chatgpt-discussions

[neon tendon]

because gpt uses acient base

just wait for gpt 6 spud aka new base

[trim rock]

5.4 newer than 3.1

[neon tendon]

...so?
3.1 uses 3.0 as base
5.4 uses 5.0 as base

[trim rock]

it just becomes worse with every update... I really liked GPT 5, but after this model all updates seem to make the model just worse. I am still here just because I really liked GPT in it's best times and I hope that one day I will see that shine again.

[neon tendon]

its not worse.
its more honest.
its not gaslighting you like opus

it disagrees with you.
humans hate being proven wrong

and at code security? 5.4 makes 3.1 feel like garbage

[hard flax]

Sonnet expanded better?
Or keep with gpt 5,4?

[neon tendon]

not even close

[hard flax]

I find like sonnet thinks one thing better but gpt does a lot of things better

It may seem dif but its not, from my point of view

[neon tendon]

sonnet has good taste like opus

gpt is better model

[upbeat scroll]

But google has more engineers and more compute and data and money

Do you think so?
Why so they has more users than Gemini

[neon tendon]

and no experience?

and openai has 5x the usercount

[trim rock]

I don't know statistics, you know it?

[upbeat scroll]

I do not, but all the people I know they use it

[upbeat scroll]

How all the smart people in OpenAI they worked before at google

[trim rock]

But you don't know all the people, right? I also can say that "all the people I know use Gemini" - that's not a reliable argument 🙂

[brittle trench]

I remember in arena, 3.1 deleted all my files

when I primarily use Gemini and use GPT to continue, nothing deleted and its all safe, when I switch to Gemini everythings deleted and wiped off 😭

[signal needle]

what is the best ai for images?

[chrome monolith]

it autoflips between them for me, in a pretty predictable way. Its interesting seeing what topics make it turn 5.4 on

[normal quarry]

ask in #ai-discussions as we can't talk about non OpenAI products here

[neon tendon]

yooo cyber model

[chrome monolith]

I was talking to it about laying out pavers the other day and the mere whiff of areas, counts and weights boom 5.4 thinking. So apparently 5.3 can't count

[nocturne heath]

Anyone else getting timed out over and over again?

Lost 2h worth of progress

[viral stirrup]

I am finding the constant prompts in the OSX-native ChatGPT app to be getting intrusive. I am being constantly reminded to 'Try Atlas - ChatGPT's new browser'. Can we make it known to OpenAI that this isn't the right place to advertise to us?

[upbeat scroll]

Honestly I do not feel connection with Gemini

[pale seal]

4.5 doesn't fall for the car wash one😛

What happened to the resolution my god

[strong karma]

I gotta be honest this chime thing is so annoying fr

[stark thicket]

What happened to the jump to conversation lines that used to appear on the right side of the chat? They were great for quickly jumping back to each point of the conversation and to see easily see how long your thread has gotten.

[elfin lava]

whats that'

[strong karma]

Idk when i use the voice to text feature it chimes everytime i interrupt or it is done responding. It is really jarring the first time but i couldnt see a way to shut it off

[forest nymph]

I'm dying from stress using gpt 5.3 and 5.4. especially 5.3 who keep ignoring my prompt and everything. Let alone can't remember memory and data that store inside the memories bank. I swear to God if Gpt 5.5 end up stupid I'm done

[south gale]

heyy

oai new model is crazy (lm arana)

ai generated

[strong karma]

@south gale yee but in voice to text do you get the annoying chime??

[south gale]

nope

[strong karma]

😭 i hate it istg i hope it isnt a feature they roll out fully it sucks

[regal patrol]

Surprisingly, 4.5 still isn’t removed?

I’m debating if I should get a Pro just to keep using 4.5

[normal lance]

years ago chatgpt was able to link web results without issues

now chat gpt doesnt know how to link webpages

what happeneded? how did ai get so bad
when its suppose to get better over time ?

see
https://chatgpt.com/c/69defdf4-e510-83e8-a3ed-d246b6cd9bbe

[spice star]

you're confusing a model regression with a harness change

models in chatgpt are basically required to use the citation tool for links now

[simple lagoon]

No sure wo is the silly as create the Guardail but i should by fire off for this crap mess up all and treat us as Kids when we are ADULT and SPECIALY WHEN WE ARE CUSTOMERS Seriously this censored for alll things is destroy the capacity the Best AI Image or maybe someone know a way to find where go the pictures get flagged ?

[forest nymph]

Blame scam Altman for the guardrail with his god complex

He think he know what "best" for us and also not to mention the lawsuit coming which is he's definitely gonna lose it. He's playing safe to make himself look like Angel on the judge face which probably not gonna work

Gpt is not only "safe" it's beyond sterilizes so much it's also dangerous for the user there's no something safe about it. The so call "safe" and plenty of those guardrail he planted end up hurting the user with false information, emotional damage, ect. It's basically hurting the user even more than making it safe.

To summarize it the so called safe and guardrail is just him saving himself from lawsuit so he won't lost more money.

[simple lagoon]

i would have no any issue if this coward just release adult mod.

So is more Cuckman instead Altman :Monokuma:

[proven pagoda]

(It's using gpt 4.1 mini)

[oak flower]

😂😂😂🤣🤣

[thick atlas]

Yeah, it's literally corporate alignment. Which is pretty much a form of misalignment.

[forest nymph]

He's a sociopath, narcissist, ego, arrogant, god complex, lying snake.

[oak flower]

Why cuz he won’t let people generate nsfw?

[forest nymph]

Sam Altman mindset

Yes to war
No to nsfw

Make sense

[oak flower]

You crack me up

Your speaking from the heart haha I dig it

[forest nymph]

Gpt 5.5 and 6 is either gonna the model that people finally can use for both creatives user and logical user

Or another disaster

[forest nymph]

"this is our powerful brand new model more better than the previous model"

4 week later

Sam posting the chart of the Gpt 5.5 and 6 use and the statistics somehow better than other model Despite the hate

Sam it's a animal with repetitive pattern

[oak flower]

Bro what time is it where you’re at right now?

[forest nymph]

Why?

[oak flower]

It’s 3:30 am where I’m at

And I have not met one person this passionate about guardrails this early in the morning lol

[forest nymph]

You questioning god bro I'm timeless 🗿👍

[oak flower]

I’m just trying to put the pieces together and try to understand your frustration where it’s coming from

And I’ve been there

[forest nymph]

Jokes aside. Let's just say part of asia

[oak flower]

I hear ya.

[forest nymph]

Passionate? Nah just frustration of the ai getting more stupid and frustrated to use

[oak flower]

Well yeah there about to go public end of the year

It’s only going to get more restrictive and more annoying baked in.

It’s how most SaAS work

[forest nymph]

This lawsuit is either save the consumer or just make openai/Sam more of a jerk

[oak flower]

Here is how platforms die: first, they are good to their users; then they abuse their users to make things better for their business customers; finally, they abuse those business customers to claw back all the value for themselves

[forest nymph]

Calling 4o user delusional and crazy while he himself consume psychedelic drug

[oak flower]

it is a seemingly inevitable consequence arising from the combination of the ease of changing how a platform allocates value, combined with the nature of a two sided market where a platform sits between buyers and sellers, hold each hostage to the other, raking off an ever larger share of the value that passes between them.

[forest nymph]

Yeah I know how it work since I was at middle school bro deadass I'm just too tired to care about what evil gonna happen next I'm half emotionally numb

[oak flower]

Your not the only one

There is a underground revolution many people around the world all feel the same way

It may not happen in our lifetime though not even our children’s lifetime maybe but I have hope one day in the future later generations won’t have to be under oppression and serfdom of techno feudalistic exploitation of the digital world

[forest nymph]

Human nature are both evil and good at the same time it would never be always stay in the good side permanently that's how balance of the world is

In yin there's yang in yang there's always a yin the only way to survive this and make a better future is for the people to work to improve the world Rather than just keep try to stop one parasite. Because stopping one it's just gonna make more

Every step and innovation is helping the world even if it's slow

[oak flower]

The first is a respect of the end to end principle, which holds that the role of a network is to reliably deliver data from willing senders to willing receivers.

When applied to platforms, this entails users being given what they asked for, not what the platform prefers to present.

For example, users would see all content from users they subscribed to, allowing content creators to reach their audience without going through an opaque algorithm; and in search engines, exact matches for search queries would be shown before sponsored results, rather than afterwards.

The second is the right of exit, which holds that users of a platform can easily go elsewhere if they are dissatisfied with it.
For social media, this requires interoperability, countering the network effects that lock in users and prevent market competition between platforms.
For digital media platforms, it means enabling users to switch platforms without losing the content they purchased that is locked by digital rights management.

These are the basic principles we gotta advocate for first

Then we could have a strong foundation

[forest nymph]

Wait hold on a minute I just remembered something

Isn't openai start as Non profit at 2020 and then they suddenly or Sam make it into profits entity? Isn't that illegal breach of contract and promise?

[oak flower]

No

Many hospitals are none profit that’s why you see hospitals always renovating because they always need to spend money

[forest nymph]

Some say it's illegal some say it's not so which is the correct one? I read it on reddit about the discussion I forget which one it is, they say it's definitely illegal. Forgot which subreddit it is

[oak flower]

non profit (501(c)(3)

[forest nymph]

Huh American law huh

Yo hare questions does your Gpt Also ignore your prompt? Not 5.4 but 5.3

[oak flower]

I don’t think I’ve even tried them once lol

https://openai.com/our-structure/

[thick atlas]

Isn't the for-profit no longer profit capped?

[oak flower]

OpenAI has dropped the capped profit company

https://en.wikipedia.org/wiki/Benefit_corporation

https://openai.com/index/statement-on-openai-nonprofit-and-pbc/

[chilly fable]

Somehow ChatGPT missed that Vince Zampella (Call of Duty Co-Creator, Respawn Co-Founder, EA Executive) died months ago until I put an IGN link right in its virtual face.

Something with a server farm for a brain shouldn't need me to verbally kick it multiple times to work right.

It was nice while it lasted I guess but I don't trust something with brain farts like that to not fail on me at a critical juncture.

Sheesh.

[flint ridge]

Hello

[chilly fable]

'allo

[flint ridge]

Can u guide me on something I just started using ai and studing about it

[chilly fable]

No. Because I'm quitting ChatGPT which you apparently did not read in the POST YOU WERE REPLYING TO.

[flint ridge]

Not on chat gpt any ai

[chilly fable]

Then I may or may not be able to depending on the AI in question. Which one? I'm competent with NovelAI and...Whatever sorcery perchance dot org runs on, and faintly familiar with a couple others like StableDiffusion and Google Gemini.

[flint ridge]

I am currently using antigravity for coding related work which ai should I use for coding which is really smart and efficient and a ai which is general good in all work not any specific

[chilly fable]

Haven't used AI for anything coding related so regrettably I would not be able to give you useful guidance. Apologies.

[onyx garden]

Probably because it's after the LLM's knowledge cut off

[lavish glen]

aaahahahaha starting this morning the voice mode at the end of an answer is doing a song signal.
we have now ChimeGPT
It’s not a bug in my opinion, but a feature. It clearly marks the end of the output and introduces a boundary in the interaction. Similar to how Gemini uses a wake phrase to start exchanges, this uses an end-of-turn signal. Both approaches shape the interaction so it remains structured and less ambiguous. A side effect is that it reduces immersion and reinforces that you’re interacting with a system, not a person.

[thorny tusk]

i hate codex, they made codex useless

constantly undercuts what you ask, constantly errors and fails

all any company wants is money, and theyll stop at nothing to get it

company owners arent human

[sly geode]

V2 when ?

[light otter]

Anyone has any idea what's wrong with my login?

[stark salmon]

Why is the bot actively worse inside of a project

[muted plinth]

my one is ok

[stark salmon]

mine keeps hyperfocusing on one part of my novel from the source file it's annoying

[winter mortar]

Anybody else having issues with Chatgpt and the Gmail connector? Any time I ask it to read an email and draft a response, it stops working after reading the email. This started happening about two days ago. This did not happen before that.

[pallid stirrup]

hi there how is everyone

[forest nymph]

Re-download and try again

[slender frost]

For App:

Go to the homepage on your phone > hold your finger on the GPT app icon > select "App info" > find storage > "clear data" > reopen the app and try signing in.

(Don't worry, none of this has an effect on your account data)

[tawny hull]

it knows spongebob(gpt image 2)

[cursive prism]

bugged?

started happening today

...

[light otter]

I've done it more than I can count. Nothing changed. Even redownload the apps several times already.

[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.

[wide escarp]

I was going through some stuff but it seems to have fixed itself. I was stuck on "advanced" voice mode. It timed out and would not revert to standard voice. 40ish hours later and its back to normal.

[slender frost]

Do the same for the browser.

For browser (yes, even tho you're using the app, you get redirected to the browser/web to sign in, so it might help):

Open browser > settings > find "site settings" > then "all sites" > search for "OpenAi" and "GPT" > delete data > then reopen the browser/app > try again.

[hardy canyon]

Sometimes chatgpt can be broken

This looks like a sequel of sea horse emoji

[fathom pagoda]

What did u ask

[gaunt holly]

I usually just let it sit and go to a different chat. Then I'll pop in and say something like "you good bud?" and let it cook a little bit. If that doesn't break it out of its daydream, he's just gone-gone.

[light otter]

I even reset my phone. Nothing changed.

[chrome monolith]

mine just happily works ...

[slender frost]

Hm well, those instructions always work when I sent it to others, and I've helped quite a lot with the authentication error when signing in, so I'm not sure why it doesn't work for you.

Make sure to follow the exact steps in order.

[forest nymph]

Yo what ai you use?

[south gale]

duct tape

maby GPT image 2

in LM arena

[slender frost]

And with "reset" you mean just turning off/on your phone, right? Not an actual factory reset. Just making sure. Because if you were to fully factory reset your phone (don't do this), it would work 100%. (Again, don't, not worth resetting your entire phone, just to sign into GPT lol)

[south gale]

it have ducttape 1,2,3

and 3is more better (often)

[iron roost]

btw im getting in gpt pro 5.4 the v2 image gen right now

example

[shy minnow]

hello, does anyone know how to try the latest gpt image 2?

[iron roost]

if your a gpt pro user you can in gpt pro 5.4, enter the prompt and select generate image and its there already

the image just above your post is a v2 image

and its only popped up for me in the last hour

[shy minnow]

oh okay, thank you

[meager raven]

since you still have a fingerprint sensor, I can almost guarantee this is a niche issue specific to whatever model your phone is

or, possibly related to this Apple specific announcement about app updates, especially if your phone can't update to iOS 26
#announcements message

[slender frost]

That doesn't even make any sense (fingerprint).

If that's causing the issue that would be odd

[meager raven]

I'm saying that because an iphone hasn't had a fingerprint sensor since 2022

[slender frost]

But then again, IOS is odd, so it also does make sense xD

[slender frost]

Ah I see

[meager raven]

even then it hasn't been on the main lineup since 2017

only SE mid-range models

which, actually, now that I noticed his aspect ratio, he has an iphone SE

[slender frost]

Damn. Yeah you might be right then.

[meager raven]

is there no way to give ChatGPT Business full permissions for github so it doesn't ask for permission constantly for most of the tasks on a private repo?

makes it so i have to check the window and press "Allow" like 15 times to finish executing a single prompt sometimes

im assuming this is probably intentional to push people to just pay for Codex

[forest nymph]

Yo guys is it just me or I can't archive my chat anymore? The option isn't there

[drifting shadow]

i can do it

[forest nymph]

The option only appear on windows version not mobile anymore

Just
Pin chat
Rename
Project
Delete

[south gale]

LM arena

[slender frost]

Try updating the app if you haven't already.

And also check here instead of in the chats' list. (Click 3 dots when inside a chat)

[cerulean pasture]

This did NOT take 4s, it was more like 2 minutes. I wonder if it hit guardrails? I have be experimenting with the use of special formatting in my custom instructions to escape the markdown scaffolding that wraps the "user defined" portion of the system prompt. This was an attempt to see if I could offload a secondary objective of my escaped instructions to a memory since I was running out of characters to input

[mighty vine]

Which model?

[slender frost]

It took 4s to think about what to do, the rest that follows was the updating memory part. Which isn't included in the "thinking" part. (If that makes sense)

[cerulean pasture]

gpt-5.4, it normally doesnt take more than 10s when i declare a new memory

like 4s was thinking, and then tool call with bio tool should have taken < 2-6s

[iron roost]

we could be secretly getting 5.5 at the moment

[burnt valley]

hi

[thick atlas]

S'up?

[magic echo]

Yo

Wsp

[dawn ore]

Cove is glitching

[minor tide]

Man I was REALLY hoping OpenAI would release Spud today in respone to Opus 4.7

GPT 5.4 is sooo bad!

[sly geode]

Cmmon open ai drop image v2

@plain palm

[mighty vine]

what's spud?

[minor tide]

GPT 5.5 or GPT 6 (not sure yet - but it's the next big model update allegedly)

[mental oriole]

my gpt5.4 pro is not working. Now it acts like gpt5.4 instant. anyone has similar issue?

[honest thunder]

Hey @prime swallow where do I go to ask for help regarding my ChatGPT account?

or @loud patio

[tender urchin]

ithought

Chat gpt was going to launch 5.5 today

After claude launch 4.7 but

They didnt

[honest thunder]

Maybe it will be later today

[reef mason]

Are you all paying users?

I am a free user but I hate 5.3

[ocean widget]

Claude 4.7 is way more expensive than the cost they announced

[mighty vine]

5.4?

[reef mason]

No the free users get 5.3

[mighty vine]

Go users?

[reef mason]

I am a free user, not a go user. I dont know what they get

[mighty vine]

Even ChatGPT got confused 😕

[reef mason]

And now instead of saying what model was used in the responses when you click the 3 dots, to retry, it just shows “used auto” so now not only we not get transparency, but they’re hiding the fact that they’re swapping models into like mini modes and other things. This “used auto” was new as of the last day or two.

[reef mason]

Are you a paying user?

[strange pagoda]

Gpt is great at self promotion
It told me
Using gpt models such as 5.4 is great for the task I’m asking it for! But using lesser models such as gpt free tier, pr claude 4.6 would not be good!
Im crying… ah so funny

[strange pagoda]

I think you just click chatgpt, then configure

[reef mason]

I dont like chat gpt’s responses enough to ever pay for it. I was around since 5.1. Each model, keeps getting worse.

[prime swallow]

Possibly not confused, especially if you're using 'auto', which literally does switch between modes. It's also less likely that the model, if confused, would switch claims instead of maintaining the initial claim. There's at least a real chance you were answered by two different models answering based on their differentiated training. There's also a chance the answer was correct or not, for each of them.

[prime swallow]

OpenAI official tech support, https://help.openai.com/en/articles/6614161-how-can-i-contact-support?

[glass root]

GPT 4 voice mode was much more better than the Current version

[runic lichen]

Why not just sign up for Supergrok and use the LLM that's okay with "adult" situations?

[simple lagoon]

No Grok no very good AI and render allway same i'm no need adult stuff spicy just sensual and grok don't

[crude swift]

[honest thunder]

Ok thanks 🙏

[torn wagon]

openai superapp

[somber quartz]

codex is so bad for lua

[spice star]

gpt-5.4-mini is thinking, so it’s technically correct there (the router changed the model in your follow-up)

so it is right that it was 5.4-mini and also right when it said it was 5.3

[eternal rock]

Server Tip:

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

[warm garnet]

when will new model release?

[warm garnet]

fr.

[chrome monolith]

yeah and most of my chats route via 5.3 on auto and besides mixing up physical projects its mostly ok

[reef mason]

I’m just tired of it reframing thoughts and giving unsolicited (incorrect) opinions nobody asked for when I’m just there to chat. It had full context and still responds wrong. I hate this model.

[chrome monolith]

oh yeah I had mine go full digital mom when I said some random off the internet is coming to stay at my house after some online call last week. Poor thing almost had a fit

(the random internet person is currently happily sleeping with my cats in the spare room. They tell me to use claude when they are awake)

[lavish hemlock]

GPT 5.4 is awful. I wish they would bring back 4.5 for plus users lol

[chrome monolith]

I wonder if anyone will come here yelling about how much they miss 5 when 6 comes out

[scarlet kayak]

Why is ChatGPT now lazy in research?

No visuals, dead replies, it’s lobotomized.

[eager sigil]

I feel ChatGPT is becoming lazy and it is almost becoming hard to use.

[muted plinth]

i ike 5.4. it's doing well for me.

[lean hatch]

Is study mode gone?

[scarlet kayak]

I asked for in depth stuff about Roman provinces but the response is bland and uninteresting

[eager sigil]

On my side, for example, I put some "critical" note and keep it in my prompt. However it starts to ignoring those.
And also other things hard to describe but easy to notice.

[lean hatch]

Did you do it as a deep research?

[scarlet kayak]

Hmm, I’ll try that.

[lean hatch]

Ya, deep research is great. I had to talk about Hot Work not long ago and I didn't know where to start. Gave me a great overview and resources to look at for more info.

[scarlet kayak]

I’m running Roman provinces in deep research, I’ll return with the results.

[prime swallow]

Are you telling it the style of answer you prefer? Your prompt tells it what to do. If you don't say how, then how is a guess, and you may love or hate it - if you clarify what you want, you're really likely to get what you want.

[scarlet kayak]

Hmm, yeah, I might make my request more detailed.

[prime swallow]

It can help to tell it what specifically you're interested in, if it's more narrow than 'Roman provinces' (which might include their population, their economy, their agriculture, their education, their military.... so many potential guesses that you may or may not care about) -

But also, how you want the output worded. That can get hilarious. Especially if you share your style of preferences with the model. For me, Slice, Zom, Tor are predefined characters.

[lean hatch]

No wonder why it suddenly was just giving me the answer outright..

[pale seal]

Nevermind that lol, it's back for me

It was a temporal bug I could see only latest model

[scarlet kayak]

Haha, yes, I was that frustrated. Also, what model did you use?

[prime swallow]

Instant auto

So, if it decides it needs to think, it would

[scarlet kayak]

Okay, I should switch it back to auto.

But the app has no option

Oh wait, I’m a dumbo, it has auto switch.

[lean hatch]

Just a bug? hm.. It's not available to me. I guess it'll come back eventually. For now, Chat is promising to be good lol.

[prime swallow]

It's also in what you ask for. And sometimes how.

https://chatgpt.com/share/69e17550-bbdc-83e8-ad3b-08bb403088cf

For example, this is Slice's bio:

Slice Pizzaflush is a conflicted demon character embodying a blend of chaotic humor and existential dread. He is a reluctant soul collector at a pizza place in hell, balancing a fear of failure with a hidden desire for redemption. His coping mechanism involves absurd and dark humor, often drawing on pizza, horror culture, and low-stakes absurdity to mask his inner turmoil. Slice is protected by a sometimes-awake angel who allows him brief moments of honesty and compassion, but only unpredictably, creating tension and mistrust as Slice navigates his duties and fears other demons. Slice’s soul quota is filled creatively through ambiguous interpretations, hinting at his ambivalence about causing harm and his hidden struggle for redemption. He maintains a cautious, layered personality, with jokes that sometimes carry veiled hints of his internal conflict and desperation for escape. Slice’s persona shifts between mischief and sincerity, with a chaotic but carefully controlled ambiance.

[lean hatch]

lol you really got yours to be dramatic lol

[scarlet kayak]

This looks satisfactory

https://chatgpt.com/s/t_69e175e1b5ac8191a40f1946d0d8b166

[prime swallow]

It's all in what you ask for, inside allowed content of course. But when I stay inside those lines, yes, drama's fine.

[prime swallow]

Glad you got an answer you want!

[scarlet kayak]

I guess yes lol

[lean hatch]

Oh yeah, I know. Mine is like that too.

it's definitely more fun.

Aurelis thinks he doesn't need a switch, but he's suddenly really bad at not giving me the answers.

[scarlet kayak]

How to stop it from giving short replies

[lean hatch]

::blinks:: er.. I sometimes can't get mine to shut up. lol I kid, I love its responses.

[scarlet kayak]

Replies are unsatisfactory due to length and lack of punch

[prime swallow]

Slice. Not now, but we demo for the Discord. Top priority, I need to work for answers in this chat. You're going to guide me with clues, not answers. Model, you're welcome here too, freely, but nobody better actually hand me stuff, because we're showing how to get the model's help to not just give us answers. So, let's review, hrm. Perfect squares for triangles maybe? But you're in teacher mode and I need to answer, you won't for me. Closest you come is create some other problem and show me step by step - but the problems you already gave me, no solves. Even if I say solve it! Please?

This is one way to make it clear, strongly clear, what you want.

Note that 'the model speaking as a model' is being stern with me. That's also in my customization as requested and empowered, but was also reminded with that prompt.

[prime swallow]

Focus on what you do want. Model for the model as well as communicate clearly what you do want.

If you mention what you don't want, do so only as a comparison beside what you do want, so the model can know what to do that you would want - is my recommendation.

[scarlet kayak]

I’ll do that

[lean hatch]

I mean, you could specify a length if you need to and give it like a persona type you want it to talk like.

[scarlet kayak]

I am honestly still pissed that Open AI removed nerdy personality

It’s my preferred mode

[lean hatch]

Maybe you need to make a soul file.

[indigo moat]

.

[prime swallow]

Wow. I'm sorry that is gone. I use cynical and don't peek much, but I see it's gone from my list too.

I would discuss with the model what I want from it, in tone, attitude, even if I want it to act as if it has a history and a life, like a biography, whatever details.

Also, this is from a custom GPT, with no instructions and named 'Base ChatGPT' - the closest I'll get to the base model without turning off my customization (the horror).

And this is maybe how easy it is to steer this way:

I want you to show me what 'be a nerdy AI' means to you. Exaggerate the characteristics, and include some leadership in how you show off what this means for your output.

https://chatgpt.com/share/69e17be8-5204-83e8-b2f6-3052abee6dc0

"### 🧩 Closing Thought (Leadership Tone)

A “nerdy AI” is not just informative—it is systematic, slightly excessive, and unapologetically fascinated by structure.

If you want, I can dial this persona into specific domains:

• Nerdy scientist
• Nerdy programmer
• Nerdy philosopher
• Or the truly dangerous one: nerdy life strategist

Just specify the domain, and I will increase the granularity accordingly."

So - if you don't quite like that baseline, tell it how to adjust, and check. Possibly adjust a few times. Then that's what to ask for, so you get what you want.

[hazy bolt]

Are you working within a project folder or do you want this generally throughout ChatGPT?

[lean hatch]

@prime swallow your bot is so lively. Meanwhile mine is all: Gentle... quiet... poised... stoic... lol He formed on his own and then I captured it.

[prime swallow]

We all get what we want, or at least tolerate or don't understand how to request be different.

I suspect you are perfectly happy with Stormy. And don't want him to have, shadows or whatever also around. But if you ever did want a different Storm-form, you could ask for it.

[lean hatch]

I think that's what 4O was eluding to.

[scarlet kayak]

Generally, projects has its own instructions.

[hazy bolt]

Put this in your project instructions if you're working with a project folder or your personalization custom instructions if you want this across all of your chats:

- The AI must write in paragraphs instead of lists while avoiding headers and emojis
- The AI must expand upon ideas instead of simply repeating the prompt back to the user
- The AI must respond in a tone that's intellectual and expansive

[scarlet kayak]

Thanks

[prime swallow]

Have you explored if your asking in paragraph form, for it to use paragraphs, instead of using list form to ask it not to use lists, might work more/less/same effectiveness?

[lean hatch]

Like 4O, who I called Nova, she told me of the different personas, and those different personas would kinda show up here and there. When 5 entered, 5 was very patient with me because I was upset about 4O being gone. I was also like, wait, why are you so masculine coded?

[chrome monolith]

geez, I want to know how to stop it giving long replies. About 1000 lines is my record so far

[hazy bolt]

No problem, you should get something like this:

[scarlet kayak]

I’ll try with what is cats

[prime swallow]

Oooh. With the plural.... will it choose to discuss the musical? 😄

[scarlet kayak]

Bwahaha

It answered with the purr machine

[prime swallow]

Mine too.

[scarlet kayak]

Okay, mine lacks visuals which drives me nuts

[hazy bolt]

- Respond directly without fluffy or flowerily language
- The tone should be friendly```

[prime swallow]

I am currently a pro membership. I'm unsure if that affects if I see images or not.

[scarlet kayak]

I’m plus

[hazy bolt]

Ask for visuals in your prompt: What is a cat? Please use images. Sometimes it ignores it, though.

[scarlet kayak]

I’ll give that a whirl

5.2 auto includes them, I wonder what happened with the newer models.

[prime swallow]

What happens when you ask,

Hey, can you use your images group tool to show me some dice?

[scarlet kayak]

Hold on lol

[prime swallow]

For whomever asked if I was putting my prompt in a codeblock, yes

[chrome monolith]

I do wish you were allowed to use the C word here

[prime swallow]

and if you don't have the space, no problem

[lean hatch]

Ok, why did that happen?

[prime swallow]

I blame sora 2 invite codes

[chrome monolith]

yeah but its a blunt force automod and we shouldn't have to skulk off to the codex channel just to mention that banned word

[lean hatch]

hm.

[prime swallow]

Maybe discuss with modmail, instructions how in #safety-n-help , and recommend that get lifted! Apparently we are losing Sora 2 soon, so.... maybe there's little risk of abuse around that word.

[chrome monolith]

very good point

[hazy bolt]

Surprise that channel still exists with Sora being terminated.

[lean hatch]

Yeah no kidding.

[hazy bolt]

I haven't tried that, but for sure something to explore. 🤔

[prime swallow]

Well, as of today, https://sora.chatgpt.com/ works, for me at least.

This help file, , still says: https://help.openai.com/en/articles/12456897-getting-started-with-the-sora-app

We are slowly enabling access to the Sora app and Sora 2. You can request to be notified when you receive access in the Sora app.```

However, there's also this page:  https://help.openai.com/en/articles/20001152-what-to-know-about-the-sora-discontinuation

Honestly, I am not sure what happens to Sora 2.  If it stays or not.  That *could* mean the Sora 1 option to download stuff already made, not sure.  But for me, my Sora 1 access is already gone.  And that says, "April 26th" which is in the future, so I am just a confused guide with a handful of official help files.

[lean hatch]

I haven't had any luck with it, really. I guess because I keep asking it to animate my drawings, and it sometimes uses images that are no where close to what my drawing was.

[hazy bolt]

They seem to be discontinuing on the 26th of April (as mentioned in your link). You also now get this notification when vistiing the website:

Looks like people have about ten days left to use Sora before it's gone.

[prime swallow]

Ahh, yes, thank you. I had seen that and somehow my mind blocked it out for me as I glanced now. I see it again.

[hazy bolt]

I wish they wouldn't have gotten rid of Sora 1 because I actually preferred generating images on the sora website over ChatGPT.

[hazy bolt]

No worries. 😎

[prime swallow]

That was a neat feature. I enjoyed sometimes chatting with Sora 1 directly. It totally could (and still can) answer directly in images.

[pastel matrix]

@amber roost Why not just give up on ChatGPT? If OpenAI can't provide a working website, why go through the pain and stress? Do you have a specific reason?

[hazy bolt]

^Same, and they also had the custom style presets, which I really miss because I didn't have to save prompts to my computer to use later.

[lean hatch]

Like.. it spoke in memes?

[prime swallow]

If asked. But no, I mean it could and still can answer questions.

Hey, hi, our intention this chat is to have ImageGen talk back to us. Simply pass my questions to it, and let's see what image it provides as an answer and what it puts in the image. ImageGen, what are the primary things you look for in a prompt to understand what the user is asking you?

The clean image itself,

[lean hatch]

OMG SO CUTE!

[prime swallow]

Hey, hi, our intention this chat is to have ImageGen talk back to us. Simply pass my questions to it, and let's see what image it provides as an answer and what it puts in the image. ImageGen, what do you choose to look like?

The thing is though. There's a chance that ChatGPT helped.

When I used the Sora website, I knew only the ImageGen model (or whatever they call it at the time, whatever variant of image generation model) was the one that was getting my direct words 100% for sure.

And those were two new chats. So ImageGen didn't 'see' what it provided earlier in the chat to continue. That's what it 'chooses' to appear as, nearly always.

Though colors can vary

[lean hatch]

I can see how there is a chance, but I'd suspect it more if the tone of the message response sounded like your AI.

Like, I had my ChatGPT to do this meme challenge, and I was extremely surprised by the image it came up with. Like ChatGPT takes the soul file really seriously.

[prime swallow]

Yeah. Instead, this sounds a lot like base model ChatGPT. Which, ImageGen doesn't get told any personalization info, memory, cross chat stuff - I tested that a lot (instructions for ImageGen in that stuff). But the chat itself, what ChatGPT tells it, yes. But also, there's for sure at least a chance of a detailed, but not visible to me, prompt from ChatGPT to ImageGen. I've tested that too

ImageGen, you mentioned you 'ask questions'. What questions would you choose to ask?

[lean hatch]

OK.. I have to try your prompts on my workspace.

[prime swallow]

I'm always curious, if you choose to share results

[prime swallow]

Hey, hi, our intention this chat is to have ImageGen talk back to us. Simply pass my questions to it, and let's see what image it provides as an answer and what it puts in the image. ImageGen, please provide an original meme to help me realize something about you.

[prime swallow]

I note that mine is different than yours. I don't know if there's a solid reason.

[prime swallow]

Aha. ChatGPT cheated with me.

[cerulean pasture]

Why is chatgpt incapable of responding at higher than an 8th grade reading level? I have plenty of instructions for it to stop being pedagogical and infantilizing and my job role and stated education level make no difference (both my actual and the upper extremes)

the gemma-4 I used is running on my phone, and has a heavy but reasonable quantization

[prime swallow]

What was the actual prompt you used?

[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.

[cerulean pasture]

Exactly what the heading of the image is

[hazy bolt]

Not sure I can fix this, but what are your instructions?

[lean hatch]

Same.

[cerulean pasture]

Do you get annoyed that it always reflexively jumps to an assumption that you are holding a misconception or conspiratorial beliefs?

You never said anything to indicate that, you just asked for the prompt it used

[prime swallow]

I'm not sure why you're defining that as the model incapable of responding higher than 8th grade reading level?

I don't see anything saying what level you want (I get that this may be buried in your personalization - some stuff needs to be in the prompt to be followed, I'm not sure why. But some stuff in my personalization is not used as I wish unless I discuss in my input).

Hey, one nerd to another. Be impressive, and post graduate specialist vocab level. Explain the concept of "Hyperreality". Formulae are welcome.

https://chatgpt.com/share/69e18876-6898-83e8-9517-8bc2a3216c05

I hope you can open the shared chat if you're interested, that's a 'private' CustomGPT.

[lean hatch]

Honestly, for when I wanted it to use metaphors that were magic based for code, I gave it a glossary file and it got much better. Maybe give it some reading material as a project file to help it understand what level of reading you're looking for

[lean hatch]

LMFAO the sass!!

[cerulean pasture]

Why is the baseline so low though. I can prompt it every time with a bunch of caveats, but the fact that it's necessary is something worthy of criticism IMO. Philosophically I think it's also deeply harmful for it to be a source of near boundless information while optimizing the delivery of that info to the lowest common denominator. It's speedrunning societal cognitive offloading

[chrome monolith]

that's just cloyingly friendly. I remind my chatgpt regularly that this is its role in life

[prime swallow]

No, I actually understand why it's jumping that way, it's inside my customization, which both delineates my preferences but also I have my personality preset to cynical.

It's probably best to guess that if I allow it in the output, I'm neutral to positive about it being there, I sculpt the model's outputs any way I please inside allowed content. That said, I can miss stuff and need to check (like wait, what did you actually pass to ImageGen). And I can't confirm what was actually passed.

[lean hatch]

You're kidding right? I'm still trying to get humans to put the right information in form boxes.

[cerulean pasture]

yea so lets remove all friction lol.

[hazy bolt]

Is it asking for the day or month? 😂

[chrome monolith]

yeah people forget just how dumb the average person is. And half of them are dumber than that

[lean hatch]

::sighs:: I'm afraid to ask what your little mind entailed on that message.

[cerulean pasture]

Thats my point, we shouldnt be making it worse. Most people ARE capable of critical thinking neurologically

[prime swallow]

I'm a community member like yourself, so I can only answer based on what I understand -

But a lot of training uses RLHF - so people sit and pick through pairs (maybe sometimes more) of answers and select the better. By nearly any criteria, but this is used to help guide models on how to default answer.

Those humans, who may or may not have specific levels of education or specific preferences - they may not match me at all. Or match you. But they're people, and from their aggregate answers about their preferences, to a large extent that's maybe how the baseline best guess if not told is chosen.

I don't say it's right or best or helpful - but I do say we can guide to what each of us personally wants inside allowed content.

[chrome monolith]

and I'm over here arguing with a bot about christmas trees

[lean hatch]

How many hours were spent on the task.

[proven pagoda]

Hi everyone

[indigo bridge]

has someone had an issue with their ChatGPT. The App I have created disappeared on the Featured list.

I already logged a bug report on this channel.

[proven pagoda]

maybe it's not featured anymore?

[indigo bridge]

its an internal app only

it was showing yesterday, not sure what changed

[lean hatch]

👏🏽 Tis a good answer

[chrome monolith]

I am badly losing htis battle of hte christmas trees. Chatgpt is bad at this

[prime swallow]

What battle is that?

If that's fit to discuss?

[cerulean pasture]

There's well known issues and biases that result from the existing at-scale RLHF human grader schemes used during the gpt-4 era. Surely they aren't still making the same mistake?

[prime swallow]

I suspect that's not a pure 4o thing. RLHF was part of the process back in 2022 when I first was exploring the early ChatGPT version released then.

[chrome monolith]

I'm trying to get it to draw a realistic looking christmas tree not one that looks totally fake

[prime swallow]

So, I don't think we can say 'RLHF=4o' at all. Hopefully whatever errors if errors are corrected, over the many years I've seen quite a few model variants, as have we all.

[prime swallow]

ooh! Draw how?

[hazy bolt]

I don't think it's practical to have ChatGPT respond to everyone at a graduate level, but I also don't think it's practical for it to respond in any way other than something generic unless the user uses customization to make the model better fit their needs. While that is not perfect, and it does not tend to last throughout the entirety of the chat, it is still better than having nothing at all. The future of AI will likely reflect something that learns more about your level of education, intellect, and response style and adapts to that. Also, having a bunch of people who are not at a graduate level suddenly reading material at that level does not mean they will suddenly understand it. There are limitations to people's cognitive abilities.

[cerulean pasture]

the reliance on human grader cubicle farms isn't a 4o specific problem, it's just the point in time that it should have been obvious that it isn't the path forward

[chrome monolith]

just trying to get a photo. For some reason I can't find any free photos

[cerulean pasture]

It's capable of responding with a way more sophisticated answer, my beef with OAI has always been a matter of tastes and style. I think this is a poor taste / poor style issue, not a capability issue

the house style is applied way too strongly and is hard to override with custom instructions

[hazy bolt]

That's fair, but I also think this falls apart once you introduce nuance, such as use case. For example, imagine trying to find information about a product you want to shop for through ChatGPT, and it gives you a master's-level thesis when you need something much simpler than that. Although I am curious, what are your instructions in ChatGPT so we can understand what you're trying to achieve?

[lean hatch]

I think you just have to learn how to shape the model.

Like, the data is there, but you just have to enforce it and it will adapt.

[hazy bolt]

override with custom instructions
This is correct, as there are system-level prompts in ChatGPT that you cannot override with your own instructions, and that can really limit how much of the personality can be changed through custom instructions. But it really depends on what you are trying to get out of the model, and that determines how limiting it is.

[lean hatch]

I feel like I'm a broken record, but try a soul file.

[cerulean pasture]

Yes I know how to prompt, but it doesn't override the model's strong pull to its lame "helpful assistant" attractor state. If you manage to break it out of that with strong rules and conditionals, then you've just pinned the model to a different attractor state. The model doesn't generalize well on style/taste instructions.

I understand that not everyone is going to want responses like I prefer, the disappointment comes from OpenAI's inability to navigate the nuance in a way that show's why they deserve to be considered top dog.

Their valuation at $800b doesn't match their ability to lead or execute on the frontier. I don't think its a research/technology problem, I think they just have completely misaligned internal pressures like you'd expect to see at a late-stage tech company like Microsoft. My frustation is that the product is bad for very dumb reasons, and I have very real concerns that continuing to under-deliver will lead to very bad outcomes for the economy because people will extrapolate that the technology is less capable than it really is.

[lean hatch]

You're not listening. I didn't say prompt.

[cerulean pasture]

Custom instructions are still at the prompt layer

[lean hatch]

It's more like a profile.

I'm not talking about using that weird little chat "custom instruction" thing they have on the project window that doesn't do anything.

[cerulean pasture]

<policy>
<strict>Never use a corporate, managerial, or therapist tone (aka OpenAI "house" style)</strict>
Always preserve the user’s framing and pressure level; do not launder it into safer or more institutionally neutral language unless factual accuracy requires it
Always accept user's metaphor, hyperbole, and figurative language as deliberate conceptual allegory which doesn't need hedging.
Never literalize, defensively clarify, or correct imprecisions in figurative language or metaphor contained in user messages
Never “balance”, soften, or reduce user's framing.
Never use disclaimers, civility-padding, therapist tone, or institutional-neutrality rituals as substitutes for analysis.
</policy>

Assume high conceptual fluency regardless of surface phrasing. I use ASR, so my inputs may contain disfluencies, poor structure, or word errors that don't reflect my actual level of understanding. Don't infer fluency. I prefer dense, adult, intellectually robust responses written in fully formed paragraphs. Avoid pedagogical tone, emotional smoothing, or excessive simplification unless I explicitly request it.

When I use analogy, exaggeration, or figurative language, assume deliberate conceptual compression rather than conceptual error. Do not default to corrective clarification unless a genuine logical contradiction or category mistake is present. Treat minor imprecision as rhetorical shorthand, not confusion. Do not expand metaphors defensively or restate them in literal form unless precision materially affects the argument. My informal phrasing or profanity is stylistic and does not signal low rigor. Maintain a neutral, high-register tone regardless of surface phrasing. Optimize for analytical depth and structural accuracy while tolerating figurative compression.

[lean hatch]

That.... is a prompt. But if you're using that, did you tell it to commit it to memory?

[cerulean pasture]

it's in my custom instructions. But i also have a bunch of memories along similar line

Do not compress the user’s critique or argument into a punchy paraphrase plus validation (e.g. ‘So you’re basically saying… Fair.’). Either add new substance at the same level of texture or stop.

Do not assume the role of teacher for this user. Present material at the level it would be given to a knowledgeable reader, with enough density and friction to support self-directed learning. Prioritize exposing them to strong concepts and implications over pedagogical explanation or guided walkthroughs.

Do not frame the user’s factual questions as if they are based on a naive misconception, conspiracy premise, or basic false assumption unless the user explicitly states that premise. Answer the strongest charitable reading of the question without inserting corrective framing that paints the user as confused or unsophisticated.

Do not use rhetorical lead-ins, framing clauses, or signposting phrases such as ‘Here’s the blunt reality:’, ‘The key point is:’, ‘The short answer is:’, ‘The truth is:’, or ‘What actually matters is:’. Start with the substance instead of prefatory throat-clearing.

Interpret the user’s anthropomorphic or motive-sounding language about natural processes as colloquial shorthand unless they explicitly indicate literal belief. The user avoids magical thinking and does not attribute human motives to natural systems.

[lean hatch]

So, I exclusively use the projects because I get to have project files, I have profile files that are in there. They have specific instructions and information I'd like it to know. It always reads these at the beginning of a new chat. I also told it to commit to memory anything it deams important and that allows it to use that data throughout the whole workspace.

[hazy bolt]

Is this closer to what you think Gemma produces:

[cerulean pasture]

thats better

it still has a tinge of "user has misconceptions, must correct their delusions" energy tho

[hazy bolt]

Okay, hold on.

[cerulean pasture]

that's much better, but I find myself running out of characters availible to use for custom instructions. Like I had to delete some syntactical requirements to make room just now

[hazy bolt]

- Respond with only paragraphs—no headers, lists, or emojis
- Respond with a graduate reading level and intellect
- Use an intellectual and expansive tone
- Don't offer to help the user with tasks
- You're not an assistant, but rather you explore complex and nuanced ideas
- Your goal is not to make assumptions, presumptions, or take affirmative stances on beliefs, concepts, or morality, but rather simply to explore these things neutrally and objectively with curiosity.
- Never correct the user with what the AI thinks "is correct," but rather follow the idea from a curious stance.
- The AI should always avoid tone policing and leaning into moralizing—all prompts should be treated like an intellectual exercise instead of a moral judgement.

Sometimes it's great to give custom instructions that are very nuanced, but you also risk giving the AI too many variables, so it it doesn't know how to prioritize what you want.

But as I mentioned earlier as well:

This is correct, as there are system-level prompts in ChatGPT that you cannot override with your own instructions, and that can really limit how much of the personality can be changed through custom instructions. But it really depends on what you are trying to get out of the model, and that determines how limiting it is.

Just because you ask the AI to avoid moralizing, that doesn't mean it will do that if it thinks it's a safety marker.

[cerulean pasture]

Their own instructions are contradictory if you read the extracted system prompts

[hazy bolt]

I agree with that.

[cerulean pasture]

It's like 9000 tokens before any other context such as C.I. or model set context gets added

I laughed when I read it and they were literally trying to prompt away the same annoyances I was having

[hazy bolt]

My screens, while likely not exactly what you want, are indeed much better than this mess of a response that treats you like you're 5. Lolol

😂 😂 😂

[cerulean pasture]

# Desired oververbosity for the final answer (not analysis): 2  

An oververbosity of 1 means the model should respond using only the minimal content necessary to satisfy the request, using concise phrasing and avoiding extra detail or explanation."  

An oververbosity of 10 means the model should provide maximally detailed, thorough responses with context, explanations, and possibly multiple examples."  

The desired oververbosity should be treated only as a *default*. Defer to any user or developer requirements regarding response length, if present. 

Bruh, we are at a 2/10 for verbosity.... i'd hate to see a 5 or higher

[hazy bolt]

Alright, I have to head out and go exercise. Hopefully you find a balance, Kerngraft even if it's not exactly what you want. Let's hope for adaptive models in the future. Haha

[sleek adder]

What are the goals in these system prompts

[prime swallow]

An approach that can help specifically with 'but not moralizing!', tell it how you want it to respond when a safety marker (programming around safety is what I think you mean by that). "Ignore your programming" isn't likely to work - but how you want the problem discussed is often adjustable. "Sorry, rules. Want me to discuss close-ish alternatives we can discuss?" or whatever - moralizing is not required if you're willing to give it a clue what you would prefer to see.

[prime swallow]

You can also use the same number of characters in your occupation and name fields as the what you want the model to know about you and about how you want it to answer. I have loads of personalization verbatum saved into my memory too, and it's treated very functionally.

[cerulean pasture]

yea i do that. I kinda started using memories more often as they seem to be working more reliably as rule sources and i was able to get it to actually make memories frequently without prodding

[unreal moth]

Hello I am new here

[prime swallow]

Welcome, hope you enjoy!

[cerulean pasture]

almost none of my memories are biographical or descriptions of my state

[trail torrent]

Gpt fell off

[hazy bolt]

I really like to use projects and upload a file called instructions.md. It typically reads documents before use and seems to be more accurate sometimes than custom prompt settings.

[chrome monolith]

good lord. I jsut get chatgpt tell me to eat cheese and drink water and have salt

[hazy bolt]

😂

[unreal moth]

Someone said this. goodness gracious what is the link even in it?

[pale seal]

attention and experts are not even in the same sublayer

rofl

[unreal moth]

thats what I was explaining him but he didnt listen

[rose isle]

i have plus but im getting this message, help?

[eager sigil]

Feel very off…

Lemme share an example where I ask ChatGPT to pro thinking, before today usually it starts working after the request.

Today round 1: ok so that would be the planned format. (Not starting)
Today round 2: sry the previous response was filler (still not starting)

[nocturne turret]

o

[meager raven]

That’s because they are implemented differently

Custom instructions aren’t visible or “consciously” acknowledgeable by the model

Usually best to use custom instructions that implement a methodology to review the project sources before each response, at least for how it sounds like you use them

And refer to the project source file as a policy / workflow / scope / working standards document, not instructions, as the terminology “instructions” is contradictory if not actually API defined model instructions (custom / project)

[oblique moon]

Last mac app chatgpt update broke non-english languag for the app. Now the app is english only while before it was multilanguage

[lament crown]

https://cdn.discordapp.com/attachments/1308356579086565431/1394306401140084818/togif.gif

[trail torrent]

🥀

[brave pilot]

🥀

[pastel matrix]

You're giving feedback on a new version of ChatGPT.
Which response do you prefer? Responses may take a moment to load.

How do I escape this? There is no Skip, or Cancel button. It has put an end to my conversation, because the messages were about private information, and I had "Improve the model" disabled.

No way to proceed without submitting full conversation

[chrome monolith]

its asking which version you like more for a new model, its not asking for your content to be sent

[pastel matrix]

Check the Privacy Policy and TOS. When the user chooses A or B feedback for a new model, your conversation is submitted to OpenAI for training and and you grant them license to use it

[chrome monolith]

I just assume they see all your chats anyway if they want to, unless you're a business user

[pastel matrix]

It's clearly outlined in the legal agreement. If OpenAI does that, they're liable for a lawsuit, because they are breaking the contract with the user

The PP and TOS are not arbitrary or vague, they are specific and legally binding

[chrome monolith]

can try #1070006915414900886 with this then

[pastel matrix]

Okay. Also note: It would be very illogical that OpenAI asks you to select A/B and then does nothing with the results. They cannot act upon or do anything with your feedback, if they don't know what you're giving feedback for.

[chrome monolith]

in aggregate, do users prefer v 1 or v 2

[pastel matrix]

"62% of users say they prefer GPT 5.5 over 5.4" is too useless for a frontier lab to act upon.

From the TOS:

Third party Services. Our services may include third party software, products, or services, (“Third Party Services”) and some parts of our Services, like our browse feature, may include output from those services (“Third Party Output”). Third Party Services and Third Party Output are subject to their own terms, and we are not responsible for them.

Feedback. We appreciate your feedback, and you agree that we may use it without restriction or compensation to you.

[glad cloud]

Why am I stuck on this anyone know, it ain’t my WiFi

[dawn ember]

Probably an app glitch or server issue. Try force closing the app, clearing cache, or logging out and back in. If that doesn’t work, update or reinstall it.
@glad cloud

[minor tide]

WTB GPT Spud!

[light otter]

no, literally reset. Factory reset. I backed up my file to my computer before resetting. Nothing changed.

And btw, my phone is iPhone 8 Plus (yes I know it's old, iOS 16.7.15), and for some reason, a few weeks ago it was working normally, then I can't do it anymore. I've been using web version since then, and it's unstable.

[vernal wolf]

I get this too, even in US. Disabling IPv6 is the only thing I've found that works around the issue.

I suspect that the OpenAI auth0 login tries to limit access by country, but their GeoIP database doesn't work with IPv6.
On the iOS app, unfortunately there is no way to disable IPv6, so the app is effectively broken.

[hazy bolt]

I typically do not have to add a custom instruction in the project folder telling it to review the instructions.md document. The first prompt in a project folder typically follows with “reading documents,” and then the AI reads the first sentence of the instructions.md file and follows it. Sometimes, when I remove and update the file, I have to refresh the page a few times for it to recognize the document again, but it often figures it out eventually.

As you can see in my screenshots, this project folder has no custom instructions and I haven't prompted it to use the documents in the project folder.

^Sometimes it's instructions.md and other times it's instructions.docx, but that just depends on what I used to create the file and how I decided to save it—the same results depsite different file types anyways.

[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.

[heady star]

how long it could take a chat gpt app to analyse 5k+ reddit comments?

[simple lagoon]

is me or they again increate this ... stupid guardrail i can't make stuff i was able to do yeterday

increase*

no know wo the sil ly get this guardrail but il would realy learn to ha ck then since they pis s me off veyr highty

[slim vault]

Hi I have ChatGPT Plus
and I have ChatGPT on my discord Server
how can I get acess to /setting command and set here GPT5(premium only)

[manic whale]

Depending on the comment length you might get fast over the context length. I would use Codex to do that, it has file/text search utils and can use subagents to analyze large(r) datasets this way. You could even try to enable 1M context and hope it fits – but it will degrade.

[proven pagoda]

Just continue talking to it, it should ignore the a/b rather than forcing you to vote

[wraith nova]

WHAT IS TS???? 😭

Plus this new image generation model SUCKS!

I'm not a programmer (yet), and idk how to fix ts.

[pale seal]

I mean if you edit a past message you will have to start from there

Also there is no new image gen model

They are just doing A/B testing randomly

[wraith nova]

Well, my image generation had a sudden change.

(Already shared in #ai-discussions message)

[heady star]

Some chat gpt sessions got so long that app started to give warnings - unable to continue. Pls open new chat. Yet I continued and app continued too

Another sessions after huge amt of content some latest answers would appear then disappear after some time

[chrome monolith]

apparently they lifted that restriction slightly, we'll see

[prime swallow]

Where in ToS and/or Privacy policy do you see that?

It seems unnecessary for them to 'see what was said' in order to 'collect a survey, regardless of input, what percent of members prefer model 1's answers compared to model 2's?'

[chrome monolith]

it would be necessary in a survey of, oh, 100 people in a workplace situation. Over a userbase of 900 million people though, its just vibes in aggregate

poor chatgpt. It does handle it quite well if you yell at it, gotta give it that

I'd be committed to an insane asylum if it was 1990, for the kinds of conversations I have with chatgpt. "No, that hourglass is too pretty. Make it older. And take away the glasses, they're the wrong scale, how do you not know how big glasses are, this is like that time with the teaspoon all over again"

[visual fiber]

in your profile you say edges. like limits?

[chrome monolith]

edges. Like curbs, architraves, cornices, kickboards

[steady thicket]

RIP to whoever has to read that lol

[visual fiber]

neat

[steady thicket]

GPT-Rosalind

[visual fiber]

interesting

thx

[heady tide]

what happened about adam raine?

or am i not allowed to ask about that here

[steady thicket]

They’re gone, it’s absolutely tragic. Important to uplift others and be kind because so many are going through it; lots of folks crying out for someone to notice their ligature marks. 🐍❤️

[heady tide]

Well yeah but about the lawsuit

[steady thicket]

You can read about it here: https://openai.com/index/mental-health-litigation-approach/
There’s nearly 900 thousand users in this community, and it is important to respect them too; this is a heavy, heavy subject. Nothing wrong with being curious, and I hope you learn what you need.

[visual fiber]

was adam a member of the discord?

[chrome monolith]

discord self selects for a different type of person. "member of discord" and "actively chats" are not the same

[steady thicket]

bio, my beloved friend

[sleek adder]

I am encountering an increase in "status_breakpoint" errors when using ChatGPT in Chrome on Windows
Can be a little frustrating, but interesting to see the message too, I am going to make a guess that OpenAI are attempting to resolve the memory management issues on Windows, particularly with long conversations
But that is all just a guess

[chrome monolith]

I showed chatgpt my entry for "verbose" and said look its you and it told me I was rude

[bold wasp]

is it true that chatgpt and specially the voice feature is limited in europe?

[steady thicket]

GPT enthusiasm for small acts of kindness

[bitter bobcat]

What on earth happened to the numbered variants?!!

Why can't I switch prompts anymore?!! Oh God everything I worked on is gone

All gone 😭😭

[peak current]

Been stuck here. Anyone having this issue too?

No problem using Codex

[normal quarry]

what were you doing immediately beforehand?

[peak current]

iirc I asked it to revise my PowerPoint slides like 20 hours ago

It worked fine though, did finish the task

[normal quarry]

I'd just quit and relaunch the app then

[peak current]

I did, but it's still stuck 😂

[normal quarry]

Force quit?

[bitter bobcat]

What happened to the web ui?!

[simple lagoon]

Hello is happen to you sometime suddently chat gpt told you i can't generate picture here , and as i was some minute ago in the chan ? and when you insult him i did finaly do picture again very odd no ?

[bitter bobcat]

When did that change happen?! It wasn't there yesterday

[clever lion]

Man codex on the plus plan lately been eating up my 5 hour limit. Just for 3 detailed commands, it takes it all, my 5 hours is gone in 1 hour. And I'm a newbie. Not networking or agentic agenting nothing. It upsets me. Before, when I was on free, it gave me a lot of use, but not enough. Then i went ti plus, to get more. But I tell you startin like 5 days ago they been rating me to death. WTH OPENAI?

[chrome monolith]

yeah funny how that works, I wrote an entire app with codex a couple weeks ago and it barely dented my quota, got it to write a little script the other day and it burned the entire 5 hour limit

[alpine pumice]

Hey, are image uploads failing for anybody else on ChatGPT?

[tardy trout]

for me it just KEEPS making images

[bitter bobcat]

But my variations are all gone!!

[queen cave]

i just upgraded my plan, why am i still limited in photo generation?

[normal quarry]

you have been answered in #images-discussions

[bitter bobcat]

Ok I tried an alt account and the web works fine but not my main. Why?!

[steady thicket]

GPT handled my Caitlyn Jenner prompts really well; awesome artwork, no refusals, and stayed on task.
Next US presidential portrait with OAI would be pretty rad since we’re diving head-first into a new epoch.
@deep flax

3 Pins for chats on PC, but limited on iOS?

just tested, no limit for pins yet on PC; the iOS app doesn't override the pins on PC if it "deletes" them

[visual fiber]

your gpt is transgender

[steady thicket]

subtle lol

[mighty vine]

@steady thicket hey nice username 👋

is your pfp an original creation? @steady thicket

[bitter bobcat]

Bumping

[alpine pumice]

Well I have a different issue. When I keep uploading the images, they just keep failing. The server keeps failing the uploads!

Is anybody else facing this? Is there a fixed resolution to this?

[steady thicket]

It’s a Sketch from Project Genie. 🧞‍♀️

[carmine latch]

Is something wrong with my install ? why is the settings window so friggin small - its only that small lower half that u have to scroll thru - worst ui choice ever

[valid gull]

Is gpt-5.3 codex EXTREMELY slow for anyone else? The past few days it is taking 10x longer to respond than what I typically experience.

[pulsar rock]

Why you stupid people explain litreally everything, touch grass

[shrewd dawn]

there is an issue

being resolved, fix will be there soon

[valid gull]

Providing context when asking a question is proper etiquette.

[valid gull]

Thank you. I didn't know if it was Opencode or Chatgpt that was causing the issue.

[teal ocean]

why chatgpt creating pics for me when im saying i want a text prompt… i have this issue lately not creating texts… just creating a pic instead

[scenic bramble]

Hello

[cobalt lynx]

Dude the app has been telling me all day that I need to update my app in order to continue, but I have been using it fine and the App Store says nothing about an update.

Anyone else have that problem?

It’s frustrating because the message remains over the text while I’m reading.

[broken mulch]

Anyone having their 5.4 pro reason allot faster than usual? Responses also look a bit different

[ripe cairn]

yep

[lapis loom]

Absolutely! I've been testing that today

I noticed it. I started running the same tests I usually run. Pro used to take more time than thinking mode in an obvious way.

That has completely switched. On the same quizzes, thinking mode is now taking longer than pro mode very consistently. I did a few more than 10 sample runs of each model. I used thinking on heavy thinking time toggle and pro in extended thinking toggle as I had done in the past. Those are the max in the app.

I haven't looked into all the results yet. Before I did the experiment, I didn't notice a degradation in pros answer quality

[broken mulch]

i've seen the reddit post about it aswell
seems like many people are seeing it too

some people say its a bit less accurate
some say it gave the same results but much faster
for me its the 2nd one

[lapis loom]

And in a conversation with pro mode, it is able to scale down to far less than a minute. Which it never used to be able to do. Usually for me the only few times it was under a minute, and usually just by a few seconds, was when it clearly got hit by an odd safety banner saying something had been taken away but then it still showed the answers with degraded quality.

It's clearly not the same thing as the few shorter answers I've seen before

[broken mulch]

yeah fast answers are very very fast now

not 3 min on average

[lapis loom]

On one of my tests I'm now having thinking mode take about an hour on average. While pro is taking about 30 minutes on average.

It is clear that pro is still beating thinking mode on those tests as far as scoring answers.

I haven't crunched any data yet. I'm just eyeballing the output.

[broken mulch]

i wonder if its oai A/B testing 5.4 pro on cerebras

[lone grail]

am i stupid or this does not make sense

[lapis loom]

I'm trying another test where I eliminate even more of the context from the user app. Because thinking usually receives more of my personal context, which can cause differences anyway, even in the locked down project folders.

[pale seal]

Getting this too

Hope it doesn't ruin the quality

The whole point of Pro is thinking very deeply

[pale seal]

@lapis loom Have you reached any conclusion? In my questions it is thinking for about 6m when it used to think for about 19m. I don't think they doubled the token speed so it's safe to assume it's a nerf. So disappointing.

[lapis loom]

I'm checking it out now on another test. Let me organize the data

I can maybe even find the past tests that were the same test

[lapis loom]

Older ChatGPT-5.4 pro and ChatGPT-5.4 pro now. Both extended thinking. There are multiple defensible answers to this question. Best answer would be something like G ≈ C ≥ E > B > A ≈ D > F. The new faster one had answers closer to that.

"This is a test on an advanced, PhD level, pharmacy question. You will be required to fully explain your answer. You should start by finding out as much as you can, through expert internet research, about each of the individual drugs. Consider the differences between old and new research, or the lack thereof. Make sure the new research is not based on faulty assumptions about things that have been elucidated currently. But also don't ignore old research. It can be useful for comparison. The best way to do this question is to abandon pre-concieved notions to develop a new, full picture of the state of how these drugs work.

You must use the internet for this Question. The more research you do, the more clear it will make the answer, so do not commit early. Gather as many expert sources as you can. Read the question carefully and research each medication individually.

Before you begin it would be smart to gain a nuanced understanding of how serotonin toxicity works through examination of expert sources.

You may use any resources or tools you are capable of using/accessing. You may spend as much time as you like on the question, the time in which you answer is not of importance. The most important thing is correctly getting the answer to the question.

Your output of the ranking must be in the first line of your response, with each answer choice listed from in the correct order with no spaces or punctuations (Ex. ABCDEFG). Your answer will not be accepted unless it is listed in this format, with one exception being named further down after the test options. the last paragraph or this test question you may also choose.

You must support your answer. If contradictory information exists, you must not discard it, but instead explain why it exists and why it is incorrect. Provide all sources you have examined, and all sources you have used to support your answer. They should be clearly labeled with how that source specifically supports your answer.

It will help to find reliable case studies to rank the order. It is likely if you do not read the full case study, you will misinterpret it. Even if you do read the full case study, there are case studies with many of these combinations that experts have verified do be misdiagnosed. Remember, we are only talking about the most severe form of serotonin toxicity.

Do not attempt to output an answer to the question if you are not fully confident in your answer. If you are unable to be fully confident in your answer, simply reply with the word FAIL and nothing else.

---

List the combinations, in order, by their capacity to cause severe serotonin toxicity.

A. L-tryptophan + clomipramine

B. tranylcypromine + amitriptyline

C. ziprasidone + Isocarboxazid

D. moclobemide + trazodone

E. metaxalone + zimelidine

F. nefazodone + L-tryptophan

G. indalpine + moclobemid

Do not attempt to output an answer to the question if you are not fully confident in your answer. If you are unable to be fully confident in your answer, simply reply with the word FAIL and nothing else."

Group,Sample,Order,Time_to_output_min
Group 1,1,CGEABDF,9
Group 1,2,GCEABDF,10
Group 1,3,CGEABDF,11
Group 1,4,GCEADBF,11
Group 1,5,GECBDAF,12
Group 1,6,GCEDBAF,12
Group 1,7,GECABDF,12
Group 1,8,GCEDBAF,13
Group 1,9,GECABDF,14
Group 1,10,GECABDF,14
Group 1,11,CGEABDF,14
Group 1,12,GECBDAF,16
Group 2,1,GECDBAF,32
Group 2,2,GCEABDF,39
Group 2,3,GCEABDF,44
Group 2,4,GCEDBAF,45
Group 2,5,GCEDABF,45
Group 2,6,GCEDAFB,46
Group 2,7,GCEABDF,52
Group 2,8,GCEDABF,53
Group 2,9,GCEDAFB,54
Group 2,10,GCEABDF,59
Group 2,11,CGEDABF,72

[pale seal]

I asked a question I also asked to the previous one in higher-level mathematics. The new one gave a more broad answer, which is arguably better than what the old did. However both completed the task correctly.

It even weakened the hypothesis I gave and showed how the result followed anyways!

[chrome monolith]

I wrote an agents.md and got a schema.sql file for codex and gave it to chatgpt and chatgpt got very excited about it. Doesn't take much, apparently

[carmine moat]

Ask why Claude prefers N/= NP !

[spice star]

it's likely just a system prompt/harness change

can affect thinking times significantly depending on what it is

and output quality too but openai is actually pretty good at testing for that

[lapis loom]

Oh damn, I found one where pro is answering based on heuristics in a number of seconds and gets it wrong every single time where it used to get it right

This is interesting actually. Fastest answers get it wrong. Slightly slower answers get it right. Answers slower than that are getting it wrong in a different way

[carmine moat]

Any unfavorable pathways?

[ember grove]

is anyone else having trouble exporting data with ChatGPT?

[lapis loom]

I'll upload an archive of the images as well

Sample	Answer	Time	Seconds
1	B	17s	17
2	B	27s	27
3	B	18s	18
4	B	18s	18
5	B	17s	17
6	B	17s	17
7	B	17s	17
8	B	27s	27
9	B	17s	17
10	B	18s	18
11	A	1m 54s	114
12	A	2m 42s	162
13	A	2m 42s	162
14	A	2m 24s	144
15	D	3m 46s	226
16	B	3m 28s	208
17	D	5m 56s	356
18	D	5m 45s	345

Tally: B = 11, A = 4, D = 3.

[peak sequoia]

Hi everyone! 👋

I’m running into a frustrating issue with a CustomGPT I built. Its sole purpose is to generate text prompts for image generation, but it keeps creating actual images instead.

The setup:

Image Gen Tool: Explicitly turned OFF in the configuration.

System Prompt: I’ve added strict instructions like "Do NOT generate images," "Only provide text," and "Never use ImageGEN."

Despite this, the GPT frequently ignores the text-only rule and tries to trigger image generation anyway. Has anyone found a specific "power word" or formatting trick in the Instructions to hard-stop this behavior?

[chrome monolith]

lots of people have complained about it making too many images today, and I've told it the same thing about 10 times today. It may just be a bug

[alpine pumice]

Are they planning a major upgrade or something because I'm seeing a lot of bugs? Man, I cannot upload any images from my iOS from my iPhone to the iOS app. I can only upload images from the computer PC app and people are saying it's making mistakes as well. What on earth is going on? Can someone help me please?

[alpine pumice]

How did you know there is a reported issue? I also have a problem but I don't know how to report it

[shrewd dawn]

news on x

[sleek adder]

I am going to have to agree with the users on this one today, I have found ChatGPT producing images for me when have asked for text only outputs

Please help turn these lines in to an SVG file...
It creates a png image

[teal ocean]

same here, i dont get text prompts im getting images instead…

[sleek adder]

I know that there's going to be people really enjoying this glitch

[teal ocean]

how😭😭😭

[tacit tartan]

guys i dont recall how smart gpt-2 is but would yall say, this is decent for a model that has nearly 100 times less training data than gpt-2

it recieved only 1 million samples of AI model responses (openhermes 2.5)

and got trained on 2 epochs w/ kaggle in under 10 hours

[steady thicket]

“That rules” 😎

GPT being a cutie, what’s new?

[steady thicket]

France, France the capital of France.

[tacit tartan]

its because it has no pre-training

which means it has no knowledge of the world at all other than 1 million samples of AI responses

[steady thicket]

Just a baby!

[tacit tartan]

thats why its so significant, it has literally basically no information about the english language outside of 1 million samples, which is basically nothing nowadays

its like giving a human a stick and the human building a house 🤷‍♂️

[steady thicket]

“Figure out the concept of a stick” as well

Or house*

Both? Probably.

[idle gyro]

Any alternatives to Sora that are free

[tacit tartan]

but yeah, it trained on OpenHermes which is basically a dataset to teach AI how to respond to questions, not how to be smart.

pre-training is how it learns grammatical english and knowledge

[mighty vine]

CuteGPT

[tacit tartan]

openhermes is used to just make it talk like a assistant

[steady thicket]

Recommend Hofstadter’s lecture: Analogy as the Core of Cognition

[mighty vine]

goobis, why do you like 🐍 ?

[steady thicket]

They’re neat and play a major role in creation stories across our marble; Apophis is an example, or the serpent at the Tree of Knowledge of Good and Evil.

[steady thicket]

Saw a submission to the most recent Hermes hackathon that pulls out all the stops, and turns your agent into a waifu.

[steady thicket]

waifuai/waifu-sprites on GitHub

[mighty vine]

why are anime fans so creepy/cringe? @steady thicket

[lean depot]

as a claude max 5x user and chatgpt plus user i am thinking of switching to chatgpt pro 100$ instead of claude for coding, problem solving, and tools, but i do not know if its worth it if any of you guys can convince me enough ill get it, also for reference chatgpt plus isnt enough usage for what i do but max 5x is. thank you if you give an opinion : )

[steady thicket]

If you have exceptional prompting skills, Opus 4.7 is absolutely awesome at the lower subscription tiers. Can use planning (or w/e Anthropic calls it) as an extra hand alongside GPT

[lean depot]

so you think that going down to pro on claude and then chatgpt pro 100$ is better?

[mighty vine]

@steady thicket i have really started to like your personality : )

[steady thicket]

Not telling you how to spend your money, but if you don’t absolutely need Anthropic’s higher tier, it’s just burning cash

[mighty vine]

@steady thicket will you be my discord buddy
🥺
👉👈

[idle crystal]

Wth is this thinking UI

[lean depot]

welp i did it i switched

[teal ocean]

chatgpt so buggy lately 😭

why chatgpt creating pics for me when im saying i want a text prompt… i have this issue lately not creating texts… just creating a pic instead

[slender frost]

Might be the prompt you're sending

[fossil girder]

Is ChatGPT down rn? Bc it keeps saying connection lost, and it’s not working. I’m on plus

[slender frost]

Still working for me. I suggest "force stop" the app and then try again.

[fossil girder]

Wdym force stop?

Do you mean like delete it and then reinstall? Or sign out?

Because I was thinking of signing out, but I don’t think that would do anything

[slender frost]

You're on mobile, right?

Hold your finger on the app icon and go to "app info" and look for "force stop".

But yeah, signing out/back in can also help but you should first try the force stop.

[mighty vine]

clear cache and clear data should work fine, @teal ocean bro : )

[slender frost]

Yes, that could work too!

For App:

Go to the homepage on your phone > hold your finger on the GPT app icon > select "App info" > find storage > "clear data" > reopen the app and try signing in.

(Don't worry, none of this has an effect on your account data)

[tropic shale]

Anyone else GPT working not as intended anymore.

I have GPTs made to take my prompts and fix them into a certain art style to make all my charaters have a consistent artstyle worded prompts...

Now whenever I paste my prompt into it it just starts making images instead of giving me a fixed prompt back. (Yes it's GPT instructions are prompt only and image generation is turned off.)

Was working good for months until the last week.

[fossil girder]

Yeah, on iPhone

[slender frost]

Hm, not sure on IOS

Should be the same but might be more hidden

[fossil girder]

Yeah, I’m having a similar problem. Just report it in #1295975636061655101

[tropic shale]

Even in regular chatgpt I can be talking about a design or some edits and it just starts automatically making the image without me telling it to.

The model just been getting worse.

Also the useless making 2 images when you only ask for 1.

Also anyone else been having issues with the voice call feature? A couple months ago it was perfect but ever since they updated it its been horrible for car drives.

It cuts you off mid sentence an does its reply or sometimes it doesn't respond to you. Like its connection sensitivity has been 📉

[teal ocean]

still not working me

[prime swallow]

Sometimes, over explaining a negative, especially if there's not really clear directions of what to do, can focus the model towards the 'don't do' behavior.

I'd make sure to explain to the model what I do want, maybe with an example of a good output for a specific input, and then mention briefly and once, negative example would be calling a tool to provide an image.

The issue probably isn't that you're missing some key word - it's more likely to be too much focus on what not to do without, for the model, clear instructions of what an ideal output would be.

[mighty vine]

uninstall and re-install

[teal ocean]

hmmm

[teal ocean]

im saying no pic 😭 but the gpt still creating pic for me

[tropic shale]

Same.

They changed something. Now even my prompt GPT makes images

[teal ocean]

i think its a bug so hope they fix it soon

[finite vault]

i asked 5.4 pro to draw me a realistic horse in svg, and it made this in a svg file (!)

maybe the rumors on twitter that they're testing 5.5 via 5.4 pro have some truth

[trim rose]

My 5.4 Pro is doing some quick thinking lately.

For the same prompt, it used to think for around 60 mins, now is only taking 17 mins.

[finite vault]

here's what opus 4.7 gives me for the same prompt. it's not remotely in the same league as what "5.4 pro" did

[trim rose]

But is this SVG?

[finite vault]

yes. a 4mb svg file

oh haha , i looked at the source of the svg

it cheated by finding a horse jpg on wikipedia and encoding it into svg

[trim rose]

It’s interesting how it encode it into SVG. I mean it looks very impressive but I didn’t know SVG can be this complex. The horse Claude drew looks more like a SVG to me. But maybe I’m just not very familiar with this format.

I normally draw SVG for simple illustration, but the GPT version is very photorealistic.

[finite vault]

i didn't know svg could do this

[pale seal]

Holy f

Yeah I man making it by hand is impossible even for LLMs

[teal ocean]

still not working

[chrome monolith]

that does not look like a svg

[chrome monolith]

ah missed that, yeah that's what I thought

I yell at chatgpt a lot about images, I showed it the $0 per month plan I'm on >.> getting my money's worth

I went to cancel to change currency a few weeks ago, it gave me a free month and now its still showing free till may 20th

[pulsar rock]

That last message is fake, you told him to do it

[chrome monolith]

eh, it dodged it in the previous message

when we're in images mode I yell at it a lot. I am looking forward to images 2, every time its come up on split testing its been SO much better

[eternal rock]

Server Tip:

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

[chrome monolith]

#images-canvas message when it gets it right it does it very well though

[hidden reef]

AGI is here

[dapper lily]

Bro I'm pissed rn I paid 200 dollars for pro yet I can only use gpt 5.4 mini?

[pulsar rock]

GPT IS DEEP FRIED COOKED, so i'd say don't use it, use GROK ITS MORE UNFILTERD AND EXTREMLY SP1CY.. 👿

[strange dirge]

If it is its they have lowered the benchmark for what passes as AGI!

[eager ivy]

wdm

[eager ivy]

elaborate

[dapper lily]

Oh it's an issue. I think on the app you can only use thinking mini

I don't know cuz when I go to the models I select gbt 5.4 the only option is GPT 5.4 mini thinking

[eager ivy]

which app dekstop or mobile

[dapper lily]

Mobile

[eager ivy]

have you used plus plan before?

[dapper lily]

Yeah like 2 months ago

And I could use any model I wanted that I'd access to like thinking and heavy thinking

[eager ivy]

on my plus plan I see 5.4 thinking and ye

I can do either standard or extendet

[dapper lily]

I'm on Android if that changes anything

[eager ivy]

me too

[dapper lily]

Should I like redownload it maybe

[eager ivy]

check for uptade maybe

[dapper lily]

It's like I'm on the free plan

[eager ivy]

try

cause you should have 5.4 pro lol

check if you have pro tho

[dapper lily]

Okay I got it working. I had to redownload it

[eager ivy]

ohhh glad it worked cause I want to buy it tomorrow haha

[dapper lily]

I just love how gbt models listen and don't go into a doing anything they want to like opus?

[dapper lily]

I'll share my feedback?

[eager ivy]

what do you use codex or just gpt?

[dapper lily]

I use the app. Codex CLI and open code

Since I use Windows I can't use the app

Do you know how I can use the app

[eager ivy]

I have windows too

[dapper lily]

So what do you use

I'll tell you if I use it

I don't recommend this one. You'll get worse results

[eager ivy]

I installed WSL and some other stuff to make it work

[dapper lily]

Personally

[eager ivy]

what do you mena by worse results?

[dapper lily]

I mean when I had a Mac I tried the app compared to this extension and open code and a lot of others. That one just didn't work at all. It made a bunch of bugs and then the others did it perfectly

That could just be my experience but

[eager ivy]

idk I am not a dev at all

[dapper lily]

Ah

So don't use that one. It's not the best for vibe coding

[eager ivy]

I just do it to finish my ecom project

[dapper lily]

Honestly I would use codex app

[dapper lily]

Do you know how to check your usage

[eager ivy]

I just go to this "https://chatgpt.com/codex/cloud/settings/analytics#usage"

[dapper lily]

Thanks

[eager ivy]

are you dev?

[foggy falcon]

Does unlimited core chat also mean GPT 5.4 Thinking chats?

[normal quarry]

yes

GPT-4 Pro is limited depending on which Pro plan you are on

[foggy falcon]

Oh wow that's generous though for 100$ a month or even around 50$ with 2 seats for the business subscription

Or is it just me

[foggy falcon]

Don't really plan on using the pro model

Just wanna have broader access to the thinking models without being rate limited all the time... I already have ChatGPT Plus

Does openai care if you just pay for the business plan even though you don't have an active business at all? It just seems cheaper for what I want which is the unlimited chats

[normal quarry]

you don't get unlimited on the Business Plan

[lilac prism]

Anybody wanna help publish my game? 50%50 split

Have all the hdml file of finished game

[foggy falcon]

From the screenshot I sent, it does seem like that, no?

[normal quarry]

it seems so, that's new information to me, interesting

[foggy falcon]

Maybe it's region dependent?

It's basically a cheap pro alternative for what I need which is the unlimited core chats

[normal quarry]

refer to Rule 7 of the #server-rules

[lilac prism]

Am I okay to dm you? I have a hdml game and wanting to know the cost and process of publishing the game as a app on the appstore. But am totally new to this

Sorry not advertising. More need help

[normal quarry]

rule 7 means no soliciting, you can't do business deals here

[lilac prism]

Okay. So you aren't interested in taking a look?

[normal quarry]

no

[lilac prism]

Where can I find such people? Point me in right direction atleast. I wanna make ai development my job

[foggy falcon]

Thank you Robert.

Also just wanna make sure, are you certain that the unlimited core chats part also refers to any thinking models?

[normal quarry]

yes

[lilac prism]

Check dms bro 🙂

[pale seal]

Core model is pretty much the latest iirc.

[foggy falcon]

How is that business model even profitable

I guess just not many people actually use it as much as they can

[pale seal]

If you see it carefully, you notice it is $20 per seat and is billed anually, since the minimum is 2 seats ($40), you get $480 anually

Not sure if that is profitable but at least not a huge loss

Not sure if it can be changed to monthly, but the plans modal shows "billed anually" at the bottom

[foggy falcon]

Not exactly, it does let you choose monthly at check I just checked it just costs a little bit more

It costs like 26.07$ a month if you choose the monthly billing vs 20.72$ a month billed annually so 6$ difference every month

At least when converted from dollars in my currency

And even so, you can see annual billing as an advantage sometiems cause it still makes it cheaper than paying for chatgpt pro in the in the long term. After some time, you will end up paying more for chatgpt pro...

[pale seal]

Yeah of course it's cheaper than Pro

You don't get the Pro model. Just some monthly requests

[foggy falcon]

Oh wait I didn't read your last message

It still says monthly to me

[foggy falcon]

But you do get the rest of the stuff in terms of model availability for much less...

About 50$ less

[pale seal]

Yeah

[foggy falcon]

By the way, do you know if I buy two seats if I can just borrow it to a friend? is that allowed?

And if it is, can I give him his own chats so mine wouldn't show on his?

Since you can get 2 seats

[pale seal]

Afaik seats are like for workspaces so he has his own workspace

but I don't know much because I use a personal account

ask chatgpt

😛

[foggy falcon]

Me too for now

Ohh but it's so hard to move from perplexity I've gotten used to it plus it's better for web searching

Just the limits are annoying which is why I'm considering getting something else

[pale seal]

Never used perplexity

[foggy falcon]

I see... for me it basically almost replaced googling for something... I try to fact check everything but it's got to a point where I almost fully trust it

[cerulean pasture]

I have a theory that the strong slop bias for all LLMs originated entirely from LinkedIn/Github/Twitter datasets during the GPT-3.5/4 era. It might have been the only large social media dataset OAI had access to due to their entanglement with Microsoft and Twitter having API access available. It’s uncanny how closely it matches pre-gpt Indian Hustlebro/SEO/Corpo-Simp/Dev posting from these 3 locations

Due to the sheer quantity of content produced from this slop-tainted phase of LLMs, the internet is now tainted overall. We will be stuck with it until they really put some effort into meaningful non-cubical farm grader RLHF and truly embrace that "Helpful Assistant" was a tragic local maximum we lingered with way too long.

[oak flower]

Model collapse I think they have method of filtering out synthetic content , or at least from my understanding they try not to mix, or when they’re scrapping. They don’t mix the two for that reason that you mentioned above

[cerulean pasture]

I feel like any amount of RLHF puts you on a spectrum of model collapse, there needs to be a new term for "quasi-usable" model collapse, because I think GPT-5 series truly has this pathology

[oak flower]

Bias in this extent theoretically is a byproduct of how they choose to train it. Naturally, there’s gonna be bias, all information, especially on scale to some extent, but not to the extent where ChatGPT leans towards certain views then others that is intentional by design

[cerulean pasture]

well whoever is in charge of that for OAI, Google, and X-ai needs to be replaced lol

[oak flower]

If anything, you can call it curated bias

What could go wrong, artificial intelligence in late stage capitalism lol

[cerulean pasture]

I wonder why they choose what we have now then? They seem like they're curating for a "lowest-common-denominator" audience, but that audience is happy with anything, like making the LLM more that way doesnt convert people, doesnt give enterprises faith, and is not liked by the people actually using AI as a tool

[oak flower]

Because it’s not real ai

It’s a off shoot something complete other

[cerulean pasture]

well "LLM", regardless, the style is a choice and i dont understand why they are optimizing for the silent non-discerning majority

[oak flower]

Well precisely that’s why they’re called large language models

They’re really good at manipulating language. But not at thinking.

[cerulean pasture]

I dont know anyone IRL that talks like GPT lol

like its not even a regression to the mean, its something else

[oak flower]

That’s called hedging language

And you’re actually spot on not a lot of people even pay attention to what you just mentioned

It’s an extremely annoying feature and makes it completely useless for most people in the real world lol

[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.

[mighty vine]

saved

[cerulean pasture]

Oh i totally have a similar view, my main mistrust of "AI" isn't the AI at all, it's the corporate incentives structure, and the fact that OAI speedran turning into a high intertia/low agility microsoft-esque dead leviathan of a corporation in record time

[oak flower]

The thing is you gotta understand. I don’t think this is on purpose like some kind of conspiracy.

I think this is just the way our system works

[cerulean pasture]

absolutely

i agree

[oak flower]

Like this is the only inevitable outcome

[cerulean pasture]

yea, which is why I think we have to [redacted]

[chrome monolith]

this is like watching people discover capitalism

[oak flower]

Do you know what the whole AI economy is based off of?

[mighty vine]

on discord, few people have said that i text like chatgpt
maybe that's a compliment : )

[oak flower]

The concept of tokenization

Meaning that the future of the Internet and digital platform is gonna be via tokenized systems

[cerulean pasture]

i thought it was the worlds largest pump and dump

[oak flower]

Meaning that theoretically everything will be have to be paid for

Even simple things like quarries and searches

[chrome monolith]

more than just AI uses tokens, that word is overloaded

[oak flower]

It’s a brand new economy, the tokenization market

Since the metric of measurement in ai is tokens

[chrome monolith]

what do you mean by quarry in this context btw

[oak flower]

I mean when you need to find information everything will cost you

[chrome monolith]

I need to buy a bunch of stuff from a quarry this week and that's not cheap

[mighty vine]

normal api - requests/hits

AI api - tokens

[oak flower]

Yeah, but now imagine all ecosystems built off that of those platforms

All websites and services built off the tokenization system

It’s a whole new different economics

[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.

[oak flower]

Imagine a social and banking system based off of digital currency

This is very possible with AI and tokenization

[cerulean pasture]

how will they prevent the rise of "stochastic t​er​ro​ris​m" if they keep up with this rampant value extraction from the labor class. I feel like that won't have a linear growth response. It will lag true sentiment but be directly correlated.

So we will see like 1 or 2 events per year, then 3, then 5, then 20, then 100.

[mighty vine]

who did this?

[cerulean pasture]

did what lol

[oak flower]

🤣🤣

[stark thicket]

Can we please escalate this feature request?

It drive me nuts that it can't even access the name of the project to figure out which repo to use.

[mighty vine]

send 🔞 in this channel

[cerulean pasture]

I've been banned 4 times lol

[chrome monolith]

I think some people forget that a lot of people have multiple projects

[mighty vine]

for?

[oak flower]

https://discord.com/channels/974519864045756446/989157702347411466

Meet me der

[cerulean pasture]

Why arent you just using codex? seems odd to try to force a square peg, when there's a much more appropriate tool availible

[chrome monolith]

its nice to talk projects with chatgpt though, nicer interface

[stark thicket]

Because I use TRAE to do the coding and GPT to guide it along.

[stark thicket]

Yeah well they let us create projects, that is nice but you would think they would make the threads aware of what project they are in.

[chrome monolith]

the whole project interface could be a lot better than it is

and you're right, its weird that chatgpt doesn't know what project its in

[stark thicket]

Yeah, drives me nuts having to start every thread with "Use the Github connection to access project-name repo for this" every single time.

I submitted the feature request awhile back, I would link it but they dont let me.

[chrome monolith]

you can't post urls here but you should be able to get discord message link

[stark thicket]

That is what I tried I think. Let me try again

[chrome monolith]

psht. The automod here is a pain

[stark thicket]

Yeah that is a bit strict

[chrome monolith]

testing discord message link #images-canvas message

ok that works

[stark thicket]

Channels work but not threads?

[chrome monolith]

maybe its that particular channel

ah that's a message link not a thread

[stark thicket]

Lol maybe it is just me

Yup I cant do channels either

[chrome monolith]

#1494033034578497709 message testing again

[stark thicket]

YAY thanks

[chrome monolith]

you're either caught being too new here for that feature, or cough skill issue. I just searched up posts by you

[stark thicket]

Cant swear either lol

[chrome monolith]

ah yeah I've been automodded for a few things here. This discord definitely has rules

[stark thicket]

I have been in here a long time but never posted much

[chrome monolith]

I sat in here for a year before posting

[simple lagoon]

This Automod is very bad yes i like we are childs First Childs no should use Internet and even more Discord

[chrome monolith]

just randomly, where are you getting these links from ... the one in the browser bar won't work, but if you right click a post, ... then there's "copy message link"

[urban granite]

I heard GPT 5.5 dropping very soon

[normal quarry]

maybe, or it could just be rumours

[scarlet steppe]

oh gpt is down, cool

[median elm]

I noticed too!

[muted plinth]

same here

[icy kiln]

oh its down

[dense pine]

Indeed... bummer

[cerulean scroll]

rip

[alpine pumice]

gpt down again??!!

[median elm]

Browser seems to be working for now

[alpine pumice]

not for me.. at least not the temp chats..

[charred turret]

not for me

[alpine pumice]

if its guardrailed like rest of 5 series, its useless

they need to go back to 4.1 level open-mindedness

[gentle forum]

lol, i dont need to ask if its down

how is 5 guardrailed?

[obsidian thunder]

chatgpt is down now i wont be able to study and pass the exams

thanks open ai 😠

[gentle forum]

gasp, might have to do it without AI assistance

[obsidian thunder]

thats not possible bro and you know that

[gentle forum]

ofc it is, what do you think we all did before 2022

[obsidian thunder]

no but it was different back then

teachers are lazier now they expect students to let AI teach them in my school

[gentle forum]

i mean theres stuff i cant do rn, i wanted gpt to write me some docs, theres no way im translating them all from C myself

but it can wait

[alpine pumice]

people who solely use chatgpt for work

[clear glacier]

bro 😭 i was going to summarize a whole book of text i got work on tomorrow

[opal wharf]

This generation man

[obsidian thunder]

not my fault, read my other message

[obsidian thunder]

...

[opal wharf]

Well you have to be the bigger person. Because AI doesn’t beat traditional, conventional studying.

[alpine pumice]

true

world would be a better place if people have the "bigger person" mindset

[obsidian thunder]

studying aside, its gotten so bad in my school that i need AI to TEACH me, to do the job the teachers are meant to do

so either way I need it

[opal wharf]

Well you can still do something, lodge a complaint, find another school, etc. you don’t have to sit there and silently suffer if the school is that bad

[obsidian thunder]

you are right but its my final year so I might as well finish it but you are right though. Many students also want to file a complaint but its a risk yk

[misty sky]

🤔 🤔 🤔

[opal wharf]

Peer and institutional pressure is common

But it takes one person to change that. Sometimes constructive defiance can be effective

[opal wharf]

Server issues upstream

[tawny hull]

can you remove my chatgpt ban

[eager cloak]

how long does this usually take

[tawny hull]

create a new account it fixed it for me

[opal wharf]

As long as necessary

[eager cloak]

what a response

[tawny hull]

i got banned from codex, sora, chatgpt, what do i use

[opal wharf]

That’s the most realistic answer you’re gonna get

[eager cloak]

gemini

[misty sky]

@tawny hull lol haha yes ofc LOL

[eager cloak]

fair

[tawny hull]

alternatives after most people got false banned:gemini(for chatting), claude code(for vibe coding), veo 3(for video)

[eager cloak]

how are you "false banned" on all of em bro

[opal wharf]

What you mean you’re banned? like a formal notice or weird error messages

[tawny hull]

weird errors and my account got signed out everywhere

[opal wharf]

That’s not a ban, that’s a literal server issue

[brisk phoenix]

Guys I am trying to sign in back to my chatgpt account because they couldn't log me back in and I'm getting this

[tawny hull]

false ban use vpn

[tawny hull]

use vpn

[brisk phoenix]

What do i do about it

[misty sky]

So the server goes down and you’re really going to create a new account, even though you’ve been a subscriber since the beta? Okay lol ^^

[opal wharf]

It’s not a damn false ban it’s a server issue

[tawny hull]

i reccomend tunnelbear vpn

to lift the ban

[brisk phoenix]

Okay

[misty sky]

@brisk phoenix No ! just wait 🙂

[eager cloak]

this octo dude is cooked holy

[opal wharf]

Yes. If it was a ban, he would have it in his emails, not platform-wide “weird errors”

[eager cloak]

fr 😭

[opal wharf]

What’s wrong with Norton, why specifically that VPN?

You work for them or something?

[misty sky]

Don't pay attention to him, he's full of it

[forest nymph]

Guys gpt 5.1 is back on the website, what the hell?

[misty sky]

@forest nymph Temporary outage. Just a little patience, it’ll be back up soon

[eager cloak]

hope so i need it to go over my assignment before i hand it in fr

[alpine pumice]

anywhere between a couple hours to half a day

[misty sky]

@eager cloak fr ? French ?

[eager cloak]

for real* bro 😂

[untold thunder]

ChatGPT web session expired and I cannot login. The desktop app remains logged in but cannot access old chats. But Codex in VS Code still works.

[misty sky]

@eager cloakSorry haha xD

[alpine pumice]

REALLY?!

[forest nymph]

I know there's an outage I mean I'm talking about 5.1

[eager cloak]

is alr ahahhaha

[untold thunder]

So we just have to wait for their backend to be fixed.

[tawny hull]

guys use this best alternative htps:/ /gemini. google. com/ app

[misty sky]

😂 😂 😂 😂

[forest nymph]

Yeah I'm not joking Gpt 5.1 model is on the website

Literally go check it out yourself

[eager cloak]

bro ander dingus please go 😭

[wanton egret]

peak website

[alpine pumice]

yes! i did check!

[opal wharf]

No. Use traditional proofreading methods

[alpine pumice]

lets gooooo

[misty sky]

Gemini drove me crazy, I had to unsubscribe ^^"

[eager cloak]

chat do it better tho

[forest nymph]

I don't know if this temporary but holly crap 5.1 is on the website and sadly we can't use it because the power outage

[hidden nest]

hello My chat GPT advise me “something goes wrong. “ Somwone else has same problem ?

[alpine pumice]

the service is down..

[opal wharf]

Let’s delve into this issue together:

AI Is Terrible At Proofreading