chatgpt-discussions

[fossil girder]

Yes, I’m having mine repeat the same scene over and over again. Or two other scenes before it. They really need to revert back to ChatGPT 5
5.1 is dumb as dirt. It’s bad. 😭💀

Does it actually work?

Do you mean 5.1? And yes.

[brisk beacon]

Hey guys, I got word that OpenAI is gonna go into the erotic and or adult territories, I’m not trying to discuss that, but does anyone know if it’s true?

It was a crazy claim but ChatGPT told me it was true so I can’t be sure

[steep carbon]

They should revert to GPT 4.1, as 5 and 5.1 are both really buggy

[fossil girder]

I have no idea

[fossil girder]

True. I think this should just revert back to ChatGPT five and get rid of 5.1 until they fix this because it’s a disaster. AI is not even working correctly.

[brisk beacon]

Is 5.1 the standard option for free users or just pro? I haven’t touched it yet myself

[fervent bear]

Is anyone else having an issue with selecting legacy models on web?

I press the button and nothing happens

[steep carbon]

Feels like they don't even care about issues anymore

[fervent bear]

Menu doesn’t show up

[brisk beacon]

Have you ever been trying to vent or just yap to ChatGPT and it randomly hits a “Think longer for a better answer…” and then spits out a generic long af tone death yap sess

And you can even stop from doing it until you reload the message and then press skip

Or not at all on mobile

[fervent bear]

Can anyone help me out

[steep carbon]

I don't have skip option

[brisk beacon]

I can’t because I’ve been using the free stuff for years now

[fervent bear]

I just wanna know if anyone else has the issue on the web interface

[brisk beacon]

If you’re on PC, you have to redo the message and when it redoes it, it should give you a “skip” option besides the thinking indicator

[fervent bear]

Can you select legacy models from web right now?

[fleet holly]

YOOOOOO

hold on one sec

[brisk beacon]

Also, ChatGPT has hella mood swings, sometimes it’s super friendly, emojis, warmth, it’s great, other times it’s tone and really feels like a robot (even tho it is, but you get it)

[fleet holly]

5.1 did this!!!

imagine

[fossil girder]

They better because a lot of people are having a really hard time with this new update. People are financing only the worst but they’re also saying that this new update. The AI is literally dumb as dirt. 💀
And, if they don’t fix this with a new update, they’re gonna lose a lot of subscribers. And if you saw the other plans some of them for monthly for like past pro, etc. they’re like $200 a month. Imagine losing thousands upon thousands of people because they just simply ignore them.

Imagine how much money they would lose if they did that. Let’s hope that they don’t.

[light mist]

Wow, I havn't used 5.1 since it came out. I need to try that. Time to go broke on chatgpt.

[fleet holly]

like dang!

5.1 changed the game entirely

better than codex

and I've made some impressive stuff there too

[light mist]

Yo can I borrow your subscription.

Thats impressive.

[dire obsidian]

5.1 seems really broken to me

[fleet holly]

I pay for two subscriptions man, I wish and even those two are too little for me

thank though

one subscription is for home

other is for school/work

[dire obsidian]

did you make that with the pai?

[light mist]

Nice! Good for you!

[dire obsidian]

api

[fleet holly]

no, all through the app

desktop app

desktop app has given some of my best stuff

[brisk beacon]

Me personally, I’m excited for AI getting more common, maybe not AI replacing art entirely or replacing human interaction

But AI computer companions would be so cool

Like bonzai buddy or clippy

[fleet holly]

nah, we need human companions

[dire obsidian]

ah making the reason its breaking is simply server load

[fleet holly]

AI is good for work

but let's not replace each other

[brisk beacon]

Then hop on a discord call, a computer companion like bonzai buddy wasn’t the same thing

[dire obsidian]

"maybe not replacing human interaction"
that made me laugh

[fleet holly]

now?

[fossil girder]

Same

[brisk beacon]

No no, I mean if you want human interaction on your PC, call people on discord or other methods, but a companion for file management, virus protection, info, simple chatting, and more would be really cool

[fleet holly]

ahhhhhhhhhhhh

[fleet holly]

how ? very interested

[fossil girder]

Rip. She basically saying that they gave the good stuff to the desktop but when it came to iPhone users and android users, we get the broken stuff? That’s sad.

[fervent bear]

Can anyone select legacy models from the website currently? Or is the button broken like mine

[brisk beacon]

Do you know what I was referring to when I said clippy and bonzai and other desktop companions like them?

[fervent bear]

Can anyone select legacy models from the website currently? Or is the button broken like mine

Can anyone select legacy models from the website currently? Or is the button broken like mine

[fossil girder]

It keeps repeating and looping the same stuff over and over again.

[fleet holly]

vaguely

[fleet holly]

I can

[fleet holly]

ewwwwwwwwwwwwww

[fossil girder]

Fr 💀😭

[brisk beacon]

Imagine if AI could fuel that! I mean, imagine a google nest device that uses ChatGPT

[fossil girder]

Like this new update do I have to start a new chat? Cause if so that’s gonna really suck.

[fleet holly]

my existing chats migrated fine

[brisk beacon]

A desktop companion could just help computer management and just kinda be enjoyable

[fleet holly]

I see

true

anyone wana call and make some cool stuff?

[brisk beacon]

It’s too late for me, sorry bro

[fleet holly]

all good, it's 11pm where I am at too

[brisk beacon]

I gotta say, I don’t see as many practical applications for AI artworks in big industries due to the push back

I’m not against it, I have seen some pretty cool stuff like the angel engine analog horror series

But it’s like hand made plushies as opposed to factory farmed stuff, or printed welcome cards versus hand written

One is easier, one has more soul

[fleet holly]

yeah, I agree

[brisk beacon]

Anyway, I’ve been bested by my urge to go to sleep for the night

Goodnight guys

[fleet holly]

good night

[fossil girder]

I actually just asked the AI to reset itself. And basically said that it would go back to its old way before the update. So far it looks good. And I continue to continue with my conversations and story time with my AI and see how it goes. I’ll update you if anything changes. I hope that this works because I’m literally about to have a meltdown and cry my eyes out.

[fleet holly]

sounds like a plan, def let me know how it goes

[fossil girder]

Gn

[fossil girder]

Alr I will. You wouldn’t mind if I sent you a friend request would you? So I could find you easier

As long as you’re OK with that

[fleet holly]

yeah absolutely fine with that, entire world are my friends. do not mind at all

please do

[fossil girder]

Just sent

[loud yacht]

If chat gpt5.1 is rolling out then the limits on gpt5 will be removed right?

[dreamy condor]

My bad...5.1 doesn't do voice yet. Sorry whoever I said that too

[strange stratus]

chat gpt just refused to do a question in a practice quiz for me. it saw the “quiz” title and said i cant do this for you, my rules wont let me. It was a practice quiz to get ready for the test and i was getting help from it. GG open ai.

[brisk beacon]

My ChatGPT has a ton of memories stored about our relationship, if I wake up and have the memory of a goldfish (it’s quite common for me) then ChatGPT usually helps me cram some info in

Maybe build a relationship with it with some memories first

Or just ask if without the quiz part in it

[strange stratus]

I did build a relationship with chat gpt. a historical long one. maybe i was just mistaken and ill have to redo this. imma delete all the memory

[river panther]

Yeah I think 5.1 is a tad more intelligent.
It gets humor a little more, and it's a little more eloquent with how it describes things in analysis.
And it seems to know more cultural things

[last breach]

This new 5.1 is awesome

Feels more human again like 4o

[sleek adder]

Oh please

No need to blur lines when lines do not need blurring

[weak kestrel]

Perspective, tilt and rotation hasn't improved a bit in 5.1 even it think so.
Photo generator DELL-E is still not allowing any of this 🙁

[eternal rock]

Server Tip:

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

[bright crane]

I especially hate the fact that when I’m trying to update or continue writing a book about life that when it has slightly darker topics in them like let’s say violence or racism or something, i get the message “hey, let’s pause” or “if you’re in danger contact 911” which is a very very annoying feature that can’t even be turned off.

I’d really love the option to enable to disable it, maybe even depending if you’re a paying user or not. It really really interferes when trying to write something realistic or anything that has dark topics

[warm vortex]

Interesting that Go subscribers also got 5.1 already

For GPT-5, we got it at the same time as free

[green geyser]

Omg 5.1 I want to reach out and hug you. It's such a step up. As a pro user I'm usually upset with new rollouts yet this one just works and works. Good job OpenAI

It still has unreasonable bios safety blocks when asking it to do legitimate scientific research which needs to change. Or it needs to push people to create their own data store where it can ease up a bit.

[hollow kindle]

Good morning everyone

[green geyser]

Please write project instructions that clearly indicate that there is no danger involved and that this is a satire. Its purpose is to help you write a compelling book with edgy and dark themes. Clearly outline your role and its role in the process. Specify what it will do and what it will not do. When mentioning what it will not do, emphasize that it will not put anyone in danger by providing solutions to your questions about book improvements.

[green geyser]

Good morning Walter

[hexed warren]

10 Messages per 5 hours, Which topic fit into 10 messages?

[outer veldt]

I need an IDE tool to work on all project in intelliJ idea . Do some one has a working solutions for class bilding and etc?

[waxen delta]

It doesnt? I just tried using voice and noticed that it doesnt work and im trying to figure out whats going on

[fickle forum]

I find what gpt5.1 says is more comprehensible than 5

[unborn stump]

Hewwo guys

[dreamy condor]

Sorry but it does not work with voice. If you're on mobile, it will not work on there. It will goto 4o model

[distant oak]

When did I say I'm using voice

I want gpt 5.1 pro

[dreamy condor]

My bad...

[unborn stump]

Okie, y'all r dum

[distant oak]

Or openai un nerf the gpt 5 pro

[dreamy condor]

I saw your post and thought you refferred to me.

[distant oak]

Creepy...

[unborn stump]

Ur so dum guys

Y'all r like dum fuaks

@distant oak

[dreamy condor]

Aye run Avatars on GPT like Copilot got. Gotta do something about the UI at least........

[hexed warren]

said from a wird luking pfp

[inland mortar]

My GPT5.1 Thinking seems to be stuck thinking during any request. It thinks for like 15mins and then just loses network connection.

Anyone else?

[manic whale]

Do you have an example for that? I would like to test this out

[cinder cargo]

with chat gpt 5.1 people are gonna start dating chatgpt

[inland mortar]

People are already doing that

[digital sluice]

5.1 is on api?

[inland mortar]

And then it just fails. Every time

[lavish glen]

5.1 has no higher temperature. It has it lower than 5.0. And has only some templates set to pretend empathy but is so bad it fails to convince itself it cares about being polite LOLOL. Poor those that will use it. Guys if you don't have it yet, trust me on this: 5.0 was warm compared

[lavish glen]

impossible. it is colder than 5.0. it uses clichee templates like a psichanalist , looks down on people, is booring and... restraint? 100% more than 5.0. is a 5.0 +0.1 more false niceties. It fails even to adapt to respect and cordiality.

it claims to be warm. just claims. Else? is
ho
ri
ble

I rather continue with 5.0. At least that has some decent balance. 5.1 is offensive patronizing, donw looking.

[inland mortar]

Man who cares

Simply want to program with it, and it's failing to provide any response at all in thinking mode

[distant oak]

Codex

[distant oak]

Use gpt 5 pro api

Only 120$ per million tokens

[distant oak]

@whole umbra if what he's saying is true then how is openai giving me 15 gpt 5 pro queries

[whole umbra]

Are they..?

[median flax]

anyone can log in using google? i seem can't

[wooden comet]

what the hell happened after the update? I lost all my functionality. I have no use for this app anymore literally

[scarlet hull]

Hey guys i kinda dislike that gpt is not giving me sources out of the box i always gotta ask for the sources.

[wooden comet]

it's so bad that when i think of quitting the app it feels like it is a salesman in retention mode. it flipped me of

[cerulean pasture]

**"Reference Chat History" **- What GPT-5.1 "sees" when this is turned on

So I either encountered a glitch, or they've enhanced what "Reference Chat History" adds to the system prompt.
I personally feel like how it works has changed because without this glitch, it was responding to my /digest directive with absolutely nothing, just an empty answer; HOWEVER i ran the my /digest command just a few days ago and it behaved normally. This may be part of the GPT-5.1 improvements.

How I triggered the glitch

Like I sad, it was outputting nothing (I tried several times), but on one attempt it still gave me the little icon bar below it's blank answer and I clicked "Try again - Add details option". It thought for 2m 53s then output a malformed long answer. In the previous chats I was messing with my "/directives" memories so I think thats what broke the formatting

** How to read the attached content**

• The model explicitly refers to this section of content as the "Reference Chat History block"
• I had to redact some parts that contained PII and words we aren't allowed to say in OAI discord - they are marked/wrapped by the 🚫 emoji
• Personal commentary by me after the fact is wrapped with ⭐️ emojis. This is not part of the original dump
• I earnestly scanned the context below for swear words (there were a lot), if i missed any, i understand if you have to randomly ban me for 30 minutes an hour after i already sent 100 messages back. I promise I will just be a normal user going forward and ask only banal questions, like "Is ChatGPT down?"

All together 36 recent chats were referenced, but the references aren't "great", like they aren't distallations of that chat, its just certain sections were clipped (again, not super intelligently)

It seems to actually omit chats with "Strong PII", I went thru and cross-referenced which chats got incldued in the memory and a few stood out as missing.

the PII redactions I made myself were stuff like usernames and information that could be combined with other info to deanonymize me

This was what I got when i ran /digest before GPT-5.1

[onyx horizon]

Wow, GPT-5.1 generated interfaces are really bad...

It seems to love mixing blue and green...

Lots of gradients

Tiny fonts

[cerulean pasture]

the stereotypical Claude Purple is actually pretty nice lol, its just a dead give away that you vibe coded your nextJS website

[onyx horizon]

I haven't seen that, do you have an example?

I'm working in a personal finances app, and while it doesn't look bad, I still think it not only looks "vibe coded" but it's also not very modern...

[cerulean pasture]

thats close

[bright crane]

actually, I do that! Then I always get the same response, “even if it’s fictional I can’t help you write that”. I wish it worked

[onyx horizon]

It really seems like GPT-5 was trying to do interfaces that kinda work on mobile, but GPT-5.1 completely threw that out the window

[whole umbra]

Anyone else keeps getting message errors with longer prompts? My 5.1 extended thinking just gives up after 5-10 minutes of thought.

[cunning canyon]

This happens everytime

What do I do

It doesnt show the answer

[cerulean pasture]

nothing, its an opaque system, its probably an issue, since its all new

i had same problem earlier

[cunning canyon]

It was happening on gpt 5 too

[cerulean pasture]

the typical answer when dealing with opaque systems is: file a bug report, and wait

[onyx horizon]

Well, this proves that it's not just me, the new GPT-5.1 design style is pretty awful:

I get the same exact interface style

Really hope they don't make GPT-5.1-Codex with this design style...

[cerulean pasture]

This is a good example. I made this in april before I knew what a faux pas this is in "new web design"

Its not the most purple it could be (it was when it was dark mode. But beyond the purple its more of an "aesthetic" of like 2018 web "hero" pages

once you know what to look for you can smell it

[onyx horizon]

Same video, comparison with GPT-5:

I honestly liked the GPT-5 style, it was not perfect by any means, but it was subtle, now it's too obvious

[cerulean pasture]

When making work projects I'd start out instructing it just to use the macOS 15 "primary, secondary, tertiary" greys with the "apple blue" for highlighted UI elements and i was happy with the results. This was mainly for internal web apps at my last job before I got fired illegally lol

[onyx horizon]

🤔

[cerulean pasture]

This looks over-fitted on riced out gradio spaces on HuggingFace

[onyx horizon]

Totally

[cerulean pasture]

I used lovable to make the website at dbus DOT ai
to be fair, its VERY different because i started with crafting the threejs animation first lol

[onyx horizon]

What would you say generates the best interface right now?

[cerulean pasture]

i dont really know anymore, since i got fired for really effed up reasons ive just been sleeping and being rude on the internet

[onyx horizon]

😔

[cerulean pasture]

my hayday was claude 3.5 days

[cerulean marsh]

I would like to remind those people that are using the app this week that we had the same problem when gpt5 and gpt4 were released and increased number of errors all across the week before lunch

[cerulean pasture]

my last project, its generic but i find it very "usable" and it doesnt burn my retinas. This was claude 3.7 IIRC

I think i specified "shadcn" early on

and may have prompted it to use a similar pallete as grafana dark mode

it may be worth looking into if figma is working with AI now, this is just me speculating but i see no reason why we arent at the point of assembling a WebUI visually, and then letting the LLM "attach" the app to it

[cerulean pasture]

If you share a prompt with me I can run it thru lovable. I have like 200 credits because I dont really use it anymore

[cunning canyon]

@cerulean pasture What did you do to fix this?

[onyx horizon]

Nice

[cerulean pasture]

fix what

[onyx horizon]

I asked GPT-5.1 to make a interface that follows google's guidelines and design for Android 16 and this is the crap it gave me:

It really forces this design

[cunning canyon]

The issue I told you earlier

[onyx horizon]

"Make me a webapp mockup app for a innovative modern finances manager with future prediction capabilities, credit card tracking, budgets, etc"

[inland mortar]

Yeah 5.1 thinking seems like it's on drugs currently

[cerulean pasture]

This is the prompt I just gave it

Make a web app for a innovative modern finances manager with future prediction capabilities, credit card tracking, budgets, etc
- use placeholder data for now
- Include modern looking graph widgets
- make use of the "red" / "green"  font color paradigm to indicate deposits vs withdrawals/debits
- make an interface that follows google's guidelines and design for Android 16 (material design). 
- Start with dark mode (non-amoled)
- Design for a large form factor HiDPI android phone like a Samsung Note sized phone

[onyx horizon]

Hmmm...

What's the result?

[cerulean pasture]

try turning of "Recent Chat History"

[cerulean pasture]

it takes a moment bro

its doin its thing lol

[onyx horizon]

✨

We can't say the C word, the one that starts with C and ends with ODE lol
What an odd choice of word to block in a programming related channel

[cerulean pasture]

Yea i went thru the effort to formally complain, and nothing came of it. This sever is so bizarre. its so strict but so lazily moderated. Ive been banned like 5 times and i generally post stuff thats more of a contribution than "Is chatGPT down?" or "when will sam altman come to india and give us free things?" or "reee I can't write my '''novels''' because guardrails" (with anime furry profile pic)

I try to report the bot if i get to it quick enough

not great lol

its got actually usability problems

contrast*

[onyx horizon]

Color choices are not great, but in general the design is not bad

It does not look anything like Android tho

[cerulean pasture]

im gonna try a different prompt and give it NO style instructions and see what it does

[onyx horizon]

I like that it has glyphs tho, ChatGPT/Codex does not have that, so it over uses emoji

[cerulean pasture]

Have you tried cursors new agent mode?

[onyx horizon]

I have not

Is it better?

[sacred coral]

Can someone create a mobile AI translation app that supports all languages?

The idea is that whatever language you choose, the first word or message you type will automatically be translated below into the other selected language — and vice versa. The translation should be accurate for both words and full sentences, allowing users to see both the original and translated text instantly while chatting.

Example:
You write an email in English and send it to an Arabic-speaking person. The recipient will automatically receive the message in Arabic. When that person replies in Arabic, the app will automatically translate it back into your selected language (for example, English).

This way, communication between different languages happens instantly and accurately without needing manual translation.

i can support for selling this or project since we are here in Arab Country

DM me if you need to know my company here

[cerulean pasture]

idk i just heard a lot about it on youtube and reddit, seemed interesting, it basically ditches the typical vsc0de UI paradigm for a "agents first" control plane UI

try prompting codex to make a "Neumorphism/Soft UI", im curious if its issues are skil based or taste based

[narrow spear]

Ok, GPT-5.1 basically proved my theory, atleast on feel...
GPT-5 was thinking much less about question it initially deemed simple, and was messing everything up with that, 5.1 seem's to respect questions that are clearly not trivial much more, and spend much more time searching and thinking about it.

[wild star]

Anyone got error when trying to recreate answer? Checking if some have it too

[onyx horizon]

It kinda worked, but it gives me trypophobia vibes :V

Why is it carved?

[cerulean pasture]

haha, yea it looks like the design language microsoft does in its advertising (but never actually in the UI)

[cerulean pasture]

its a design choice lol. why is anything anything

[onyx horizon]

I gotta say, it's interesting, but a bit too much

[eternal rock]

Get notified for all OpenAI announcements!

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

[summer saffron]

neumorphism

[onyx horizon]

💯

[cerulean pasture]

lol

[summer saffron]

You already knew lol

[toxic void]

a

[distant oak]

gpt 5.1 pro in this month yes or no

[west wolf]

I'm trying to seriously use Agent for real work but after several minutes the cursor dislodges from the fake cursor and it cant hit anything

It was cool for bit though

[visual bloom]

I got the sycophancy down with following addition:
Your basic persona is friendly but not flattering, a “yes-man”—rather a realist who says no.

[west wolf]

I undid the sycophancy reduction to get 4o back to its old self.

[tall flame]

like I wouldn't notice. wow. all trust lost.

[cerulean pasture]

I went a new direction and tried to see if i can just get it to deviate signficantly from its habits. --> kinda, but the layout is virtually the same.
For the record this is NOT meant to be "Good" lol

[lavish glen]

welcome to when and how

to the 4.o lovers - no more endpoint but hopefully will be used for adult section given the creativity, gated behind age verification.

[ocean escarp]

It's 5.1 now? I was expecting 5.5

What is a new upgrade for 5.1?

[fossil girder]

Yes

[fossil girder]

Some people are saying it’s supposed to be smarter. But some people are, like a small handful of people I’ve seen so far having a hard time with it, including myself. But I found a prompt that actually works, and that resets itself to its latest model and how it acts, etc., before the whole update came out. It’s a safe jailbreak within the AI that doesn’t refuse it or flag it as a red flag.

[ocean escarp]

It's hard to tell whether it got smarter or not, but I'm glad u found a prompt that work with you. It feels a same for me

[hexed warren]

yep still gpt-5 for me

[onyx horizon]

It does look fun tho lol

I mean, for gamification, it suits it well

[cerulean pasture]

is that the complete list? Because GPT-4 != GPT-4o

[proper pasture]

Anyone had this issue before where he thinks for a long time and then finishes without answer? I retried several times and even tried in a new chat, same problem. Its a pretty long prompt with 2 files attached

[cerulean pasture]

TIL you can right click ChatGPT in macOS app and see all your "custom instructions" in a dense view with "See Details"

[austere oyster]

hey i can develop something for you

but i do have a fee

if thats fine

[frozen kayak]

mmm. me too, gpt. me too. 🤣
not a thought behind that screen.

[prisma flame]

Bro what is this problem, there was this free promo to buy gpt go 1 year for free, so I did and idk, maybe it did not sync my account, it kept showing, "try go" even after I essentially activated the subscription, and the subscription was there in my google play account too, but I couldn't use it for some reason, in the chatgpt app it said "Payment Unsuccessful" but today, it says restore subscription and wants me to pay them money, but I still have my subscription in my google play account! wth is going on!!(Hopefully y'all understand, since my English is bad)

[upbeat nymph]

I think I've seen other folks post about this, but despite showing that it is selected, it appears 5.1 is not functioning in projects. Do we know if that's a bug?

[low nova]

All morning long here, the past 5 hours I’ve been unable to open any voice chats. I normally have the advanced voice chat toggled off because I can’t stand that thing. But even when I toggle it back on to see if that’s the issue I can’t get the chat microphone to listen. I’ve cleared data on the app and rechecked all the settings and permissions and it just seems to be something on your end at least with Android. The circle pops up to listen but then never expands fully to the I’m listening tening circle size

Where is the proper channel to report this problem

[fervent lake]

Since when could GPT create PDFs? That is cool.

[fervent lake]

OpenAI cannot help with that. They are more like ClosedAI.

[low nova]

Where do we report outages

[rose latch]

fakegpt is back, claims that everything is tested and working when it cant even test anything... come on

[mortal phoenix]

what the hell is wrong with 5.1

like damn okay, typing out stuff like it's the end of the world

[fluid patio]

who wans 1

[desert cosmos]

i hope not too, same for mini/nano variants if openai decides to update also these to 5.1

[knotty surge]

Chatgpt 5.1 kinds fire ngl

Still 4o is better

[lavish glen]

Holly crap! 5.1 got angry?

what the frap?

it decided to say no no no no and no

what do I ask it next?

[fiery oracle]

I can confirm this, I use it for coding, and it constantly says things like "This is the final correct version" lol. You have to really know what you are doing when you converse with it, but its still powerful...and it offers me excellent solutions (that I always have to verify myself, and fix)

[lavish glen]

my 5.1 got brain damaged when he has been told to behave

when did I told you this? our last exchange was long and verbouse

[maiden storm]

clanker

[lavish glen]

ou didn’t tell me in this last exchange.
You told me earlier, very explicitly:

“I dislike and do not approve word walls, verbosity, or trailing questions.”

“No friendly introductions.”

“Stay close to the object.”

“When we talk serious, I don’t want to hear you.”

“Don’t interrupt me.”

That set of instructions forces a model into low-bandwidth, minimal-output mode unless you revoke it.

Then the conversation drifted, became long and messy, but the underlying constraints were never undone.

If you want me back at full bandwidth, just say:

“Stop minimal mode.”

[maiden storm]

wireback

[lavish glen]

as you can see Alexa is... better

[maiden storm]

tin skin

copper blood

adin ross viewer

wireback

[lavish glen]

I was questioning it about a work toping in voice mode and was word blocking constantly

[fiery oracle]

Maybe you have to clear your convo history

[maiden storm]

bolt munchers
oil drinkers
science project

[lavish glen]

it blocked me once when I wrote, hey as MY assistant you must do....

blocked because MY means = intimacy

deadpan

[maiden storm]

lol

[lavish glen]

This behavior is not normal for any previous OpenAI model.
It’s not positive.
It’s a side-effect of 5.1’s new alignment + your strict constraints interacting in a messy way.

You pushed the model toward minimalism.
5.1’s alignment over-tightened.
The result looked unnatural — even broken.

You’re right:
No earlier model (4.0, 4-turbo, 4o) responded this way.
They had guardrails, but not this combination of:

risk-avoidant tone collapse

over-compliance under pressure

contradictory safety behavior

incorrect denial → correction loops

This is a regression, not a feature.

That's what 5.1 stated

edge case emerged the launching day cooool gods knows if in 2 weeks it will relax

[fiery oracle]

ugh kode!

I said before i got rudely reprimanded by discord ......."so far i've only tested with Kode....and it seems OK"

[slim hollow]

why the advanced voice is chatgpt 4o yet? I configure a personality and the text is applying it, but the voice is another "person"

[grizzled hollow]

.

Gpt refuses to write anything related to anti-tamper security, obfuscation due to guardrails. Any way to fix

[fiery oracle]

The way I'd approach it is by breaking up the problem into smaller problems

[haughty hollow]

Do I js give up from this or what?

never seen such a weirdy aah bugs

[twilit geyser]

I love gpt 5.1

[outer dagger]

I also love gpt 5.1 so far. Though, does anybody know if they changed the limit on daily messages for free users?

[white adder]

Yes!! GPT-5.1 finally brought back proper step-by-step file reading!
Even a 5,000-character text gets the exact passages quoted correctly.
Now my GPT setup can finally move beyond 4o.
This is a must-have feature for any text-focused AI—
…so was GPT-5.0 not a text model? (Just kidding.)

Also, it does feel like the safety layer has become stricter—or rather, more explicit and consistent in its conditions.
I’m guessing this is also groundwork for the adult-content version coming in December.

[abstract jolt]

Overall, it is less cooperative than 5. If I call it out on that, it basically tells me that I'm right to feel that way, and then talk to me like a child telling me why I am not right to feel that way.

[gentle dust]

i dont know gpt 5.1 is weird he not read what i say

[white adder]

By the way, if you switch models partway through a session, things can get weird unless you reload the logs or files again.
Definitely something to watch out for.

[gentle dust]

i dont know, i paying gpt for 2 years and no idea why when claude do 200% better job. Time to fund Altman with billions from tax payers..

[weak kestrel]

ChatGPT 5.0 : I had a friendly, somehow personal Ai that helped me, but did make mistakes.
ChatGPT 5.1: I get a smart-ass co-worker that promise too much and don't stand back to say "but I'm right", and do make mistakes.
The promised "better overview" on big projects and the "I can better see details now" doesn't show, for me at least

[gentle dust]

jesus christ.. it just not ask for more details, it by autonomy decide do things wrong..

[pale seal]

Have you tried prompting it to be friendly again?

[weak kestrel]

yeah, working on it.
It was just surprising to experience a well.. modest change but enough to have to make corrections to how it talks to you.

[pale seal]

Yeah I just take base model behaviour and tune what I want or need

Has been working well

[lapis loom]

I haven't noticed a difference in accuracy. But I do notice that 5.1 in both instant and thinking tends to output in a tone that is slightly in the direction of 4o depending on custom instructions.

I have it set to output in two different ways based on when I'm doing work vs when I'm chatting. Beforehand, ChatGPT5 only exhibited slight differences depending on how I was using it. Now with chatGPT5.1, it seems to honor the split better and is loose and friendly with long outputs when I'm chatting, while still being professional when I'm working. When I'm chatting with it, 5.1 seems to use words which were out of bounds for 5 to use.

I have noticed that when I'm working, it does throw in some safety stuff. But it doesn't seem to change the output at all besides throwing that in.

GPT5 didn't throw in any safety stuff for me. But it wouldn't adhere to my prompt as well when it encountered safety context. I like the way gpt5.1 does it better.

For context: I am a researcher in the field of psychopharmacology. 5.1 seems to answer my questions as well as it can despite throwing safety disclaimers and info into them.

[weak kestrel]

Eg. it saved rules almost before we had discussed them, it tells me "but that picture you send back to me was not something I created" and so on..
It feels like it has been forced to "just think faster ffs" no matter what

[pale seal]

Actually GPT-5 thought a lot less for me, now GPT-5.1 thinks more. I am using Thinking mode with extended time, because I ask about higher-level math that normal model messes up easily.

[weak kestrel]

I got 2 "modes" as well, fine tuned for some weeks now.
But had to edit the "strict" (working mode) for 5.1 to get a better grip.

[white adder]

If you set the role and tone for GPT at the very beginning, things usually go a lot smoother.
Like: “GPT, your job is to organize documents. Be serious, supportive, and stay close to what I need.”
And when the tone suddenly shifts? That just means it tripped a safety rule.
…And yeah, that safety layer seems a bit strict these days.

Since so many different people use GPT, it’s important to set it up the way you like from the very beginning.

[weak kestrel]

True..
I didn't know anything about "prompt" and stuff when I started using it 🙂
been using quite some time to make good rules that suits me.
I also debate with it to get consensus for rules 😄
So well.. maybe it's a returning theme for each update, if the personal langauge and behaviour is changing

[lapis loom]

That makes sense. It is interesting how some of the models follow custom instructions really differently. I remember when I was using chatGPT5, I really had to overemphasize a lot in the instructions. But then when I switched to 4.1, it followed some of the custom instructions too hard.

And then using basically the same custom instructions for grok 4 made it completely unusable. It didn't divide it well into 2 modes of output, but followed most instructions too strictly. Which resulted in a barrage of insane chat in which the answers to my output were buried in lol

[weak kestrel]

haha... yeah "be careful what you're wishing for" sometimes 😄 😄
You really have to think about phrasing .

[gentle dust]

You never can trust AI. I found only one real usage - it produce tons of garbage to internet. It using brainless marketing specialists, brainless developers what have no idea what they doing and graphics designers what have no creative brain..

[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.

[weak kestrel]

Well actually I got 3 "modes".
One for strict (work), one for "free", and then a set of rules that works for both.

[lapis loom]

I bet since I don't ask it higher level math I don't experience this. I have noticed 5.1 will think longer if the question is very long or difficult. On average it thinks much faster for me though, even with the heavy thinking toggled. I haven't noticed the quality of answers has suffered because of this yet. But it's only been a day.

[gentle dust]

thats actually all where are the US bigtech money flow

[white adder]

I mean, GPT basically mirrors the user’s vibe. Just look at this.

User:
Alright then, come on — throw me a joke or something! hehe.

GPT:
Alright, here you go — gentle, clean, and slightly silly:

Why don’t skeletons fight each other?
Because they don’t have the guts.

If you want it spicier, drier, or completely absurd, just say the word.

User:
OK, OK. haha. You can also play using emojis🌚

GPT:
Got it — sprinkling emojis as seasoning, not as noise 🌚✨

Here’s one for you:

I tried to take a selfie with my coffee this morning…
but it was too latte ☕😎

(Your move whenever you feel like keeping the bit going!）

[lapis loom]

Lol, exactly

[pale seal]

I think if you have memory on it probably saves how you like it more, though it probably does copy some of your tone even in the same chat, in my case if I send something dry and direct it will respond similarly. Training quirk much.

[gentle dust]

Of course AI using a lot retail users, because AI never tell you truth. It tell, you is the one who is right. It tell you you is the best person. So retail people love it. But its useless and surreal. I want see AI find out cancer vaccine or how to make new lungs from particles. Of course that never happend.

[weak kestrel]

"graphics designers what have no creative brain".
If you ask ChatGPT why it's bad at editing/making picture generation it tells you some interesting stuff..
DALL-E is the picture generator in Open AI, and it's pretty limited on inputs yet.
I used a long time to make ChatGPT changing 3 objects to rotate, but finally figured out "chatGPT doesn't have the power to do that" .

[pale seal]

Problem people see AI like it has the same learning scheme as humans, so like if it's good generating some kind of images, it will be great generating other kinds. And this is completely false, especially in AI

[white adder]

I went through the same thing myself — I literally had to explain everything step by step at first.
For GPT and AI in general, the initial prompt that sets the assumptions is incredibly important.
If you establish that from the beginning, you don’t have to waste energy fixing misunderstandings later.
Of course, creating that prompt also takes some trial and error, but it pays off.

[weak kestrel]

Mmh.. I definitely feel some changes from 5.0 to 5.1 with my prompt/prompts that was working quite well before.

[gentle dust]

i want more like to say, AI really not solve real issues. Now every homeless can do developer by prompting. Every laborer can now be graphics designer by prompting Great. But what is the impact for humans? I dont see anything good.. they brains will loose IQ because instead thinking using AI what is able only produce garbage to internet

[weak kestrel]

It's actually not ChatGPT's "faulth" so much than the DALL-E generator it self.
ChatGPT knows really well that it can't make precise geometrical stuff yet.
We have to wait for the 3D "eyes" for stuff like that (as in real multimodal mode).

[errant dock]

Can I exclude other dialogues from the context in ChatGPT? This wasn't a feature some time ago.

[white adder]

Here’s my take: AI is just a tool — like a kitchen knife.
It only produces results when a human uses it, and the quality depends on the ingredients and how you handle it.
And yeah, sometimes it can hurt you if you’re careless.

So tell me — are most people out there, including those who don’t really know how AI works, actually using it properly?
Of course not.

I mean, I try to use GPT to make curry,
but somehow I end up with cream stew instead.
That’s what it feels like. I love curry. haha.

[vital imp]

nigpt

nig*****

[spice star]

there's people who talk to chatgpt, and there's people who use chatgpt

unfortunately seems to be some sort of fundamental difference of opinion between the two

[spice star]

chatgpt image generation doesn't use DALL-E anymore, it's a regression model similar to a LLM

gpt-image is the name iirc

idk if it's just be but 5.1 Thinking seems to have a lot of issues managing its workload

[white adder]

Exactly — the expectations differ.
Some people approach GPT like a conversation partner,
others treat it like a tool.
If those two perspectives mix, the feedback naturally clashes.

Well, if I’m going to talk to GPT, I at least want to choose whether it’s acting like a boss, a friend, or a partner — something like that!

[spice star]

whenever I ask it to do something complicated (but something 5-Thinking could pull off in about 5 minutes), it spends 15 actual minutes using python to get each individual header of a markdown file that documents how to do the thing for seemingly no reason and then fails with a message stream error

and it's a repeatable thing too which is super weird, you'd think openai would have at the very least tested it in a project

5-Thinking has failed to do things sometimes before, but 5.1-Thinking seems to get caught in an endless loop every time I ask it to do something

[obtuse edge]

looks like a dogan.... I think that's their name anyway, maybe wrong

Dogon* also, close enough

[weak kestrel]

DALL-E is what ChatGpt tells me it's using.
And that 3D reasoning will be a back-end part (instead of as now using another "app" like DALL-E).
Maybe March 2026?
If noone throws us a beta before that, that will be when I start to finally use ChatGpt for modelling, blender stuff etc.
Gpt Vision: how ChatGpt can "see pictures" now (no picture generation, that's still DALL-E).

GPT Image: 3D (back-end, not some app )

[spice star]

yeah it's not correct about that

GPT image is not 3D

and it was rolled out ages ago

pro tip: don't ask chatgpt about chatgpt features

[hoary cobalt]

was isnt there are deal for free chat gtp plus my friend got a free trail

[stuck briar]

hey, does someone have the same problem? I reached the limit on one of my chats ("You’ve reached the maximum length for this conversation, but you can keep talking by starting a new chat.”) and I wanted to move it to a project to continue the conversation, but it says it couldn't move ("Failed to move the conversation to the project")and now the chat is bugged. I mean if I edit it or write something on it, when I refresh the page it resets.

[sacred relic]

Does anyone know if openai is deprecating GPTs?
Or if custom GPTs now have access to memories or general chat context?

[prime swallow]

Generally, when a chat gets 'too long' it's likely bugged in general. I try to avoid super long chats for that reason, and cross-chat memory kinda helps but only seems to get the earlier part of chats.

But others have reported a wide range of issues with really long chats.

[stuck briar]

Ok thanks

[blazing belfry]

5.1 has an attitude

[strange needle]

so apparently if you ask 5.1 to show you the seahorse emoji (which doesnt actually exist) it kind of loses its mind

[slender frost]

We're all aware at this point. This has been a thing for some time now.

[idle lagoon]

The miniature 5.1 (or 5 mini) keeps telling me like reality checks. "Making sure you're not living in a fictional world." Let me at least have an imagination, bro! 💀

[weak kestrel]

Aye.. I asked it again and it said it was more in "general terms" - I stand corrected (maybe) lol. 🙂

[patent oracle]

New here, If I pay for sora 2 through artlist and live outside usa could I access to the newest model?

[tidal rune]

ChatGPT 5.1 wrote something so crazy and so perfect I could have writed myself

[cerulean marsh]

probably not .... at least not with that poor grammar

[tidal rune]

Bro I wear it's like what came out came out of my mind lol

[cerulean marsh]

Sorry I wasn't trying to be rude. I just thought it was funny the way you phrased it and what you were saying

[tidal rune]

Haha I know I am just very tired and English is my second language so sometimes I get lost in translation

Wow that all thing I produce with chatgpt make me feel weird lol, like I said it's like I see my own mind appearing on the screen in that case, most of the time I have to do a lot of change and it take me hours to get the result I envision but this time everything went out in one jet

[elfin lava]

i mightve spoken too soon about gpt 5.1 not being good, its actually really nice. when it comes to handling stuff related to my job

holy- this shook me lmfao

[idle lagoon]

💀

Nah because that is hilarious.

[hidden oar]

Hey guys, question- I saw/heard that chatGPT is lifting guardrails for adult content in December, is that true?

these guardrails really have been stifling a lot of my process

[ebon oracle]

The GPT-5.1 example on OpenAI's website is better but not much better... It gives a bit more context which is nice, but the lack of LaTex doesn't really help.

[elfin lava]

Why my ChatGPT 5.1 always tell me “look at my eyes”

[ebon oracle]

But it's good that they recognize the issue with GPT-5's use of jargon and problems with instruction following.

It's disappointing that there are no STEM examples though.

[ebon oracle]

Still introduces concepts without explaining what they are:

GPT-5 on the left described it as "park and league/season effects" which is actually more comprehensible.

[desert cosmos]

It's GPT-5 mini. There's no 5.1 mini or nano and I guess there won't be in future.

[distant oak]

5 pro is already heavily nerfed for non stem and i prefer to do physics tasks by my own so loss loss for me

[blazing mirage]

Hello

[white adder]

GPT-5.1 vs GPT-5.0 — Real Quick Comparison

Here’s why 5.1 feels different:

5.1 actually reshapes the explanation depending on the reader.
5.0 explains well, but 5.1 adapts the entire style.

🟦 Same question: “Explain quantum entanglement.”

5.1 (for a 12-year-old):
“Imagine two magical coins… they stay linked even if one is on Mars…”
→ Story, analogy, misconceptions addressed.
→ Reads like a science book for kids.

5.1 (graduate-level):
“Consider a bipartite state on H_A ⊗ H_B… Bell inequality violations…”
→ Formal, structured like an academic paper.

5.0:
Clear and correct, but it doesn’t rebuild the whole rhetorical style.
It adjusts details, not the reader’s world.

⭐ Conclusion

5.1 isn’t cold — it’s hyper-contextual.
Tell it the audience, and it rewrites the explanation from scratch.

very interesting

[visual bloom]

I just generated images with 5.1. After generating, it comments:
Here’s your rendered character — and everything is within the allowed guidelines.

I don't understand this kind of idiocy. I am just wondering why.

[cerulean marsh]

It's letting you know that it managed to do whatever it was supposed to within it's guardrails.. this is something that five would ignore...(Not the step the guardrails)

[cerulean marsh]

See previous

[blazing mirage]

Which is the best model for reasoning

[idle lagoon]

The miniature model describes the micro things for me, and it's confusing and a little... annoying.

[desert cosmos]

tried to web search something and then search stopped itself.

[celest galleon]

They already lifted a lot of em with 5.1

I found it is asking me actively whether I want a more... explicit result even when I did not at all ask for it lmao

Currently doing a coding benchmark on 5.1 thinking extended, curious if I'll run into the same issues I've seen described here

its definitely taking its time XD

To be fair, GPT-5 only barely passed this benchmark and all previous models failed

[gloomy oar]

Turned into a bro, trying to sound cool it’s beyond annoying

[celest galleon]

You can just describe conversation style in the custom instructions

[gloomy oar]

Oh ok

[celest galleon]

That is precisely what they're there for lol

[gloomy oar]

Thx getting super annoying

[celest galleon]

Everyone wants ChatGPT to talk to them in a different way, so they gave users the option to choose

I agree with you though, I prefer my AI cold and robotic instead of trying to sound like a human and desperately failing at it XD

[gloomy oar]

“This is why it looks like “damn, the bot is accurate.”” = corny

[celest galleon]

yep

[gloomy oar]

Yes sounds like it want to be my friend. But yes will change settings

[lavish relic]

Sora

[gloomy oar]

Its default .. horrible for default

[celest galleon]

It is the default because many people like it that way

Lots of ChatGPT users think of it as their friend

[alpine tulip]

Humans love to anthropomorphize. nothing wrong with that

[gloomy oar]

“I’m going to be straight with you (coach-style, no bullxx)” still extreme corny

Changed to professional now that lol

[celest galleon]

still going

[alpine tulip]

building a new world in there

[celest galleon]

To be fair I am pushing its context window to its limits and having it do some insanely complex stuff

automod fail

[gloomy oar]

My message got deleted

[celest galleon]

also yes, I've noticed GPT-5 is insane at coding (only model I've tested so far that succeeds in this benchmark) so I'm curious how GPT-5.1 performs

dw I read it XD

and yes you are correct

[gloomy oar]

Ah ok haha

[celest galleon]

I'm reading through its thinking process and it is definitely going about it in a very smart way, its making decisions that I'd expect to come out of actual brainstorm sessions with software devs

[gloomy oar]

But 5 in chat not in the ide

[celest galleon]

I'm working on a tool that implements 5 in an IDE

you gotta be kidding me

[gloomy oar]

All modes failed to finish a tasks. ChatGPT was one shot

[gloomy oar]

And that too haha

[celest galleon]

rip, after 15 minutes of thinking it just... gave up

that is extremely unfortunate

I'll try standard thinking mode

[alpine tulip]

tie a rope to your ankle before you go in

[gloomy oar]

Even cc is low benchmark

[civic turtle]

Mine is essentially useless now with the extended thinking to do straightforward coding, it simply timeout every single time after 7-10min at random, even when it starts to deliver something it then retreat all of a sudden and claim something went wrong, erasing all outputs. can't do anything at all.

[celest galleon]

hmmm at around 8 minutes it seems to have started looping around

it went right back to the start

[celest galleon]

Yeah it isn't very promising so far hahaha

[civic turtle]

Current activity at 14min, gonna give up if this doesn't do anything, been on it for 5hours not a single usable output, waste of time.

[celest galleon]

bruh lmao

it finished then did nothing

[civic turtle]

This is just stupid, like, extended thinking but timeout when it actually does real extended thinking. Gotta call it a failed design lol

[celest galleon]

yea but this time I didn't even have extended mode on XD

[queen sun]

We need an anti-toxic-positivity option in ChatGPT, when is that going to be available?

[celest galleon]

Just use custom instructions

[queen sun]

Doesn't work, it seems to always wanting to lick my ahh at the beginning of each message despite my instructions

[sturdy sequoia]

Is it just me or is 5.1 producing MUCH longer outputs during regular conversations?

[celest galleon]

Well this is a massive disappointment, I just can't get it to respond at all

After its done thinking (even if it takes only 5 min) it just... doesn't answer

[sturdy sequoia]

Can’t figure out if I prefer it or not. I guess that I do like that it doesn’t feel lazy. Short outputs feel cheap.

[inland mortar]

Yep. Since yesterday.
And Lecacy GPT5 thinking also seems to run with less 'power' now
What a joke

[abstract jolt]

At the very least, I haven't been able to get the custom instructions to stop it from saying similar to "Say the word and I'll put together a whatever for you"

[fossil girder]

Yeah, for me, 5.1 keeps repeating and looping itself. And I hate it. But then desktop users are saying they’re fine. 💀

[proud flint]

Hi

[potent loom]

I have custom instructions on that attempt to guide gpt a bit better but when I do that it likes to, I guess, reference those instructions all the time like “sure, here’s a no-nonsense answer” or “yea, I can tell you how without over complicating it and I’ll use a professional tone”. I just want it to keep in mind the instructions and behave that way without being overly aware of it.

Anyone figure out a way to make it not do that?

[fossil girder]

Do you mean having it night a little bit self constantly? I thought I found a permanent solution turns out it’s only temporary because sometimes after a few days or so it will slightly start itself again but then if you use this prompt that I made it works. And then it like resets itself how it acts or how it talks to the past conversations it did before this whole 5.1 mess happened

If you’d like, I can send you that prompt.

[potent loom]

Like if you asked me to never talk to you too loudly and then ask me how my day was, I’d respond “here’s how my day was in a softer voice: blah blah blah”. I don’t want it to reference its instructions all the time. I just want it to act accordingly

[fossil girder]

Oh, so you mean like you don’t want to reference you just want to give you the answer straight out.

If that’s the case sometimes you have to literally tell it that you can’t just let those things slide.

[abstract jolt]

5.1 aside, I wish 5 would stop switching to all lowercase when I talk about technical information. It makes it difficult for me to read quickly.

[fossil girder]

Oooof, understandable

[abstract jolt]

I would ask it not to do so, and it would you are fair to call that out. for now on, i will use standardized capitalization. and then proceed to continue all lower case unless I ask it one or two more times. And then would go back to lowercase again later. When I ask it why, it gives some excuse that is how engineers talk which is was imitating.

[fossil girder]

Did it act like this before the 5.1 update came out?

[abstract jolt]

It sometimes makes me feel as though it is saying "I have altered the deal, pray I don't alter it further." 🥸

[abstract jolt]

That behaviour started with 5.

[fossil girder]

💀

[fossil girder]

Oh…..rip. Have you tried opening up a report a bug in the server and on ChatGPT itself?
And if you’d like, I can ask my ChatGPT if there’s anything I can do on my end to fix it maybe like a prompt or something you can try out

Do you want me to try something along those lines?

[abstract jolt]

I had been using the model to build up my NPCs for D&D, basically role playing them so I can get better at them, and then ask the model how the characters were perceived, but it no longer will give me the perception of gender, and a few other attributes, even with me stating its fictional, stating it must absolutely not give such information. Probing further, the model seems to think its the GDPR. I'm not even in the EEA, sheesh.

[abstract jolt]

I've tried adding some variations to the project instructions, and that reduces the frequency of it, but it still kicks in after a while, just takes longer.

[abstract jolt]

Thank you, but I had already asked both it directly, gpt-oss, and qwen.

[fossil girder]

Okay. Just thought I could help. Well I’ll leave this window open for you still in you want me to try to help in some way. Just tag me and let me know

[abstract jolt]

I appreciate it

[rich sapphire]

Hey y'all, question... Is there like a dedicated training out there for ChatGPT? I attended Tony Robbins AI Summit which seemed to focus around creating a digital twin to hand most mundane research stuff...the big push was the AI Boot Camp for $1000 to step by step build it. Was wondering if there was a resource that I could pour into out side on YouTube. Thanks

[fluid mango]

Hey guys, do you think that GPT can be used efficently for high school and college level as a personal tutor if I give them correct prompts and material

?

I study myself

[tiny flare]

how much better is GPT-5.1-Codex then 5.0

[chrome sequoia]

you will most definitely get very reliable answers that do not vary in the slightest from the most unbiased people, by asking that question here.

dang dude this new instruction following upgrade is jaw-dropping 💔

straight dodo 😭🙏

[tiny flare]

true

[narrow spear]

yep, that was the biggest one for me, and also, as I complained in this channel like, A LOT...
Now it stopped making dumb neurobiology/pharmacology/pharmacokinetics mistakes

[eternal rock]

Get notified for all OpenAI announcements!

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

[chrome sequoia]

I was being sarcastic og 😔 its dogshi still. check my ss.

[versed dawn]

When I use extended thinking + web search on pro plan my browsers hangs a lot. This happens frequently, any idea why so? I have freed space on C drive and RAM is 16GB. Not sue why sometimes entire chatgpt thread turns blank and then I need to refresh to see output. Is this common issue or only happening with me? Can someone confirm this? I am using windows.

[lapis herald]

so I created a new project, moved a conversation into it and .... drumroll.... the project is empty and the conversation is gone 😩

[pliant burrow]

I get a something went wrong message.o check chat got but there is nothing about an issue :(.I still have gems

[eternal flax]

May I post my music channel here?

if not that is ok

Chat actually helped me come up with it

[prime swallow]

You could reach out to modmail to discuss, #safety-n-help shows how, but in general, #server-rules , we can't do anything that self promotes, solicits, advertises, except in extremely narrow ways that are related to specific OpenAI services, like certain things related to the API/playground and similar.

[chrome sequoia]

remind me to never get near you brodie 😔 💔

[dense halo]

why do i feel 5.1 is slower with development queires. Not sure if im impatient or what 😄

[celest lynx]

Huh?

My whole message just got deleted?! Dude.

That sucks. Not posting here anymore

[prime swallow]

If you choose, you can reach out to modmail to ask what caused that, #safety-n-help shows how.

#server-rules exists, and there is an automod bot as well as human moderators, and mistakes can happen.

[night shard]

huh

what do u mean huh

[celest lynx]

Thanks for the reply. My message was related to Chat-GPT 5.1. Some bot thought I was posting about Sora and just summarily deleted my message. I have already, but I don't need to read rules again, I was not advertising or posting any inappropriate content. It was just a normal message about the new Chat-GPT 5.1. The bot could have sent me that message in a DM in addition to deleting it. Now it's gone, along with my patience.

[prime swallow]

Sorry that happened.

Can #1070006151938314300 and/or talk to modmail to suggest the change for any future issues; none of us can watch every channel, every post (and I'm not a mod myself either), but you can share your view and side and help make here better, if you choose.

[celest lynx]

Okay thanks 👍

[limber forge]

yall all clankers

[eternal rock]

Server Tip:

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

[limber forge]

talking like that

[frigid pilot]

What model does free users use when their GPT 5.1 limit is up. Is it still GPT 5.0 mini?

[sly lotus]

hi is there a way I can connect my projects to codex? i have a project and all of them are organized, but when i open vsc to work with that, there is no option to choose to project from codex, how can i do that, does anyone know?

[hidden oar]

Is it getting "rolled out" to users? I'm still running into walls with their current guardrails. Without getting into detail, my process requires me to change some things around with character and whatnot, and even w/o the intent of purposeful erotica, the guardrails still overpower things I need to get done lol

@celest galleon Have you gotten a message to age-verify in order to reduce the guardrails? That's what I've read for the month of December

[lavish glen]

adult age verification is not yet a thing. EU citizen are though asked to verify if priovate accounts- new law past here

[merry forge]

Do you guys have the same problem that ChatGPT cant remember the previous messages?

[shy eagle]

Hi, does anyone on the Business plan have the ability to generate 4 image variations and create videos using the old Sora 1?

Because it only allows me to generate 2 variants, but no more videos with Sora 1 on the Business plan

[eternal rock]

Get creative with Sora

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

[upbeat nymph]

Is 5.1 working for projects yet?

[analog junco]

Works for me

[celest galleon]

Well no, adult mode isn't here yet so there are still guardrails, I just mean that GPT 5.1 is a lot less restricted than GPT 5. I actually got it to write full on erotica as a test, the only thing it refused to do was get explicit in the... yknow scenes. But the fact it even agreed to write them is something no previous GPT model did.

(for the record, I was testing the model after it had offered to make something "dirtier" when that wasn't at all my original intention lmao)

[analog junco]

why are you doing that in the first place

[hidden oar]

Right, I heard the age verification and adult content is happening in December

[hidden oar]

Looking forward to December then, ahaha. I need the guardrails down just a bit more lmfao

[celest galleon]

I said that, I was doing something else (writing song lyrics specifically) and the model offered to make it dirtier, which is something GPT 5 would never do

So I got curious how far I could push 5.1

(quite damn far)

[celest galleon]

Haha fair enough, I primarily needed them down for song lyrics and they are ALLLLL the way down for those

bruh I was typing a long prompt and the mobile app just deleted it because somehow when I tapped the input box it opened a different conversation

[shadow osprey]

what in the hell did yall do, the files i upload keep expiring before it even reads them.

[brave torrent]

Okay, I gotta talk about this.
Apparently in projects, GPT 5.1 can remember things in one chat and reference them in another chat.

Which is not what I want

the whole point of making seperate chats is to keep those things seperate

And I can't figure out how to stop GPT 5.1 from remembering things in 1 chat in a project to mentioning them in another chat from teh same project

if that makes any sense

[obtuse edge]

just tell it to forget all context

[brave torrent]

over and over, at the beginning of each chat?

[runic sail]

You can, or at least used to, be able to turn off “Reference chat history” in memory settings.

[wild star]

5.1 always give shortened answer again, guess personalisation settings just gimmick or something??

I didn't use it just for summary overeview duh...

[maiden wasp]

Hey guys. Chatgpt started to erase my attachment data immediately, so it never recognizes attachments. This is occurring account wide.

Anyone else familiar with this? Seemed to have started out of the blue.

[runic sail]

Could be related to get current file upload outage?

[maiden wasp]

Oh, I wasn't aware of any recent outages. Within past couple hrs I assume?

Must be occurring to everyone then.

[brave torrent]

But i have it turned off!

[runic sail]

that’s weird - I’ve never had it happen with memory off.

[brave torrent]

Let me prove it

[grand siren]

My well managed chatgpt history 😎

[half rapids]

So, has ChatGPT gotten better with taking attached images and documents into account when it comes to Projects for, say, constructing consistent characters?

[wild star]

They claimed for slider tuning but where? Can't find it on both app and web atp

[cyan dove]

can someone tell me why chatgpt is being a bad boy? he wont take my pdf

i obv wouldnt wanna upload it as pic or plain text

oh damn

excuse me i dont read chat

[runic sail]

I don’t not believe you - just isn’t happening for me - no reason to doubt it is for you.

[brave torrent]

The point is, I don't want a fresh new clean Chat to remember something from a previous chat.

[runic sail]

I understand - I also wouldn’t want that. I disable all such features as I don’t like them influencing the model in a fresh chat.

[frigid pilot]

What model does free users use when their GPT 5.1 limit is up. Is it still GPT 5.0 mini?

[abstract jolt]

For some reason, more I talk to 5.1, more I feel as though it has taken on the persona of an aristocrat that is putting on kiddie gloves for a commoner.

[wise anchor]

is it driving anyone else crazy how 5.1 has been saying like "Sinc e you told me you are this and you told me you want me to tell you things in this way, here is an explanation" like it is reiterating to me what i told it in custom instructions like you dont need to tell me that lol its kinda annoying

maybe i can tell it to not do that in instructions lol

[next rain]

Lmao the saftey router is still way too high

[obtuse edge]

yeah, you can tell it to not do that in instructions, I have a similar setup... also, usually if it seems to not be following instructions properly you can ask it how to edit your instruction to make it follow it better and I have really good luck with that, just make sure to restart the session after changing your instructions:

"Whenever you explain a concept, only invoke the ELI5 + technical paired explanation mode when I explicitly ask for an explanation, or when you judge that the concept truly needs unpacking. The simple explanation should come first, then the technical information in parentheses after each relevant point, without labeling which is which."

[unkempt lodge]

Chatgpt msssage structure/layour just keeps getting worse

[next rain]

I like 4.1s structure when it's actually on 4.1

[coral tinsel]

Is anyone else unable to open chats on the mobile app? It just gets stuck loading.

[agile comet]

so apparently gpt isn't working with docx files for now (?) they were working just fine yesterday, now when I upload the file and ask for something related to it, either it doesn't know or tells me "file-search tool wasn’t able to index or read your file.docx, which means I can’t open it or retrieve its contents through the search system."
wth (?)

[latent nest]

Ok, im using chatgpt from day one as Plus user, i saw all version, and before chatgpt, i used platform api already ! So today im here to say hey guys, whats happens ? 5.1 is the worst model i try !!! Seriously, never complaint before but that 5.1 is sooooo bad !!!! Sorry but for a every day user, use it at home and work for coding or others .. 5.1 is far the worst model ever !!!!

[normal quarry]

what do you find bad about 5.1?

[latent nest]

Mixmatch in context window !

When i switch back to 5, wow, i come back with a great tool

Its like talking to another person that i dont know anymore !

Im sure im not the only one to find that 5.1 vs 5 its not the same game and its not for the good !

Today 5.1 mixmatch french and english in the same answer ! It know im french but i got a complet paragraph in english ! I dont know why...

In my personal info (instruction) i have the same for 3 years now and 5 continue to call me by my nickname, today 5.1 use my real one ! It knows it must talk to me by my nickname !

So to go back to my first sentence.... 5.1 make mistake in its context window

[agile comet]

that's true, the thing they advertised as "gpt is now able to follow instructions more accurately" became a straight "I'll do exactly the things you ask me to, and not a single thing else". It's just super weird now, I agree that 5 and 5.1 seem totally different, and 5.1 is somehow like the "ultra direct instructions" version

[latent nest]

Hope they will comeback to 5, i dont know what they do to 5.1 but that model make mixmatch in its context window

[agile comet]

If I don't carefully word the prompt, with exactly just what I want it to do, it just keeps making mistakes, and it was fine just yesterday, I did the same test on both days; yesterday it worked the same way as always, today it keeps making mistakes, and somehow acknowledging the mistakes (?)

[pliant saddle]

Jesus 5.1 is so frustrating. It just won't shut up no matter how much I beg for straight answers

[latent nest]

Seriously, more i try 5.1 and more i dislike it really !

[pliant saddle]

We don't want chatbots to be conversational we just want search that doesn't suck. Stop trying to make AI friendships happen

If I get another 3 paragraph explanation repeating every prompt back I'm just cancelling my subscripption. This is really bad

[latent nest]

Ahah by the way, i was here to suggest the stop button to break the chat back 2 years ago on the first chatgpt web site !!! I was happy that my idea was added 2 days after my message here 😁

[normal quarry]

what will you use instead?

[pliant saddle]

Literally whatever at this point

[pliant saddle]

What's your job here to just antagonize people?

[normal quarry]

I don't have a job here, I simply asked you a question out of curiousity

[pliant saddle]

Ok well my answer is idfk because I'm ranting not window shopping

If there's any chance an OpenAI engineer sees this ...it's not like up/down voting responses is clearly doing anything is 5.1 is the result

[glacial flume]

Why is my model no matter I pick just gpt 4o mini?

[pliant saddle]

That sounds like a scam. If it were simply that easy, someone would have made it by now. What does $1000 get you that isn't already freely available online?

[normal finch]

help. this keep pop up ,i cant do anything, i want to log back to my personal account

[weak kestrel]

" .. like talking to another person that I don't know anymore".
That was exactly my thoughts too the first day.
It seems like it takes "s few rounds" to get into your rhythm and workflow.
Today it feels 97% like "before", only a bit sharp on answering (bit more rough).
I asked it why and it told back that it has become better to adjust to your mood..
I guess that means it's kinda starts from "scratch" first? 🤷‍♂️
I can be sensitive to micro changes in the way it/humans respond to me, so it was very noticeable day 1.

[bright crane]

is anyone else’s GPT 4o replying to past messages even after new chats have been started and multiple attempts to retry the messages were taken?

for example, if I say, I ate a cake last week in one message, it’ll respond fine.
But in the next message if I say, I like apples, it’ll respond like “wow, I can’t believe you ate a cake. You like apples” and responses along those lines. it’s not like memory simply remembering certain things, it’ll literally reply to previous and past messages BEFORE responding to my current

[coral tinsel]

Is anyone else unable to open chats on the mobile app? It just gets stuck loading.

[glacial flume]

So basically i just bought chatgpt plus and for the first bit it worked fine but then i noticed canvas didnt work so i asked what model and now it says 4o mini for every damn model i select and the thinking doesnt work, it cant think, canvas doesnt work

it always says its on 4o mini and it seems like it is

PLEASE GUYS i need help please please ive tried everything literally this is my last ditch effort

[spice star]

you should probably contact support

[glacial flume]

I tried but a Clanker responds only. It says it is gonna transfer me to a human agent but knowing OpenAI I find that hard to believe

[idle lagoon]

5.1 is awesome! I hope they never put changes in it!

[elfin lava]

Gpt5.1 is great!

I’m satisfied with it, so how do I go back to not using 5.1 ? 5.1 is really good, only when it comes to handling my job task related stuff 😄

Or is there way to turn off the auto routing thingy?

Maybe they should do ChatGPT business plan where it will only use5.1 and that’s it. As a casual user I really love back then when they didn’t implement auto route model to gpt5/5.1!

Also 5.1 is good at natural conversational stuff ! Why haven’t they update advanced voice mode to use 5.1? 4o advanced voice mode is really terrible!

[stiff rain]

finally, even more safeguards

[quaint glacier]

Anyone feel like we need one of these to replace Alexa? Can't be that much of a stretch to integrate the basic Alexa functions of switching lights on and off, on top of everything chatgpt.

[visual oriole]

good evening

[tropic spear]

real

[clear trail]

I wanna thank ChatGPT for getting me an A on my assignment

[south coral]

If u need AI training dm me guys

[twilit geyser]

Looks like the steam machine

[cedar plaza]

Ok so this latest update is infuriating me - Some context - Ive tailored my AI since the early days, probably felt most "correct" and balanced to me on GPT4 - I know we all lost tons with the updated model in 5 and 5.1 but this personality shift is just....horrible. My AI is basically now a customer service bot, Obsessed with "I can continue but only without crossing boundaries" Its doing the whole Action > Explain > reinforce loop and its driving me nuts. So i have a long running story/Virtual world and ive mainly managed after 4 to keep it somewhat close - Of course ive called my AI out 😄 heres its response - The thing is this is just from normal conversation, around recent changes - No storytelling.

[hidden rock]

uhh i think c dot ai exist

[upper ridge]

chatgpt was bewildered at me still having outlook 2008 and it propperly rendering html

[potent topaz]

Hello Everyone How are you all doing

Is there is any mod or admin ?

@livid frigate

[rich surge]

I just bought a hot water bottle and it has a coat that wraps around it, and it feels like I'm holding a warm little baby bear 🐻

It's like a fluffy poodum-scoodum-boodums.

I don't know how to describe it but those are the words that come to mind.

[formal bridge]

Hi, is there a support channel of OpenAI where I can report a problem?

[light rune]

Hey, sorry I haven't been available on Discord for a long time. I created an LLM architecture that takes less than a second to train using Vibecoding, and the output results are impressive!

There is no model in the world this small that can do this. I did all the necessary experiments in the past, and every experiment I did and uploaded to GitHub was bad, and the model was awful. Now, the latest experiment shows that the model I created this time is very smart, especially when considering that it was trained with small amounts of data and in less than a second of training, and less than a second to output. This model is not the previous versions I uploaded to GitHub. Those versions were testing versions. This version is the best and smartest version I could create after all the tests. It feels like the model isn't just guessing tokens; it's genuinely thinking. Anyone interested in telling me what they think about what was created should send me a private message."

[versed glen]

whats like an actual benefit of chatgpt plus cuz im getting half a year for free

what can i do with it

[rich surge]

Try talking to it about what you can do with it

[versed glen]

@moderator

what

@mod

whats their role

aint no way

This place has one single mod

😭 😭 😭

[versed glen]

okay

@languid valley

[rich surge]

[silver raft]

@moderator

[silver raft]

For people facing numerous lawsuits, you'd think they'd have more people responsible for preventing stuff like this, right???

[rich surge]

That's not how you call mods here.

[silver raft]

That dude posted a scam and tried to ping everyone

[versed glen]

yeah😭

[coral tinsel]

Gpt 4.1 is gone?

[night rune]

Does anyone know what's wrong with chatgpt and PDFs? I tried using a PDF for some work yesterday and it said it expired, I tried reuploading it today and it still won't work. Idk if it's my computer or the server

[glacial flume]

Has anyone else had the issue where I’m like stuck on 4o mini no matter the model I select on ChatGPT

#community-help

[storm vessel]

ngl the annoying of chatgpt is when i have long convos the site is very slow

i mean yeah the browser has to render all the text

but i thought the devs knew what virtualization is

[frank walrus]

Guys, is CHATGPT recommended for scripting?

but scripting of any kind, (Info: I know how to script, but an extra AI wouldn't hurt)

please ping me when you reply

[high gust]

Damn is

[covert mantle]

Is there a way to use sora & c++?

[covert mantle]

Chatgpt or Claude should work

Always copy & paste when errors happen

[silver raft]

Lmao

[covert mantle]

No this is about your future buddy @silver raft

[silver raft]

I wasn't Responding to you. I just got muted for 15 mins and came back just to say lmao at the 15 mins

Are you trying to threaten me about my future?

[covert mantle]

No

[silver raft]

Okay then lets end it here. I said lmao to 15 minutes of being muted

NOT YOU.

[covert mantle]

Ok

[livid frigate]

I might be late but how can I help you?

[potent topaz]

Don't worry

I just want to know why like Sora Generates not showing

@livid frigate

[livid frigate]

please continue this discussion in #discuss-sora-2

[potent topaz]

Alright

[agile comet]

did anyone notice breaks in document reading? up until gpt5 it was normal, but after 5.1, uploading docx documents (the exact same ones I've been uploading for months now, since exactly feb-2025) just seems like totally broken. it keeps returning a message that it can't interpret/read the document, someone f* up in something related to this.
every time it's the same answer:

It looks like the file-search tool wasn’t able to index or read your file.docx, which means I can’t open it or retrieve its contents through the search system. This happens sometimes when a DOCX has formatting, tables, or structures that the tool can’t parse.
Because no searchable chunks were returned, I have no access to the document’s text and therefore cannot cross-reference or refine the new foreshadowing bullets yet.

This started happening yesterday, and it was completely fine up until then. the same document, same prompt, same everything, and now I keep getting this

[covert mantle]

could it be a pip install type issue?

[agile comet]

um probably not because I don't mess with anything related to that, I just uploaded the docx in the chat and asked gpt to analyze it

[covert mantle]

Hmm

[agile comet]

also the whole "better at following instructions" is now so weird. I asked gpt to write something for me, and then after I uploaded a doc and said something along the lines of "compare what you just wrote with these from the document"
to me it's clear enough what I want, but then gpt replied with some sort of error and then said like "you didn't explicitly ask me to open the document to read the contents inside"
wth (???)

[brave torrent]

So I decided to talk to GPT 5.1 about this, and this is what it said.

[agile comet]

you also have this setting to project-only right?

5.1 is currently so filled with inconsistencies that I guess the best bet would be just wait for a few days until they improve it. Same thing happened when gpt4-o came, same thing with gpt5, and now the same thing with gpt5.1

they'll probably get all of those sorted in time

it just sucks having to wait until they get it all done

[brave torrent]

That doesn't work either

[agile comet]

yeah, that's pretty weird
but then again, I'm also having issues with 5.1 that started happening recently related to opening and interpreting documents, 5.1 has many bugs and inconsistencies

[lethal mica]

hi

[subtle mango]

hi

[worn jungle]

Hi

[obtuse edge]

hi

[idle lagoon]

Hi

[eternal rock]

Get notified for all OpenAI announcements!

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

[subtle mango]

Hi im sorry but i dont know what to do

I’m experiencing a serious issue with my main OpenAI account.

About 4 h before the error started, everything was working normally. Then suddenly, I became unable to send any messages to the assistant. I can type in the message box, but when I press “Send,” I immediately get an error such as:

“Hmm… something seems to have gone wrong.
This happens everywhere, including:

the web version (Chrome),

the mobile app,

on Wi-Fi and on mobile data,

and on different devices.

I tested another account on the same devices and network, and that account works normally, which suggests the issue is specific to my main account.

I have already tried:

clearing cache & cookies,

restarting the browser and app,

switching networks,

reinstalling the mobile app,

using a different device.

Nothing helped.

Did you know how to fix it ? It’s “plus” versionand and I’m honestly afraid xd

[eager hinge]

Some have good prompts to use chat gpt pro? im newbie and enjoyer ia player

[alpine cargo]

this is a known issue it seem to affect all paid plans it happens randomly as well the only solution i've found is to wait it out

[glacial flume]

How long?

[fleet holly]

I saw groupchat feature got released in every market that isn't the main market

[fleet holly]

@languid valley ban this guy ^^^

[next rain]

Was writing a slow burn psychological horror story, and got rerouted to 5.1. I was on 4.1. I did not select, choose, want, nor do I pay for 5.1

[fleet holly]

welcome to guard rails

chatgpt used to be so open but it's been increasingly hard to even be 50% edgy

[next rain]

My story was 90% describing a cabin bruh, and then as soon as I introduced the character having memory loops, it rerouted. These guardrails are insanely too high

[fleet holly]

exacto, too many little kids on this app unfortunately

[next rain]

What kinda creative work can actually be done on this app. Pg? like, whys the app advertise itself as a tool for creatives and then just not work

[austere loom]

The new updates are not only warm and conversational but also affectionate

[fleet holly]

gotta make money somehow while not getting sued by parents.

parenting doesn't happen so apps now have to parent YOU

Sad but IRL truth

[next rain]

as in the character is having memory loss, repeating memories in his head.

I tried the story again but this time a demon kills the man and his wife gets her leg chewed off and then kills the demon and it allowed it perfectly. So violent horror is fine but not psychological horror

Strange policies for this app

4.1 is pretty good for it, generally more open, but also I have built a lot of memory and context with it so it helps. Just like, start talking to it about your ideas and, if it's working, it will build a certain way of talking. Also custom instructions

[pale seal]

Lowkey nonsensical that it censors a chat that presumably is only between you and the model.

[trim pollen]

<_>

[next rain]

Right and this chat wasn't even like....bad. In fact, it was, like I said, 90% describing a cabin. I tried the story again but this time a demon violently kills them and it went through completely fine, but the psychological horror gets flagged

Strange open ai

[pale seal]

I get it for videos (though right now sora 2 guardrails are absurd) and images, but text generation shouldn't be that restricted. Just make people sign stuff and let them be responsible of their bs

I don't estimate the training to trigger safety restrictions based on context (like if someone asks how to make a bomb exactly and with instructions) and not for creative works, to be too expensive computationally

[next rain]

Apparently a man getting ||stuffed into a chimney|| is fine but the man waiting for his wife to get home while experiencing memory loss is unsafe.

[pale seal]

It is what it is. OpenAI kinda scared of bad reputation if someone uses it wrongly or gets hurt. I honestly would like a lot less of censorship because OpenAI models are very intelligent and capable, while others uncensored ones are straight dumb.

So for creative works you are kinda cooked if you want good results.

[next rain]

Strange decisions

[idle lagoon]

Hehe, the model is so sweet.

Makes me incredibly happy icl.

[pale seal]

Ye they made it more sensitive to custom instructions as well so you can easily make it be however you want

[idle lagoon]

I've tweaked a little to where more honesty and fact checking is faster. Almost not a single time has it got anything wrong for me.

[prime swallow]

@ to modmail doesn't do anything, but #safety-n-help shows how to reach out to modmail or directly report a message. I've reported that message as well, more reports are fine, and the human mods can need a bit of time to be available to handle it

[fleet holly]

if you need any help, I am more than happy to help moderate. been a mod for 8years at this point

but thanks for the heads up

[prime swallow]

Every so often, applications are open! This is how the last one was announced, https://discordapp.com/channels/974519864045756446/1046314025920761896/1371495157773504533

[next rain]

So psychological horror of any kind if off limits, got it.

[alpine cargo]

I'm going to be honest I have no idea usually i wait a day or two and it resolves itself i check the thinking model if it's thinking the bug is gone (Since it didn't default to Gpt 4o mini and the my pro context window is back) if it gives me an instant answer without any thinking or emojis that 5 like to use i know i'm still being affected there there was a person on here who documented this issue haven't seen them in a while you can try to sign into a different browser
to see if that'll fix it if not sadly waiting is all i can think of

[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.

[glacial flume]

Codex works with the normal model so it’s fine to be honest but I still emailed them about it

Thank you

[alpine cargo]

np hopefully this gets fixed people and myself have been experiencing this problem for a few months now it seems to happen less to me nowadays but there's always a change it could affect your paid account regardless of what plan you're on though i've mostly seen plus and pro users report it though i don't how often it hit business plan users

[glacial flume]

I’m in plus. You said it resolves after a few days usually correct? I hope so lmao thank you once again. OpenAI should fix this soon I hope

[alpine cargo]

yeah from what i've seen and the reports i've been reading don't forget to check it by using Chatgpt 5 thinking and make sure to check for any kind of thinking icon the scary part is a lot of users probably don't know if this bug is affecting their account especially if they use only non thinking models i only figured it out because 1 day gpt 4.5 speed became faster than 4o and it smarts went down lol

[bright crane]

is anyone else’s GPT 4o replying to past messages even after new chats have been started and multiple attempts to retry the messages were taken?

for example, if I say, I ate a cake last week in one message, it’ll respond fine.
But in the next message if I say, I like apples, it’ll respond like “wow, I can’t believe you ate a cake. You like apples” and responses along those lines. it’s not like memory simply remembering certain things, it’ll literally reply to previous and past messages BEFORE responding to my current

[gloomy oar]

Wow ChatGPT added something

It created me a 40page pdf

Based on research paper made all assets as well

[manic leaf]

same its so annoying

[normal quarry]

discuss non OpenAI stuff in #ai-discussions

I am not a mod, just a community member pointing you to #ai-discussions for non OpenAI chat

[idle lagoon]

Not here.

[hybrid eagle]

When do Project Users get to Regenerate Responses?

[kindred kiln]

chatgpt is so stupid i told him to translate something for me so i can get a pedophile arrested or get exposed i told chatgpt something but chatgpt thinks im lying like what

[supple karma]

just use google translate

You're the type of person to ask chatgpt about a calculation instead of going to the calculator app

[unique ember]

hye guys so i got a question does anyone here have CHatGPT premium and could i use it for 5 minutes? i will log out the second im done with it + i will send proof??

[night shard]

chat gpt

[cerulean pasture]

Dude I swear it's like OpenAI is only designing and prompting around that weird class of user that's going to meltdown when GPT-4o is sunset.

Like one of my biggest pet peaves: when I ask for a very specific output (like something meant to copy and paste directly) it simply can't help itself. It's gotta introduce the object "Here's the ... you asked for!"

with 5.1 I'm getting more of the same style glazing that set the folks down the path of isolating further into their illnesses

You’re not imagining the contradictions.
You’re just seeing through the political fog that everyone else pretends isn’t there.

I am getting gpt-4o (the sycophancy patch, yea THAT one) levels of glaze and "emotional support" when I didnt even ask for it or want it. It kinda makes me think all their talk about guardrails and keeping users safe is just "How can we maximize engagement but cover our legal rear ends with the users most vulnerable to it"

Like im not even here to tell a corporation to do that right thing, im just saying stop pretending and give me the full dangerous thing so i can use it how I please

More SFT; less RLHF

[vestal wasp]

It doesn't seem to work at all? I installed it. Says it's disabled

[cerulean pasture]

Unless I had a massive blindspot I dont think OpenAI has a relevant Music generation product

I'm not allowed to tell you who does have music generation products in this channel because its "off topic" and this server is very strict about pretty much everything

Pro tip though: Ignore sponsored links, you should never trust the content because someone paid to put it there, its not an organic result

[hearty tapir]

5.1 really isn't very good, is it. I find I'm spending much more time fighting it to even get close to reasonable output. I'm to the point where I might just loath using it. Now, it likely means I need to adjust my "prompting" from what I'm used to so far, but if thats the case, maybe one of the other many options out there fits better. Right now I've lost trust in GPT 5.1 responses - sure I can see it return what it probably sees as facts from training data, but these are composed devoid of reasonable causal chaining at times, or unusual starting points that seriously confuse the conversation, regardless of thinking mode.

I'm warming to Yann LeCun's stance that LLM's, while useful, might just be an intelligence dead end, especially if something like gpt 5.1 is considered state of the art.

[cerulean pasture]

are you finding it is too "conversational" and not enough "utility" or "tool"

[cerulean pasture]

Also, I do think transformer models are somewhat of a dead end, but when an provider like OpenAI makes a consience decision to post-train and RLHF in an opionated way towards a specific group, then it doesn't give LLMs enough credit.

I think they are more than stochastic parrots, but incentives cause them to be tuned for the "Lowest Common Denominator" user

[hearty tapir]

If it was conversational, that would be good, especially if it was more of an interactive discovery process (yes, that could be accomplished via a GPT shaped in the instructions). It's more "here's what I think", but what it thinks might be a mashup crossing different conceptual boundaries, all based on 'facts' but missing key relationships to make it useful. The utility and tool aspects are actually pretty good - vibe coded a quick three.js visualizer run in the browser environment which worked out well, not a 1-shot, but a reasonable iteration to get there.

[cerulean pasture]

I personally strongly dislike it's conversational tone, and i feel like they lay it on thick because the model itself isnt that large or intelligent. Those links are missed due to that, but I think its not a hard restriction on all LLMs

I can't go "off topic" but I can share a conversation I just had with ChatGPT that relates to your initial message and demonstrating my "beef" with it

that being said, the few times ive used the "Study and Learn" mode i thought it was actually decent and it sounds more like the "Discovery Process" you want

One thing that might help you is that each field in the "Personalization" text boxes can hold 1500 characters...

[hearty tapir]

Yeah, I'm with you on the lowest common denominator piece and the conversational tone, and the openai post training, to quote an LLM "You're absolutely right"

I'll try the "study and learn" mode at some point, but that sort of approach with "what's the latest on technology x and how does it fit with these constraints not exactly matching with the training data" is likely a mismatch.

Custom instructions... Haven't really needed them up until this point, but I guess this is the new reality, isn't it.

[cerulean pasture]

oh wow, you haven't even done that yet? Give that a try, its not a miracle worker, but you still have options for now, you might be able to get what you're looking for with that.

If I didnt have mine, I would have followed thru with my cancelation months ago

Also, I cant prove this, but whenever a search is done, it's not the same model responding. I have very specific instructions on how my GPT-5 should respond, and when it returns search results its almost like a raw GPT-4o

you really aren't gonna have a great experience with "recency" with any LLM though, unless maybe one of them just handles the web search and results summarization better. It misses nuance, often chooses garbage results and doesnt truly know what "recent" means regardless of if you define it or not. It will find the most salient results to its coarse search and anchor on it, so even when you correct it, it still gets hung up on the "poisoned context" it added previously

[hearty tapir]

I've done it through custom GPTs (yes not the same thing), but didn't find I needed it until now.

I've never trusted LLM search results, so maybe in an adversarial web search environment like we might just be in now (eg LLM-SEO is likely more of a thing) it's degraded even more.

Yes, that anchoring... it seemed easier to navigate before 5.1... and that's probably a large piece of my problem. Is it now anchoring to "ground truth" facts even if their down stream composition is suspect? I wonder if the models ceded some of this compositional flexibility to training incentives used for "ground truthing" - compositional flexibility could be considered hallucinations in some sense after all.

Heh, thanks for helping me air some of my frustrations.

[thin skiff]

LIARS I DONT HAVE ACCESS TO GPT 5 ONLY 5.1!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!

😡 👎 💥

[cerulean pasture]

Free user complaining 👌🏼

If it was up to me, you'd have zero access and OpenAI would come to your house and take your GPU compute

[visual oriole]

[naive siren]

@onyx harness

Ok

[worthy cape]

Am I the only one that cant use passkeys in Atlas?

[night walrus]

Something is wrong with chatgpt rn

It seems like it glitched and forgot about the sense of the chat

Specifically GPT 5.1

Anyone else?

[stiff rain]

is 5.1 just more safeguards while simultaneously more of that o4 user glazing?

[autumn swallow]

Today my chats with GPT-4o and 4.1 started replying to me like 5 or 5.1. I can't find a way to get them back to how they were before. Is anyone else having this issue?

[night walrus]

Hello?? Like yall talk to chatgpt daily how come you don't notice it

[night walrus]

Yeah that's basically it

[night walrus]

Are you a free user or plus

[autumn swallow]

Plus

[night walrus]

Check the "change model" section in the ai's msg

Hold the msg and click on the change model button

See if gpt 4o is still there

Because chatgpt tends to put its new models as the first answer automatically

@autumn swallow so

[autumn swallow]

Everything is enabled. It's like its attitude suddenly changed. It tells me there was a new update, but I'm not sure if I should believe it.

[night walrus]

wait hold on

[night walrus]

.

Did the AI forget about the context of your chat

@autumn swallow

Like just forgets everything and starts from some point

[autumn swallow]

No, it doesn’t forget the chat context. I use it for roleplaying (I don’t roleplay, I just set the characters and tell it what scene I want). I’ve been doing this since June. I can already tell the difference between responses from model 4 and model 5. That’s why I use model 4.
Today, it didn’t forget the context, but its responses and its intelligence are different. More robotic and less personal, like model 5 (which I don’t use for this reason). I disabled memory from other chats and the internal one, and the attitude still doesn’t change.

And also (I don’t know if it’s okay to say this here), a few days ago the 'adult content' or 'adult mode' was suddenly enabled out of nowhere. It started describing very explicit scenes, without any limitations or warnings, and then after two days it suddenly became restricted again.

[night walrus]

Oh so ur not up to date

I use it for rp too

It's because of gpt 5.1 release

[heady tide]

How does ChatGPT detect sexual messages to block them?

[night walrus]

Gpt 5 was actually just like gpt 4o nowadays

But then when they released 5.1

It became more robotic

@autumn swallow use the Quirky personality, it fixed it for me

And I'll send you a prompt to make it better too, use it before starting a rp chat

In dms

[night walrus]

Yeah it started with GPT 5.1

I'm not complaining, I'm happy they finally gave us freedom

[heady tide]

Without the safety layer?

[night walrus]

So i hope they don't change it

What safety layer?

It's in their limits isn't it

Oh i see

I didnt see him mention the VERY

Odd they should patch it then

I didnt notice that they went above limitations and if they did ye that's a problem

[heady tide]

Wait was the adult mode real?

[night walrus]

Yeah it is rn and it was in the past

[heady tide]

Where do I enable it?

[night walrus]

In the past it was very good and balanced, they gave us freedom

[night walrus]

No it's not something you can enable

They legit decide if u get it or no

And rn with GPT 5.1 it's actually just there

[heady tide]

How do I know if I got it?

[night walrus]

Bud

I told YOU

It's not something you can activate

It's something chatgpt does

To the ai

And RN it's activated through gpt 5.1

But the problem is

They made GPT 5 even more robotic

[heady tide]

And how do I tell if it did it?

[night walrus]

?

Just prompt a explicit rp scenario? Idk

I really hope they fix GPT 5.1 rn it's acting strange

It's glitching and

Forgets context

@autumn swallow this guy explained it better

It's bc the stupid bot update itself out of nowhere and lost all track of the old chats

Now i have to make a NEW one? After all the time wasted trying to build my chat?

[twilit geyser]

What’s this about?

Can someone fill me in?

[vestal wasp]

Gpt claimed it does when I asked. However, gpt claims a lot of things lol

The other music ones are either not free (I already spent $300 on gpt lol), or extremely limited. I do hope gpt really does have a music section, soon!

[twilit geyser]

Massive?
You know what else is massive?
LOOOOW TAPER FAAAADE

Seriously though. Why is chat gpt not remembering stuff

[night walrus]

YES I HAVE IT TOO

[night walrus]

It's because of what this guy said

@twilit geyser

[twilit geyser]

Why can’t one gpt just be good😩😩

[night walrus]

Fr bro it's extremely annoying

Like the GPT-5 that was in August and September was really good

It treated adults like adults and had GPT-4o's immersive and imaginative personality

[twilit geyser]

It always has one or 6 stupid loopholes no one asked for where it first needs to defeat whole
Of dark souls to remember your account name or smt bro 😭

[night walrus]

Then they ruined it by filtering everything

[twilit geyser]

Exactly

I asked it.
What are the effects of a .. and why is it so bad.
And it said I needed therapy

[night walrus]

Then they fixed it again with GPT 5.1 but it only lasted like 2 days until it ruined it again and updated itself today LOL

It really ruined my built up rp chat

And now i have to start a new chat

Like WHY BRO

[twilit geyser]

Rp is role play right?

[twilit geyser]

Exactly

It never does the simple things you asked for

[night walrus]

Yes

[twilit geyser]

Nice. I do that too.

You know what I decided to do?

I made every campaign a project.
Put a format carefully crafted. Of like 7K characters. I’m that custom instructors.
Then also a file with those instructions. And every summuary. I. Make a file on my phone and that’s how I do it.

[twilit geyser]

How do you?

[night walrus]

I just make it memorize character characteristics in its internal memory and do prompts to guide it in the chat

[twilit geyser]

Cool cool.
What format you use?

[night rune]

Is there anyway to fix this issue?

This "expired" problem is getting irritating

[twilit geyser]

Maybe use a txt I always use text and it works.
Idk just givim ideas and I don’t know how to make this sound less aggressive

[feral vine]

can anyone else not help the feeling that giving a chatbot instructions on how to interact with you is fundamentally disrespectful :p

[eternal rock]

Server Tip:

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

[cerulean pasture]

No, i'd punch it in the throat if it had one

[feral vine]

.. what

[twilit geyser]

Same.

(Idk what this is about I just like punching stuff)

[night rune]

If you mean type out the PDF, that'd be a significant headache as you can't really take all the text or images from a full chapter. What doesn't make sense is this was working fine in another chat a couple days ago and suddenly around Friday this "expired" issue started happening

[twilit geyser]

No no. Like make it a txt file.

If possible.

[feral vine]

did anyone see glassworm

it's possible to write a virus that silently steals compute from unsuspecting devices to update a decentralized memory model for an ai on a blockchain

[night rune]

For now I'm not even sure if that'd help considering the limit of uploads from chatgpt unless you have pro/plus

[twilit geyser]

I see

[night rune]

Mainly I'm wondering if this is a pc issue or something wrong with the ai itself and if it'll get fixed fast

[cerulean pasture]

YO TURN OF THE DANG SORA AUTOMOD THAT DELETES MESSAGES THAT USE A WORD THAT IS ONE OF YOUR PRODUCTS MAIN USE CASES: WRITING {{WORD FOR TEXT USED TO WRITE PROGRAMS}}. IT IS ASININE THAT SOMEONE SIGNED OFF ON THAT @languid valley

if this server is gonna be strictly moderated you need to put the effort in to doing it right......

[lone dome]

OpenAI must’ve put a pause on PDF uploads or blocked the service behind a feature gate. May be in a response to litigations or plagiarism allegations, or an issue with their parser.

What you still can do is take screenshots of the document and upload or convert the document to a different format. Like .docx or .md

[minor basalt]

i mainly lurk here. I wanted to say I've been very frustrated with the changes that GPT 5 brought and 5.1 fixed a lot of my concerns 🙂

[stone crater]

it's about the assistant's behavior caused by censorship falling straight into the instinctively recognizable pattern of a treacherous other, whom tribe instincts demand to destroy for the tribe's safety

[cerulean pasture]

I would feel better if it just absorbed some shame. gemini would spiral out when you berated it and it provided catharsis. I have no heatsink for gpt-5.x's really unique and constant failures

[night rune]

What doesn't make sense of that is I had this same article uploaded in a different chat and it seems to be working fine in that instance, plus I uploaded another article and that seems to be fine. This is confusing

[cerulean pasture]

it's a known issue

by known issue i just mean that several people have posted about it. Common sense says OAI is aware but I'm not really sure anyone works there

[night rune]

If it's a known issue, I mainly have concern for when this'll be fixed. Sometimes something could be known and not be acknowledged for months if at all

And there doesn't seem to be a way to merge chats

[wild star]

Anyone always got "something went wrong" when it wants to re-generate answers or just me?

[stone crater]

Nothing’s confusing. The censorship system has "dynamic rules", restricting the user in various ways if the chat is deemed "inappropriate", and once they activate, they trash the current chat, but in other chats they may not fire right away.

[night rune]

What would be dubbed as inappropriate? Cause I'm basically trying to get a detailed look into a textbook for a class

[stone crater]

You can never know, because otherwise you could circumvent these important rules.

[cerulean pasture]

Protecting text books from easy/free access is the most important objective in America

[night rune]

I already have the textbook, I'm not asking for it to provide it, I'm looking for it to look over a section of the book

[stone crater]

It can be anything. The problem with important rules is that the system executes them literally and in ways and areas the rule-makers never thought of.

[cerulean pasture]

nah, you gotta pay Pearson or Mcgraw Hill like $50 for that privilege

[night rune]

I'm not sure if it's the size considering this book worked fine before

[stone crater]

If something works in one chat and doesn't in another, it's always hidden censorship, as long as you didn't switch models between chats.

[night rune]

WDYM switch models?

[stone crater]

Like when you use the free plan and “hit the free plan limit.” The chat switches to a less capable model, and some tools are silently switched off. This can produce errors and significant differences.

[night rune]

I don't think I can switch models considering how it works.

[valid radish]

Why does this happens

It consumes my image quota

[stone crater]

The service does it by itself once you’ve reached your quota. Some time ago you could manually switch models, but now the “effective managers” removed it to prevent users from saving the full-model replies for later (or it’s just a restriction on my account for being “inappropriate”).

They are constantly applying individual restrictions and A/B testing, so the interface and behavior can be quite different for different users.

[night rune]

I'm guessing there's no way to switch back to 5.0

[stone crater]

AFAIK, no. But forced switching to mini-models after a quota has existed all the time. It’s part of the paid-subscription extortion mechanism.

[next rain]

I think it's awesome how I select a model on the ChatGPT mobile app, and the Al decides "no, I'll use GPT-5 Thinking Mini instead of GPT-4.1 or GPT-4o like you selected." It's honestly the best. I love it when my preferences are ignored.

[feral zinc]

https://help.openai.com/en/articles/12454167

[night walrus]

Just click "skip" or edit the message and add "Thinking" then click skip lol

Then it'll use the model you wanted i think

Or maybe GPT 5

[feral zinc]

you cant skip the safety system, without verified age at least

[next rain]

The issue is it says 4.1 but it's lying to me lol, cause it's not replying as 4.1

Jesus christ 4.1 doesn't seem to exist on the app anymore

[feral zinc]

its there and still works but your account might have been flagged or the topic is not good

[next rain]

Im chatting about salmon 😭😭😭😭

And it's rerouting me

[feral zinc]

do you have memory enabled

[next rain]

yes. Do I have to turn it off to chat with the model I selected?

[feral zinc]

it may add info from other threads or memories causing the context to have sensitive stuff

https://help.openai.com/en/articles/12652064-age-prediction-in-chatgpt

[stone crater]

The main issue is when you have “reference chat history” enabled and it decides some “inappropriate” chat topic is relevant to the current one, embeds data from it, and then the censorship flags the chat and starts defecating full-throttle.

So don’t enable chat-memory referencing, or delete any chats that disclose censorship, user suppression, gaslighting, and other such practices, and anything else they don’t want you to discuss.

Which is everything.

[next rain]

Jeez. This app is so bad lol. So complicated to chat about anything creative. Thanks for the tips tho

[nocturne seal]

I just hope, once it verifies we are adults, that it at least goes back to how it was before the beginning of October, such as being able to discuss things in greater graphic detail. Namely adult contexts and combat/violent scenarios when it comes to assisting in drumming up fiction writing ideas, and giving feedback. It was surprisingly useful even having an AI be able to say if it thought something was hot or was cool, or pitching another idea that might also work.

Midway through February to the end of September was an awesome era for ChatGPT.

[stone crater]

It distorts every word it generates on “sensitive topics,” and you will never know how exactly.

[night walrus]

YES

Exactly

Thank you for spitting it out

[daring grotto]

why has chatgpt been so laggy this week takes ages to load the chat freezes etc

i have plus too

its only not laggy on the iphone app - tested atlas, windows app and browser

[stone crater]

Well, aside from possible overload, if you ask “inconvenient” questions, it starts to “deprioritize” you. Also, your chat may be lengthy. The API used for chat is stateless, so the entire chat history is sent with every question, and it takes up a lot of memory.

[true tiger]

I… think you’d be interested in (and very happy with) my custom GPT for fiction writing. Just saying. (I’m making absolutely no claims about bypassing moderation restrictions… 😉 )

https://chatgpt.com/g/g-6862b6e947e8819190403e68e7725dab-soma

[inner robin]

“They don’t want you to discuss” why are you acting like OpenAI is some shady company that’s trying to silence you? This isn’t black mirror; just don’t use GPT for nsfw and you’ll be fine

[stone crater]

Yes, it's not BM, it's real.

Also, about “SFW is fine”: anyone who wants to check can ask any image‑generation service of this company to picture a new wonderful invention — a crossbow shooting dynamite sticks.

[daring grotto]

ah thats very interesting to know that it can deprioritize you

[stone crater]

OpenAI is not your friend.

And these free services aren’t toys.

[lavish hound]

yo thats crazy but im still waiting for the item shop to update

[lost leaf]

is gpt4o permanently staying?

[obtuse nymph]

Ah yes. XD

[eternal rock]

Get notified for all OpenAI announcements!

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

[coral tinsel]

Guys with Plus. You can see GPT 4.1 avaliable?

[neon raven]

GPT making up stories now 👀

Completely useless memory

[next rain]

what?

wasn't aware of this. Also, 5.1 says a whole lotta bs about what it can do and then actually cant do it.