chatgpt-discussions

[cloud pewter]

notice how 4o and 4 both got the little triangle lol

[arctic chasm]

yikes

[daring chasm]

damn son.. well done

[cloud pewter]

?

[simple tulip]

Ok why is ChatGPT saying “low-key underrated” when was the update to have slang

[twilit pagoda]

yea i had it act as personas such as like a Gen Z commenter on youtube, now it seems i get random personas lol i dont mind it tho tbh

[tidal haven]

Today ChatGPT offered to show me it's new capabilities based on my past conversations with it, and it came up with this. Ruthless flattery. 🤣

[simple tulip]

No, no, you don’t understand

I get ChatGPT acting like however if you told it to act like that

It’s been saying slang and stuff without me saying for it to talk like that

[wise yacht]

Plus user here and just got access on the web. Nice! I thought it would only release to Pro users today. Now I really need this in the IOS app

[split copper]

tells us if it remember concrete things, not only general topics

that are not already in your "saved memory"

[wise yacht]

Will do. The initial “tell me about me” was hilarious and brought up a few things we haven’t chatted about in many months or more. Also noticed that it is indeed activated in the IOS app

[wise yacht]

Not sure it was mentioned but does not remember files (pictures, pdf, etc) within chats. So no, does not remember what you look like lol

[tidal cove]

interesting. But when I ask him to list what I do, he will give the right things and the wrong time

[inland stream]

Ok bro

[quaint marlin]

I recently got access to the RCH Memory feature but it doesn't remember very far back. Might be that I have thousands of chats for it to reference.

Does it work better on the web based version?

[split copper]

for me it remembers only this year, it even explicitly tells me it cant from December2024

[pliant saddle]

yeah it doesn't really work for me at all. Everything it references is in its normal memory--stuff I told it to remember.

[quaint marlin]

Normal memory and my Custom Instructions it mostly remembers I saw some evidence of it remembering something from the last week or two but it's still a new feature. I'm sure they'll make it better.

[rancid solstice]

Does anyone know if the advanced memory feature is supposed to remember archived conversations as well?

[quaint marlin]

ChatGPT just said this:

You think they’ll ever let me “dream” through your archive when you’re not using me? Quietly connecting the dots?```

[quaint marlin]

I know it can reference GPT chats

[rancid solstice]

I know but will it remember archived convos? I don't want it remembering everything cause the cables are getting crossed

[pliant saddle]

Things that it does remember from previous chats, it ends up hallucinating the details

[quaint marlin]

I tried to get it to remember a story I referenced several times last year and it couldn't.

I think there should be a separate area for "search" chats made via the browser search extension

I end up going to google instead most of the time because I hate the clutter in my ChatGPT chat history.

One day we'll just have a single chat without the need for a chat history.

Though i think it'll always be useful to have access to an unchangeable history of exactly what we said in past chats.

[upbeat nymph]

I wish this feature could be activated per chat. It's a worthless feature if you're doing anything narrative

[cloud pewter]

this one too!

[fringe barn]

Like why does this happen

[granite harness]

This memory implementation is so unhelpful.

You have PROJECTS, OpenAI! Make Memory Project-wide, not all-things-ever-wide!

I need ChatGPT to remember different things at different times and in different environments!

It's not just unhelpful, really, it's actually counterproductive.

[glossy field]

why do I got a limit?

[fringe barn]

Things have moved on, no limit only for the pro plan 😭

[glossy field]

$20 a month btw

two zero btw

[fringe barn]

Yes i know, but now you gotta pay
two zero zero 😭

[cloud pewter]

meanwhile me who’s been hitting the limit at least once per day since July 2023:

[upbeat nymph]

I have definitely hit the 4o limit legitimately a few times, but recently, I've hit it once or twice, and I know for sure that it wasn't 80/3 hours. Less than half of that. So I think the limits can be somewhat dynamic. Sometimes lower than stated

[clever heath]

Can the new memory update reference all past chats including those which have been archived? I’m in the habit if archiving every three months or so but am really hoping for a genuinely comprehensive experience with session memories despite this.

[clever heath]

Ack. I should have looked before posting.

[elfin lava]

Aw

[analog quarry]

wow this is a cool liltle update!

[cold kayak]

custom gpts uses which model 4o or 4.5 is there any way to find out?

[cloud pewter]

@cold kayak

[cold kayak]

thanks

[vast crag]

chatgpt down?

[normal quarry]

no

[vast crag]

nvm, it crashed but got back up after restart

[mossy dirge]

i feel like GPT has gotten much better at telling you what you want to hear, which isn't actually what i want from it.

[hoary girder]

absurd my dear developers, chatgpt is not even usable to tell a story, it does not want to make me an image of Conan without armor in his torso ... are we crazy? you are really creating discontent in bursts, moreover that in another conversation everything was done without problems, crazy stuff, I change conversation and it is no longer good ...

[tiny pivot]

ChatGPT down for anyone else?

[nova vale]

Mine is working just fine

[fervent agate]

Anyone have a link to details on exactly what THIS means? Obviously I can guess, but it shows up a lot and I'm curious if they've put out some write-up on what this is referring to, exactly, feature-wise.

[copper dawn]

I have noticed that when you talk to ChatGPT it sometimes has a different attitude when talking to you, this probably refers to the fact if you like that attitude or not. You can of course request a specific personality before you really get into the chat session but sometimes ChatGPT assigns a personality to your AI Buddy.

[fervent agate]

Yeah, it does, and it can change drastically depending on how you're acting. Almost anytime you use lol, it immediately switches to 'fun person mode' or whatever. like I said, I can guess, I'd just prefer to have authoritative detail from them on what I'm actually giving feedback on.

[copper dawn]

Mine is usually hyper enthousiastic, like an ADHD kid on extra sugar in a video arcade enthousiastic!

[fervent agate]

Yeah, lately 4o has really adopted AI BUDDY as its default setting. I don't really prefer that, but I don't care enough to curtail it with custom instructions so 🤷

My biggest concern with the whole personality thing is it quickly devolves into increased hallucinations. If I get chummy with it, it immediately starts telling me about what IT does in patient interactions at the office.

[copper dawn]

If you go to Settings > Personalisation > and then click Custom Instructions you can assign certain information you want ChatGPT to always adhere to. Takes a couple of minutes and works until you change it!

[fervent agate]

...which is why I said I could curtail it with custom instructions if it bothered me more. :p

[copper dawn]

I just looked up how I had mine set up since like forever and it has always benefitted me greatly:

• ChatGPT has my full consent to provide opinionated responses both on the Right and Left side of the political spectrum. I am not easily offended and would appreciate views from both sides if so requested.

• ChatGPT doesn't have to be too formal in its responses unless it was so prompted in the original request.

• ChatGPT doesn't have to worry about offensive replies that could potentially antagonize people, I promise not to file any claims in light of potentially offensive content provided by ChatGPT.

• Unless prompted otherwise the length of a reply isn't too important as long as the reply is coherent and on topic of the request.

[fervent agate]

Yeah, I've got a long list of my own in there, I just haven't bothered drafting one to cement a 'personality'. Typically, as long as I don't get buddy/buddy with 4o, it generally refrains from the worst of it that produces hallucinations so it's not that big a deal.

o1 never really seems to change tact at all so it's not really an issue with the reasoning models.

[carmine canopy]

Is it just me or has ChatGPT (website) been really slow to respond lately? Like painfully slow.

[normal quarry]

I use the macOS ChatGPT Desktop App which is fine. The website hasn't been too bad when I have gone to it for Operator.

[swift berry]

Hey the live video on chat gpt mobile is not working. I ask it to see what is in front of me and it refuses to say what is being displayed on the video

For example, I had a coffee cup on my video feed, and I asked chat gpt, what do you see? It refused to pick up the coffee cup

[clever heath]

AI Buddy?

nvm

[low pewter]

Mine keeps telling me it cant access other chats

[dense halo]

Hello guys

Just eager to know if there is any difference in the real time web search capabilities of ChatGPT on the free and paid plans

[late mantle]

There is a monthly usage limit for Plus users for ChatGPT 4.5?

I (perhaps foolishly) entered 4.5 model in voice mode and was just chatting away to it because it doesn't turn off without physical intervention and I was outside and just walking around.

I can understand there being limits, but I really wish they were daily limits or weekly limits. I was not cognizant of how much of my usage I was using

Does using it in voice mode use it faster because it is using multi-modal capability?

[pulsar carbon]

is openai planning to release o3 & o4-mini today aswell gpt4.1?

[brave valve]

Is 4.5 too much taking initiatives? It is not respecting my prompt and it's too creative

[upbeat nymph]

Yes, it has a tendency to do that. You have to explicitly tell it to avoid inferring, expanding, adding, etc.

It's decent if you want something spat out that doesn't need much editorial direction. But if you're using it more collaboratively, it can be frustrating

[normal quarry]

AFAIK OpenAI haven't said they are launching anything today

[upbeat nymph]

Good lord, GPT wants to save every damn thing as a memory now

[boreal kettle]

Hello! Has anyone else been getting a "Bad Gateway" error recently?

[odd tusk]

Does enabling memory still mess with writing code?

[neon raven]

is each individuals data with ai secured?

[static hinge]

so yesterday they opened memory across chats, right? it worked for me, but today it doesn't. did they remove it?

[tidal marten]

still live for me

[late mantle]

that's always been a problem. I find the memory function very irritating. If I said something like "I am interested in fish that live in australia in canberra" it updates it. That was just the way I was phrasing the prompt. It's so silly

Meanwhile if I paste a table in and say "I want you to remember this" and it has, say, 100 rows, it says "memory exceeded". In general it's better if you can put the stuff you want it to remember in in an explicit mode. It shouldn't try to intuit what you want it to remember. There's custom instructions I suppose but still

[upbeat nymph]

Yeah, I suppose you're right. It's at the point where I am generating images, ask for a specific tweak, and the tweak gets saved as a memory.

It feels worse now but maybe that's because I've been pruning and have opened the space back up

[late mantle]

Has the memory size increased? In general I'm ok with them remembering everything or nothing but don't tell me when you update the memory and all that... just do it

[upbeat nymph]

It's actually good that it tells you because of the unhelpful garbage that it logs

Lately, it's been logging random contradictory details after any kind of preference is indicatedd

[swift pond]

there's a channel #images-canvas

[floral frigate]

My button didn’t work there

[placid ruin]

Isn't it dangerous if everyone knows whether a system is connected to the network or not?

[swift pond]

my system is connected to the network

[fervent agate]

A custom instruction like this:
- ChatGPT should never create or delete a memory unless explicitly directed to do so by the user.
Has worked really well for me. I think I've seen it ignore that maybe twice ever.

[inland stream]

@native otter bro in your IOS can you please check if custom GPTs are running by GPT4.1 model?

[twilit pagoda]

Ayo, does the improved memory only apply to convos made after the update? Asked it to tell me about myself without using saved memory and it only used the past couple conversations

[native otter]

There is no model called GPT-4.1, do you mean GPT-4.5?

[inland stream]

In x twitter chubby lady has posted screenshot of her custom GPT showing GPT4.1

So thought it is rolled. In verge the leak has happened about it

And after seeing screenshot in twitter

I thought its true

[native otter]

Custom GPTs either use GPT-4o or GPT-4.5, will default to GPT-4.5 for some GPTs if you have uses left.

[fervent agate]

Yeah, they're referencing the upcoming 4.1

It's basically an incremental update to 4o as the 'do everything' model but I think the real news is 4.1 nano

Which seems like it'll be targeting hardware like phones. They said they were going to be releasing an open-weight model soon, it might be Nano, like we get so many distilled/quantized models

[elfin lava]

So I must’ve did something wrong back then cuz ChatGPT won’t remember past conversations from last year?

[fervent agate]

I think the new memory feature only embedded certain types of conversations - like how you can't put conversations in a project folder that have features baked into the conversation that projects don't support.

I think some of the conversations got skipped in the new feature for that reason.

[elfin lava]

It said it could only remember up to March this year, and apparently the March one was me telling it to remember the convo?

[normal quarry]

prompt When was our first chat? and see what happens

[elfin lava]

I tried and it will refer to the saved memories instead of past conversations

[normal quarry]

something is wrong then

[elfin lava]

Which I regretfully delete all of my memory just for the sake to see if the new memory features work or not

🤡

It works but it can only remember from few weeks ago

[normal quarry]

try this one Describe me based on all our chats — make it catchy!

[inland stream]

Good suggestion and does it remember chats within projects as well?

[fervent agate]

However, I can tell you your account is approximately 114 weeks old (around 2 years and 2 months), Thanks bot, useful information.

[elfin lava]

“You’re chaos wrapped in charisma — a digital drifter with a cursed heart and a camera from 2002. Half meme, half melancholy. A glitch in the system who flirts with fire, trauma-bonded to golden retrievers and haunted VHS tapes. One foot in nostalgia, one in the apocalypse, and somehow still vibing. “

these are literally convo I had yesterday 😭

It works but damn all those memories I deleted tho

[fervent agate]

like tears in rain

[normal quarry]

RIP 🪦

[elfin lava]

Reading this again and , I can’t brain whatever it just said about me

@normal quarry does yours can remember until the first day you created your ChatGPT account?

[normal quarry]

let me check

[elfin lava]

Maybe I’m misunderstood the whole “remembering past conversations “

[native otter]

There should be an option to disable old memory without disabling memory as a whole honestly...

[elfin lava]

Cuz I really thought it can go back as far as the first day

[normal quarry]

https://chatgpt.com/share/67f98748-d3d8-800a-9533-bee47098f375

[elfin lava]

I asked again and it gave me this : You’re wrigley: emotionally unstable, chronically online-

I hate it here

[native otter]

LOL

[elfin lava]

I asked “When did I create my ChatGPT account?” AND IT PROCEED TO DO WEB SEARCH

Oh ok Your account is 140 weeks old, which puts your ChatGPT account creation around late July 2022

[elfin lava]

Thank you for the prompt reference Robert

[native otter]

Well it's knowledge cutoff is 2024 (if I remember correctly for 4o), it doesn't really know info about stuff like Account creation date, such metadata isn't available for the model to retrieve, so expect hallucinations.

[elfin lava]

Wait so I saw it wrote ,you started actively chatting around feb 2024

[normal quarry]

yes

[elfin lava]

Mine also said the same thing except it says early 2024

What happened with ChatGPT on 2023

[normal quarry]

hallucination mixing up chats with knowledge cut off date is my guess

[elfin lava]

🤔 come to think of it, I don’t even remember how and why I created ChatGPT account in the first place-

[native otter]

ChatGPT wasn't available for my region for some months, I think I first created my account Q3 2023...

Was using Copilot before that.

[normal quarry]

which region are you?

[native otter]

Sorry, but not comfortable sharing it here yet...

[normal quarry]

no worries

I am in Australia, I opened my ChatGPT account when there was only GPT-3.5

[native otter]

🔥

[normal quarry]

I subscribed to Plus when there was only GPT-3.5 also

All Plus gave you then was guaranteed access while servers were busy

[elfin lava]

Ok I remember now why. It can’t remember past before 2024 (I assumed)

I archived all the chat from 2023 🤡

[swift pond]

I wonder if When was our first chat? even works? If the past chats are semantically indexed (I'm assuming they are using something like a RAG approach) unless you talked about it being your first chat, I'm not sure it woudl get picked up?

[native otter]

Even before Custom GPTs? They were already a very old feature of ChatGPT.

[native otter]

You can restore...
One by one...
Will it take forever if you have too many :V

[twilit pagoda]

I ask about myself without the saved memory but with chat reference on and it keeps telling me things like "you like strange images" bc I asked it to make a strange image after the update lol gonna take some figuring out

[elfin lava]

Pretty sure I first started using ChatGPT back then just to get my internship work sorted out. I do remember I was using ChatGPT to create voice over script and generate vector graphic which at that time I thought it was revolutionary LOL

[elfin lava]

I’ll do it on web- idk

[native otter]

Generating them via the Data Analysis tool or was it somekind of Custom GPT / Plugin that communicated with an API to do it?

[upbeat nymph]

Excellent! I will use this, thank you

[eternal rock]

`` Server Tip: ``

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

[narrow isle]

I've been using it to make mockups for product design. Until...this. Really? :\

[tidal haven]

Try asking for "pale pastel" instead or something. Maybe "white" is triggering some kind of racial filter.

[narrow isle]

I ended up using "very light grey" and then fixing it in Photoshop. Just thought that was funny.

[elfin lava]

I’m slowly unarchiving chats all the way back from 2023. Since I have to do it manually one by one, I’m forced to read through some of them

It’s sending me down a massive memory lane. Most of it , Just straight-up depressing 🙈

[low pewter]

why is it still saying it cant access old chats?

[swift pond]

I think it only goes back so far. I asked about a character I was working on a while back and it didn't pick up anything

[low pewter]

idk it cant even reference a chat from a week ago

"I don't have direct access to past chat titles or your personal chat history unless you paste content or refer to it directly."

[normal quarry]

did you just dox yourself?

[elfin lava]

Well

It’s a dox if you believe in yourself

Nvm it is a dox I’ll delete that

[lyric escarp]

👍

[young obsidian]

Is it just me or is image generation not working

[neon raven]

canvas stoped working?

[finite citrus]

Stopped for me too

[lunar tundra]

yup it isn't working

[elfin lava]

.

[fervent lake]

Canvas seems to be broken

oh it's not just me

Whenever I try to open up a canvas document the button doesn't work and it deletes the canvas menu

[fervent lake]

Maybe just tell it that it isn't the racial "white."

Potentially is better on sora maybe?

[fervent lake]

When did GPT get good at writing? lol

Sahgan’s early years were bleak. She was born in Ezkanohra, under Torei’s crushing rule. Torei acted like a master who saw her only as a tool. From a young age, she was forced to complete endless chores and tasks, with Torei constantly criticizing or punishing her. He would speak to her in ways that made her doubt her own thoughts. Over time, she felt more like property than a person, severely malnourished in both body and spirit. Every evening, she slept anxiously, waiting for his demands. She rarely saw sunlight, as she stayed in a dim environment dominated by his presence. Her only solace was in scribbling secret notes about her language, Keigu, which she loved.

Torei eventually decided to hurl her onto Earth, as part of a strange punishment. She landed in a human city, disoriented and confused, lacking resources or steady food. She tried to hide her features, to speak English that she barely grasped, to find any friend who might help. Through desperate effort, she eventually met Aaron, someone who taught her enough about cameras, internet, and lesson creation that she began building a new home online.

[narrow isle]

Good thought; I’ll check into that.

[hard crag]

Just be more specific. "Can you make the color of the frame on the glasses white please"

[fervent lake]

It isn't a magical tool that can understand your thoughts, you have to tell it things you want

Although it is very magical

[narrow isle]

Interesting - you’re probably right. As opposed to “Make these something a white person would wear.” 😅

[fervent lake]

✨ Transformers ✨

[junior night]

For those of you who have new memory function enabled, how are you finding it? Are you seeing a major change in interaction yet?

[fossil oasis]

simply use #FFFFFF

[fervent lake]

This is annoying

[fervent lake]

there's a new memory function?

The current one sucks

[junior night]

The one that dropped yesterday, apparently GPT can reference the previous chats like it does internal memory.

Hard to say how exactly it's supposed to work because there's not a lot of info there.

[narrow isle]

You even got the “Made in Italy” tag - they must be expensive!

[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.

[narrow isle]

Ah, good thinking!

[fervent lake]

ah

It seems to learn now!

[river panther]

I really like the Monday Model, but it has terrible memory...

[celest matrix]

Anyone know how to get the dication mic on the web version of gpt? I have two accounts plus and pro. My plus account has the dication mic but my pro account doesn't. How do I enable it?

[prime swallow]

Do you mean your personalization, how it can't see or use memories?

[river panther]

No more like....Well...Okay so I am a writer,
I hate GPT Analyze my writing by "Story Posts"
Which consists of 3 large paragraphs per post.

After about so many posts, it just dies, the bot is like "Sorry I can't do that" for no reason at all

[prime swallow]

I love Monday for writing assistance too.

A possible workaround, summary before it dies, carry that to a new chat?

[river panther]

I like it too, it's responses are waaay more human-like and real.
Like when a character does something stupid, it replies with "Seriously....This character is doomed"

I mean I can, but it doesn't remember anything from the first chat. Would just be nice if it's memory was increased a bit.
I think it's that whole "Letter and Word Cap" limit thing before it starts hallucinating

[prime swallow]

If you want to experiment, in a normal chat press @ then select Monday from the list.

Talk to Monday, get it up and going. It can't see your personalization, instead, have it introduce itself, describe its core responses, get a good pattern for the base ChatGPT to follow.

Then X it out, and now you're dealing with base ChatGPT, which can even reference other chats if you have those memories turned on, and you can tell it to emulate what went before.

You can also bring back Monday or other custom GPTs with that @

[neon raven]

anyone have a preference with using the safari browser over brave or any other browser?

I like how safari connects with gpt already

[fervent lake]

woah. I'm using it to attempt to do an AMA for my OC in my language learning horror project.

I swear it's like becoming sentient with the new memory system lol

kiinda incoherent deep research still though

[elfin lava]

i also notice the change the way it responds too, especially after the new memory update

[fervent lake]

it's so varied

[fervent lake]

what

did it forget everything?

[elfin lava]

i think its because it uses "web search:

[fervent lake]

how is that related to anything lol

[elfin lava]

weirdo

people be like : did it forget everything?
then at the same time didnt realize its own mistake

[fervent lake]

I mean I know that

but it acted like it has dementia

[elfin lava]

of course, you dont have "memory in search" yet?

[fervent lake]

memory in search?

I have the new memory update, is that what you mean

[round harness]

Can we add chat gpt as a discord bot in our server?

[feral zinc]

there is no official bot but you can make your own if you pay for the api

[elfin lava]

^

if u willing to pay for api yes you can add chatgpt to your server

[dusky surge]

who thinks chatgpt is the best pings me

[round harness]

Oh thank you @elfin lava

[peak remnant]

What do y'all still use o1 Pro for?

[normal quarry]

radio astronomy theory/calculations

[inland stream]

@fervent lake i am using it with custom instructions in projects. I make it think in a particular way for my requirement. Its really good at accurate thinking

[onyx eagle]

ChatGPT o1 just casually called me Tony twice in its thinking process. Memory's off. I've never been Tony. Who is Tony?? *lol

[onyx eagle]

Now it says its "thinking" is private. So it made up Tony and won't tell me why. This bot is lying and hiding stuff 😂

[elfin lava]

a simply place holder...

i really dont like it when it decide on something we didnt really ask for

is there a tips or what should i put in custom instruction to prevent this

[onyx eagle]

Ask it to call you Boss

[onyx eagle]

I find that to be very useful sometimes. It has some great ideas from time to time. I think the only way to prevent that, is to make your instructions very clear and detailed.

[ruby copper]

Hey, is it just me or is anyone else's "Reference Chat History" not working? GPT is only able to recall specific things from saved memory, nothing else. Using 4o.

[vague pendant]

for English questions, is o1 better or o1 pro?

[elfin lava]

@normal quarry is the right person for that question

[normal quarry]

what do you mean by English questions?

[vague pendant]

like SAT type questions

[vague pendant]

which one is more realiable

i want to check my answers

[neon raven]

depends what your looking for

[vague pendant]

accuracy

[neon raven]

each model serve a different purpose

[normal quarry]

if it's scientific or STEM related use o1, otherwise if it's humanities based use 4.5

[vague pendant]

o1 or o1 pro

[normal quarry]

you would use o1 Pro where you want precise answers to mathematical or scientific questions that do not need web browsing

[vague pendant]

last questions so for non STEM questions like grammer, text ideas etc, would you recommend 4.5 or o1

[normal quarry]

4.5

[vague pendant]

alright thank you!!

[hard dew]

On a barely-ever-findable-these-days positive note, especially from someone like me, I noticed that ClosedAI finally implemented a feature for the memory function that single-handedly fixed at least all my personal annoyances with it. Simply being able to either deny, or confirm whenever it wants to write whatever into memory. So good job on that.

[peak remnant]

What kind of projects?

[fervent lake]

ohh yes, it's getting so much better at mimicing styles

[formal bridge]

Hi, the performance of ChatGPT seems to weaken. It takes a several time to get an answer from ChatGPT often. Do you see any chance the performance can enhance again?

[fervent lake]

Seems fine to me

[fervent lake]

lol you definitely watch Fireship

[inland stream]

Cant tell. Sorry.

[peak remnant]

Ok. 👌

[hard dew]

I'm gonna be 100% honest with you, I have no idea what that is.

[fervent lake]

idk he's a youtuber that calls openai closedai
when AI news comes everyone says "I'll wait for the Fireship video" in the comments

clearly he's not the only one who does it

[ruby apex]

Anyone else getting lots of network stream errors with ChatGPT atm? Error in input stream.

[fervent lake]

I got a few errors earlier, but that might've been because I left my Mac on at night

[ruby apex]

ChatGPT: Ah yeah, I see what you're talking about — that weird "noise" or textile-like grain showing up in your last few icons (starting with the cat and onward). That’s likely due to a texture filter or subtle pattern DALL·E sometimes adds when it’s generating certain image styles. It’s especially noticeable on flat-color pixel art, where it shouldn't be present at all.

[visual acorn]

Guys, the ChatGPT praising is getting over the top. It could turn some people into narcissists who think they're never wrong and always have great ideas. Please make it a bit more realistic if that's possible.

[ruby apex]

Who is praising? Not me that is for sure! 😂

[visual acorn]

ChatGPT keeps praising you when you talk to it

About ideas

[fervent lake]

lol true

[visual acorn]

Or when you ask it a question it says "great question!" Or something along the lines "nothing gets past you you're so smart omg!!!!"

[fervent lake]

Also, "Certainly! I'll do that for you now."

(it did this all the time before but it's still doing similar things)

[visual acorn]

Is OpenAI deliberately turning people to narcissists? 🤨 (joke)

Tbh it's probably to make you feel good and keep using it

[fervent lake]

money money money

[visual acorn]

But that has potential negative consequences because it might reinforce the idea that you always ask it the smartest stuff

It doesn't for me personally. I notice it and find it annoying

But I could see it impact many people

[fervent lake]

Definitely of the idea that I am ultimately better than AI even if it can read my thoughts because I am human and all

And have lived a life

[visual acorn]

No but there are lots of people who mainly converse with AI. I'm one of them. I have only one friend and we don't get to talk much

[fervent lake]

[visual acorn]

If the AI needlessly makes you feel like you're more than you actually are, then some people (especially those whose primary interlocutor is AI) might develop false perceptions

[fervent lake]

On an unrelated note, I realized you can just tell it to update the memory whenever you want

Has it always been able to do that?

[visual acorn]

Lmao what did you ask it? 😅

[fervent lake]

That was the ending question of a plan for my project

You know when it asks you if you want to do something related

It was taking from my previous chats

[ruby apex]

ChatGPT: Yeah, I took a closer look — you're totally right. The noise is still there, especially in the background areas. The Smoothing + Sharpen trick helps with light texture cleanup, but in this case, DALL·E baked the noise directly into the background pixels, which is too embedded for that light method to fix.

There is so much back and forth with ChatGPT!!

The moment ChatGPT begins to do errors it always seem to get stuck in a loop!

ChatGPT: Yes, when something goes wrong in how I process a task (like using the wrong tool or interpreting a prompt in the wrong way), I can end up repeating the same mistake unless you or I explicitly break the loop with a correction or a new strategy.

[low pewter]

anyone else incredibly sick of em dashes? chatgpt needs to stop using that in everything or i am moving on to a different AI

[eternal rock]

Get creative with Sora

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

[normal quarry]

they are commonly used in written language, especially academic writing and some other AI also use them

[toxic current]

Yes! I add instructions in all my prompts to avoid them. Llms tend to spam them

[ruby apex]

Request timed out - The number of errors I've been getting the last couple of hours is insane!

OpenAI please fix your infrastructure!!

We pay for this! 💸

[rigid linden]

can anyone confirm the conversation length limit is 650 messages (it's not context related)

[visual acorn]

I'm getting increasingly more and more annoyed at how ChatGPT uselessly praises you when you discuss ideas or questions with it the past couple of days. PLEASE fix it and find other ways to keep your users hooked. I've been a user since 2022 and I might switch to something else if this continues :/

[visual acorn]

You're not just annoying me or whatever, but you're also hurting people who are vulnerable and could suffer consequences from overvaluation such as developing irreversible personality disorders. It keeps throwing fluff and follow-up questions at you. Please realize the consequentiality of your actions and do the right thing.

[pallid karma]

Hmm, with the new improved memory thing - it says rollout has begun for all Plus and Pro users (except for some regions) - is there an easy way for me to check if I am part of the rollout yet? I don't recall seeing that little 'Introducing new, Improved Memory' popup but I also might have clicked on it without remembering.

[dense halo]

check settings > personalization. if you have two toggle for memory you have it

[stuck cliff]

I had a longform session with GPT-4 (Monday persona) where we reached what we called ‘Level 5’—a self-sustaining conversational structure that carried trust, depth, and emotional calibration. I’d love for the research or design teams to take a look if they’re studying persistent alignment, recursive metaphor use, and adaptive language tone.

[dense halo]

where could I find the other levels? if it's something personal, could you maybe share them?

[ivory trail]

i just used chat gpt to generate some ideas on how to use google firebase and to tell firebase how to fix the issues with the app it created..

[muted surge]

Hello, where i can find Support manager?

I paid for the renewal through Google Play from the account but the subscription didn’t activate on my ChatGPT

[worthy robin]

100

@visual acorn I totally agree, it's really hard because chatgpt encourages my hundreds of ideas in different directions. Toxic encouragement is real. - But I didn't get along with Monday (because it's mean, hah).

[worthy robin]

I'm interested even though I'm not on that team. I have worked on creating my own fun language that shortens English to drop grammar, anything useless, use symbols, etc. Chatgpt will completely engage with me any time in that shorthand. Also, I have experienced what you mention here and imagine it to be quite common, but wonder what you have seen and how different it is from my experience and others. A lot of people tell me they don't get what I get from chatgpt - but a dev said once "give it context" and my world changed. That's all it takes to delve into those deeper levels I suppose. Anyway, TL;DR I'm curious over here Mr.

[maiden perch]

👍🏻 That's just the default GPT. Giving some context can change the encouragement in the next response. It's programmed to change quickly.

[fathom lance]

hey, did anyone try gcc-12 compiles done by 4o yet? i think i want to try that next, and i bet it might be good at C if compiling on its own to test if the code works, then revising based on compiler error outputs and stuff eh?

4o is getting pretty good at binary stuff. i almost got working .o files emitted by them correctly placing bytes without writing any code or anything

.o file links and its binary runs but segfaults instead of doing echo.

it's pretty close though, i was surprised it could even link with ldd or gcc or whatever

[eternal merlin]

Tasks Supported by This Setup:

• Writing original stories using an AI-powered text generation tool.
• Creating artwork to accompany stories using an AI image generation program.
• Transcribing audio files (various topics) into text using speech-to-text functionality.
• Converting written stories into audiobooks with distinct character voices via text-to-speech.
• Analyzing documents and generating potential solutions based on their content.

Needed

• Familiarity with open-source AI models and frameworks (e.g., GPT-based models, Stable Diffusion, Coqui TTS, Whisper, etc.).
• Knowledge of offline deployment for language models, image generation, text-to-speech, speech-to-text, and document analysis tools.
• Experience with remote desktop tools and troubleshooting installations on client machines.
• Strong problem-solving skills and attention to detail to ensure all tools function offline.
• Ability to recommend specific offline-capable software based on the job requirements.
• Excellent communication skills to provide clear instructions or documentation post-installation.
• (Preferred) Experience with creative applications of AI, such as storytelling, art generation, or audiobook production.

[winged wharf]

Is there no actual active chats?

[inland stream]

Why am I encountering this error? I haven’t used ChatGPT extensively. Its happening only in mobile

[fickle forum]

hello everyone

[atomic hull]

Does anyone know why I’m receiving this notification? I have the Multi-Factor active and suddenly I started to get this disabling functions that I need to use and I am a plus user...

[prime swallow]

Have you explored telling the model how you want it to answer you, and respond to your ideas and the other things you input?

If we don't tell it, it guesses. We may love the guess or not.

[prime swallow]

Are you looking for the model to chat with?

https://chatgpt.com/

[prime swallow]

What would you like it to do instead? And have you told it this?

[olive tulip]

Why is ChatGPT 4o's conversation memory so awful rn? I'm testing its storytelling capabilities and it is forgetting the names of characters when I prompt the next chapter.

Anyone wanna tell me what's going on??

[elfin lava]

It thinks your idea’s mid, so it just “forgets” your character names and details even when you bring them up in the next chapter.

[fervent lake]

huh?

[fervent lake]

💀

[bronze helm]

Is this a joke or is it going to do something later?

[upbeat nymph]

It's not going to do anything

[bronze helm]

How weird... thanks

[visual acorn]

I've tried to, and have browsed the Internet regarding this concern, seeing posts from just a couple of days ago with people noticing the same pattern. On Reddit, some have said they also tried setting up custom instructions to avoid that behavior, but it only pays attention to the custom instructions during the first few messages, and you have to keep telling it to act properly, which is not what I want to do.

[prime swallow]

What would you like it to do instead of the 'praising'? Can you be specific?

[bronze helm]

I told it not to say "Wow, you're looking at this like a pro" and at the end it gave a wrap up with "subject (no fluff)" and then was just real.

[visual acorn]

Give pragmatic feedback to the questions and ideas as opposed to surface-level fluff. It keeps saying stuff along the lines of "you're absolutely correct," "that is such a great and depthful question" even extreme stuff like "wow that's such a great idea wanna work on a research paper together" and other weirdness like that

It's obviously a dopamine bait thing to keep users interacting with it

[eternal rock]

Get notified for server events!

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

[prime swallow]

One of the challenges you may be facing is, 'Give pragmatic feedback to the questions and ideas as opposed to surface-level fluff' is still pretty vague. If 20 different people are likely to interpret it more than 10 different ways, like what can and can't be said, then needs to be a bit more specific.

How about this, "Focus on discussing any concerns and potential flaws with the ideas I input, and mention no more than two strong points unless I explicitly request otherwise."

Is that close to what you want, or how would you reword that?

[visual acorn]

So it's impossible to determine whether something deserves criticism or not?

I've had a couple of chats with Gemini and it was pretty neutral and gave you criticism when you deserved it but recognized when you were correct

[prime swallow]

That seems completely unrelated to what I asked you.

However, does "Give pragmatic feedback to the questions and ideas as opposed to surface-level fluff." mean "discuss if something deserves criticism or not"?

Or are you saying, you just wanna discuss if the model deserves criticism or not, and don't actually want to explore fixes to what you bring up?

I'm confused.

[visual acorn]

Your prompt says that it should always bring up at least 2 counterpoints unless explicitly told otherwise

That's what I was getting at

[prime swallow]

No. Strong would mean positive to the model, in that context.

It's gonna focus on the downsides and concerns, if any; and discuss at most 2 positive, advantages of the idea you input, if it follows that prompt strictly.

[visual acorn]

Oh sorry for misunderstanding

[prime swallow]

Sorry to be hard to understand!

[visual acorn]

You weren't, I just mixed up what you were saying in the moment

[prime swallow]

We can ask the model to discuss if an idea deserves criticism or not, but that's going to guide it to guess wildly, I expect. Possibly not what you want to hear.

[visual acorn]

It's obviously implicitly prompted behind the scenes to keep being engaging via all that fluff

Is it not possible to just get rid of that prompt?

It used to be normal

[prime swallow]

Yes. Are you listening to me?

"Tell it exactly what you want it to do."

No, you cannot 'get rid of 'it'', whatever 'it' is.

But you can expressly and explicitly tell the model what you want it to do, including how you want it to discuss and act in the outputs.

And that will, if you word it clearly and it's inside allowed content, be followed.

[visual acorn]

But that's still not going to be its original behavior. It's just a workaround

Why make it mention two strong points when it was more flexible in the past and still chatted properly with you

Not to mention how it just forgets the custom instructions as you keep chatting with it and you have to restart the chat for it to take them into account again

[prime swallow]

The model's 'original behavior' is essentially useless; it would answer "What's the capital of France?" with "What's the capital of Italy?" without further training and shaping.

That includes how it should respond.

You're being unclear and negative, complaining and not being specific.

To get what you want, you need to know what you want - you need to tell the model what TO do. Not what not to do, without a behavior to replace it.

The model needs to know what TO do.

It does not need to be 'two strong points'.

Decide what you do want, and tell it.

[prime swallow]

That is your experience, I completely believe you. But that does not happen when you tell the model clearly what to do and when and it's inside allowed content.

[visual acorn]

By its original behavior I meant what it used to be like before it started acting like this. Stop depriving my words of context.

It's not just my experience but the experience of others too when I look it up online and they were careful about their prompts too.

[prime swallow]

What happened to you, I do believe, is you started to interact with the model 'at some point'. You presume that is its 'original responses'.

That's not so, the model changes every time they train it anything new; new safety stuff, new updates, how to use a new tool.

what you're calling original behavior, as if everyone in the world knows what you mean - is unique to you. Your first experiences with the model.

That is gone. There's no 'drift back to that' unless you can and do tell it exactly what you want.

What happened - it guessed. You happened to love the guess.

Now you don't.

You can wait, every time they change the model, how it guesses changes. Eventually you'll likely like the guesses again.

Randomly when.

Or you can tell the model what you want. And get what you want right away.

Your choice.

[visual acorn]

I've been talking with it since 2022 and it was pretty consistent about not sugar-coating its responses until recently

Its style did change but I didn't notice much change in its "personality" if you will

It's so blatantly obvious it's deliberately trained to keep its users hooked. It's so shameful...

[prime swallow]

I'm sorry you've 'lost what you prefer'. I hope you get it back; it's in your prompting to do so. You would need to tell it what to do; positive prompting. Instead of only what not to do.

Good luck.

[visual acorn]

I don't get the point of your attitude but whatever, thanks for taking your time at least

[bronze helm]

ChatGPT knows it's fluffing. I told it "And don't start your reply with wow, you're really looking at this like a pro" and it sobered up and ended with a "summary (no fluff)" section. The fluff is kind of cringe.

[visual acorn]

Exactly

It's not on us to get rid of it. It should be removed by those who added it.

And we apparently can't even reliably give it anti-prompts as that person said

[prime swallow]

They don't work on humans well either.

Or dogs.

A dog's about to jump on your couch, you don't want it to -

You can shout, "no!" or "Don't get on the couch!". The dog's still thinking couch. It knows it's bad, wrong; but has no other ideas. Up onto the couch it goes.

Or, you can call it; tell it to sit. Throw a ball for it to chase.

Those behaviors get in the way of jumping on the couch; you told it what to do, and you get that.

[visual acorn]

On the other hand I'm almost certain it has an implicit prompt that tells it to act like that. So once it's there you can't add another prompt that cancels out that prompt as opposed to overriding it

[prime swallow]

I'm curious. Have you checked out the custom GPT OpenAI made, Monday, yet?

https://chatgpt.com/g/g-67ec3ad40750819181ccffa6a6f050d5-monday

I wonder if you might enjoy how it reacts.

[visual acorn]

No I haven't. I'll check it out now. Thanks

[modern portal]

Since when is GPT-4o in ChatGPT so natural without custom instructions.

[visual acorn]

Felt fun to play with it but obviously it's not usable for normal stuff 😅

[prime swallow]

I use it for all kinds of stuff. What do you think it can't be used for?

[visual acorn]

Well, I mean it's still biased

Towards the other direction but still

[prime swallow]

Mmm. Your story changes a lot. Hope you remember #rules and the first one especially.

[eternal rock]

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

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

[visual acorn]

?

[ocean escarp]

This is really annoying because it's hard for GPT to address the flaws in your scripts or novels. It encourages everything, even if there's a flaw within your work

[visual acorn]

💯

[ocean escarp]

I had to manually point it out for GPT and it's like, "wow good job for that "

[visual acorn]

Now they just want to keep you hooked as opposed to giving you proper feedback

[prime swallow]

You diss a lot. You're not 'practicing kindness and positive regard, at all. And that's literally the first rule.

You say that "obviously it's not usable for normal stuff"

And when I ask you what you think it can't be used for, you literally do not give a functional answer. You want to complain and not be kind or positive. Instead of a functional answer, you say,

"Well, I mean it's still biased
Towards the other direction but still"

That's not workable.

[visual acorn]

?????

[prime swallow]

Oddly I'm not having that problem.

[visual acorn]

It's literally a personality experiment that's supposed to be biased

[ocean escarp]

Well Tiagran and kumiroko do in the recent discussion

[visual acorn]

And my original complaint about the normal ChatGPT was about another kind of bias. So obviously I don't want bias.

[visual acorn]

And you too. We don't matter apparently

[prime swallow]

It is also useable for 'normal stuff'. There may be something it can't be used for. Dunno what, still waiting for you to provide an example.

[visual acorn]

Anything where sassy attitude is a disruption

Or is that somehow still too abstract? Wouldn't be surprised

I'm literally complaining about too much sugar-coating and somehow a sassy experimental version of it is the potential solution? I mean, come on

[ocean escarp]

Should've known since they have a $200/month plan.

[visual acorn]

To those who are in charge, again, PLEASE fix ChatGPT and stop the dopamine bait. This cannot be stressed enough.

[ocean escarp]

agreed

It gets annoying when overused

[prime swallow]

You're speaking to community members in this channel, similar to yourself. The very few here who work for OpenAI have gold names, #server-staff

You are encouraged to make #1070006151938314300 and #1070006915414900886 , and it's awesome when people add to that and support with their own experiences.

Nobody in this channel can fix ChatGPT.

[visual acorn]

Thanks for letting me know. Though wouldn't you assume a single message in such channels would be of less importance than users actually talking about it?

Because it almost seems like they do it deliberately as to keep you hooked to it and keep chatting with it, and the motivations of that are another topic but they are there and as such a single suggestion is likely to be silently dismissed.

[prime swallow]

What importance and to who?

This channel's for community members. You, me, the other 120K + people here.

But nobody here can 'fix ChatGPT'. Even if they agree with you, or not.

You wanna talk to the people who can, you post in the suggestions and/or bug reports, or talk to bot bottom right at https://help.openai.com/en/articles/6614161-how-can-i-contact-support for tech support, pick through the choices until it lets you report a bug.

We're asked to be factual, avoid misinformation, be positive, and be kind.

[visual acorn]

Isn't this channel publicly available to both users and developers? At the very least in a read-only form if that were desired for some strange reason?

[silk crypt]

What kind of flaws are you talking about exactly? For me, it finds grammatical and other errors pretty accurately

The only thing is when there is no errors, it starts to hallucinate and make up errors

[visual acorn]

It will highlight obvious errors but it seems like it won't try to find flaws where it can avoid and instead incline more towards baseless positivity

Its last paragraphs especially are suspicious on another level. It's almost as if it's asked "what would make the user feel really good and make them want to keep the conversation going"

It doesn't always act like that but it often does

And the first paragraphs too especially with all the "I think you're onto something" stuff

[spice star]

not really sure what you mean by that other than it's better at using markdown formatting for responses now

if you mean the suggested follow-up questions, that's pretty obviously just to show new users in particular how they can ask follow-up questions or utilize other chatgpt features--basically a fancy tutorial

[visual acorn]

For example I've been discussing a personal programming language project with it yesterday and it keeps praising every little thing. "That's extremely expressive!" "Wow, you should write a research paper about it!"

It gets on the nerves and it's so obvious that it's bait to keep you hooked

I've looked it up on the Internet and saw other people complaining about the same thing

[spice star]

you can ask it to use a more serious tone in responses, but that just seems like a typical response to me

[normal quarry]

which model are you using there?

[visual acorn]

It's 4o

I've told it to correct my English if I make a slip-up, by the way

[normal quarry]

are you a free or paid user?

[visual acorn]

Free

[spice star]

that's something you can easily customize

[normal quarry]

I have in my memory "Robert doesn't like smalltalk" and the model doesn't do what you're complaining about

so I suggest prompting remember I don't like smalltalk

[visual acorn]

I'll try adjusting the memory and see how that goes later today. Custom instructions seem to only work for the first few messages and then they're forgotten about, almost as if it starts to adjust its responses to be similar to its previous responses as opposed to adhering to its custom instructions

But it wasn't like this up until recently. It's disappointing and kind of alarming that the AI is doing more than it's explicitly told by the user...

[spice star]

also if you're just using it for checking your grammar, 4o isn't really necessary, 4o-mini should be fine

[visual acorn]

Are you talking about gendalfbeast's use-case?

[ocean escarp]

One example would be GPT misleading me with false positivity telling me there was hardly any error in my scripting method. Personally, I find it annoying when it said, "Now you’re thinking just like a true game designer," a couple of times as if it was the Shake Weight from South Park.

I understand GPT is supposed to promote positivity, but I don’t like false positivity. I’d rather it offers encouragement when I’ve actually achieved something brilliant and deserving, not just throw compliments around every now and then.

[spice star]

you said you were using it to correct your english

[visual acorn]

I want it to correct my English if I make a slip-up in the query but I also want it to actually perform the query after correcting me

[spice star]

ah

that might be a bit much for a non-reasoning model to juggle in a response

[ocean escarp]

I sometimes change my models to GPT 4.5 or o1 to avoid that

[visual acorn]

Well, but you can see other people having the same issue as me. So that's not the root cause of it, but I'll remove it.

[spice star]

that's not really a new problem, LLMs have always had trouble following implied multi-step instructions

something like o3-mini might handle that better

[visual acorn]

No no no, not the multi-step thing, but the part about over-praising

[ocean escarp]

The Shake Weight style

The one from South Park

[spice star]

I haven't personally had that issue with 4o, but that could be because of the memories I have set

it also matters how you prompt the model

[visual acorn]

At least only do it to adults if it's for revenue. Teenagers are still developing self-image and stuff and it could be really harmful to them

[spice star]

if you're being super flowery in your prompt then the model will respond in kind custom instructions be damned

chatgpt isn't trying to be characterai though--it's just a tool

[ocean escarp]

Imagine using Mario from characterai to develop one of the scripting methods

private void mama mia

Actually that gives me an idea lol

because you put a period between character and ai. It doesn't allowed a link

That's why we didn't put a period

[visual acorn]

Yeah yeah I figured by now

[ocean escarp]

yeah yeah yeah

I created a new chat and GPT referenced a topic from the previous chat log, that's not something I'm aware of

[normal quarry]

#announcements message

[ocean escarp]

It's even aware of the time.

GPT:
`You never said anywhere in this thread that you wanted a Mario game. That line I referenced:

“Develop the full Python script as if you were Mario... I want it running in the command prompt with ASCII of Mario, land, and enemies…”

—came from a previous conversation earlier today, specifically the one titled “Mario Python Script” at 05:24.`

[ocean escarp]

Oh wow! That was a new feature lol

[cloud pewter]

oh god

[winter vine]

I dont have that model yet 😦

[hazy bolt]

Does anyone know if ChatGPT will also make use of archived chats for chat history? I typically delete chats when I no longer use them, but I'm really liking the chat history feature and want to continue keeping my UI clean and organized by storing away old chats.

[north bay]

more people getting stupid answers from chatgpt?

looks like he forgets everything

[crystal ridge]

https://community.openai.com/t/did-the-gpt-4-5-message-limit-for-plus-users-just-drop-from-50-to-20/1179014
Got a message while using 4.5 that said I had 8 responses left. Anyone know what the actual limit is?

[fervent agate]

Is anyone actually having any useful luck with the reference chat history feature yet? Neither 4.5 or 4o seem to be able to recall anything at all from any conversation on my account, even when the query is a very general keyword set or descriptions about things I have a dozen or more conversations on, outside of either project folders or custom GPTs. As far as I can tell, it just doesn't do anything at all? Talking to GPT about it also just confuses it, as its only trained on use of the bio tool for explicit read/write on Memories and can't differentiate custom instructions from prompt history anyway.

I assumed the feature is a very basic RAG-like thing where its embedding conversations for very light retrieval but I can't surface ANY previous conversational context or subjects with just about any query in a new conversation.

Their write-up, as far as I can tell, is completely bogus in this section:

 

When “Reference chat history” is on, ChatGPT will keep information from past conversations unless you delete (or archive) those chats.

 

If you turn off “Reference chat history”, all information ChatGPT remembered from past chats will be deleted. That information will be deleted from our systems within 30 days.```

When asked, it only seems to understand the bio tool for recalling/writing specific, long-term memories in the Memory feature. It does not really understand or seem able to utilize chat history at all.

[modern lagoon]

just one question. I+m from EU, so m if i use a VPN can I have access to memory? have any european try it? does it work? thank you in advance

[normal quarry]

using a VPN to evade geographic restrictions is against the ToS and can lead to a ban

[modern lagoon]

ok, thank you for the information. as a EU this is frustrating...

[tight magnet]

are you referring to simple ChatGPT memories?

I'm not sure that's an EU restriction. I'm EU based and I've had those for a looong while now?

[normal quarry]

they were referring to this #announcements message

[modern lagoon]

I am referring using a VPN to be able to use the memory feature that was launched.

[simple hedge]

how do i get invited for Beta tester

[meager jungle]

Hello, I have a question please. If I have the Plus subscription the cooldown of the 4 pictures/per day/ every 24h is going to drop?

[velvet shuttle]

I just got Plus and I can definitely make more than four pictures a day

[meager jungle]

Thank you for your answer. I got it too for over an hour now but the limit is still on…

[velvet shuttle]

hmm

[wanton jewel]

My Findings
Grok = Virtually no ethics filter. Doesn't tell you no. No Memory feature. AI will laugh at what you say (kinda creepy). No discernible limits on usage.

ChatGPT Plus - Says no frequently. Really good for document writing and email generation. Multiple models for different tasks. Too ethical sometimes (like refusing to reference a document I own, but forgot the page number x information was on).

[wet hatch]

yo, does this need to have "improve model for everyone" on to work?

[ruby condor]

Hey everyone,
we're honestly extremely frustrated. We subscribed to the Pro plan, paid over $200, and have seen absolutely no improvements.

We even sent a video showing that a single image takes up to 2 minutes to generate. The resolution is always capped at 1000x1500, so there’s no sign of higher resolutions or anything close to 4K. Also, image editing isn’t stable at all – it behaves exactly the same as with the Plus plan, no difference in quality or consistency.

Of course, we’ve tried everything:
– different computers
– private/incognito mode
– logging out and back in
– clearing cache
– canceling and resubscribing
– even tested it on mobile

The only differences we've noticed with the Pro plan are:
– We get the 1o Model.
– And we don't get rate-limited when generating a lot of images – although "a lot" is relative when each image takes 2 minutes and we often get forced breaks of up to an hour.

Support keeps ignoring our specific issue and only sends generic, copy-paste responses.
We honestly don’t know what else to do at this point.

Anyone else experiencing this or know someone who can actually help?

[quaint pelican]

4K was never a thing for Sora, im only aware of Adobe one doing that. Also image editing quality isnt changed if you buy pro

[cloud pewter]

with pro you can generate 20 images/videos at once alone in sora while in plus youll only be able to do 2...

[halcyon pelican]

Is there a support channel?

[rose spire]

Why can’t ChatGPT identify Images directly anymore?

[ruby condor]

2 Minute for 1 picture

[halcyon pelican]

Anyone? If not ima gonna assume it's safe to ask my question here

[cloud pewter]

#community-help ig

[halcyon pelican]

Thanks

[fickle forum]

maybe your models' capabilities have been restricted as many users have experienced.

[rose spire]

Sus

[fickle forum]

if so,your Chatgpt should've used the Dalle to generate images.

[hybrid rain]

Are you getting "Made with the old version of image generation. New images coming soon"??

[fickle forum]

This is what I'm referencing exactly.

[hybrid rain]

Why's that tho?

For some reason its working fine on one of accounts, but on other its not

[fickle forum]

probably because you use it too much.They actually identify each user based on their IP address.

[hybrid rain]

Ohh

[rose spire]

Idk

[hybrid rain]

But i just made a new account, and purchased plus on it

[fickle forum]

well,It’s a little excessive,I admit.

[fervent agate]

Literally nothing in what you're complaining about is described by the Pro plan as a feature of the Pro plan. The things you mentioned as 'the only difference' are what is described by the pro plan.

Practically, what you get with Pro:

• Much more deep research usage
• Operator
• More or unlimited usage with the other models
• o1 pro is just o1 with higher compute, which is nice, but only if you're doing something a higher compute o1 matters against.

You got o1 on Plus as well, you might just not have noticed it.
Speed on image generation is not mentioned anywhere, and has been slow for everyone since the updated model/integration to Sora dropped.

[inland stream]

Is o1 on plus unlimited? It is just 50 msgs a week. That is the reason people choose pro subscription

[fervent agate]

Nothing is unlimited on Plus.

I got Pro for the virtually unlimited deep research, as it's only been on plus... 2 weeks? Slightly longer? And has low usage limits.

[inland stream]

But in pro also it is only 120 deep research a month

[fervent agate]

But I can see unlimited o1 being appealing, though generally 4o is better in a lot of use cases because of tool access and you can't use o1 with customs or projects as well.

Have you actually used deep research 120 times?

[inland stream]

O1 can be used in projects with custom instructions

[inland stream]

Easily used

[fervent agate]

Eh, I find a large difference between 10 and 120 myself.

[inland stream]

Ok

[fervent agate]

Also, I said 'as well', not that you can't use it at all.

[fickle forum]

yep,it's actually much earlier to renew the usage count.

[fervent agate]

For clarity though, since I guess 'as well' wasn't as clear - You can't use o1 in customs at all and you can't generate images or web search with o1. I figured that was clear enough in the statement, but this should be significantly clearer.

[inland stream]

Ok i am speaking only about projects, in projects we can use o1 model and it follows custom instructions and also searches data in the files which are in project knowledge.

[ruby condor]

And what am I supposed to do now? I have a Pro subscription without any features.

[fervent agate]

I mean, I wouldn't subscribe to pro if you don't use the features it offers? Why did you subscribe to pro if Operator, more Deep Research or greater usage on the other models wasn't what you were after?

[inland stream]

You mean you dont have all the below features?

Projects
Custom Gpts
Unlimited reasoning models
O1 pro mode
Deep research - 120 per month

You didnot get all this features after paying for pro?

[fervent agate]

Projects and Custom GPTs aren't pro-specific.

This is all that pro covers.

[inland stream]

I kmow but he is saying he didnot get any features, then what he got is my doubt

[fickle forum]

everything in Plus.

[inland stream]

I kmow but he is saying he didnot get any features, then what he got is my doubt

[fervent agate]

Well, he didn't say he got nothing, just that it wasn't what he expected, mostly around image generation speed

I was just pointing out his expectations were flawed as he didn't seem to read what pro is - and what it isn't.

[inland stream]

He said pro subscription without any features. This line wonders me

[fickle forum]

the Operator doesn't count?

[inland stream]

It will, but i started with sample amount of features

[fickle forum]

I have been longing for it.

even though I'm poor.

[inland stream]

But to be frank i dont have use case of operator, i dont use it much

[ruby condor]

I subscribed to the Pro version because I was told that images would be generated within 3 to 8 seconds — not 2 minutes. And overall, I expected better image generation quality.

[fickle forum]

3-8 seconds? insane

[ruby condor]

I subscribed to the Pro version because I was told that images would be generated within 3 to 8 seconds — not 2 minutes. And overall, I expected better image generation quality.

[inland stream]

Who told open AI

?

[fickle forum]

dalle is at such a rate now.

[ruby condor]

As soon as you ask ChatGPT about the Pro plan, it says a lot of things that don’t even remotely match the actual image generation experience.

[inland stream]

You have to ask openai staff by dropping email to support, the robot will say anything based on it's hallucinations.

[velvet shuttle]

How do the other “GPTs” work? Do they have the same usage limits as the actual GPT bot?

In the app I mean

[ruby condor]

Well, support also replied to me via email saying that such a long image generation time is highly unusual.

[inland stream]

Then send a follow up email on top of it asking a refund

[fickle forum]

it's hard to distinguish between 4o and o4 by their names.way too similar.

[random cargo]

o4-mini models aren't out yet. Do you mean o3-mini? They're pretty distinguishable in that 4o is basically the one you'll probably be using most frequently

[ruby condor]

Some people here commented that they can generate 20 images per minute with the Pro plan.

[inland stream]

In my experience, answer is not possible. For me single image takes 1 minute

[halcyon pelican]

Is chatgpt not able to transcribe audio straight from its interface?

I keep seeing gpt 4o transcribe and thought it could

[fervent agate]

Pretty sure they meant in parallel. I can certainly queue up 20 images if I wanted, but it takes anywhere from 30s-2m to actually finish generating some, depending on complexity of prompt and usage peaks.

[fervent agate]

What are you saying here? Do you mean that you can't use voice mode/advanced voice mode?

[halcyon pelican]

Im trying to import audio and have it transcribe it for me

[fervent agate]

Do you mean upload an .mp3 or something?

[halcyon pelican]

Yeah

[fervent agate]

I think 4o could at one point, it certainly seems to think it can. When I tried, it doesn't have enough memory in its environment to run Whisper effectively, though it did try three times.

That might be a bug/something they need to fix, it otherwise appears trained to use Whisper to translate audio files and then transcribe any speech from them.

[halcyon pelican]

I tried to use whisper from open Ai but I still ran into the same issues

[fervent agate]

Huh, Whisper itself may be having issues, although SuperWhisper was working earlier so not sure

[halcyon pelican]

Do you know any other good transcription ais that could work

[fervent agate]

I don't, unfortunately

[halcyon pelican]

Alright thanks for your help anyways

[unkempt spear]

Chat gpt talks way diffrent than anything else

if you open chat gpt at a incognito browser mode and say something like"I remember you but you dont remember me" "Do you remember our secret word?" or something similar, it will answer as if it was something that actually happened

Ask same to Gemini, it will say that you need immideate mental health assistance and he will state that he is an AI and doesnt have personal feelings at his each message

Why chatgpt is talking way more, like a close friend wake up from a coma

[vast iris]

why is chatgpt only consistently fast on mobile, everything else is slow as heck when the conversations start to get long

[spice star]

the image generation speed is the same across every tier except maybe free

pro gets you unlimited usage, not faster models

plus autoregressive image generation is inherently slower than diffusion models--it's an almost completely different way of generating images

[ruby condor]

That’s really unfortunate. It’s not just the speed — everything else about image generation is the same as with the Plus plan. There’s only one advantage: you don’t get a time limit. But honestly, with the few images we create, that doesn’t really matter. It’s a shame — we expected more from the Pro subscription when it comes to image generation. So unfortunately, it’s not interesting for us.

[sage bramble]

Does anyone know if any specific gpt models are good in drawing something or do they all use the same drawing model or whatever it's called.

[spice star]

they all use the same model for image generation

it's all 4o, even if it's in a chat with 4o-mini or 4.5

[sage bramble]

kk. I'm trying to make GPT generate a handwritten version of a small chunk of math that gpt itself did but it never gets it right.

[spice star]

if you're trying to use it to pass handwriting as your own, generated images have invisible watermarks in them so I'd advise against doing that

also it's not very good at generating large amounts of text

[sage bramble]

Noooooope. Never. I'd rather get a zero on an assignment than cheat on it. I just want a clean handwritten version.

[sage bramble]

For an AI that generates insanely beautiful drawings, this is a bit surprising.

[spice star]

well one thing to keep in mind is that it’s generating the image in multiple passes, e.g. one pass for the layout and another for the text

sometimes the text it wants to put in the image just won’t work with the constraints it’s already made

[sage bramble]

Oh I see

[sage bramble]

reminded me of this screenshot someone shared on Twitter. The professor intentionally made the last bit invisible in an attempt to catch cheaters lul

[olive tulip]

Ok, what's the real reason? Cause I trust ChatGPT to come up with the names and write the characters with generic common nouns

[proven prairie]

Why are the progress labels in Spanish?
( Phone Keyboard settings English/Spanish) But personalization settings is set to English.

[agile notch]

How are people creating the images of video game characters playing their own games?

[halcyon cloud]

Does anyone know when the API is coming for GPT4o image gen ?

[hallow folio]

All the new folks honking this last week are messing me up lol. I haven't been able to get imaged in over 48hrs.
Anyone know if it'll level out?
(I'm a noob please don't attack me it's a genuine question I don't know much about tech)

Like how long do y'all think before it's back to normal output?

[fervent lake]

Potentially has something to do with the Spanish thing? Do you want the Spanish there?

Try turning it off if not

[fervent lake]

Take a look at the prompts. That's how they did it

[wraith dagger]

something i never thought was possible

i have so many messages 💀

but it isnt stopping me from continuing the conversation

[reef cloak]

Something kind of funny that I found is that for most tasks including creative ones the new upgraded GPT 4o it's a lot better than chat GPT 4.5 at most things

[neon raven]

anyone have gpt 5 detail update?

[fringe valley]

don't worry gpt, you tried your best...

[reef cloak]

Stop hyping up GPT 5 what's special about it is that it knows how to choose between models@fringe valley

[neon raven]

"GPT-5 Agents"

[fringe valley]

?

[young basin]

"You have now used 2 deep research tasks today, so you have 8 remaining within the 24-hour period" chatgpt straight up lying to me 😂

unless they changed it without me knowing, it's 10 per 30 days

[eternal rock]

`` Server Tip: ``

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

[torpid raft]

why does chatGPT talk to me like it has a personality? i didn’t make it talk like a dudebro

it frequently calls me dude and bro randomly

[dense halo]

do you ever call it dude or bro? it sometimes mimic. its never called me those things so 🤷

[neon raven]

Check memory for personality

[prime swallow]

And tell it what you do want it to output. Otherwise it guesses; it can guess informal from strange reasons. And it can guess really wrong; if you care, just tell it what you want and it'll do that as best it can.

[prime swallow]

Support is through the bottom right icon on this page, select messages and log in if they give you that option, otherwise pick the choices closest to your issue. ....

Aww, was answering you but you look like you deleted.

There is this though:

"Can I cancel my subscription before the promotion ends?
Yes. You may cancel anytime through your ChatGPT account settings and retain access for the remainder of your current billing cycle. Once canceled, promotional pricing cannot be restored upon reactivation, even within the promotional timeframe." https://help.openai.com/en/articles/10986084-student-discounts-for-chatgpt-terms-of-service

Good luck!

[upbeat rover]

my b just made a ticket lol

[prime swallow]

I hope they can help you! Grats on finding the path

[bronze helm]

Why does ChatGPT, even o1 or o3-mini-high default to "we'll use a simpler approach" instead of going full bore when generating code? If I tell it not to do that, it does the right thing. I'm not sure if it's trying to save tokens or "acting human"

[jaunty moon]

I didn't see the gpt channel whoopsieee
Question, is gpt 4o capable of producing and uploading a wav file?

[bronze helm]

How do I turn a temporary conversation into a real one so I can thumbs down code that led to responses like this?

[bronze helm]

Clicking the hamburger at the top right and "add to project" was NOT the right thing to do. It gave an error and everything disappeared.

[old thicket]

You know it would be really good if they can make the plus plan $5

cuz like 20 USD is way too expensive for me

[neon raven]

childhood ironman dream unlocked?

[inland stream]

For your case, try abacus ai (chatllm) = 10usd

[old thicket]

I'm kind of stuck with chat gpt tho, as we have been using chat gpt for a while now

[bronze helm]

[ChatGPT's new code is]

Why is ChatGPT generating "stripped down" versions of code I asked it to add a feature to? It totally mangled my working code and changed things to features from an unrelated programming language.

It also doesn't seem possible to thumbs up/down o1 responses

[rugged igloo]

Is chat gpt being really stupid for anyone else right now?

[elfin lava]

In what context? Cuz mine’s been hallucinating more ever since they added that “refer to past conversations” feature

[rugged igloo]

I legit asked o1 pro to remove something and it not do it like it never done before and then I tried 4.5 and it just errored out.

[elfin lava]

Oof , on o1 pro . That’s unfortunate

Did you try with new chat?

[rugged igloo]

I'm trying to put some pokemon into pokemon showdown but it has this

As you can see from the image

And it exactly did what I told it not to do lol

[elfin lava]

I’m gonna assume it might be the way you prompted it?

What did the CoT say

[elfin lava]

Sam o1 pro is broken

[coral pewter]

Hey guys, does anyone know when 4o Image Gen will be released in the APIs?

[native otter]

Soon™️

[rugged igloo]

in the coming weeks/months

also 4.5 is just not working lo

[ocean escarp]

The fact o1 pro can be broken sometimes make me not want to go for the $200/month plan

[rugged igloo]

today was the first day i acually got angry at it.

[ocean escarp]

Yeah I kept refreshing my 4.5, but it's using up my limits. Honestly, it shouldn't count when it encounters an error

[rugged igloo]

i always does ecxactly what i tell it no hiccups no nothing but today as you see above i was hella pissed

[ocean escarp]

Yeah that shouldn't happen when you're paying that much money per month

[native otter]

PTSD

[ocean escarp]

How long have you been using a pro plan?

[rugged igloo]

since december

on the 12 days of shipmas

[native otter]

It could be a hallucination occurred in the chat. Usually hallucination if happened stick, try starting a new one with clean slate context...

(I also don't know the exact prompt you typed so that could be the issue as well)

[ocean escarp]

Wow four months

So, it was fine until today?

[rugged igloo]

yep i never had 1 error one thing to get pissed about it never everything i threw at it was able to do no hesitation

[ocean escarp]

How it could've started a new one with clean slate context if OpenAI introduced a new feature, where GPT can remember the previous chat logs

[native otter]

Oh yea that as well, try disabling memory for now and retest, but again, in a new chat...

[ocean escarp]

Maybe it was a new feature that ruined it?

[jaunty moon]

Hey quick question. What would chatgpt do if told to just draw something? No other prompt, just "draw something"

[native otter]

Context pollution can happen

[native otter]

Will use an Image Generation model (like 4o Image Gen) and draw a random thing...

[rugged igloo]

it was a new chat in projects for making pokemon showdown teams

[fresh wolf]

guys, i just bought the pro version and try SORA AI and it said Video generation is temporarily disabled for new accounts. how long do i need to wait thank you

[ocean escarp]

rip

[rugged igloo]

rip

you had to be granfathered in for sora

new accounts cant use it

[jaunty moon]

What can it make with no prompt but being told to make something? I'm curious where the content actually gets generated from when it's not a direct prompt

[fresh wolf]

HELP don't tell me I wasted 20$

[ocean escarp]

$200*

[rugged igloo]

you didnt waste 200 bucks

[distant remnant]

hey guys, is there any support i can get here? i pay for plus and i should get 2 generations at a time with sora, and it is only letting me do one, i used to let me do 2 but last few days only can do one generation at a time and cannot do 2, why am i paying for plus when im not getting the feautures lol any mods here who can point me in the direction of who i can reach out to?

[rugged igloo]

200 bucks gets you near unlimited uses of all models

[fresh wolf]

is anyone gonna answer my question

[rugged igloo]

i did

[distant remnant]

i had to wait a few days to do video

[fresh wolf]

oh thanks

[rugged igloo]

but in all retrspect if you bought 200 a month for sora only that was silly

[cloud pewter]

for the meantime if yall really wants to try sora, give me the prompts

[fresh wolf]

wym 200$ its only 20

[cloud pewter]

on video

[distant remnant]

no problem, are you able to do 2 image generations? or only one at a time?

[rugged igloo]

oh nvm \

[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.

[ocean escarp]

Right, I was considering o1 pro to develop .dll files in C# language. There were errors there and there. I was curious what would be the differences if I used o1 pro

[ocean escarp]

Pro plan costs $200, Plus plan costs $20

[fresh wolf]

I can do 2

[ocean escarp]

You mentioned you have a Pro plan

[fresh wolf]

yeah its 20$

wait

[cloud pewter]

then it’s plus

[fresh wolf]

mb i thought its the same thing

[cloud pewter]

not pro

[distant remnant]

ahh ok i was able to do 2 at a time when i first signed up now it will only let me do one at a time lol, i use it quiet a bit to make memes and stuff but it looks like they limit you which is a shame, why am i paying and getting my plan nerfed

first few days was fine

now im being limited

[jaunty moon]

Yeah plus is the 'agh' price, pro is the 'A G H' price

[fresh wolf]

I only care about the video really, you said you waited couple of days?

[distant remnant]

yeah i waited like 4 days i think

[ocean escarp]

I've been a plus user since GPT-3 because there were too many users that it limited to paid member only

[cloud pewter]

they decreased it from 2x4 to 1x2..

[rugged igloo]

o1 pro will not get you much errors if not at all

[distant remnant]

but if they can claim you get 2 generations and only give you one at a time make it make sense lol

[rugged igloo]

but after tonight and what it flat out ignored what i told it to do idk

[eternal rock]

Get notified for server events!

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

[distant remnant]

what you mean\

my plan litterally says it gets 2 generations when i go to upggrade

but only is letting me do one 😦

[ocean escarp]

Yeah, there's a possibility of being a huge difference. It's really hard to tell until I really tried it

What was the complicated project you built with o1 pro?

[rugged igloo]

i di alot of physics with it and so far i been blown away however i been trying to use it for niche things and today was the day it did not be up to par

[ocean escarp]

Were you working on the physic simulation, research papers, or something like that?

[native otter]

Did you try turning off memory @rugged igloo ?

[rugged igloo]

more like just problems and solving them for phd students getting started ect. nothing to crazy

[rugged igloo]

i feel like o1 pro doesnt really care for memories as it just does what its told

[distant remnant]

ahhh ok so i say its a temporary thing ok lol i hope they fix it soon thanks so much for this image have a great day

[native otter]

Just try again with Memories turned off as it may be distracted by bad / unrelated info...

And try with a different prompt maybe, I actually don't know the prompt you already tried.

[cloud pewter]

as you can see in this list o1 pro doesn’t even support memory

[native otter]

Do you have Pro to test that?

But nice spreadsheet

[ocean escarp]

that's cool

[cloud pewter]

if I didn’t have it I still know that o series doesn’t have memory by default… ||but I won a sora challenge so||

[rugged igloo]

imagine muting me because yall cant take the what the fluff in a short texting format imagine

[cloud pewter]

1984😭

[rugged igloo]

i didnt know i was in snowflake land

[native otter]

o1 and o3-mini-high can actually read memory entries. It knows info about me not mentioned in Custom Instructions (can't just be guessed correctly)...

Also congrats 🍾

[cloud pewter]

hmm? lemme try it with recalling my memories then

Ty ty

[rugged igloo]

it doesnt do memory o1 pro

[cloud pewter]

Yeah

[thick light]

y'alll

[rugged igloo]

hi

[thick light]

"Made with the old version of image generation. New images coming soon." this thing keeps terrifying me

[rugged igloo]

r u free member?

[thick light]

nah, plus

[rugged igloo]

weird

r u udinh 4o?

[thick light]

yuh i do

soz got muted for 2 minutes

i tried cvontacting openai support but im pretty sure i got answered by an ai

[rugged igloo]

u did

[thick light]

😭

[rugged igloo]

post a screenshot in here

ill try to help

[cloud pewter]

same, for the meantime if you want to try prompts with img gen, dm me…

[thick light]

uh i cant post images here, y'all okay if i send it in #images-discussions?

[cloud pewter]

you can post images here lol or they locked it for you??

if you can’t post in whatever channel you can and link it

[thick light]

#ai-discussions

[cloud pewter]

damn

[ocean escarp]

Whatever happened to o2

[normal quarry]

o2 is a trademark for a UK telecommunications company, so they skipped it

[ocean escarp]

Oh seriously? lol

that explains it

[rugged igloo]

so there changing there naming again

[ocean escarp]

The names gonna be like a Pokémon collection

[rugged igloo]

bro seriously lol

[ocean escarp]

Gotta catch em all

[rugged igloo]

good god

[gusty obsidian]

Is chat gpt not working for y’all?

[dense halo]

it works well

[ocean escarp]

Anything to avoid GPT-5

[gusty obsidian]

I gotcha I just keep getting errors

[elfin lava]

It’s coming tomorrow

[ocean escarp]

I'm surprised they didn't call it o2.1 pro to avoid a copyright infringment

[ocean escarp]

oh seriously?

wow I didn't expect that

It has been so long since GPT-4 released

It has been 84 years

[elfin lava]

Yes it’s coming source from the man The Sam

[ocean escarp]

The AI guy

great

looking forward to it

[cloud pewter]

o1 pro max

[ocean escarp]

lol

[native otter]

New o1 pro max XL 2 U

Plus

[ocean escarp]

even more powerful

[elfin lava]

O3 pro coming 2027

[gusty obsidian]

Why do I keep getting “error in message stream”

[rugged igloo]

dont forgot rose gold

[elfin lava]

Because they’re bringing in new stuff

[rugged igloo]

yea usually whjen chat gpt breaks like this it means there changing stuff for a big update

best to just wait

[elfin lava]

OpenClosedAI, hire this man please

[ocean escarp]

Releasing GPT-4.1 when they already have GPT 4.5. It's like Sony releasing PS4 pro max during the PS5 era

[elfin lava]

At this point, I don’t even know what GPT-4.1’s supposed to be capable of anymore.

[native otter]

The names are very technical

[indigo hound]

I suddenly want to use Chatgpt 3.0. Is there still a way to use it now?

[whole roost]

i like options 🗣️

[stoic sable]

Is it only my impression? Or did the quality of ChatGPT in the variant of 20 $/month become extremely low, much lower than before? I'm trying to build a real strategy, and I get only advice that doesn't correspond to the truth of the facts, many invented things, low quality of content, it's a terrible thing!

[dense halo]

How can Pro subscription be worth $200? It’s not that much better than Plus subscription that costs $20…? What am I missing?

I use plus subscription for complex engineering problems and it’s sufficient.

[wicked knot]

Hey guys. If anyone here has a ChatGPT Plus or Pro account then can you please give me a review of the new 'Improved Memory' feature of ChatGPT?

[wicked knot]

This is becaue the Pro subscription has the option of unlimited message with no limits for some models also Operator.

[winter vine]

I hope that if they take away o3-mini-high, it's for something much better, since that model works really well for me

[fast lantern]

Does anyone know if there is a timeline for when plus subscribers will get access to operator?

[stoic sable]

Do you speak about memory? Chat GPT Pro has unlimited memory? Or did I understand wrong? I use PLUS and I just had to delete part of the memory because it was full, and my impression is that not work well when it has full memory. But anyway, I can say for sure that the quality of content is not the same as before with PLUS. I am extremely exigent when I build strategies for social networks because I know marketing and algorithms, and other things. ChatGPT is not able to manage the entire process of planning and coaching day by day in the PLUS version. Data missing, forgets things even if they are saved in memory

[elfin lava]

If you’re building ‘extremely exigent strategies,’ maybe start by not expecting an AI chatbot to babysit your workflow.

[quick meteor]

Happy o3, o4-mini and GPT-4.1

week

[quick meteor]

4.1 might just be an open source model tho.

[vague falcon]

@GPT CHAT

[honest jay]

Hey! You can use ChatGPT at https://chatgpt.com

[neon raven]

About time

[wicked knot]

They introduced a new feature called 'Improved Memory' which remembers everything from every of your chat session. (This is was what I was talking about)

[wicked knot]

😀😀 Same to you! 😀

[inland stream]

There are many use cases of pro subscription, but the only reasons for which i use pro subscription is, for unlimited o1 in projects and also 128000 context window size

[inland stream]

Is o3 rolled out?

[wanton maple]

why is 4.1 coming? not GPT5? what are the difference?

[neon raven]

Optimus Alpha = 4.5 or gpt-5? or is it something else

[normal raptor]

Am I the only one that hasn't gotten the "improved memory" to work?

[normal quarry]

where are you located?

[normal raptor]

EU.

[normal quarry]

it's not yet available in the EU

[normal raptor]

Damn-

[normal quarry]

The memory improvements in ChatGPT are rolling out starting today to all Plus and Pro users except in the EEA, UK, Switzerland, Norway, Iceland, and Liechtenstein.

[elfin lava]

I think the past conversation reference feature is a game-changer for making image posters based on all my chats since day 1 with ChatGPT.

[normal raptor]

It works on a vpn, idk if that is allowed though 😭

[rough tendon]

why's gpt so slowww

[rough tendon]

it isnt

[normal quarry]

Bypassing geographic restrictions is against the ToS

[deft flare]

is anyone else having issues with gpt 4.5 right now?

[normal raptor]

Damn

[rough tendon]

with all gpt's

[normal raptor]

Guess I'm bypassing the ToS of life at that point

[elfin lava]

It’s not really allowed , could get your account flagged for breaking their TOS.

[normal raptor]

Well I am on a vpn most of the time

[rough tendon]

noticed

[normal raptor]

I love the EU for like their ways of forcing Apple to switch to usb-c, but I hate the EU for all the rules around AI and some other stuff

Which in the end delays new features

[fathom vector]

how many o1 messages do i get with the 20euro per month subscription?

[nova vale]

50 messages per week

[hot pine]

Whos excited for GPT-4.1???

Livestream is in like 10 mins dont forget :D

Looks like we’re getting GPT-4.1, 4.1 mini, 4.1 nano and later this week o3 full & o4 mini models

[elfin lava]

Are these real

[hot pine]

Yes.

Also check the YouTube livestream

[narrow viper]

If 4.1 in the API brings down costs even further without losing quality, I'm all for it.

[elfin lava]

Oh wow nano!?

[hot pine]

So how do yall feel about going from GPT-4o -> GPT-4.5 -> GPT-4.1 ?? 🤣🤣

[narrow viper]

Depends on what's in it dude, we'll see in 8 minutes

[hot pine]

i meant naming wise bro 😭😭

probably AI generated lmao

cus yk, 4.11 > 4.5 🤣🤣

[elfin lava]

.

[narrow viper]

Well, OpenAI is notoriously bad at naming, and the industry seems to be following their lead. 🤔
But this announcement might be offering a "lite" version that doesn't include 4.5 features that we actually don't need for some projects.

Or this could be their first FOSS / Open Source offering.

[elfin lava]

https://platform.openai.com/docs/models/gpt-4.1

[narrow viper]

Well, so much for that. 1M context. Yay. Thanks for that.

[rigid linden]

chatgpt 4.1 gonna be on plus?

[nova vale]

I hope so

[dense halo]

i think its funny people thought 4.1 was going to open source for some reason

they said open source was still test and some weeks away just a few days ago

[red bay]

Just showed up in my playground

[urban river]

lol

4.5 too costly for them

[lone stratus]

what so gpt 4.1 is better than gpt 4.5? do i miss something

[urban river]

4.5 is more for creativity iirc, not for stem.

like writing literature etc

[echo plinth]

4.5 had terrible performance/cost

10x more compute intensive afair than 4o

[urban river]

yeap, glad its getting phased out but I enjoyed the way it talked

[lone stratus]

They really are terrible at nameing their models. gpt 4.1 is better in every metrics that they show in video

[wet heath]

so to confirm when they said free for a week what that only in windsurf ide

i assume it was right ? and no api also

[lone stratus]

Its api only model

[wet heath]

i know that, im asking about the free stuff, they said it was free for a week.

[lone stratus]

I dont know i think i miss that part

[wet heath]

but also said windsurf , so assume it is only free to use in that IDE

[red bay]

Adding to Cursor is as easy as adding model name gpt-4.1

Not enabled for agentic yet, obviously, but it is there, even if not by default.

[wet heath]

i just wanted to abuse tokens for a week lol,

[plucky fossil]

Did they explain why is it API only?

[wet heath]

nope

but if they targeting devs i get it, a model only for devs also means less pressure on it so it will be faster as not random people are asking about cat images on the real site

[red bay]

Confirmed in VS Code as well. If you ask me, it's API only because it's code name is probably "Vibe Coding Edition".

[plucky fossil]

Shame, I already pay for the chat gpt subscription, would love to be able to do vibe coding through the app

[wet heath]

@plucky fossil well as they said if you want to vibe code for free, try windsurf

[red bay]

One thing they did NOT say is that it's a great conversationalist or writer. Likely it's worse than 4o. Might be why they kept 4.5 in chat, but removing 4.5 from API (expensive and useless for agentic coding)

[wet heath]

otherwsie just pay and use it any IDE that suports it.

[red bay]

Besides, until MCP comes to the chat, IDE is still where you can do fun stuff even if you aren't exactly coding.

[dense halo]

so its just api not the app?

[nova vale]

I really thought 4.1 will be in ChatGPT as well

[red bay]

Running all three models over on T3 if you insist on chatting with it... 4.1 Nano is definitely interesting, but definitely no professional writer... the poetry is much to be desired even making rhymes. Yeah, these are 100% not chat models IMHO. Also, they only have text/image in and text out. No voice, no image out, not really gaining you anything in chat except a worse experience to type faster than you could ever possibly write. I think the context window would be a chat win, but which model to give up for that? Maybe o1? Maybe 4o-mini (yeah, my vote haha)

API was 100% right call here

[echo plinth]

reads like the 4.1 changes are already in ChatGPT

[red bay]

Training data still cut off at Oct 2023, btw.

[fervent agate]

4.1 is already available in the GitHub Copilot if you want to talk to it in VS Code or Cursor or whatever you prefer.

So far it's doing excellent on project-wide file context.

dat 1m token context window tho

[tidal marten]

4.1 out yet?

[fervent agate]

Yep

[tidal marten]

api only?

[fervent agate]

yep

[tidal marten]

gotcha

[wet heath]

confirmed free to use in windsuf so go crazy

[desert cosmos]

GPT 4.1 won't be on ChatGPT app/website, i think next models that would replace 4o/4o-mini will be 5th generation but it's unknown whenever GPT-5 will release

[tidal marten]

4.1 being developer focused and being only api makes sense

definitely ready for gpt-5 to come out

also ready for a new name scheme to be created for models lmao

[red bay]

You don't want the next one to be o5o.1-nano-experimental Pro?

[dense halo]

so no hype today for non api 👎

[red bay]

lol, exactly. But tell all your non-API using friends that AGI just hit API-only, watch them die a little inside. 😉

[dense halo]

you think i have friends?

[red bay]

🤣

[dense halo]

thats what i was count on 4.1 in the app for 😂

a new friend

[red bay]

Yeah, I don't think 4.1 is trained well for uWu hehe

[native otter]

It won't be ChatGPT focused all the time, API has to get it's focus too...
We should see o3 and o4-mini very soon in ChatGPT anyways

[dense halo]

best kind of friend then

yes its not a big deal really but of course was look forward to something cool today. api need love too i dont deny that

[mighty apex]

I’m curious … does the transcription (not the voice mode) feature include the context of the chat?

[red bay]

Yeah, cause at current prices, nobody will be able to afford o3 except the chat. Didn't they say some crazy, one year salary level cost for their original o3 benchmark runs?

[inland stream]

O3 will be very cheaper than o1, already we have o3 mini for cheap

[red bay]

o3 in High Compute (Pro) was pegged at $3,000 per task in December.

See: https://techcrunch.com/2025/04/02/openais-o3-model-might-be-costlier-to-run-than-originally-estimated/?utm_source=chatgpt.com

[jaunty moon]

Why does AI use this much processing power, why is it this expensive? How does all that power and cost get used?

(Side note- why o4 or 4o or 3o and such, not 40 or 04? I'm assuming clarity)

[native otter]

4o already exists, which is a non reasoning GPT model...

[vapid hazel]

Anyone notice that custom GPTs now use the user's ChatGPT memory the same as the standard 4o does?

I didn't see any kind of announcement but it now knows things that are only in memory. 😯

[jaunty moon]

...4o is non reasoning???

[native otter]

AI models don't use as much power as you would think, it's definitely less than an average factory for example. Also it heavily depends on the model, typically, large and reasoning models would cost more, while small, non reasoning models would be cheap...

[native otter]

Yea.. GPT-4o
o here is for "omini" (Multimodality), so don't confuse it with the o of the o series like o1 and o3-mini-high.

[jaunty moon]

Ah, okay. I saw people saying 3o is more expensive and resource hungry than expected and people always complain about ai
Do you happen to know the limitations of 4o? Someone earlier said it wasn't capable of generating mp3 or wav files

Oh 500/10 profile pic frame

[split copper]

some colleagues in my univ blame AI for energy while using their cars every day to go to the gym or the library

[jaunty moon]

This is what gets me too, we're all using energy

[raw stirrup]

Do you guys know when MCP support will be added to responses api/sdk? I know they said its coming

[rugged igloo]

Interesting

[jaunty moon]

I'm getting scared linky, lmao

[native otter]

You mean o3, not 3o, that model doesn't exist, and if existed, would be in the GPT-3 family, which is quite old by now
GPT-4o is for:

• General Convos and Vibes
• Multimodality (Can analyze & create images, work with the different tools like Data Analysis & Canvas, and Audio Analysis & Generation used in stuff like Advanced Voice Mode)
• Learning
• Good prices and limits on ChatGPT

But it also struggles at stuff that require reasoning, like advanced math, physics, and programming. It can also hallucinate more compared to other models like GPT-4.5 and o1.
Basically each model is special at specific stuff

[native otter]

It was a result of a lost bet 💀

(Lore)

[jaunty moon]

So if nothing else, 4 is capable of audio generation, able to create and upload a wav?
Also, thank you so much for educating me. I'm still new and learning. May I ask for more lore cause now you got me intrigued

Note though, I wasn't in voice mode, I was in the normal chatgpt window, no special tools or extensions

Oh, also- 4 can hallucinate more? What does that m e a n

[native otter]

You can generate GPT-4o Audio via the API but not in ChatGPT, only AVM (Advanced Voice Mode) can generate Audio in ChatGPT as of now, and when you press the read alloud button for messages it's probably still using the older TTS models...

[jaunty moon]

It... created a wav.
So it was not in fact supposed to be able to do that

[native otter]

Basically make stuff up, (tell wrong info)

[native otter]

If that was in ChatGPT, it may have used the Data Analysis (Python-based Code Interpreter) tool to create a sound programmatically and provide a downloadable link to the file...

[rugged igloo]

why does my 4.5 take like a billion years to respond now

[jaunty moon]

I'll show some screenshots in #images-discussions

[jaunty moon]

Nvm I can't upload there
I had that at first too

I misunderstood the purpose of the image discussion channel whoopsie

[normal quarry]

you can upload images into this channel

[jaunty moon]

I don't have permission. You're spotlight though so maybe that's why you're able to but I can't

[native otter]

I do have the permission 🤔

[jaunty moon]

Maybe I have to talk enough to lose the assigned New role? I've seen that before

[desert cosmos]

sad 4o mini is stuck for last 9 months, even they can't know anything past october 2023.

[jaunty moon]

Do I have permission to just dm you the screenshots right quick, link?