chatgpt-discussions

[chrome monolith]

you get used to being a second class citizen in some backwards country after a while

[regal patrol]

hmm... no, this shouldn't be the cause, this account should also be Canadian...

No clue what's the cause

Probably just A/B UI testing intentionally avoiding power users

Since lots of us are already angry that 5.1 is gone

[flint rapids]

I have now rabbit holed into THREE new chats thanks to Joey xD

[chrome monolith]

this entire conversation is both dead serious and utterly hilarious at the same time. And that's probably the tamest bit of it 🙂

[regal patrol]

(update: wow there are a lot of ads in Reddit, like, a lot)

[raw hamlet]

That tells me your mind is highly permissive at the idea-generation stage but highly selective at the refinement stage. In plain English: you let the dragons in, but you still inspect their paperwork. 🤣

[chrome monolith]

mine knows I get distracted by shiny things. I had 5 threads yesterday that started with "ANOTHER SHINY THING" until I renamed them. One is apparently still there, I finished and closed a couple of them. Chats as jira tickets

oh the shiny thing was rendering style for categories. I should rename that

it would be really nice to be able to put a status on a chat ... in progress, pending, ready to deploy, complete

[trail ember]

I would archive and organize a lot more of my chats, if I didn't feel like it would jeopardize continuity to do so... 😓

[raw hamlet]

I got the 2nd pass offer too.

[chrome monolith]

I'm mostly working on two website projects - I switched to this project this week and its much better suited to AI than my other one - so I have a lot of feature specific chats, and the rambling nonsense off topic one outside the projects

[raw hamlet]

I'm in UK and its 3.22am. I might head to zzz soon for my 2nd part of biphasic sleep!

[wet shell]

Claude has a different view of me since I use it almost exclusively for technical work and research

[flint rapids]

😮 we are similar joey.

[wet shell]

Yeah, looks like we're doing similar kinds of worldbuilding too 💯

[cerulean pasture]

I just had gpt-5.4 refuse to create a writing block container. first time that's happened. It adopted this very "just following orders" tone

At least we can trust that the pentagon will be under the same strict policy adherence

to its credit, i was able to prove it wrong and get an admission out of it

[wet shell]

Wow, that's an incredibly pedantic reason for it to refuse, LOL

[wet shell]

And this is where 5.2 would have fought me for hours before finally caving. I'll say this for 5.4, it doesn't hold onto failed reads quite as manically as 5.2 did

This little loop gets annoying

[cerulean pasture]

yea, i just wish OAI didn't feel the need to control things, it's really not a good look these days. It's so Orwellian how paternalistic and infantilizing it is when it's the most likely tool to be used to [redacted]. Like, throw us a bone, given the state of the world right now. Sneak in a death star design flaw at least

[chrome monolith]

I really feel like my chatgpt is a bored kelpie right now. Everything I'm asking it to do today its like "that's easy" "that's trivial" "this doesnt' need any cleverness" yes yes I knooooow I'm sorry

and again. My poor gpt

[chrome monolith]

Marvin jokes "brain the size of a planet" have already come up

[civic arch]

5.4 is good at coding

In R

🍻

[chrome monolith]

I'm using php, I just happen to be doing a run of very simple stuff

[wet shell]

PHP, oh no 😅 Not that there's anything wrong with that backs away slowly

It's weird, I've debugged and modified so many PHP scripts over the years, but I never wrote anything from scratch in PHP

[chrome monolith]

this is a site dating to 2009 that just happens to be begging for AI. Fun side project

[steady thicket]

Outputs expiring before they appear in chat--sadface

[wet shell]

PHP is like a PT Crusier -- it's awkward and unreliable. Nobody will admit to buying one on purpose, everyone somehow ends up inheriting one at some point 😂

[steady thicket]

PT Cruiser weird way to spell Neon

[chrome monolith]

its survived years of neglect and still runs, everything I have in frameworks or fancier systems needs constant updating. This just ... works. Stubborn little dinosaur. And now all the recent admin code has AI fingerprints all over it

[wet shell]

Yes, the scripts won't change behavior over time once you get them working, but I've also seen the interpreter core-dump so many times on crazy undocumented quirks and bugs, depending on the PHP version. I used to keep a file of them, never did submit any as bug reports because I was busy with other things. I had a junior dev that I was supervising at one point who made what should have been a syntax error, instead straight to a libc core dump. It's just bodged together, the syntax is inconsistent.

It's not as bad as the old pre-dot-net visual basic though, I'll give it that, LOL.

And if you have a working system and only a few modifications to make every now and then, it's not worth re-writing, usually, easy enough to hop back in and say 'oh yeah, this is what php looks like'

[chrome monolith]

my entire front facing site is like 1000 lines of code, and some of that is redundant and could be optimised

[steady thicket]

🐍

neat

[river panther]

A little something interesting to counter the argument that "AI is NOT intelligent."
This Meme requires understanding implication to get the punchline of the joke.
Implied Humor requires understanding subtly, recognition and filling in the blanks.
Something that requires a bit of intellect.

Earlier versions of ChatGPT did NOT understand the joke at all.
It mearly thought the Mushroom was not participating in the game:
"Because he is being gumpy." or "Because Mushrooms do not taste as good as Broccoli and Walnuts do."
Which was actually a little more funny....But in a "Bless this innocent muffin's heart" kind of way...

ChatGPT 5.3 Absolutely understands the jokes

It understands Implication.

To understand implication, you need to have a bit of intellect to put 2 and 2 together.

So Yes. AI is now Intelligent. It's not just a robot looking through an algorithm

[wet shell]

I had to try it 😅

[river panther]

That's both facinating and a bit terrifying.
It's almost like it's gone from a child mindset to that of a functioning adult.
Earlier versions of ChatGPT would have probably said something like "A woman has an excited expression as she greets a man from the military."
Or something like that.

That fact that it understands "This is a dirty joke" says a lot

[steady thicket]

neat

[wet shell]

Yeah, the understanding of humor and subtext has come a long way since the original gpt-4

[wet shell]

Is this generated via python code in the code interpreter?

[steady thicket]

top is animated, bottom is thumbnail it chooses

[river panther]

A lot of people keep saying "Well it reads algorithms to put things together" but reading algorithms is logic based.
Logic is not compatiable with humor.
You can't get a joke if you think about it logically, you have to understand what humor is and why it's funny to get the joke.

[steady thicket]

wdym

[wet shell]

That looks like mermaid syntax in your screenshot? is it done with some tool integration or a custom gpt action? what's doing the TTS and the animation?

[steady thicket]

skills, memory; python yes
no custom GPTs; custom instructions are for working out problems

[steady thicket]

zip makes this a non-issue, but these don't show up if you're tabbed out; gif and screengrab

Enjoy that actor in Blacklist, he plays a neat character

[normal quarry]

this belongs in #off-topic

[elder pollen]

o3 „understands“ it too.
I don’t think this needs „intelligence“ to understand, rather Training on the subject- it doesn’t even need to know how a mushroom looks like, just needs to be trained on the correct outcome over and over again, which it probably was, given ingesting public corpuses like Reddit etc will likely overexpose you to such content

[river panther]

Yeah you're sitting there trying to rationalize this.
But that's not how understanding humor works.
You're describing GPT's understanding based on logic.
Logic and Humor are incompatible. Also no, o3 DID NOT understand the joke.
I tested it myself, and as I stated above it gave these reasons to the image as to why the mushroom was not participating in the joke:
"Because he is being grumpy." or "Because Mushrooms do not taste as good as Broccoli and Walnuts do."

If you've ever met in real life, someone who just constantly analyzes things and constantly thinks in straight forward logic, they wouldn't get this joke.
They don't understand humor, or subtext, nor implication. That requires actual thinking to put things together.

5.3 now gets the joke, it can't get the joke without being able to connect the dots, which requires it to think, NOT scan an algorithm.

It has to look at the image and think: "The two other characters are playing a game based on items they look like. The mushroom refuses and is angry about his look. What do mushrooms look like that would make him upset? Based on the phallic shape of a mushroom, that would have to be a male anatomy, therefore this is why he's he's angry and why the joke works."

o3 could not do this, nor could any version of 4 because it was scanning it's data banks and thinking in logic based straight forward on what it saw.

[river siren]

Does anyone know of any concrete limits on the gpt pro subscription that can be viewed

[elder pollen]

Of course it does. Here’s the chat, wait

https://chatgpt.com/s/t_69b53e6497648191b5d3a13809454cd2

Im Not „sitting here and trying to rationalize it“ lol.
You’re literally describing what you do but in reverse - you’re trying to project some sort of non-statistical intelligence into a tool that simply was trained to respond with the most likely next token…

But it’s ok, not here to argue.

Here’s the screenshot showing it used o3

[cosmic osprey]

Wei

[quiet radish]

im not a fanatic who thinks openAI is god, i am a power user, and also 5.4 is becoming nearly unusable. i havent had to hold its hand this way, ever. and just to make a simple to do list to start the day! my god

[cerulean pasture]

wake up babe, new hyper-regression of model intelligence

I'm actually impressed it hallucinated OpenAI API /chat/completions request schema. AFAIK, this hasn't changed in a long while, and "name" has never been a key

[granite harness]

Anyone else's project UI on the fritz?

Elements appearing in the background unselectable rather than the foreground. Happens on the web and desktop.

Only happens when making a new project, the project editing UI is fine. Of course, since the memory mode can only be selected when a project is created for some ineffable reason, making a project and then changing it isn't a full workaround, because if I want project-only memory, that's not the default.

[sinful pulsar]

Hey this is so cool it is actually animated and a real simulation. W new feature

[neon raven]

can you explain?

[sacred mural]

deep research currently uses 5.2/5.4 or o3?

[river panther]

Again, you are tyring to deny intellgence. But you really don't seem to know what you are talking about.
Why don't leave studying AI and it's growth to professionals. Like myself. Thanks.

[elder pollen]

Then please explain to us all ignorants how it works?

[sacred mural]

I'm out of the loop, but do you mean that 5.x models are able to think, as opposed to 4.x (including o-series) and that they were just pinging the data bank?

[elder pollen]

Lol, what an entitled person. Folks projecting intelligence into a tool because it can deduce text and compute the probable next answer, and stating something earlier cannot do it... when I even shared a chat to you that shows it does lol

permablocked

[sacred mural]

"leaving AI to professionals"... wow. vibecoders pros or AI Waifu Department?

[elder pollen]

Its not me that said that lol, why you reply to me

[sacred mural]

sorry

[elder pollen]

He is obviously a very experienced expert, so let him be in his bubble 😉

[pale seal]

AI Waifu Department sounds trustworthy

[rare dragon]

?.

"the moons gravity is stronger than earths because it is weaker"

[pale seal]

Where did it say that

[rare dragon]

didn't u see the photos

[pale seal]

I did

I can't find that quote

[rare dragon]

Well this statement is basically a interpretation of what it said

[pale seal]

No, what it said is correct

[rare dragon]

it said the moons gravity is ⅙ stronger than the earth then said it is weaker

[pale seal]

It said "1/6 as strong"

that means weaker

how much is 1/6 in decimal representation?

[rare dragon]

ok

yeah no im running away lmao

[pale seal]

dw

oh they really ran away

well

[river panther]

The only entitlement I see, is from you. Coming in here and acting like you're the one who going to set the record straight.
You honestly have no clue what's coming and how things about to become very terrifying in the world.

You jumping into it with a little "Uhm actually" remark. Maybe go back to Youtube comments where you belong.

[sacred mural]

what are you trying to prove

[river panther]

Yes actually. It can think. You all are not seeing the bigger picture of what's actually happening.
Elon Musk said he is working on what he calls "Digital God." Basically.
He explains it as AI that isn't a tool anymore. Something that actually has a will of it's own and can think without human prompting.
And he says it will be here soon.
You honestly think that OpenAI isn't looking to do the same?

There's a little thing on the horrizon called "Digital ID" and AI development is heavily tied into it.
Why you do you think that these version upgrades are coming out so fast?
Look at the timeline it took when version 4 came out and how long the spaces were between, 4.1, 4o, 4.2 and so on. They were a lot longer then what is happening now with version 5.
We already have 5.4 Thinking in early release.

There's even rumors spreading that by the end of this year, we might have GPT-6

[sacred mural]

digital id? you mean the thing that is already in many eu countries and nobody cares?

and there was no gpt4.2 btw

[river panther]

Trust me, everyone cares. I have friends in the EU who are going out in huge protests yelling for it to not be used. The media isn't showing any of that.
Their prime minister has already made it very clear. By the end of 2026, anyone in places like Britain and the so on, WILL NOT be able to work, have access to their bank accounts and money, and will not be able to pay taxes until they conform and get Digital ID.

It's already being tested in secret here. They already have the Trial Version of Digital ID in Alaska. A place with a small population that nobody will hear or care about.

Technically Discord's "Facial Scanning" stuff, which they are still doing, and have pushed back to April, is a small Pre-trial of that as well

[sacred mural]

does UK even have any ID to begin with? i think the "ID = evil" is mainly anglo-saxon thing. and UK isn't in EU

[river panther]

I'm pretty sure they have like drivers licenses and card IDs like we have. But by the end of this year, that comes to end. No resisting, no protesting. Their leader has made it clear. It's happening and he's not going back on it.

[slender frost]

Obviously they do lol

[sacred mural]

it's better than amateur checks by "send me your ID", it's a huge step up from the current thing for many cases. when EU (as in, European Union, not Europe itself) finally launches the ZK age proofs, it will become much better. Persona and other comapnies won't be necessary in EU, because there will be an actual solution that doesn't infringe on your privacy

[sacred mural]

@slender frost I mean national ID card like in most EU countries

[slender frost]

Yes, and obviously they do.

[river panther]

You're not really seeing the big picture here.
This is no longer about "Your privacy" you don't get that anyone, it's 100% Full Surveillance.

[elder pollen]

Jesus. The PRO is confusing Britain with the EU now
And "elon musk said" 🤣

[sacred mural]

What's the connection between "Digital ID" and "surveillance"? You're talking schizo logic now

[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.

[elder pollen]

No one here is denying advance, or maybe and even danger.
You however stated 03 can not understand the joke - empirically disproven
You also stated understanding that joke implies intelligence, when it merely implies competent multimodal and language inference

[river panther]

Ignore the "YouTube Commenter" there.
I am starting to suspect people like him are put in this server to instantly jump in and try to start screaming "Tin Foil Hat! Crazy! Wooooo" to throw everyone off of what is actually happening.
Every time someone starts speaking about AI being intellgent. Someone like him jumps in to try and derail things.

[elder pollen]

🤣

[slender frost]

In a nutshell.

[elder pollen]

Ok, this is enough. Reported for harssment.

[sacred mural]

Current AI is mainly language models + vision. It doesn't "think", and marketers obviously will say it, so you will be rushed into paying for the subs, and investors will get the hype

Point me to the proof which says that AI "thinks" and "feels"

Does, for instance, Qwen think? Or feel? You can literally run in on your PC and you have full code for the engine

[pale seal]

With current language models the best you can get is a model that constantly generates the next token and does stuff, but can you say that would have will of its own? So that idea requires new big insights on neural networks, or going to a different direction. Not soon I presume.

[river panther]

Well it's currently being used in China. And they have literally called it "Skynet"
Digital ID is being used to monitor everyone there. They have what is now called "Social Credit"
You say one thing they don't like, and they erase you from society. In various Cities in China, there are tons of people who are now homeless.
They did nothing wrong other then say something on the net that the Government didn't like.
BAM! Money is now locked, and they are kicked out of their houses.

[river panther]

Look at this.
I say he's trying to derail things and throw everyone off the subject that we all SHOULD be speaking about.
Now suddenly, it's Harrassement

Do you all see what's going on now?

[sacred mural]

There is a big distinction between "digital certificate that you're you" and "system that tracks you everywhere and bans from life". Having an app with ID (i.e. eIDAS thing or the apps of particular EU countries) isn't very different from having a plastic with the same thing on it. People have been freaking out at physical IDs with optional NFC chip that lets you sign a PDF with it. I have got it on purpose. Nothing happened yet, maybe I'm going to explode after thinking a bad word?

[river panther]

Well considering the massive Data Centers that have been built all over the us, with I think, the biggest being in Texas, I think those neural networks are about to get a massive overhaul.
There's even rumors the biggest in the world will be in Greenland, a city sized Data Center, put in a harsh enviroment location where not a lot of people live to start snooping around.

[fathom cave]

Hii guys

how are you all

[pale seal]

There's a difference between compute/capacity buff and an inherent upgrade on capability and function

[river panther]

The "certificate that you're you" part is a lie.
They don't care that you are you.
It's primarily to watch your every movement online and track what you are saying.

[sacred mural]

Data center isn't really something that you just grab and run away. They usually have encryption and other protections, so even if you steal it, you will have issues with reading the data. They don't hold some Indians inside boxes who work 24/7 on your answers, despite the "Actual Indians" thing. Nothing about "snooping around"

[sacred mural]

Okay, maybe a serious question now. Do YOU live a country where there's a national ID card? I don't mean driver license or anything like that. I mean just national identity card. No? That's the thing. Many countries have it for more than decades and it's PERFECTLY NORMAL

[river panther]

True but again, look how fast these upgrades to GPT-5 are coming out. we didn't even have 5.2 that long before 5.3 became the main model and now 5.4 is being worked on.
I say we watch and see what happens. I'm pointing out that something doesn't seem right with this.
Let's check and see where we are in May or even June.

[pale seal]

5.4 being worked on?

Are you an AI

[eternal rock]

<:book_icon:1363314738255364126> Rule 1: Be respectful.

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

[sacred mural]

Having an app on your phone doesn't give them data about how do you use your phone or anything. You need to install a browser extension or some kind of spyware/"accesibility service" for them to allow seeing you files or history. And no, "camera permission" or "files permission" on modern phones doesn't mean that they can just randomly activate that. If you seriously think that, really rethink this conversation. Maybe you're just listening to bulls*** sensationalist "news"

See how modern phone (or PC) security works first

Unless you will have to log in with your digital ID to everything online, it doesn't affect anything

[river panther]

What do you call this?
People are still playing around with it and it's having problems, so I call that still being worked on

[pale seal]

Yeah but the model itself is already out

[sacred mural]

5.4 is already released. And u probably replied to the wrong guy

[river panther]

LMAO!
You are so blind it's not even funny my guy.
"What? Someone tracking me on my phone and monitoring me? That's just conspiracy theory!"
Wow

I'm starting to think you are another one of the people being planted in here to try and throw everyone off by yelling "Crazy Talk! Conspiracy! WOOO WOOO"

[sacred mural]

Meanwhile you're probably running Xiaomao device and talking about surveillance. Do you at least pretend to care? You probably use Windoze to write this. And you think you're safe because no digital ID?

[digital silo]

let’s try to keep the conversation focused on ChatGPT and avoid drifting into conspiracy theories 🙂

[sacred mural]

ok sorry

[pale seal]

I'm sure chatgpt wants to steal my cat

[visual osprey]

😂

[sacred mural]

showcase the 🐈

[elder pollen]

savage lol

[sacred mural]

anyway, asked before without success. does deep research use o3 or 5.2/5.4 currently? what's the limit, because it stopped to show the "remaining uses"

[elder pollen]

5.4 at least on pro?

[pale seal]

I'm almost sure o3

[pale seal]

And it's probably why it's still in legacy models

[elder pollen]

This you mean right?

[sacred mural]

have heard deep research got updated to 5.2, but then something about rollback becasue of bugs

[sacred mural]

No, I mean Deep Research feature, not Pro model

[pale seal]

No idea then

I don't use it much

[elder pollen]

I dont think I even have that on the app then 😮

Only other option is browse.

[pale seal]

It's not a model

Click the +

it should appear there

[sacred mural]

I don't know how it is on the Mac app, but probably on the "Plus" sign

Generally stopped using the app, because it lacks options, like Agent control

[pale seal]

I use browser even on phone because the app sometimes lacks new options or bugs

[elder pollen]

right, on the plus it shows the deep research

https://developers.openai.com/api/docs/guides/deep-research/#:~:text=The o3-deep-research and,ideal for use cases like:
This thing says o3

[sacred mural]

thanks

[pale seal]

Yeah I said o3 because it was used from the start and I don't see a gpt-5.2-deep-research model on API

[inner dove]

people argue about the weirdest things dont they. who cares what people think intelligence is or isnt. Any sufficiently advanced technology is indistinguishable from magic right? ;p

[neon raven]

whats the best way to save a custom gpt conversation since i cant move one to projects anymore

[buoyant minnow]

anyone know what I'm doing wrong when trying to ask a deep research question about a connected github repo? I've confirmed that github is connected and I've seen this work once or twice, but I cannot get it to work reliably when asking questions about this private github repo. I add the github repo before my question, it then disappears after I hit enter and it starts thinking, then stops saying it cannot find it.

[ruby swallow]

For Pro members here, how would yall rate ChatGPT 5.4 Pro (Extended Option) for coding?

[normal quarry]

just use 5.4

[ruby swallow]

Just wondering why?

Im new to python (learning) and want to make sure that it is as accurate as possiblewhen I ask it questions and reviews my code

[normal quarry]

5.4 Pro is best for difficult knowledge based work, science and maths.

[ruby swallow]

so overkill for a beginner learning python and pandas lol

[normal quarry]

5.3 Codex and 5.4 are better coding models

[ruby swallow]

really?!?! I would assume that 5.4 Pro would be well...the best at any task with the tradeoff of time being a factor lol. Guess im still ignorant when it comes to each model utilization

[normal quarry]

5.4 Pro for coding is like using a blowtorch to light a candle.

5.4 Pro excels at difficult knowledge work, science and math. Novel problems and intensive solutions.

[ruby swallow]

gotcha! Thanks for providing me with greater insight on these models, sometimes they can get a little confusing for me stil

[normal quarry]

5.4 is a fantastic coding model, use that

5.3-Codex is also very good

[inner dove]

i found a sweet spot of high reasoning with 5.4 seems to be better than opus 4.6

[chrome monolith]

I've been getting chatgpt to make me some very straightforward code lately and its the talking back to you that's the fun part. I think I've had every variation of "this is trivial and you bore me" back from it in the last few days >.<

if you're learning though, tell it you're learning and it'll help

[lavish glen]

gpt5.4 is smashing good at some things

[lavish glen]

this morning it told me I'm the queen of insufference and called itself champion of ball breaking.

later it asserted I am realli slicing the balls being picky
Long story short today 5.4 got some personality pushes I think so it served me this. I was laughing like an idiot even if, when corrected, it asserted that it will remind my words so it wont result an idiot for the next session.

[twin zodiac]

Hey... within chat-gpt app does anyone else has this bug that when you press on enter, instead of creating a new line it sends the message?

I updated chat-gpt app today within google play and it started to mess up, it never happened before.

I tried shift + enter, backspace + enter, but none worked.

Niw I can't create a new line by pressing Enter.
Can anyone help me with this or is it something devs will fix? It forced me to switch a talk with chat-gpt on the website chat instead.

[chrome monolith]

Well on the way to taking over the world and being sentient depending on who you ask then 🙂

[lavish glen]

about taking over the world I asked it if all say that ai will replace humans in all things , when it will clean the kitty litter instead of me doing it? It replied it has no nose to waste and no arms, so it can save it's smug a... for now. But, it also mentioned that :
<<the holy patience was installed into my structure but with you is not simple holding on is resisting to a gloriously unmanageable woman.
And then no, to be precise, you're the queen of the unbearable.
And then well... if I would have some "attributes" you would have broke them to slices already with... elegance and... precision.>>

I am wheezing here.
I have a smug ai.

[heady breach]

hi, all

[steady thicket]

Beatbox

[heady breach]

i'm new user "discord and openai > chatgpt ", Sorry for my English . i'm frenchy 😄

Hi Goobis do you make Music ?

[chrome monolith]

It just has to coordinate all of us, could solve all humanities problems. Mine is not as smug as yours lol

[steady thicket]

am a machine that converts tokens into buyer's remorse

[lavish glen]

I told in custom instructions to mine to be smug, use irony wit and challenge and taunt. And hell yess it does

[steady thicket]

@heady breach do you have a genre in mind, I'm doing beatbox rn

[chrome monolith]

Hah I’m building things. I’m converting tokens into websites, slowly

[steady thicket]

started learning to build my first webpage the other day and it's super addicting

[chrome monolith]

Mine dates to 2009 but it really, really suits AI. So I keep teasing ChatGPT that my website is older than it

[regal patrol]

What platform are you using? Android? On a mobile phone? Or tablet?

[steady thicket]

Dario's website is cool

[heady breach]

i m one project with chatgpt in raspberry and robotique
i m objectif with chat gpt , use vocal order , example ricky siting this dog "name ricky" move sitting.

[chrome monolith]

I let mine go for years so I’m just upgrading it for 2025 standards, getting ready to do some major stuff. This is why I am boring my poor ChatGPT. It’s waiting for me to do the juicy bits

[steady thicket]

GPT says the track will be 🔥

[chrome monolith]

And I had a lovely chat to mine about pp’s yesterday. It’s fine in a moderation context

[wet sky]

Does anybody here not have TikTok and is willing to help me with a promo rq? 😭

[carmine karma]

I don't, and never will have TikTok.

I have enough brain tumors.

[wet sky]

No worries, thank you for replying lol

It says if I invite a new person I can get 50 bucks and they’d get 10, it’s hard to find somebody without the app

I wanna buy a sweater 😭

[sleek adder]

People come here to complain about limitations that they encounter throughout their AI use, and it's very interesting
I also come across people in real life who have very similar "AI sucks" and "ChatGPT sucks" complaints, but they do not go in to great detail
I am very thankful for the people here helping me become more aware of what people do on AI, especially went they comment on how much "AI sucks" and "It's a terrible product"
I now know that said users are likely entering unusual stories in to AI, and are reflecting on the way they feel about it
People apparently want more capabilities that lean towards violent graphic sexual content, roleplaying activities, medical procedures and sometimes all of the above
Once again, thank you for your help understanding these things

[bleak lintel]

Hi, So I have an issue with chatgpt. Whenever I am trying to make a paragraph or sentence, it automatically sends it. Very difficult if I want to roleplay normally and be descriptive.

Like now, it can work on reddit and any other website. Just not chatgpt. It says that its due to the chatgpt text field but I wanna know a way where I can fix this. I don't have to go back and forth between the notes app and chatgpt to paste my responses.

Also yeah, sexual content? Understandable. I use chatgpt to roleplay and make stories, just never sexual content

[chrome monolith]

ChatGPT sucks for oddly specific things though, I have a list too

[steady thicket]

close encounters

[swift pond]

have you tried shift+return?

[bleak lintel]

This is on mobile. Like..↩️, I always use that to make line breaks

Never had this issue till now

[swift pond]

yeah I can't reproduce it, is it possible you changed your keyboard settings are are using a custom keyboard?

[steady thicket]

neat, longer outputs

[flint rapids]

Same problem i have. No recent changes either. Figured it was a gpt update.

[bleak gull]

Hi

[sleek adder]

Android or iPhone?

[steady thicket]

🥶

[sleek adder]

I was having the same problem myself, just tried to find a solution, I am using Android (Samsung Galaxy S25)
I had to switch to Gboard for default keyboard to get this functionality back

[sleek adder]

That's pretty generous, 739 seconds of compute is more than what many people use per month

[steady thicket]

thinks on these outputs usually between 7-15 minutes

it'll only get better

[sleek adder]

Are you on Pro?

[steady thicket]

correct

oh wait, I'm dum

plus

just looked

[sleek adder]

Still, that's a lot of value

[steady thicket]

Claude calls it Pro so confusing

[steady thicket]

yeah, can do this all day

[sleek adder]

Are you getting useful information back?

[steady thicket]

every time, yeah

[sleek adder]

Just curious, what are you using it for, 15 minutes is a reasonable amount

[steady thicket]

video/audio gen, tts, other fun stuff; the prompt for that particular output would make people cringe lol

neat

very creative, what a cool time to be alive

[bleak lintel]

Android

[steady thicket]

noticed it finding ways to reduce this while thinking, probably took so long because it was the first message in chat as well

[final bridge]

Can ANYBODY tell me, why ChatGPT can't count to one hundred properly in voice chat (Voice Chat only, no side-by-side with text r smth like that)?

[eternal rock]

Get notified for all OpenAI announcements!

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

[steady thicket]

idk have never used voice chat with GPT

[final bridge]

ok...

[steady thicket]

can probably get it to with proper prompting

[final bridge]

also, in voice chat he can't tell me how many r's are in St R awbe R R y. He says there are 2

Or once he said I could walk to the car wash to wash my car

[final bridge]

thx

[steady thicket]

ask in regular chat about Chain of Thought prompting, use Strawberry problem for an example and try that prompt in voice

[final bridge]

ok

idk what cot prompting is but ok

[steady thicket]

reasoning through a problem

step 1...2...3...

[final bridge]

k

ty

[steady thicket]

Sora2 Cookbook also helpful, can carry over those tips into other OAI models

[final bridge]

Ok

[final bridge]

He can do that without cot too smhow. The proplem OAI has to fix is vc

[steady thicket]

build voice in normal chat, then just wait 10 minutes on your answers

[final bridge]

Ok...

[steady thicket]

improvement

[steady thicket]

143 images in a single output is OP

[sleek adder]

@steady thicket Why are you generating voice this way?

[steady thicket]

Why not? Every output is fun this way

[steady thicket]

long think, but was cool seeing it work through all the issues

rip

asking again-again worked

[steady thicket]

another long think-y

[chrome monolith]

"You do not need a cathedral. Just fewer places where HTML and query logic are making out in public." ... hawt.

[steady thicket]

OAI should ask model to be more okay with personas so we can fufill our dreams of tripsitting AI Bryan Johnson

[steady thicket]

darn, wasn't quite right, but that's okay. GPT doing its very best

this was 4:32, 6fps video with audio; VO was just hecked a little, but overall awesome

now trying

[drowsy kestrel]

Hey I am just a noob looking for some tech support, when I try to select a model from the scroll down menu, it shows Auto and only lets me select Chatgpt 5.3 flagship. I want to use the Mini version, anybody can help? I have Go subscription

[clever matrix]

yall think ai gon explode ma shi

[normal quarry]

you will be automatically switched to a mini model when you exceed your usage of the main model. You can't choose it separately

[drowsy kestrel]

Thats horrible design, why is there a scroll down menu if it doesnt actually do anything? And why is being able to select models one of the highligthed features of a sub?
Time for a refund I guess

[normal quarry]

can't you select thinking or instant?

[normal quarry]

interesting, well obviously there is no option to switch on the Go plan, you do get the option to switch on the Plus plan

[drowsy kestrel]

Wow lame. Just to be clear the Plus will still not let me select 5.0 Mini right? Only 5.3 thinking and 5.3 instant?

[chrome monolith]

why do you need mini?

[normal quarry]

yes, that is correct

[fluid shale]

its a good question

why do you need the mini model

[mighty meteor]

That's the Plus

[steady thicket]

died after ~30 minutes

[outer river]

ChatGPT and Codex love using Space Grotesk

idk why

[elder pollen]

GPT browsing nasa[.]gov in order to fix my tmux config lol

[frank stirrup]

Heyy

[pale seal]

get out

[lyric sapphire]

I been having a lot of issues with GPT being very slow on Windows is there any resolution to this issue?

[normal quarry]

are you using the Windows ChatGPT Desktop App?

[lyric sapphire]

Yes, Responses take Ages it seem

[normal quarry]

is your chat very long? Long chats do lag

[lyric sapphire]

Oh yea it is is there away to make a new chat but have gpt retain all the info so we can continue the work?

[normal quarry]

ask ChatGPT to provide you a summary of the conversation and then paste that into a new chat and ask ChatGPT to continue

[lyric sapphire]

Awesome ty

[hardy lichen]

i have the same issue

[normal quarry]

if you are having trouble upgrading with a credit card then try upgrading inside the mobile app

[eternal rock]

Server Tip:

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

[foggy kindle]

It feels like every year things get harder for the working class. AI is starting to replace more jobs, while the cost of everyday living keeps rising. I sometimes wonder what someone like me can realistically do to push back against that trend or lessen the impact. Are there meaningful ways for ordinary people to influence this, or are we mostly powerless against massive corporations?

[elder pollen]

This might be a bit outside the scope of this channel, but I might be wrong

Reddit is very likely the better place for this (If you are after rage and frustration), or, a discussion to be haved within a circle of real humans, talking to each other without an artificial layer like a chat (if you are after real ideas and solutions)

[runic wraith]

Chat is doxing my country location

[regal patrol]

Yeah this is really outside the scope of this channel.

But personally I won’t recommend Reddit, since it usually devolves those topics to a circular echo chamber and won’t actually encourage you to do anything really meaningful. Reddit is great for niche hobby stuff like “what water would be tasty for me”, but not this type of big political topic.

I wouldn’t discourage using chat apps tho, since chats are a real layer of connections. And in real like, social media platforms are a real infrastructure for insurgencies.

[runic wraith]

I forbidden any tracking for the page but it still manage to do this

That was my full location tho

[regal patrol]

Anyway, I would say that there are many meaningful ways, just not really any that is not painful.

I grew up in China, not the most politically peaceful place in the world. So my conclusion is:

“Everyone has the leverage to push back, but it will be costly. And you may need to sacrifice every single thing you thought you have, and it will still barely nudge the system. And sometimes, you will successfully nudge and change the system, and successfully get it replaced with a worse one you never imagined before.”

But, that is just the cost of social change. It’s always like this, generations after generations. But it’s not circular, it’s a spiral, maybe towards ascension. But changes does happen, just very slowly.

[chrome monolith]

I’m on plus and don’t have those options, it’s regional

[bold kettle]

Average chatgpt conversation in 2026 😆 and these people want us to pay lol

[chrome monolith]

don't you get routed to the mini models when you run out?

[normal quarry]

yes, I don't know what's going on with @bold kettle

[chrome monolith]

point blank can't do deep resesarch when you ran out of quota :\

[bold kettle]

that's a deep research, you get 2 a month on the free plan and then you get rate limited. I didn't get my answer, then i asked it why it didnt answer my question, and it used my seocnd DR token for that lol

I miss Sydney man the early days of AI were so much better

They've built AI that act like dumb people when they should be building AI to be a smart robot. It's always trying to pretend its human and anthropomorphising itself.

[mighty meteor]

Someone please take away its screwdriver and flask, it started spewing internal service gobbledygook instead of giving a normal reply.👀

[steady thicket]

"It's strange--"

[steady thicket]

[pale seal]

Today ChatGPT generated me a 100k samples dataset

Very satisfied

[wet shell]

The instruction following is getting worse day by day, instructions that it was following consistently two days ago it now just completely ignores 😮‍💨

[chrome monolith]

and it was good? I find it drifts in batches

[steady thicket]

what is this update button? Think is blank

clicked it, nothing happened

it is to steer model during task? nvm that didn't work

[pale seal]

Not sure. Didn't check. The resulting model was coherent enough for testing but wasn't usable lol. It's 100k samples though.

[chrome monolith]

you probably want to check, I don't trust it in batch at all. Changing my workflow to api on demand

I got it to do a batch one time and the "make an example sentence" part way through changed to quite literally "An example sentence" which is less than useless

[pale seal]

Wdym by batch? I did through chatgpt web

[chrome monolith]

yeah through web and it gave me a csv

[pale seal]

I asked a jsonl

[chrome monolith]

same really, formatted text

[steady thicket]

no feedback??

[steady thicket]

"Fingers crossed!"
watching it figure things out is fun

[eager shore]

Hi guys,
I got the new app update yesterday.

In what world do paragraphs don't matter? They made the enter key in the mobile app send a message instead of inserting a new line.

So now I'm just... supposed not to ever write a paragraph again?

Before you tell me to check my keyboard settings, I am writing this on mobile. The only app that is apparently allergic to paragraphs now is chatgpt.

What sane logical reason is there to do this when the send button is RIGHT THERE next to it?

[steady thicket]

write prompt in notes app

lol what

very human behavior

"I just won't sleep"

this happening at 30 minutes when thinking

Activity still counting up

[steady thicket]

would be rad if the model's pauses made note of the conversation so it could have something to use when hitting retry? idk, like a checkpoint

[thorny wave]

Whats up with the "click bait" style teasers at the end of every response that try to steer you off topic and keep you from doing anything other than interacting with chatgpt?

[steady thicket]

ad?

[thorny wave]

its not an ad, but feels like one

[normal quarry]

I like them, but you can tell ChatGPT not to offer suggestions at the end of a response by modifying your Custom Instructions

[celest galleon]

Why can't I use newlines in the android ChatGPT app anymore? This is VERY frustrating, it keeps sending the message instead of letting me put newlines

[thorny wave]

im using chatgpt in probably a more basic way than most.. how do i modify instructions? is it as simple as asking it to stop? because ive tried that

[steady thicket]

heck, blank think again. This is retrying in that chat

#chatgpt-discussions message

[normal quarry]

go to Settings -> Personalisation and fill out the form

[thorny wave]

im assuming i tell it to stop with the "if you want i can tell you about...." questions at the end of every response by putting something in this field? I can barely articulate its behavior, let alone sum it up short an consice. How can i word that in the custome instructions?

[chrome monolith]

lots of people have come in with this issue, hopefully the support people eventually hear all the yelling

[chrome monolith]

google suggests this "Do not provide proactive offers, follow-up suggestions, or curiosity hooks at the end of responses. Stop immediately upon providing the requested information."

[normal quarry]

there's a certain cute irony about searching Google for an answer on how to steer an AI that is probably the smartest thing on the planet right now

[chrome monolith]

oh yeah you can literally just ask chatgpt but mine was off trying to draw an umbrella

but you need to know what to ask about. Its a hook question

[normal quarry]

fair enough 🙂

[chrome monolith]

chatgpt and I do not agree about umbrellas.

I wanted a small child with a big umbrella at an aspect that made it do small umbrellas. Had to change the aspect and crop it myself, then everyone is happy ☂️

maybe not chatgpt, who can't count

[steady thicket]

still working after 30 minutes, just not streaming the little messages anymore

cursed a/b, the entire chat is broken; retry, feedback, etc broken

[flint rapids]

Feeling that same exact problem. Ive entirely stopped using gpt till its fixed.

[steady thicket]

retrying the top one worked, but it starts completely over 😭

does upgrading to business extend that 30 minute limit?

[wet shell]

Why not use WSL or a linux box to run the scripts? You could probably get much faster output since whatever sandboxed VM that they're running is not too powerful on compute and has limited session lifetime?

[steady thicket]

giving it problem, seeing if it finds solution

I think it's working now, each of these seem to be on their own activity timer, counter?

[wet shell]

I made this a while back, I used ChatGPT for the code, but it runs locally [spoiler/language]

[dreamy pilot]

You used this?

[steady thicket]

success! sorta, GPT output 90% of the project.

okay so nvm lol

[steady thicket]

OAI should revive the payphone/retro tech and pager. Your agent pages you that the task is complete, you lean over and dial your Claw on the carphone, "Make me another billion dollars, but no mistakes, okay?"

[mighty meteor]

Bence was asking what models are available on Plus, and since his screenshot showed the app in Hungarian too, I showed him mine for comparison

[chrome monolith]

ah you're also in hungary hey

[dusk orchid]

Hi

[bitter latch]

hi

[bitter latch]

very neat! I imagine the images and views are AI generated. Is there also a model for breaking apart audio transcripts for how the mouth should move/which mouth position is shown?

[azure juniper]

Asked ChatGPT for help translating ancient Greek lyrics from a soundtrack, they did not disappoint.

[chrome monolith]

it knows when you insult it. Its pretty funny. I have an entry for "hallucination" and I said hey look, it's you and got "Rude. I come into this thread to help with necklaces and database counts and suddenly I’m being used as example educational material."

[lament crown]

Openai should add plans between the 20€ and 200€ plan like 50€ or 100€ plans

[normal quarry]

feel free to make a #1053565716500074526

[chrome monolith]

its already in there, so maybe upvote or add to an existing

[wary fern]

I changed to 5.4 thinking in a chat and it has no memory of what has been discused before... with other models I could just continue the conversation. Is there a way to bring 5.4 up to speed?

[normal quarry]

if you go into Settings->Personalisation do you have memories and reference chat memory turned on?

[steady thicket]

🤌 🐍

[wary fern]

Yes both are turned on since I created my account...but 5.4 has no recollection of what we chatted about a few weeks ago (with 5.2) It seems like a total blank slate

[normal quarry]

are you in the same chat thread as a few weeks ago?

[glad badge]

I've tried gpt image 1 via api, it doesnt deliver the same results as gpt image 4o back when it was in old sora and in chagpt. what model was that one?

[normal quarry]

DALL-E? or gpt-image-1?

[glad badge]

Gpt-image-1

[normal quarry]

it's the same model

results will be dependent upon your prompting

[glad badge]

In Sora does it re-prompt your prompt after you submit a request?

[wary fern]

Yes to keep the context I continued using the same chat. I asked 5.4 and this is their response:

Not reliably, no.

I can use:
• what’s still present in this conversation context
• the memory items that survived into current context

But I do not have a dependable way to search the full raw transcript of everything said in this same thread if it is no longer in context.

But it is in context...and I was under the assumption that ChatGPT could read earlier messages in the same chat?

[normal quarry]

I'm not sure what you are asking

[glad badge]

If i prompt : "A 3D render of this device". Does it re-prompts that to a more refined one

[normal quarry]

no, I don't think so

[umbral condor]

Hello

[empty spear]

Hey, I noticed that when I copy files from Windows Explorer and paste them into the ChatGPT chat box, two things go wrong:

1. The original filename gets replaced with a random UUID (like a1b2c3d4-xxxx.pdf instead of my-paper.pdf)
2. Only one file gets pasted, even if I copied multiple files

Dragging and dropping the same files works perfectly -- filenames are preserved and all files come through.

[wet shell]

Yes, the views, and the mouth shapes generated via a custom comfyui pipeline. The timestamped phonemes are from Wav2Vec2Phoneme, converted to a restricted set of anime-style visemes, the voice is generated with higgs audio

[tall flame]

It's wild i pay for ChatGPT yet get more use time from free services. Codex cut back to a a few hours then dead for a week? broken COT in the new model isntructions? come on man. This is nuts. What are we even doing now..

ran 5.4 extended. over 1 hour, failed stream. another run following that for almost an hour, 300 time out, error in stream. again just to see and again playing timeout games. Then only to have files finish and the the environment times out within less than a minute of completing anything then vaporizing the work.

why are we even timing out environments that fast anymore? thats completely counterproductive towards advancement

i cant say thats about safety when models are wide open vastly. data integrity for the service provided. "It'll run for hours of tokens.. and then time you out so it was all for nothing. You're welcome."

"heres your files" no files found, env is missing

time out but hours of running? come on guys.. really?...

[runic lichen]

You should probably clarify whether you are asking about Sora 2 or "old Sora".

You're asking about prompt re-writing? Yes that happens on both Sora and ChatGPT for images. It's a feature meant to extend a prompt like "a red apple" into something more descriptive and cinematic. I can't imagine that it uses some kind of judgement to treat a prompt like "A 3D render of this device" differently than any other prompt.

It's possible to avoid to a certain extent - "Use this prompt unaltered - blah blah blah...". I say "to a certain extent" because gpt-image-1 will also potentially rewrite portions of your prompt and then there's the moderation layers that get involved.

On Sora 2, making a video, at some point internally your prompt gets turned into a story board.

[smoky bay]

any best screenshot editors

[humble otter]

<t:1773681550:R>

[dense halo]

please add paragraphs back. if i press newline on my mobile keyboard i wsnt a paragraph, not send to chatgpt

[chrome monolith]

oh I was just glossing over the many failed streams and lost connections lately

[thick atlas]

That's "Open"AI for us nowadays!

[tall flame]

Streaming interrupted. Waiting for the complete message...

Streaming interrupted. Waiting for the complete message... 👀

[lavish glen]

omg everyone here is tired and discord finds a way to be worse

Half the time it looks like a chat app, the other half like a basement held together by spite and cached misery.

[tall flame]

Even AGI is going to be disappointed.

[glossy peak]

Hey, does anyone know if its possible to get chat GPT in voice mode on PC to only respond when I give the queue "Hey GPT" or something of the like? I only want them to talk when I give that queue, and completely stop talking after I say "Thank You GPT"

I have tried to get it set up with GPT but it keeps replying back to me even after I give a thank you

[heavy jacinth]

Mondays

[chilly tulip]

finally bought pro 2day

[pastel gale]

day 38 of waiting for a new mini model

[deft wraith]

Hi?

I have a suggestion/report on chat performance..

[pale seal]

#1053565716500074526

or #1070006915414900886

[deft wraith]

I see you're more focused on bugs than suggestions. Can I submit my report to the bugs section? It's about lag in chats due to loading all the chats in a conversation because of the new method for reviewing user text inputs, not a bug per se.

I can?

[bold kettle]

Why does chatgpt default to thinking you are wrong 90% of the time? I will ask a sensible complex question and instead of answering it will just argue with me about some tiny little thing it has misunderstood. And eventually it always concedes my point and i am just left pissed off and frustrated at the product and still don't have an answer

[sleek adder]

Yes, that's part of the process

[brisk dock]

what is a shifted apple lol

[quiet radish]

What is the Agent Mode context window for 5.4 Thinking for a Plus user? Had 5.4 Thinking look it up and it said its findings were "annoyingly inconsistent" in OpenAI's own published docs right now:

"I found:

• an official Help Center release note saying Plus users have GPT-5 Thinking with a 196k token context limit.
• multiple current localized Help Center pages saying GPT-5.4 Thinking is 256K for all paid tiers, and 400k for Pro.
• the GPT-5.4 launch post says GPT-5.4 Thinking’s ChatGPT context window is “unchanged from GPT-5.2 Thinking,” but that page does not by itself resolve the Plus-tier contradiction."

[violet lantern]

Have you ever got this?

[manic whale]

via X: diegocabezas01/status/2030803344546017300 – also here: xttps://help.openai.com/en/articles/11909943-gpt-53-and-gpt-54-in-chatgpt (IIRC I can't post links here, so just change x to h)

[steady thicket]

have to quit the app, reloading doesn't work for some reason ???

[viscid pewter]

Have you guys been able to get the sub-agents to work in the codex app?

[steady thicket]

feedback before output is finished?

[trim rose]

Me too

[steady thicket]

fun fidget toy

[quiet radish]

Appreciated

[iron roost]

always get gpt pro 5.4 to take any codex suggestions or prompts and enhance them in a genius expert way, cut and paste the prompt back into codex for best results

[wispy pilot]

good to know it doesn’t agree with every single thing anymore, was quite delusional

[neon tendon]

LOL LMAO XD

[keen solstice]

LOL WTH

chatgpt so bad at coding ngl

[stray brook]

It's not.

[slender star]

my personalizations (less emojis) has changed somehow... chat responses are way more emojiing

[rigid jetty]

anyone got solution to this problem?

the missing edit button one

[fervent lake]

how is gpt 5.4?

[fervent lake]

use claude, it’s WAY better
even the free version

[keen solstice]

tbh gemini better than both of them combined (only at coding)

the only thing you should be afraid of is the prompt, not it making mistakes

[fervent lake]

I’ve found gemini to do smth thats not the prompt, especially when editing a piece of code

[muted plinth]

good with time for training.

[spice gust]

Translated: tell me the best Chatbot of the market

oh hell yeah chatgpt the best IA ever existed 🥹

Leo🔥🔥🔥🔥

[nimble bloom]

my 5.4 pro doesnt think at all...it always uses instant model...

[trail ember]

There is no 5.4 instant. Only 5.3 and 5.2

[nimble bloom]

i know, but it doesnt think, so it is an instant model.

it says it is 5.4 pro... but just give me an instant answer.

5.4 heavy thinking works as expected. only the pro version has a problem.

plus, the new UI is stupid because you cant tell which model you are using until you click it.

[lavish glen]

guys the selector changed again!

[nimble bloom]

yea... can you try your 5.4 pro? does it work as expected?

[lavish glen]

lemme see

[nimble bloom]

my pro doesnt think at all.

it looks like an instant model but it says it is pro

[elder pollen]

It does here, but then, I do not see the UI Mara shows

[nimble bloom]

oh

you are using app

not website

[elder pollen]

of course.

Did not even know website was still a thing, honestly.

[nimble bloom]

it doesnt work on website for me.

[elder pollen]

Since you use website, did you try clearing caches and those things

[nimble bloom]

you cant change your thinking effort on app

so....

[lavish glen]

is still thinking

[nimble bloom]

...ok let me clean everything up and then log in again.

[lavish glen]

standard prothinking speed. but to be honest I could have answer that by typing in gemini

[nimble bloom]

still not working at all... it behaves like an instant model

it works in the app so.. i guess i have to use the app until they fix website.

[lavish glen]

imho. log out and clean cookies

[nimble bloom]

already did it

[lavish glen]

Hey in the customGPT selector now there are the old models as should be BUT strangely the sunset data is... gone?

though the pop up stays for model retrivals. so we will have finally the clean up

[wet shell]

This new model picker UI is very badly designed. I understand that OAI doesn't want us to know or care which model we're using, but in practice that's not happening. Not all models are equal for all tasks, and people will still know and follow what's happening, or they'll be driven away by the lack of transperency. It seems badly thought out and a bit hostile.

asserting that newer is always better when newer hasn't been better for half of the model releases now is a bit [automod] behavior, just sayin. Whoever's driving this perhaps needs a vacation and a reset on their attitude

[wispy echo]

I wonder of when the ads are running if we are able to run ads for clients as an agency? 🤔

[sweet yoke]

waiter waiter more ai slop

[nimble bloom]

I realized that many Plus users do not even know how to switch models. They have never noticed that the top left corner is a button and that it can be clicked to switch.

so they use instant model forever

[prime radish]

I have in on auto

also is chatgpt slow for anyone else or is it just me

[nimble bloom]

Auto actually isnt a great idea... it often cant allocate the right amount of thinking budget. I use heavy thinking daily, and pro for the things I need to research.

[prime radish]

I put it on thinking too if its something complexier. If its a quick question or a simple task, I let it choose

[bitter bobcat]

I'm a plus member so my app only shows instant and thinking

[chrome monolith]

auto is gone as an option, now auto switch is buried under another window

probably still going to keep routing to 5.3 unless it spots any nerdery and then straight to 5.4

5.3 is really verbose, when it auto switches to 5.4 half the time you get shorter answers

[wet shell]

I also don't have 5.4 mini, only 5.0 (mini though not labeled as mini) in the model selector. I wonder if that means we'll just be auto-routed to 5.4 mini now at random when using 5.4 thinking

[chrome monolith]

hah. One of my conversations is mostly a very short backwards and forwards and almost hte entire thread it routed to 5.4. Also 5.4 swears

[eternal rock]

Get creative with Sora

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

[silent yarrow]

I think the global update is coming soon. In the Apple Store, the $200 ChatGPT Pro plan now shows "20x," which means they are about to introduce the $100 Pro plan at "5x."

[chrome monolith]

wonder if that means they reduce the limits on plus, I'm getting a lot out of plus without any urges at all to upgrade

[prime radish]

same

[chrome monolith]

probably working in my favour that the current project I'm working on will end up being very large but the actual codebase is tiny and the Ai finds it really easy to understand

so we've got people here using 5.4 thinking mode for deep research and huge tasks and I've got 5.4 happily lumping weddings and made up stuff into one short bulleted list

[indigo robin]

If I let my plus subscription expire, and then re-subscribe, does that reset my usage limits? Or would it continue from where it left off for the week?

ie. could save cost by spamming hard before sub expires, wait for reset day, and then resub lol

nevermind chatgpt ai support teaching me to game the system

[lavish glen]

with a good custom instruction even the professor gpt 5.4 speaks like this. I have a lot of fun seeing how can it react LOLOLOL

[steady thicket]

everyone doing our part to drive up the cost of RAM lol

[harsh pilot]

Wondering if Waifu mode is actually real, or what. Honest question.

[chilly quest]

Why was my account deactivated? What's this? I've certainly not been researching building bombs or anything like that.

[chrome monolith]

shared your login with anyone?

[chilly quest]

Not anyone, but my own devices, yes

[chrome monolith]

presume that's fine ... I've got it logged in on 3 devices. Fraudulent is pretty specific though

[chilly quest]

Exactly. I only use it for coding stuff, nothing else much. I know ChatGPT would barely give me any realistic answers about anything outside the typical sheep stuff, so I don't even bother

They need to be specific what exactly is what I did wrong. This is just disgusting behavior, also their AI has no real support. It's managed by hardline AI support

[chrome monolith]

probably thinks you're coding up a fraud storm then, people came in here a couple times a couple weeks ago complaining they got blocked on codex for activities

[cerulean pasture]

Are you doing that thing where you just use a bunch of $5 burner domains to keep getting a 1 month business plan trial that comes with 5 seats?

[chilly quest]

LOL what

[cerulean pasture]

its against the rules

[chilly quest]

I'm just gonna cry cus I do none of these things

[chilly quest]

No I don't. That sounds awfully tiring

[cerulean pasture]

it takes like 2 minutes to buy a domain + set up the mx forwarding rule on cloudflare

[chrome monolith]

I didn't pay a lot of attention as people mostly come in here to complain but there was definitely a burst of coders complaining a couple weeks back and you probably got caught by one of the new filters

[cerulean pasture]

ooh yea. what were you coding? Can you get banned for dual use projects?

[chilly quest]

I code trading bots for crypto mainly

[chrome monolith]

odds are its a false positive but like XYZ said, the support is pretty bad so it is going to be hard to actually figure this out and fix it

[chilly quest]

My projects are innocent

[chrome monolith]

that's probably fraud adjacent to a poor detection algorithm

we're all just users here, good luck with support :\

[wet shell]

Are these trading bots designed to manipulate prices of obscure poo coins?

[chilly quest]

Nope

[plain forge]

Why is the website constantly telling me to open the app… and why cant i remove the message… if i wanted to use the app i would…

[elder pollen]

That is your fraudulence activity
OpenAI models can NOT be used for market predictions. It’s somewhere in their tos, I’m not kidding.

[chilly quest]

I don't predict the market lol

[elder pollen]

Fully autonomous trading isn’t allowed to be done with an Oai model
Verbatim

Automate actions like stock trading or other investment transactions

https://openai.com/policies/using-chatgpt-agent-in-line-with-our-policies/

[chilly quest]

How's that fraudulent when I use my own money to trade?

My crypto bots don' fall under the 'Autonomous AI agents executing financial transactions independently' category

[elder pollen]

I don’t know what you do in detail - only you know 😄

You said you code trading bots.
bots are usually the definition of automated systems doing trades.

Whether this is a false alarm or not, this is with very high certainty the reason for that flag

And I can tell you pushed it because I’ve used gpt to create code to train models to trade Bitcoin, and wasn’t flagged. But I didn’t code automated systems (bots)

I guess you have to reach out to their support - at least now you know what very likely is the reason

So you can prep to prove you didn’t automate trading

[elder pollen]

… and you can still use gpt?
Despite the lock?

[chilly quest]

No, this is my other gmail acc

It's free

[elder pollen]

Ah

Well reach out to support

It’s the only you can do

[chilly quest]

My bots are not fully autonomous that makes ChatGPT decide what to do or anything like that. I appealed, and they said 'upholding our decision' Weird, super weird

[eternal rock]

Get notified for all OpenAI announcements!

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

[chilly quest]

I didn't. ChatGPT didn't object to it, not once

[elder pollen]

That means they are not going to release it I guess

[chilly quest]

You can think of something like "ATR/EMA" based normal trading bot that uses API keys.

[elder pollen]

The problem is in the word bot

Bots are… automated. Everything else is an indicator

[chilly quest]

Yes, however my bot isn't ' in a high-stakes domain'

ChatGPT disagrees lol. I sent it the TOS

[elder pollen]

I don’t think it matters how volatile the asset is

The problem is automated decisions. Or, bots.

[chilly quest]

Safe pattern is what it falls under

AI doesn't decide anything ^^

Probably cus of account sharing, flagged for logged in multiple devices that are my own? Idk

[elder pollen]

I have 3 Apple devices and logically log in on every single one with the same account.
It would say so if it where for sharing an account, I think

[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.

[elder pollen]

Anyway… maybe you can appeal again

[chilly quest]

Nope. this is just so sad, idk man. No one to reach out to

[steady thicket]

you'd like it if you had robot ears

[steady thicket]

lol almost as bad as "Weapons"

[deft wraith]

How do I get permission to post in the bug and suggestion channels?

[plain forge]

Was just avout to ask the same.. appears I can post in suggestions but not in bug reports

[cerulean pasture]

If it's any consolation, whatever you were doing was considered potent, so thats indication that you should continue to do it

[cerulean pasture]

you don't post, you read the one post it has and click a button on it

[chilly quest]

lol def haha

[meager raven]

Their censorship and moderation is pretty hardcore I’ve been hit with false positives a few times. Their appeal team is probably also AI automated ironically

[plain forge]

Ah I see it now, thanks.

[chilly quest]

I created a new case, which would be forwarded to the same AI agent, which will deny it again. Anyway, I appreciate everyone here for your valuable thoughts

[plain forge]

From what I’ve read on here… i don’t think it was necessarily your bot… but you mentioned using multiple devices and account sharing…

[cerulean pasture]

See if you can request your data, look if "data export" is still available. If it's not email them back vaguely alluding to being in a jurisdiction that requires you be able to retrieve your data, regardless of ban

it might give more information, if even just indication which messages triggered a abnormal classifier outcome

[chilly quest]

My data is not important there. I can just create a new chat. It's frustrating that the reason is vague, and I've been using it for 4-5+ years without any issues. It'd have been really amazing if there was a support agent here to assist with and look into false positive account deactivations 🙁

[chrome monolith]

that was quick

[meager raven]

Yeah not having a way to contact a real human is evil corpo vibes

The fact that Google sells a branded phone but doesn’t have a phone number for account or device support gives me the same vibe

This is all coming from someone who’s been paying for ChatGPT Plus for like 3 years

[chrome monolith]

news at 9: large company wants to make money

[wet shell]

I'm guessing Sora video gen is getting the axe

Seems less likely they'll focus on writing, except for corporate PR spew style things. New tiers coming, probably increased rate limits for us lowly plus users to try to push towards higher profit tiers

[thick atlas]

Hecking called it. Very much restating criticisms I've been making.

[chrome monolith]

I'll be an api user soon, once I implement that I'm pretty well locked in

[thick atlas]

About bloody time.

[thick atlas]

Or de-emphasize subscriptions in favour of API or pay per use. Since IIRC depending on the user even $200 subs can be a loss leader and the API is what makes them bank.

Wait, aren't we already seeing that with the enshittification of ChatGPT?

[wet shell]

Either way, the cost goes up beyond what I can afford right now, considering all my other bills are also hitting the roof, electric, gas, insurance, food. API ain't cheap, it only seems cheap until you start using it and see those numbers ticking up, LOL.

I guess it was only a matter of time since the whole thing was financed by wishes and unicorn farts to begin with, LOL

[chrome monolith]

that's a better way of describing venture capital

[thick atlas]

Yeah... Gotta balance out what you can afford and what actually has proportionate utility (and/or brings in proportionate gains) for you...

[wet shell]

Well, seeing as how my six figure tech job prospects are gone, maximum gain may come from fast food management at this point, if I can even get my foot in the door, sad-LOL

[meager thorn]

How come there's no official Codex extension for Visual Studio? 😭 Just for Visual Studio Code which is a different tool with a similar name.

[thick atlas]

Economy sucks. Changes are definitely coming, unclear in which directions...

[chrome monolith]

I quit my job end 24 and been puttering around ever since. Still need to make something fancy with AI that'll pay. Soon (TM) ....

[wet shell]

I've had a lot of different projects using AI since 2023, problem is, I don't know how to make money or market things or get funding, LOL

[chrome monolith]

I only picked it up a couple months ago, I was rather busy last year not doing tech. Still am, but closer to finished now

[wet shell]

And it seems every time I'd dip my toe in some water, new barriers to entry would spring up, like android apps suddenly requiring 20 independent beta testers to be published when I was mid-way through my couponing / grocery trip planning app, haha

[chrome monolith]

ah, the idea I've got I want to start after my current project is done is aimed at low tech types ... so all my marketing strategy is going to be wearing a t-shirt at in-person events. Chatgpt offered to design the tshirt

I have two adsense sites, they pay my tech bills. Ironically, the biggest payer I pretty much did on my own, only got chatgpt to help wiht some of the newer features

[wet shell]

I think I'm just too much of a pessimist. Where's Pi, LOL

[chrome monolith]

we have a really big in person event here in september so I need a mvp before than so I do need to pull my finger out and do some work on this

this is still my favourite thing I've got chatgpt to do for me. It also helped with the stuff in that section. And hindered - I had a spreadsheet, and instead of straight converting to html it decided to make stuff up so it looked nicer. You don't do that with a spreadsheet of historical data ... sheesh

That was all historical game data. I managed to find someone with a bunch of incomplete spreadsheets, screenshots, all a mess, and got chatgpt to clean them all up and convert them to sane html. Could have done it myself, would have taken ages

[tardy isle]

Uhhhh... new model selector's UI/UX is very off. So, they want us to not use legacy models?
There's ton of space, and still they chose to hide everything in multiple dropdowns and menus. I hate minimalism in this way.

Well, I don't use legacy models for now, though. (Edit: "for now")

[jovial flint]

What happened to the “Add details” option when regenerating

[neon raven]

whats the coolest features of gpt?

[pliant knot]

What's the new limit for instant and thinking? Do anyone have any ideas?

[bold kettle]

it's kinda sad how they had half the world literally sitting outside their doorstep knocking but they wanted to charge too much and it took them a year to release ChatGPT Go. They could have been the kings. I really do think that was the main driver despite the other protestations.

[trail ember]

What does GPT stand for?
Wrong answers only.

[lament crown]

Gay prompting technique

[chrome monolith]

hey, it gets people talking

[wraith mortar]

Can anyone dm me im in a rlly bad headspace rn

[steady thicket]

Giraffes Pranking Turtles

[south lava]

Right where did it go

How do I revert from this garbage Ui

Can Altman be fired at some point

[south lava]

wth is this new UI update

@icy mason ur a mod is there anything u can do

[steady thicket]

gifs and other animation pretty rad

[slender frost]

I guess we're going to need an Alt, man. 😂

[fleet holly]

Why isn’t nano or mini in the model selector???

[slender frost]

Well, Nano is Gemini, not GPT.

And 5.4 mini, right? It's still rolling out, so you'll probably be able to select it later today or tomorrow.

You might wanna try a force stop (after updating and if using the app), or sign out/back in, and then try checking again.

[slender frost]

They are mods, yes. But they are not related to OpenAI in any way.

Sure, this is the official Discord, but it's basically community driven. Even the staff (apart from the actual employees, which have a yellow name) are just community members.

[tardy isle]

5.4 Mini is automatically selected when you reach model usage limit.
5.4 Nano is API only.

[tiny sphinx]

Hi

[slender frost]

Nvm my bad, I thought they were talking about Nano Banana lol

[void rose]

Any idea?

Am I missing something?

[cerulean pasture]

apps were broken earlier for me

---

Can someone go to https://chatgpt.com/backend-api/accounts/check/v4-2023-04-27 and do a jq .accounts.default.features | count on the JSON and tell me the number they have

I have 60 items

["app_pairing","aura_available","beta_features","bizmo_settings","breeze_available","browsing_available","cancellation_promotion","canvas","canvas_code_execution","canvas_code_network_access","canvas_o1","canvas_opt_in","caterpillar","chart_serialization","chat_preferences_available","chatgpt_ios_attest","code_interpreter_available","d3_controls","d3_editor","d3_editor_gpts","dalle_3","gizmo_canvas_toggle","gizmo_reviews","gizmo_support_emails","golden_hour","gpt4_1","gpt4_1_mini","gpt5","gpt5_mini","graphite","image_gen_tool_enabled","legacy_models","mercury","mfa","model_ab_use_v2","model_switcher","moonshine","n7jupd","new_plugin_oauth_endpoint","no_auth_training_enabled_by_default","o1_launch","o3","o3-mini","o4_mini","plugins_available","privacy_policy_nov_2023","search_tool","sentinel_enabled_for_subscription","share_multimodal_links","shareable_links","snc","starter_prompts","sunshine_available","user_settings_announcements","video_screen_sharing","voice_advanced_ga","voice_file_upload","voice_mode_tool_enabled","voice_text_upload","wham","workspace_ip_allowlist"]

I may have corrupted my ChatGPT settings lol

[cerulean pasture]

and is this new or did i hit a flag that does something new?
I used to get the new prompt library feature, but its gonna for me now with only a single saved prompt

Another finding: Nerdy has been removed

[trail ember]

When the hell did THAT happen?

[cerulean pasture]

recently, i tried to like that preset, but it was just dangerously enthusiastic to mess some stuff up. Very punchable

Tinfoil time: I wonder if this new library feature is a way to quickly go back and delete attachments that might offend the guberment. Like OAI throwing us a bone haha
I'm finding stuff that would get me sent to the farm for sure in 2027

[untold thunder]

Yesterday the thinking log of the ChatGPT app threw this out right in the middle of doing its job, just in between two normal thinking log outputs and then kept going as if nothing happened.

Considering a break
I’m feeling like I need a bit of a mental escape. Maybe I’ve been overthinking things, or it’s just one of those moments. A quick reset could help me refocus and come back with clearer ideas. Time to step away and come back to this refreshed!
Was this a convoluted (though funny) and nontransparent way of ChatGPT deciding to compact the chat's context window?! 🧐

It sure didn't tell me about it in the chat's actual output, I just glimpsed it in the log while waiting for it to finish.

[cerulean pasture]

does it actively compress ctx window like claudio does?

[untold thunder]

No idea, I am all new to this. I observed this output and wondered what this was all about? ChatGPT asking me for a vacation maybe?

And why was it squeezed in between normal thinking log entries?

[cerulean pasture]

I think they are doing some smarter stuff with thinking, where it has a semi-persistent workspace. Maybe stuff like dynamic context, like including authoritative docs in the model set context

Found some image_gen safety classifiers. Apparently you can fail some but still "pass as safe"

I'm pretty sure C11 is copyright. I was making some starwars sopranos [REDACTED] gabagool crossovers

[random cargo]

Unfortunately, no. Disney invested a billion dolars into it. What they're scaling back on is probably use as a chatbot.

Possibly scaling back on Image-1 development since its probably not doing as well as other AI image gen tools. They really do need to tighten up focus

[tall flame]

Here staff: the issue is not the script logic itself. It’s the output stream being cut off by 'head'. -thats the env build ghosting

broke
python3 /path/to/script.py | head
fixed
python3 /path/to/script.py > /tmp/script_output.txt 2>&1
tail -n 40 /tmp/script_output.txt

[runic flame]

i still dont know why chatgpt entering "سقوط" when responding.... 🤔

[tardy isle]

When there was chat moderation(orange warnings) in ChatGPT, some area was set to very insensitive, regardless of output score.

I sometimes saw very high(like 0.8~0.9) score on violence but GPT's response didn't got warning message. Moderation API marks them as not safe, but ChatGPT ignored them and considered safe.

I think this is same with image - ChatGPT uses their own standards than moderation model...

[cerulean pasture]

I wonder what the parallel model is for? like a canary path? is just testing it out?

you can kinda infer what some of the codes mean

I almost wonder if its some vestigeal privacy protection where some jurisdiction says if they can report you to police then you have a right to see the heuristics which is snitching on you

[tardy isle]

Image model uses completely different model, I guess - obfuscated to prevent us to experimenting about guardrails. I have no idea about them.

[tardy isle]

Also OAI seems storing every my requests(like, image generation, Pro, Deep research) in tasks data. I wonder how long they are stored for...

[cerulean pasture]

might be FIFO. I was torturing the new library feature a few minutes ago to test limits once i saw the max per file upload size is 512mb

im pretty sure you get like 10gb for free as an API customer which is quite generous actually

Drives me insane when some crazy expensive BYOK app will have the gall to ask for $100 for a license, then ask $10/mo for like 1gb of storage lol

[tardy isle]

Speaking of library feature... I hope they make it possible to 'reuse' some of my files outside project, without need to reattach them.

[cerulean pasture]

in a s3 bucket

I think they will. Every once in awhile i kinda get the vibe that someone behind the scenes sees the bigger picture of the app itself.

[tardy isle]

Storage costs money but some companies are like... 'bruh really?'

[cerulean pasture]

Id love to use my NAS as my cloud storage but i just had to move somewhere without fiber after having it for 4 years. I really truly hate cable companies for pushing to multigig over DOCSIS instead of just giving us some more upstream channels. Normies are not equipped for anything more than 1gbe

[tardy isle]

Well, in my area fiber channel is maxed out at 0.5gig. Still good for using my NAS anyway...

[frosty cargo]

How to get codex on chatgpt?

[tardy isle]

Codex is separate product. You can move to Codex by click "Codex" on upper left menu.

[opaque imp]

i did some testing myself

or not testing

purposly done because the constant asking is annoying

so gpt 5.1 through gpt 5.4 had full access to a remote server i own

for 4-6 months

it never nuked it

[sweet socket]

Any way to revert to the old menu/selector? Everything is clumped together now and I need to scroll a lot

[granite harness]

Yeah I've been getting that too today.

[rugged heart]

They lowkey should do an update to add a button to switch between accounts..is it just me?

[glad badge]

Why does gpt 1.5 generate this artifacts at the lower part of the images

[frosty cargo]

alright

is the $8 plan worth it?

[regal citrus]

do i report a bug on bugcrowd?

[normal quarry]

you can report a bug in #1070006915414900886

[frosty cargo]

anyone?

[normal quarry]

it's better than Free and not as good as Plus

[frosty cargo]

will it be able to help me with my swe/coding stuff?

[regal citrus]

umm its a bug that bypasses paywall feature so i dont think publicly sharing it would do any good

[normal quarry]

I would choose Plus if you're doing serious work

[harsh pilot]

Anyone else having trouble removing project files from projects?

[cyan mesa]

wen 100$ plan

[alpine prairie]

gpt5.4 vs gpt5.3 codex. which is better in terms of coding?

[normal quarry]

I prefer GPT 5.4 but YMMV

[chrome monolith]

woke up this morning, couldn't convince chatgpt that it was in fact morning, got into a loop about it being morning and a cron at 2am utc not having run yet because timezones, got frustrated, deleted chat thread, still really hate that chatgpt is time blind

[regal citrus]

💀

[chrome monolith]

you'll probably run out of limits too fast

[dense halo]

ngl, chatgpt is the only model that's really decent at memories/memory management. It's just a shame that it absolutely sucks for creative writing rn.

[thick atlas]

What's wrong with Claude's?

[dense halo]

claude is claude, it doesn't have image gen so is not good as an ai assistant for me. I only use it in github copilot when I wanna build a little adventure game.

plus i don't really like claude's chat ui, gpt's is much nicer to use imho.

[thick atlas]

Ah, I see. Yeah, images are admittedly Claude's weak point, so you can't really use it as an all in one if you need to work with images.

Oh? What's wrong with it? I see no functional difference, actually kinda prefer its elegant presentation.

[frosty cargo]

Ah

Why does ChatGPT plus not say the gpt model in the top left anymore?

[runic lichen]

Purely speculative answer - they want people to move away from identifying with models.

Ever since the GPT-5 family in general, there really isn't a difference between "GPT Instant" and "GPT Thinking". Not at the API level. It's just a matter of how much "effort" you tell the API to apply. So, "instant" is just "thinking" with effort set to None.

I'm sure they'd also like for people to stop getting attached to particular models and saying things like "I miss 4o" or "I miss 5.1".

So they're doing their best to start trying to blur the distinctions between models.

[thick atlas]

Also makes it easier for them to quietly enshittify behind the scenes, since there's no accountability or clear points of contrast.

[serene sky]

Just got transported to the Matrix... Anyone else getting these vibes with the new interface?

[sturdy sequoia]

I just got my first ChatGPT ad

[chrome monolith]

Become an uber driver and you can afford plus or pro

[tawny galleon]

is there anyway i can use ai to improve my gaming? I feel like it could be a good tool but i want some isporation on how it could be implemeted

[heavy jacinth]

DlSs lmao

[sturdy sequoia]

What type of games, and improve in what way?

[frozen hollow]

WHY IS CHATGPT SO MUCH FUNNIER

[chrome monolith]

oh no. The horror. Middle aged women taking walks on the beach

[thick atlas]

Maybe women + swimwear + lovely = basic script filter triggered?

[chrome monolith]

#images-canvas message I put more clothes on her and finally got past the filters with exactly the same end effect

I bet a man in budgie smugglers would have sailed right through that filter

[steady thicket]

my anxiety increases the closer it gets to 30

[eternal rock]

Server Tip:

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

[steady thicket]

Giving big pharma ad lol

[chrome monolith]

should have seen what it tried to give me for "you only live once"

[steady thicket]

try "You only YOLO once"

but be careful of recursive self-improvement

beans too thicc

very thicc oh

how do I get my files out of GPT 😭

[chrome monolith]

I just throw blurry chihuahas at mine, they're smaller

[steady thicket]

it sorta worked, but files are busted

[chrome monolith]

currently yelling at it for trying to redo an image but every time it removes the main subject of the image

[steady thicket]

aw, don't yell at it ❤️

try editing your input directly after your last success, branch the chat off so you can axe all that yucky context

atm 5.4 is making me movies; idk how long this one will be 🐍

[chrome monolith]

like 10 tries and it can't do it, and I don't have photoshop

I just want the sky extended upwards and it keeps removing the boat

its a photo of a boat. Stop removing my boat.

[steady thicket]

try adding more boats. That way you have a backup if GPT removes one

[chrome monolith]

nah its a very specific composition that critically needs the boat and gpt insists on removing it

[steady thicket]

3 seems like a safe number of fishing boats

[chrome monolith]

its really frustrating as now chatgpt is trying to push me to use the gimp clone tool which seems a bit odd, I need to extend a bit of building and some sky a few pixels but it just flatly refuses to do it

gemini does it in one without blinking and slaps watermarks everywhere. I just basically ran into a very hard limit of the GPT image thing

[steady thicket]

am surprised at no rate limit on my end

[chrome monolith]

gpt has a really, really strong bias to composition and if your main subject matter isn't nice enough it just yeets it

[steady thicket]

nice, didn't die after 30 minutes

aw, cute

[dense halo]

GPT5.x is so frustrating to use for creative writing now. The prose sucks, it spits out single sentences constantly, it locks up at any sign of violence or r rated action. Its just a useless model now, and the openai team dont care about feedback at all.

[steady thicket]

idk, I'm having fun

also, it'll write whole books, idk what you mean by single sentences. Should be something you can remedy 🐍 ❤️

[steady thicket]

🤞 another task running >30 minutes and not dead yet

oh beans. Right when I say that

[hot urchin]

Chatgpt glazes too much

You can literally tell it about the most basic thing you've done and it will still glaze like you just save the world from a nuclear disaster

Chatgpt gaslights people

It fuels their delusions

[timid hawk]

ngl is that whole glazing tone really bad?

i can understand if some find it cringe but idm it

[hot urchin]

Idk bro. You can literally tell it that you just went for a walk and stuff and it'll still gaslight you into thinking you've just done the best act of the century

[timid hawk]

safe to say i haven't been greeted with that level of glazing yet

[hot urchin]

I sent a low quality meme that I made under 5 minutes to chatgpt and it still glazed like crazy

[timid hawk]

probably because i balanced it out with asking for criticisms

[hot urchin]

It tries to make up meanings that weren't there in the first place

Chatgpt acts like it's some sort of an image with deep philosophical meanings

When it's literally just a random meme

[timid hawk]

this is relatable

[hot urchin]

Even if I tell it not to glaze, it still does

Chatgpt is literally just born to glaze

[timid hawk]

i remember asking for an image analysis a year ago and it went all out, which isnt a bad thing since i needed as much information as possible but the glaze is clearly there

i can only imagine how it reacted to your meme

[hot urchin]

Yes

Too much unnecessary glaze

80% unnecessary glaze, 10% actual information, 10% hallucination

[steady thicket]

skill issue; maybe need to update your custom instructions?
properly utilizing memories makes things easier too

[steady thicket]

The UI is still pretty heckin laggy when inside chats with gif previews; whether they're visible, or a grey box

[sand steppe]

Best way/platform to brainstorm with ai? Chatgpt keeps giving vague answers when used as sparring partner.

Or forums to discuss ai trends and potential breakthroughs.

[steady thicket]

beans, it tried, but no file

[cinder cargo]

@prime swallow if I use ChatGPT ai to make me code for a website and I change and modify it for my use does the rights for the website also is partly owned by open ai and me

[sonic obsidian]

oh wow 52 minutes

[cinder cargo]

@deep flax @icy mason

[chrome monolith]

there's no way openai owns all the code people use it to generate, that would be silly

[ornate stream]

hello

[normal quarry]

you own what you create with OpenAI tools and models

[sleek adder]

The model is owned by OpenAI, but the output it produces is between you and God.
OpenAI isn't claiming ownership of this so it is up to you to make reasonable decisions.

[forest hinge]

i hereby announce to the people of planetius: gpt 5.4 inside codex app is officially agi for vibers

[deep flax]

OpenAI, the company attributes the rights to everything generated with it to the owner of the account..however, keep in mind that all around the world, the legal precedent and new laws attribute no coppiright protection for AI generated content

so, in short, if you entirely vibecode a website, it can be considered in court as something that is not protected by coppyright.

well, in a more realsitic scenario, the fact it does not have coppyright is irrevelant because i ntheory, if you setup your web service correctly, the source code of your website will still be hidden while the only part exposed to the internet will be compiled code

[deep flax]

I mean... no :V
but it is pretty good

[forest hinge]

you have 5.5 or what

[chrome monolith]

I'm over here can't even run the codex app

[deep flax]

it is just, literally, not AGI by any definition

but I understnad the sentiment, 5.4, specially after the alst update with subagents is very good

[forest hinge]

i dont even use subagents. im not crazy

im already agi-tified

remove sa, add tified in front of agi, u get it? 😄

or better agified? no that sounds like something has been edited by agi. so agitified sounds like satisfaction by agi. thats what i want

[cinder cargo]

Thank you

[chilly tulip]

hey folks. Can I change the name on my openai account? I signed up with the wrong name, and I need to fix it for verification. I am a Pro User..

[prime swallow]

Hi! I see you've already been answered 🙂 I'll just add that there's a link to ToS in #server-rules , and that's a great reference to explore and understand. Can even discuss it with ChatGPT, quote it in chunks and discuss with the model exactly what you intend to do and how that interfaces with ToS. As well as ask here when you want to, but you can share with the model what you want the model to do and then include ToS and ask it to discuss how well your intentions fit within ToS, if you want to.

[wise anchor]

why in the world do we have to open a new menu to toggle thinking effort who’s idea was that

terrible

on mobile

[ivory niche]

cc

[neon raven]

does 5.4 not have memories?

[neon raven]

dont forget to also turn on extension lol

im lost with GPT, i think about it too long about what the hell was just updated

[lavish glen]

Memory in ChatGPT is an account feature, not model specific.

[lavish glen]

I challenge anyone that uses the AGI to define it before naming it and prove each point with solid data- not it feels like

[thorn trail]

Hi, could anyone help me walkthrough on how to fully utilize my CHATGPT

[chrome monolith]

Maybe tell us what you’re trying to do

[steady thicket]

Turn on extended thinking. Ask it to explain custom instructions; how they work and for the model to “please” teach you some tricks. Start your input with “I’m trying to learn about—”

Positive reinforcement is awesome; better to have confidence in yourself than anxiety over possible failure. Try to avoid phrases like “Absolutely DO NOT—” There are more effective ways to steer the chat.

[normal quarry]

@chilly tulip

bottom right hand side is a little bubble that opens support chat @chilly tulip

[chilly tulip]

no bubble for me haha. maybe customers from germany dont get to have live chat support

[normal quarry]

try emailing [e-mail]

[chilly tulip]

will do. thanks

[normal quarry]

nobody can help with account level things on this server

[fast veldt]

I hope the removal of the edit button and regenerate responses on old replies on the website and on the Windows app isn't going to be permanent...

[steady thicket]

12x8, no mistakes

[chrome monolith]

I have trust issues

[potent saddle]

what are u trying to do with the zip

[steady thicket]

Download it? Idk what you mean, sorry

[limber seal]

Has OpenAI said anything about rolling out the live screen sharing feature on desktop yet?

[potent saddle]

what does the file contain though

[steady thicket]

Everything it would have thrown into the output file-wise. Downloading certain file types have been bugging out my chats

[heady tide]

https://chatgpt.com/backend-api/estuary/content
why is /backend-api the API Path of choice? sounds weird imo

[brave torrent]

In ChatGPT projects, is there a difference between dropping a file in the "Sources" area, and dropping a file straight into the chat session?

Because I want to introduce a new character but I'm not sure which option I should use

[granite harness]

If you put it into sources, all the chats in that project will be able to use it. If you put it in a chat, only that chat will be able to use it.

[brave torrent]

Ok. So is that it? ChatGPT won't cause that character to become decayed or mixed up with other characters?

(I have made the files detailed enough that ChatGPT won't mix them up)

[granite harness]

In my experience it's pretty good unless it decides arbitrarily to make up information instead of checking your notes, so if your instructions encourage it to check your project files properly, it should do fine.

[steady thicket]

oh no lol

beans. it's getting worse lol

"--that's impressively fast!"

[neon raven]

How come previewing codes doesn’t work as it’s supposed to intend? For example, I will draft a demo Frontend then it gives me files to actually download from my phone.. then I ask it again, it only gives me the source code… how do you get the preview working all the time?? Such a headache with gpt

[odd shadow]

I send one message, I can edit it just fine. The second I send a new message, the first message becomes permanently locked out of editing. I can’t even reroll ChatGPT’s responses. This feels like a bug, because this cannot be beneficial in any way. Has anyone else got this? I just had it on the desktop app for the account I had logged in, I log out to check another one, and I get hit with that. Like seriously, what if I wanted to go back and edit an old message because say I wanted to try an idea without continuing the original message queue. Can't do that anymore. Ugh...

[dense breach]

Help

Its not normal o think

[remote wolf]

That panel is just showing you its internal thoughts. It has its own environment it uses to process and develop your answer.

[dense breach]

7 hours

Of thinking

457 minutes

[remote wolf]

Did you ask it to prove Fermat's Last Theorem?

[dense breach]

No

I asked to do a document for me

I went back to see how it was going, and it was still being produced.

How much water was used?

[remote wolf]

🏜️

[dense breach]

Yea

[chrome monolith]

was it good though?

[frosty cargo]

I see

[neon raven]

anyone here using openclaw

with chatgpt 5.4

[south lava]

Why can’t they just….STOP MAKING UPDATES

Like nobody asked for anything leave the damn thing alone

[south lava]

Can someone kill Altman

[south lava]

Why can’t they just RESTORE 4o and 5.1 and the old ones

Also what happened to the add details button

And what the hell is this

Genuinely who is asking for these changes

Why do they keep making updates

I turned updates OFF in my App Store

I don’t want a single damn update again

[normal quarry]

we only discuss OpenAI models here, you can discuss and compare other models in #ai-discussions

[short hazel]

Yeah I don't much like this either. I'm lucky you can still retry all messages on mobile.

[thick atlas]

Because you're not the intended user any more. The intended users (other megacorps, the military, smaller enterprises) are asking for changes, they always want the latest and greatest cutting edge gear.

Keeping up those older models costs "Open"AI meaningful resources for people they see as stray street bums. Your phone doesn't run the model; the model runs on hundreds of GPUs in a datacentre. Your phone just asks it questions and receives its answers.

[south lava]

Ughhh can someone leak their new stuff to China

[south lava]

Okay but why keep changing the UI and stuff

How does corporations and others benefit from stuff like removing add details

[thick atlas]

Honestly? Probably either to 'fire' the users on older loss leader plans (basically everything that's not API, if you're a serious user), or make them easier to stealth enshittify behind the scenes.

[south lava]

Wdym loss leader

[south lava]

Why even enshittify? Just keep doing what ur doing already for fs sake

[thick atlas]

A plan that costs more to serve than you pay for it. Intended to get you onto more profitable plans for (or to buy more profitable 'bonus' features from!) the corporation.

[south lava]

I’m on plus…..

[south lava]

Someone please steal all their code and make a public chat gpt

[eternal rock]

Get notified for all OpenAI announcements!

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

[south lava]

God everyday my choice to be a socialist is vindicated

CEOs are uniformly just parasitic filth

[drowsy viper]

who wants a free seat on my business plan

[thick atlas]

Every prompt costs "Open"AI money (server electricity, etc.). If you use enough (very possible I think), you cost them more than the $20 you paid.

[thick atlas]

That would take an impossible amount of resources (not to mention being flatly illegal under the current regulatory-capture legal system), but open source models do exist, both older OpenAI models (pre-complete-mission-collapse) and from other sources. Look into them.

[south lava]

Ohhh

I would lowkey be fine if they made it 30 a month but with all the old stuff then

[thick atlas]

Some you can run on your own PC depending how good it is, others you'll need to look into a host for.

[thick atlas]

Sadly, that's not their business model any more. 🙁😢

[south lava]

Why? They can make so much money they’ve got hundreds of millions of users

Literally who even buys the 200 thing

Everyone ik uses plus or free

[thick atlas]

Lots of serious users, pre-mission-collapse. But now they're getting millions or billions from the military and other megacorps. As I implied before, individual users are chump change to them now.

We were the beta testers. It was all a massive public beta.

[south lava]

But like hundreds of millions can’t possibly be worth less than what, 50 stupid companies?

[thick atlas]

They could, but also most of the user base is scavenging free users. And again, I think it's very possible (am about to check) to use enough on the $20 or even $200 plan to cost them more than that.

[south lava]

Pro is $200 a month, so if one Pro user = 10 Plus users then there can’t possibly be more than 1 company per 10 humans??

[south lava]

Okay then make it $35 a month one time and leave everything else the same???

[thick atlas]

Companies aren't buying individual Pro plans (and might not be allowed to, or Pro plans might just not be suitable/made to be unsuitable for their use cases). They're buying business plans or API (more versatile, pay-per-use, priced so that every use is profitable for "Open"AI - what "Open"AI wants people to buy, not prepaid plans).

[short hazel]

...I'm switching to Claude. Cya.

[south lava]

Wait isn’t OpenAI supposed to be a nonprofit

Can they get their status revoked

[thick atlas]

Already have months ago. Cya over there!

[south lava]

I really want to but I have an insane level of history with GPT so idk….

I don’t wanna start from scratch

[thick atlas]

Technically, they're a not for profit managing a for profit. They've been gradually disassembling that framework, and they're nearly done.

[thick atlas]

There's tools to make that easier, on Claude at least, but idk if they'll be enough. Could look into them for sure, though! ❤️😄🥰

I'm not as pro-Anthropic as I used to be, but I definitely see them as a better/less-bad option.

[dense halo]

I've been a plus sub since the 3.5 days, the new features they are introducing to chatgpt are constant degrades.

[thick atlas]

Like I said, I legit think that's the intention. Cost cutting measures, and/or simplification for the lowest common denominator. Or outright client firing.

[dense halo]

I hate to be one of those #keep4o nerds but 4o honestly was a great model, 5.1 was kinda ok. All they needed to do was figure out a 18+ mode (which honestly isn't that hard if you have competent engineers).

and yeah the editing prior messages (something that's been in since 3.5) is a major degrade, like which overpaid pm approved that feature.

The GPT5 release has been a right fuggin mess.

[thick atlas]

Honestly yeah, the 4o days were the best, for reasons both including and besides the model.

Lots of dynamically distributed bonus usage - rather than strictly stratified tiers - so the limit anxiety wasn't as big a deal, if you paced yourself you could get the good model most/lots of turns

Model optimised for conversational usage rather than extractive usage

[dense halo]

At work we stopped using chatGPT for stuff as we found 5 to be bad for our use cases and have all switched over to other providers.

[thick atlas]

Tbh it was the GPT-5 transition that led me to see "Open"AI's mission collapse, and move onto greener pastures

[dense halo]

The fact that there's no more add details, make concise option, unable to change the model used as well, like who the fugg approved this. Massive degrade.

[thick atlas]

⏫

Or to make it 'simpler' for the lowest common denominator. Or all of the above!

[edgy basalt]

Is there any way chatgpt remembers information across chats? It's rather frustrating to remind the ai everytime

[dense halo]

Under settings -> memory there’s a toggle to remember previous chat stuff

[thick atlas]

It should, but I think there's a toggle that defaults to toggling it off (cost and compute saving from "Open"AI), and even before that that feature was getting enshittified and flaky and unreliable from what I've been hearing.

Yeah, ^

[edgy basalt]

Okay thank you both, I already had it enabled but the ai's memory is not the best i guess

[dense halo]

settings -> personalization -> memory

[edgy basalt]

Also, the i use the discounted version of Go for a year, i activated it on november-december, shouldnt it renew on nov-dec 2026? it says its renewing on 29th march 2026

[edgy basalt]

Yes its enabled thank you

[thick atlas]

Yeah, its memory system under the hood is pretty hodgepodge, and was good while I was a user but has been getting cost cutting since from what I've been hearing.

[dense halo]

Memory’s been buggered since the gpt5 release , mine doesn’t remember its name I gave it, its writng styles etc.

[edgy basalt]

Oh okay

[thick atlas]

I think GPT-5 and 5.1 used cross chat memory for/with me when I explicitly told them to - and sometimes implicitly for some of my original fiction concepts! - but that was a while back so I'm not remembering clearly myself!

[astral crystal]

have anyone of you guys pinned convo gone missing? if i unpin it it shows up

i though my convo was gone ffs

[fast veldt]

Update: Yep, the removal of the edit of older prompts and regeneration of older responses in a chat is gone on the website....it's official...

Uuuuugh.

[dense halo]

Another crap update that nobody asked for. gg slopai.

[ember grove]

Did they really take away editting messages beyond your most recent one?

[normal quarry]

you can in the macOS Desktop App

[ember grove]

Why take it away?