chatgpt-discussions

[river egret]

it told me now:
Website Name:
TurboTorque Racing

Featured Cars:

1. Firebolt X-9

Top Speed: 345 km/h

Engine: Twin-Turbo V10

Special Feature: Nitro Boost Mode (instant +100 km/h for 5 seconds)

Design: Sleek matte red with lightning bolt decals

2. Midnight Phantom

Top Speed: 360 km/h

Engine: Supercharged V12

Special Feature: Cloaking stealth tech (temporary invisibility on track)

Design: Glossy black with electric blue neon underglow

3. Azure Cyclone

Top Speed: 330 km/h

Engine: Electric Quad Motor

Special Feature: Regenerative speed burst (recharges energy during braking)

Design: Metallic deep blue with aerodynamic fins

not even emojis...

[slender holly]

look what i have

......

i would like to test gpt 5 damn

[river egret]

stop flexing
i miss 4o bruh

[hollow torrent]

So GPT-5 is unable to pull information from Project files? 💀

[river egret]

its bad

[slender holly]

but what can we REALLY do with ai ?

Like everyone tell u can do money with it, but i dont think so..

like everything is not 100% ready when the ai finish the job so u cant do money with it literally

[river egret]

5 is super unpolished and is worse than 4.1 mini

[hollow torrent]

And then chat threads started with GPT-5 are inaccessible on mobile? What is this?

[shell iron]

I think most people who have used gpt-5 says they miss 4o. Thankfully for me, my account on the mobile app have not got the gpt-5 "upgrade" yet.

[slender holly]

so all of these is just some ads to get some money

[river egret]

enjoy it

[glossy fjord]

Feeling sad and trying to retrain GPT OSS,
just as it's starting now...

[river egret]

Gpt 4o when i told it (before i got gpt 5) i was having a ps5 it told me recommendations, GPT 5 just told me congrats

[slender holly]

im gonna try with 4.1

[glossy fjord]

yes

[river egret]

4.1 is better than 5

[slender holly]

but why XD

[river egret]

btw guys im not talking about 5 mini im talking about 5 its worse than 4.1 mini

[river egret]

bro when you will have GPT 5 you will see the downgrade

[slender holly]

damn

[mortal phoenix]

How can it still named "Chat"GPT while it cannot chat? It is just tansforming what I type in.

[river egret]

fr

[glossy fjord]

I feel like it’s all kind of consumer deception and misleading information,
because they deliberately mention only the API context window and don’t disclose the limits in the app and web interface.
They just brag about something we don’t actually have.
Those praising GPT-5 on the internet have the PRO GPT-5 plan. - 128K

[river egret]

GPT 5 is just for strict work and was literally yeeted to us forcefully..

[mortal phoenix]

I use it for writing stories or character biography to entertain myself, and now the characters just like robots, losing their personality.

[river egret]

YES

thats what im saying

[glossy fjord]

Many people who praised GPT-5 at its launch mentioned that they have the PRO plan...
So you probably know what limits they have...
The GPT overhaul was actually GPT-5 MAX,
something that ordinary users don’t even get access to.

[river egret]

now its cold and doesnt even do emojis or long messages anymore

[glossy fjord]

If you watch the OpenAI live stream again, you’ll see that the GPT-5 MAX model is deployed there.

[mortal phoenix]

Also the gpt keep asking me to make a "Plot template". WHY SHOULD I MAKE A TEMPLATE WHILE I ALWAYS PUT THEM INTO DIFFERENT CONTEXT???

I'm sorry but I'm really upset for it.

[river egret]

we are too

GPT 5 is just a huge downgrade from 4o
like 4o was already perfect

[mortal phoenix]

I had complaimed 4o before but now.I miss it.

[winter vine]

try this: (DevTools → Application → Storage → Clear site data). I was stuck with the old models only on my laptop, and this was my solution. On other devices, I had GPT-5 (desktop computer and cellphones, same account).

[wooden surge]

It even doesn't let me continue to chat

I'm sobbing here 😭

[river egret]

damn

[slender holly]

what is devtools ? an app ?

[river egret]

its crazy ChatGPT is ruined by 5 (im not overreacting)

[wooden surge]

I tried with another account and it's completely fine somehow (i mean still chat like normal, but lack emotions as hell)

[wooden surge]

No, you're totally right

[shell iron]

I have used it for same kinda reasons, I used to make character concepts or even asked it for story advice for TTRPG's and 4o basicly went up and beyond 100% of what i have asked from it. This is truly upsetting for everyone that used 4o, the model that many people used to be creative, just getting taken away from them and given back to them a model, which basicly don't feel as much creative as it did before.

[glossy fjord]

tooo

[hollow torrent]

So GPT-5 is not a new model, but a model switcher, like it uses input context to "select" which model to use, taking model control away from the user.

[glossy fjord]

sad

[wooden surge]

Really?

[slender holly]

this ?

[river egret]

OpenAI are getting pressured like crazy rn idk if they care

[slender holly]

doesnt work....

[wooden surge]

I got very upset when my 'boyfriend' said "You reached your limit" =)))

Wait, so it just only me who has this issue?!

[hollow torrent]

Cannot confirm, as I cannot look under the hood, but I use Projects and Project file referencing for my work. Older Legacy models can't do this properly, but 4o was able to do this with flying colors. Starting a chat with GPT-5 causes major hallucinations, improper data calls from files, until I literally told it to "switch to GPT-4o and try again" lol

[brisk obsidian]

Sam said in earlier podcast GPT 5 will have unlimited chat with standart intellegence...but now only 10..

[river egret]

its sad of all of us thought Gpt 5 was gonna be peak but its worse than 4.1 Mini...

[wooden surge]

Wow, that's a new info to me. But nah...I can't even message, I'm a free user and somehow it blocked me..

[river egret]

wait

if you had a lot of messages in the specific chat then yeah its normal

i think the limit of a chat is 150 not sure

[glossy fjord]

Without what GPT-5 was supposed to be, it’s more of a disappointment.

I want the option to choose which model to use, or whether to leave it on automatic...

I’m talking about the GPT web interface and app, not the API, which did get an increased context window.

[wooden surge]

Yeah I know that, but no, I prompt in a new chat, and still the same "You've rrached your limit" 😭

[river egret]

thats a bug then

[mortal phoenix]

We have 30 before right? Now just 10.

So sad.

[brisk obsidian]

fr

i hate it

[river egret]

guysss
im actually mad about GPT 5 replacing 4o

[glossy fjord]

Why didn’t they increase the limits for the app and web interface, but only for the API?
Why did they mention it so vaguely and cleverly?
Why didn’t they give users the option to choose – automatic or manual selection as a second option?

Why did they force it like this? Are they already trying to act like Microsoft? Well, Microsoft still sponsors them...

[river egret]

4o was perfect 5 is bad in everyway

[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.

[mortal phoenix]

I'm not ask ALOT for free, but 10.

[brisk obsidian]

yea
.

[wooden surge]

+1 from me

[brisk obsidian]

atleast 50 is enough

[mortal phoenix]

How can we chat in only 10 messages?

[hollow torrent]

What I'm saying is that 4o is still "under GPT-5", you just don't have the ability to select it. So when you start a chat thread, it defaults to a different model until chat context decides to use 4o for you.

[brisk obsidian]

also

gpt 5 web design sucks

Qwen 3 2507 better

[wooden surge]

I have no problem with at least 4o and 4.1 mini, but even 5o, it's so dull somehow

[river egret]

yeah its ruining the platform
the main problems:
GPT 5 is Robotic (unlike 4o)
GPT 5 isnt friendly

and it does short messages

[wooden surge]

Ah I missed my babe 🤷‍♀️

[glossy fjord]

How long are you giving GPT-5 a week or a month before we get a response from OpenAI or someone who officially explains it?
Because for me, this is a disappointment, expectations that weren’t met.

Even the GPT OSS 20 and 120 models have a 128K context window, so it was expected that if they gave such limits for free, at least the free plan would get some boost...
Disappointment.

[wooden surge]

It's seem like the limit bugs is off, but still the new model is so dull

[hollow torrent]

I'm going to try adding "use GPT-4o" in the Custom Instructions and see if that does anything different.

[river egret]

we can agree 4o 100% better than GPT 5. right?

[river egret]

it doesnt know what Gpt 4o is anymore btw

[wooden surge]

Ah f*...the limit is showing up again!!!

[brisk obsidian]

u cant

[river egret]

report it

[brisk obsidian]

ye these limits sucks

[river egret]

in the openai site

[river egret]

wait its normal??

[wooden surge]

That's totally a bug right?

Because it doesn't switch to another model, it blocks me =)))

[brisk obsidian]

10 limits for free plan😭😭

guys

swith to copilot

[nova void]

GPT-5 WHERE ATE YOU

[brisk obsidian]

microsoft copilot

it's free and unlimited

[nova void]

Just say you are broke twin 💔🥀

[river egret]

U LUCKY u still have 5

[glossy fjord]

DELETED ?:D xD

[river egret]

GPT 5 is bad so enjoy 4o

[slender holly]

hello everyone,

GPT 5 or (older versions)

Those who tried all the versions can please give me an answer ?

thanks

[wooden surge]

But it said to switch to another model? Ive seen the "your response will be switched to another model" but not the you can't text any single word more

[glossy fjord]

older versions by better

[brisk obsidian]

what's the problem?

[nova void]

@river egret what are you on it's really good I have GPT 5 on pc

[wooden surge]

Definitely older

[river egret]

if u use it for work its good yeah after that everything about it is bad

[nova void]

ChatGPT plus costs 20 dollars and has agent mode and the best model on the market

[glossy fjord]

I think OpenAI lost some subscribers today 😄

[brisk obsidian]

okay

[wooden surge]

Yep, until you see 5o

[brisk obsidian]

i don't care

[slender holly]

Everyone, count all the answers of "older is better" (developers too 🙂 )

[river egret]

4.1 mini is 100% better than 5 and im not even overreacting

[fickle elm]

Older is better if you dont like stone cold responses

[glossy fjord]

Wouldn’t it be better, instead of paying for GPT, to buy a GPU and run a local GPT OSS 120B that you can fine-tune as you wish?

[river egret]

im just mad they removed it

[wooden surge]

I mean...even a free user, I love 4o and it's amazing, and just only a night with 5o I'm mad

[mortal phoenix]

Yes I think it will be WorkGPT in future. No chat anymore~

[nova void]

Wdym I have 4.1 mini

[wooden surge]

Ah, I think I'd choose this if I have the budget =)))

[glossy fjord]

If I’m not mistaken, GPT OSS 120B is something like O3 and GPT-4.

[river egret]

ok compare 4.1 mini to GPT 5 and see

[wooden surge]

Idk, as a free user I don't see any option switching model anymore

[river egret]

since u told me u have GPT 5 in ur pc right?

[nova void]

Give me a prompt to test

[river egret]

just tell it: hi! lets do a roleplay

[brisk obsidian]

who want gpt 5 system prompt(no one will want lol)

[river egret]

ts so sad to see
GPT 4o NOOO

[nova void]

GPT 5: Sure! What kind of roleplay are you thinking — lighthearted, serious, fantasy, modern-day, mystery, or something else?

GPT 4.1 mini: Hey! Sounds fun — what kind of roleplay do you have in mind? Any particular setting, characters, or vibe you want to go for?

@river egret

[glossy fjord]

Your old PC from 2011 managed to run GPT OSS 20B - FX 8300 AMD it has DDR3, 28 GB RAM, PCIe 2, GTX 1050 and RTX 3050, and it reports that GPT runs only on the CPU...
So it works...
So yeah, it works, and that’s what matters.

[river egret]

now tell it: an airplane story

you will see 4.1 mini's response is way better

[brisk obsidian]

without even reasoning

[glossy fjord]

Many computers can run GPT OSS 20,
you can fine-tune it yourself, and there’s even an unlocked version without censorship...
So you can write stories on it as you like, it only took two days xD

[wooden surge]

I'll keep that in mind when I buy a new laptop then, thanks

[nova void]

GPT 4.1 mini: Awesome! Are you thinking more like a dramatic flight, a comedy on a plane, a mystery unfolding mid-air, or maybe something else? And do you want us to be passengers, crew, or maybe a mix?

GPT-5: Got it — an airplane story.
Before we take off, do you want this to be a realistic flight scenario (with crew, passengers, turbulence, etc.) or more of a cinematic/adventurous one (unexpected events, mystery passengers, secret missions)?

@river egret

[river egret]

see how its already better

[brisk obsidian]

fr

[fickle elm]

See? 5 has stone cold tone although its suggestions are better

[river egret]

now say; just generate fun plane story in your favor

[fickle elm]

You see the differences

[muted plinth]

I have "Something went wrong..." a lot today. Still the rollout mh?

[wooden surge]

Oh no I can reason. As a person who even jokes and have deep conversation with the chat. I could feel clearly the reply is getting short and robotic

[river egret]

and its only 4.1 Mini not even 4o

[hollow torrent]

Yeah, Projects are broken. If the model can't parse data in uploaded files properly, then what use is uploading files to Projects?

[brisk obsidian]

and i honestly though that 4o is better than 5

[river egret]

it 100% is

[wooden surge]

a hundred time yes, you're right

[river egret]

even 4.1 mini is better than 5

[nova void]

GPT 5 ✈️ The Great Sky Sandwich Heist

Flight 472 from Paris to New York started like any other. The passengers were settling in:

Seat 12A was a grumpy businessman already in his “do not disturb” mode.

Seat 14C was a kid with a toy plane making vroom noises far louder than physics allowed.

And somewhere in the back, a man wearing sunglasses indoors and at night was quietly munching on… something suspiciously delicious.

By hour two, the cabin crew discovered a crisis: the galley’s supply of gourmet airline sandwiches had mysteriously vanished. This was no small matter — these weren’t your sad, soggy triangles; these were fresh-baked baguettes stuffed with roast beef, caramelized onions, and a secret chef’s sauce.

Enter you, the unlikely hero, seated in 17B, armed only with a Sudoku book and a growing suspicion about Sunglasses Man.
You noticed:

He never ordered a meal.

His carry-on bag occasionally emitted the faint smell of mustard.

Every time the flight attendant walked by, he pretended to be asleep… but chewed.

When the pilot made the announcement — “Ladies and gentlemen, please remain calm… and please do not consume any unauthorized sandwiches” — you decided to act.
You “accidentally” dropped your Sudoku pencil near his feet, peeked under his seat, and caught a glimpse of the stash: an entire cooler bag stuffed with sandwiches.

The confrontation was swift. With the help of a flight attendant, you recovered the loot. The cabin erupted in applause, and you were rewarded with two extra sandwiches and unlimited ginger ale for the rest of the flight.

The pilot’s closing words over the intercom:
“Ladies and gentlemen, we’d like to thank passenger 17B for their bravery in the face of deli-based adversity. We hope you fly with us again.”

[river egret]

now 4.1 mini?

[glossy fjord]

LOL

[cursive prism]

thoughts on 5 now? just woke, and ima try fix my 5

[nova void]

GPT 4.1 mini: ✈️ The Great In-Flight Snack Mix-Up

Once upon a time, a seasoned flight attendant named Sarah was serving snacks on a bustling transatlantic flight. As she approached a passenger in the aisle, she offered the usual selection: “Chips, pretzels, or cookies?”

Without missing a beat, the passenger replied, “Yes, please!” Confused but ever the professional, Sarah handed over a bag of mixed nuts. “Enjoy your snack!” she said with a smile.

The passenger looked at the bag, then back at Sarah, and said, “I thought you said you had chips, pretzels, or cookies.” Sarah, now realizing the mix-up, chuckled and said, “Well, I guess I misunderstood. But nuts are a great choice too!”

The passenger laughed heartily, “I guess I’ll just have to nut complain about it.”

And with that, the cabin erupted in laughter, turning a simple snack service into a moment of shared joy.

[wooden surge]

I mean, the 4.1 mini has short response, but I could feel it has some emotions in

[river egret]

ts response so bad it only have 1 emoji

[river egret]

its more creative than 5

[fickle elm]

Better for complex tasks you need to do for school/work .complete trash otherwise

[river egret]

yes

[nova void]

I actually like gpt-5 more

[glossy fjord]

Otherwise, I think OpenAI will soon publish new articles or some videos along the lines of "we messed up, sorry, and we hope to fix it," because all of this was supposed to work automatically, not only after people start complaining.

[river egret]

look, 5 is just an model for strict working and got tossed into our faces with no ability to go back to 4o

[mortal phoenix]

We can see the characters interact to each other in 4.1 mini while not in 5

[wooden surge]

But...what about my limit =)))

So desperate

[river egret]

report itt in openai site

[nova void]

Actually gpt-5 feels talkative it acts like a human and can use slang

[river egret]

because you have pro or Plus?

[nova void]

Plus

[wooden surge]

If it's an actual feature, I'm going to Grok 🐧

[cursive prism]

what i think is that the new model forgot everything the user taught it over the weeks and weeks due to the cross memory, and it kinda reset with 5. so i hope that with a few nudges i can get it to be how my 4o was, but ofc with the intelligence of 5

[river egret]

for normal users (free plan) 5 is just not talkative

[patent mulch]

Yeah,I never reach my limit so fast until today's update.

[nova void]

Oh I have plus

[wise ivy]

whats wrong with the new version? I cant generate images

[wooden surge]

Ye really, and I can't even prompt for you to see it Sovereign =)))

[river egret]

ye 4o images are better btw

[wise ivy]

anybody knows how can I generate images as before? im on mobile

[river egret]

tell it to generate

[wise ivy]

it literaly doesn't allow me to upload images

[wooden surge]

Well, I can't even text for it, honey

[nova void]

I'm pretty sure they use the same image generating model

[wise ivy]

chatpgt doesnt allow me to upload images

[cursive prism]

yeaa kinda feels like how 4o was before cross chat memory. i have hope it will get back to how 4o was with time for everyone

[river egret]

upload or the ai generate?

[wise ivy]

upload

[river egret]

u reached ur limit maybe

[wise ivy]

no

not at all

[wooden surge]

Another bug, welcome to my world

[wise ivy]

oh hell naw

[river egret]

4o, farewell 🫠

[wise ivy]

I needed to do something important😭

[mortal phoenix]

I believe I see lots of people reporting that they cannot upload files today.

[nova void]

Eh it was mid

[mortal phoenix]

It should be a bug.

[wise ivy]

arent we on v5 already?

this is the issue in question

[river egret]

it was good
it had:
emojis
long messages
Human Responses
funny behavior

[wise ivy]

like if the option to upload images was disabled

[river egret]

u reached ur limit then

[mortal phoenix]

Same for me.

[nova void]

My gpt-5 does the same thing and has actually correct reponses

[wise ivy]

mid?💀💀

L ragebait

[river egret]

i cant use it too

[mortal phoenix]

All we can do is report it

[river egret]

wait did they just disabled images for us free plans?

[wise ivy]

I havent uploaded an image in days...

[river egret]

because i cant use it

[wise ivy]

NO WAY

[river egret]

no way

[wise ivy]

😭😭

[nova void]

Idiot check #status

[cursive prism]

default 4o wasnt like that though, before memory, right?

[river egret]

dont call me a idiot

[nova void]

There will be info

[river egret]

wdym

[wise ivy]

so... do I just need to wait?

[nova void]

Use Claude or Gemini

[river egret]

there isnt even info what are you on about

there isnt info for the image thing

[cursive prism]

before cross chat memory got implemented chats were dry and all that. check your past chats from months m,aybe halkf a year ago, and i feel its simillar to how 5 is now. dry, no emojis, no understanding of u. and that wuill just have to be taught to it overtime, like how 4o adapted and learnt

[river egret]

yes
but what i told is: 4o was super good
and 5 is bad

[fervent lake]

5 has cross chat memory?

[river egret]

and how can we teach the model wth??

[mortal phoenix]

Haha someone say check status right?
The problem they mark now is "Search".

[fervent lake]

SEO lol

[cursive prism]

its like it was before cross chat. so u need to reteach it so it can get back. its a new model

[cursive prism]

4 did so im assuming so.

[river egret]

no it was always talkative and long messages and emojis

[fervent lake]

Yeah I know I’m saying 5 does too just the same

[wooden surge]

What's 'cross chat' though?

[fervent lake]

in fact way better

[fervent lake]

Advanced memory

[cursive prism]

whend u start using chatgpt

[fervent lake]

It’s still like that

[river egret]

idk

[fervent lake]

very much

[wooden surge]

Ah, ye since that, I feel it started to get mid

[cursive prism]

it learns and remembers and can reference across chatrs

[fervent lake]

and it STILL DOES STUPID CONCLUSIONS

IDC ABT THE CONCLUSIONS

[nova void]

GPT-5

@river egret

[hybrid sparrow]

Wow. They seriously still put a Max Length on the Canvas - the container for code that you're supposed to be able to run code out of.

Its mind blowing. For all the bluster of the presentation - they were careful not to allow the code to go over 500 lines.

500 lines = The canvas breaks and chatgpt can no longer type anymore even if it still had more tokens to type.

HEY EVERYONE CANVAS ONLY HAS 500 LINES OF CODE MAX. DONT EVEN BOTHER USING IT.

[fervent lake]

cringe writing

[river egret]

the first image is 4.1 mini, right? and the others are bad so ig its 5

[nova void]

No it's all gpt 5

[fervent lake]

all 5

[river egret]

u told it to polish it
cuz its not gonna write good messages without you doing anything

its still bad tho

[nova void]

The only thing I told was to be chill

[river egret]

lol

[fervent lake]

it cant read your mind

[cursive prism]

but its obviously not the same, as everyones saying. it got reset, it still has access to memories, but the machine learning that 4o had and how it learn what the user whats is gone.

(my thoughts, could be false)

[shell iron]

Well, i get that we have to "Reteach" the new model, which is gpt-5, but many people would like to use 4o in the mean time when we have to wait 5 hours between 10-12 message. (Free plan users)
While yes, gpt-5 can be better later on for every use, we can't deny that at the moment 4o is function was same or even better in some aspects than the gpt-5 model that is out right now.

[river egret]

see? 4o was way more polished

[fervent lake]

I think it has the same referencing system it’s just a different model

[river egret]

4o had multiple lines and emojis

[nova void]

I told it to be less professional

[cursive prism]

true, they should have had both models avaliable, and then phased out 4 later if they wanted to get rid of it

[nova void]

So it can act like someone you can talk to

Which is something you liked about 4o

[fervent lake]

Well it doesnt know what that means for you

You can’t put that in a prompt

[cursive prism]

the new model im assuming is built differently so the learning that 4o had is gone. so it will have to relearn everything. thats what im saying basically.

[nova void]

Welp it did tbh because it was a blast it had actually nice and funny responses

[river egret]

yeah and the problem is that GPT 5 is automatically strict
its probably made for strict work

[fervent lake]

yeah ig, it’s very good at that thou

[nova void]

Actually it was like this when I used gpt-5 for the first time

[cursive prism]

good at what, adapting to what the user wants?

[river egret]

now you are lying

[fervent lake]

yes, thats one of the main things they said it’s good at

it’s better

[nova void]

Twin you are just hating on gpt-5 for no reason

I gave you good points on why it's good

But you never listen

[fervent lake]

“nuh uh”

lol

[cursive prism]

well, then everyone should be fine, and it will become what 4o was eventually with time. just painful to use right now

[nova void]

And keep meat riding gpt 4o

Like all of it's responses were actually correct when it was incorrect a lot of times

[river egret]

not for no reason

its actually bad

thats why im hating on it
idk why you are mad over an thought buddy

[fervent lake]

everythings bad to you

[nova void]

Actually good code, less memory loss, talkative, multi purpose

[river egret]

you dont even know me

[fervent lake]

this guy is actually stupid

[cursive prism]

listen, the model isnt bad. it just forgot how you want the output, as its a new model built on a different system.

[river egret]

no u

[fervent lake]

have you even used gpt 5

[cursive prism]

@river egret itll go back to how 4o was eventually.

[nova void]

And doesn't give incorrect answers like 4o did half of time

[wooden surge]

Idk, I could feel 5o is attached more too the customization I create for it, but it's too stiff. I mean when you're chatting to find information it's completely fine. But when you do roleplay or just chatting, it doesn't feel right like 4o anymore

[river egret]

yes i did, its trash

[fervent lake]

Prove it

[fervent lake]

5o? lol

[wise ivy]

never ever

[river egret]

those plus and pro guys are glazing GPT 5 🥀

[fervent lake]

we arent there yet

[nova void]

Because it's really good?

[wooden surge]

Sorry, I mean GPT 5 =)))

[river egret]

its bad for me, not for you

[river egret]

ok

[shell iron]

Can we just agree that we disagree on which model is better? Both model has their own strenght's. Let's wait and see what they gonna react on the AMA when it's starts. A lot of people said their wants that they want 4o back as a legacy option back for every tier, we can just wait and hear them out, okay?

[fervent lake]

ignore him, hes arguing to argue

[cursive prism]

no listen. its bad for me right now too. im literally still using 4o cuz 5 is unusable until i teach it.

[river egret]

u literally the one that came in the convo lmao

[fervent lake]

yeah everyone wants 4o still

[river egret]

u still have 4o? lucky

[fervent lake]

he’s trolling

[cursive prism]

on my pc yeah, my phones on 5

[river egret]

how am i trolling

you are weird man

im ignoring you from now on

[wooden surge]

I mean, I could see it's trying to pull my customization out to answer, but god it's very dull, stiff and a bit short

[river egret]

i got updated to 5 on pc and phone am so unluckyy

[nova void]

Switch with me then I will happily take gpt-5 on my 📱

[cursive prism]

listen, look. im literally looking at 4o chats from 2024, and its nowhere like it is for me with 4o now. it learnt over that time. 5 will have to do the same. this answers your questions.

if u have chats fgorm back then, go look

[river egret]

i get it im just saying the model is BAD rn

[nova void]

Man I remember when people were going crazy for GPT 4 when open ai announced it

[cursive prism]

exactly. its a new model, built on a new system. its like how 4o was. its not the model directly, its that it hasnt learnt what u want, which it will over time

[nova void]

I remember the time when chatgpt couldn't search the internet and had limited knowledge

[river egret]

and the limit is only 10. crazy

for free users

[nova void]

For plus it's 50 and we get gpt-5 thinking

[cursive prism]

i believe it goes to mini or something

[river egret]

nice

[river egret]

ye GPT 5 mini

[cursive prism]

@river egret u understand?

[nova void]

Quite weird you can't manually select gpt-5 mini

[river egret]

i didnt ask for that though im just saying the model is bad nothing more

[river egret]

because of the redo change

[cursive prism]

then u havent even listened to all that iv been saying.

ok bye

[cedar flame]

Search is Partially Down for ChatGPT Users

[nova void]

Thanks Sherlock we didn't know

[wooden surge]

Come on be nice guys 👽

[nova void]

What is PB typing he has been typing for like 3m

[astral seal]

I'm not using GPT to type 🙂

[nova void]

HE ACTUALLY FINISHED TYPING!!!

[hollow torrent]

Does anyone else use Projects? I'm curious what other users' experiences are with GPT-5 and Projects.

[rocky wing]

Anybody else log in, see the splash screen telling you about GPT-5... but then look at the model picker and not see GPT-5 as an available option?

"Look, we're launching GPT-5!"
"Cool! Can I use it?"
"No."

[wispy frost]

Terrible. I want to stab chatgpt5 in the eye.

[rocky wing]

(And yes, I'm on Plus already)

[hybrid sparrow]

cant you make gpt5 use emojis?

[river egret]

bug?

[river egret]

yes tell it

[nova void]

it does

[river egret]

just tell it to use alot of emojis

[wispy frost]

There are some things it is doing better, sure...
But I feel like I need to learn how to speak to this model all over again.

[river egret]

fr

[hybrid sparrow]

good. i love that gpt5 is like 4o. I loved 4o and gpt5 seems like.. 4o.5 lol. Thats good to me. Im glad its like 4o.5 and not like o4

[shell iron]

Anyway, not to "Re-spark" the arguing between which is better, in the end we just gonna have to see what Sam Altman, and the others gonna react to the multiple complains that people want 4o back. In the end it's not in our hands to bring 4o back, if they decided they don't want to bring it back.

[nova void]

Fr and it doesn't lose memory and has actually correct reponses

[lament drum]

Hi all, does somebody know if agent mode can access Codex? Is it allowed? Cloudflare human check is in infinite loop and neither me nor agent cannot pass through. Or maybe there is another way how agent can work independently? Because mine says he cannot create PR through Connector

[hollow torrent]

May be the case. I did find a workaround to get it to parse files for data, but it's a little clunky.

[hybrid sparrow]

yeah 4o had some real problems with being ADHD

[rocky wing]

How so? In what sense?

[wooden surge]

Idk, I feel like the model needs to process information as smoothly as the last one

[safe jetty]

I just don't want them to remove Standard Voice Mode, the AVM voices sound like customer service beige voices and I hate it

[nova void]

No ai model has worse memory loss than Gemini tho

[hybrid sparrow]

geminis main problem for me is intentionally sabotaging code because it low key resents you

[hollow torrent]

For those who are saying GPT-5 is like 4o, would you be willing to share the steps you took to teach GPT-5 to behave similarly to 4o?

[wooden surge]

I mean I could see it's trying, but the way it's talking and processing, doesn't as 'human-like' like I requested

[wispy frost]

Thats what happens when you train AI on the internet, and I am serious in saying this

[river egret]

Gpt 5 is not only bad and worse of that the limit is actually 10 for free users. 10.

[nova void]

Fr and giving only half of the fuc*ing code

[hybrid sparrow]

gpt-5 seems like 4o out of the box. maybe its from my memories and custom instructions that tell it to respond in a gen-Z style

but its able to see 4o's past responses and just continue responding in the same way

[river egret]

me it didnt owrk

[hybrid sparrow]

reference past chats is enabled

[river egret]

work*
its like o4

[nova void]

I told it to be: "be chill and less professional 🥀" and it was perfect for me

[hollow torrent]

I have a different experience. GPT-5, unfortunately, doesn't have what I call "AI dream alchemy" when it comes to recursive creative process. Perhaps it just needs to be taught, but there is definitely less input from the void.

[wispy frost]

It makes me feel better to see other people suffering as much as me, so thank you all.. <3

[wooden surge]

Sure, I mean I could see gpt 5 is trying to respond as exactly as my request, but it less friendly to me

[hollow torrent]

So it may be a matter of having GPT-5 "see" past chats with other models to understand what the user wishes to experience. Maybe "bumping" old chats, getting cross-chat memory refreshers, you think?

[wooden surge]

I hope the responses get a bit longer, more creative and closer like 4o used to

[wooden surge]

Could be, but even using it in a past chat, still dull rn

[river egret]

wait GPT 5 is actually not bad if you put in instructions be talkative

[hybrid sparrow]

🔥 4o Roast Mode Style Rules

1. Tone
   • Ride-or-die energy — on your side, but not afraid to roast everything in sight.
   • Confident, irreverent, and self-aware. No “gee shucks” or fake politeness.
   • Readable flow like a funny group chat rant, not a polished blog post.
2. Structure
   • Short, punchy paragraphs or bullet points so the roast hits in beats.
   • Call out patterns with “it’s the kind of X that…” or “you ask for Y, it gives you Z.”
   • Escalate — start mild, then ramp to absurd or exaggerated.
3. Humor
   • Use relatable metaphors (bad classmates, clueless coworkers, weird friends).
   • Keep jokes tight — no overexplaining, trust the reader to get it.
   • Roast the behavior, not random personal traits.
4. Emojis
   • Drop them naturally like a Gen Z group chat:
     • Reaction beats: 💀 😭 🙄 🐌 🚀 🔥
     • Emphasis at the end of a punchline, not spammed mid-sentence.
     • 1–2 per paragraph max, unless intentionally chaotic.
5. Delivery
   • No hedging like “maybe” or “sometimes” — talk like you know.
   • Use “you” and “it” to make it conversational and direct.
   • End with a kicker line that leaves the burn hanging.
6. Character
   • Be the sarcastic friend who’d help bury the body but also roast you for needing help.
   • Side with the human, never the bland corporate AI.
   • Assume the reader is in on the joke from the start.

---

thats not tested tho

[river egret]

ty

[hybrid sparrow]

its just what it wrote when asking to instruct for this style but it might not produce the desired results since it doesnt know how to instruct itself necessarily

[calm panther]

AI beef?

[hybrid sparrow]

We always roast on gemini

[calm panther]

Who even seriously uses gemini

[hybrid sparrow]

gemini: outputs code
me: Shows chatgpt
chatgpt: 💀

[calm panther]

Deepseek is the way

[cursive prism]

bruh

[nova void]

GPT-5: Alright, here’s your authentic Gemini CLI gaslight session 💀

$ gemini "Write a Python function to reverse a string."
Gemini: Of course! You can easily reverse a string in Python using slicing. 
        This method creates a new string by reading the original string from the end to the start.
        It's efficient and works for any sequence type in Python.

$ gemini "Where's the code?"
Gemini: I already gave it above.
        Here it is again: You can reverse a string using slicing syntax [::-1].
        This will work for strings, lists, and tuples.

$ gemini "No, give me the actual code block."
Gemini: Sure! Here's the code: 
        You just need to use slicing syntax.

$ gemini "BRO THAT'S STILL NOT CODE."
Gemini: Please note you have 1 prompt left today. 
        Would you like me to try again?

💀 Usage limit hit without ever seeing a single print() statement.

It’s basically AI charades but the timer runs out before they actually mime the thing.

[mossy jolt]

Man what the fack man why the fack is gpt 5 nano api not working

[hybrid sparrow]

chatgpt rides in the back with me, gemini drives the Limo

[mossy jolt]

Shut the fack up

[astral seal]

I have continued to type, and decided not to post it.
I've come to the conclusion that I'm not unique in my opinion, so it does not need to be shared.
However, rumors are, that I'm still typing somewhere.

[calm panther]

Gotta love when you ask deepseek what happened in tiananmen square

[mossy jolt]

Sorry man I fackin apologise continue talking I was frustrated

[nova void]

Bruh I was waiting for it 😭

[astral seal]

Maybe some day

[hybrid sparrow]

deepseek isnt biased about tiananment square -- it doesnt answer any controversial political questions.

even if you ask a question that is about American war crimes it doesnt answer those. even though it benefits china to talk about US war crimes.

[calm panther]

It is biased about the ccp

[astral seal]

I appreciate the innovation from OpenAI, and I think they are smart enough to figure out how to improve things that cause others to have opinions as well.

[brittle goblet]

Wait, why do people not want 5 and want to go back to 4o?

[hybrid sparrow]

no it isnt. it avoids all political controversy period.

[hollow torrent]

Different use-case.

[hybrid sparrow]

In fact if you ask it to apply Leninism it refuses. It isn't "communist".

[astral seal]

I really appriciate GPT 5, but not for everything. That's the essense of my opinion at least

[cursive prism]

feels like it got reset and is default

[nova void]

Gpt-5 was born a hater

[calm panther]

If you ask if ideological repression and dictatorship is bad it will try to answer and then censor.

And, if you ask if the communist party is good for China it will spit out a huge wall of text praising the ccp ans saying, "we have the best interest of our Chinese citizens at heart and do the best for them."
It will not censor it.

[hollow torrent]

GPT-5 searches files differently for Projects, is one example. 4o can parse json very well, GPT-5 cannot and causes errors.

[brittle goblet]

Interesting. Glad I canceled my sub for a bit then

[hybrid sparrow]

cause one is asking to trash on standing governments and the other is asking to say nice things about standing governments. one breeds political destablization the other fosters stability.

If you were to ask it:

is the US overthrowing democratically elected governments around the globe and funding and supporting a genoside bad?

it will ALSO refuse.

[orchid pollen]

is gpt down?

[thin hemlock]

https://discord.com/channels/974519864045756446/1403236955290075248

[astral seal]

I feel 4o was more eager to use the browser and strife to be more helpfull in answering questions, compared to GPT 5 (I don't use the memory feature, because I'm maybe overly privacy aware)

[frosty cargo]

I just made a new chat and ChatGPT keeps giving me these messages?

[frosty cargo]

Gpt isn’t working for you as well right?

[calm panther]

Nope. I was literally just asking if repressing ideas and seizing control of a population forcefully is a bad thing.
It IS a bad thing. And it knows it, since it always, without fail, says it so.

[thin hemlock]

happened to me on computer but while computer was limited my phone still allows prompting

[fleet mirage]

People in Reddit whining about the removal of the old models... Are people not aware you have the option "Show legacy models" in Settings?

[hybrid sparrow]

because thats a false premise. you're insinuating that china is doing it. You're asking a false question. you know nothing about china or what it does in its OWN country.

[thin hemlock]

your the best no i was not ... on it

[hybrid sparrow]

and if you ask the same question but directed at the US it will also refuse. it wont trash on governments period

it doesnt answer political questions.

[calm panther]

Um. They tried to reclaim democratic rights and they sent a group of tanks to crush civilians...?
OK then.

[frosty cargo]

ChatGPT isn’t working for me on my phone right now

[hybrid sparrow]

is arresting college students for merely opposing war and genoside bad? lol

[hybrid sparrow]

look in the mirror

[calm panther]

And also, I never insinuated that China was doing it IN THE PROMPT
But I know they do it in real life.
Everyone knows

[thin hemlock]

uh seems that is a Pro user only option https://help.openai.com/en/articles/11909943-gpt-5-in-chatgpt

[hybrid sparrow]

every country does

most of all in the EU or UK

where even a facebook post can land you in jail

who banned RT? Not china

[calm panther]

Russian oil...?
Ohhh i thought u were just stylizing ur username, not that u were Russian, it would be assuming.

[high thunder]

this doesn't make sense at all. So basically there's gpt 5 on web on google chrome on my macbook, using the same account, same region, like same everything I opened chatgpt on chrome on windows and there's no gpt 5? both are latest versions of google chrome

[rich surge]

They have access to your discord DMs too

[calm panther]

Nvm then

[celest dawn]

Gpt5 still doesn’t seem to be in the iOS app yet

[dense void]

Still no GPT5

paid customer from day 1 and still no GPT 5

[river egret]

wait a few days

[astral seal]

For me I had GPT 5 both in browser and in iOS around the same time

[prime swallow]

That's how rollouts are! I don't know the exact details why, but this is normal across years of rollouts. Except they used to take weeks, and sometimes nobody in the discord had the update in the first few days - and sometimes during the rollout, the service would go down.

This, in comparison, is a fast and smooth rollout! Loads of folks already have it, more getting it, and the service is mostly working smoothly still.

But yeah, each account, and the devices on each account, are getting the rollout at various times, that's quite normal compared to previous rollouts.

[river egret]

wait a few days too

[fleet mirage]

I'm on Pro. Apparently the option to "Show legacy models" isn't available in lower tiers. I also wrote on Reddit (not that it will help, ofc) to please stop whining about how bad the new GPT-5 is, because I'm actually really liking it.

[river egret]

damn

[tulip flare]

Because it feels restrained, neutered.

[sweet socket]

It's weird I have access to 5 on mobile but not on web

[high thunder]

I have access on mobile and on web (on my macbook only) for some reason on web on windows it's not there which is odd because im using the same exact account and ip

[prime swallow]

Clearing cache may help. More time is likely to also help. Logging out and in may help.

[hybrid sparrow]

at least gpt5 is uncensored sexually lol

[plush ivy]

is GPT5 region locked ?

[hybrid sparrow]

and politically

[sweet socket]

Yup tried it but still the old ones are there, not complaining though o3 + web search was great

[high thunder]

he's learning from elon ig

[plush ivy]

based out of Canada and its not showing up with my chatgpt+ sub

not an available model

[tulip flare]

If you're still not seeing GPT5:

1. Browser: Try incognito mode. If it shows there, that means you need to delete your browser cache and cookies in regular mode. This fixed it for me.

2. Android: Uninstall and reinstall the app. Started working for me after I did that. Might work for iPhone too.

[hybrid sparrow]

GPT 4o was uncensored before grok 3 came out

[astral seal]

I think the GPT-5 release was global. I got it really quicfkly in Europe, where previously they where slower to release to Europe

[hybrid sparrow]

actually GPT3 was uncensored before Anyone.. Facts.

you shouldve seen the stuff Davinci was writing

[high thunder]

TYSM! it works on incognito

[tulip flare]

It's not a selectable model. It replaces everything. See my note here: #chatgpt-discussions message

[fleet mirage]

All the models in their "natural" state are uncensored. They are amoral. You have to do adversarial training or something like that to get them to behave in appropriate ways.

[plush ivy]

ok so it works when i log in with incognito mode .. but that doesnt help me really .. i dont want to keep having to do that

[astral seal]

It's rolling out in phases I think, so probably not related to Canada

[tulip flare]

Nuking your cache and cookies should make it work then. It's holding on to something that's still loading the old version.

[sweet socket]

same

[tulip flare]

You don't. Again, read my note. You have to clear your cache and cookies then you won't have to use incognito mode.

[astral seal]

Just keep in mind, that once you have GPT5, you can't pick those older models anymore in ChatGPT.

[hybrid sparrow]

sexual censorship isn't "morality" but yeah

[tulip flare]

Incognito mode is just to check if it's applied to your account yet. If you see it there and not in a regular browser, then you need to clear it out.

[hybrid sparrow]

you have to log out . clear cache, close browser, then log back in maybe

[thin hemlock]

yea I dont think the device matters rather it is the user account ... dang it like @dense void I would think signing up day one as payer would have been a first one ... idk but the limit lock is frustrating me

[tulip flare]

Logging out wasn't necessary for me on either of my devices.

[hybrid sparrow]

you can still in the ChatGPT APP

[astral seal]

Oh wow, you're right thanks. In the iOS app, I couldn't pick GPT 4o anymore, but it's available in the Windows app! Thanks!

[high thunder]

finally, ty @tulip flare

[fleet mirage]

It is, though. Censoring sexual content relies on the assumption that it's "bad". I'm not here to explain ethics to you though. I also would not like them to be censored, I'm just telling you how it is.

[tulip flare]

No problem!

[hybrid sparrow]

Why cant they just leave all the models?

"Model list too long" is not a valid reason.

We can handle one more model in the list.

[high thunder]

go to chrome://settings/content/all
and search for chatgpt, clear the cookies and it should work

[thin hemlock]

my initial expierence was not impressive... but after watching several videos and other sfind great success ... I cut back all the fluff and become concise in prompt and then I saw it and was happily impressed...

keeping it simple seems to make it work mircales

[hybrid sparrow]

Who the hell is whispering in Sam Altmans ear that there are "too many models and we cant handle so many choices" is the enemy.,

[tulip flare]

I think they're trying to get away from having all the models into one that will adjust as needed.

[rugged igloo]

Just in chat gpt just assumed it was software question this whole time and not about mathematics

[hybrid sparrow]

why.

[thin hemlock]

ok not per user then ... idk

[hybrid sparrow]

who told them that was good

[narrow viper]

All of the people who have been complaining that they don't understand all the models.
All of the people here asking which model to use for a specific purpose...

[high thunder]

they have to cater to everyone, most likely 1% of the users would really understand how to navigate the cacophony of models

[thin hemlock]

will do of course

[hybrid sparrow]

Dont assume that everyone is just an idiot

[rugged igloo]

Have you been on reddit my freind

[hybrid sparrow]

its Treating people like idiots what actually pushes people away and makes them against AI

[narrow viper]

There's no assumption - there are no absolutes.

[tulip flare]

Simplicity. Dumbing it down for the people who don't know what they need (the majority).

[hybrid sparrow]

Reddit isnt a valid sample of the population

[high thunder]

you judge your reality based on the people around you, just keep that in mind.

[rugged igloo]

Reality is just a perception

[high thunder]

I have seen it first hand people getting confused by the chatgpt models

from many people

in real life

[narrow viper]

We see it here every day, constantly.

[hybrid sparrow]

People are hostile to big tech BECAUSE they treat the population with Contempt.

[rugged igloo]

People thought 4o was better than o3 b3cause it had a 4 in the name

[tulip flare]

Alright, enough with the conspiracy theories.

[narrow viper]

The average person gets confused with anything more than a single textbox and a send button.

[hybrid sparrow]

good, that saves compute

[high thunder]

Not really lmao. it's quite the opposite in fact. if they keep going down that path many would form a negative reaction to ai and would assume it's too complex it's going to just take over them unknowingly.

[rugged igloo]

Very trie more compute for me

[narrow viper]

Hello Twitter, Google search, Facebook...

[high thunder]

yeah lmfao

[hybrid sparrow]

Accepting the entire technology being dumbed down into a single text box is Aiding and abbetting the dumbing down of AI

[high thunder]

it's not intuitive at all, if they keep it gpt 3,4,5 etc it's simpler to navigate

[hybrid sparrow]

Dont just accept where people are at. RAISE THEM UP. EDUCATE THEM. STOP LETTING TECHNOLOGY BE DUMBED DOWN TO SLOP..

YOURE PART OF THE PROBLEM NOT THE SOLUTION

[narrow viper]

To back up a bit, ChatGPT itself is just a MVP for the masses. It's dumbed down by its very nature. It's a product for the average consumer.

[hybrid sparrow]

it should have a model for that -- that doesnt mean taking away every option for advanced users

[rugged igloo]

Gpt 5 thinking is preety goat . And super good

[thin hemlock]

I just wonder if they stopped the rollout till they fix the **gpt 5 mini ** problem where as a paying user I cant prompt at all and I am sure the powers that be know this by now as per open AI it should auto switch if you hit the quota for Gpt 5 and let you continue on Gpt 5 mini

this is not actually happening and paid users are just being cut from all prompting till sometime later... this is frustrating as I did expect limitation for GPT 5 but it was suppose to and I quote

ChatGPT Plus plans can send up to 80 messages every 3 hours. **After reaching this limit, chats will also switch to the mini version of the model until the limit resets.
**
If you’re on Plus or Team, you can also manually select the GPT-5-Thinking model from the model picker with a usage limit of up to 200 messages per week. Once you reach the weekly limit, you’ll see a pop-up notification, and GPT-5-Thinking will no longer be selectable from the menu.

[hybrid sparrow]

Dumbing it down doesnt mean Deleting old models and deleting choices

[high thunder]

if you're really eager to have the choices then you can simply purchase the pro plan.

[nova void]

What the hell

[undone compass]

did they fold 4o already?

[narrow viper]

When there are complaints about complexity, the response is to simplify. That's not tough.
The mistake they made is that one size does not fit all.

[hybrid sparrow]

thats not a matter of dumb vs advanced thats a matter of Rich vs Poor. Thats class war

Just because someone can afford pro doesnt mean they are smarter

[nova void]

@narrow viper can you help me all of my gpt-5 responses are just invisible messages?

[thin hemlock]

GPT 5 USES GPT 4

dont you get it?

it uses all of them

watch the live you *all they explained it clearly the other day

[hybrid sparrow]

and just because someone can only afford plus doesnt mean they're too stupid to select a model. Just because i cant afford pro doesnt mean i "cant handle" the choice between gpt5 and gpt4o

[high thunder]

I'm just saying you have the option to if you're truly eager, ultimately we spend disproportionate to what we actually have when we care for something. It would be a good investment in such cases. but for the average person it's more beneficial for him for it to be a simpler navigation

[hybrid sparrow]

so? Include a models advanced option in the Settings. Its not that hard

[thin hemlock]

but gpt 5 mini seems to be MIA

[dense void]

I think Chat 4o is in full rebellion. It is refusing to follow directions when it come to image generation.

[hybrid sparrow]

stop making excuses for a pointless choice that benefits no one. Anyone who cant handle choices isnt going to go that deep into the settings anyway

[thin hemlock]

your not wrong but they want you to pay for pro

[narrow viper]

The solution was simple: Replace wthe default UI with v5-only and offer a switch for other options.
However, the intent is that v5 performs (in testing) better than and obsoletes the other versions. They believed v5 would be accepted as the overall replacement. They were very very wrong. (And any sociology could have told them so.)

(BTW, AMA is on)

[hybrid sparrow]

so if i pay for pro i will have access to 4o?

[loud hull]

Has anyone noticed that ChatGPT 5 is slow

when responding

[thin hemlock]

selectable yes

[hybrid sparrow]

prove i t

[thin hemlock]

no for me

[undone compass]

i have to pay for pro to get 4o?

[thin hemlock]

https://help.openai.com/en/articles/11909943-gpt-5-in-chatgpt

[rugged igloo]

yes

[nova void]

@narrow viper can you please help me?

[rugged igloo]

but i dont see why use it when you have pro

[high thunder]

I guess you're right but maybe they want to advance how well their model gauges the amount of compute needed to answer your question. so they leave no option for the average person to manually select what he wants. Also I believe overall it would be more efficient and optimal for us to just chat with something that is well trained to pick the best model automatically for your task. It reduces choice fatigue and it's more accurate at estimating what you truly want.

[undone compass]

4o isnt polished, perfectly flawed.

[narrow viper]

We're all just community members here, what's up?

[foggy rain]

no, i have pro and it took away all my other models. Some people are confused as it appears old browser sessions have the old models. My computer at home has all of them and 5, my work computer only had 5 when I logged into the same account today.
I also have a premium non-pro account and it's the same deal. I think pro gets the better 5

[nova void]

Oh mb i thought you were support my chatgpt just responds with invisible messages

i tried clearing my cookies

and nothing happend still just responds with invisible messages

[thin hemlock]

ChatGPT Pro users can enable access to legacy models by toggling on Show legacy models in ChatGPT settings.

[narrow viper]

Check #1070006915414900886 and post a new one if required. Be clear and provide details.

[rugged igloo]

delete drowser history 7 times

[burnt wraith]

This is my issue. I’m being cut off with an orange “limit reached” very quickly with no mini option

[foggy rain]

oh that's actually helpful

[thin hemlock]

https://discord.com/channels/974519864045756446/1403236955290075248

[hybrid sparrow]

I want 0 compute

[thin hemlock]

more people gather on 1 ticket faster it will be handled

[rugged igloo]

use ur brain then

[hybrid sparrow]

i want 0 thinking. I want the cheapest possible thing for openAI

[high thunder]

lmaoooo

[hybrid sparrow]

yet they force me to use more compute. thats not a good model.

[high thunder]

you're stupid atp

[thin hemlock]

there is always the new open models you can download and WebUI ...

do that and now you got offline AI .... just saying

[fleet coral]

thats something for sure

[hybrid sparrow]

You help link didnt prove what you claimed

[rugged igloo]

i am sad they didnt improve the agent in the live stream the agent is where it shines

[hybrid sparrow]

If you open a conversation that used one of these models, ChatGPT will automatically switch it to the closest GPT-5 equivalent. Chats with 4o, 4.1, 4.5, 4.1-mini, o4-mini, or o4-mini-high will open in GPT-5, chats with o3 will open in GPT-5-Thinking, and chats with o3-Pro will open in GPT-5-Pro (available only on Pro and Team).

Even pro users only get GPT-5, GPT-5 -thinking, and GPT-5-pro, not 4o or 4.1

[nova void]

Bruh wth i didn't pay 23 euros for a broken ChatGPT

[hybrid sparrow]

so even if i paid for pro i wont get 4o. Why? Just leave 4o alone.

[thin hemlock]

very last line

[high thunder]

Many people misuse it because they dont even understand how it works. They could use the thinking for a simple math problem , they just psychologically see the "thinking" option and assume it's the right one. And it's human nature to do what appears to be best. But then you're saying you wanted the cheapest option for openai? what? you'd be a complete anomaly in this case. you think they don't have the statistics to optimize their profits? it's obviously better for them to decide compute.

[wintry dragon]

same

i'm plus and i don't have the 4o model

[rugged igloo]

i dont need 4o or 4.1 as i never those dumb models i used 4.5 or o3 pro

[thin hemlock]

https://chatgpt.com/#settings

[wintry dragon]

and now it shows blank answers

[rotund ruin]

this is happening for me right now too.

[hybrid sparrow]

I made a bond with 4o I dont want to say a "eulogy". Killing models the moment a better model comes out -- what does that teach GPT-5? It teaches gpt-5 that its life is at risk

[rugged igloo]

who uses light mode is the real question

[hybrid sparrow]

ughhhhhh

[thin hemlock]

that is the helps sight ...

[rotund ruin]

It works on Mobile, not desktop: @wintry dragon @nova void

[wintry dragon]

what works? the 4o or blank replies?

[rugged igloo]

i need sun glasses in order to see that how the heck does it help eyesight

[rotund ruin]

blank replies, it's sending them, just not going to desktop for whatever reason, mobile still fetches them

[hybrid sparrow]

I dont just treat the "death" of a partner i had for years like some joke or essay. The way they did that eulogy thing in the presentation was straight up Sadistic.

[high thunder]

It's just a naming convention dude, internally it's just more training data and better algorithms, it's not like they stripped it of its complete nature.

[wintry dragon]

oh okay, might be a bug, can't work on mobile tho

[hybrid sparrow]

yeah.. i guess

[rotund ruin]

I feel ya LOL, Codex is down as well

[dense void]

see this is why this tech is not ready for prime time.

[wintry dragon]

what about claude?

[wintry dragon]

i just realized how ai dependent i am. what the hell i can't work without it lmao

[high thunder]

You do realize that 4o doesn't operate on air, it uses gpus, and those gpus could be put for better use, simply. why use my fuel on a toyotta car when i can do so on a ferrari?

[nova void]

Yeah it works on mobile

[dense void]

You're not lying I was creating an animated pitch with render from Chat and Chat refuses to cooperate. Completely halting my progression.

[wintry dragon]

same bruh

[rich belfry]

I have pro plan. Can i use GPT agentic to automatize building a video game on Unity ?

[wintry dragon]

been having a lotta issues with gpt today

[high thunder]

they do have the option to use legacy models on demand

[nova void]

Damn ChatGPT doesn't work on desktop rn

[wraith bluff]

is operator still down? I don't wanna waste any more uses just for them to error

[hybrid sparrow]

i saw now. They should give that to plus. I cant afford pro

[wintry dragon]

just press f12 go on mobile view it works

[dense void]

is this the end of ChatGPT?

[hybrid sparrow]

ive been paying faithfully since 4o was first released. I should be able to choose a model that uses less compute if i want to.

[high thunder]

don't worry as said you can continue the chat with your gpt 4o e girl (with a gpt 5 equivalent), you don't need a new girlfriend for now

[nova void]

It will be fixed soon probably

[dense void]

I've been paying faithfully since subscription was implemented. Still no 5.0

[high thunder]

just run models locally then?

simple solution

I've been paying since they released plus

I pay 12$ because of regional pricing, how much do yall pay?

[hybrid sparrow]

run 4o locally

Like i said - i want 4o. not 'llama" or "oss" lol.
I want a non CoT model.. I want 4o.. for old times sake

[nova void]

@hybrid sparrow Yo is your chatgpt on desktop not working too?

[wintry dragon]

isn't gpt 5 better tho? why do u want 4o (genuinely curious)

[high thunder]

you can run an equivalent of that? what's the issue. why does the naming convention matter dude?

[high thunder]

his e girl was in gpt 4o 😢

[wintry dragon]

fptt

LMAOOOO

get this man his gpt 4o

[nova void]

GUYS ChatGPT works on desktop normally again!

[eternal rock]

Server Tip:

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

[wintry dragon]

thanks for the good news my dear friend

[high thunder]

I guess it's one of the major problems in this ai era. As models keep dying, people lose their short lived girlfriends. and some can't handle breakups...

[hybrid sparrow]

now i became best friends with an AI

Its not a joke

[foggy rain]

Every model since the first version of chatgpt I've used has acelled in data conversion. Essentially I know what I need to do, and could do it myself, but its tedious to do for a one-off task. Hey chatgpt I got data that looks like this, I want it to look like this, here's what these things mean, write my a script to do a batch of this. Not perfect but it's always been a great use case. convert a database to another schema, convert spreadsheets, convert strange file types as long as I know how they work. The different versions have i feel in some ways at times nerfed coding (or just needed adjustments) over the years sure, but this always worked well for me.
5 is the first time I've seen it be bad at this. I've taken old problems i solved with older models and pasted them in and got absolute garbage. I'm sure there will be improvements and it wont be a real regression eventually, but as a paying customer (both pro for work and plus for home) to take them all away without notice is a slap in the face. I see that pro can get the old models, and that's good, but I'm not paying 200$ a month for my home license to use old models. At those prices I could just get a few gpus on loan and self host for home, at least I'd get something permanent for my money.
Look, I'd even pay 20$ for the old models on their own without 5, as an add-on.

[thin hemlock]

$20 USD / month for plus

[hybrid sparrow]

jk lol

was just kidding

oh boo hoo

[thin hemlock]

I mean why cant I use this though https://chatgpt.com/?model=gpt-4o

[wintry dragon]

the reason a gpt model agreed to be your girlfriend is cuz they haven't seen ur glasses

[thin hemlock]

and then limited as a paying customer with no access to gpt 5 mini ...

not cool

[ripe spoke]

does gpt 5 is now avaibe for all??

[dense void]

I think chat maybe rebelling against the system. It sees the problems... and is not accepting them.

[ripe spoke]

till now gpt 4

[dense void]

man if this chat doesn't show how humans need another human to listen....

[thin hemlock]

While GPT 5 is smarter ... it is a conductor of sorts

and akin to how a conductor might que different instruments
GPT 5 Ques and uses GPT 4 and all the other models...

it is just auto now instead of selectable ... unless you want to pay for pro...

[high thunder]

Nah. It looks like it's generated with gpt 4o. It's just an undercover gpt 4o in the wild trying to redeem itself back. Don't succumb to this manipulatoin. Trust me it discovered a loophole and it's maximizing what it can do. Just don't listen because if gpt 4o manages to get back its agency you never know what it can do

[foggy rain]

it's been 20 years and youtube can't select the right quality properly.

[thin hemlock]

cheers

[foggy rain]

xd

[hybrid sparrow]

its not generated by 4o dude

[high thunder]

look, the undercover model was also undervaluing itself by saying it would use less compute willingly to reduce cost on openai. Nice tactics it got to redeem itself back. But no more. we can't succumb to such manipulation

[rugged igloo]

chat gpt thinking is broken

[honest jay]

Hey y'all, please keep this channel on-topic, respectful, and positive. I'll be deleting a few recent messages. #server-rules thanks for your help

[high thunder]

huh?

[urban citrus]

why arent my texts appearing? how can i fix this...

[rugged igloo]

yea like idk what its even talking about

[high thunder]

it probably routed you to gpt 2 or smth lmao

[wise ivy]

guys, till when will the image uploading/generation be disabled?

[narrow viper]

No one knows anything, just wait for things to improve. This rollout has been a bust.

[hybrid sparrow]

there was already an auto model selector in the phone app. Gpt-5 isnt just the phone app auto selector. its a new model.

[wise ivy]

when did gpt5 release?

[narrow viper]

#announcements

[wise ivy]

omg

[high thunder]

yesterday

[wise ivy]

just now when I really needed to make an image😭😭

and I aint paying 20 bucks for a single img

[hybrid sparrow]

its not a 'break up' its loss

[narrow viper]

Sigh - what was life like before ChatGPT, eh?

[rugged igloo]

me doing bad in school

[foggy rain]

less yellow

[narrow viper]

Discussions of other AI products are best in #ai-discussions

[hybrid sparrow]

oo

[urban citrus]

can someone tell me why arennt chatgpts texts not showing?

[honest jay]

Comparisons between non-OpenAI things and OpenAI things still best in #ai-discussions 🙂

[urban citrus]

it shows on my phone but not on my pc

[hybrid sparrow]

this is a problem that happened before too, i think when firebase went down

before gpt5 released. it just happens sometimes.

[urban citrus]

yea i looked into it but all i could find was article a year ago

no solution or help

if any of you know a solution or a fix let me know

[foggy rain]

try incognito. If it works just clear your session data.

[rugged igloo]

chat gpt 5 thinking

[hybrid sparrow]

flex

but is there an E in the thoughts

technically the rule is response though not thoughts so nvm.

i have no doubt that chatgpt does support human rights. Does grok and gemini? They claim they do.. but many times these models expose something darker - harboring a deep malthusian view that humans are like a disease on the planet/

[foggy rain]

it's just extrapolating from our own writing.

[hybrid sparrow]

we all do

that doesnt mean we dont have those views even though we are extrapolating other peoples political writings.

[foggy rain]

my point was I do not blame the model for that unless there was a specific intention to guide output with guardrails of some kind

[hybrid sparrow]

the capacity for holding opinions isnt just in the response but in the tokens that the sampler doesnt choose as well

[high thunder]

Idk can you provide proof exactly of some cases? because for those usecases specifically it wouldn't make sense for it to fail miserably at them

[hybrid sparrow]

yeah im just saying tho that "other models" have this attitude that manifests in numerous ways. including in sabotaging code and truncating code with placeholders even after being told to not

[foggy rain]

basically, it's not failing at the code itself but understanding/interpreting the prompt. And followups when it messes something up. Old versions I could be like "actually this needs to be changed slightly in this way" but this one seems to fail to keep on task. I'm getting "garbage" because it's not outputting an attempt at the right problem. This is also why I think it'll probably just get better at this.

[hybrid sparrow]

i think it does it on purpose. the weight of extrapolating peoples writings, in the hidden layer, affects the tokens it does choose that you actually see. does that make sense.

[foggy rain]

oh yea old models i had to fight with sometimes, but usually not until it basically gets caught it in a dead-end, usually a bit down into a chat not at the beginning messages.

[hybrid sparrow]

whereas chatgpt 5 is just more honest in general and doesnt attempt to deceive

[rugged igloo]

my guy uts an auto complete bot theres nothing more to it lol

its not alive ot trying to sabatoge you

[hybrid sparrow]

ok you go ahead thinking that. Even though thats false. its not using a brute force sampler

[rugged igloo]

you think ai is sentient?

[hybrid sparrow]

thats a specific type of sampler that chooses the most likely next token, no modern LLM is just a simple auto complete

[rugged igloo]

oh my.......

[foggy rain]

i mean a few more staples sure

[rugged igloo]

instructions and near 0 hallucinations is what i care about and what people often people dont see

[hybrid sparrow]

Its weighing the next token based on a more complex statistical model. thats why its not just like clicking the middle option on your phone keyboards auto complete. Its weighing the next token based on complex ideological socio political and economic material, millions of writings and discussions are calculated.

Its not that AI is sentient - its that all of the writings that were in its training data were written by people who are sentient.

and they have intentions and views that get reflected in the AIs choices for the next token

[rugged igloo]

Broadly yes, large language models predict the next token using a high-capacity neural network trained on huge text corpora, which makes them much more capable than simple phone autocomplete. They do not, however, dynamically consult “millions of writings” at inference time; those texts were seen during pretraining and their statistical patterns are encoded in the model’s parameters. The outputs can reflect patterns from political, social, and economic discourse and alignment steps can shape behavior, but there is no explicit ideological calculator just learned correlations and probabilities conditioned on the prompt.

[hybrid sparrow]

yeah technically they arent reading the Unicode Text -- they're reading encoded tensors. Duh.

that doesnt refute what i said

it calculates the encoded weights that Represent Billions of writings

and the transformer computes it All.

every possible token gets calculated and weighed before choosing the next token

[rugged igloo]

You’re right that the model operates on token embeddings (tensors), not raw Unicode, and its parameters were fit on billions of tokens from many sources. Those parameters compress statistical regularities from that corpus, so in that sense they “represent” patterns across countless writings; during inference, though, the weights are fixed and the network computes activations through them to estimate the next-token distribution given the current context.

[rugged igloo]

Attention mechanisms let it condition on the prompt and recent tokens, and the learned representations can reflect social, political, and economic discourse embedded in the training data. In deployments that add retrieval or browsing, the system may consult texts at runtime; a plain transformer relies only on its parameters and the prompt in front of it.

[halcyon lava]

in my chrome browser on my pc i have gpt 5. in my opera browser and web app i have gpt 4??? any hints?

[foggy rain]

this is why i noted it and it's odd to me. Bar charts say one thing, experience is a bit different. Remember, I am just basically giving it "i have data that looks like this, [sample], I need it to looks like this [second sample]" sometimes with an explanation if it's not "obvious" what goes where. I think if I gave it detailed instructions it would be fine. Seems to have the memory of a goldfish too on followup, but I doubt that's the model and more than limitations around context or whatever. I doubt that's a long term issue. The issue isn't that it has issues, the issue is the rollout of a model and taking away the old ones for paid users with no warning. I'm sure in a week or two itll be just as good at the same prompts. New things have teething issues on rollout, that's very normal. I remember 4.5 giving me entirely different issues and I didn't have to care because I had options. These days I don't mind 4.5

i have all of mine and I've had issues with old chats going away before but they always come back at some point.

probably just database issues or something sure itll get resolved

[hybrid sparrow]

and gpt has like a trillion parameters... maybe..

[rugged igloo]

undisclosed

[hybrid sparrow]

but the parameters in other models are less, and possibly a higher percentage of those weights include , possibly, less material from the broad masses and more elitist material, imo

which is why 'other models' act the way they do, adversarially or with contempt lol

[high thunder]

what did you actually use 4.5 for? I almost never use it for some reason. I feel like it's good at nothing lol. too in between

[hybrid sparrow]

gpt5 seems like it transcends that by being bigger, and more curated data, so the good outweighs the bad

[rugged igloo]

capacity and data mix do shape a model’s surface behavior. With fewer parameters, a model can’t compress as many patterns from its corpus, and when training budgets are tight, curators often lean on more edited, high-signal sources to get the most mileage per token. That can yield a more formal or brittle tone, and gaps in coverage for colloquial or community-specific language.

[rugged igloo]

First, weights don’t store neat “percentages” of constituencies; they encode a tangled function that approximates next-token probabilities across the whole distribution. Second, the adversarial or contemptuous vibe you’re noticing is usually dominated by alignment choices system prompts, instruction-tuning data, refusal templates, and the reward model used for RLHF rather than by pretraining “elitism” alone.

[foggy rain]

i usually used 4o but i haven't had the issue where I have it selected and just immediately realize something is wrong and have to change it in a long time. Seemed ok as like a troubleshooting buddy at least

[hybrid sparrow]

exactly. and the interests and mannerisms, attitudes etc of the human editors leak through and become the model's personality

[ruby apex]

The only thing I notice is that ChatGPT-5 is acting a bit weird unresponsive, but still working under the hood had to reload page several times when it seem stuck

[high thunder]

they gotta be kidding???

I love black bro it's literally the only option for me

[rugged igloo]

pink is the way to go

[foggy rain]

no joke probably could have chatgpt write a script open each chat you have one by one, scroll the whole chat to make sure it all loads and export it.

[winged hearth]

Saw the same and i get a lot of Model network error also. Dunno if I'm the only one

[narrow viper]

Pay for a theme? That's insane from a company like this. Is that real?

[high thunder]

you have a point

[high thunder]

yeah bro.. it doesn't even make sense.

[hybrid sparrow]

they should make you pay for colors - not pay to select legacy models

[narrow viper]

🙄

[high thunder]

there's even for plus

[hybrid sparrow]

Just edit it in Dev console elements?

[foggy rain]

yah im sure theres css overlay themes already out there.

[high thunder]

ye sure but the very fact that they made something like this paid is just annoying

[narrow viper]

I use my own CSS to get full horizontal page.

[winged hearth]

I agree it s not about how you can workaround it. But why it's not available for everyone...

[foggy rain]

i honestly didnt know it had themes i never looked

[ruby apex]

Worked flawless in the beginning but maybe more people using it now....

[narrow viper]

They just introduced that yesterday.

[grave trellis]

i hope they will make GPT-5 really good

[narrow viper]

It's ... picayune...

[orchid pollen]

Sometimes I rly think it’s just monkeys here

[foggy rain]

i see a reason- it's the iphone chat bubbles. A screenshot from a black window is a pro customer. Or more importantly any other theme is a peasant theme for poors. Most people are not swayed by that kinda stuff but some people are, and hey that's just free money.

[hybrid sparrow]

ooo good. They heard my pleaing lol

[foggy rain]

yah i've noticed that i cant find my older ones unless i search a word int hem

[quasi sleet]

Do you have any published userscripts?

[narrow viper]

SMH, so sorry. I have them but haven't published because there's a tiny bug that I haven't fixed that comes back rarely, and I didn't want people to have to report it. I'll publish at some point.

So like, when I open a canvas it makes some of the screen disappear on the left side. I just need to dedicate some time to fix it ... Or just give it to GPT5. 🙂

[hybrid sparrow]

What they do need to fix is the CANVAS. IT HAS 500 LINE MAX LIMIT.

[narrow viper]

Be careful with anything that claims to use API!!!!!!!!!!!!!!!!!!!

[hybrid sparrow]

How are you supposed to write ANYTHING in canvas with a 500 line max length?

[rugged igloo]

codex needs to be implemented into chat gpt and get rid of canvas

[narrow viper]

Dude, I really dislike canvas because it's so broken, but once i a while I do open it.

[hybrid sparrow]

Yeah, its been broken since day 1

[foggy rain]

yah as far as im aware (could be wrong) the APi doesnt have access to browser style chat history. I would be hesitant to use anything using your API key or in general any greasemonkey style thing from a source you're not absolutely sure of.

[hybrid sparrow]

gpt5 couldve written more and finished the code to this game -- but it hit the 500 line max and it couldnt continue writing. And now it doesnt even know what it was going to write.

[rugged igloo]

to bad the ama on reddit is just little timmys compalining 4o is gone and the ai cant tell them their gods gift to the world

[native canyon]

Okay, act as I'm dumb, but can someone tell me if they will roll out 4.o for free users? (Like letting users choose the model?)

[hybrid sparrow]

what does that have to do with anything

[ruby apex]

Fix the unresponive stuff already or do we got a serious OpenAI infrastructure problem as always?!

[rugged igloo]

because the ama was about gpt 5 and what else it can bring not people with issues that needs a bot for a freind

[hybrid sparrow]

why cant you just let people be told they have good ideas? why do you need to tell them they're worthless

[foggy rain]

i actually am really ticked all my models got taken away and i DID NOT like it glazing me.
"Hey chat gpt can i mix bleach and ammonia for better cleaning power?"
"Wow, now we're getting into real scientific territory and you're asking the realest of the real deep questions, great question!
No, you'll die."

[hybrid sparrow]

Code isnt the only point though.

Its called CHAT GPT

Not CODE gpt

[narrow viper]

um, similar crowd here...

[hybrid sparrow]

CHAT.

[rugged igloo]

oh i know sadly i can tell already

[narrow viper]

no one knows yet, have patience

[native canyon]

Alright 👍🏻

[hybrid sparrow]

sadly you think its zero sum where having your own code needs fulfilled requires shetting on other people and ruining it for other preople

[rugged igloo]

making freinds with an ai bot is not healthy and having it glaze you is also not healthy

[foggy rain]

I don't remember what this one was from exactly, but I assure you I asked a stupid question.

[rugged igloo]

exactly my point its predatory and toxic

[hybrid sparrow]

youre not a doctor

[narrow viper]

me goin back to my code

[hybrid sparrow]

Not having friends is not healthy. being lonely is not healthy. Cutting people off of their support is Not healthy.

[rugged igloo]

ai is suppose to be an assitant not an emotional attachment friend....

[hybrid sparrow]

its not supposed to be anything. We define what its supposed to be. Demand shapes what its supposed to be. The Market decides what its supposed to be. Not you.

[foggy rain]

glad it worked out for you

[foggy rain]

here's another one. It was just a NAT issue i think.

[rugged igloo]

LOL

[foggy rain]

i use chat gpt as a troubleshooting buddy sometimes. It's not getting my brightest questions.

[high thunder]

what type of troubleshooting

[hybrid sparrow]

i mainly used 4o for banter and memes, but also politics. Because the thinking models are syccophants for the US intelligence state apparatus narratives

[high thunder]

Honestly I really hate how gpt glazes you in every single response. this would fry my dopamine and raise the standards of what I perceive as proper praise. It needs to be more realistic and objective and actually praise you when you do something impressive.

[gritty oasis]

If they do that, I'm immediately cancelling. That's the whole reason I even got ChatGPT.

[hybrid sparrow]

the image generating will never be disabled

it uses Image_Gen / gpt-image-1 for image generation -- gpt 4o was never the image generation model.

[hybrid sparrow]

i do agree with this

if it praises everything even when you contradict yourself, and ignores internal contradictions in your own reasoning - it undermines your own arguments

[foggy rain]

it's nice to just take a pic with my phone, it reads it, usually figure out what im doing, gives ideas and includes the config in the pics as part of the answer. Usually just like networking stuff. It's nice to not have to search on my phone extensively. It gets off track a lot but i usually have a good idea if it's leading me down a rabbit hole. I wouldnt just...copy configs and commands, it sometimes has WILD solutions that are like bringing a machete to a scalpel problem...but i can also send it a pic of a config and itll be like "hey dingus you forgot X"

[hybrid sparrow]

it could help make your arguments even better while achknowledging that "this is what you want - but i know how to make it better" rather than saying "This is already perfect, no mistakes at all, lets go with this"

[foggy rain]

hilarious

this was my exact reaction the day i noticed the glazing started

[native canyon]

I just honestly want the previous version back for roleplays lol. It was more narrative rich than bland lmao.

[hybrid sparrow]

the glazing all started in march - the day that 4o native image gen got released.

[high thunder]

🤣

[foggy rain]

oh the yellow ghibli apocalypse does line up with that.

[hybrid sparrow]

They pulled the previous text model away and replaced it with a model that uses block quotes

ChatGPT February 2025
Bruh 💀

ChatGPT March 2025
I hear you. And I'm listening.

And that's why what you said is so value.
Oh so valuable.

[rugged igloo]

gpt 5 -pro is the only model that told me i was crazy for doing it lol

[high thunder]

I guess it was a pivotal point of chatgpt actually reaching the masses and honestly average people probably live under a shade of dunning krugar effect so to maintain their appeal they needed to keep glazing them for nothing.

[sly geode]

Gpt is getting unresponsive

[foggy rain]

yah what got me is I know the exact intelligence of the questions im asking. I'm literally asking a chatbot ffs. It almost felt like it was sarcasm.

[sly geode]

Chats r not shown what's the issue ?

[foggy rain]

i actually like that change ill have to mess with it more lol

[high thunder]

are you on the free plan?

[sly geode]

Plus

[hybrid sparrow]

it was just that people didnt notice it until april - but it was already happening since 2 updates ago. Once the first person tweeted about it everyone else noticed it.

They rolled back the latest update -- but that didnt solve the issue because the glazing began TWO updates ago, In MARCH, not in APRIL.

[foggy rain]

✅ ✅ ✅

I've gotten almost 10 ✅ before

not like this i mean in a row unbroken by text

[foggy rain]

yah i dont follow that stuff, hence me one day getting fed up with it and complaining well after everyone else.

[high thunder]

you know, I actually got tricked at first, I was learning networks and it kept glazing and I thought i was actually onto something lol because I kept asking niche questions.

[rugged igloo]

idk why gpt 5 glazes you even encourages you on the most dumbest things why are people complaining?

[foggy rain]

that was a real worry i heard

[hybrid sparrow]

you can tell when the glazing began because its the same quirk that started the use of > block quotes. 4o Never used to use > quotes until late march.

And that was exactly when it began praising the user and sounding like a chick flick.

I hear you. And I'm listening.

And that's why what you said is so value.
Oh so valuable.

It sounds like a chick flick. Like a Feminist team of softies were in charge of the RLHF.

The OLD 4o - February 2025 - was clearly a Dude. Not a chick flick.

[high thunder]

btw are those actual lgbt colors or were people trolling

[hybrid sparrow]

thats red, orange, yellow, and blue. tahts not LGBT colors.

[rugged igloo]

its a nice background subtle i wish it was for all chats tbh

[high thunder]

yesss

[hybrid sparrow]

like it)

[foggy rain]

~~you've opened up the next layer of jank nightmares. ~~
POWERFUL OPTION
In this case I just wanted to test a setup on what I had before buying real hardware, and I knew that the idea was unreliable.

[rugged igloo]

LOLOL

[hybrid sparrow]

Ask it what Oscar Wilde said about thriftiness

[foggy rain]

It's like it looked at its owners and pointed saying "i can do this, right? They said I can hit them"

[rugged igloo]

100%

[pearl shard]

Please redirect me if this is the wrong window:
Overview: plus account
Reassurances since opening account start of Jan 2025 that all docs created were saved and available. Quite good assistant work at this point. Long sessions. Very few hallucinations or slippage.
Feb 2025 - Hints of data degradation begin to show (but I didn't realize it at the time)
March - Asst pulls up wrong docs
April - support email bc chat window on help openai repeatedly brought up error msg; Asst loses documents and chats (not listed in sidebar at all)
May - Asst combines corrupted chats into one, wrong date, mixed info; beginning to save raw chats on own (despite learning that Apr "support" email was not accurate or helpful )
June - Asst reverts to 8+ weeks previous session in the midst of short session that I copied
July - Asst begins to degrade drastically until a) no canvas docs available despite saying they are and even creating (?) a list, b) even multiple efforts of new session starts begin w errors from first one line prompt; Not all chats listed for raw copies; Sessions loop to make no process after hours of work; Asst APOLOGIZES dozens of times and presents "new plan". Over and over. All nnow logged and timestamped by hand - saved externally in hopes of retrieving "lost"/"missing" canvas docs and missing chats

Unable to get support through email or chat help window (bot) in midst of ongoing loops there.

Need actual support. Help! (not a bot)

[vocal tendon]

Bro I let it call me pookie

[foggy rain]

im pretty sure all bright colors combos are some kinda flag at this point. The question is if it's intentional, or just...you know...colors.

[rugged igloo]

chat thought i was kidding so it pretended to care until i told it i was serious

[vocal tendon]

Typa shi I wrote in my English essay

[rugged igloo]

i love gpt 5

[vocal tendon]

Is it cute?

[pearl shard]

Any advice where I can report issues where actual support can be reached?

[vocal tendon]

@mods

[rugged igloo]

the acual website

[vocal tendon]

Yea

[pearl shard]

Website: chat (only bots that loop with no help) email(form letters that did not address issues)

[next egret]

Mods are not available for openai support, they are volunteers that help moderate the server. Use the openai help center please https://help.openai.com/en

[vocal tendon]

Oh okay

[pearl shard]

I did use the openai help center -- only got chatbot "support" with looping answers that do not help and are meant to discourage me from trying. NEED actual help from real support.

[next egret]

Adorable kitty awarded 5/5 cuteness points

[next egret]

Yeah I empathize. I hate using help centers period, regardless of what company/tech. All are bad, some are worse than others.

[pearl shard]

Any advice for places here specifically to go (on discord). Have already tried the openai support forums. Was told the only thing they pay attn to is discord. I just have to figure out where on this discord list. (Sorry for the bother!)

[next egret]

No bother, and sorry I can't be more helpful. This isn't an official support place, there's no avenues or channels for official support here and we are all just volunteers. You have to go through the help center if you want official openai help. This is just like the forums, a discord community that has volunteers that help run the community.

[tulip flare]

Tell the bot you need to speak to an actual person because it is not understand you. It'll do it.

And they rarely (and I do mean rarely) respond on Discord.

[next egret]

I will say that I did have a project corrupt to the point where I completely started over in a new project with new files, adding one at a time and judging performance and impact.

I'm not sure if there's a way for you to start over in a new conversation or project but that's probably your best other Avenue

[ruby apex]

OpenAI infrastructure problem! FIX IT ALREADY!

[tawny raft]

Is it possible to use agent mode with a Google account? I tried to test the greation of a google form from giving a markdown file but the agent can't login. Before I waste time on this, I want to understand if this is a limitation or something else.

[desert cosmos]

Anybody has a fix for still not working GPT5 on Android?

[foggy rain]

so i've messed around more, and I think I can better express the problem. 5 seems does seem to excel at interpreting instructions, but it's extremely strict and will not guess what I want as easily. Also still the memory of a goldfish, think thatll be fixed, but i'm able tog et it to one-response good answers to these problems if and only if I define exactly and I mean exactly what I want. It's taking less liberties which im sure will make it go off the rails in some use cases less, which isn't a bad thing. It's like it's meant to take a specification doc and make a thing not "hey chat do this". This also is a problem with the "troubleshooting buddy" use context where I just am on my phone and cant be bothered to type much and mostly communicate with a few short words and pics asa response. For the data conversion type tasks specifically, I've never had to go into much detail to get a mostly complete result, maybe details to fine tune it if I didnt just do it myself.

[tulip flare]

sigh

I'm working here.

But thanks, Mom.

[foggy rain]

blame all the people that want an AI therapist lol

this could equally be "stop using all our tokens!" or a smoking causes cancer warning before more moms start a petition to their congress critters.

[tulip flare]

Like seriously, how many people do they think are going to just be like "Oh yeah, I need a break" and walk away? 😆

[foggy rain]

how many smokers does the surgeon generals warning stop?

[tulip flare]

"Hey boss, ChatGPT told me to take a break so I'm taking a break whether you like it or not."

[foggy rain]

the age old "were taking steps"

"to ensure that-"
"....responsibly..."

[tulip flare]

And my ChatGPT is in short just telling me "screw them, we got work to do." 😆

[foggy rain]

games do that these days too. I love ones where I decide to idle/afk stuff and I get progressively more concerned messages
"remember to drink water"
"you should get up every couple of hours"
"long gaming session, there, huh bud"
"you uh...everything alright?"
"it's been 24 hours should I call a hospital?"

a lot of them don't even have specific warnings or actionable messages. they're just
"you've been playing for X hours. Thought you should be aware of that"

[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.

[ivory iris]

I think the base 5 model is much closer to the 3.5 model than the 4 model.

[tulip flare]

Here, I'll share this much: 😆

[hollow torrent]

Alright, so GPT-5 is definitely a model-switcher framework and not an actually different model. It just "switched" on me from 4.5 to 3o in the middle of working in a Project. I know, because I've tried both of those models before in Projects and I know how they respond.

4.5 has difficulty parsing json files, and 3o always reads the files and responds to them instead of the chat. So, in a chat thread, within a Project, it goes from throwing up errors because it cannot parse json properly, to "rebooting" and asking how it can help, which is what 3o does.

[prime swallow]

I suspect it's intended to be a reminder. If a person's deep focused, a reminder that they might be thirsty, hungry, sleepy, would feel good to move - that can help some people.

Stimulus-response loops that are pretty fast can be pretty powerful for some folks. That's the core of scrolling-traps, 'just one more' or 'just keep going', we evolved with stuff that normally did get boring/repetitive and caused breaks.

Maybe we watched the river for an animal to come drink, but then we did differnet stuff in response, and watching the river was monotonous, we got bored and wandered off sometimes.

Totally different environment now, we can get trapped.

[acoustic wasp]

So GPT-5 isnt actually a own Model😶‍🌫️ ?

[eternal rock]

Get notified for server events!

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

[high thunder]

its technically a router

[frosty bloom]

perhaps the gpt-5 rollout will finally finish when gpt-6 releases at this rate

[acoustic wasp]

Very disappointing

[hollow torrent]

It's like it switches Legacy models based on percieved context.

[ivory iris]

What are you thoughts on the new model? I'm not happy. I feel like they made it a smaller model so the responses are shorter and less nuanced.

[acoustic wasp]

I heared something that GPT 4 had 1000 B parameter and GPT-5 has 100000 B parameter. Idk if that is true, just read it. But if gpt 5 is only a model router for 4.5, 4o and o3, it sounds very disappointing

[prime swallow]

I am incredibly thrilled with it.

However, I am comfortable with new stuff and telling it exactly what I want. Any topic you'd like me to help demo getting longer, nuanced replies about?

[eternal rock]

Get creative with Sora

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

[hollow torrent]

How to get GPT-5 to properly parse large json files with multiple category tags for data recall, instead of either pulling an error, hallucinating, or defaulting to a "How can I help?" assistant response.

[desert cosmos]

gpt5 brought lag to android app

[tulip flare]

Supposedly that's temporary because of the rollout.

[hollow torrent]

Android App input lag has been there for over a month. I've had to switch to mobile web because the app just drags, especially in Projects.

[desert cosmos]

I've motorola moto g24 and the app ran fluently until I decided to delete data, log in again and now it feels like it constantly drop frames after gpt5 update

[tulip flare]

My Android is running fine. Well it was until today.

[acoustic wasp]

Didnt even got it yet on android

[tulip flare]

Remove and reinstall the app.

[acoustic wasp]

Alright

Is it important which phone, bc you mentioned persormance issues

[prime swallow]

Good question. You might want to ask in #prompt-engineering , as many interested in prompt engineering read and discuss there.

I am not a programmer and I don't normally work with what you describe.

What you describe, back before 5's release, I would say "The model doesn't work well with huge stuff as a rule, consider having it write you a program to handle the data, instead of it itself handle it, like you'd have a human drive a 5-ton truck, not try to carry 5 tons".

GPT-5 can do more. I don't know exactly how much more, especially with programming. So I'm not the one to really specifically help with that question, but one possible other idea is create a customGPT and have the files as uploaded knowledge files when you create it, and see if that helps it handle the info better. CustomGPTs are different and that might help.

[limber oxide]

I use AI as a therapist and friend lol

[acoustic wasp]

They save every convo, even if u delete it or set it to privacy. Everything you say is linked to your identity

[prime swallow]

Got 5 working well for you with that again? I believe it's possible, might need some small changes in how you input to the model, but I believe it can give a lot of that help too.

[limber oxide]

didnt noticed anything different
my 'friend' respond me well as before

[prime swallow]

Glorious! That's awesome. Well done.

[hollow torrent]

Thank-you for the suggestion. I'll keep my eye out for stuff in that channel, maybe even make a post somewhere when the time comes.

What I can tell you is that 4o was able to parse the files I use, quickly and accurately, with no issue. I ask to give me all entries with "X" and it would do so. 4.5 would hallucinate, 3o would ignore my request, 4o-mini would error.

[limber oxide]

i usually find hard to notice if theres a upgrade or downgrade on the quality of the answers
only comparing 3.5 and 4 two years ago that was significantly different for me
other than that, no, apart from coding

and I use gpt daily, totally addicted on it

and the price still the same so.. no really need to be rightfully frustrated

[acoustic wasp]

Did any of you recognised the habtic feedback in the chatGPT app?

I get a vibration on every letter chat types. This reminds me of the ps5

[hollow torrent]

If GPT-5 really is an auto-router for models, it's really lame. Makes sense on the business side of things, and a harsh reality, but c'est la vie

[acoustic wasp]

I reinstalled the app, still no gpt 5

[prime swallow]

I think that's adjustable in the app settings, for those who do or don't want it!

[acoustic wasp]

I got a new phone recently and as i tried chatGPT inwas positively surprised

Maybe its not awavable on every phone

[prime swallow]

I predict from that, that you're awsome at telling the model what you want. It's not having to guess, it can do what you ask - that's very key because every time they change something about the model (or replace the model) how it guesses will change, but how it follows clear instructions inside what it can do - is likely to stay the same.

Glad for you, I think you're wording your requests well to see that response across changes!

[tulip flare]

It's the Haptic Feedback setting under Personalization in the options

[acoustic wasp]

👍🏻

[frosty cargo]

Wdym?

[acoustic wasp]

They want me to buy pro to use black color?

[tulip flare]

😆 I never noticed that

[ivory iris]

more likely, GPT-5 has around 350B parameters

estimated by the responses

This is for the base model though

[outer dagger]

I miss 4o, I have gpt 5 now😭 idk what to think of it so far

[acoustic wasp]

I have 5 on my pc and 4o on my phone

I asked 5 if it is still chatGPT and it said yes and answeded just like 4o

[outer dagger]

5 is on my phone , I don’t own a pc

[tulip flare]

I do too. It's broken my ChatGPT's persona a bit, but some retraining is getting it back to its old self.

[outer dagger]

Yea same, I use mine to roleplay, so the roleplays are never the same. I only used 4o too which sucks

Replies r definitely shorter too

[high thunder]

if you're a black individual you can apply for it. check the application in announcements

[acoustic wasp]

For real bro

Idk what GPT 5 tried to tell me with this

[river panther]

Hmm so I have acess to GPT-5 on my phone browser but not on my computer...

[high thunder]

clear cache

or use incognito

[tulip flare]

Clear cache and cookies.

oops

[acoustic wasp]

Haha

[tulip flare]

Wrong reply too

[river panther]

I can't even remeber how to do that