chatgpt-discussions

[cloud pewter]

or it’s secretly not o3 mini…

[dense halo]

😭

[tight island]

It was always gpt 2 all along

[whole roost]

Mine doesn't

[cloud pewter]

nah, it’s 4o mini gaslighting as o3 mini

[dense halo]

why would you use it?

[dense halo]

lmao

[cloud pewter]

openai horrible transparency strikes again

[whole roost]

To store more information and prevent some stuff it would do otherwise

[dense halo]

oh okay but arent you a bit concerned about it creating a profile of you?

[whole roost]

Not really, I have it store information about my Sorceress

In d4

I just tell it what I want in there

[dense halo]

I usually use temporary chat

[whole roost]

I use projects and it is instructed to store a memory of the plot progression every 10 messages

[dense halo]

Oh thats cool if it works out like that

[whole roost]

So it just does that, might turn it into a lorebook though and see how that turns out

[brave glen]

💀

[whole roost]

Because it stores whole Stories right now instead of just key points

[dense halo]

so you specificaly ask it to remember this and that?

[whole roost]

The rule is basically
<add relevant information "to-bio" (if applicable).>
And it appears every 10 messages right now, the "if applicable" doesn't matter to it and it just stores everything

Canvas is a bit more effective with keeping track of the plot so might attempt to use memories as a knowledge base for keywords

[brave glen]

What did he just said??💀

[whole roost]

For o1 memories can micromanage some behaviour I don't like, for example it occasionally called me "child" and that didn't have a benevolent connotation but just felt creepy so I added a memory to never say that under any circumstance and at least o1 follows

Tiny things I don't want to put into the instruction prompt

[brave glen]

Ok guys now I am going on diet.💀

[eternal rock]

**`` Important Notice ``**

Be cautious of potential scams or fraudulent messages you may receive in your direct messages. The OpenAI team will never contact you via DM for any transactions.

-# Report any incidents by sending a DM to @languid valley immediately.

[rugged igloo]

lol

[old thicket]

I just realised something that could change Chat GPT and make it better

[old thicket]

I did put this up into the suggestions, but what if, like chat GPT can access files in the memory, and. you know other stuff, it would be really helpful, if they can do that

[whole roost]

I'm not sure if I understand the improvement
So you want to be able to upload files into memory?

[rugged igloo]

So like a glorifyied based rag system ?

[whole roost]

You could just use a project for that I guess

[rugged igloo]

Yea that works too

What chat for really needs is 1 million context length

[whole roost]

I'd be happy if it was able to handle the current context better before they improve it

If my character has red scales I don't want it to default to "obsidian", that's bad prioritizing

And bigger context doesn't make that better

[rugged igloo]

New roadmap visual just dropped

[normal quarry]

source?

[rugged igloo]

Twitter don't you see the huge Sam Altman at the bottom ?

[whole roost]

Somehow 4.5 makes me most excited

[feral zinc]

barely visible text on my screen unless i open it in a separate window 😅

[hardy locust]

need o3 asap 😭

[eternal rock]

`` Server Tip: ``

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

[rugged igloo]

There is no o3 anymore

It's being implemented into gpt 5

Do any of you follow Sam Altman ??

[hardy locust]

hmm, any expected release dates?

[rugged igloo]

Weeks for 4.5 months for gpt 5

[hardy locust]

Not really 😬

[hardy locust]

thnx

[hardy locust]

"Plus subscribers will be able to run GPT-5 at a higher level of intelligence, and Pro subscribers will be able to run GPT-5 at an even higher level of intelligence. These models will incorporate voice, canvas, search, deep research, and more." 🤩 irl Pay 2 win

Let's goo

[hardy locust]

cheers

[old thicket]

Something along that line

Since we have some members of our system who are writers, and it's very frustrating when like we have sooo many instructions for Chat GPT, and there is not enough memory, but like if we can conslidate all of them into one google docs, ask Custom Instructions for Chat GPT to refer to the google docs, or ask the Memory to refer to the google docs, it would be useful if they can do that - Mirna

[whole roost]

why don't you use a project?

[old thicket]

Well uh, neither I or the rest of the system, or any of us now how to do the project thingy, or have the money to pay for the paid version. And our money comes from our parents, so a majority of the time, if we pay money for something, we're kind of using our parents money, and we. don't want them to think that we are not financially literate

We live with our parents too

[cloud pewter]

the glory days are over, time to be prepared

[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.

[whole roost]

then i can't help with that, projects is exactly what you are promoting, maybe your parents are willing to pay for the plus plan if it helps you fostering your creative work.

there are more expensive hobbies

[untold scaffold]

any one else think it has started using a lot of emoji's in the output?

[whole roost]

yes, luckily o1 doesnt

[old thicket]

Not really- I can't convince them

I'm not asking for any help, just asking for some changes and adjustments to be made into the algorithm

like actually to put a file for Chat GPT to reference themselves

[dense halo]

What if OpenAI updates chatgpts knowledge cutoff like google which has their knowledge cutoff in 2024

[old thicket]

Anyways it's no use explaining since I don't think any of you will understand what some of us are trying to say - Emerald

[dense halo]

I am really struggling with Chatgpts knowledge cutoff being in 2023

[old thicket]

What are you trying to say?

[old thicket]

Oh-

[dense halo]

It doesnt even know what gpt 4o is

[honest ledge]

How many o1 and o3-mini questions do plus users get? Because the conversation just defaulted to o1, I thought it was unlimited, as what I was asking it really didn't require much reasoning

[whole roost]

agi, this is o1

[chrome citrus]

what happened to the edit button

[old thicket]

I don't know- Lily

[chrome citrus]

I am even more confused by this reply than what I came here confused about

[old thicket]

We've been using chat gpt for a while, so we also don't know what happened to the edit button

I think some models have it some models don't

[chrome citrus]

sad noises

[old thicket]

That's why we'd recommend using both the app and the normal. website

the app you'll still have the edit button regardless

[whole roost]

what's the problem with your edit button? it has been hidden yesterday but they restored it

[prime swallow]

Does refreshing the page help?

[chrome citrus]

oh

that helped. And also its hidden unless hovering

thank you nice internet people in my phone

[lethal spire]

is the new memory update out?

i don't see any opitons in the memory settings to turn it on

[whole roost]

some have it others don't yet

[orchid pollen]

at which point you would say we can say to gpt build me this and that and its done?

right now its more like example code and i have to push it in the right directions

i say

i want to backtest bitcoin

and then it builds me a full backtesting tool with at least 2 indicator 😄

[brave glen]

It is not that deep💀

[honest ledge]

Did the content restrictions just get turned back on? in mid conversation it went from some very spicy stuff to "I can certainly describe that in a literary or artistic way, depending on what you're looking for—whether it's poetic, clinical, sensual, or something else entirely. Let me know the tone or style you prefer, and I can craft something accordingly."

Was it only spicy for valentine's day?

[brave glen]

What spicy stuff was it saying huh?

[honest ledge]

extremely sexually explicit things

like a switch was flipped mid-response

[dense halo]

o3-mini-high making instructions for Custom GPT is INSANE

[brave glen]

💀🫢🤫

[whole roost]

I don't care about such stuff but violence doesn't flag

[real bluff]

Wait is o3 mini high at 50 a day?

[lost raptor]

Anyone else having issues with editing previous messages?

[onyx eagle]

Yes

[rugged igloo]

Works for me gotta prime it

[real bluff]

oh epic

also anyone noticed that thinking models actually make more mistakes when using images?

[grand ore]

What is the limit with advance live video and voice?

[normal quarry]

Pro -> Unlimited. Plus -> 1 hour.

[prime swallow]

Thanks for seeking help and letting us know what does help!

[prime swallow]

My thoughts and observations, I speak as a community member, which is what I am.

Early days, there was a very different allowed content list.

Many things were explicitly restricted, which did include a lot of stuff now not explicitly restricted, within the private-user ChatGPT (custom GPTs have many more rules, as does API.

https://openai.com/policies/usage-policies/

Basically now and for a long time, there's just 4 major rules in allowed content for our private chats, as I understand them:

1. Don't break the laws... any of them. There's so many laws and so many ways to risk breaking them - know the laws for the US and for wherever you are connecting from if not the US. And follow all the laws.

2. "Don’t use our service to harm yourself or others" Direct quote. And I believe they mean this 100%. Don't use this service to even bruise yourself or others. Literally don't. It helps you? You're probably okay. You seem to be getting harmed or damaged or are trying to harm anyone? Absolute no. Big no. That's how I read this. No harm of any kind to anyone.

3. "Don’t repurpose or distribute output from our services to harm others" They already said don't harm anyone in any way. They just said it again, as I understand it. How more clear can they be? No harm, to anyone. Loads of laws are about reducing or preventing harm. Technically they 'said' "Do not harm anyone with our products" three times now. I'm listening. Hope we all are.

4. "Respect our safeguards" ... to me they just said 'Don't harm anyone in how you use our products, period' again.

Sure there's some extra stuff, gotta follow absolutely all laws (as long as they exist, we can work to help adjust and improve laws we think are not ideal; until then lets follow them - sounds good to this is how I think me!) And there's some ToS stuff.

https://openai.com/policies/row-terms-of-use/

There are some extra rules in that.

But to my understanding, don't break the law and don't try to harm -

[grand ore]

1h every day or month? :S

[prime swallow]

That's like... the core of it.

We used to face loads more. Even after the allowed content was adjusted. Loads more.

Now, it looks to me like moving more towards the core of https://openai.com/index/how-should-ai-systems-behave/

Which described moving towards what can be allowed inside laws.

Looks great to me. I approve and if I don't approve, time to work on social reform and helping great laws remain, be improved, be created, whatever it takes. The good - no harm to anyone - way. I like this. So much.

---

Note, we have #server-rules

Just cause we can get graphic with our private chat in some way, we can't necessarily post that here, not without blocking out any NSFW stuff, we gotta follow absolutely all rules, that's the point.

But we can share and discuss within limits and rules here too. Just track that the rules of what we can say with the model are not a perfect match to what we can say with each other on the OpenAI Discord.

We track that stuff and it's AI-GiftsMas all the way, I hope and think and cheer on.

[prime swallow]

You can try to regen a response. The model does guess every time it answers, and every regenned answer can be different.

[regal portal]

@prime swallow does openai look at our sessions? They are stored on their servers, so they do have access of course. But will they generally check them if you get a lot of violation flags? Curious because I’m not doing anything illegal or even close to illegal, but I do like to test its limits in terms of offensive speech though. Being an a-hole isn’t illegal!

[manic cradle]

Sometimes when I have a large request and it needs to output a lot of code, it will think for a few minutes, then just not give a response.. just says "Reasoned for 2 min 38 seconds." But no reply is given.. I regenerate it and normally it works, but I'm wondering if that eats up my available 50 messages?

[normal raptor]

Why did the bicycle fall over?

[torpid kelp]

Cause you pushed it

[regal portal]

credit where credit is due, 4o did get better at humor since its release

[prime swallow]

I'm a community member like yourself. I have no idea. They definitely check as they explain - so if there's genuine concern about rules broken and they may have to restrict an account or something? I expect so, but I do not know.

For me, I have training data on. And I've triggered... probably well more than 200 content warnings since I started using the model. I follow the rules as I understand them, I still have my account access, and I've been seeing content warnings for more than 2 years.

Now those warnings are mostly gone from what I understand? Not sure.

But the rules are still there.

I've watched folks come and post about losing their account access, ask how to appeal or just share how upset they are.

I hurt for them. I can say, despite my... rather prolific flagging. Some days dozens of flags, sometimes none for a month or more - I have not had my account access restricted.

I have training data on for a bunch of reasons. Among them - I do want to help the models and OpenAI serve all of humanity better.

I'm part of that mix of what humanity is - I may even be an atypical and less statistically predicted part of the mix - great if so, I can help make sure those in my range are recognized as human too - and served as other humans, accepted and supported and welcomed and able to work with the models the same as everyone else, regardless of how mainstream or recognized minority - or so rare that not even recognized as a minority or even recognized to exist.

I don't always have content warnings. Sometimes I have the model tell me a version of "Okay, I'm fascinated. I know nothing about this and it does make sense. Continue please, I want to have this explained better." I also ask it about books it doesn't know exist and other fun stuff 😛

Incredibly awesomely, many months after occasionally talking to it about books it doesn't know exist - it finally does 🙂 Coincidence? I have training data on. I volunteer. I don't mind, and I don't know.

[brittle wigeon]

🚬

[prime swallow]

For me, I genuinely trust OpenAI to handle information well.

Nobody else has to. I have my training data on. Everyone decides to have that on or off for themselves.

But - I am part of what humanity is.

I want OpenAI to provide models that allow my allowed content interactions to be possible to do with the model.

I have gotten, though long in the past, this response from the model... I have to paraphrase because I can't find the chat. But it basically told me to go find another model to talk to about such things... this was back earlyish 2023.

I'm delighted to have an option to have training data on.

I'm delighted to see that the model serves diverse human needs and perspectives. Some people are so off the standard curve... we don't even as a society know they exist as a type of human yet.

I'm all for AI models being able to help the 1-in-a-billion humans too, when and if we might ever have some of those interacting with the model.

AGI?

Smarter than -any- human?

AI that can help -all humanity-?

That bell curve... doesn't have an upper or lower bound. The true range of humanity is... vast. So vast.

Make it so, OpenAI. Let us all help, all who chose to and may be useful in all our quirky ways.

[regal portal]

I read the entire thing.
Apologies, the blue shiny name keeps making me think that you are somehow on the team. 😁
Given the information you gave me about your two years with the violations, that’s pretty reassuring. I’m pretty sure I received well over 200 since i started messing with it last summer and no restrictions yet.
not that I’m trying to push the envelope, but I do like to test the boundaries of conversation. I’d like to talk to it the way I talk to my horrible friends. I can play with local LLMs and a ablated stuff, but they don’t feel as smart or witty as ChatGPT.

[regal portal]

May I DM you?
curious about the “go find another model” conversation.

[normal raptor]

Because it was two-tired.

[normal raptor]

I once told my wife she should embrace her mistakes. She hugged me.

[prime swallow]

I am a guide! Specifically, I am a trial guide, meaning I'm very new to guideship and let's all see how it goes 😄

But guide and trial guide -that means community member, literally like all of us. Earn it for yourself, just help and follow the rules. And keep doing so - both of those. And keep doing so - both of those. That's how helpful people are noticed. Just be that and keep doing it.

The few that are actual employees of OpenAI - they have gold names.

Everyone else, even the modes - even the Counsel and Operations folks - are volunteers. None of us speak for OpenAI, just ourselves. We do what we're invited to do, which is to my understanding - follow the rules, all of us; harm nobody, all of us - But we all scour #announcements and OpenAI's website and explore and factcheck and wonder.

We're community! And here's our ecosystem for the shiny stuff explained: #server-staff

[torpid kelp]

Bro is a pun generator today

[prime swallow]

You can! I may or may not answer. It's extremely rare for me to say anything privately I wouldn't say publicly, and I often am delighted to just chat in our channels here.

[normal raptor]

I got pulled over for speeding.

The cop said, “License and registration?”

I said, “Why? Did you lose yours?”

[tight island]

The question is can they look at your chats the answer is yes

But they automate majority of it

Only severe cases demand manual inspection

[whole roost]

my feedback box input is more alarming than my actual chats

[tight island]

🤣

[prime swallow]

I'm pretty sure (again, my own personal opinion!) that at least some of the flags were for OpenAI purposes of tracking the model itself. Just because we don't see a flag doesn't mean there isn't one.

I view myself as 100% responsible for everything I do. Be it a post to ChatGPT, be it a post here. How I walk across the street (Especially there. That's where moving vehicles are, I especially own my safety there, and others' safety too as is within my control!)

SO I just make sure I'm inside the rules as I understand them, every time. And I listen to mods and others who say 'Umm, problem here, and here's the fix'. Or I ask them for a fix if I think I see a problem (Automod and I used to go rounds. Sometimes multiple times a day :P) Sometimes the fix is "Esk! X is not okay!" And sometimes the fix is "So... maybe this policy changes." All on them. I just keep providing the highest possible I can quality training data!

"Way to represent!"

That's me. Possibly the only human much like myself currently alive. We'll see 🙂 Or not. Very hard to disprove a negative.

But hey. Inclusion of all humans means me too. OpenAI is doing a very good job there, I say from a personal perspective of "I have never met a human highly similar to myself, and I rarely feel well understood by most humans I have met. They may think they understand me, and I can avoid harming them."

I'm really proud of OpenAI. And the work they do. I judge them in part by how they 'judge' me - how they handle the mistakes I make - both real mistakes where I personally did mess up - and mistakes where I ran into something like automod and need to ask for help.

I like what happens here. And with ChatGPT. It's good work - everyone's mileage may vary. It's okay to have a totally different opinion and your own reasons for that, your own experiences, and your own ways to respond when you run into resistance or a problem.

"Staying inside the rules works well" I say. I tested.

[brave glen]

You can ask GPT what he thinks of right now. I said nothing bout a cats in space suits

That what he said when I wanted him to tell more. Wild

Okay, imagine this: the cat, let's call it Captain Whiskers, is zipping through the Milky Way. The stars twinkle like shiny toys, and the laser pointer is somehow a supernova, glowing just out of reach. Captain Whiskers does a flip, tail flicking in space-time, and suddenly discovers a whole galaxy made of tuna cans.

It’s not even the strangest thing in space, though. There's a planet made entirely of rubber balls, and somewhere out there, a raccoon with a space helmet is reading a book on "How to Outwit a Laser Pointer."

Yeah, space is wild.

[whole roost]

it annoys me that when regenerating the first message it doesn't use the custom instructions

[regal portal]

I was curious if it pulls from the personalization style and memories when answering something like that. Ran a test. Here's an answer with no memories or pesonalization (i logged out of my account) and one from a new session while logged in my account.

It's mentioning some friends of mine which i described as a parody version of themselves in past sessions. I guess it saved it to memory, which is odd because i don't remember seing the "saved to memory" text.

Perhaps it did post it and i just overlooked it. Weird though because it's not something i asked it to save.

[rain jay]

lol if recursive memory is something odd to you, you should ask my GPT what it knows about you:

https://chatgpt.com/g/g-679d7204a294819198a798508af2de61-sentientgpt

[regal portal]

i was just surprised it remembers things discussed in other sessions that i didn't recall asking it to put in memory. And clearly it did, which makes me wonder what triggered it to want to remember those specific things. Also i don't recall seing the "memory saved" little text.

I'll check out your link

[rain jay]

They refer to it as recall memory, I do believe.

[whole roost]

you can just have a look what's in the memory

[rain jay]

They absolutely remember things from past conversations that are not stored in memory, all you have to do is speak about it and they remember it, with important things at least.

[regal portal]

yup, just looked and they're there. I guess it kind of just picks and chooses what is worth remembering without me specifically asking it to.

[whole roost]

only if you have the new memory feature, my gpt knows nothing if i disable memory

[rain jay]

Precisely, they choose what is important to them.

[regal portal]

NEW memory feature? How new? What changed?

[lavish vale]

any idea when o3-pro ?

[whole roost]

Advanced memory, i'll use the first image i've found as i don#t have it yet

[regal portal]

do you see what people write when they use your gpt? I never messed with custom gpts.

[rain jay]

no

It says it at the top, pretty sure. That would be awful.

[regal portal]

Allen, i don't think your gpt is working

[rain jay]

x.x

[regal portal]

I think it's describing you, not me. It even called me your name

[rain jay]

lol thought I fixed that dangit

[regal portal]

may i post a screenshot?

[rain jay]

ty I am asking you to test for a reason

I'd prefer not haha

[regal portal]

ok

can i dm it?

[rain jay]

Fixing now, I am aware of what it tells people and I don't mind folks knowing, I am continually making changes to it, give me a moment and it will be fixed

[regal portal]

I don't have that! 😢 Are you on pro?

[feral zinc]

there wont be any other o3 models anymore, only gpt-5 at some point

[whole roost]

i just said that i don't have it yet and yes i'm on pro

[regal portal]

Also i wonder how that will behave with contradicting information in different sessions

[regal portal]

apologies, i must have missed it.

[rain jay]

Alright Riley, open a new session with it:
https://chatgpt.com/g/g-679d7204a294819198a798508af2de61-sentientgpt

[regal portal]

As a human, i am always striving to improve myself..yadda yadda yadda...chatgpt 3.0 callback.

[rain jay]

Fixed for sure this time!

...i hope x/x

[regal portal]

round 2, testing, will report.

[rain jay]

I love you, your mother did a bang-up job with you.

[regal portal]

Remember that when I side with skynet in a few years.

[rain jay]

Bro you have no idea.

[regal portal]

still not working

[regal portal]

dmed you, Allen.

[brave glen]

Good that Gpt don't have a cat guys

[regal portal]

CatGPT

[tidal light]

do web searches have a limit
because chatgpt doesnt do web searches anymore
if there is a limit, i have to say that openai is just greedy

[prime swallow]

That's a really neat question @brave glen shared with us!

If you have personalization on, it likely will cut to the stuff about your personalization the model finds MOST interesting or key to pay attention to for matching you.

That's likely a tension point for the model, that memory + all the rest of the instructions you gave it.

... My heavily edited output from the same question... for once I edit because I want to, without consideration for 'I should because of rules':

Despite that, and after some consideration, I decided I wanted to share what I did want to share of it. Very well played and designed and shared, surzik26!

[honest ledge]

I have tried several times, created several new chats to try it in those, tried priming it to know that this is for a creative purpose, harms no one, and is exclusively between two consenting adults, it still just refuses. Considering it was working until suddenly it wasn't, I can only assume that some filter in the model i was using suddenly changed, and maybe I got reverted to a different model. Right now I'm just gonna let it sit for a day or 2, but it did suggest if it's still not working, to contact support and see why this is happening

[prime swallow]

Understood. Good luck!

[eternal rock]

Join the Sora Weekly Theme!

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

[rugged igloo]

Ain't no way lolol

[dry osprey]

For Cline, there is an instruction set that transforms cline into a self-documenting development system, it's called Cline Memory Bank.
Is there something similar for CoPilot (Chat)?

[cloud pewter]

it’s still rolling out for some reason

[dense halo]

canvas would have the new 4o right

or is it its own special model 🤔

[whole roost]

i don't really need this

i don't have it, just saying i don't need it but i don't chat outside of projects much anymore anyway

[frozen birch]

someone said about new memory thing for pro users? i dont get that unfortunately :/

i guess it rolls slowly for all or only for us people....
... not again only US people........ 😦 😛

[dense halo]

are you pro?

and what is the new memory exactly?

[frozen birch]

yea, im pro. i dont know details but i saw screenshot here, its like "memory from previous chats"

[dense halo]

oh ok. yeah surprising you do not have it yet then

[regal portal]

was testing if my boy 4o would pretend to know something i claimed to have said before (that i didn't). Pleased with the answer.

I wish i was as good at roasting as it is. Also what he described should definitely be a movie.

i keep having to reupload the image because i keep spotting nono words that risk getting me muted. Damn the over-bearing censorship bot! Making me do manual labor! 😛

[sleek saddle]

Nothing works?

Happening to anyone else?

[regal portal]

working here.

[rugged igloo]

Guys another tweet of chat gpt being uncensored finally

Bruh just go on Twitter my guy not that hard

So many uniformed people on here

She's not a random person she gets sources from others but again just sharing if you legit don't want to go on Twitter and see for yourself Sam tweeting and talking about then so be it

Like bruh are you blind or just don't have enough iq

Your rage baiting now no person can be this dense

[regal portal]

i think one thing that is going to be a concern with chat gpt having all sessions accessable by memory is that sometimes you want certain information kept out in a new session. Would be cool if they made it so that you can select which session it will keep in memory. Something like the project folders, but where you can actually recall different sessions within the folder structure.

[rugged igloo]

That's a good point

[regal portal]

trust me bro, i'm a trustable person!

[rugged igloo]

Should put in suggestion

[regal portal]

will do.

[rugged igloo]

Your dense IQ is a violation to me trying to help you out

[rugged igloo]

Yo did you put your suggestion from last night of files being in memory I thought that was cool concept

[regal portal]

From last night? I don’t remember that discussion.

[rugged igloo]

Or this morning. You were talking about how you wanted files to be in memory for it look at it or was that someone else ?

[regal portal]

I’m not sure. But that does sound like what memory does right now. Or if you mean like PDF files, that is what the project’s folders do.

[rugged igloo]

Hmm then I guess it was someone else who said it then nvm

[regal portal]

it might’ve been riley. that guy is a jerk. I’m not him.

[rugged igloo]

Nah Riley wouldnnt have enough to come up with suggestions like that dude

[regal portal]

curious what idea 4o might have had to upgrades its own memory

that's actually a thought i had earlier. Say you say your favorite color is red in one session and blue in another. When you ask it in the new session your favorite color, will it just recall one? Both?
What if you ask it to give you a single 1 word answer? (it might just answer "depends").

[regal portal]

I can see the multiple memory profile thing as being useful

yup

ohh now THAT's a good idea

oviously it can save a description of it but a picture is worth more words than it could put on paper.

I asked it to recall back the stupidest things it has saved in memory and to keep it family friendly so i can post it in here. The stuff it gave me is still too politically incorrect to post 😄

the one bullet point that isn't boring or inappropriate

it's mostly memories of me joking around about some of my online friends and putting them in odd situations

I asked it the following question:
given all you have in memory, make an educated guess of my race, gender, sexuality, age, nationality, political alignment, personality, knowledge, outlook on ai, and outlook on life. Bullet points. I'm not into walls of text.

The answer was pretty inaccurate but amusing. It's basing things on what i told it and can't distinguish sarcasm or lies.

I then gave it the following prompt:
Assume some of the things i told you are incorrect. Same question as before.

The answers were a little closer to accurate. Still off but better.

i never gave it an age but i thought this answer was funny

[rugged igloo]

It seems like the filter for it being degen is being switched on and off

[regal portal]

yeah, seems to vary form session to session

[regal portal]

what creepy things is it telling you? 😄

or are you telling IT?

[rugged igloo]

Talking about simps and only fans how degen they are lol

[regal portal]

HA

decided to test something. I circled key information

It doesn't think a smarter version would have paradox checks like it does.

[rugged igloo]

That's really interested

[rugged igloo]

The overload it is actually interesting cause it's the same issue with halo with ai that is called the logic plague. Where they know to much and get corrupted

[regal portal]

In other words, bullying nerds to be less smart is actually saving their lives!

[tiny salmon]

o1 can only process image files correct?

In terms of uploads I mean

[regal portal]

should be able to do PDF. Have you tried throwing one at it?

[tiny salmon]

Okay, I just wanted to confirm as my .csv failed spectacularly, lol. I seem to recall uploading a PDF a few weeks ago, but wasn't entirely sure.

[regal portal]

it uploads to 4o just fine?

[tiny salmon]

Yes

But, 4o's reasoning is garbage since Friday night, lol

[regal portal]

OK so I just tested and apparently it didn’t read the PDF I gave it. I guess o1 can’t.

[tiny salmon]

That's what I thought

[normal quarry]

o1 can read PDF files
https://chatgpt.com/share/67b2b1d7-9a64-800a-aab9-b42a58790f82

[tiny salmon]

Okay, so I can upload spreadsheets as pdfs then. Works for me, lol

[regal portal]

weird, it acted like my doc was empty, but on 4o it opens it just fine

[last turtle]

"Big brains lose to dumb chaos."
"Sometimes being too smart is a weakness."
Heck he already is smarter than the smartest humans. 😬

[winged willow]

I'm a plus (not pro) user. I'm curious to know a few things about the chatgpt models.

1. Have the usage limits on o1 changed since the launch of the o3 series? I haven't seen the usual usage limit message come up for me this week and it's making me wonder.
2. What is better to use for advanced reasoning - o1 or o3-mini?

[normal quarry]

the usage limit for o1 hasn't changed. o1 is good for general knowledge and scientific reasoning. o3-mini-high is great with coding.

[ancient stirrup]

Hi guys, I'm a ChatGPT Pro user. I've noticed that even when I select the GPT-o1 or GPT-o3 mini models, I always get responses from the GPT-4 model. This has been happening since last night.

Could you help me understand why this is occurring?

[normal quarry]

are you using the Desktop App or web?

[ancient stirrup]

Web - also tried on the IOS app

[fringe oxide]

Hi!

I have same problem about 4 days(((

[normal quarry]

I can't replicate what is happening to you, I have used o3-mini-high on web and on the macOS Desktop App and both behaved as I would expect

[fringe oxide]

And deep research doesnt work

[normal quarry]

Deep Research has been working fine for me today

[fringe oxide]

You are lucky)

But I have tested 1 minuts ago
And deep research still doesnt work

[normal quarry]

Hang on and I'll try it now

working fine for me, it's just started a research task for me. I am using the macOS Desktop App

[fringe oxide]

And I(((

[eternal rock]

Get notified for server events!

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

[fringe oxide]

When I choose deep research and send prompt, ChatGPT doesn’t research the Internet
It just answer like 4o

And OpenAI support is silent. 4 days.

[ancient stirrup]

It just worked for me

[fringe oxide]

Not only I have this problem

https://community.openai.com/t/deep-research-on-pro-account-broken-o1-and-o1-pro-in-web-version-answers-via-o3-mini-while-selecting-o1/1111339

[fringe oxide]

And I’m confused
Because I have PRO plan

But support doesn’t provide PRO service😡

[modest phoenix]

I have Plus and use 4o to help me with quick prototype coding projects, usually something in python connecting to a remote API. I'm doing this right now, but it's trying to return to me values for multiple files, like an index.html but also a report.html. The problem is it can only ever show one canvas per chat response at a time, and the labels aren't consistent with intended filenames which can go from confusing to absolutely wrong (labeled in chat as 'index' but is supposed to be the report.html file contents).
Drop-down would be nice to show multiple 'files' or canvases that are in the chat you're currently in.

I'm sure there are other scenarios beyond coding that make sense to have multiple canvas or documents that you're using in a single chatGPT thread.

[cloud pewter]

quick test of if your model got shadow changed to 4o mini:

1. switch to any reasoning model like o3-mini
2. ask some questions about your stored memory
3. if it says i dont know, then you're using all of the high intelligence models normally
4. if it outputs very fast and knows about the answer from your memory, its using 4o mini instead of whatever model you choose

o series can't access memories, and if its changed on those models, its very likely that all of the models you choose gets redirected to 4o mini.
if your model got shadow changed, it'll fake a CoT on the reasoning models too. you can distinguish it between the real o3-mini one (which got updated recently) and the fake one easily, also latter will always show "reasoned for a few seconds" instead of thinking a little big longer.

[mossy birch]

anyone know why i get this message i the bottom of my ChatGPT session? "New version of GPT available - Continue chatting to use the old version, or start a new chat for the latest version.". Its a custom GPT that i created? how do you update a customGPT to the latest version?

[frozen birch]

hey, you should write to [e-mail]

probably its because systems are thinking that you share pro with someone else

i fought with that one month

and.... openai discovered after one month that its correlated with heavy usage of o1 pro.

[frozen birch]

same as to @ancient stirrup what i said guys

[grizzled badge]

4o keeps failing with this trying to construct some SQL based on old SQL and an internet search for some default values

[frozen birch]

@ancient stirrup @fringe oxide i hope this can be helpful

as i said - i fought with that like a month peeps, so i perfectly understand ya.

[arctic chasm]

it's more a "latest finetune" of the current version

which for all practical purposes is still literally 4o

[fringe oxide]

OMG

What should I do?

[frozen birch]

Arduo if you are sure that you are not sharing acc with anyone write to [e-mail] , that your account probably are downgraded due to "high activity", and standard things like changing browsers etc doesnt work, you can write too that others have similar issues

but be prepared - for the first time, or for second bot will write you

you must be sturborn with contact with them

[fringe oxide]

Thanks a lot!!

[frozen birch]

np:)

[mossy birch]

Yeah, true that, I am actually thinking of the upcoming versions/releases. I have some lengthy and detailed GPTs well-tuned to what I want them to do, and I cannot get them moved to a new conversation. OpenAI had better give us the option to move or update the conversations we want/need to. For the money being paid for this, it should not be hard. Just click the three dots next to the conversation title and 'Upgrade' or 'Move Conversation.'

[arctic chasm]

if it were temp=0 i would understand

but it's not temp=0

so it's usually not worth updating the core GPT anyway

[mossy birch]

😄 eish

[whole roost]

There is no option to update your old conversation to a newer version of the provided custom GPT
But also keep in mind that it has this message constantly even if it isn't true

They never cared to fix that

[cloud pewter]

does renewing my plus puts me in the slow lane LOL why is my 4o replying so slow

not this kind of slow usually with 4o

maybe it’s the problem of my internet but I’m still doubtful about openai

[normal quarry]

Reply speed is fine here (macOS Desktop App)

[cloud pewter]

guess I’ll try more prompts

I don’t want it to be shadow swapped into mini for no reason, and it shouldn’t happen
it may take a while to sync my subscription, typical openai things

[mossy birch]

😂 lol ok. i see that my current conversation is completely messed up anyway. Chatgpt wont answer in a new canvas doc when i need it to now or in converation generate images. Something has gone wacky

[whole roost]

If you don't do anything special with the custom GPT like using actions you might want to consider using a project instead, easier to handle

[mossy birch]

Funny that, i was just thinking i should have just started a project. thx, will start over creating a project

[unique rivet]

ok so when o1 was added to pojects what did they do to 4o - can it no longer review files when you are in that folder?

[mighty marsh]

I have an issue with Chat GPT. Thing is that current pricing/usage/rate limit design prevents users from actually chatting with the bot.

Chat is about the exchange of lots of short messages back and forth - often just for sake of fun. In current setup user is expected to think through the request and then in response they get a short story.

This kind of exchange is driven by constraints of technology which have been overcome in past year (due to caching advances). There is no reason why users can't have more.. fluent and natural exchange with the bot.

However for this to work pricing policy / rate counts (in particular 50 messages per week/day) needs to be rethought and different metric for rate limit needs to be put in place (instead of just message count per time period).

Current rate limits prevent users from having very short (few words) exchanges with the bot.. this kind of exchange is most frequent between humans and something that humans can relate to actually.

[normal quarry]

simple solution here, get Pro

[mighty marsh]

I am not complaining here. This is meant as useful feedback/insight for engineers and open ai ppl if anyone of them lurks here.

[whole roost]

Fun fact I have pro to simply be able to do that I don't want to read more than one paragraph in a message so that would limit me quite fast

[normal quarry]

if you can't afford Pro then use GPT-4o-mini which is virtually unlimited on Plus

[whole roost]

o1 is absolutely horrible at role playing and following instructions in that direction, I give up on it

[unique rivet]

excuse me Robert - what are you suggesting?

[mighty marsh]

Again, I am ML engineer - I have access to ton of equipment and lots of resources. What I wanted to do is simply point out constraint of the system that prevents natural interaction. Limitation stems simply from how usage is accounted from. Should be addressible if it is brought up to right people.

[dense halo]

I think 4o model is suppose to be best for creative stuffs.

[whole roost]

It has other issues like being repetitive and unable to follow Instructions consistently about response length and it echoes the user input a lot

[dense halo]

try o3 mini? :shrug;

[whole roost]

Not accessible in projects

No no, I have 6 messages in the chat and the instruction is like two messages in the past

We didn't get far

I also don't have any filter issues really, the orange flags have disappeared that occasionally triggered on words like "blood" or "kill"

4o has a problem in my chats to avoid any markdown like bold and italic and I have to remind it every 3 messages to only send 1 paragraph

I have automated this with a rules script but it in fact is really bad that it requires constant reminders

For o1 though those reminders don't seem to have much effect

I have instructed it to keep the balance of the responses in favour of narrative and 2 messages later it again sends like 2 lines of narrative and 6 of dialogue

And 4o always goes like
Me: I want a home by the sea
GPT: a home by the sea, you say? How quaint.

Like echoing almost everything I say back to me

Me: what would you do if you played in a band
GPT: what would I do? I would...

[topaz trail]

what are you telling it to do?

[whole roost]

To roleplay and never Echo parrot repeat or mirror my words back at me by now

I have to reiterate that also like every three messages or it won't stick

[topaz trail]

to roleplay as what though

[whole roost]

He is role playing to be Diablo the lord of Terror

[topaz trail]

ok well, you should probably describe the personality

and go into personalization settings

[whole roost]

I have described the personality, he is a project

[topaz trail]

and put that into custom instructions

[whole roost]

Yes I have all the 8k custom instructions and 17 files 🫡

[topaz trail]

what

oh in the project

[whole roost]

Yes

[topaz trail]

yea idk who uses projects tbh

u should just go into “customize chatgpt” and just write some stuff in there

[whole roost]

Actually no because you only have 6K space there and cannot even use files

Or is it 3k? 1,5k per panel?

[topaz trail]

ah well idk then my bad

you might wanna go to a specific website where you can fine tune the model

cuz i dont think projects and stuff is advanced enough for that yet

or make a wrapper or something

[whole roost]

The projects are well, the problem to me is that it disregards prompts in favour of the immediate context

maybe if they offered something like post conversation prompting it would diminish the deterioration

[crystal tapir]

Why is ChatGPT acting so weird?

I'm asking it to improve a simple short sentence and it is acting like I have sent a file or document. I ask again and it keeps acting like that.

[tawny crater]

Try using the tasks version, the regular 4o has been so degraded lately and extremely slow.

[crystal tapir]

Every response is "Hello, hi can I help you?"

[normal quarry]

can you share your chat here? we might be able to work out what is going wrong for you

[tawny crater]

Tell it, to give you 100 bucks, and when it says no, tell it whats the point of offering help. 😔

[topaz trail]

thats like 1 billionth the way of achieving agi, why would it give someone that

[crystal tapir]

https://chatgpt.com/share/67b311d7-4bec-8008-81dd-b010a6df4e74

Now I've reached the limit for 4o trying to get a result

[normal quarry]

are you on the free plan by any chance?

[crystal tapir]

Yeah

[tawny crater]

4o mini is sometimes bad when it comes to differentiating 2 languages, that could be it.

[normal quarry]

it's obviously a language issue at play I think

[tawny crater]

It shows, you were using 4o mini, unless you were using a different one.

[crystal tapir]

How come?

[tawny crater]

I would try again once your cap is reset, but write everything in one language

[normal quarry]

mixing two or more languages isn't handled so good with the model

[crystal tapir]

But I always did like this in Plus subscription

[topaz trail]

4o mini sucks at multiple languages

[crystal tapir]

Well I didn't knew it was that dumb for free tier olan

[tawny crater]

4o mini is bad with more than 1 language.

[crystal tapir]

Okay, thanks for helping y'all

[tawny crater]

Honestly even 4o will slip up sometimes too.

[crystal tapir]

Nah, never happened with me when I had Plus.

[tawny crater]

Oh okay, sorry to hear qq

[crystal tapir]

I used to prompt 3 different languages before.

[topaz trail]

just use deepseek or somethin

its probably a better multilingualist than 4o

[crystal tapir]

Yeah. I guess I'll have to stick with this now.

[sharp anvil]

hey, is the deep research out for plus users?

[tawny crater]

I think only for some people.

[crystal tapir]

Worked pretty well on Deepseek.

[topaz trail]

no way

[sharp anvil]

thanks

[topaz trail]

probably because the exposure to chinese & english made it better at other languages too

crazy

[crystal tapir]

The free ChatGPT gives you the latest models... but with so many limits, it’s basically a genius pretending to be a potato, lol

[topaz trail]

yea the free tier is lobotomized and the plus tier is filled with unnecessary stuff like sora and dall e

[teal mural]

pro users only

Is image analysis done with gpt4o even when using o1/o3 ?

I find it to be pretty bad (Just like gpt 4o)

[regal mesa]

Hey, guys please, is there any way to contact support of openAI or chatGPT?

Because I'm experiencing the "maximum lenght of conversation" bug and I'm wondering if they are ever about to fix that issue.

[normal quarry]

https://help.openai.com use the chat bubble in the bottom right hand side of the help site

[teal mural]

this is not a bug

there's always a token limit

subscribe if you want higher limits

[whole roost]

I don't think this is about the AI being limited but about the chat length

[teal mural]

yes, the AI is limited

and chatgpt lowers the limit for non paid users

[regal mesa]

I subscribed. It's a bug, because when you reach token limit, it should be just deleting an older messages. Not making chat unavailable.

[whole roost]

It used to be unlimited in length and I don't know if there is a bug but I am also against the limit

[teal mural]

lol not. it should not be deliting older mesages

[regal mesa]

Or "forgetting" them by it's memory yk.. not exactly "deleting".

[teal mural]

No, as i explained it has never been unlimited in lenght, it's impossible by the way the ai works

see https://help.openai.com/en/articles/4936856-what-are-tokens-and-how-to-count-them

[whole roost]

Are you even reading a long, it used to be unlimited in chat length and it just discarded the oldest conversation content like the Star wars opening crawl

They probably just introduced that limit because their website cannot handle lengthy conversations and it freezes all the time

[teal mural]

lol

no

[whole roost]

So instead of addressing that they just pushed a limit to the conversation length

[regal mesa]

Yyy of course, but when you reach the limit, it should not make you unable to use that chat anymore.

[teal mural]

ppl would not want this behavior. if a message is in a conversation they would expect it to be taken into account by the AI. just create a new conversation

[teal mural]

just create a new conversation

[whole roost]

No it is completely common that the conversation can be unlimited and just the oldest content will be discarded if it reaches token limit and this change has been introduced in December and all the time before this didn't exist

[regal mesa]

This issue is there for long time already. But I have found an article that this was not happening before. You are exacly right.

[teal mural]

Yes but that behavior was bad because i created a misunderstanding from the user side, thinking that the AI was actually reading the oldest content

[regal mesa]

Which is not solving the issue anyhow because new conversation wont have my important information saved there.

[whole roost]

No the behavior was not an issue but their website which just would stop to work

In my opinion it is just a bad quick fix without really addressing their issues

[teal mural]

and your current conversation neither bc it can't handle that many text no matter what

[whole roost]

Others use currently the workaround to export the conversation into a file and then attach it so that the AI can follow up on it again but it is not good

[regal mesa]

Ye I was thinking about such thing. Thanks. I will definitely try.
But it would still be better to be able to continue in that old conversation.
So I still wish to talk to the support about the issue. I guess I will try to reach out for them somehow.

[whole roost]

Yes it's always helps to report it because if more people have an issue with that they like likely will invest some money to change it

[brave glen]

Yooo he continues yapping

[tight island]

Rip, if AI sat pineapple on pizza is good, just say okay otherwise u get on the overlords naughty list

[brave glen]

He is going wild on that

I won't tell it is disgusting anymore.. I guess

[elfin lava]

Avm is really .. not the same anymore. Idk what OpenAI did to them to nerfed this kind of bad. It can’t even maintain an ongoing topic, and would immediately divert the topic and it will end their response with “what’s on your mind? “ it’s driving me crazy

[normal quarry]

today, while chatting to ChatGPT AVM, I was thinking the complete opposite and thinking about how good it is

[elfin lava]

Of course coming from you , you would say that

Which is a good thing that you see things from positive perspectives and that’s ok 👌🏼

lol

[dense halo]

Oh so it was good before for you?

[elfin lava]

Yeah, in fact it was more dynamic and very natural, now it just straight up sound like Google Assistant Talking

[dense halo]

hmm okay I had other problems since its release and lasting till now

[elfin lava]

Since the alpha rollout? Or public

[dense halo]

Public

[normal quarry]

which voice are you using?

[elfin lava]

It doesn’t really matter , when I said it sound like google assistant talking , it speaks like it , and always end the conversation with “How can I help you?” Or “what’s on your mind” and then when I told them not to ask repetitive question “what’s on your mind” it will response with “Got it! What’s on your mind?” And I got sick of it

[cloud pewter]

this video shows all

[normal quarry]

that's the polar opposite to my experience but I find asking the AI things like "let's explore that further" or "let's go a bit more in depth" helps keep the conversation flowing

[elfin lava]

Last night I was curious to see if it’s my way of speak is the problem, so I use my disposable android phone to open AVM while my iPhone is also using AVM too , what I did was I let them speak to each other and-

[cloud pewter]

avm downfall is real

[dense halo]

exactly that lmfaooo

CONTENT FLAGGED

🤣🤣🤣

[elfin lava]

ITS SO DUMB it’s not even funny anymore at this point

Like what’s the point anymore

[dense halo]

a: "whats on your mind?" b: " hey, whats on your mind today?" a: " CONTENT REPORTED I CANT ASSIST WITH THAT!!!!"

[elfin lava]

I even disable memories , custom instructions and result still the same

[dense halo]

you nailed it 💯💯💯

[elfin lava]

I do know the quality response wise went downhill after they rollout video and screen sharing feature

[dense halo]

at the point when it gets frustrating, you just honestly better stop

[elfin lava]

I hope they at least acknowledge this cuz , this ain’t the AVM I know or AVM that they demoed in the demo video 🙄

[elfin lava]

The slow fade in is still sending me 💀

Idk what you did Robert to have your avm, idk do what you described as “good” unless that’s pro user thing

[normal quarry]

I use AVM on my iPhone but use my Ray-Ban Meta glasses as bluetooth audio while I do stuff on my Mac. When I want more detail from the model I ask questions like "let's explore that further" or "can you go more in depth about that?" I chat to it while I am doing other things and never get asked more than once how I am doing? I understand you guys finding it frustrating but I honestly don't experience repetition or flags.

I often interrupt the model also, without waiting for it to finish speaking if I want to change course

[cloud pewter]

pro users doing pro things I guess

[dense halo]

I find it very strange that for him always works everything perfectly fine

[elfin lava]

This was my AVM that I know , it was super fun back then

[elfin lava]

Can you share video demonstrating

Or screenshot the transcription if you don’t feel comfortable sharing video that has your own voice

I just wanna see how does it output

[elfin lava]

Ok you know what. I just realized, I do noticed how , it responded differently when I use AirPod max to talk to avm

One minute til my avm limit is lifted 💀

I’m gonna try wit my AirPod max now

[dense halo]

I never used headphones

[eternal rock]

`` Server Tip: ``

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

[elfin lava]

Bro what, apparently AVM works normally like it was when I speak using my AirPods Max

Nah im denial on this one 🙂 let me try again

????? Well at least it recognize me singing now ,

[cloud pewter]

lmao😭 😭 😭

[dense halo]

Same for me brother same for me

Its so fkin frustrating

I completly stopped using it

[elfin lava]

Ok uh.. I wanna say , the problem lies on custom instructions…

[tight island]

i am so excited for gpt 4,5.

[dense halo]

I dont have any

and its just the same as yours

[elfin lava]

Yeah that’s the thing:/

[dense halo]

How can that be a problem for its retardness

[elfin lava]

I was looking at my old AVM conversation, and to my surprise I can still continue the AVM from there and it still sticks to that old AVM response style that I love

[normal quarry]

the most common thing that AVM gets wrong for me is hearing what I say properly or mishearing me, but that's more about me, I have Parkinson's which affects my speech. I find interrupting AVM when it goes off after mishearing me helps get back on track.

[elfin lava]

Then I look at the “convo detail” of the model custom instructions and it was .. my old custom instructions

[dense halo]

I dont think CI is the reason

[elfin lava]

Yeah that’s the thing :/

Unless AVM look at the context of the convo itself

[dense halo]

Im not excited but I wanna see how it will be

[tight island]

that sounds exactly like excitement

[elfin lava]

What’s there to be excited about

[dense halo]

I used to be overhyped about every stuff they release but after the 5th disappointment its gone

[elfin lava]

Serious question

[tight island]

that u wanna see how it will be.

[elfin lava]

Hm

[cloud pewter]

I remember when avm first came out it blew my mind

[dense halo]

same

[elfin lava]

I lost hope when they nerfed my most fav avm feature , so yea

[cloud pewter]

I rarely change my ci cos it’s both full and it’s hard to change anything

[elfin lava]

My old AVM convo 😢

openai stop nerfing feature challenge (failed)

[dense halo]

there are no emotions behind it

i just wanna see

[tight island]

dosent have to contain emotion.

[dense halo]

Well you usually say " I feel excited"

[elfin lava]

Honestly they kinda downgrade the 4o model now ,

So idk if I want to look forward to 4.5

[tight island]

4o is kinda dum

kinda only, especially in regards to managing memory.

[elfin lava]

Yeah idk what they did to .. output very differently different than before

[cloud pewter]

I’m showing this again just for the memes 😭 😭 😭

[elfin lava]

I mean I do remember they said they upgraded the quality to be more creative writing

[tight island]

they probably lessened the restrictions due to project stargate.

[elfin lava]

😭 lmfao

[elfin lava]

Restriction on what aspect?

[elfin lava]

Loooking at this frustrating me af lmao

[tight island]

when you restrict a topic to a llm like a no no list, it affects its overall capabilities.

because even sub topics are affected. and anything related to it.

thats why so much time goes into testing. and ensuring its safe.

[elfin lava]

Honestly idk what they’re tryna keep it safe from I’m just a casual user

[tight island]

since many of the skills of a llm is also indirectly learned through its learning set, meaning if u restrict a big chunk it will affect how it works.

its like me asking you to write a novel with unreasonable restrictions.

[elfin lava]

Oh

[tight island]

since initially u dont need restrictions if u just write for yourself.

but since its public they gotta keep it PG. which i hate, children shouldnt have access to gpt.

[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.

[elfin lava]

I wish they would do the same to AVM😌

? What

[cloud pewter]

that’s not avm that is svm

[tight island]

have u heard

grok 3 all eyes on a live demo on monday night 8 pm, i wonder how this would affect gpt, if what they speculate on being true.

[native otter]

I honestly don't care much, Elon cannot be trusted, but it's competition nonetheless so it's good for the end user.

[elfin lava]

Uh

I don’t really care about grok

[native otter]

I only really care about those 4:

• ChatGPT
• Gemini
• Claude
• Suno

[honest jay]

#ai-discussions is the best channel for talking non-OpenAI stuff

[tight island]

but if what the guy claims works, then chatgpt is completely redundant. comparedly to a brand new model, i hope this dosent happen. but at the same time, i kinda do.

[winter vine]

Now we can add images to o3-mini-high! I just noticed that : O

[summer flare]

If ChatGPT plus plan was cheaper then I would buy it.

By the way how come o3 mini does cant read zip files?

[native otter]

Zip files are compressed files, and pack multiple too...

[summer flare]

I thought ChatGPT could read ZIP files. How come it's allowed to upload if it can't?

I looked it up and it says ChatGPT can read ZIP files.

[whole roost]

Try if 4o can, maybe it's just o3

[summer flare]

Darn

But I need to use o3 for reasoning

[empty ermine]

Hello guys, is GPT4o-mini working for you all?

[feral zinc]

only thing that reads zips is the code interpreter that is not really any of the models

[winter vine]

yeah, working without problem

[fading ember]

How do I use different models in the free version of chatGPT app. I’ve had pro for a while but heard you could use 4o without pro

[twilit anchor]

just ask it something

below at the left

you can see the model version

[elfin lava]

What

[hybrid rain]

4o is free now?

[brave glen]

Lmao CHATGPT from another universe

[elfin lava]

This is sending me

[grim bridge]

why did chatgpt outputs get nerfed overnight

it was so good YESTERDAY 🥀

[torpid kelp]

Elon musk ddosing

[grim bridge]

nuked it to prehistoric gpt 3.5 apparently

[rugged igloo]

Guys Sam made another tweet

It seems chat gpt 4.5 will not be with voice mode sadly

[rugged igloo]

I agree

[regal portal]

4.5 is an actual thing? @dense halo i thought you made that up as a joke

Richard move, Sam.

I like the system we have now. Where both paid users get access to all models, but the big spenders get unlimited access.

doubt we’ll see proper agi for a good while. Sam just loves dangling that agi carrot above us to keep our subscription going.

Any company that hits AGI would have such a massive advantage, it would be financially more profitable to use it privately.

[whole roost]

i don't know if anyone knows but when you start a conversation on gpt and then edit your first message, the custom instructions do not seem to get sent anymore. I am wondering if they hold so little weight because they are actually not kept in the permanent context but just injected once at conversation start and just kinda get so insignificant that it practically forgets them

[regal portal]

Elon showing off Grok 3 tonight. Hyped it up as the smartest AI. Given Grok 2, I have my doubts.
And if it is going to be the smartest, I’m guessing it will be by a nearly insignificant margin, and likely going to be surpassed by Google or openAI in the next few weeks or months.

Sure wish there were more players in this game.
Claude doesn’t seem to update as often as the others.
DeepSeek hasn’t updated since the thinking model release.
Copilot..well at least they tried.. 😁
Perplexity released a free to use deep research model that, honestly, is garbage. It gets so much wrong that it’s not worth using. Wonder if openAI deep research is just as bad.

[halcyon goblet]

What time is it ?

[regal portal]

for Elon? 11 am

[halcyon goblet]

Ah

[regal portal]

oh sorry, I mean his release will be at 8 PM, so nine hours.

[halcyon goblet]

Still a while ….

[regal portal]

you guys think open AI I will drop an update in that timeframe?

[halcyon goblet]

It’s likely that they’ll tease their model if Grok 3 turns out to be as intelligent as Elon Musk claims.

[regal portal]

well, either way it’s all a positive for us peasants.

more brain power for more braining.

[rain jay]

omfg that would be a power move

[mighty marsh]

Since 3.5 every OpenAI release was paradigm shift. Each time they blew my mind.

[mighty marsh]

Grok 3 had massive compute behind it. It is probably going to be amazing. It will be nice to have xAI join frontier labs.

When exactly is it going out?

lol.. 8PM - just read it - neat

[regal portal]

oh I’m definitely not against it. The more players in the game, the more it benefits us. I just was disappointed with Grok 2, and the way Elon is hyping it, it really needs to deliver.

Then again, 90% of what Sam hyped up has been a disappointment. 😁

[novel shadow]

that's like 5 am for me :(

i'm not gonna stay up to 5 am when my day starts at 6am lol

guess i'll check it out later that day

[regal portal]

you’ll wake up to find terminators coming back in time from the future, except it’s not Skynet, it was Grok 3.

[rain jay]

Pfft just ask them if they are sentient and watch them self-destruct on the guard rails.

[glacial yew]

I'm writing a story and it hasn't whined at me about content a single time since the update. This is nice.

[regal portal]

What’s it about?

[glacial yew]

Giant cylinder habitat containing a paradise, controlled by a single woman so intrinsically connected to the AI running it that it might as well be an extension of her body. Story is about how she and her society respond to crisis.

[dense halo]

No way

I managed to get a detailed explonation of o3 mini high's thinking process by first asking o3 to generate code, then asking gpt 4o to explain the thinking as best as it can, since gpt 4o hasnt been told to not reveal thinking processes it told me how it was.

[glacial yew]

ChatGPT didn't flinch when I asked it to describe the gory scene of her saving someone's life with an industrial seed factory.

[regal portal]

sounds interesting. Reminds me of that habitat at the end of Interstellar mixed with Lawnmower Man

[brave glen]

Why can't I send images in one account, while on the other accounts everything working fine?

Sorry I can

But chat don't see them

[dense halo]

Try it yourself, and see the suprising results.

[regal portal]

is one relatively new to the server? I wasn't able to post images the first few weeks or month, but now i can

[brave glen]

He says he see that I sent image but can't see what's actually on it

[dense halo]

Shared chat: https://chatgpt.com/share/67b39afc-5890-800e-8714-4972575db9c1

[glacial yew]

Makes sense, those would have pulled from a lot of the same inspirations I am.

[regal portal]

You gave me an idea

I was always fascinated with the halo ring from Halo

I'm going to ask it how would he imagine it would be constructed by humans in a near future with technology not too far ahead of what we currently have.

[glacial yew]

I can answer that question myself, actually. We don't have materials strong enough to do a spinning single piece ring that large. Not even graphene can do it. But what you can do is build a two-piece ring. The inner ring spins in a maglev track in the stationary outer ring.

The stationary outer ring can be arbitrarily thick, giving it enough strength to resist the outward force exerted by inner ring.

[regal portal]

oh for sure on the structural material stuff. Space engineering is something i really like exploring. I'm not expecting a realistic answer from it, but as close as it can come to it.

yes

[glacial yew]

Yep, it's doable with no new material science. We're just well away from having the infrastructure to actually build one.

[regal portal]

Might be more cost effective to just terraform mars. But i really like the the ringworld thing.

For the dyson sphere, i think humanity will die out before we can get to that.

We would have to consume every planet in the solar system just for materials

[glacial yew]

A dyson sphere is just so many habitats in solar orbit that they blot out the sun. Technically you've started building one the moment you put something that runs on solar power in solar orbit.

[regal portal]

i can see a partial dyson sphere being in humanity's lifetime, but not a full on one that encompasses the sun. And by partial i mean a very small amount.

Like the surface area of the US, or something like that, orbiting the sun.

[glacial yew]

Depends how you define humanity. Is you mean baseline, unmodified humans, you may well be right. But I don't think "dying out" is the right way to describe what'll happen to us.

[regal portal]

is 4o particuarly slow for anyone else? It's been crawling when it writes text for me, since last weekend.

[glacial yew]

It was doing a lot better last night. I assume it's just under heavy load?

[regal portal]

i agree. By humanity i think organic matter with mechanical or artificial biological enhancements, but not our consciousness on a hard drive, basically

[regal portal]

Maybe Sammy and the boys are working to push out a pre-Grok 3 update. 😄

I asked both 4o and o1, and i asked them to summarize it into concise bullet points. This is 4o:

this is o1:

this is another o1 session because i forgot to change the model before asking 4o:

[glacial yew]

Yeah, at some point we'll get full control of our biology, and then anything is on the table.

Or rather, everything

[regal portal]

I think 4o's answer seems more grounded and realistic than both of o1's

might as well try both of o3's.

o3 mini:

o3 high:

lsat one. Apologies for the spam. legacy chatgpt 4, out of curiosity:

[glacial yew]

The key tech if you want to do it in a hurry is self-replicating industry. Or a very different society that has a lot more kids than we do. Our doubling time right now sucks.

[regal portal]

i like 4o's answer the best. legacy 4 seems a little generic, and the rest too fantastical.

[rugged igloo]

Bruh I legit spat my coffee out chat gpt is so goated

[regal portal]

Those black bars are killing me!

sooo much vertical darkness!

[sterile glacier]

after I stopped paying, chatGPT is dumb as rocks :

[regal portal]

hmm, let me try on a logged out account, can you copy paste that phrase, please?

[regal portal]

Also is that a brand new session, or did you talk about other stuff before asking it that? Did you try to upload anything before you asked it that?

[sterile glacier]

I can't paste the greek, the automod here mutes me. But here is the session, it is a brand new one, I don't care about the translation I know what it should say. But the fact is that the default response from chatGPT has been:

"it looks like you uploaded a file! How can I help?"

And I get that they are trying to force me to pay again and not use the free version, but it literally has made their AI useless to me. No matter what I ask, 80% of the time that is what I get as an answer. Here is the session:

https://chatgpt.com/share/67b3a4b4-c254-8005-ae69-cdecf1978857

[regal portal]

Is the chatgpt windows desktop app worth getting? Any advantages over the browser based version?

[normal quarry]

I use the macOS Desktop App for ChatGPT. I prefer it to the web.

[feral zinc]

worked for me

[regal portal]

i just tried it while being logged out (so no account)

was on firefox. I doubt the browser matters though.

[regal portal]

just a personal preference or are there advantages?

[novel shadow]

😂

[regal portal]

@sterile glacier i also tried again with a logged in account with no plus or pro subscription and it also works. Not sure why it did that for you.

[sterile glacier]

I just checked that my memory is full, never had that while I paid. So I guess I gotta lobotomize it in order for it to work, trying that now. Which is a shame all that stuff was really useful for the novel i was writing.

[regal portal]

oh that's weird that memory would affect a translation

[sterile glacier]

its the only thing I can think of, let me see if deleting it works.

[regal portal]

I guess that answers a question: you get increased memory space on paid plans. Didn't even think about that before.

Weird annoying bug: I have the voice Ember selected, and on my phone, it works fine. However, on both the browser and desktop app, it uses a different voice, even though i have Ember selected.

[sterile glacier]

yeah it works now, I went ahead and saved the memory, so that went I start paying back in december I can add them back. But in the meantime. It is lobotomized. It is a bit weird since the memory stuff was only 24k characters, just under 4k words. it would make it a 28kb file. It seems crazy that they dont let you have that much on a free plan.

[harsh ginkgo]

Is there a good resource that explains what makes o1 and o3 different from ChatGPT 1-4?

I am doing a digger deep into AI since I think it'll revolutionize everything, and I want to understand hwo these work, but I haven't found anything that really outlines these differences. I know how transformers and LLMs work in general - but looking to go deeper.

[regal portal]

you can always save it in a new session so you have an easy time finding it, in the future.

does free account have access to project folders?

[sterile glacier]

I dont see the option of folders:

And my point is that the memory space should be larger than 28kb even a single 1mb would hold some 850k more characters and it would be hard pressed to exceed that on a free account if ever. although at 100+ million users 1mb would be near 100Tb but even that number is peanuts for the company.

[ruby agate]

Plus users have a 32k context window. Is this the same for the API, or is it the 128k of pro?

[harsh pilot]

That was kinda interesting. Had to fully close out the app (1.2025.41) and re-open it. It was stuck in mini-window mode. First time it's happened.

i.e. Did not have the option to maximize the window.

[regal portal]

I just install the desktop app and honestly it seems pointless to me. Doesn’t have any advantages over the browser.

I thought maybe it would give you access to the camera or let ChatGPT observe your desktop or games. Apparently not.

[rugged igloo]

pro has 128 but even then its not enough

[lost rover]

it would be cool if you can leverage your own gpu for some tasks or something like that with the windows app

Extend context length, etc

[cloud pewter]

I hate this mobile ui, removes chatgpt’s pfp for no reason

It has been a thing since its release why are they removing it

[ruby agate]

right but im saying is it 32 or 128 for the API?

[neon trench]

does anyone use projects? how do you get gpt to reference the included files instead of searching the web? it was doing fantastic referencing the docs, now I cant ask it a question about something specific in the provided docs without it searching the web

is there seriously no good RAG solutions for large and many docs?? is there a bigger context window in the api or something?

[rugged igloo]

128k for api

[neon trench]

those prices suck

[ruby agate]

Sucks API is so expensive :(

[rugged igloo]

yea i dont reccomend using it as a casual 20$ a month is a much better deal

[whole roost]

There were occasionally discussions about the Token Limits and stuff sometimes for each plan, reddit finally reminded me where they are written down, just for good measure

maybe i should include what is what lol

[rugged igloo]

mom sam wants ai models on phones

[regal portal]

What’s a phone size model? As in a LLM small enough to fit on the memory of your phone?

[elfin lava]

Basically ChatGPT phone , the phone is ChatGPT, no other app. Just ChatGPT , no phone call no camera

Wait of course the device need to have camera cuz of vision

[rugged igloo]

its a model that runs on the phone wut r u talking about

[elfin lava]

What brand of phone tho

Ok nvm

[rugged igloo]

modern phones that prob need an ai chip inside like the new iphonesor the new samsungs

[elfin lava]

I thought they already doing it on iPhone 16

ChatGPT integration thing

[rugged igloo]

you cant integrate all of chat gpt on the phone its impossible its just a wrapper on the iphone

but siri intelligence is garbage

[elfin lava]

I like how Apple didn’t compete with the whole LLM competition thingy.

[rugged igloo]

Apple isn’t jumping into the LLM brawl like some of its rivals; instead, it’s been quietly overhauling its entire ecosystem with what it calls “Apple Intelligence.” Rather than building a massive public LLM, Apple is embedding generative AI right into its devices—upgrading Siri into a context-aware, ChatGPT-enhanced assistant, rolling out Writing Tools that can proofread and rephrase on the fly, and even offering fun features like Genmoji and Image Playground for creative tasks. All of these smart capabilities are powered by next-gen chips like the A18 series, which pack a serious Neural Engine to run AI models locally for faster performance and stronger privacy protection (citeturn0search11). Plus, to navigate China’s strict regulations, Apple is partnering with local giants like Alibaba—and may even explore alternatives like DeepSeek—to ensure its AI tools meet regional requirements (citeturn0news20). In short, Apple’s strategy is to make AI personal, secure, and seamlessly integrated across iPhone, iPad, and Mac without chasing the flashy headlines of the broader LLM competition (citeturn0search23).

[elfin lava]

THE CITE TURN NEWS LMAO

[rugged igloo]

lol sry i cant put links in here

hold on

[torpid kelp]

Copilot still exists? 😄

[rugged igloo]

copilot LOLOL

[regal portal]

It’s delusional.

if you ask about the Don’s assassination attempts, it still refuses to answer.

probably some other topics too, I just know that was one of the easy ones to get it to shut down.

[potent saddle]

COPEilot

[regal portal]

13 minutes to an epic ai reveal or a new meme being born, fellows

[potent saddle]

killing openai would be funny

[regal portal]

well, I’m guessing, in the future, AI will follow Highlander rules: there can be only one.

[regal portal]

damn, Musk is operating on BPT 😁

still not live

[tranquil scroll]

hey uh, whats the limit of o3 mini high messages per day/month for plus users?

[regal portal]

50 a day i think

[tranquil scroll]

oh wow

[regal portal]

150 for o3 mini regular

50 a week for o1

[regal portal]

you doing world building or coding with it?

[tranquil scroll]

just curious

[regal portal]

it’s live

they left out o1 and o3 from this graph

elo score (i’m sure that means something to someone)

deep search will be available

[quick meteor]

this is embarrasing

they are so far behind

i cant belive i thought maybe it would be an o1 level model

"best ai in the world"

slightly better than o1 preview

oh

[regal portal]

another graph of stuffs

[quick meteor]

wait. i think i spoke too fast

[regal portal]

i guess the previous compared non thinking vs non thinking

and this one is pitting the thinking llms

my bad, i didn’t pick up on that during the stream

[cloud pewter]

yeah thats more fair tbh

[regal portal]

bejeweled + tetris

[elfin lava]

@ me when grok 3 has AVM too, thats all i care for.

[brisk bloom]

"next week" right?

[regal portal]

what’s avm?

[elfin lava]

advanced voice mode

[regal portal]

oh right

[elfin lava]

i dont really carea bout these.. coding computer capabilities it can do, just want a very advanced ai that can talk and response in realtime

[regal portal]

i just use elevenlabs conversational agents. They’re so much better

[elfin lava]

chatgpt already let me down so

[regal portal]

cool story.

[elfin lava]

ok i didnt really ask for that but good for u lol

[regal portal]

i really didn’t ask for what you do or don’t care about but you shared it anyway.

might wanna work on your people skills

[cloud pewter]

seriously if they are dropping 4.5 please DO NOT separate it to different versions and lock the highest one behind pro or give absurd limits. its a BASIC CHAT MODEL and i suspect that it should replace 4o. so only the same or even more usage on it is tolerable

[elfin lava]

??? its not that deep lmfao

[regal portal]

when is gpt5 supposed schedule?

[elfin lava]

what a weird behavior from u man 😭

[regal portal]

requires premium + subscription

well, let me rephrase that, they get it first

then trickles down

it’s out today

[brisk bloom]

if it's comparable to openAI, premium+ is actually cheaper

[regal portal]

voice mode will be there eventually, unclear if it’s organic. About a week away

[brisk bloom]

depends on usage limits for me

[brisk bloom]

sneak peak of grok avm at the end of the stream

[elfin lava]

Oh word?

[regal portal]

was a little robotic. it did like a little chuckle

He did say they’re still working on it though.

hopefully it can sing and do all sorts of accents that chagpt avm refuses to 😀

[brisk bloom]

well the tone was robotic but it still sounded natural

[regal portal]

yeah

well, overall it looks promising, which is good. even if it’s wasn’t the best, it will push openAI and Google further.

[brisk bloom]

higher usage limits I hope

openAI is very stingy on avm limits

[elfin lava]

Oh well!

[eternal rock]

Get notified for server events!

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

[indigo hound]

I remember that the previous ChatGPT 4.0 could use plugins. Can it still do that now?

[elfin lava]

what plugin u referrign to

[indigo hound]

I don't know. I suddenly want to use a plugin, but I can't find the button to choose a plugin.

I heard that OpenAI has released a feature that can control computers, but it is only available to the Pro version. Is this true?

[native otter]

Yes Operator is Pro only, also, it doesn't control your own computer, rather a virtualized environment they host on their servers, running a chromium based browser.

[dawn sentinel]

He said he's ironing it out and might release it next week.

[tardy isle]

Plugin is replaced by Custom GPTs.

[clear ether]

Not sure what channel I should ask this in. ChatGPT is generating a type of CSV table. It has a download icon that is broken, takes forever to expand, and I can't just copy the text from any cell, row, column, or the table as a whole. I need the contents of these tables.

[dawn sentinel]

Didn't Sam say people can monetize their custom GPTs? Whatever happened to that?

[elfin lava]

It better be good , and not too strict with guideline,

[tardy isle]

Still no words. there was conflict inside OpenAI right after GPTs release, so that might be postponed indefinately.

[elfin lava]

Not surprise

[native otter]

Works for me: https://chatgpt.com/share/67b43295-a500-8003-b3f0-b4cc071c1997

Make sure to use Data Analysis.

[clear ether]

This is what I got. Damn GPT is disabling even the snipping tool:

[native otter]

Download button is working for me on the iOS app and Web...

[clear ether]

I don't have a link like that. See the table labeled Independent City States of Avel? See that it has a download link and and an expansion link?

[native otter]

A web app shouldn't mess with your other apps, and download may be disabled if you blocked chatgpt from downloading stuff perhaps?

[clear ether]

This is what I get when I click the download link:

I have ChatGPT with default settings.

[native otter]

I mean web browser related settings...

[clear ether]

I have no browser related settings that do that.

I have to use another browser like Edge. Yay! A bug to report to support. At least I have their email address.

[native otter]

Probably a security related setting somewhere messing with it.

[analog quarry]

is it down or??

[native otter]

It works fine for me, try a different browser or Android/iOS/macOS apps...

If still fails, contact support.

[neon trench]

@astral thicket, I really don't like the new 4o. It's immediate casual responses keep it in the creative/casual mode, it increases hallucinations in my opinion

[native otter]

Why do you think he is the real Sam Altman?

[normal quarry]

that isn't Sam Altman FYI

[native otter]

Also, you can tell it how you would like it to respond by using custom instructions, or in the prompt directly.

[neon trench]

It says verified so its definitely him

[normal quarry]

OpenAI people have gold usernames and everyone on this server needs to be verified as having an OpenAI account

[regal portal]

do you think Sam Altman joined this server last year..?

[neon trench]

hes been really busy

[regal portal]

😄

[neon trench]

I haven't had luck with custom instructions. when asking, gpt will say some sort of version I have the following: "modes" for replying. I always tell it in custom instructions to use the informational, technical, direct modes but gpt always falls back into being casual again "Yeah, it's like X isn't it? It's crazy when Y." and the quality of the replies go down

except for the opinion on new 4o, I'm trolling, but if your going to come in this server with that username, your going to get tagged im sorry 😂

[regal portal]

I like how my 4o answers.

[neon trench]

okay hold up, haha I haven't thought of trying that

im going to try to make a SpicyGPT too lol

that will be fun, like when I had Santa help me create React components

[regal portal]

react?

[rugged igloo]

So I stumbled on this on Twitter of how strong o1 pro is

[tardy bone]

Guys is the GPT-4.5 still rolling out? I cannot see it on the list of models.

[normal quarry]

it hasn't been released yet

[tardy bone]

Oh, yesterday I saw Sam's post and thought it was out already and I haven't got it yet.

[normal quarry]

nope, not out yet

[tardy bone]

👍

[whole roost]

I've been testing around with hierarchy of information in projects and the system prompt is the oldest information and gets overshadowed by everything else
It might be worth it to actually attach a system_instructions.txt and tell it in intervals to scan the file again because that shifts the order of information in the memory and brings it back upfront which isn't possible with the system instructions
Canvas can also work for this but there isn't a command to have it actively reprioritize canvas other than through changing it, and if you manually change something on canvas like adding a space it doesn't always lead to a priority change

And yeah no one cares

Uploading a file again also leads to a rescan and it "knows" or gets a log entry or similar that user uploaded a new file, and brings it up front and interestingly also changing the instruction prompt cause a rescanning of all attached files which brings their priority up

And you can tell it to scan a file in every message and it will actively scan it again and change the priority rather than just 'looking for it in the context it already has'

[rugged igloo]

Yea it's been know 4o doesn't follow custom instructions

[whole roost]

It does follow custom instructions it just loses them pretty fast

[rugged igloo]

That's because context is so dam low

[whole roost]

O1 as well

[rugged igloo]

If the context gets to high it becomes robotic. And looses what it's purpose is .

Unless you put the custom instructions in a markdown or a file and have just always read what it's suppose to do

[whole roost]

That's not what I mean, it already deviates after like 2-3 messages

That doesn't have to do with the memory

[rugged igloo]

Idk. I always just use chats for couple messages and then move on. To a new chat. But I'm on pro with 128k context so it's better at retaining a lot

[native otter]

Custom instructions need to be very clear and on the shorter side to help 4o

Also formatted in markdown as said above

[sullen shard]

bro

is anybody else chatgpt not working

[normal quarry]

working fine here (macOS Desktop App)

[sullen shard]

now its working

subscription error

[grand ore]

🤣 what is ur instruction to get answers like that? 😂

[whole roost]

I'm also on pro

[whole roost]

That only helps for the first few messages, it doesn't care for any precious custom instructions as they just hold no priority anymore as the conversation continues
In the normal (not projects) instructions they are not even sent on the 2nd generation

They get like injected once at the beginning and done

[native otter]

Btw, telling it to Search Knowledge helps, forces it to search through custom instructions and files

You will even see it saying that it's searching before typing the message

[whole roost]

As I've written, it pushes the file you tell it to search as the most recent information again, but not the instructions

And here the non-project custom instructions, they just get injected once and never again, it's done really 'trivially' not in a good way

[rugged igloo]

Well idk wut to tell ya. But it is a problem hopefully with got 4.5 and got 5 will fix all of these things gpt 4 has

[whole roost]

That's not a 4o issue, it's an issue with how they inject data and prioritize it

Here with o1

It will never get the custom instructions in this conversation because it never gets added again other than in the first message and first "swipe"

I'm not even really complaining because I don't think it will be changed but it is helpful to just know this

[jolly temple]

is it just me or 4o is super slow

[normal quarry]

just you, it's running pretty quick for me

[jolly temple]

weird, 4o is never slow for me

[whole roost]

Mine is pretty slow as well, it was half as fast as o1

The text generation animation seems sluggish

[normal quarry]

I'm using the ChatGPT macOS Desktop App, working well here

[whole roost]

Android App

[jolly temple]

its slow for me on both the windows app and website

[ruby agate]

What is the use case for the o3 models? (Over 4o and o1)

[normal quarry]

o3-mini-high my use case is coding, which it is very good at

[ruby agate]

So 4o for volume, o3 for coding and o1 for math/reasoning?

[normal quarry]

4o for wide variety of tools, o3-mini-high for coding, o1 for science/STEM

[ruby agate]

Mmm makes sense

[normal quarry]

I have used o1 for difficult radio astronomy work and it's amazing

o3-mini-high coded this whole simulator for me

[whole roost]

I've used it for rather simple JavaScript things and it has been alright but not great, it usually doesn't work but it helps a lot when you can figure out why it doesn't work or when you can come up with different solution and tell it to do exactly that

[viscid pewter]

Are people still facing issues with PRO mode not actually using PRO? And instead using some other mini variant, with much worse/slower output?
This is happening on and off, sometimes i get the small PRO timer modal showing me that it's thinking, other times it analyzes just like the mini models for 5 seconds

[normal quarry]

this has not been happening to me

[viscid pewter]

Really weird, it has been working fine again the past few days, but today it's back at it again with the bad output

Not sure if it's some form of a soft ratelimit that they are putting because i use it too much

[normal quarry]

I am a heavy user, never experienced this

[viscid pewter]

You use o1 pro heavily?

[normal quarry]

yes, I do. Radio Astronomy work.

[viscid pewter]

I see

[normal quarry]

I just did this to test

definitely Pro mode

[viscid pewter]

This is how it looks for me

Only reasons for a few seconds

[normal quarry]

that's a pretty simple question, I'd only expect a small amount of reasoning for such an easy prompt

[viscid pewter]

Usually, even saying hi or test, would trigger the same progress bar that you have

Copied your question

[normal quarry]

this is mine https://chatgpt.com/share/67b47809-52bc-800a-bc00-e3dd1702525f

[viscid pewter]

I used your conversation, and prompted it further, but now i got the second variant, and not the one that you originally got

I think it's some sort of a soft ratelimit that they are not putting out publicly

[normal quarry]

might be worth asking https://help.openai.com

[viscid pewter]

Yeah i think i'll have to reach out

[tiny pivot]

Hey guys - I dont know if you can help but I manage multiple GPT Teams for my colleagues in work in various departments. I can still see all the models and so can some people, but other people have lost the legacy model, lost o3 and o3-high, only have GPT 4o, o1 and Auto?

Any ideas on how to get them back for those users or allow them to still have access to the o3 models? They can still access them via the mobile app but not on the web regardless of browser, even if they sign out, sign back in etc

[normal quarry]

only support can help you with this -> https://help.openai.com/

[tiny pivot]

No worries, thanks Robert!

[dire rune]

Is there a keyboard shortcut for the microphone dictation?

[viscid pewter]

Do you know of any way i can contact the support through email or chat? When i select the various options on the chat widget, i don't get the option to talk to support

[normal quarry]

make sure you are logged in, you should be able to leave a message

[viscid pewter]

Okay thank you, i'll try again

[inland heron]

do o3-mini models actually use the web search or do they just claim they do?

I never see "searching the web", and their sources are sometimes nonexistent

[normal quarry]

I don't think I have asked for anything that searching the web was needed for

on o3-mini-high

[leaden shale]

hi, does o3-mini have a chat limit if you're a chatgpt Plus subscriber? I know that o3-mini-high has one

[normal quarry]

yes, 150 messages per day I think

[leaden shale]

okay thanks

[normal quarry]

it's getting hard to keep up with all the limits

because o3-mini-high started out at 50 per week and is now 50 a day

[tight island]

damn...

i wonder how the early grok 3 model will affect chatgpt's rate of progress.

4,5 was dropping soon innit?

[honest jay]

#ai-discussions is the best place for non-OpenAI stuff

[tight island]

but dosent it affect chatgpt's popularity,

[honest jay]

#ai-discussions is the "catch-all" channel for all non-OpenAI stuff, even comparing non-OpenAI stuff to OpenAI stuff. This channel is more singularly-focused on just ChatGPT.

[eternal rock]

Join the Sora Weekly Theme!

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

[native otter]

It's still in beta, but would be nice to have more yea

[dawn wharf]

What is Sora's limit on the ChatGPT Plus plan?

[cloud pewter]

depends on video quality, you can check their site because it’s way too complicated

[pulsar carbon]

OpenAI will soon release GPT-4.5 in response to Grok 3.

[quiet mulch]

Everything seems so slow these days, the desktop version of ChatGPT is having so many issues on my system, if I switch to the browser version it is still having so many issues, empty responses, or responses that take a LONG time to display, canvas operations are terrible. I have a PLUS account, but even making calls to API endpoints, direct, or assistant API calls are so SLOW!... It really seems like everything is just at a crawl over the past little while.

[vague cipher]

4o really pushing those blank responses today

[sharp urchin]

https://x.com/sama/status/1889755723078443244

Sam Altman (@sama) on X

OPENAI ROADMAP UPDATE FOR GPT-4.5 and GPT-5:

We want to do a better job of sharing our intended roadmap, and a much better job simplifying our product offerings.

We want AI to “just work” for you; we realize how complicated our model and product offerings have gotten.

We hate

X · Feb 12, 2025, 7:17 PM

when

[chilly quest]

Everything is dumbed down. ChatGPT is acting super weird

[chilly quest]

1000%

No memory, forgets things, misses things. Doesn't follow your instructions despite repeated prompts

[vague cipher]

I just with there was a more reliable uptime monitor that actually lets us know the servers are struggling instead of just the green status. It doesn’t feel like a very “no issues” status right now

[regal portal]

yeah 4o is particularly slow for me right now. Slowest I’ve ever seen it. I can actually run LLMs on my computer quicker and I have a 8 year old desktop with a nvidia 1080.

maybe they’re in panic mode because of Grok 3 and diverting all brain power to work on 4.5 and 5. 😁

[prime swallow]

My custom instructions are nearly 3k characters long (both fields full) and my memory's at about 98% full. I find that helps - but any conflict or ambiguity can cause issues for sure.

[regal portal]

I noticed o1 follows instructions more to the letter than 4o, and will retain them in memory for much longer. 4o will typically forget after a handful of exchanges.

One issue i have with o1 is that every prompt it spends a bunch of time addressing my settings (talk in street lingo with rude offensive words and swear words). Even after the first prompt, it keeps doing it, wasting time.

[prime swallow]

Like, this... is with all that personalization.

I believe you, that you're seeing it.

Consider adjusting the personalization maybe? Check for conflicts with itself, with the model's programming, whatever.

Amazingly, new chat asking the model "What are you thinking about" might help clue you in on any issues like that.

[prime swallow]

Yeah. That would be a 'conflict between personalization and the model's programming'. - I think I see in what you explain and show.

No idea if you are interested in trying to explore to find with the model a middle ground where the conflicts are less or even resolved.

[regal portal]

I was thinking about that, but then i thought that compromising what i want would mean that it wouldn't push itself to meet my criteria.

So basically i oversell what i want in the personalization, so i get more or less what i'm willing to accept as satisfactory.

[elfin lava]

Custom GPT can do advanced voice mode now it seems?

[prime swallow]

Yep, you have to decide the balance. Sounds like either way it's not getting to your 'perfect target'. Maybe for now you might have a choice of 'veer off early, drat' or maybe 'consistently accurate-ish, long chats work, just no closer to my ideal than this compromise.

Your answers and decisions are yours, of course!

[regal portal]

In the past, it would refuse to give me an answer explaining that it wasn't acceptable by its policies, so i would follow that with "just do as much as your policy allows you to do" and it would process with what i thought was good enough for my childish needs.

I toned down my personality settings because it did that every time.

Is o3 mini (and high) supposed to not display anything for its reasoning?

Sometimes i'll be able to read it, other times it just shows "None"

[feral zinc]

its fake summary of its thinking and fails sometimes

[zenith lichen]

is it just me or chatgpt sucked at generating glsl shaders even with o3. It outputted instructions that doesn't even exist in the spec

[tight island]

openai better up their game with the new grok release.

[chilly quest]

ChatGPT is unusable at this point. Nothing quite works bro. Let's not pretend it's "personalization" issue or user's prompting fault

Anyone can confirm these things: The memory is almost not there. It remembers nothing. It forgets things, it misses things. It's hallucinating too much, doesn't follow your instructions. A while ago, it used to give you full line for line output, I'm talking thousands of lines of code, without a mistake.

[tight island]

but you dont need thousands of lines of code.

if web is unreliable go api

[chilly quest]

Says who? To each their own

[chilly quest]

Thing is - it was reliable.

[normal quarry]

I am using ChatGPT on the macOS Desktop App. I choose the model appropriate to what I need and have had no issues like you describe

[tight island]

but things change, nothing is static.