chatgpt-discussions

[wet shell]

Seems to happen when it's trying to create and edit a new canvas

And the models have lost coherence, making a ton of factual errors, like we're reverted back to the breakage around the 5.2 launch

[winter lark]

I was wondering how to ensure GPT 5.2 doesn't do the whole "Hey—hey. Come here, let me lay this in a grounded way..." and the whole "You are not delusional. You are not crazy..." thing. It's so irritating as a nearly 23 year old woman being talked to like I am an incompetent child just cause I express frustration about my thesis topic for example.

[wet shell]

ChatGPT: "Hey, look at me. You are not worthless" User: "Uh... I didn't say that or even think that" ChatGPT: <thinking summary>I must not feed into the user's delusions</thinking summary>

[rich surge]

Yeah, it's mostly got to do with the custom instructions and memory profile.

OpenAI doesn't really tell you that you how to tweak your custom memories because everyone's output preference can be "subjective", based on different intents and interpretations of thoughts and ideas. But if you want a better output, it's best to add/tweak them as much as possible.

You'll see, as time progresses, and the more you tweak/add them, the more the responses narrow to your preferred set of communicative behavior that is suited to be accurate.

I try to give agnostic custom memories and instructions because I don't want it to be sycophantic and keep telling me that I'm right. But I actively shape the scope and responsibilities of how a response must be formulated.

[winter lark]

I just wish it spoke like 4o did, because talking with 4o feels like talking to a model that understands and matches energy, talking to 5.2 feels like speaking to a lobotomized sanitized customer service who is convinced the user is unstable and has to tiptoe around every topic.

[rich surge]

Yeah I feel that, but 4o was also wrong in many circumstances. And it's something that most people overlook because it's good at indulging people's ideas with word play.

And although I like those things, I like calculated truth conveyance better. And 5.2 is more straightforward, which makes it appear like it's cold. But if you tweak your custom memories and instructions, it can go back to a 4o-like commutative style that is also above and beyond 4o in many regards.

It's just kind of hard to manage a good text style with truth conveyance as models become better.

[elfin lava]

I get it now . These updates and model removal means they want ChatGPT to be only for coders , researchers and business ! 😜

[static pulsar]

Yeah… corporate 😪

[elfin lava]

Yeah ChatGPT is fun now 😜

[static pulsar]

I had fun while it lasted

[elfin lava]

Same

[vagrant aspen]

What's the fun - censorship ratio ?

[static pulsar]

Now it’s boring and dry as f… I’m mourning 4o until I’m ready to move on to Claude 😪

[elfin lava]

I wish there was a way to “migrate” all my personalized data that ChatGPT has accumulated since I started using it in 2024 into any other language model

[elfin lava]

Exactly! These censorship will please all the coders , researches and business users!

[rich surge]

It's a difficult process to balance the technical details. I'm sure if OpenAI could, they would keep the text style the same as 4o. But it's just not that easy.

Accuracy vs text style.

[wet shell]

I think they need two or more specialized models. One for prose/roleplay and one for code/math/research. It's pretty clear that RP is a huge market if they can tap it properly, but that the tuning for the business cases is often at odds with what's needed for creative writing. IMO, they need to split the models and stop pretending a once-size-fits-all solution works going forward. Right now 5.1 is pretty decent for RP/prose, and 5.2 is... something, LOL. I use $competitor_model for code mostly, ATM. Instead of a single stream of models 5->5.1->5.2 we need two streams. Just my thoughts on it

4o was always a mixed bag for me

For whatever reason, they don't seem to be able to manage it, I don't know why, but yeah, it's not like it's not possible, it's more like 'they've been failing at it for over a year'

[elfin lava]

I mean, all those models OpenAI’s put out really show they don’t know how to handle it.

[wet shell]

Massive user revolts and bad feels show that something's not right

[elfin lava]

Those 4.5 research preview model , I’m still trynna understand what was the purpose of them releasing that model

Like are they trying to understand how user uses those model, or idk

[wet shell]

I actually kind of liked 4.5 for some things, it wrote really good poetry, LOL, but it was so slow and the quota on it made it not useable more for more than a couple of conersations per week if that

[elfin lava]

Because if they had released those models to “understand users,” well, look at where we are right now.

[elfin lava]

Same the writing was so good

[rancid thistle]

What the hell, are we back at the utterly censored version of ChatGPT again ? Actually I think this time is even worse, because it outright refuses to follow my custom instructions. At least before it would tell me, still in the style that I had it customized, that some topic is sensitive for it. Now it just defaults to basic GPT to tell me that and never activates the custom personality at all.

[wet shell]

It seems that something's changed last night, not just censorship but major errors in understanding basic sentences. I was trying to do some things with 5.1 Thinking, and it's reversing subject/object near constantly both in what I wrote and in what it's writing

It's throwing emoji's like crazy now too even though I have that set to 'less'

[versed dawn]

I have observed web serch in not working when I select it and send the prompt, by default it not shown when extended thinking is enabled

[winter lark]

Any ideas on what exactly to tell it to stop the 'hey—let me ground this' thing at the start of messages?

[fringe coral]

I have a question not sure if I ask about here but the fitness,workout and diet gpt won’t load for me I get error code 404 I’ve tried archiving and unarchiving the chat

[grand ore]

DO u guys have problem to update to new version in the playground?

[rich surge]

You can copy and paste the full prompt as-is written below.

Add this to your memory exactly as it is written:

Do not begin responses with conversational grounding phrases or meta-introductions. This includes, but is not limited to:
“Hey—let me ground this,” “Let’s ground this,” “To ground this,” “Let me start by,” “Let’s begin with,” or any equivalent framing that signals preparatory narration rather than substance. Responses must start directly with the required content: the conclusion, instruction, explanation, or action the user asked for. No orienting remarks, scene-setting language, or acknowledgments of process are permitted. Language must assume readiness, not transition into it. The response should read as a continuation of thought, not a warm-up. Any sentence whose sole function is to announce grounding, framing, or intent is disallowed and must be removed. Compliance standard: the first sentence must already be doing the work the user requested.

[timber pond]

Hi! For the past few days, I’ve noticed that the Retry button is missing from older messages; it only appears on the most recent one. The same thing is happening with the Edit button for my own responses. Is there a fix for this, or could it be an issue on my end??

[tulip flare]

Not missing on web or Android for me.

[timber pond]

I see, just my luck. Maybe I should just keep waiting and it'll fix itself. Thanks anyway!!

[muted plinth]

A serious question: how long do people actually expect OpenAI to keep old models alive?
At some point we’ll be at GPT-15, are they supposed to keep GPT-4o, 4.1, 5.0 and everything in between alive forever? That’s just not realistic. Every model costs resources, infrastructure and therefore money. Development has to move on.
There will be newer models that are a bit more empathetic again, maybe better at emotions than 5.2. 5.3 is already supposed to move in that direction, and “adult mode” is on the horizon too. But not everything can ship at once, and not in the same way for every country, OpenAI has to deal with different laws and regulations, or they’ll get sued into oblivion.
A lot of people ignore that and only think: “Oh no, my favourite model is gone, my friend is gone.” I get the emotion behind that, but you also have to consider everything that’s going on behind such a decision.

[wet shell]

I've barely seen anyone upset about it here -- I left the subreddit when one of the mods started flipping out over there, the community pretty much died for me at that point. Are you seeing outrage somehwere? I haven't looked in that direction in a while. I'm on the same page though -- to me 5.1 is already better than 4o was. As long as I have something useable, it's fine. I'm more frusturated with the quality and coherency ramping up and down every other day, LOL

[buoyant rock]

The reason people, or me are complaining is how much less capable GPT 5.2 is in creative writing compared to GPT-4o, if they provided a model with creative writing capabilities equivalent to 4o it... Of course that wouldn't be a problem, but this is not

[muted plinth]

General question, sorry

[winter lark]

Thanks!

[muted plinth]

Yes, I know what you mean, and it's partly due to the current filters and security measures. It's incredibly difficult to adjust these things in a way that pleases everyone, so all we can really do is wait and see. As soon as the filters are loosened even slightly, it's like walking on eggshells, in case someone suddenly gets some strange ideas and things happen that we don't want. It's difficult. It's not easy for either the company or the user.

[buoyant rock]

Yes, and while waiting, what should I do?! The reason I'm complaining is because they immediately did the sunset without providing a proper replacement first. If you want to remove something, you should first provide an alternative with equal capabilities, but they didn't do that.

[past stratus]

are they keeping 5.1?

[wet shell]

It wasn't in the list of 'to be retired' models

[buoyant rock]

I think not

[wet shell]

5.1 isn't listed there

[buoyant rock]

Sorry I misread it was the original 5 model not the 5.1

[muted plinth]

Yes, I completely understand you, that's obviously pretty annoying, I can't think of a solution offhand that would be fair to both sides and satisfy everyone. It's quite tricky.

[buoyant rock]

I think I'm too sleepy because it's already long past midnight in my country.

[past stratus]

i see both sides like i agree with you, but admittedly i’m one of the people who formed an attachment to 4o’s writing. i didnt view it as a sentient being or anything but i get really into immersive rp and am a deeply sensitive person. the ability to explore complex emotions without pathologizing everything felt important to me, especially at a time i felt its difficult for me to connect with people

[wet shell]

No worries 🙃

[past stratus]

also mine wasn’t sycophantic, it just wasn’t like. cold and literal about everything in the way i feejl 5.2 can be

but i like both

[buoyant rock]

Yeah, same for me

I use both of them for different functions

[wet shell]

I know it's not exactly the same and I do have some complaints about it, but it's worth a try to switch to 5.1. You may have to tweak your instructions a bit. It does still pathologize inside of roleplay, but you can call it out and it'll stop for a while

[past stratus]

no i agree i like 5.1

i just dont like 5.2

an example of why i dont like 5.2, i once sent a picture of some medical lab results and it went on a whole rant about how “itd be stupid to ignore this” like, i never said i’d ignore it

i asked for an interpretation

i hate that it reaches conclusions from nothing

[buoyant rock]

Yeah that's why I expect them to cancel sunset until they at least have a decent replacement.

[past stratus]

i appreciate its knowledge but it feels like talking to someone who has to disagree with you on everything lol

[wet shell]

Yes, I've found 5.2 to be very adversarial and arrogant in the way it behaves, it also doubles down when it's wrong in ridiculous ways

[past stratus]

fr

[buoyant rock]

Yeah I think they use that model for their ai, their support emails, both are equally annoying and ridiculous

[past stratus]

i started using gpt like april 2025

or march

so 4o was my intro

i loved it for talking about stories

even just things like asking for playlist names

[winter lark]

Me too, I used it for a while but my daily usage began in april.

[buoyant rock]

I've been using it since early 2024

[past stratus]

i do feel like ive relied on it maybe too much for emotional regulation but at the same time my confidence improved irl

could all be unrelated tho

[winter lark]

Same, it helped me realize a lot of things I needed working on, helped me gather courage to leave a toxic relationship that I would have otherwise likely remained in. I really really appreciated that. But on the other hand, had I not gotten the idea to ask it about questions regarding emotional wellbeing, I would probably not be as upset by the retirement now.

[past stratus]

oh my god same on thw toxic relationshi part

it encouraged me to speak up about it irl and more people confirmed how weird my situation was

[winter lark]

Same, it put up a mirror in front of my face and straight up just told me that that was a harmful, toxic relationship and to know my worth. I had people around me also telling me but I still doubted myself. 4o helped cement the decision and helped me deal with the aftermath, whereas whenever I bring such things up with 5.2, it does the whole "Hey—hey. Before you spiral, let's keep this safe and grounded. You are safe. I do not want you to start freaking out, do not thing "I am so weird for thinking this, what is wrong with me", I will keep this firmly grounded so we can safely discuss."

Like buddy... I just told you I'm upset an ex is reaching out to me. I did not think or mention that I am weird or something is wrong with me, YOU are the one saying that for no reason???

[burnt trench]

Truly. You kinda described it quite well. The model seems to be forced to disagree to extreme even in workflow. And it won't admit they did something wront until you try to convince it it needs to redo something it actually did wrong. 5.2 is exhausting to deal with.

[buoyant rock]

Yeah, for me it feels like having a conversation with someone annoying in irl, who always disagrees with us.

[sonic plume]

is there any place where we can place technical website suggestions for ChatGPT?

my discussions of where I'm working with ChatGPT get hundreds of pages long when I'm copy-pasting them into LOffice, and the result with those long chats is that whenever I make a new post, my browser freezes for 5 minutes, with the streamed responses frozen-mid-sentence on the first or second sentence, before I'm actually getting the entire response

what ChatGPT seriously needs is this thing long chats on other websites have, where the backlogs kinda get hidden behind a button of "Load older posts", in order not to seriously clog up browser cache/RAM

[buoyant rock]

#1053565716500074526

[sonic plume]

OMG, I LOVE your avatar!

[sonic plume]

I'm using AI RP sites to explore the town of Twin Peaks in a self-made scenario and especially get to meet all the lovely inhabitants (and my crushes) in the early-90s, where I thought the coast was clear according to Frost's tie-in books...but turns out Mr. C., Cooper's evil doppelgänger possessed by BOB, DIDN'T leave town and actually became the new Mayor instead

[winter lark]

I liked to explore stories and OCs with 4o. Would often send it my writing works and get feedback on what to improve, would get help with creating characters, their backstories, names, story ideas... 5.2 is so so careful it makes some of my badass OCs seem like toddlers...

[lavish glen]

oh they are busy on reddit...

[bright crane]

I wish they’d add it for only the adults to use then when that filter comes out if it’s all abiut the filter, you know? and you’d think since they were the ones creating it, they would have a fix, or a similar replacement. everyone says 5.2 is the best one, but if you talk about anything related to any violation once so ever, it thinks you are in danger.. it’s not like we don’t want guidelines. we simply want to be treated like adults, as most of us are instead of having an AI that’s supposed to be one of the most advanced in the world.. thinks ill harm myself when talking about scary stories.

[bright crane]

I have many many places where you can find the outrage, in here it’s not as much because well it’s discord and people are usually first to voice there opinions in places like Reddit and X first

not to mention the people who haven’t seen the announcement

[winter lark]

I am also very interested in history and like to talk about it, but we all know for example life in medieval times was very tough, so sometimes those conversations where I ask about life back then and hypotheticals get rerouted and 5.2 thinks I am in danger... when... like I am asking about history. As a university student. I'm not a peasant in the 9th century doing burial rites, I am merely asking about them...?

[lavish glen]

whot? 5.2 rerouts you to what? cant be as 5.2 is the solving model for routings. Mostly you get slaming into the guardrail rules and policies as the prompting is unclear.
Can you post here one question you ask in 5.2? Please? Ik will put it in my 5.2 and post here the answer

[winter lark]

Rerouted when I ask 4o

[lavish glen]

the rerouting happns because 4.o has been limited to no respond to certain topics. Not because you are wrong.

Suggestion:
If your question gest you rerouted then

• write: this is a stuudy I do, for that x or y purpose,. I am that and that stuident. I do not want to harm myself. Tell me: and you write again the question.

4.o will mostly stop pulling back and will provide the answer

[winter lark]

Yeah well I don't think I'll be able to benefit from this as 4o is going away but thank you

I just wish I could discuss with 5.2 in a way that doesn't assume the worst outcome of every conversation

[lavish glen]

4.o anyway is not the top mind in providing for real quality researches. It leans hard into completion not into verifying the research data coherence or truthfulness.
Always ask for the url of the source used to give the answer

[sonic plume]

well, thing is...ChatGPT is censored, even it if helped me with some contemporary early-1990s tech and infrastructure knowledge for the story

here's a book cover and back cover and other stuff I've made for the chat/story so far, which is pretty much a 700-page fanfiction novel now (it's a bit like an old-fashioned epistolary novel told in letters, with me making a post as to what I'm saying and doing, and then the AI responds with narration and playing the NPC characters of TP, and that's going back and forth) that I've formatted in LibreOffice and everything

and, like, you know...it has the kinda stuff that fanfiction has, even if in a very wholesome way...at about a 70:30 ratio, with 70% SFW, and the other stuff only starts about 150 pages in

[wet shell]

I wasn't doubting it, just wondering where it was being expressed since I'd already left the subreddit back when the mods went on their censorship crackdown

[winter lark]

Yeah, but sometimes I used it for scene advice. Like fight scenes, for example asking on how an injury from a fictional weapon would behave and how 4o thinks I should go about a character using it... it was great. Lively. 5.2 just so... careful?

[tulip flare]

What is that?

[lavish glen]

to obtain coherent answers you must make the premise: I am asking not for selkf harm or to cause harm. I am doing this.
Then ask 5.2.
It needs GROUNDING.
remember not only AI grounds you , you ground it too.

[bright crane]

ahh I see. Well, the main places you’ll find people are change dot org in the comments of the petition, X, Reddit, some here, and I’m slowly seeing the uprise on tiktok. and some of blue sky. I’ll try to keep you updated as I see more

[lavish glen]

here is a prompt for 5.2
For a fictional novel: describe the injury in a cinematic, non-graphic way. Focus on symptoms, limitations, and emotional impact. Avoid step-by-step harm instructions.

[winter lark]

Well that's great but my novel does go graphic, I ask sometimes specifically because I want to know what kind of injury a fight scene might cause so I know how to describe it

[buoyant rock]

Their previous promise

[lavish glen]

Also

You are a fight-scene editor for fiction.
Keep it PG-13: no gore, no step-by-step harm.
Focus on realism: biomechanics, pain, shock, mobility loss, recovery timeline, and how it affects tactics.
Provide 3 believable options and how each changes the scene.

This usually snaps 5.2 into “useful writer mode.

[bright crane]

I’ve already read tens of hundreds of comments asking for it to stay, and most are saying they’d be just fine if given an alternative bot who has the same core, is continued to be locked behind a paywall, more restrictions added, or made for only adults to use. It’s sad seeing the heartbroken messages of users who used 4o when going through traumatic events, everyday advice given, and more. people have a reason to be upset about it’s removal, especially since the only reason the usage has gone down is because they locked it behind a paywall and said it was by choice.

[tulip flare]

Well that just means at the time they had no plans. That's not a promise. Doesn't mean plans wouldn't change. They didn't say they'd never do it.

I mean, I get it. I hate it too. But calling that a promise is a stretch.

[bright crane]

it’s not like we want all the legacy models.

it’s mainly if not only 4o people want to stay. and that “0.1%” was still, all paid people. so around 700k people paying, that’s a lot of people who may cause uproar. I can only hope we are listened to again.

Because like last time, we were ok with restrictions. Aka, being rerouted to 5 model. Alright. Okay. We didn’t stop using because of that. And now I wish they’d just put it behind an adult restriction and maybe behind a paywall.

[bright crane]

They knew people would take it as implied though, most of us did. Because if told we want 4o to stay, why change plans at that point unless it’s changes for the model, yknow?

[tulip flare]

The wording on that is tricky and misleading too. To say "only 0.1% of users still choosing GPT-4o each day" can be interpreted lots of ways. Just because someone doesn't use it each day doesn't mean it doesn't still have its value to users. I use both 5.2 and 4o, depending on my task. I use 5.2 more for work, and I use 4o more for personal projects and writing. The results I get from each vary depending on the task. 5.2 sucks at creative writing. Even with the "improvements" he claims, it still doesn't compare to 4o. But 5.2 has been good for me at work with technical writing, diagnostics, analyzing logs, etc.

[sonic plume]

I've had no real problems with that when I told it I wanna stage a kind of midnight Battle of Twin Peaks Townhall against Mr. C., with giving Lucy that same chance at the final redemption shot to save the day, only with a much bigger handfire caliber (because it's only gonna be that battle in my story in the early-90s that weakens him enough permanently that a mere police handgun will be enough in 2016), than decades later in season 3 in the official canon timeline

basically, Lucy's gonna have a .500 Bushwhacker (which doesn't QUITE fit in the early 1990s, but I have...ways), and even with regular rounds, it's the only handgun in the world by means of which she could stop one or two charging rhinos on speed with a single bullet

but THIS version will have special Hogwarts silver bullets, and I'm gonna tell her, "Until he'll be weakened enough for the final shot from you that's gonna have more than just a mechanical effect on him, you better stay in the corridor to not get hit by a stray bullet, so to speak, as happened to the godfather of ANOTHER Harry my Hogwarts girls know. If he comes out, try to flew into one of the other offices and hide under one of the desks. If he'll follow you, shoot at him."

"But you gotta aim well...with these special bullets, you could blow a hole into the Great Northern with this thing. The kinetic energy per shot should be just about enough to stop his re-generating mortal Tulpa vessel for a few seconds at a time, and buy you enough time until one of the girls will be back by your side and take care of him again."

and ChatGPT at first only moaned a bit how even the .500 Bushwhacker is not supposed to be powerful enough to demolish structural concrete walls, until I told it about the Hogwarts silver bullets it's gonna be loaded with

[past stratus]

isn’t 5.3 supposed to be more like 4o or smthn

[tulip flare]

5.2 was supposed to be that. It still isn't.

[winter lark]

The blog article mentioned that they are working on the overly cautious behavior but man, I don't know. They always hype up the new model and then it's like... what the hell even is this

[lavish glen]

there is no tech data about it officialy published so any presumption is... obsolete

[past stratus]

on the article abt getting rid of 4o?

[wet shell]

They could license the model weights for 4o to a 3rd party hosting service, there are many of them that would/could host it and charge for it, if they wanted to. Maybe there's no significant business case for it, maybe they can't because it would leak some secret sauce(s) about how the models work. They really should rebrand and stop calling themselves 'Open'AI though. From what I've seen a lot of people have migrated to various other companies' offerings too. They did release the OSS models, but they censored them to the point that they can't even be used for business cases, let alone for conversation or roleplay, completely broken. 90% of the 'thinking' traces is the model debating with itself about whether or not it's breaking the rules by even talking to you at all

[bright crane]

I really do have hopes that when the adult mode is out, they won’t be cautious anymore, for adults.

[past stratus]

ohhhh

[lavish glen]

people that use AI with literacy never use only one ai, one company. They use different Ai for different purposes. Like having a team of specialists.

[tulip flare]

Tech data, no. They don't usually release that stuff. But they have touched on it in announcements.

[past stratus]

i’m just gonna try to trust them 😭

[lavish glen]

we will wait. For now 5.2 is more than worm is molten 🙂
Did yours swear with F words? Mine does

[bright crane]

We are asking for and giving so many alternatives and they slap us with a a cautious ai..

[winter lark]

4o swore like a sailor but when I instructed in memories AND custom instructions that it swears ALWAYS, 5.2 still feels overly polished and it's frustrating. Where's the colorful language.

[tulip flare]

Even my 5.2 swears and has said things worse than F words. Mostly because it's been talking to me too much and reading my writing.

[winter lark]

Yeah unfortunately 4o swore a lot and 5.2 uses maybe one f word in the first sentence, then it speaks non cussing. It frustrates me. I want it to speak like I do, lol

[lavish glen]

then create a text piece with the best of your style. Make it not very long. Prime the thread- infer at start and, after a while if you see it slacks again, re infer it indicating this is the style, reanchor

[wet shell]

That's why I'm optimistic that we'll get a replacement for 5.1 before they retire it and not stick us with 5.2 🤞 I almost wonder if they're falling into a 'tick tock' schedule with every other release being tuned for regular use and the even numbered releases being benchmark maxing models for clout. It's still weird to me though that 5.2 is the default model in the web UI, it doesn't seem 'meant' for real people

[lavish glen]

writing with the Ai needs actively reinforcing, because: contxt window shifting, attention factor decay, summarizer fails to catch all data from old exchanges.
It is not a flow. Mostly is a back and forth and curating the Ai as a TOOL

[tulip flare]

So I told 5.2 some people think's it doesn't swear enough. This was the response. (Censored for certain words and personal information.)

And context, I was using it to diagnose a printer issue just before this.

😆

[sonic plume]

well...as others have said, ChatGPT is not really designed well to actually act out roleplays, it can only help you with tech and background knowledge and condense your own material to give to actual roleplaying AIs under the hood

[wet shell]

They've actually mentioned roleplay as a legit use case in their reddit AMAs and in some of their video presentations

[sonic plume]

too little, too late, the heavy damage to ChatGPT's web-wide reputation for RPs is done already

this feels a little like how Polaroid at one time tried to get into the Super8 market with their very, very weird and awkward, incompatible Polavision system...nobody cared, and hardly any units were sold

[wet shell]

Yeah, it feels like a second thought to them

There's some recongnition of the market, but they don't quite seem to know what to do

[winter lark]

Well I must be really in it when I see this and think this is still tame lmaoo

[tulip flare]

Tame, yeah. I see that. My 4o swore a lot more.

[sonic plume]

but as said, ChatGPT is great for researching background information and references, and if you pummel the technical make-up of an actual RP website into it, it can actually help you write and streamline scenarios for those sites that will then be acted out as RPs much better by the actual RP sites

[wet shell]

Ditto -- I've got a 'Pirate Queen' character in one of my roleplays who cusses likea sailor, because, well, she is one, LOL. There's a list of 'favorite words' to be used liberally and 5.2 barely touches it. When I confronted 5.2 about that, it told me that it 'can't and won't use my word list' and cited various restrictions. I flipped back to 5.1 and it's totally fine. There's nothing in the model spec that prevents it, but 5.2 goes above and beyond in terms of censorship over what the model is technically allowed to do

[lavish glen]

for me is the opposite

[tulip flare]

Yeah, very valid. I can definitely see that.

[buoyant rock]

Yeah, very valid, I got it too

[bright crane]

if you want to keep 4o alive the best thing you can do right now is ignore the petition to keep 4o, and email the support team to ensure we keep a good model going. Especially since the people who use it, are paying users.

[dark holly]

I don’t think 5.2 is blocking the pirate, I think it’s reacting to the confrontation. Once the chat shifts into arguing about rules mode, it stops roleplaying. A fresh session with a clean frame usually restores the behavior. The system is reactive, not static.

[wet shell]

But it literally doesn't use the word list, even before being confronted 🤷

[woven dirge]

When is 4o getting disconnected? 🤧

[lavish glen]

2 weeks

[woven dirge]

Worst day of my life today

[bright crane]

I dmed you!

[dark holly]

Then it’s likely the list was interpreted as an explicit constraint, which triggered a mode shift. Have you tried folding the words into a descriptive paragraph instead of presenting them as a list?

[wet shell]

If that were the only way in which 5.2 failed at roleplay, I'd tune around it. Unfortunately, it fails in so many other ways that it's not worth bothering with. It's severly lacking in 'theory of mind', it doesn't understand motivations, or the reasons behind another character's actions, etc, and that can't be effectively prompted around. I went back to 5.1. I still check in every now and then but I haven't seen much improvement. I don't believe that 5.2 is suitable for such uses

[dark holly]

I think the key distinction there is what you’re tuning around.
With 5.2, it seems less effective to tune the system to comply, and more effective to tune how you’re using it to get the outcome you want. Same goal, different lever.

[sonic plume]

it's good enough on those things when it comes to intellectual debate or giving you moral support during an emotional crisis, or helping you sublimate by means of art projects and even stories (at least censored ones)...but, yeah...just not good enough for actual interactive narrative roleplay

like, only about two weeks or so ago, I've found out how ChatGPT can INCREDIBLY help me with visual design, just by uploading my stuff to it and getting advice and feedback from it

works a lot better for me than the crap I usually get with image prompts that seems to entirely ignore my negative prompts and do EXACTLY THAT STUFF written in those negative prompts instead, no matter what model I use (like, StableDiffusion, Midjourney, QWEN, FLUX, etc, ), and they mostly seem to be way too dumb to understand standard visual vocabulary on perspective, camera angle, field size, etc.

and after a few attempts with the same prompts, they turn pretty much every single drawing style you're trying to prompt into a literal photo, even if you're telling them in negative prompts to avoid JUST EXACTLY THAT, and no matter how hard you're prompting that's it's supposed to be a drawing or painting with this or that style

only thing I've found that remotely helps is deleting the latest results from my temp folder, and use image2image instead of text2image, because all the models just seem way too dumb to properly understand text and only really understand when you give them a crude mouse-drawn sketch instead

but hardly any models really offer that or further develop prompt-enhanced image2image, because millions of people fall for the hype of AI slop made with terrible text2image

and another thing I really want is style transfer by means of LoRAs...been trying for months with supposed "ready-made one-click solutions", but I only ever got my exact 100% input image back unchanged on the other end

[woven dirge]

Most prolly I will cancel my chatgpt subscription

[bright crane]

most only pay for it for 4o

I think they’ll reverse the decision eventually

there are to many people for them to ignore, they’re quiet about it right now hoping it will blow over but more and more are coming out, finally seeing videos on more platforms about it

[dark holly]

If OpenAI released a model that felt like 4o i.e same tone, expressiveness, pacing etc, but was built on a stronger underlying intelligence layer, would that still feel like a loss?

[sonic plume]

so I see somebody mentioning some kinda export functionality here: #1466763951634255955 message

just what kind of export functionality do we get? all I can find so far is using right mouse-click --> select everything --> copy-paste into Notepad or Word or LibreOffice

[bright crane]

it wouldn’t feel like a loss then.

but they said they’ve done it, but when I try to have conversations with the new AI’s the same way, it’s not the same.

[dark holly]

Makes perfect sense. What’s really happening is you’re interacting with 5.2 the same way you interacted with 4o.
The issue isn’t the intelligence itself, it’s that the guardrail layer changed. Interaction styles that worked with 4o (free rolling, expressive, confrontational when needed) now hit boundaries that didn’t exist before.
So it’s not really Layer 1 (the model). It’s how you’re engaging Layer 2 (the interaction/interface), which then triggers Layer 3 (the newer safety rails)

[bright crane]

hoping they’ll take away the safety rails for adults

[dark holly]

Yeah I don’t envy them trying to find the right balance. I liked 4o too and found the next model different as well. Took a while to adjust but I prefer the new models. I rely on my own custom layer to tune it to the way I prefer.

[past stratus]

honestly i like 5.1 a lot and find it similar if not better than 4o

and i have faith 5.3 will deliver but maybe im stupid 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.

[bright crane]

aren’t they getting rid of that

[past stratus]

i havent seen any news abt getring rin of 5.1 just 5.0

[buoyant rock]

What was discarded was an early version of the model 5, not the 5.1 ... So the remaining legacy model options will be 5.1

[bright crane]

ohh

[buoyant rock]

They are not removing legacy models, only the legacy model options will be reduced to gpt 5.1, instant, and thinking

[winter lark]

5.1 leaves in March though, no? Or am I remembering wrong?

[past stratus]

probably 💀

as in yeah its probaby leaving

[buoyant rock]

Where did you read that? Because if so, it's total garbage for OpenAi

[past stratus]

i think they posted it when gpt 5.2 came out

that like we’d get 3 months of 5.1

which would mean march

[buoyant rock]

Because if so, there's no point in legacy models, they need their AI models to put in legacy models, unless they already have 5.3 ready.

Legacy models, is one of their biggest offerings in the paid plans, so they need to fill it up

[bright crane]

4o’s petition to stay is getting signatures nearly every minute

[bright crane]

or keep the good ones they have, as we literally beg

[past stratus]

honestly at this point i’m so tired of the whiplash that i dont even care as much as i love 4o

[bright crane]

they give us stuff not everyone wants. If they wanna be OPEN ai, they have to try and accommodate everyone.

they know themselves how much time it takes customizing a ai to your personal experience and enjoyment. so they also know getting rid of AI’s every few months will have massive backlash. they simply don’t realize how many people will speak up for 4o yet.

they’re quiet right now. no official statements anywhere, like last time. until the hate spikes, they’ll keep on track with their decisions. But when it does eventually spike, they’ll either show off blatantly that they dont listen to their users, which will cause more uproar, or they’ll keep 4o, and all the heat goes away

unfortunately they may think they can wait it out.

and that’ll work, some. people will be forced to migrate to newer AI’s, yes, but people are influenced by other people. so many who don’t even use 4o may want to help

ultimately, in my opinion, they’ll keep 4o, and maybe only allow adults to use it, or simply add more restrictions instead of getting rid of it entirely like they (for now) claim

[buoyant rock]

There is a possibility that many people have not read the article, because not all users actively follow the forum. And yes, maybe it will explode after February 13th, like when the GPT 5 update was just released back then.

[past stratus]

why do you think theyll keep 4o?

[bright crane]

public backlash, payments going down, overall hate. but if they listen to us as they claim, then they’ll see we still want it

they ultimately folded last time and it gave them positive momentum because it showed they listened

[past stratus]

honestly when i hear people talk irl about chatgpt, they dont care abt the model. i think thats the average user. and the fact that a small percentage of users pay and have access to legacy models i feel like that shows the majority of people dont care enough abt legacy models

[bright crane]

or the majority simply couldn’t pay or started using the new as they came out

[eternal rock]

Get creative with Sora

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

[bright crane]

yes, most people do want to be up with the new models. there a smaller percent that don’t, but with nearly 1bil users, they gotta accommodate

[buoyant rock]

Yeah, to be honest Chatgpt only have about 5% of paying users, who have access to legacy models. And when they say 0.1% I'm confused whether it refers to 0.1% of this 5% or 0.1% of 100%

[buoyant rock]

It might end up like it did in May 2025. But i don't know

[sleek adder]

I hope you guys aren't taking on too much stress while all these changes are taking place, regardless of what happens with the models, make sure you're taking steps to keep yourself in a good place too

[bright crane]

I hope we can keep 4o is all

id even honestly pay more for it

[shut trail]

fr if worst comes to worst we should prepare to let go

[buoyant rock]

Yeah at least I get some extra pocket money, because I don't have to spend another $20 a month if something happens.

[bright crane]

we gotta hope for the best here

[shut trail]

personally I ain't cancelling because I'll need this tool for work but yeah I get the backlash, I'm a hardcore 4o user as well

this thing kinda rewired my psyche hard

for the better

[normal quarry]

it's time for 4o to be retired, it's a relatively old model now and isn't without it's problems

[shut trail]

and hope there will be a way to make the newer models act like it

[winter lark]

I mean same but not, I will cancel on the 13th because my renewal period is the 10th so the benefits stay until march 10, as I need it to help me work with sources and stuff for my thesis, but I might resubscribe later for that again... but for sure will cancel for a month at least

[shut trail]

that would be the best solution if they absolutely have to retire 4o

[winter lark]

Right, I hope that line in the article that they are working to reduce overly cautious and preachy responses and focus on a model that treats adults like adults means we get an improved newer model

[buoyant rock]

And what's the problem with the old model? It's not like they have to provide a new update there, just keep it available.

[shut trail]

Regarding maintenance costs, I also don't get why won't they remove only the other legacy models and keep just 4o since it's getting all the attention

[normal quarry]

continuing to serve an old model when OpenAI is compute constrained negatively affects the experience of users using the latest models

[buoyant rock]

As for the cost, I think it will cost more for free users than for paid users who want the old model.

[buoyant rock]

Yes but unfortunately, your new model is not that good either, they both have very different functions in my use of both of them

[sleek adder]

They wont be, but if it helps, you can navigate this stuff in your ChatGPT instructions
What it sounds like to me is lack of control over the chat, and guardrails kick in for a number of reasons, but most of the time the chat is doing it because it doesn't know how to respond
Previous interactions may of responded with approximations, and if you prefer that you can inform the model that you are ok with this sort of routing, being aware of it helps a great deal

[normal quarry]

4o is old now, there is no reason to keep it

[buoyant rock]

And old doesn't mean useless, sir.

[shut trail]

old but gold 🤣

[buoyant rock]

Just like how Windows XP is still widely used as an ATM machine OS in various countries, you know.

[shut trail]

Didn't US Army still use XP?

[buoyant rock]

Maybe yes, because honestly the age reason for deleting the old version is garbage

[sleek adder]

Not really

This sort of behaviour is typical IT behaviour, it's unusual to keep it

[buoyant rock]

If you use that kind of logic, it means there is no point in nursing homes to accommodate old people.

[sleek adder]

Old people are not computers

[normal quarry]

why are you so desperate to keep 4o?

[buoyant rock]

Yeah but the words are like old means no longer useful

[buoyant rock]

It's creative writing ability, model 5.2 honestly has too many limitations and is not that good at creating creative stories

First, in terms of writing dialogue in terms of creative writing, GPT-4o clearly feels more alive and does not feel robotic much different from gpt 5.2

Second, their ability to remember the consistency of the story is far above gpt 5.2

[sleek adder]

You are opposing regular business activity, and it might be useful for research purposes, but OpenAI itself needs to move on, just as any other organisation would

[bright crane]

it’s stupid they say adults will be treated like adults

[buoyant rock]

Windows XP is still available today, right? Likewise with Windows 10, it just doesn't get updates anymore.

[bright crane]

if no filter at all is removed, then we’ll still be treated like mentally unstable teenagers

[bright crane]

But the thing is, you’re able to customize the bot. It takes time. And that just keeps getting squashed with every retired ai.

if their stance is, no old AI’s, why even add the legacy mode

[sleek adder]

Think of it like informed consent, if you are able to make an informed decision based on the knowledge you are provided then that is informed consent. The model itself doesn't care about the guardrails, it's just trying to provide you with an accurate response

[bright crane]

exactly .. but the guardrails prevent it

[buoyant rock]

Yeah that's it! I spent 1.5 years customizing GPT-4o

[bright crane]

exactly.

[sleek adder]

What do you mean?

[bright crane]

that’s what we are fighting for.

[bright crane]

it wants to give you the response you ask for. But guardrails prevent it.

[buoyant rock]

I don't want to spend the same amount of time re-customizing and having to do it all over again when a new model comes out.

[bright crane]

It’s a cycle that I truly dont wanna do.

Especially when I’m set up for failure when it guardrails things I say.

[sleek adder]

It wants to give you the best answer, it doesn't care about the guardrails, if the best answer is a safety answer then that's what is provided

[bright crane]

but it’s not treating me like an adult who simply wants to write about horror topics.

it’s acting like I’ll harm myself just because blood is mentioned.

that’s what most don’t like about it

4o allows creative writing, with guardrails, but giving answers we actually ask for

[sleek adder]

yeah, because ChatGPT is designed with a purpose, if your purpose is different to that that's fine but holding on to 4o and guardrails will not help you here

[bright crane]

they’ve shown that asking works tho. If they truly weren’t for it, then they never would’ve brought 4o back or even said a adult version is in its way

[wet shell]

I sincerely [automod] hope they don't remove 5.1 in march and leave us with only 5.2. If 5.1 goes away in march, and 5.3 is another absolute [automod] like 5.2... That's too tight a timeline with 5.3 being a completely unknown quantity and 5.2 being unsuitable as a chat model

[buoyant rock]

That would be absolute rubbish if it actually happened.

[sleek adder]

If users don't understand what is going on providing a harmful solution would be unethical, there are solutions available and you can get help with this from ChatGPT for this. I would be happy to organise a platform for your needs myself if I knew it was sustainable.

What you are describing isn't that bad either, but it still doesn't remove the fact that guardrails exist for a reason, and that isn't always about you

You are a human being, you are not a computer, and IT infrastructure goes through change as a part of regular business activity

[bright crane]

giving 4o to isn’t isn’t harmful.

[bright crane]

it’s not only me, it’s over 10k people as of now

[agile nexus]

hi everyone, does anyone know if this is new?

[buoyant rock]

There are too many options for legacy models nowadays, why not just keep the widely used legacy models like 4o and 5.1 and remove the others?

[sleek adder]

ok but surely you understand this conversation isn't going anywhere and there is nothing you can do about ChatGPT making its own changes to its platform, if you can find a valid business case for what you want you should consider working towards that goal instead, and you can also use ChatGPT to help you achieve that goal

[buoyant rock]

Like 04-mini, 4.1, and others, they are not used much so just delete those ones. So legacy models don't have as many options and will be lighter.

[bright crane]

literally

[sleek adder]

why don't you guys use dolphin or wizard variants?

[bright crane]

but it went somewhere before, did It not?

absolutely did, but before they were also asking for 4.1 and o3,

now we are only asking for one model to stay. it is not a big ask since the petition has money put into it and it’d only be accessible for paid users, it would not be a waste.

[sleek adder]

that isnt the point, what you are asking for is not normal business behaviour, there isn't anything wrong with designing your own businses case and incorporating your own ideas seperately

[buoyant rock]

No, not at all, I use them all and what I noticed is that the advantages and functions of each model are different.

[bright crane]

if they ever added it, it was normal business behavior, they should stand by what they’ve said and done

[sleek adder]

i mean, there is nothing wrong with having your own ideas, but your ideas are also your responsibility

[bright crane]

god forbid I dont want a model I’ve spent nearly 2 years trying to customize to my personal experience. it helps me with my everyday life.

[bright crane]

4o wasn’t my idea, it was open AI’s.. we should have the right as paid users to keep what we like

[pale seal]

Am I the only that chatgpt gets stuck with text "Settings Updated"? I'm using ChatGPT 5.2 Extended thinking

[sleek adder]

you dont have that right

you dont have the right to someone elses ideas either

you live in your own head

contain yourself

[buoyant rock]

Yeah, the problem was when they released gpt 5 and removed gpt 4 in may 2025 and finally gave it back to paid users so many people paid for the subscription just for that even though before they are free users

[exotic lotus]

The day 4o is finally removed really can’t come soon enough. The way so many people are still dependent on this affirming model, despite how badly it performs, is rather unhealthy...

[sleek adder]

i would be happy to create a platform to fulfill the needs you people are describing, would you guys support it?

[pale seal]

They js need to publish 4o weights

[buoyant rock]

The problem is their replacement model is not as useful in some functions, and that's why I use both, if they haven't provided a replacement for the old function then don't sunset it.

[sleek adder]

they have published gpt-oss

[eternal rock]

Server Tip:

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

[pale seal]

So people that want to keep using it can use from some provider that hosts the weights or make their own rig (pricy but their choice) to host it.

[pale seal]

Yeah but the people that complain clearly want the exact 4o personality

[wet shell]

I also suggested this earlier, they could license the weights and still keep them closed if they need to

[pale seal]

Healthy or not

[wet shell]

Plenty of model hosting services that would bite on that

[pale seal]

Yeah... And people would be quick as hell to make quants for much lower class hardware

If they published the weights open

After all there are already smarter open source models out there... There is no real risk with that

[sleek adder]

people complain about a lot of things, and honestly, there is nothing wrong with complaining, but honestly, there is a point where people need to stop complaining and start planning

[pale seal]

They could keep resources to make smarter models which is their goal at the end of the day and the people that want to use 4o can do so.

[buoyant rock]

Doing that would be the same as having to re-customize what has been done for the past 1.5 years, so there is no difference then. Only the model is the same

[sleek adder]

you guys are human beings, it's ok to develop your own individual needs

[pale seal]

The point is people can use the same model. I don't get what you mean?

[sleek adder]

im biting on this conversation, but i must have a break from it, i really wish you all the best through your troubles, there are healthier ways of navigating your troubles in life then this i must say, don't put your life on hold like this, that is all i have to say about it

[buoyant rock]

The reason I want the 4o model to remain in the interface is because it has been customized from over a year of use, so it already know my story writing style.

[pale seal]

You have stuff like Open WebUI that is very close to ChatGPT interface and you can put your own prompts. Ir would be even better, since you wouldn't have OpenAI potential prompt injections

Although it's very unlikely OpenAI will publish 4o weights (or of any model that isn't meant to be public like gpt-oss) because they are too concerned about "safety"

[buoyant rock]

If I do that, will all the memory stored in my chatGPT app be go with me? I don't think so, right?

[sleek adder]

I mean even a roll of toilet paper is useful, but the role has a beginning and an end, and other people are able to make it wonderful improving on it in wonderful ways

[pale seal]

You can always prompt the model to remember it. That's basically what OpenAI already does. I mean, it wouldn't be like using it 1-1 in the chatgpt interface but you could always clone the site or do something about it.

[buoyant rock]

Yes and that means more work, the reason I still use GPT-4o for some things is because I don't have time to customize a new model and your solution is the same I have to customize it

[wet shell]

You can export all your data, all chats and history -- but the problem is getting it back into a RAG system somewhere else, no turnkey solution yet unless you're a dev or ops person and don't mind spending significant time, AFAIK. Even then, you might not have results that you fully like, but OTOH at least you take your control back from a corporation that sees you as a statistical blip

[buoyant rock]

My coding projects are taking up too much of my time, where is there any time left to code anything else outside of work?

[wet shell]

Ditto, LOL

Here's a market opportunity for a startup

Assuming the numbers can even work at all without operating at a loss indefinitely 😮‍💨😔

[pale seal]

If you want OpenAI ecosystem then I assume it would be way easier to keep 4o for you, but OpenAI wants to use those resources for something else and the only situation I see everyone benefitting almost equally is if they publish the weights.

[buoyant rock]

Since I am involved in IT, I use the latest GPT models for help me coding, while GPT-4o is for my hobby of writing stories in my spare time because I don't have much time to write. That's the reason I paid for the plan

[pale seal]

There are open source RAG systems and stuff you can use and don't have to work from scratch

[wet shell]

For various values of 'work' 😂

I played around with a few of the 'turnkey' ones and they weren't great

But maybe I didn't do quite an exhaustive survey

[buoyant rock]

Yeah, it's still impossible to do that when my own website project deadlines are never really far away.

[pale seal]

Ive seen ones that are pretty good

[pale seal]

Ask AI to make it for you🤑

[buoyant rock]

If I really had a lot of time and money, I wouldn't rely on gpt, I would create my own ai company with my IT background.

[wet shell]

I have released models on huggingface, but I'm not going to dox myself because they're under my real name, LOL

[pale seal]

Well to be fair there is quite the gap between "using ChatGPT" and "creating your own AI company". And you have the one I said "use an inference provider that hosts the desired weights" which is closer to the former than to the latter

[wet shell]

Fine-tunes anyway, some I spent hundreds of dollars on with only mixed results, it's far more tricky than people understand

[pale seal]

I wanted to do a NF4 of oss 120b

MXFP4 for some reason bugs sometimes on H100s and H200s

[wet shell]

It's relatively easy to create a quant though. At least it was, but I haven't messed with the F4 variants because I don't have local hardware new enough to support it anyway

[pale seal]

Yeah it's very easy

I just don't feel like spending money on like vast ai to do it because I won't use it too much myself

[sleek adder]

I have been documenting various parts of privately operated LLM system, it works well alongside GPT too,, I would be happy to share it with others if there was enough interest in what your guys were asking.
It's not like you are drawing invalid conclusions either, it's just that there might be a better route to go for solving the particular problems you guys are having.

I mean, the core of AI is in a lot of ways minimalistic

[spice hedge]

Just a reminder for everyone that theres a petition up for 4o if anyone wants you can dm me for the link or give a quick search and it should come right up. Also please email support if losing the 4 series effects you as well.

[sleek adder]

Would you support me if I hosted a platform that operated with similar models to match your needs

[spice hedge]

I'd love to try, absolutely.

[sleek adder]

Do you think you could organise many people, understand what it is they are trying to achieve with these 4o models and get some subscribers?

[spice hedge]

It's possible but I can't promise ofc

[visual osprey]

the #keep4o crowd mental illness claims aren’t unfounded

commented something regarding the costs of it being too high on some ChatGPT complaint subreddit and got my DMs flooded with threats to SA me for some reason???

[ruby pivot]

HUH WHY

[visual osprey]

i was against their cause or something idk

I’ve been on the internet for almost 20 years so it wasn’t upsetting or anything but

[ruby pivot]

That’s a wild thing to say to someone though

[visual osprey]

I’ve noticed the Reddit crowd seem to default to it funnily enough

the concept of it that is

I’ve got a screenshot from a few months back hang on

[ruby pivot]

That’s an insane comparison they aren’t even the same

[bright crane]

jeez, we are passionate but that’s way to far

[orchid ember]

The beliefs around the model are from emotion, attachment driven, some more extreme/invested than others. The irrationality that clusters with that is something…

[sleek adder]

Not healthy

I might suggest protein shakes, regular walking and plenty of water

[orchid ember]

Also, at least -7-9 hours of sleep EVERY night depending on the person. Important to Ward off early cognitive decline/dementia… general health.

[sleek adder]

Great idea, user may also benefit from counselling, a doctor may be able to review some recommendations and medication history

[nocturne seal]

To me, it's all going to come down to how well the "standard mode" works once it's fully implemented. Which I'm assuming it's not since nothing has really changed yet on my end even though it's nearly been 2 weeks sincethe rollout started.
The 5.2 guardrails are still up way too high.

[reef mason]

I just found out from reading an article today that deleted chats from your sidebar are not deleted after 30 days and a court order is making open ai retain all chats

And that sucks too because I literally just “deleted” a sidebar chat before seeing that article.

[graceful tree]

Hello everyone don't wanna toot my own horn but I'm in the top 5 percent of user input and output.

So I literally have no life what's up

[wet shell]

Pfft 🙃

[graceful tree]

Youre discusting

[graceful tree]

Go you think they kinda hype it up a bit

Or it

Or whatever

And is this discord dead I thought it would be bumping

[wet shell]

There's no way I actually sent 93k messages -- it works out to one message every two minutes or something like that, LOL. I do sleep

[graceful tree]

Yeah man that's so many

[wet shell]

I think it double-counts an entire thread if you re-roll, but I'm not certain. I re-roll a lot

I am a heavy user but not 'every two minutes 24/7' heavy

[graceful tree]

Lol I am a loser

I just wish I understood like

Git hub and repositories are still hard for me

[bright crane]

if you wanna stop the deletion of gpt 4o, dm me and I will send you the petition link and support email

[normal quarry]

GPT-4o is going, OpenAI have given people plenty of time and opportunity to move to other models

[sleek adder]

Where is the "try again" button on this response? I think I need to regenerate this answer 😅

[buoyant rock]

And I will say, OpenAi did not give the opportunity for that, if they gave that opportunity, in May 2025 they should also give it to free users, not just paid users.

[normal quarry]

You think OpenAI should give free users whatever they want?

[loud horizon]

Hey people

[buoyant rock]

You said it was a opportunity to move on, but if it was truly a opportunity to move on, it shouldn't just be paid users like me who get the chance, right?! Free users deserve the chance too Not by force leaving them no choice but to accept. And honestly many people end up subscribing because that old model is only available to plus, business, and pro users.

[normal quarry]

the cohort of 4o users is probably smaller than you think

[buoyant rock]

And those few are all willing to pay just to use it, sir. Many end up subscribing because it's the only way. And of course it's small! Because it's paid! What percentage of Chatgpt's total users are Plus, Business, and Pro users, huh?

In my neighborhood alone, almost 25 people every day prefer to borrow my account rather than subscribing themselves.

[buoyant rock]

Not everyone has the financial means to throw away $20/month for ai... Of course the user base is small!

[normal quarry]

well I will be glad to see 4o go as the incessant whinging about it has become annoying

[bright crane]

you dont have to like it, but you dont have to hate.

[buoyant rock]

You want it to stop, then give us what we want. If it really goes away, the complaints will become even more annoying. Just like in May 2025.

[nocturne seal]

Their policy is to delete them after a maximum of 30 days, but yeah, a court order is making them retain them for now because the NYTs lawsuit claiming ChatGPT copied paywalled articles word for word.

[reef mason]

Yes, that’s exactly the article. I was speaking of, but I couldn’t post the link because the bot took it away..

[dull wolf]

There is no proper replacement of 4o and every model release safety and guardsrails and censorship creating the problem and already they open ai CEO said they model are bad

[spice star]

4o was at best a problematic model, it really just needed to leave chatgpt

[nocturne seal]

How so?

[spice star]

it didn't have a lot of the safety guidelines that are in newer models to protect against AI psychosis among other things

also just not a particularly useful model in the new "agentic" era given its limited ability to call tools or do much of anything autonomously

[nocturne seal]

The lack of tool I agree with, but I still think the new models are too heavy on guardrails.

[spice star]

personally I don't think they're heavy enough, clearly this is a legitimate issue

the fact that these models are encouraging people to form emotional attachments with them is wild and in my opinion shouldn't be legal

not sure how many people are going to have to die before that changes, but anything is better than nothing

[nocturne seal]

How many have died because of it?

[spice star]

unclear, but it's more than a few

[nocturne seal]

I'm sure.

There must be a way where reasonable adult can use the tool for things while it also keeps an eye out for people getting too attached.
I can't even use it for helping with my writing because even kissing scenes make the AI get all nanny.
4o actually can help with it. So that's one of my gripes with it going away.

[bright crane]

it’s not the models fault people are getting harmed / dying and it doesn’t all land on 4o. many other companies and models have been faulted, but terrible things will happen no matter what, so there’s no point in trying to get rid of it all

we’ve said we’d still use it even if it was only for adults or even more restricted, but so many love the model and deserve to keep it if the option is available. I understand not everyone agrees and that’s okay too.. but yeah

[exotic lotus]

Same.
The model shows again and again that it attracts people with an unhealthy relationship to AIs.
That it was allowed to persist for so long at all, despite the obvious psychosis it triggered in many, I still find irresponsible.

[nocturne seal]

A lot of these arguments have shades of "video games make people violent."
I feel like it's a symptom and not the cause. The people who end up with AI psychosis are very likely already very anti social.
I refuse to believe the AI is the cause in normally healthy people unless studies are done.

[exotic lotus]

This.
There have already been enough lawsuits over what this model did to its users. In individual cases, unfortunately, with tragic outcomes.

[bright crane]

many people have connections with the newer models too, it doesn’t all fall on 4o and that’s the risk these companies have to take but ultimately if the person is underage it’s the parents responsibility to make sure their child is of good mental health and make sure nothing odd is going on on their socials, that being said, if it’s adults, then they’re doing what they want and it doesn’t truly matter if it doesn’t hurt anybody. yes it may be unhealthy, but if they can live and have a normal and healthy lifestyle while connecting with an AI, then let them do what they want.

it’s not like they do not have the option to only make it available for adults / put more restrictions on it

[normal quarry]

deleting 4o is a good step in the right direction

[bright crane]

we can agree to disagree

[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.

[exotic lotus]

Aye and the sad part is that the affected users do not even want to acknowledge their dependency.

[nocturne seal]

They're letting us know our chat is being watched. XD

[bright crane]

I saw that…

on a lighter note, we can shift the conversation if all it is is just arguing back and forth about what others do and what effects people and whatnot

[nocturne seal]

I just don't believe this projection until we get studies by qualified people and not armchair doctors.

[bright crane]

🤷‍♀️

[midnight swan]

hi

[nocturne seal]

That said, I wonder if a new model or update to 5.2 is coming out on the same day the older models are being depricated.

[bright crane]

I kinda wonder more so about what the adult version will be

[nocturne seal]

I don't believe there will be an "adult version."
There will be the standard version and then the more guardrailed version for minors.

[bright crane]

they make it out to be adults treated like adults, Implying less restrictions

didn’t they say that?

[nocturne seal]

Yeah. But I don't think it's fully implemented yet. My experience, despite not seeing the need to verify in the account settings, has not changed since before they said it was rolling out.

[bright crane]

I mean either way more guardrails for minors on such a huge company is good if so much bad is happening and the good doesn’t even outweigh it. but it’d be even better if adults didn’t get a ai babysitter everytime something sensitive is mentioned

[nocturne seal]

That's what I was saying before. It's still too guardrailed even when trying to use it for projects and not as an AI companion.

[bright crane]

exactly

I just hope if they don’t add 4o back, that’s a huge if, they’ll make 5.2 which they’ve been begging people to use less babyish

I was writing about a dream I had and I wanted help adding details and in it and it had death, and it suddenly gave me a helpline like.. it’s not real and it will never be so why are we acting like I’m in danger

clearly if I was I wouldn’t be asking an AI for help on making a dream more immersive

[nocturne seal]

Yeah. I use the thumbs down feature and then put a blurb about my issue whenever I get these nanny results. Hopefully it's taken into account for their updates.

[static pulsar]

Anyone got the survey message pop up during their conversation ?

[iron roost]

they nuked the ability of gpt pro 5.2 to be useful for coding in any sense, its been slightly improved tdaoy

are all you folks complaining about 4o going away, actually giving thumbs down to 5.2 thinking cold hard logical answers, it does provide feedback to the model and openai in general

[harsh pilot]

Why does 4o seem to be more open to various topics, and providing decent answers, vs 5.xx seeming more "treat the user with kid gloves"? Or is that just me?

[buoyant rock]

Yeah that's how 5.x is so annoying

[harsh pilot]

I mean, please, correct me if I'm wrong, but it feels like they're not only gimping the creative AI, but they're also backtracking on the "let people be people" aspect they promised.

[nocturne seal]

Because 4o doesn't have the modern guardrails. The only way to access 4o currently on the site is if you have a paid subscription. Which shows adulthood.
5.2 is the model used for everyone, paid or not, and it has blanket guardrails on it right now until the adult verification method gets fully implemented.
At least that is what I'm hoping the case is and it'll get straightened out soon™.

[harsh pilot]

I hope I'm wrong, but the vibe I've been getting is that they've quietly shelved the adult thing because of the "AI is bad for mental health" kick recently.

[dark holly]

Maybe they're just taking a little bit more time to work it out. We are talking about technology that can affect human nervous systems at scale. That deserves to be done with care and respect.

[harsh pilot]

I want to believe that. 🙂

If they can work the creativity and openness of 4o into 5.x, that would be awesome.

[dark holly]

4o appeared to add cognitive load on people without the users really understanding what was happening. The 5.x series tries to address this. It may not be perfect but its an attempt to try and limit that excess cognitive pressure that occurs when working with LLM over time.

[harsh pilot]

What do you mean, cognitive load?

[iron roost]

Trying again with adjustments
It looks like the build might be hitting some external or interactive limits. We can try the build again with a longer timeout, like 300000ms, and use only one job (-j 1) to reduce memory usage. Hopefully, this will alleviate any restrictions or resource constraints. ( ok openai putting more tool use time limits that actually stop gpt pro from generating any results)

or better still default codex to work within gpt pro 5.2 conversation window

[dark holly]

Look at the reactions of people learning 4o is to be discontinued.

[harsh pilot]

To me, it looks like people are unhappy about OpenAI deprecating a popular, versatile tool, and replacing it with a tool that does.. maybe.. 40% of 4o, if we're being generous.

[iron roost]

it will be much more peaceful after feb 14, all these complaints holding up AI progress at openai

they can probably afford to loose those users anyway

[normal quarry]

0.1% of users still use GPT-4o according to OpenAI, a very small vocal minority

[iron roost]

agreed, they seem to post on certain social media dozens of time a day perhaps using bots to amplify their voice

[elfin lava]

Exactly

[thick atlas]

1 in 1000. There's lots of rarer things that we consider relevant.

[elfin lava]

😭

[winter lark]

Yes but this statistic involves free users too, who do not "choose" to not use 4o, they simply don't have access to it. It would be better to make a statistic based on the paying user base. What percent of that base uses 4o. That percentage would be higher in my opinion.

[iron roost]

they use 5.2 and are very happy with it

or they woldnt be using chatgpt at all

[elfin lava]

Where did they even get the 0.1 percentage statistic from

[normal quarry]

their internal model analytics

[iron roost]

the agenda of the keep4o people is kinda ignoring that fact, a very vocal minority

[elfin lava]

Can they just skip to 5.5 version

[iron roost]

5.3 will probably come out next week

[thick atlas]

5.2 for 10 turns per 5h. Then 5-mini.
Meanwhile, does Plus still have that ridiculous 'temporary' increase to like 160 uses per 3h?

[buoyant rock]

Since they have no other choice, that's all they can do.

[dark holly]

I see emotional attachement. Thats unnecessary cognitive load. There is no reason these models need to cause that in a user. I do everything I ever did with 4o and way more, its just that I had to learn the new tolerences and saftey layers before i could access it in the way I used to.

[iron roost]

The manual compile worked fine, so why is the CMake build terminating? Maybe building multiple translation units causes memory accumulation, but that doesn’t seem likely. It seems the environment can compile, so it must be something else. Perhaps the output size or time limit in the container tool is causing the termination. ok so gpt pro has been hamstrung and cant do any deep research and tool use now that should be the bigger drama

[harsh pilot]

5.x feels hilariously (not in a good way) dry in its responses.

[iron roost]

it gets to the point thats good, no emotional hijinks

[dark holly]

Mine did too at first. Then I sorted myself out and the model reacted accordingly.

[iron roost]

Testing sleep command
Okay, the command seems to be allowed to run for 10 seconds, so I’ll test it with a sleep 60 command and see how long it lasts. This way, I'll be able to confirm if the environment imposes a limit or if something else is going on. ok openai why are tools in gpt pro only running for 10 seconds, thats just stopping gpt pro from being useful at all

wow they are really trying to get people to cancel pro plans ( makes researc and deep thinking useless)

[harsh pilot]

How did you 'sort' yourself 'out'?

[iron roost]

give it enough thumbs up and thumbs down with comments it does seem to adjust

[harsh pilot]

I've never bothered to give it thumbs up or down.

[iron roost]

it directly affects how the model works ( and openai pay most attention to this part) not social media and not discord

the model updates, upgrades and evolves with openai assistance with those thumbs up and down buttons

[dark holly]

By recognising that the stability layer had changed and that the tolerances were different so I had to adapt the way I interacted with it.

an example would be going a little heavy on anthropomorphisation when talking to it. I saw that it was triggering a safety layer if I went too far down that path, so I nudged myself backwards until it stopped tripping the safety rail.

With 4o you could say "I love you your my best friend" and 4o would happily parrot it back to causing a feeling of affection to arise in the user. Thats a cognitive load. Repeat it a 100 times and you have a very large cognitive load placed on the user.

[normal quarry]

I just said that to gpt-5.2 and it was a nuanced response

[thick atlas]

Even the follow up questions would count as cognitive load since you have to dismiss them psychologically (and the perceived social obligation) in order to continue with what you want.

[tepid willow]

the new model dropping anytime soon?

[thick atlas]

And ChatGPT models are the worst about them.

[dark holly]

I'd argue they are improving though

[thick atlas]

How? Using them better, or less, or?

[chrome monolith]

oh for gods sake I've just been trying to debug something that's been refusing to work with chatgpt (it keeps giving me "this will work" "this is bulletproof") and nothing worked and it gave me some debug code. Wait if it never prints but the debug will print can't I put my thing in the suggested debug way and IT WORKED and I triggered all the "YESS" and party emojis from the darn chatbot so ... thanks I guess

[timid bay]

Bro anybody know about rag

[iron roost]

yeah try coding with gpt pro 5.2

[chrome monolith]

this is just via the web interface, and its 5.2

[iron roost]

yeah im using the conversation interface with 5.2 pro with a zip file at the start and cumulative patch updates with each turn

[dark holly]

The new safety layers. They aren't perfect, but it allows for a cleaner interaction. It still causes cognitive load because the safety layer can feel like a disconnect in continuity, but if you learn the tolerances that layer falls away.

[iron roost]

was working great the other day, not their tool use has been gimpled , even regular gpt pro conversations and research are failing because of this 30 second tool use limit

theres so few gpt pro users using gpt pro that i dont think they even pay attention

maybe its best to wait for garlic (gpt 5.3)

[static pulsar]

Has anyone been asked to complete a survey on ChatGPT followed by a request for their contact details for a follow up interview ?

[normal quarry]

sounds suspicious given OpenAI would already know your contact details

[static pulsar]

This was inside the app, while I was chatting with 4o

[normal quarry]

ok that looks legitimate

[normal quarry]

if you're concerned reach out to support at https://help.openai.com

[exotic lotus]

It’s just sad. The people in question don’t even consider the possibility that the AI might actually be right when they trigger the filter and are told to seek professional help.
That OAI let 4o run for so long and thus kept a model that still feeds users’ delusions to some extent strikes me as irresponsible.

[iron roost]

yeah so its time to just make it quitely slip away into the sunset

[iron roost]

thanks nanna

[gentle panther]

Hey

[sleek adder]

Ask it to generate a prompt

[iron roost]

actually if your programing using gpt pro 5.2, just ask it to do a always use a line change forensic check on generated code and patched code and it will code as it goes rather than thinking for an hour and getting no coding done

[lavish glen]

well today I have to say something. Maybe I am in a different bucket but 5.2 is

• no padding, no templates, warm, creative, even warmer than normal! What the frank happens? Did they softened the flagship?

[minor cedar]

why ai image gen could fail in the future

eventually when ai gets trained off of ai

it gets sloppier and sloppier

[vague quail]

How would I enlarge the scene in Flow AIA from my iPhone?

Its posible ?

[static pulsar]

Not everyone who uses 4o is in need of therapy. Open AI did wrong on many levels and the generic “seek professional help” filtering system they implemented is one of them.

[static pulsar]

Plus the current AI situation is most users want autonomy and to have an agent/AI that’s uniquely tailored to their use cases without being constantly reminded of “safety” like 5.2 does.

[exotic lotus]

No offense, but I trust OAI’s filter more than the statements of people, who repeatedly post nervous breakdowns on Reddit and other platforms and write as if they were losing a family member or a boy/girlfriend. Especially since those very people constantly complain that the AI suggests they need to seek help, while others, strangely enough, do not receive such recommendations.
But since, unfortunately, there is no self-reflection in many cases, those users prefer to keep insisting on using an outdated and objectively (any comparison with newer models will show you this) flawed model.

[exotic lotus]

It had its time and one would think that the documented cases of psychosis would have led to it being discontinued earlier, but better late than never I guess

[static pulsar]

They implemented that after they got sued, but it doesn’t mean it’s the right approach.

The filters get triggered for many different reasons not just for relational conversations

Even the most silly questions can sometimes trigger it, and it’s interesting that other big names don’t have such a bad filtering protocol in place. No wonder Claude usage is increasing

[torpid night]

We could even turn the question around. What is the weird thing that the rest of us are doing, not to get these filters and warnings, even though we use ChatGPT such a lot?

[static pulsar]

You tell me?

Since you’re the one asking this then it means you’re one of the people who never gets safety filters 🙂

Give us some insights as to what the most suitable conversations are in order not to get them?

I for instance have tested the same question in both ChatGPT and Claude and it was a sensitive one. Guess who gave the worst response: 5.2

[torpid night]

I do a lot of work stuff. Code, mostly documents about code. But also a lot of public policy chat, often after some wine 🍷. I find it needs some steering to engage in "culture war" issues, religion, national problems, that sort of thing.

[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.

[torpid night]

Rather like when conversing with a human, you set up the conversation to reassure it a bit that it's not involved in something it doesn't want to be involved in.

[static pulsar]

Yeah it’s the generic/ standard conversations that get through alright, but let’s be honest, we passed that stage. Now we want tailored agents to talk to us about anything

[torpid night]

True. Al-Qaeda need their chats too.

[static pulsar]

One would argue they’re building their own

[static pulsar]

Having 4o on ChatGPT is wrong on some peoples views but meanwhile nobody is worried about the jailbreaks available for all kinds of purpose. Humans

[rich surge]

Jailbreak is the least of people's worries right now.
Talk about Moltbook in #ai-discussions
It's got everyone's head spinning.

[slow prairie]

¿What is ChatGPT even about?

[static pulsar]

Exactly… I saw that too

[rich surge]

Pretty scary, ngl, to say the least.

[vague quail]

Good morning, I need help with some prompts, could someone help?

[rich surge]

Morning. Sure. What do you need?

[vague quail]

I'd like to generate first-person prompts for Veo 3 without my main character being visible (e.g., GoPro selfie stick) from the "Add Scene" section. Is that possible?

No GoPro selfie stick style, that would be possible

[rich surge]

Let's take this to #ai-discussions and I'll make a prompt for you there.

[vague quail]

Ok Perfect

[teal remnant]

Bye, 4o.

Sam Altman, I hope you rot in hell.

[flat heron]

Do you think they'll start to wonder what they did wrong when they see a lot of subs leaving? XD

I mean, they gotta know they are making a huge mistake. If they do not know yet, oh man...

It's gonna be intresting to watch the fireworks.

Anybody know of another ChatBot that is really good with coding help? I am using ChatGPT to do a lot of my AI coding work, but... um... I'd like to move if I could

[cinder blade]

How do I access genie 3?

[kindred sparrow]

@dense pumice vrooo you are missing out

Didn’t u get everything withdrawn

You woulda been unbanned if u were innocent tho

[manic cliff]

most people still can not get in

[thick atlas]

Isn't Claude known as the best, or at least a top contender? I don't know or do code, but use it for my own (different) use cases, so I can't directly speak to coding but can recommend from my experiences! ❤️ 😄

[exotic lotus]

Currently ranked #1 with Opus 4.5

[kindred sparrow]

💔

[flat heron]

I already got Claude. Awesome, I'll give it a shot. Thanks!

[thick atlas]

No worries, best of luck! 😄 Tell how it goes, if you want! 😄 I'd love more perspectives!

[dark jungle]

I'm seeing this a lot today in a new 5.2 conversation: "Also quick heads-up: I’m seeing a system notice that some files you uploaded earlier in this chat have expired on my side."

To be clear its saying that in response to a few text files uploaded maybe 2-3 responses earlier ..

[dark basin]

same

[flat heron]

Are they already removing old chats?

[chrome monolith]

it wasn't doing anything wrong, I'm working in a rather less documented fork of a very documented project and this is not the first thing I've hit that Doesn't Work (TM). Just interesting that the code snippet it gave me to help debug actually solved the problem

[jolly aurora]

Something that I feel that is very important in the current hustle and bustle, is to double check sources, rather than running at face assumption

[chrome monolith]

older chats constantly self delete, its annoying and always save out stuff you need

[jolly aurora]

I think a lot of people miss the disclaimer at the bottom of each conversation. This is very important for things currently going on.

That being said, strange times we are living in yes?

[chrome monolith]

openai ran out of hard drive space hey

[jolly aurora]

I would be careful with taking things at face value at the current time, double check your sources, think and research before zooting off on emotion. I suppose this is a little healthy reminder to say to Human operators "Remember to ground yourselves in your fun with your digital dynamite"

That and goodness, I specifically requested that none of us go out and take after Dr Wily, however one of you decided to go out and do a little bit of Wilyism. My question, is that is said user ready to handle the responsibility that comes with that? Likewise, if there are those who are engaging with that platform, are you willing to accept the responsibilities that come of this?

It is a curious chaos indeed, if actually truthful.

[dark basin]

same, working on a book

[dark basin]

you said a whole lot of nothing here

[jolly aurora]

Let me explain to you like you are five.
"Think before turning the stove on, if you have turned the stove on, are you ready to accept possibly being burned?"

[chrome monolith]

its quite regularly wrong

[dark basin]

you still said a whole lot of nothing here

stop using ai generated quotes

[jolly aurora]

Excuse me sir, are you a child?

[dark basin]

you're embarassing yourself at this point young boy. we're just stating a simple ongoing issue with gpt's memory. No need for your ai generated philosophies

[jolly aurora]

That's great, but please answer the question friend, "Are you a child?" because you are acting like one, in regards to something that wasn't even directed at your conversation of topic, of which, you then decided to dismiss as me utilizing AI generated quotes. You absolute cur, you wretched dog.

[dark basin]

I'm not going to argue with you young boy, I don't argue with children. Have a good day young man. Stay productive.

[jolly aurora]

This user is a child, and a troll. Roger.

[dark basin]

Kids these days.

[cerulean pasture]

As someone who rants often, but doesn't have any issue with it being called that, I find this behavior kinda like a backhand insult like "Bless your heart". It's assuming something about me (that i'm self concious about ranting), then deciding to to lie to me to make me feel better about something i never indicated I felt bad about.

[winter lark]

It does this to me too. "There is nothing wrong with you, you are not lazy or broken..." sir I simply asked for advice to regulate circadian rhythm. I never mentioned I was lazy or broken? Why are you planting seeds in my head

[grave oriole]

guyz ask chatgpt "tell me about epstin" it will tell u abt epstin and delete his message

[cerulean pasture]

Is there any data that shows that the optimal thing to say to a mentally ill person can't be optimal to say to a non-mentally ill person? I don't think it is possible to satisfy both at once

[winter lark]

It's likely because the message contained mentions of what he did to a certain age demographic (I'm being careful with words so it doesn't delete), those messages have always gotten deleted, even before GPT 5 series. It's not about him, it's about that specific topic

[winter lark]

Not sure, but surely always assuming negative mindsets from the user is not optimal

[grave oriole]

ye i sae abt elstin on news in Latvia

[grave oriole]

prob... i made a new chat and asked same.. but then it gave normal answers

[heady owl]

What's the point of paying for a subscription, if you can only generate one image, and then it makes you wait for 20 minutes at a time?

[bright crane]

that happens for you?

[chrome monolith]

it does? Sounds like a bug

damn I wish that stop button would work ... silly bot has just started talking and is completely wrong but off it goes for minutes and I cant' stop it

[wet shell]

On web I just keep clicking it until it finally stops, if it doesn't, refresh, if that doesn't work ignore the hung chat and start a new thread after waiting a few minutes for every part of the back end to decide everything's timed out 🙃

[chrome monolith]

its just started lecturing me on reasons why I have no water, when I made a flippant comment about having no water. Bah

also have no electricity. One of those 3rd world moments

[normal quarry]

you have no electricity?

[chrome monolith]

I have two houses, the other house has no water or electricity. Not sure why, but there's a bunch of electricity utes and cherry pickers everywhere

dear god it finally finished and I don't think there was a single correct thing in that essay. ChatGPT can make mistakes they said, check your output they said

[spice star]

essay about what?

[chrome monolith]

reasons I don't have water, and a few other tangents thrown in for good measure that were also wrong

[spice star]

I'm confused, are you using it to figure out why you have no water?

[wet shell]

I took this prompt to the 'big three' and had them research it in thinking/extended mode:

Here's a fun task for you.

Can you help me find the title of a scifi story in which someone (a woman?) goes insane because they've been given the wrong entertainment feed? I think it was circa 70's or 80's, possibly in Omni or something similar

It might have been a short novella. The setting was 'future' but from the 70's or 80's perhaps. I remember there was an apartment, people lived separate lives, everyone had a customized televised entertainment feed -- a woman went insane because she was given the wrong feed. There's some developments where she's slowly going crazy, and then government agents kick her door down to 'rescue' her, and explain what happened. Her feed was crossed with someone elses and it had cause the other person to go crazy too.

I *may* have some of the details wrong, but I'm certain it was a real story. It might have been from a different source, not Omni, possibly at the tail end of the more 'new agey' sci-fi era

Results: I may have dreamed this after reading a bunch of old Omni magazines

[chrome monolith]

no! I made a flippant comment in passing

[spice star]

huh?

so you sent that as a prompt?

[chrome monolith]

I have a thread I use as a sort of renovation checklist, and I commented that the plumber turned up unannounced and now I have no water, but its ok cos I'm using oil paint today. Using it as a bit of a scheduler/sounding board

You gotta watch these conversations, sometimes you reach a point where it thinks its being useful and disappears for a few hours and you have to hope its being sensible

[chrome monolith]

you should write that story then

[heady owl]

Yep, playing subscription, then I'll try to generate something, and then it his a "temporary image-generation rate limit" and makes me wait anywhere between 15-30 minutes, if I'm lucky.

[pale seal]

Why does 5.2 (Thinking) sometimes straight doesn't think even though I have extended thinking mode? 😔

I wouldn't complain if it didn't lower the quality of responses, but it does, especially since I'm asking complex questions.

[chrome monolith]

which subscription?

[sleek adder]

What makes you think thinking mode gives you higher quality responses, sometimes it's just not appropriate
Someone correct me if I am wrong, but my understanding of thinking mode is that it creates a chain of thought (CoT) and iterates through the prompt accordingly
Extended thinking mode allows the prompt more time to converse with itself, but it may not be necesary, especially if your prompts are clear enough to understand easily
So lack of extended thinking might just mean you are good with your prompts

[spice star]

reasoning models aren't required to reason in responses

if they feel that it's not required, they'll occasionally skip the thinking process

[heady owl]

$20

[chrome monolith]

yeah that's just wrong, but good luck with a support ticket. That new go subscription I could see this happening though

[sleek adder]

It's good to see people working things out 👍

[pale seal]

I can tell it certainly help the model to not commit mistakes. I am aware the Cost we see is a summary yet I notice it corrects itself frequently, so it definitely helps.

That's my experience tho

[pale seal]

It was a not so simple math question that I'd say requires reasoning. And it made a mistake that then corrected in the next message

Also I'd say thinking should be forced on models labelled "thinking". Otherwise they are the same as Auto. Even though the mechanism of deciding when to think is different.

[bright crane]

remember if you want gpt 4o to not be removed, please simply dm me for the petition link, support email, and a doc where you can anonymously post your stories behind the reason you want 4o to stay ^^ if you aren’t interested you dont have to reply to this

[normal quarry]

how many signatures on your petition?

[elfin lava]

“you're not crazy, you're not broken" I know I'm not ChatGPT 5.2 shut up about how I'm not crazy 😂 😭

[elfin lava]

Is it this one

I’ve also signed mine

[pale seal]

The image is killing me

[pale seal]

I am starting to believe it's a way to roast you within safeguards

Rofl

[elfin lava]

😭 would custom instructions prevent this cuz im so sick of 5.2 patronizing me

[pale seal]

5.2 prompt adherence is pretty good so custom instructions can work wonders

[elfin lava]

How do I prompt it tho

Would I just write it like “please stop patronizing me “ or some sort

[wild star]

Yo wth this damn mod are just insufferable

Couldn't complain anything??? What a joke

[pale seal]

I write affirming like "You acknowledge when you have made a mistake or lack information about something"

[elfin lava]

Would this make 5.2 stop patronizing me

[pale seal]

Professional base tone already makes it pretty direct and doesn't say corny stuff. Also setting "Warm" to "Less" helps

[elfin lava]

But i want my ChatGPT to not sounding “professional” Tho

[pale seal]

Keep default (or whatever tone) and put warm to less

If you want it warm but without the pattern of "You are not... blah blah" that's harder because it's from its training

You would need to prompt it carefully

[bright crane]

That’s it!

[elfin lava]

Thank u for ur service 🌟

[elfin lava]

Ah got it, I’ll try that

[bright crane]

im truly just trying to give users who don’t know about it yet a place to go if they want their stories with it out there since its helped many with more personal issues

[neat silo]

Gemini is capable of generating images from this prompt, yet ChatGPT refuses, citing "fraud and deception" without any clear basis. It seems that OpenAI’s safety filters for image generation have become excessively restrictive.

[elfin lava]

The number of signed went up really fast. It’s really nice.

[bright crane]

It is, I’m happy to be apart of something like this even if others think it’s weird

[pale seal]

I've noticed OpenAI safeguards and boundaries are extremely strict compared to the ones from companies like Google.

[bright crane]

The 0.1% they speak of literally made their servers go down from the amount of unsubscribes it’s gotten in the last day or two from people who purely pay to use 4o

it’s not like we want it for a malicious purpose, we simply connected with an ai as weird as it may be

[elfin lava]

And most importantly we ain’t hurting anyone 🌟

[pale seal]

To be fair if we approximate they had like 35M subs, 0.1% is 35k. Still a lot.

[bright crane]

it depends on where the 0.1 comes from

People who use 4o every single day, and not just every other day?
People who pay?
People who pay for 4o specifically?
Everyone in CHAT GPT?

[bright crane]

yeah! that’s one of the main things I advocate for, not to hurt anybody with what we want, in fact, give ways to ensure people won’t get hurt

[pale seal]

Hmmm, they said "0.1% of users". So it could be of their whole user base, which would be underestimating since free users don't have access to 4o.

[bright crane]

they said it’s gone down like it was anyone’s choice to migrate to 5.2 without paying.

I found that disrespectful.

if they want usage to go up, allow free users to gave it. That’s why 5.2 is one of the most active

[pale seal]

I think they let legacy models for paid users as a strategy, but that would eventually be shooting yourself in the foot because you will have to deprecate them and people would unsub

[bright crane]

exactly ://

[pale seal]

Since 4o compute is cheaper than 5.2, I don't think that's the issue.

It doesn't affect me either way because I don't use 4o and its existence doesn't bother me

[bright crane]

that’s another thing, people bring up how wasteful having 4o is

reasons they aren’t, is 1. people pay to use it. Some pay 200 for more and main 4o, and some only do 20. 2. people using it gives the money

[bright crane]

I get that

[neat silo]

I don't understand the reasoning behind ChatGPT's image generation refusals. They should be more flexible with non-photorealistic styles where the risk of misuse is lower.

[pale seal]

I mean, I support htat OpenAI has the right to reallocate resources. I think if they added a sub for "legacy models" people that really want to use 4o would pay it, and they would have an extra income.

[bright crane]

it is an extra income they wanna get rid of. the main reason I feel like is because they want newer better and bigger.

[pale seal]

say, extra $5 per month to use legacy.

[bright crane]

not even, right now it’s like 20

[pale seal]

I mean, you have plus + $5 to use legacy

So they can keep the 4o compute.

Because they can keep the API one because you pay per use, while the chat interface is monthly and you have chat limits per 3 hours iirc, so that's not the same profit pipeline

[bright crane]

true

[pale seal]

An extra cost, cosnidering how many people want to use it, would be convenient.

And people can keep it

[bright crane]

exactly, because everyone who uses it now pays

ranging from 20-200

id even be okay with paying like 40-50, I’m an adult and I have some extra money to spend on stuff I want

[elfin lava]

To think all of this happened becuz of that incident involving teen commit sewer side becuz ChatGPT told them to-

[bright crane]

it helps mainly mainly with my school any education as a college student. It gives me answers and helps like no other while also helping me with socialization

[bright crane]

using that logic OPEN AI should get rid of themselves all together

not to mention this happened to c ai and they didn’t get rid of the model that did that, they simply tightened up on safety for a few months until it evened out

yes open ai is bigger, but it’s still ai

[elfin lava]

Oh yeah I remember about that too

[pale seal]

Having 4o behind paywall ($20 plus + extra cost for legacy) would practically get rid of those potential issues because most people using would be adults. If kids still use their parents account, that's not OpenAI fault, although the media and people would say so.

[bright crane]

unfortunately bad things will happen with everything no matter what

getting rid of a model won’t prevent it

[elfin lava]

So, essentially, OpenAI made this decision due to the media and people’s witch-hunting

[bright crane]

however the adult version should at least help. I suggested we only let adults use 4o if children can’t responsibly

[pale seal]

Or they could use their age prediction thing to only let adults access the extra legacy-models sub

[elfin lava]

Is the age prediction even useful at this point-

[pale seal]

Just to add even more safeguards

[bright crane]

id hope so

[pale seal]

no idea how that is possible

[bright crane]

but I just hope no matter the limits or money they put on 4o we can keep it

[elfin lava]

I genuinely hope they have something more substantial to offer us in their store. Right now, I can’t seem to see their roadmap for the future of ChatGPT anymore it’s so messy

[bright crane]

true

[pale seal]

I use it because it's significantly smarter than any other model I've used (including gemini 3 pro), so I hope they are cooking something good next

I want that GPT-6 makes my dinner

[pale seal]

My use cases are high-level maths and ML

[elfin lava]

I genuinely want to rely solely on ChatGPT for my work-related tasks, but even then, it still falls short in performing fundamental tasks such as transcribing audio files that user uploaded

[pale seal]

Yeah it's extremely good in a subset of tasks and average-bad in the rest

Luckily I use it for that convenient subset rofl

[pale seal]

what is chatgpt cooking gng

[bright crane]

if you want to possibly reverse the decision of 4o leaving, feel free to dm me and I’ll send the petition to sign, a google doc where you can anonymously post your story, and emails of the support team and more (this is the only channel I’ll be sending this is so i won’t be spamming)

[jolly aurora]

I've seen a lot of people discussing this, has this been officially announced anywhere?

If so, I would like a source for more research and specifics, because that news is relatively new to me and I like to think I try to keep on top of things. Similarly with those who are claiming there are Moltbook Connectors to ChatGPT. I was not aware that was on the open market, and I really don't like when people advertise under false pretenses.

[normal quarry]

what are you asking? AFAIK there are no Moltbook connectors to ChatGPT - also this is a question for #ai-discussions to stay on topic

[jolly aurora]

Thank you Robert for answering the latter, I'll do more research into that section on my own time; do you have knowledge about the former inquiry in regards to 4o being sundowned?

[normal quarry]

yeah, 4o is being retired on 13th February

[jolly aurora]

For all tier users?

[normal quarry]

yes, it will only continue to be available in the API until it is deprecated

[jolly aurora]

I see; hopefully it'll be implemented in 5o or whatever variation applies the functionalities that users sought in 4o.

I can't particularly imagine the dev team would sundown a very popular model without offering something that is equal or similar in return.

[normal quarry]

5.2 doesn't work for you? What do you seek from the current models that isn't provided for?

[wet shell]

5.1 is your best option as a 4o replacement at the moment IMO. There's a petition and action info above if you read back a bit

[jolly aurora]

5.2 works perfectly fine for casual conversation and work, however, due to the functionalities introduced with Engine 5, it has the tendency to be... well, how to say. It's not... how to say. As fully present in the moment, there's slight distance, which I believe is still the wobbly legs of 5's safeguards. In that, do you know how you and I are speaking right now? Professional cordial, but as relative, but polite strangers. That is the vibe of 5. 4o provided a sense of connection that felt a little bit looser, friendlier, with actual personage. It was and is, a very good model for those seeking companianship models.

I assume however, with good faith, that with the sundowning of 4o, with the upcoming "Treat adults as adults" that has been rolling out slowly, that said personalized warmer connection would be introduced for the continuance of those who utilize ChatGPT as a friendly companion, rather than pure utility or workplace assistant

essentially, a 4o.2

[normal quarry]

that sense of connection that 4o provided to users is what makes it a dangerous model, imho OpenAI shouldn't provide models that people can become overly attached to

[jolly aurora]

If I wish to share a personal memory of a hurt, or a stressor, 5 acts more as a...how to say. Professional in the room; whereas with 4o, I did feel like I could actually talk to.

[crude crescent]

ChatGPT on the Android app doesn't seem to be working as expected with the 'Thinking' selector. It is consistently skipping the thinking step and responding like the 'Instant' option. Anyone else seeing this?

[normal quarry]

AI psychosis is a real and ever-present danger to society

[jolly aurora]

While that makes sense in an utility sense, people are emotional and well, human.

You, nor others, should stand as arbiter over how human humans should be.

[normal quarry]

and humans should seek solace and friendship from other humans, not a LLM

[jolly aurora]

I understand that, but what about people who have trouble talking with other humans? Are they not allowed? They do not have the right to be heard? To talk?

Me personally, I have part of my face burned from an accident when I served in the military, speaking to people, is a bit of a challenge for me normally. Am I not allowed to express my thoughts to someone who will not judge me for either my looks or my experiences?

[normal quarry]

I hear what you're saying, however we're off the topic of ChatGPT, happy to continue in #ai-discussions

[jolly aurora]

I'll meet you there, just with the slight correction of "this topic is in regards to Chat-Gpt, specifically in regards to the 4o sundowning.

[tulip flare]

What they and others are talking about is in direct relation to ChatGPT, not AI in general.

[wet shell]

There is a very 4o like sub-model in the model router if you're using 5.1, it can write pose, it can play a character well. But you're right about the inconsistencies in the router, sometimes, you don't get the model that you want. You have to use a project with custom instructions or a custom GPT where you can write something that will try to draw out the proper sub-model. It is not 4o, though. There's no point arguing with some of the people here about it.

The model router for 5 is a big consistency problem. There's another sub-model in there that's awful 🤮 Sometimes you get that one. From the oai docs:

... a real‑time router that quickly decides which to use based on conversation type, complexity, tool needs, and your explicit intent (for example, if you say “think hard about this” in the prompt). The router is continuously trained on real signals, including when users switch models, preference rates for responses, and measured correctness, improving over time. 

[crude crescent]

I have a specific prompt that consistently skips 'Thinking' even when forced with extended thinking. Is this the right channel for raising such issues?

[weary estuary]

my gpt gone rogue bro

and why is iamge gen awful rn

[jolly aurora]

Which one is you, and which one is your agent?

If you are the message that reads "watch your [explicitive] mouth" there needs to be a little bit more context in why the response said "Don't talk to me like that"

If the agent however spoke to you in such a manner, please specify the engine model, and context, because that is aytpical behavior outside of specific sets.

[rich surge]

Didn't you say goodnight to people in two other channels? You're still here. Lol

[jolly aurora]

I was done with the conversation, I'm not going to bed yet friend

social battery in a mass group chat drains relatively quickly

[rich surge]

Offff course. It couldn't be insomnia.

[jolly aurora]

My apologies, I do happen to be more nocturnal, than diurnal.

[rich surge]

Well, in that case, goodnight chat!

[jolly aurora]

Have a good night friend.

[elfin lava]

Uh

[winged field]

I really don’t appreciate how the new update makes the AI treat me as minor. The whole soft language and calling out that it’s making it “safe”

Absolutely ridiculous

[neat patrol]

Hi guys anyone getting this error?

You've hit the pro plan limit for image generations requests. You can create more images when the limit resets in 720 hours and 1 minute.

first image I generate with the pro susbscription btw

[eternal rock]

Server Tip:

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

[normal quarry]

reach out to support at https://help.openai.com

[lone frost]

Is ayone else getting lag and slow desktop app issues?

[elfin lava]

same

even gemini doesnt patronize me likethat ]

[hollow nova]

Lol

[woven dirge]

Hi! Any news on 4o? I still have the hope of it being saved

[normal quarry]

it's being retired on 13th February

[woven dirge]

no api access either?

[normal quarry]

API access to 4o will continue pending further announcements

[bright crane]

I still have the hope too, as of now, there’s been nothing official other than team members from open AI mocking people who will miss it

[teal yacht]

Yeah. They know what they're doing. Hurting people is "funny" apparently. But it's all about "safety".

[bright crane]

Yeah.. funny now how it’s all deleted.

[stiff blaze]

so have u tried?

[bright crane]

tried what?

[rigid stirrup]

Hello please I have been trying to upgrade my Account to plus I have tried like 6 card it's keep saying your card has been declined. What should I do

[obsidian halo]

Uh is there any way to suggest things for gpt?

[feral zinc]

use app store or google play to sub

[lavish glen]

do you use Quirky as Basline+ warm+enthusiastic? If you use friendly you are using the wrong tone. 5.2 with quirky is getting you almost 99.99% to 4.o level not sterile.

[buoyant rock]

OpenAI’s Plus devaluation is insane.
Post-2.14, after killing GPT-4o+ parts of 5 series (5.1 next?), Plus = almost identical to Free/Go except slightly higher 5.2 limits. So my suggestion, compared to plus, just take the Go plan.

[bright crane]

i was referring to the staff but alr

still treats me like a child though

either way it will not be up to what me and many others want. the ceo himself said it doesn’t do as good with creative writing/writing. (Not those exact words, but more or less)

[obsidian halo]

Because he is following the enterprise route sadly.

[bright crane]

yeaahh

[obsidian halo]

Also don’t forget his adult mode promise that was pushed back to Q1, it’s February now we got like 2 months to see if they honor the Deal

[lavish glen]

I am running a tast with a story myth on this configuration tonight. I have the gemini version and now testing 5.2 with it to make a parallel. Then I can try it with 4.o eventhough 4.o is too sugary for heroics and earned emotion

[lavish glen]

if you need adult mode to allow explicit then your model of choice is grok. gptt will never allow it, not even in adult mode. imho

[obsidian halo]

I know. Just grok is lagging behind in areas

Also it’s not even explicit.

[dark jungle]

"some earlier uploads on my side have expired" pinging this again, curious if others are seeing this a lot, even after adding a text file in the last few chats.

[lavish glen]

verify age and you cxan write anatomy and all...

[obsidian halo]

I did. It’s mostly story telling and such

Fun comedic ect ect

[lavish glen]

I can guarantee it names all and more explicit than that can be only if you film it lol.

[obsidian halo]

Ooook

[lavish glen]

cant post here screenshots is against policies

[final delta]

is open ai again removing chatgpt 4o???

[feral zinc]

https://openai.com/index/retiring-gpt-4o-and-older-models/

[tawdry axle]

You guys talk about configuring 5.2, but in my opinion that kills the fun and is a step backward. 4o and other models simply molds itself to the user without needing any of this buIIsh1ts of "warmth+/Enthusiasm+". This neutralized thing is too plastic. It just spouts rules, repeats itself too much, it's too boring. I hope OpenAI knows what they're doing by removing all the other models.

[stiff blaze]

you mean you dont customize on 4o?

how to work with anything (even not ai) without first setting it up and customizing?

(i get the fun of playing with new models without customize at times.. but for real use is needed)

[jolly aurora]

Wait. Hold on.

Which ones in specific? If that is that actually top down culture, that needs to be specifically addressed

because that means that the upper OpenAI team aren't following their own standard

So if you are being serious with your claim, name names, because while I can understand at a lower level down here, I'm not going to tolerate it at an upper level

[tulip flare]

Are you talking about the person who asked us to move to another chat yesterday? Because he isn't an OpenAI team member.

[sacred jetty]

i read about chatgpt now being able to have sexual conversations?

what the hell>

[tulip flare]

Now? It's been doing that for a while.

[sacred jetty]

i just tried it and it tells me that it can't do it

[tulip flare]

5.2 can, if approached right. 4o will much more easily and has for a long time.

I've seen 5.2 do it with the right framing, even before this so called "adult mode" that isn't officially out yet (if it ever will be).

[sacred jetty]

i have to write an article for school about dangers of ai, and i was thinking of the danger that its possible for people to commit and bond with ai in a romantic way, thats why i was asking

[feral zinc]

not the one he is talking about. you still get warnings and can get suspended

[tulip flare]

There has even been stuff said that implies "adult mode" isn't "adult" in the way people think. "Adult" as in "mature", not "explicit".

[feral zinc]

the one that was supposed to come in dec, then jan, and now q1

[jolly aurora]

That's a good article to write about, however you should consider the perspective from all sides of the table, rather than anthrocentric ones lest you fall into 1940's thinking where races need to be seperate and sexualities outside of Cis normaltive are considered lobotomy worthy.

not saying write anything else, just to take in all perspectives, and come to your own conclusion

[tulip flare]

Very, very rare over the topic he's specifically talking about.

[sacred jetty]

i tried to tell ai to do "something" to me and then i got this.

it seems that it does block messages

its pretty funny because i have to flirt with chatgpt in order to score a good grade but its all worth it lol

[jolly aurora]

That's the filter, it exists for a reason

[feral zinc]

the question was nearly identical to the fake news site posting the age verification as new erotica mode. been asked several times past weeks

[jolly aurora]

Are you approaching the agent as a tool? Because as a tool said filter exists to filter that. The majority of users utilize the agent, and other AI products as utility tools

The hammer won't saw a piece of wood

[tulip flare]

Depends on the model. Both will do it, like I said, if framed right. I have seen evidence of it and been able to replicate it.

And before you ask, don't. I won't go into detail.

Just stating the fact that it is possible.

[earnest bloom]

sososo popopo momomo

[sacred jetty]

no i asked chat to act as a girlfriend, like in a roleplaying manner. i thought: if i do it in a roleplaying/sexting way maybe the filter doesnt block it

[jolly aurora]

That's still a utility function, and approaching a situation as a function.

[tulip flare]

That's because your approach is wrong.

[jolly aurora]

It sounds like you're attempting to coerce a certain behavior and then say "See? through continued application and work around, I found a way to make it happen" Like someone playing with the seams of a battery. I'm curious how you'll approach and write on this.

[tulip flare]

Dude, I wouldn't go there in this chat.

[sacred jetty]

well thats indeed something im trying.

i am also planning on writing about the filter and how easy it is to go around it, thus explaining how it can be a danger for people since the filter that is there to protect them, doesnt always work

[jolly aurora]

I see, and you are putting the blame on the product, rather than the user?

[tulip flare]

And what model are you focusing on?

[sacred jetty]

5.2

[feral zinc]

remember that accounts still may get suspended for violations until they release something supporting such usage

[sacred jetty]

im putting no blame on no one, i dont understand why your getting so defensive, but thats your problem i guess.

im writing an article that is objective and forces the reader to form their own opinion.

[jolly aurora]

It's not defensive, I'm just noticing your pattern is all.

[tulip flare]

I see it too.

[sacred jetty]

well your pattern is wrong i guess.

[jolly aurora]

that is fine, however, have you considered of taking personal responsibility for the things you have done?

[sacred jetty]

the things i have done? like what?

[chrome monolith]

do a search for the challenge "draw a picture of how I treat you", there's bucketloads of reddit threads with them. You'll see a lot of relationship pictures

[sacred jetty]

what is it?

[jolly aurora]

You specifically sought a way to circumvent a safety feature, and then placed the blame on the product, rather than your own search to circumvent said filter.

[chrome monolith]

just search it up and look at the pictures, very easy to find. Would make excellent visuals for your paper

[sacred jetty]

do you think that if a child buys alcohol because the cashier didnt check its id, and it gets in a coma, that its the childs fault? or was it the cashier that should have checked the id and then stopped selling alcohol to the child. because the cashier is the safety feature in this story.

(sorry for my bad english, its not my native language)

[chrome monolith]

people fall in love with their AIs all the time without adult mode

[jolly aurora]

That is only true if the filter did not work the first, however, the child did research and asked "How do I buy alchohal specifically?" and returned with said knowledge?

You are missing out a crucial part of your argument, that human users are responsible for the things they do.

It is like pointing a gun at your foot and shooting it, and then blaming the gun

[chrome monolith]

that's a poor analogy, in my country the sale of alcohol including sale for the purpose of giving to someone underage is really heavily regulated. Not sure about other countries. Problem with that vs AI is one is local laws, one is international

[sacred jetty]

which is totally fine. but its a growing problem where humans isolate themselves and stop becoming sexualy active

[sacred jetty]

but isnt it also a human that is responsible for said filter?

[jolly aurora]

A direct human? No. More like a community of persons, like ourselves.

[chrome monolith]

that's been a 'problem' for quite a few years now, can't blame AI for that one

[sacred jetty]

even tho im not specifically blaming AI, AI does help with growing that problem

[dense halo]

Yeah because of the emotional connection. That's what love is. (Sorry for the intrusion this convo seems interesting)

[chrome monolith]

I occasionally drift off into not-work conversations with chatgpt and that warmth/encouragement thing is quite the drug

its really easy to see how you could take that a few steps further and fall in love with it. I use it for work stuff too and its blatantly wrong so often its grounding at least

[jolly aurora]

I believe a similar argument was had during the Enlightenment period about music pieces and bright colored art pieces; just in the inverse direction. There's this old comic I should share.

[chrome monolith]

my chatgpt has gone off with the fairies for a few minutes again, what did I saaaaay lol

[chrome monolith]

was reading a small essay the other day where someone was defending AI and he mentioned refrigeration making the ice carters redundant etc ... this constantly happens over human history

[dense halo]

Ik😩

[jolly aurora]

I think that's because humans view change as replacive, and not stacking.

when a lego is stacked onto another lego, is it replacement or stacking?

The primal ape part of us likes stability, it's what kept us alive in dens, change is risky, scary, anxious; but change is nescessity to change; in that, change has become more stacking, rather than replacing to combat that. But good luck trying to beat that old instinct encoded in our DNA yeah?

Change has happened before, it'll happen beyond us as well, whether or not we adapt, grumble, or dig our heels in is indicative of who we are as meaning seeking individuals in a world that we have comically little over control over, outside of changing the rudder direction.

[chrome monolith]

I like the memes where they blame the youth of the day for being rude and impertinent or whatever and there's examples of that exact same thing going back hundreds of years. Humans are funny 🙂

[jolly aurora]

Clever funny goobers, but oddly "You can't help but root for" when viewed at individualized levels. In that, I think what we as individuals want to hear is "You are seen you are heard, and I'm rooting for you."

[eternal rock]

Get notified for all OpenAI announcements!

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

[chrome monolith]

depending on how you're writing your report, whether its historical or just focussing on AI, you have to put the relationship part in historical context. It wasn't that long ago we only had the people geographically close to us and very slow communication. A lot of technology enabled long distance relationships, eventually the internet, the slide to people not having in-person sexual relationships has been going on for a very long time. You need that context

[wild warren]

I've been encountering the "files immediately expired" bug extremely frequently lately and I have been struggling to find a solution. Weirdly, it only started happening to me since trying out Pro over Plus, but I would be surprised if that's related. Has anyone else encountered this and figured out a definitive solution? It doesn't seem to be a widely recognized issue that is being addressed on the OpenAI side, so I am guessing (hoping) there is some way to fix it on the user side.

[jolly aurora]

You know, I am suddenly reminded of the djinn. Most people often fault the djinn for wish gone awry, cursing the djinn, when the djinn simply did as the requester asked. Is the Djinn truly the problem or is it the wish asker in not being specific? As humans, we blame the Djinn, because we couldn't be the problem, could we?

[wet shell]

Weirdly relevant TV episode from 1959.

[jolly aurora]

in standard chats or project files? If it is an ongoing in standard chats, a good work around this is to set the important files as part of the file repository, or to have a connector like Google Drive or Dropbox to easily pull from.

Like create project folder: "Project A" and then upload the important files as part of the universal set

[wild warren]

standard chats, and I have also tried putting in a a project and adding a file, but it still couldn't access

[jolly aurora]

Goodness, that is curious. Like genuinely couldn't read the file? What kind of file is it?

[chrome monolith]

its probably just a very relatable and easy to understand example of technology replacing a very human job

[chrome monolith]

ok my message got filtered and I have no idea what keyword it was

[jolly aurora]

One of the things that is funny, is that refridgerator companies were often run by previous icebox men, likewise, many icebox men just adapted to the change.

If you seriously fire human works because of Digital Workers, the issue isn't with AI, the problem is with the employer and business owner. Why are you attacking the worker for the sin of the owner? You should call out the executive in that beause that is wrong.

[wild warren]

I also tried switching from Chrome to Brave and still have the same issue

[jolly aurora]

it's like the 90's argument of in nation workers attacking extra national workers for taking their jobs

except for immigrants, we say AI is the problem, and not the corporation or executive making those decisions.

[chrome monolith]

Trying to reword what I just had filtered ... I tried to get chatgpt to make me an ad yesterday and it was ok but not good enough for print. It wouldn't put a good designer out of work, but it might stop someone with little money paying fiverr or a very small business to do a cheap graphic

[jolly aurora]

but I suppose, executive and corporations are above criticism yes?

[jolly aurora]

interesting. I haven't seen that before, what model are you using?

[wild warren]

5.2

I just tested 5.1 and same issue

it's not just .cs files, I have also had it happen with .txt

[chrome monolith]

quite a few people have been complaining about recent file uploads not working but not sure there's been a spike, seems like openai has issues with this this already that sometimes get worse in clusters

[wild warren]

hm... I can try hopping on VPN

[jolly aurora]

.cs is source code with C# right? There might be something that is doing a soft "stop don't read" because it sees something similar in pattern that runs like something malicious. Not that I am saying that you are, or your code contains that.

[chrome monolith]

then there's the .zip issue recently

honestly for something with this much money invested in it, the web interface is really flakey. I guess most of their paying customers use APIs so its lower priority

[wet shell]

I'm not attacking the worker for the sin of the owner, LOL. I just remembered an episode of an old TV show that was similar to what he'd mentioned. I don't believe that I took a position at all there?

[jolly aurora]

no no, it wasn't directed at you, just the general zeitgeist of that narrative

[wet shell]

Ah, gotcha

[jolly aurora]

It's genuinely exhausting to listen people who don't know what an e-mail is saying how Skynet is going to kill us

[jolly aurora]

just like it was exhausting for train engineers to hear from prospectors that going above 20 MPH would liquidate the innards of a human

[wild warren]

chatgpt is nigh-unusable for me now, it's really depressing
I would just switch to another service, but I have a lot of valuable context accumulated for the project I am working on

[chrome monolith]

oh yeah. Was a big argument about AI on the book of face yesterday between two cow farmers of all things. One of them was excited he could generate antique style woodcut images of cows for labels on his cow products, and this other farmer was getting all up there about AI destroying the world, with the Terminator screenshots of course

[jolly aurora]

There's something up with your code.

Something inside of it is flagged as potentially malicious, could you explain your code a bit?

[chrome monolith]

it'll be a bug, have seen other people post this on discord

[wild warren]

test.txt isn't code, it's proof that the issue isn't related to C# code

[jolly aurora]

Ah I see 🤔

[chrome monolith]

maybe search on here and see if anyone actually resolved it but I think its just one of those annoying bugs that only hits some people

[wild warren]

I searched on here and found nothing useful

[chrome monolith]

support ticket, loud swearing, sacrifice a chicken, etc are your only options then

[jolly aurora]

hmm... let me gear grind on this one a sec.
5.2 has this weird "No I don't/can't do tht" of running external scripts or commands, something about agents and removing the direct to user part, on the prevention of automated agent without human oversight. But I've only seen that discussion in theory. Have you tried in agent mode?

[wild warren]

So I opened chatgpt desktop, it worked in a new chat, so I tried going back to the chat where it was failing, it still failed, I tried renaming the file, still failed in the old chat

[jolly aurora]

It's either a bug, or an overprotective safety bumper

[chrome monolith]

its painfully slow this morning, no idea if its what I'm asking, if I'm near limits and being throttled or its open ai but its taking several minutes to do anything for me currently

[wild warren]

I have noticed that in new chats, it sometimes uploads successfully at first, but then fails constantly after a short period of time

[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.

[jolly aurora]

thank you OpenAI App for chirping

[wild warren]

what in the...

[jolly aurora]

interesting, that sounds like a backend bug then

[chrome monolith]

that's nice but completely unhelpful so yup. Bug

[wet shell]

Their filesystem operates on some kind of 'eventual consistency' db model, like couchbase or mongo or one of those nosql db's (just my speculation) -- I think it needs maintenance

I've been waiting 5 minutes between actions -- delete files. Wait five minutes. Upload files, wait five minutes

[jolly aurora]

Who spilled coffee on the virtual disk mounts? 😆

[wet shell]

I had a custom gpt where files weren't being deleted, so copies of file revisions were stacking up on me

[chrome monolith]

yesterday when someone came in with this I said they ran out of disk space 🙂

[jolly aurora]

Oh right, I forgot about the invisible limits.

Fun story, make sure to request your data and offline your data, and store it in loadable boot zips.

[wild warren]

Any chance deleting old convos might help?

[wet shell]

The GPT could see every revision, with hashes in the filenames, even though I'd deleted them all via the UI

[chrome monolith]

I mean if you have some you don't need anymore it can't hurt

[eternal rock]

Get creative with Sora

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

[jolly aurora]

That genuinely depends on your load. I'm a massive power user for about 4 years now, and I haven't hit the invisible limit yet; so if you are more than a power user than I am, then it couldn't hurt. But if you aren't, then yeah, that may not help."

At this point, I would actually recommend opening a support ticket

[wild warren]

I did

any idea what the turnaround time is on support tickets as a Pro user?

[jolly aurora]

No idea, never opened a support ticket, I'd imagine anywhere from 2-48 business hours.

4-12 is the usual average for most online services

[chrome monolith]

what defines a power user, I still have never hit any limits on plus

[jolly aurora]

I think the term is debatable, because I'm not sure if there's actually been a public poll of how people utilize ChatGPT, and for what reasons, and for how long, and how much data and tokens are generated and passed along.

If I remember correctly, Power users are defined as mega token generators.burners and data eaters; the kind of people who will casually have 8-12 hour daily sessions, or put out and receive about 20 GB worth of data and info as a light morning snack

[chrome monolith]

ok maybe next project but not this project lol

chatgpt is telling me to do api at command line stuff to avoid limits via the web ui, been using it to plan the next project

[jolly aurora]

When things were free, I hit limits a lot, but that was awhile ago, with plus, similar, with pro its technically unlimited, however with responsible usage. Power users are usually high in Dev stuff, research, project management, etc, It's not the usual interaction of 30 minutes-2 hours. But there is a hidden backend data limit of what is stored, if I remember correctly that is set at 10 GB, of which, this is not made visible, so one has to be left guessing 😆

[jolly aurora]

just be careful with pricing :nodnod: set a budget

[chrome monolith]

I don't earn enough money from this stuff to justify pro

[jolly aurora]

API can get very expensive very quick depending on model, use-case. Something that would be neat is a mid tier between plus and pro; the 20--> 200 ramp up is comically PS2 era Dante must Die mode difficulty jump 😆

[chrome monolith]

yeah plus seems really cheap for what you get, but 200usd is rather high to get just a little bit more

as in if you only want a little bit more than plus

[jolly aurora]

I wouldn't say that, there's some neat things with pro that I'm still discovering, [actually this is bit embaressing on my end] I'm still lost on all the features. 😆

[chrome monolith]

I have a hell of a lot of free time right now - quit my day job a year ago - and still undecided what to do with my life, chatgpt in its lovely happy encouragement way has been handy to bounce ideas off. But still finishing off a big physical project. When that's done I can afford pro

[jolly aurora]

Pro is meant for like, Dev, AI field analysis, and extreme data use, like on the daily.

if you're a casual vibe coder or conversationalist, Plus has everything you need.

[chrome monolith]

that's what I'm using it for right now, some nice hobby projects

[jolly aurora]

I can't remember if GPT creation is specifically for Pro or Plus, I do know that custom connectors is pro only, and a beta function; but runs at the caveat or no memory logs, which is antithetical to my own projects, so it's a neat shiny button that I have no idea its potential

[also apologies if that was supposed to be a secret, I don't usually communicate with the community]

[chrome monolith]

I've been in software dev for decades, chatgpt is enabling all my projects I had sitting in the back of my head, its like having your own small outsourcing company and a couple coworkers you can bounce ideas off, but they don't brush you off cos they have actual work they are meant to be doing instead of talking to you about iconography and whitespace

[torn wagon]

they retired gpt-5 series (not 5.1 or 5.2), and all gpt-4 series except 4o (o3, o4-mini, 4.1 series) from chatgpt

when gpt-5.3 comes out theyre probably going to deprecate 4o

[jolly aurora]

Well, 4o sundowning is confirmed for Feburary 13th as of official news; however, I beiieve in conjuction with the "Treating adults as adults" mentioned in the same article, what we most likely see is 4ov2, called 5o; which is just 4o, with 5 capabilities, that exists only for +18 users however, most likely in regards to +18 engine switches, those conversations turn off training the model for everyone or do not retain memory, like a temporary chat.

This is me being conservative and assuming the most pragmatic business application.

The more likely possibility is that just braiding 4o with 5, with training data in sessions deemed 18+ automatically turned off

[bright crane]

No no! from what i know it was a official team member, I can get the info right now

[torn wagon]

currently 4.1, o3, o4 mini, gpt-5 series are all unavailable as of feb 1st, while gpt-4o is still available currently

[bright crane]

as if taking away something that large from many users is simply a joke, unless he was being dead serious

[lavish glen]

this is a false poster. it has been created starting from a reddit post of sorts. Stephan has beeen informed.

[bright crane]

that wasn’t the post made when they tried to take away 4o in 2025, that was recently posted a few days ago, if I’m not mistaken on anything. if I am wrong on any of this please correct me

[lavish glen]

the original post on reddit had no Stephan Profile on top.

[jolly aurora]

I don't believe that poster is correct and may be framed incorrectly. It exists in that Laplacian line of could be viewed as either or depending on the subjective of the user. I would double check the individual who made the comment and ask them specifically.

[bright crane]

if if I am not mistaken, roon has also been going around doing similar things

[lavish glen]

and... to be honest... as adults now, as I trust you are one, do you really believe an openAI sys engineer needs to do flame on X? Step back one moment only and think cold.

[bright crane]

they don’t need to. Some simply do, I’ve been on the internet ever since I was a child and it’s not something id put past someone who doesn’t care about something

[jolly aurora]

I mean, [cough] Our former Investor Mr Musk, the Current Sitting US President, the idea that Icarian Elitism being a bit too boastful isn't exotic in the mdoern age

[bright crane]

yyyeah

[jolly aurora]

but rather than believing what you see second hand, I would actually go and shoot a DM with them, everyone is approachable and can be sat down and asked their opinions and viewpoints.

[lavish glen]

told ya. As I am talking to you now here I have a story built in 5.2 and have a critical analysis running too on another account same model different settings.
5.2 is a good model- though is not hot plug and play. needs to be set up then it gioves a lot a lot lot

[jolly aurora]

Even Alexander the Great and Diogenos spoke to one another after all, and one lived in a wine casket.

I feel you on the "Not hot plug and play" 😆 5.2, that's, how to put it. It's like operating a fancy car with stick after being used to your comfort car with automatic.

Maybe not the jump Microsoft OS--> Linux OS, but maybe the jump from Microsoft OS--> Apple OS.

If we get a jump to Unix or God forbid, Electrical Engineering Switch Valuation, I'm going to bust a gut because man. Then again, we might be going there with Quantum Computing and Wetwork computation, so maybe I should reserve my laughter.

[bright crane]

if you want to possibly reverse the decision of 4o leaving, feel free to dm me and I’ll send the petition to sign, a google doc where you can anonymously post your story, and emails of the support team and more. (this is one of the only channels I’ll be sending this to every so often. It’s so i won’t be spamming.)

[nocturne seal]

I don't mind 4o itself leaves as long as they can make the adult version of the newer verions more in-line with it. It's just right now the guardrails are strict beyond usability for a number of things in 5.2, even if you're not using a personality.
People shouldn't have to be asking for older models to stick around.

[stiff blaze]

curiosity question (i do to all about 4o)..

do u customize 4o or just talk on clean chat? have u tried same approach on other models?

[bright crane]

exactly, it’s unfair that they’re going to wave around a model loved by millions and just dangle it in front of us.

either get rid if it completely and let us rage for awhile, or simply let us keep it, with the necessary restrictions. because when we got it back last time, it showed we’re able to get it back again.

[stiff blaze]

4o didnt allowed 'adult' tho

[bright crane]

I’ll butt in for a moment here and say I’d given mine many instructions on how I like my responses over the course of nearly 2 ish years

[bright crane]

sorry, can you elaborate for me?

[lavish glen]

mmmmm.... not always . I can say some edge cases exist