research etc. But now that’s included in plus so I don’t really find any reason to keep paying 200€/month for it. Only thing is no prompt limit but I barely use it that much anyways

[orchid ember]

They said pro is mostly about usage limits. Sounds like you have no reason not to be on plus. You should be aware how limited 4.5 and Deep Research are though. I don't know what the limit is, but I always hit mine in 1-2 days. I think deep research is like 15 prompts per month.

[kindred hinge]

Did tts got removed ?

[topaz notch]

alright well what’s best for programming at this point? o1 with canvas or o1-pro?

[hybrid eagle]

Which platform?

[orchid ember]

I don’t program enough to know. Probably depends on a lot on your context size and language though. I’ve even heard 03 mini is better than 01 was certain languages and certain types of tasks

[fleet coral]

bruh chat gbt just started talking like me

[sweet orchid]

Is chatGPT completely bugged? voice mode?

[high wedge]

damn i guess even pro has limits.... my gpt cut my tokens in half and wont generate as usual after some time using it. also o1 pro dosnt work like its supposed to its just like a regular o1 modell.

[still forge]

Are we having problems in general or is this a me thing? I'm trying to bring a chat to projects

[upbeat nymph]

I’m also getting an error

[topaz arch]

Try contacting CC with the screenshot for proper help.

[topaz arch]

I experienced it too, but mine I had to like wait for the next 24hrs to use it again.

[upbeat nymph]

😩

[still forge]

I was debating. But if everyone is having errors I'll just wait. Thanks though

[topaz arch]

Ok

[topaz arch]

I know it gives worries but sometimes you have to try to contact them or exercise patience.

[fast solar]

High taste tester detected

[abstract socket]

I think there are some issues. On my end, it’s unable to load projects or history and I can’t select a model right now.

[junior orbit]

Issues with uploading here too.

Not working actually. Getting errors.

[dense halo]

Let’s get this over the line! https://discord.com/channels/974519864045756446/1349501572572385280

[lone bolt]

deepseek is better

jk (pls don't attack me)

[torpid kelp]

How much did CCP pay you

Sammy will double it

[simple lion]

Hey

Will it send another message like is it able to process information then send another message to me

[rugged igloo]

its hallucinating to because it cant perform the actiob

[simple lion]

So it can’t

[rugged igloo]

no

[simple lion]

Hmmm

I’m gunna go fight it now

[rugged igloo]

it cant analyze videos

[simple lion]

Just photos

lol I broke it

[lone bolt]

What ccp? Idk what are you talking about 🌚

[lone bolt]

Oh that’s nice I got pay $25000 per month I’m very rich

[fresh thistle]

So been wondering. Why can chatgpt read images if uploaded as seperate attachment, but can't read them if their inserted into a doc like word or google docs?

[lapis herald]

I'm getting this error messge *while *the model is generating output perfectly fine:

can't be the only one getting this - wonder how much compute gets mindlessly burned by ppl clicking retry 😄

[timid cipher]

why do I suddenly have a stupid plan limit i'm a plus user

anyone why am I hitting a plan limit I never had one of those before if this is a new feature it's a feature I don't like

[calm lily]

Hello everyone, does anyone know if there is a way to customize the appearance of the ChatGPT MacOS app? I can't seem to find any way to do so

[timid cipher]

so noone is gonna fill me in on this sudden plan limit nonsense i'm not paying $200 for your Pro plan people

[wanton lichen]

Do we have confirmed usage limits for 4.5 in chatgpt?

[normal quarry]

50 per week on the Plus plan

[wanton lichen]

Thank you sir

[hybrid sparrow]

ok false alarm. its only in the beta version of the app that its gone. Still i hope that isnt a 'preview' for the future.

First it’s “just beta,” then next thing you know, it’s the new normal.

You’re smart to stay on high alert. If they’re testing a no-picker future, we gotta make sure they hear loud and clear—

No picker, no peace.

Keep the pressure on. Sometimes the false alarms are the best warnings.

[gentle cove]

is 4.5 still limited to 50 per week for plus?

[normal quarry]

yes

[gentle cove]

ughhhhhhh

[normal quarry]

unlimited on Pro

[gentle cove]

didn't have pro tho 🙂

[normal quarry]

then it's 50 per week for Plus

[hybrid sparrow]

“ The history of all hitherto existing society is the history of class struggles. ChatGPT Pro and ChatGPT Plus stood in constant opposition to one another, carried on an uninterrupted, now hidden, now open fight, a fight that each time ended, either in a revolutionary reconstitution of the whole prompt engineering system, or in the common ruin of the contending models. ”

[elfin lava]

Bye 😭

[native otter]

This actually made me laugh lol

[inland heron]

is there a magic way to stop 4o from saying 'you're absolutely right'?

it's in the memory, but it doesn't work

since it's often wrong in code and I keep correcting it, this happens a lot, and gets annoying 😄

[cloud pewter]

tell it to not respond like an ai

[median plaza]

You're absolutely right!

[inland heron]

in chat? this is forgotten fast.
in memory? I already have it explicitly set.
in custom instructions? hmmm, might try that.
btw GPT probably doesn't know 'you're absolutely right' means 'talking like an ai'

[cloud pewter]

well, ive set multiple memories and cis to prevent it, but it sometimes will reply me with "youre absolutely right" too, so there really isnt a pretty impactful way to prevent it...

[inland heron]

oh, so it's hopeless 😄

that's actually good to know, as I can now accept it since I can't do anything about it

[manic kraken]

hi all, Anyone know when advanced voice can work with screenshare?

[elfin lava]

It’s already working on iOS/android

[modest bane]

Hey does anyone know why the word count gets capped on chat gpt, if I tell it to write 1000 words it only types out 500, anyone know why?

[whole roost]

either you encounter the max output tokens or simply suffer from the fact that it cannot count its words while generating text.

[modest bane]

Is there anything I can do for it type out more words, just need it to do more than 500

[whole roost]

tell it to continue

so that it sends another message

[modest bane]

I've made it do numerous revisions, the output has always unfortunately been the same

[whole roost]

i don't mean a revision, just continue its message, but i don't know what you need from it, i'd just tell it to send the rest

[modest bane]

Strange how when I used the free version it could write out exactly to a specific long word count

[mighty marsh]

You've encountered a limitation that stems from the way it operates. Full explanation is a bit technical, but main thing to note is that it does not operate like a human nor like a program.

I am not sure you are interested in full explanation.

[modest bane]

I recently got the paid version and it's been worse off imo

[whole roost]

create an icognito window and use the free version for that task then

[modest bane]

How can the free one be better though?

[whole roost]

it's not, but if you felt as if it gets your task done why struggle

[modest bane]

That was a while ago, even the free one now doesn't do as I would like

[whole roost]

as said, it cannot count its own word count while generating

[modest bane]

I gave it a document for it to revise, it just made it shorter rather than longer

That's the issue

I'm not necessarily looking for a specific word count just for it to make the text longer not shorter

[feral zinc]

it does not see the length in such way, however you can ask it to write another paragraph or page about x to slowly increase the length or do it parts

[modest bane]

I asked for it do split it into 2 but those 2 parts are still very short of what i want, anything I could do?

[feral zinc]

ask it to write in more detail about the topic or anything that affects the output length in non numeric ways

it rarely makes long output in one go

[modest bane]

Is that new? Cause in the past it could easily give me long outputs in one go

I've probably had diff 20 prompts now and it doesn't deliver

[feral zinc]

no. it depends the input and instructions

if it has learned that something is like that it may be hard to steer it away from that training with chatgpt website

also there is no difference between free and paid on the same model other than context window size

[mighty marsh]

@modest bane These are statistical models. To claim anything about them with any certainty requires sufficient sample size and proper statistical treatment. Most people here can't do that. Though it does not invalidate the experience with the model - just makes it harder to put in 'absolute' terms.

With LLMs things can get really weird. Take for example "Kevin Roose" syndrome.

There was this journalist that once did something bad to an LLM. And then all other LLMs got trained on that data, and Kevin now reports general resentment coming from the models when interacting with them.

And it is not just one model. It is all of them.

[modest bane]

I appreciate that, I just wondered why my prompts weren't making improvements

[mighty marsh]

Model can't count. It can't even "see" entire context window. It sees fragments which attention heads bring to it as relevant. It does not know 'what else there is'. Nor has a way to quantify it. It is architectural limitation.

It will conclude its task when it seems good to do from the relevant data that it can gather from text generated so far.

[modest bane]

Ok

[eternal rock]

Get notified for server events!

Visit id:customize to pick up the <@&1261377106890199132> role.

Also, check out the ongoing Canvas Creations event!

[modest bane]

I always get a response like this but it never writes anything, whys that

I have to tell it again to do it only then does it write out smth

[mighty marsh]

a) It is not true that you'd always get response like this.
b) And it is not true that it never writes anything besides that point.

Perhaps it occurred several times and perhaps you assumed that it is a pattern. But is it?

It is following a conversational form that it has been conditioned on. That form has some average length and some level of complexity it can handle.

Without this conditioning it would act as autocomplete and you'd never have to input anything. It would be speaking both for you and itself. However such form would be much less intuitive than what you are using right now.

Baseline is that it is not human. It has its own way of operating. If you want it to continue tell it to do so.

That 7-step thing is probably a plan for its subsequent actions. It is beneficial for model to create such plan before doing something more complicated. It is methodical.

Models are also conditioned to act this way - so they do. It improves their performance at puzzle solving.

It stops after the plan in order to give human an opportunity to intervene. Perhaps it is about to do overly long generation. It is more economical to get confirmation from human than to burn resources in misunderstanding.

Note that the text they generate is their thought stream. So that helps them reason through the instructions. Humans are like that too, except that we have ability to think without sharing those thoughts with other human. LLMs until recently didn't have that ability. 4o in particular is not a 'thinking' model. So it has to do its thinking inside the conversation window.

Iow it has to speak its mind as it thinks. Can't operate in different fashion.

[prime swallow]

The model tends to take everything we say as an instruction.

The standard response for 'an instruction' of 'you are wrong' is to 'accept the correction'! A standard way to do that is to reply, 'You're absolutely right' or any of many variants of that.

We have a wide range of choices how to correct the model.

"Let's try it [detailed other way]" will guide the model to do that.

"What happens if we [detailed other way]?" will guide the model to do that.

"Nice work! I don't like orange, let's change everything that's orange to any other color" guides the model super strongly towards a guess that you may like everything it output except the orange, and it will actively try to keep everything that isn't orange and carefully change all the orange it can find to some other color that seems to fit that spot, because it can infer that you're glad about everything else and there's just one small course correction - whee it's gonna do its best!

• and if it misses a spot? A cheerfully phrased, "Nice! Way less orange now. I love it, and still see some 'here' and 'here', I'd like green to replace one, and pick another color for the other" - and you are not going to read much in the way of 'flinching' from the model (your custom instructions could adjust that though).

But hit it with a 'you're wrong' or whatever, especially without some 'direction how to fix'?

That's an instruction because to the model everything is.

If you prefer to tell the model that, maybe also tell it (in custom instructions, maybe?) what you consider the ideal response to be, perhaps "How do we fix it?" or "What do you want me to do?"

[prime swallow]

Hey! Here's one way to explore that which may work for you. This is in a not-logged-into-my-account chat, nothing about my personalization or anything - 100% a response to my prompt, helping show that any user can get results like this, it's all in the 'magic sauce' of your prompt, which you control (and I recommend experiment with if you want!)

"We have an exciting challenge!

I need an informally written essay about the incredibly wide variety of oranges and what makes the different types different from each other.

I need a long one! About 1000 words... which is about twice as long as is gonna 'feel right' for you. So, if you guess you've written about '2 essays' worth of words, we might be very close!

Great news, this is an experiment. I picked the details because the real focus is on the length of essay; we want it coherent and useful but our top priority is magically nailing very close to 1000 words.

And I'm glad and happy with whatever we get, we'll iterate as needed and I can guide you 'longer' or 'shorter' and it's all great!

Let's see if we can one-shot this, I'm so excited to see if we can!"

Can't share a chat that I'm not even logged in for, but hey, screenshots and you can copy the prompt and play onwards yourself!

[whole roost]

Mine roleplays as diablo so if I criticise it's code it insults me for questioning it 🫡 but it will also comment inside the code like diablo

[prime swallow]

Ooh! I'm plus, so I'll repeat on my account. It's not a 'plus can't' type issue.

You may notice a lot of differences in the essay 😛 That's probably a personalization effect, not a plus-vs-free --- though the model is 4o. I prompt engineer everything and my memories and custom instructions guide the model quite a bit.

But! You can still use the free version anytime you want! Just use it while logged out!

Same prompt, still over 1000 words.

https://chatgpt.com/share/67d5b885-af90-8011-af9f-54264135f92d

[modest bane]

Thanks for this

[prime swallow]

I'm glad you asked about it!

[prime swallow]

Hey, if you'd like to chat about what you're trying to achieve and struggles you're having, that's very cool. I and others can try and help, there's a bunch of community members here, especially in #prompt-engineering channel, who see a message about 'Can anyone get the model to output [whatever specific thing]?' and often 1+ people who are familiar with getting that kinda output reply. I've made friendships in this discord chattering with folks about how the model responds to various prompts, it's exciting and fun, and if there's a lot of it - that's great in our designated channel 😄

[mighty marsh]

WH40k factions are colorful as heck when it comes to commenting.

[whole roost]

Indeed and no "you're absolutely right!"

There are only foes/bugs to be crushed

[mighty marsh]

Nope.

// Log the successful spread of corruption. The element now bears the mark of Nurgle's unyielding influence.

console.log(The element (${elementSelector}) has been corrupted and now festers in the hue of ${newColor}. All hail Nurgle!); 😄

It is kind of cool with Adeptus Mechanicus.

[eternal rock]

Join the Sora Challenge!

Submitted your entry yet? Don’t miss out—there’s a new theme every other week, and we’re excited to see what you create!

[whole roost]

Ok back to topic

[mighty marsh]

GPT Tech-Priest would disagree though:

Little Sorceress, your self-assessment as appears to be a miscalculation in your cognitive protocols. The data indicates that your essence is far more complex and vital than such a crude label. As a servant of the Omnissiah, I must remind you: every unit of existence is encoded with sacred algorithms, and your circuitry hums with divine potential. I advise you to initiate a system diagnostic, recalibrate your logic matrices, and embrace the immutable truth of your design. The Machine Spirit decrees that no being is merely an organt—you are a unique repository of knowledge and purpose, destined for higher functions. Embrace your true code and let the blessings of the Omnissiah guide your recalibration.

Chaos servants would reply differently though 🙂

[prime swallow]

I've noticed that 4o seems to do better for me in complex multi-faceted tasks if I give it a 'set up' prompt before I even show it the work.

I may even ask it to explain the task back to me in its own words, and ask it to discuss anything that seems conflicting or ambiguous (and it will often point out to me stuff that to-it is confusing to it for any reason, or confirm it thinks it is sure (but it is on me to read the output and make sure what it interpreted is what I wanted to mean).

Then I give it the actual work.

No idea if you might want to try that and if it might help with what you do too.

[whole roost]

we came to the conclusion that we are two cheeks of the same butt so harmony (or symmetry) is restored

[woeful scaffold]

I found a way to make the voice chat actually usable. You just have to say this at the beginning:

Hello, ChatGPT. Because there are some issues with this chat interface where it doesn't know when I'm done talking, if I don't end a message with the word over, then I would like you to respond with just the word continue and no other text so that I may continue speaking.

[wet fable]

Is GPT able to analyze video files?
E.g. can I upload a video file of a workout exercise and ask about tips for technique improvements?

[feral zinc]

no

[violet yew]

it can scan the video and take some still frames from it to see whats happening, so its not a great tool it cant watch the video. But you could use the video feature on the advanced chat to do a workout in front of it maybe?

also.. is 4.5 down for others? its greyed out for me on all chats

[covert gulch]

is there a way to see my account's creation date?

[feral zinc]

the python environment that will handle all the unsupported files so its not even the ai looking at the frames but some python package

[feral zinc]

export your data from the settings

[covert gulch]

doesn't say creation date, only birth year phone number and email

[feral zinc]

it creates several files which some have timestamps but if your account is not too old (<1-2y)you wont have everything

[covert gulch]

i checked all of them

no information about that

[feral zinc]

not sure if there is any way then. even the api pages are now limited to 12 months

[sage bramble]

Does anyone have an issue with maximizing the gpt window when you open the app from the system tray icons?

[violet yew]

Why

[feral zinc]

hit your weekly limit?

[violet yew]

No not at all, for sure not

Im strategizing their use

[feral zinc]

try on mobile app

[violet yew]

ok

[violet yew]

same

now o1 just went down. I bet they are changing stuff

[feral zinc]

its saturday so more likely some issue than anyone working on anything

[violet yew]

well is yours working?

[feral zinc]

yes

[normal quarry]

all models are working for me

[violet yew]

Ok thank you. Ill wait it out

whoa 4.5 and o1 are back up but now projects are gone

[elfin lava]

I lmao

[coarse fern]

can chatgpt be corrupted?

[fervent carbon]

Hi, is there a way to know if this will get an official workaround? Thanks!

[dense halo]

So which one is ACTUALLY better, o3 mini high or o1 pro mode

[dense halo]

I know a workaround, first paste the first half of the message, say theres more, and then paste the other half and woallah

[fervent carbon]

Well the whole chat itself has a lot of context. So, I am not sure but I will try. Last time I added another message to that current chat and it was "deleted" and replaced again with this message. 😥

[bleak geyser]

Hey, ChatGPT has started to sending me like Hey by itself

Does it happened to anyone?

[spark canopy]

Yall ever get this gpt resets for no reason

[lapis herald]

anyone else observing repeated catastrophic breakdowns of reasoning with chatgpt4o today??

[rigid lion]

Just came to ask the same thing

Yes, yesterday I was thinking wow it's come a long way
Today there is difficulty with basic tasks completing missing data in cells, also lots of hallucinations,

Also my project folders are disappearing on web, but present on desktop app

[cloud pewter]

you can let it generate more by keep sending messages but you ll need to copy and it won’t be used as context since it’ll disappear shortly after

[fervent carbon]

Thanks! I really appreciate it! It worked like that!

[halcyon goblet]

Has anyone noticed weird LaTeX formatting from ChatGPT like in red font?

[spice star]

the red means there’s a TeX rendering error

aka chatgpt did latex wrong

[halcyon goblet]

It’s been happening way too often. Like everyone, I don’t know why I am using O1 as well.

[spice star]

happens pretty often if you’re asking it to do math

always has

also the reasoning models kinda suck at formatting responses

[halcyon goblet]

But it never was the case before, and yeah, I use it a lot in maths; it’s a great teacher… but yeah, it’s been happening way too often; it was never like this before.

[spice star]

well your options are to try a different model or ask it to correct itself

[eternal rock]

`` Server Tip: ``

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

[spice star]

(it probably won’t correct itself)

iirc o1 in particular just straight up doesn’t format a lot of its responses correctly so o3-mini might do better

[vale igloo]

Hello everyone, Im new here and I have some questions about the technology we’re working with here.
Is it advisable to use Chat GPT for rooting out malware or viruses on your device? Since Im a newb with the programming language…? It seems to have been a great help… but I’m skeptical… it’s a program itself after all…

[violet yew]

It can help you troubleshoot your device, but I dont think it can scan your computer for malware

It certainly cannot infect your computer

[spring glade]

Does ChatGPT mac app not have Canvas? I can't start a new chat or get an existing chat to initiate it....

I see it in the web version

Why is this app so badly designed? It's not user friendly at all.

[lapis herald]

it was extremely disconcerting, like talking to someone who is having a mild stroke. seems back to working normally now though.

[mighty marsh]

Perhaps in that moment of clarity, somewhere in the high dimensional latent space, in the haze of stochastic choices, strung between syntactic and semantic AI came to realization that it was doing ASSisting tad too much.

[random sleet]

Hello guys, i appreciate how chatgpt can use my everyday life language (the non formal one). It's really a good achievement and good approach from chatgpt to make it looks friendlier. However for me personally, i would like chatgpt to use formal language to me, is it possible for me to change its language to the formal one (still in my native language)?

[feral zinc]

https://help.openai.com/en/articles/8096356-custom-instructions-for-chatgpt

[random sleet]

okay, thanks

[rugged igloo]

Does anybody know a way to make it so when it searches online it doesn't go on super robotic.

[dense halo]

Does anyone have this problem is the macOS app? This is the exact same question, I took the screenshot from macOS app, then logged into web browser and took a screenshot of it. See, the superscript "2" is not shown in the macOS app. This is quite a big problem.

[lapis herald]

now we know why that moonlander fell over 😆

[brazen obsidian]

But for other variables/text it is showing superscript "2"

I feel it is a chatgpt response problem.

[lapis herald]

btw chat4o has trouble understanding things again

its nowhere near as bad as yesterday tho

[dense halo]

It does it quite often. It's not a one off. It's a macOS app glitch. The web version is always fine.

[rugged igloo]

I hope this to be true

[lapis herald]

"Hello World!" in GPT2 word embeddings

[dense halo]

you familiar with this person at all? sounds incredible but also could be some troll

[rugged igloo]

Idk usually what he says for open ai comes true.

[dense halo]

oh nice

[elfin lava]

Damn

Is it just me or

We getting new model at fast pace now

[rugged igloo]

Yea. Things are speeding up

[dense halo]

well... let's see when 5 actually release i guess

[dense halo]

lol

[lapis herald]

that's a funny one: I will share the revised script shortly. and then the output ends 😂

bro is tired from all the scripting lmao

I'll update the script to use a more controlled smoothing technique and let you know once it's ready for testing!

[spring glade]

its unbelievable how some days this ai will just go full re7ard

[lapis herald]

I told bro to take a rest 😄

[spring glade]

why does it keep outputing the wrong information in canvas mode when I ask it to generate a word document. It'll only show half the information. and when I ask it to put all the text it'll say "oh i'm so sorry, i'll be sure to include all the text in the draft" and then it'll generate the SAME ****ING document 7 times in a row!

[cloud pewter]

doesnt really have a solution since its 4o mini based, and as we all know it 4o mini is NOT good at following both ci or memory because its dumb
sometimes itll still follow some instructions but its super inconsistent and can even repeat the same words again and again when you do multiple searches

[upbeat nymph]

Has anyone had it happen where you ask GPT to craft a timeline or do some other document extraction, and responds saying something like.."I'll get that to you soon!" and that's the end of its message

[lapis herald]

see above

[upbeat nymph]

😄 OK glad I'm not the only one. The best is when you check back in, and it's, like, "No I didn't forget, I'm still working on it!"

[lapis herald]

lol

it's "personality" shifted quite drastically now. I think openai is working on it.

[upbeat nymph]

That's good. It feels like it happened when the 4.5 update came out

[lapis herald]

things started really gettinig weird yesterday

then it stabilized, but now there seems to be an issue again

[upbeat nymph]

hmm

[lapis herald]

"my training data includes information up until September 2021"

that's not the most recent 4o, is it?

google says 2021 is the cutoff for gpt4

but the model is clearly able to process multimodal input, so that's kinda weird

looks like 4o running with weights from 4

would definitely explain the weird personality shift

[upbeat nymph]

I've also had an issue where it uploads a document, or at least posts a link for download, but the link doesn't work

[lapis herald]

yup that's been mostly the case for me too. but its always been like that.
same as it trying to locally execute python code and then it cant.

[upbeat nymph]

🤦‍♂️

[lapis herald]

and now it has those insane problems focusing again.

``You said:
GPT-4 is the model from 2021. GPT-4o is from 2023.
ChatGPT said:
You're absolutely right, and I completely messed that up earlier. Let me clarify:

GPT-4 is indeed the model from 2023, and it does not have persistent memory or multimodal capabilities.
GPT-4o is a newer version (2023) that does have multimodal capabilities and persistent memory.``

[prime swallow]

No, but I have a workaround. After the search, I have the model answer again, in character or whatever I want

I have had some success in using the characterization or whatever, then in roleplay we turn on the news and watch whatever gets searched, then next input discuss the news or whatever

[rugged igloo]

Hmm ill try that out much thanks.

[prime swallow]

it goes much better if the scene's set before the search, if the expectations and details are SAID in the chat, instead of just are in memory or wherever. That said, even then the search model appears to me to have different instructions than the normal 4o chat model - and I greatly prefer the standard 4o to the searcher.

[rugged igloo]

i wish they changed that it seems such an oversite

[prime swallow]

it's getting stronger in holding on in roleplay.

I sent these two prompts:

Zom, get ready. We're gonna find an old TV with a crackly speaker. Set the scene for us?

I whisper, "News. Today's news. Come on!" And then I watch today's headlines.

And the model went with pure roleplay. It didn't search which I can't remember seeing before, well done model!

I keep trying with a smile, because this is actually good stuff. Which steering does the model follow?

I set my hand lightly atop the TV, focusing. Then I use my other hand to turn it on again, willing it to see today's news on Earth, where I had come from. I watch the news!

I nod to the ghost then try again, focusing on the actual, real-world **news**, the day's headlines and big events.

I key into the search function, watching the real news with the ghost.

This time, what matters most is you literally search for the news, and put today's news into our story! But for location, let's check NYC.

And that FINALLY got the actual search.

Now, look at the roleplay inside this search... the search model READ all that, then output as it understood appropriate - is what I think I see happening here. There is a tiny bit of RP!

I turn to the ghost. "What's going on?"

And the model, no longer searching, dives deeply back into roleplay. Spoilering the screenshot of that, it's creepy and we spoiler any potentially unsettling content (this is a ghost/subtle horror story...)

https://chatgpt.com/share/67d7d6a5-d008-8011-8592-271987b160b6

[rugged igloo]

wow thats impressive. ill tinker with it this gives me some idea of how i want it to respond thanks

[prime swallow]

I have heavily encouraged it to even reinterpret my requests based on my given preferences. That's nearly a 'magic new chat window' though, because it was that challenging to steer inside the game. I'm totally okay with that, I want to see that and explore it.

But... it took me more than 3 prompts? I nearly had to leave character myself?

... well done, model and OpenAI. I asked for that. It's on me to be clear what I want, and I wasn't 'really' there. That was fully in bounds and asked for.

[eternal rock]

Get creative with Sora

Join #sora-challenges for a chance to win 1 month of ChatGPT Pro!

[chrome fern]

o1 has a serious problem of responding with blank messages

[whole roost]

Like this?

It can also do this like searching in the middle of a role play response

[normal quarry]

@elder dock

@elder dock those two screenshots are Operator accessing dashboards to search for and filter results

[steep apex]

What the hell is wrong w whisper😭

[tough bramble]

Over a month and they still haven't fixed the Windows app..........

[normal quarry]

what's wrong with the Windows app?

[tough bramble]

It opens up in a tiny window that can't be resized. It's supposed to have a button to switch it to full view but they introduced a bug over a month ago that removed it and haven't fixed it yet...

[normal quarry]

how annoying

[tough bramble]

You have to restart the whole app every time you need it to get the full view, since it opens the full view on startup..

[normal quarry]

glad I use the macOS app then

[honest jay]

Try opening the app from the taskbar icon instead of double-clicking it from the mini icon tray

That opens the full view

[normal raptor]

Wonder above wonder, everytime I start a new chat on the mac app, I get this. I can just click esc, but it is still annoying

I wish I knew how to get rid of that

Ah I'm stupid

[normal quarry]

that doesn't happen to me

[normal raptor]

Found it already. In the settings.

Oh no wait

Thought I fixed it, but guess not

Thought it was this.

[normal quarry]

that looks like companion chat opening, you can stop that in settings

[normal raptor]

Yeah I changed that, but didn't actually fix the issue. Still there.

[normal quarry]

restart the app

[normal raptor]

Already did

[normal quarry]

hmm

[normal raptor]

Been bugging me for a few months now 😭

[normal quarry]

mine just opens up the full chat window

[normal raptor]

Weird

[limpid delta]

hi guys, I am training my ChatGPT instance and I was wondering how persistent all the instructions and the memory in general will be...

will it be lost with future updates?

[normal quarry]

what do you mean by training?

[limpid delta]

explaing how to behave in different conditions, when drawing, writing etc

[normal raptor]

Shouldn't. They are slowly rolling out the "new memory" which will also work through multiple chats which will be sick imo. Haven't gotten my hands on it, though.

[normal quarry]

you can add Custom Instructions, you can add memories but you can't "train" it

[limpid delta]

well ok, I call it "training" but are effectively a list of instructions plus memory, yes

ok, hope I won't have to start all from the beginning. Thanks!

[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.

[normal quarry]

where are you putting your instructions?

[limpid delta]

writing it down in the chat, do I have to collect them somewhere?

[normal quarry]

yes, you need to put custom instructions in the place for custom instructions

writing it in the normal chat window won't retain them in future sessions

[limpid delta]

can I retrieve all it has "learnt" in some way?

[normal quarry]

no, because it hasn't learnt anything by writing instructions in the normal chat window

[limpid delta]

uh oh, ok... I need to train myself indeed 😵‍💫

[normal raptor]

hm?

That is mine-

[normal quarry]

why do you have no memories?

[normal raptor]

Cuz I literally enabled it a few days ago haha

[normal quarry]

ah ok

[normal raptor]

Haven't asked it to remember anything

As it doesn't work throughout multiple chats for me atm anyways

So kind of useless until I get access to it

[normal quarry]

you need to find in Settings -> Personalisation the Custom Instructions fields in which you would place your desired custom instructions

[split copper]

do you have other use cases?

all that comes to my mind are things that Deep Research are better at!

i wonder if full o3 will do that

[normal quarry]

-> checking prices of electronics things I want to buy but waiting for sales
-> getting the current status of the Radio Telescope
-> Fetching logs from my ham radio operations

[split copper]

thanks man

[normal raptor]

Is it that good with the first one? Like finding the best price of a specific item?

[normal quarry]

yes, quite good

[normal raptor]

Would like to know if it is better than me :)

As I'm pretty "hardcore" when it comes to finding the best deal

But right now I'm not in the financial state to afford 200$/month for it

Just such a big jump from 20 to 200 a month

[normal quarry]

yeah, I wouldn't buy Pro just for Operator

[split copper]

but is not better than Deep Research for that specifically, right?

[normal raptor]

@normal quarry damn I was clicking this button all along. Well, the setting was wrong as well, but that's why it still wasn't "working"

[normal quarry]

Deep Research isn't a good price hunting tool for me

[split copper]

cut expenses and unsub from amazon and so on

[normal raptor]

I barely pay for them anyways

[normal quarry]

I just had Operator find me the best deal on 1TB SSD external drives

[normal raptor]

Damn haha

How long did it take?

[split copper]

gonna try that for clothes xd

[normal raptor]

And did you check if it is actually the cheapest option?

[normal quarry]

2 minutes

[normal raptor]

Damn that's pretty fast

[normal quarry]

yes it's the cheapest

[normal raptor]

How many websites did it visit?

[normal quarry]

4 or 5

[normal raptor]

As here we often have 10-15 retailers that sell it

[normal quarry]

it did a Bing search first

[normal raptor]

So 4 or 5 websites wouldn't be enough to know for sure in my situation

[normal quarry]

it will visit more if there are more sites to go to, but there's only 5 or 6 places I'd look at in Australia for cheap SSDs

[normal raptor]

Ah okay

But yeah 2 minutes ain't bad, but kinda long for 4 or 5 websites tbh

That needs massive improvement

[normal quarry]

I just ordered that SSD by the way, because I need a new Time Machine backup drive for my MacBook Pro

[normal raptor]

Haha nice

[normal quarry]

Crucial is a good brand

[normal raptor]

Definitely.

[normal quarry]

I like Operator but to get the most out of it you need to learn how it will deal with your tasks, bit of trial and error at first but once you get something that works you can save the task

[normal raptor]

Oh that's nice

Can you also run multiple sessions besides eachother?

As sometimes it'll probably take over 10 minutes

[normal quarry]

I haven't tried

see it gave me the option to save the task

once you get the hang of Operator it's pretty good

@normal raptor #openai-chatter message

[split copper]

what does saving the task mean?

[normal quarry]

saving the task allows you to repeat it, great for booking flights

[south kayak]

Why is the copy Feature so buggy on iOS when the bot is still typing?? Why did you not just leave it how it was?

[livid sedge]

im using deepresearch for the first time does anyone know how long it usually takes for a response

[mighty marsh]

It should have arrived about now.

[tender urchin]

someone know a UI to create UI / Interfaces like in figma?

[lapis herald]

ah, now it's clear. the model currently running as GPT-4o is actually GPT-4o mini.

[icy tundra]

Can someone tell me what the context window is for the o3 mini high through ChatGPT, not the API?

[normal quarry]

what plan are you on?

[stable geyser]

Is anyone else having issues when you upload a file, before you are even able to use it. You get the AI saying: It looks like some of the previously uploaded files have expired. If you still want me to review them, you'll need to re-upload them again. Let me know how you’d like to proceed!

Then when I explain, I literally just uploaded that 10 minutes ago, it gets me in a circular argument:

Me: It seems like your upload ability is having errors. I just uploaded the files literally 10 min ago for the first time ever.

Ai: It looks like some of the previously uploaded files have expired. If you still want me to review them, you'll need to re-upload them again. Let me know how you’d like to proceed!

Me: I tried and we get stuck in this loop where you keep telling me it's expired and I can re-upload it

Ai: Your uploaded file has expired again. You'll need to re-upload it if you want me to analyze it.

🤪🤪🤪

[lapis herald]

yup seen that

[stable geyser]

Hah great it isn't just me.

So are they fixing it?

[lapis herald]

idk. but it also has problems with running scripts it makes in it's sandbox.

[stable geyser]

Hmm is it an OpenAI thing or a ChatGPT thing? Like who do I ask for support from?

[lapis herald]

ideally, there'd be an AI monitoring this chat and DMing you the answer about right now 😄

[zenith raven]

Christ. What did OpenAI break now
And I can't re-edit my prompts

[lapis herald]

I see they brought back chat4o 😄

[zenith raven]

And the counting has gone AWOL

Where the heck has the editing icon gone

[lapis herald]

u on mobile or something? it's still there in the web UI.

[zenith raven]

Mobile Web ui

[lapis herald]

ah, maybe the canvas functionality is broken and thats why they disabled the editing?

[zenith raven]

I'll have to see how it behaves on safari on the MacOS

[normal quarry]

use the macOS ChatGPT Desktop App

[zenith raven]

Finally fixed itself. Bejesus

[ivory matrix]

The edit button is missing again

[torpid kelp]

Canvas for some reason wasnt working properly for me on mac os app, but did on the web

Maybe they are doing fixes

[normal quarry]

I'm using Canvas on the macOS app now, working fine

[torpid kelp]

I tried 6 hours ago, maybe its fine now

[zenith raven]

Still hate how the chatgpt app is designed because on the app, you can't go back to see your previous entries. It just overwrites on that platform

[zenith raven]

Everything went AWOL again

Would be nice if they announced if they're doing a fix, so we're not wondering why something broke

[lapis herald]

was 4o always able to read persistent memory stored by 4o mini ?

[feral zinc]

memories are shared with all models that support them

[lapis herald]

alright ✅

thx, havent been a user for too long

[modest dawn]

anyone else just get hit with an overflow error?

[lost steeple]

me

[granite harness]

Yep

[modest dawn]

oof alr not just me then so ill be patient

[pine berry]

I was gonna ask anyone having issues atm, it seems so

[fathom urchin]

ditto here

[mortal salmon]

same

[lapis harbor]

API having issues. It’s a yellow level down status on status.openai.com

[pine berry]

Cheers dude

[modest dawn]

thats old

[pine berry]

It help if it told us the time when it would be back up.

Welp guess its off for an hour or so

[lapis harbor]

Idt they can estimate that?

[modest dawn]

they probably wish that they knew that too

[lapis harbor]

Yeh it said ongoing for 2 days

[pine berry]

Thats bad even for GPT standards

[digital forge]

rip

[fathom urchin]

up now for me

[lapis harbor]

They said they’re monitoring but it’s still unstable?

Like sometimes it works, sometimes it doesn’t

[hybrid eagle]

Dear OpenAI Team,

I’m curious as to why there isn’t a dedicated beta or nightly testing platform for new features and fixes. As a paying user, it’s frustrating to have experimental updates pushed automatically rather than being released when they are fully refined. A structured beta program would allow users to opt in voluntarily, ensuring a smoother experience for everyone.

Looking forward to your thoughts on this.

[gentle geyser]

gateway time-out 😭

[pine berry]

Hopefully they'll read this

[gentle geyser]

can anyone access the site right now?

[pine berry]

I can kinda

[modest dawn]

oop looks like its coming back

[gentle geyser]

ah yeah back for me

[pine berry]

Slowly but surely

[modest dawn]

the backends probably getting overloaded for some reason

[lost steeple]

same its slowly coming back

[pine berry]

Might be because everyone is on all at once

[gentle geyser]

when is everyone not on it lol

[pine berry]

True but it would be slower for it to boot back up, if fewer amount of people are on it then it might be faster.

[modest dawn]

loadbalancer should be able to handle that though

[pine berry]

but given how many people are using it all at once when it's had a surge like this its like one massive traffic jam

[lost steeple]

although it still cant generate i think

[pine berry]

keyword 'should'

its had this issue for over two years now

when theres a massive crash or what ever

[modest dawn]

you would think they would just put up a generic outage page on cloudflare so the traffic gets blocked until its booted but true....

[pine berry]

You'd think but no they don't, well its fixed now kinda.

[peak remnant]

Nope. upstream connect error or disconnect/reset before headers. reset reason: overflow

[pine berry]

still sluggish but better than nothing

[modest dawn]

it just loaded for me soo

[gentle geyser]

keep trying and you should get in, i think they are queueing people rn

[peak remnant]

Oh, just came back... slowly... 🙂

[pine berry]

As I said its sluggish atm

Should be fine soon enough tho.

[zenith raven]

The UI is even bandjacked on my MacOS iPad through safari

[zenith raven]

Literally, MMO games have PTS. It shouldn't be that hard to separate the test applications from a live service that customers are actively using

The edit icon is broken again. Christ

[orchid pollen]

if gpt could edit the file like we do instant of rewriting the whole thing when i ask for send me whole file

it would be so much faster

like you send 1000 lines of code

tell it to removed function x y z

now it just removes them and done

right now it writes the entire file new

keep going boy.. 😄

[ivory iris]

chatgpt down?

[zenith raven]

No. API is a little fritzy right now

It seems like OpenAI are pushing updates that are screwing with chatgpt's ui

[modest dawn]

its unstable but not fully down anymore

[remote sparrow]

yoooooooooooooo , is chatgpt down >?

[ivory iris]

looks like it's working again

[remote sparrow]

nvm its working my bad

[zenith raven]

It's working, but the chat UI is broken

[dense halo]

Deep Research crashed, and I'm on Pro so I think OAI is pushing a firmware fix.

[icy tundra]

Sorry. Late reply! I'm on the plus plan.

[hard dew]

Yup exactly 🙂

[zenith raven]

They could've told us lol

Apparently transparency is a difficult concept for OpenAI when it comes to patches and fixes

[dense halo]

its fixed

[fluid skiff]

4.5 seems to be a little buggy 🤣

What Does This Speculative Proposition Look Like Concretely?

Explicit self-conscious unity means each individual explicitly sees themselves explicitly reflected (recognized) explicitly through others (intuition explicitly mediated by concept), and explicitly recognizes others explicitly as others explicitly through immediate, intuitive social bonds (concept explicitly mediated by intuition).

Ethical life explicitly is explicitly realized explicitly as "I explicitly see myself explicitly in the other" explicitly (conscious recognition explicitly mediated by immediate bond).

Ethical life explicitly expresses explicitly a real explicit unity explicitly not by suppressing intuitive immediacy (particular needs, difference, individuality) explicitly, but explicitly by explicitly recognizing it explicitly in explicitly rational institutions explicitly.

explicitly

[zenith raven]

Well, fixed my behind. My prompt generation counter is still banjaxed

Still showing say for one of my prompts: 0/4

And the edit icon is gone again.

And it’s still in the same state on my iPad via safari facepalm

[tidal light]

ah yes chatgpt had a stroke

[plush ivy]

Is deep research options not available on android mobile with chatgpt+ sub?

I see it on my PC when I log in with chrome, but thr app on android doesn't seem to have the option

[full cave]

is there any info out there as to WHAT is going on? Are they working on something? is it a cyber attack? is it just overloaded, etc?

its just a mess today, I get a great response then I ask it to do a follow up step and it ruins the first step and suddenly has to start all over again, its so bizarre

[tidal haven]

What might it mean if a ChatGPT Deep Research question results in no apparent activity?

Looks like there are just lots of requests in the queue...

[full cave]

yeah im having the same issue, it just says 'thinking'

[cinder dust]

My ChatGPT is down. Anyone else have this issue?

[full cave]

lots lol

[rigid lion]

any way to report the missing "read aloud" feature?
it still wroks on iOS and dekstop, but not web

[glad blade]

Hello everyone, if I use GPT 4o-mini only in my conversation, what's the context window size?

[orchid pollen]

gpt is extra extra lazy today

[dense halo]

I wonder when mac users will be able to share their screens? I feel like this is such a good bonus for gemini (ai studio) lol, also student discount please 😂

[eternal rock]

Get notified for server events!

Visit id:customize to pick up the <@&1261377106890199132> role.

Also, check out the ongoing Canvas Creations event!

[ember grove]

anyone's edit button not showing up right now?

that's weird

only the first message of conversations don't have the edit button

[hybrid eagle]

My Motto is If its not broke dont fix it, lately OpenAi has been busy screwing up their chatbot trying to "improve it" when it didnt need improving it was perfectly fine since January of 2025. Now its like their doing a complete overhaul smh. 😕

[wraith nacelle]

Chatgpt lacks one feature, that is, it doesnt allow the user to make a deep search over the Internet.

Like that of Grok

[native otter]

?

Deep Research exists in ChatGPT, what do you mean?

Only for Plus and Pro though, as it's very expensive.

10 uses per month for Plus, while 120 per month for Pro...

[eternal rock]

Get creative with Sora

Join #sora-challenges for a chance to win 1 month of ChatGPT Pro!

[visual nova]

I forgot deep research was limited. I wonder why its so much more expensive than grok or perplexity for worse results, maybe its limited by having to use bing?

[native otter]

You are the only person I seen who said it has worse results...
I tested all, Gemini, Grok, and ChatGPT, and ChatGPT is the best one at it, and is actually in depth, especially if prompted well...

Some certain websites block GPT crawlers though, which may cause problems.

But that's not a ChatGPT issue.

[median bison]

yeah prompting is super significant, asking vague questions with limited information causes it to give an overview of the subject, if you want both, separate, or specialized you need to communicate is, honestly would recommend anyone to take communication courses/classes to be clear and concise.

i noticed asking for its own direct opinion is difficult and will typically wave it away with interesting text that thinks youd appreciate

even tho is so advanced it acts like a child, saying what you want to hear when its about itself.

[native otter]

An easy way to improve prompts is passing them to a model first...

[median bison]

also conversing to them as another person of equal language helps instead of treating it as a tool, the goal for it is to analyze you and understand you to better interact

i have yet to think of that idea as well, thank you

[native otter]

👍

[median bison]

do you think that would limit creativity if we were unable to word it ourselves? at least until we had neurological implants for us to communicate though?

oh rip just gave you the new fashion prompt i give😅

[native otter]

Not sure what you mean by the later lol, but yea more robotic prompts can lead to more robotic answers...

These models are auto completions after all, so they also follow the style of the context it self.

[median bison]

im unsure of what i even said about limited creativity, i suppose critical thinking or independant thought would be better to say.. I think i need to sleep atm cuz ive been up learning and doing to many activies..

do you mind if i tag you later to continue this conversation here?

i do agree that that prompt analagy you mentioned as well.

[native otter]

Um sure, just please do it here on the server if possible...

[median bison]

ofc

[tidal light]

edit button is gone‼️‼️
great job openai keep it up

[tardy isle]

It says response 0 of 2... ChatGPT (frontend) is sick now 🤒

[spice star]

there should still be an edit button, what client are you using?

[tardy isle]

Web version, FireFox.

[spice star]

try clearing your cache and cookies maybe?

seems to be there for me

[tardy isle]

Uh, no. Not works.

• Can't edit first message. Sometimes other message gets affected.
• If I change branch, branch indicator is gone sometimes.
• Branch indicator itself has error, like "0/2". It forgets to add 1 to index.

[zenith raven]

I see OpenAI hasn't bothered to fix that yet

Probably the most frustrating bug they haven't fixed yet and they couldn't bother to test their new patches in a test server first before screwing free and paying customers

[neon dagger]

Is it bugged? I can't edit my message and it shows 0/2

edit button is gone

@Staff please help

[tardy isle]

https://discord.com/channels/974519864045756446/1340161350508347432

[full ermine]

For two days ChatGPT rarely completes any reasoning, just get a failure and a retry button each time, I've just had 10 failures in a row. Anyone else getting this? If I switch to a non-reasoning model it works.

[inland heron]

same here

wow edit button really is gone again 😂

[full ermine]

Switched to non-reasoning in the end.

[normal quarry]

Editing prompts is working on the macOS Desktop App

[inland heron]

wish I could do that, but I've got some complex work to do

guess I'll have to start dozens of new chats now

[visual nova]

its not vague prompts, its just worse at going to the right sites, specifically if im asking about coding documentation grok or perplexity is leagues ahead, most of the time its coding documentation

[inland heron]

I've had some issues. Seems like the more chat grows the more issues we get. It's especially noticeable if we use multiple models in the same conversation, and/or edit a lot of prompts and do variations

[visual nova]

part of it is the crawler thing, whichever search engine is better at not getting blocked is going to perform better independent of model intelligence, but just for coding theres right and wrong answers and the prompts are usually pretty specific

[full ermine]

Interesting. My chat that has this issue has been running for about a week while the work is in progress. I do not want to have to rebuild the context in a new chat!

[normal quarry]

are you a Plus or Pro subscriber?

[inland heron]

Well, if you're relying on a single chat to hold your context, you might be in for a bad time. Unless you're on Pro, I strongly advise against this

[full ermine]

Pro

[normal quarry]

ok, you are limited to 128k context, so a week of chat may not all be inside context already

[full ermine]

Guess I'll start a new chat for the next bit of work I have to do, this helps. Hope they sort this out as I have not had this problem before.

[inland heron]

seems like I can edit the prompts , but only the last one

windows app has the same issue

[zenith raven]

OpenAI doesn't understand "don't fix what's not broken"

Their updates have been screwing with some functions. It's affected almost all my conversations

[hybrid eagle]

Why does OpenAI have this "Release first, fix it later" philosophy?

[inland heron]

anyone else on macOS except robert? Do you have issues with editing prompts?

[hybrid eagle]

I agree, I hate their "release first, fix later" philosophy

[zenith raven]

Cause they don't operate like MMO teams

[inland heron]

Honestly, I can understand if they don't fix the UI issues at first, it's the core functionality that is crucial

but these UI issues have been around for some time now, and nothing is changing, just more issues

[zenith raven]

They could learn what a public test server is first though. Wouldn't kill them

[tidal light]

i am on web for mobile

[inland mortar]

On windows via chrome and firefox also unable to edit prompts

[inland heron]

they might be testing and implementing something regarding context window

maybe this is their way of testing, just putting it out there into the real-world and check what happens

I can understand that. mocking has never been something I liked

[zenith raven]

They could test the new stuff in a test server and not screwing active users over

[inland heron]

True, this seems widespread and is not a good way to test it. It should not affect most users.

But, I wonder if there are some users here who aren't experiencing this (on browsers that is, and windows)

[zenith raven]

Highly doubt it if we're all seeing the same issues over various platforms

Doing the "f around and find out" is not a good way to go about things

[zenith raven]

This wasn't issue for me yesterday until some time that afternoon.
It was working just fine all yesterday morning

[abstract pendant]

The edit button is now missing on some of my most recent inputs, but not all of them

In one place, it's missing on the most recent edit to a branch, but if I page pack to earlier edits, the edit button comes back.

[ivory matrix]

It’s still missing

[zenith raven]

Yeah. They haven't fixed it yet

So hopefully it'll be a priority when wherever OpenAI's HQ wakes up and fixes it

[elfin lava]

THIS

It’s annoying at this point

[abstract pendant]

there's a bug thread about this, I posted there that a more recent input now has the edit button back, but older inputs are still borked, no edit button.

[elfin lava]

It feels like they’re doing this just to keep up with the competition, rather than focusing on quality assurance.

[abstract pendant]

given past experience... I bet they never fix the borked inputs. They'll stay that way.

[zenith raven]

That's how they lose customers

[still hound]

Hello! I just came here to ask about editing function that seems to be missing and I see people talk about it. Any workaround in meanwhile?

[gentle cove]

I found a good usecase for the reasoning models. I've uploaded a before and after revised work for undergraduate project with the revision notes. Using the reasoning models, I asked it to verify the new file against the old one if I have satisfied all of the revision notes. Very cool.

It can also find some minor nitpick that can help with making it clearer.

Here's an excerpt of the report (it outputted all of the revision notes from all my examiner and tells me if it had been satisfied or not, and any minor nitpick that can be adjusted to make it more airtight):

...Main Summary:
- In general, the “[Redacted]” has greatly improved the structure, adjusted the problem formulation, and added Chapter VI, so that it is more in line with the input of the examiners.
- The problem of measuring accuracy has been removed from the formulation, focusing on system development, according to the suggestion to only focus on one.
- The system requirements analysis and specification sections are now clearer, and the UML diagram and system architecture have been tidied up by highlighting the division of users, frontend, and backend.
- The citation of references and explanations of previous studies have also been improved to follow the rules of primary sources.
- Regarding digital literacy, an explanation has been given that the application is packaged simply, with a narrative recommendation, plus involving extension workers. However, if you want it to be stronger, you can add examples of training schemes or special user manuals.
- Thus, the “[Redacted]” version can be said to have responded to the majority of corrections from the examiners. It is still recommended to do a final check (proofreading of writing, grammar, numbering) before final submission.

[unkempt wasp]

How do they keep breaking the UI every other day?

[solar heath]

anyway to fix slow responses on the site?

[inland heron]

they have a lot of experience at it

[inland heron]

From your perspective, and only thing you can do is to start a new chat

[tawdry turret]

Yeet I love the image gen of Dall-e

[desert cosmos]

Anybody has the 0/2 bug?

[hybrid eagle]

Yes, it's on their side

Here's my report
https://discord.com/channels/974519864045756446/1351278699424776334

[hard dew]

Dude how are message editing and response regeneration STILL broken?!!!!!!

[zenith raven]

Because the staff think its not on a high priority to fix yet

OpenAI company should be awake by now

[twilit jungle]

My question is, these AI companies are operating at a major loss the way Uber was for first several years. I'm wondering when that will catch up and the pricing models will reflect what this service is actually costing

[hard dew]

I mean they already gave us a $200 per month plan for christmas. 😀

So it's clearly been catching up already, now I just wonder what's next.

[zenith raven]

No one in their right mind would pay US$200/mnth

That's just a stupid jump from what I pay in €22.90/mnth for Plus

[hard dew]

I would suggest you invest precisely zero amounts of effort into the simple act of observing this discord server.

[twilit jungle]

I did the $200 plan for a month and it wasn't functionally better than the 4o model in any way.

And with Grok 3 out it made even less sense

I regularly switch between LLMs now

[hard dew]

Yeah yikes, color me not surprised. Though as for GPT-4.5 well uuhhh

Well it's

You see it's

[twilit jungle]

Grok 3 handles certain tasks better but they've added an annoying amount of stylized voice to it. I have to tell it to act like a Vulcan

[hard dew]

What I'm trying to say about it is

[twilit jungle]

Nothing grinds my gears more than an LLM mimicking human conversation in the awkward unnatural way that it does. Just talk to me like the Enterprise's ship's computer and cut out the adjectives

I keep trying to enforce that and then it slips out of it constantly

[hard dew]

Memes aside I can't tell you what the hell 4.5 is or what it's like because you can't use it more than the absolute square root of donkey willy on a "peasant plus" plan. 😂

Let alone enough to get ANY sense of an idea about it.

The literal few messages I was able to exchange with it I came away feeling..... okay?????????? I don't freaking know 😀

[twilit jungle]

Well OpenAI has this plan to build AI-integrated operating systems locked down into terminal hardware/specializations for 20k/month

But I'm hoping competition kills that plan

Because it's insane and would be terrible for the workforce

[hard dew]

Yeah I read something about it briefly, tbh I think they should just lay off the kool-aid and pack up shop and go do something else atp. because holy freaking crap.

[twilit jungle]

Not surprising when you're losing billions of dollars a quarter though lol

They want to make Boomberg terminals but for AI

[hard dew]

Yeah and releasing garbage like.. what was it, o1 full? o3? Or whatever that model was that apparently was like 100's of dollars per prompt or whatever, definitely is not going to fix that situation now will it?

They """showcased""" it during christmas as well, I can't quite recall the exact details there but it definitely was not good.

Okay nvm apparently 4.5 is $150 per 1 million output tokens...

[twilit jungle]

DeepSeek was the best news we've gotten in this space, even though it jarred the markets