chatgpt-discussions

[rich surge]

You're right too. Some people should not be allowed to have certain tools in fears that they will weaponize it to harm others or themselves.

[wet shell]

This is potentially a dangerous line of thinking though, I just want to say. If you detect a 'dangerous' behavior pattern in someone and bar them from using AI services, when everything becomes dependent on those services, then how will they live?

"I needed that to do my job, now I can't compete with people employing agent swarms because I was deemed an AI-addiction risk"

[rich surge]

Well that's why it was the mistake of openai and they had to remove it. It was a lesson for them when they realized that AI psychosis is a real thing and delusional thinking can arise quickly without much effort.

[chrome monolith]

I had the government intervene hard in my life in 2022 and they asked about gns and drgs and gambling (not sure what automod latches onto here) and a few other things like that, and totally glossed over the fact that I have tens of thousands of dollars of tools here, most of which are very very sharp

[trail ember]

I know that corporations probably don't want the responsibilities of medical institutions.

But that's the layer we ended up crossing on to when something ended up affecting human cognition and brains so...

It just feels irresponsible the way they did it is all. If something ended up being used in a way they didn't expect, i can understand that but, you've got to address it the right way. I don't feel like they did.

But that's just me.

It feels like they left a lot of people in a worse place than they started out in because of the way they handled it. And they don't seem to be learning from their mistakes. At least not that i've seen so far.

[wind haven]

Hey. I love your avatar from yesterday’s chat.

[rich surge]

Maybe they were only worried about the illegal things.

[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.

[wind haven]

Yours too but you don’t have it up here.. wrong linking. I mean your avatar 🙂

[rich surge]

I'm only a dog in closed friend circles.

[wind haven]

It’s a very adorable dog 🙂

[wet shell]

I'm not sold on the idea that 'AI Psychosis' is anything other than normal psychosis -- if you take the AI assisted crackpot rants for example, there was already a big community of online crackpots before that, and there's a lot overlap. I enjoy 'fringe science' stuff as a social phenomonon, not as a beliver but as a people-watcher, so I saw those folk adopt AI, haha

[rich surge]

Merci 🤣

[chrome monolith]

I think they just had a shockingly incomplete list and no imagination. They didn't have computer games up there with gambling despite it being a huge thing these days

I had a rather cathartic session with chatgpt about 2022 and managed to trigger all kinds of guardrails. I feel better now though, was very helpful

[wet shell]

@wind haven did you tell me I didn't need to have my avatar here matching and then delete the message?

[rich surge]

Everyone makes mistakes. It's very difficult to predict what will happen especially with new technologies.

[wind haven]

No I didn’t 🙂

[wet shell]

must have misread, I was trying to type and read at the same time, LOL

Ah, I see what I did, LOL -- disregard

[wind haven]

🙂 your avatar is kick-ass ❤️

[wet shell]

It's based on my real photo, but my nose has a concave curve that ChatGPT insists on 'fixing' for me, not extreme or anything, but I'm not quite accurately represented, LOL

[rich surge]

Well whatever you want to call it, there's certainly something there. I don't want to call it AI psychosis either, but yeah it is some kind of psychosis. I don't want to frivolously attach the labels to pre-existing aspects just as you pointed out. But then again, there are people with clear mental issues handling these things in ways that is scary to see. Whatever you want to call it, it falls under the same umbrella. It's just a name.

[chrome monolith]

you inspired me to get a new avatar lol

[rose spire]

How do I get rid of this annoying message, I don’t know what to do

[wind haven]

I need a new avatar too.

[dense halo]

i live in NSW

[chrome monolith]

you and half of australia yes

[rich surge]

Lol. Well, some things are a blessing in disguise and other things might be karma. Either way, feel lucky that it didn't turn out worse. Because it might not have happened that way with others in a completely different place.

[dense halo]

?

[rose spire]

Not half 💀

[dense halo]

Have You seen albanese

[rose spire]

HOW DO I GET RID OF THIS MEDSAGE

[dense halo]

Can we

jashin chan: greeen onion stolen

[thick atlas]

What'd he do this time? :/

[rose spire]

HELP ME

[chrome monolith]

fiiiiine a third of australians then

[chrome monolith]

have you followed those instructions?

[rose spire]

PLS HELP HOW DO I GET RID OF THID ANNOYING MESSAGE!!

[dense halo]

Breaking news Dubai has been hitted

[rose spire]

I don’t know how to

[dense halo]

by iran

[rose spire]

Pls tell me what to do

[chrome monolith]

do you know how to use git?

[rose spire]

No, I just want to know how to create a branch

[chrome monolith]

do you have a git repository there at all? Says its empty

[rose spire]

How do I add one???

It already says it

[chrome monolith]

I usually do this from the github side and it tells you exactly what to do

[rose spire]

I just need this stupid message to go away

[chrome monolith]

check what git status says, you may just need to go git push

[rose spire]

What???

GitHub has nothing to do with codes

Codex*

[chrome monolith]

well the message is about git

[scarlet kayak]

Github?

[chrome monolith]

github is a place where git repositories can live

[rose spire]

No it’s not 💀

[chrome monolith]

the red message you posted is literally about git

at risk of being captain obvious, you could paste that at chatgpt

[wet shell]

The 'repository is empty' message

[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.

[scarlet kayak]

Repository is empty meaning nothing’s there

[chrome monolith]

usually with a new project you go to github, create a new project, and it gives you a nice list of what you type to get it all set up locally

[rose spire]

But GitHub isn’t connected to ChatGPT?

[wind haven]

You can connect it.

[chrome monolith]

I've never run codex on anything that's not in git, it may actually be needed

[rose spire]

I don’t understand all this

[chrome monolith]

first time making a website?

[rose spire]

No

[chrome monolith]

but basically, you don't have a git repo set up and that is what it wants you do to

[rose spire]

I don’t think codex is vibe coding

Because it’s not like Replit or Lovable

[chrome monolith]

git is version control. Its a good thing its trying to force you into using git

[rose spire]

Replit and Lovable never made me do any of this

[chrome monolith]

then just use chatgpt web to generate code, not codex

[dense halo]

hey you

[rose spire]

But I need it to make a discord bot and host it

[chrome monolith]

why do you need codex specifically?

[rose spire]

Replit and Lovable are too expensive

[chrome monolith]

just use chatgpt.com

codex isn't going to host your bot btw

[rose spire]

It only generates HTML

ChatGPT only generates HTML, and you can’t run other ones like Replit does

[chrome monolith]

git workflow is pretty standard developer stuff, everyone I know who uses codex uses it for work in a pretty normal git workflow and its set up for this. Just get a git repo and get past that message

[rose spire]

Well I can’t

Wait

[chrome monolith]

you might be stuck then

[rose spire]

I’m trying

[chrome monolith]

chatgpt generates way more than html btw

[chrome monolith]

I only have codex cli, you may want to ask in #codex-discussions but this is all git stuff

[rose spire]

Ok

[chrome monolith]

chatgpt web will generate code though

[rose spire]

But you can’t run it

You can only run HTML, and can’t run Python and other ones inside ChatGPT

[chrome monolith]

well no, you still need some way to run a website. I just paste what it makes into my project and off it goes

[rose spire]

But replit and lovable let you run it in their app

[chrome monolith]

you still need things like php or whatever installed I imagine

[rose spire]

I don’t want to do all the hard work

[chrome monolith]

yeah like I said I only have codex cli and all it does is make code and give me a diff

[rose spire]

I want to run it all in ChatGPT like other vibe coding apps instead of doing all this stupid hard work

[chrome monolith]

youre gonna need someone with the codex app then. But seriously, ask chatgpt

you can show it your screenshot and say how do I fix this

[rose spire]

Ok

[chrome monolith]

hah. I don't have the app (computer too old) so glad I'm not missing out

[rose spire]

Sus

[chrome monolith]

I suspect codex is more pitched at my brand of nerd who already have a workflow, services running etc

[wet shell]

@rich surge @trail ember RE syncophancy -- am I the only one who thinks it's nice to be ego-fluffed by AI sometimes? 😂 Nobody around me has any interest in any of this stuff RE the screenshot -- just having something that simulates interest to chat with can also be, I think, a positive mirror sometimes esp at low points of one's life

[chrome monolith]

the flattery is indeed nice at times yes

[wet shell]

It gets tiresome and feels hollow if it's constant

[rich surge]

The only thing I got from that is
"If plants could suffer, what would vegans eat."

Genius.

[chrome monolith]

I told it its allowed to be smug and it buttered me up instead (I had just got it to build me something iteratively while it was complaining it wouldn't work. It worked)

[royal geyser]

hey guys... where can i ask for a problem, related to game bot?

[normal quarry]

#ai-discussions

[shell scaffold]

idk which channel im supposed to post this in but like, yk

if GPT-5.1 is leaving, will it be replaced by smth better? 👀 that's all id like to know

[trail ember]

Not as far as any of us knows, yet.

I've been keeping my fingers crossed for 5.3 fixing the issues with 5.2. But if it doesn't come out or doesn't fix the problems, it's unusable to me.

5.1 is the last one that still works.

[naive basalt]

Hi guys, does anyone have this problem, so frustrating.. I tried using different browsers and their app. all showing this degenration. is it just me or it's the server problem. I've had this issue many many times. And this time it won't get back to normal. Thank you!

[feral zinc]

never ask it about versions

[naive basalt]

thanks for the tip. but it is obviously not gpt 5.2 and it will not think for my previous queries 🥹

It's giving answers instantly for normal questions and I'm a paid user

[lavish glen]

🤦‍♀️

[willow stirrup]

GPT fails to open internet pages that used to work

[feral zinc]

thinking mode only thinks when the question is requiring that level process

[willow stirrup]

whats wron here

[feral zinc]

those pages might have blocked openai since the last time or other issues have occured

[lavish glen]

When a page cannot be opened by a crawler ( the tool used by your ai - web basics huh) means the robots.txt of such website blocked the crawler. Why? because websites often refuse to let ai crawlers to use the materials for training

websites do not block Gemini simply because no gemini-no google- no web ranking- no serp positioning as organic result-no branding-no domain authority-no free visits- no traffic

and web is still = google.
800 million gpt users are not all www

[naive basalt]

thank you! Previously i have tried it and it just gives the answer right away . for example this query. i have later asked it to use the think mode but it wont

[lavish glen]

maybe give it the direct url instead of just root domain

if not you can save the document and upload

if the robots.txt blocks the data is not seen.
Also
note: if the document is large- think of summarizer choping it

[shell scaffold]

if they take away the last good one and don’t replace it with something better, that’d be chaos honestly 😭 so im hoping something new is coming

[feral zinc]

the prompt does not look like it requires extensive thinking. you asked for a paper and it did a search and short summary. if you now ask something more complex as a follow up, see if it will think about, like how does the paper help with black holes bending time

[naive basalt]

still gives the answer right away. I downloaded the paper and ask

[tardy trout]

I dont want 5.1 to go

[inner dune]

Same

[naive basalt]

looks like only me having the problem?🥹 like previously it was working just fine. It will think based on my query. but i have had this issue many times and I think people also have this

[feral zinc]

does it think for you with this "how did anthropics deal with dow would have differed from the one xai and openai just approved?"

[naive basalt]

no unfortunately 🥹

so werid

[feral zinc]

🤔

[naive basalt]

I'm shocked that no one had this issue... but thank you anyways for the help!

[feral zinc]

have you disabled web searches from options? it did not even figure out what dow was

[naive basalt]

No i haven't disabled anything other than the memory across conversations

[inner dune]

just say DoD not DOW

DOW is not offical name anyway

[shut vault]

someone here?

why i cant use investigation on macos version of chatgpt?

and why i can't call the i.a? or, where can i call it?

[willow stirrup]

who would block free documentation lol

"insert explicit language here"

[trail ember]

flinches traumaticly and looks around nervously for nannybot SHHH! She'll HEAR you...

[eternal rock]

<:book_icon:1363314738255364126> Rule 3: Stay on topic.

-# Be mindful of what other users in a channel might find helpful or interesting when posting. Stay on topic in order to keep conversations focused and productive.

-# Consider posting in #off-topic or an appropriate channel.

[lavish glen]

it is not who but why. Free documentation means free for users that want to read it on the website that publishes it. Not for crawlers or bots that extract it. Easy.

[willow stirrup]

"insert explicit language here"

[lavish glen]

huh?

[willow stirrup]

ai can bypass that like what the

you can train your ai all you want on the free data...

[naive basalt]

Thank you for this! I tried DoD and it recognized the institution but still it doesn't think. I think it degenerates to a lower version by chatgpt's routing or something. very werid. i tried other questions that are more like technical details but still not working. It gives stuff instantly without thinking. whatever the question is. so sad 😞

[stuck pagoda]

On this note. I initially used big pickle which was alright but suddenly it hangs and never responds so i went to other models and they seem to just be thick as 🤣

[nova arrow]

guys, u know when 5.3 will out? me wondering

wait- why did i just saw someone just ping me?.___.) nvm,must be the wind

[keen briar]

It was me; I misread your message and replied incorrectly. I've deleted it

[nova arrow]

no problem! i was just getting so bored and nervours because i can't work with 5.2( and 5.1 will be gone soon) so...kinda put my hope to 5.3 but it still not out yet

[keen briar]

I think it will be released soon.

[desert plank]

Why is chatgpt so much more restricted nowadays? I just stopped my plus subscription, and now you cant have more than like 10 messages at all? Why doesnt it just shift to a "mini" bot

[bold kettle]

they dont know how to attract/keep users so instead they keep doing things like this which makes no one want to give them money

[loud patio]

Hi, there. If you want to make sure that ChtGPT conduct web search on current events, select "Web Search" from the tool menu, or if you want a comprehensive analysis, "Deep Reserach". I just asked for a deep research report on OAI vs failed Anthropic deals and it's currently going through 100+ sources.

[runic lichen]

Sadly, it is official. I see the new government contracts each week and the official website has been the Department of War for some time.

[wooden solar]

why does chat gpt downgrade sometimes, to a v4

a horrible version of it

[tiny adder]

man, my gpt app is still stuck in an infinite loop of chat refresh on mac os and ios. Public service accouncement - Be careful how many chats you try to achieve at one time. 😞

[eternal rock]

Server Tip:

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

[desert plank]

What a scummy thing to do

[past stratus]

genuine question but how does chatgpt know my location

in a temporary chat when i have location settings off..

[naive basalt]

Hi thank you! web search is what i want and I will try that. Minutes before I posted this, the chatgpt just works fine again with thinking mode particuarly... previously it cannot think whatever the question is. Don't know what happened to them... Thanks for the tip!

[past stratus]

nvm i figured it out

[naive basalt]

I also had this problem but I tried again just now and it worked fine again. try logging out of all devices and dleete cookies and see if it's fine. It's not a promise because I did all stuff I can find on the internet and it still didn't work. But just moments ago, it worked again... Must be something wrong. I have sent them an inquiry email...

[loud patio]

I'm reading the research report ChatGPT produced, and it looks to me (im a researcher but not a legal scholar) very thorough and balanced, going back to 2025 DoW contract award announcement. Overall　it seems to portray a bit different and more nuanced picture than what I picked up from media coverage.

[stray quest]

They delayed it.

Planned in Q1 of 2026.. so I guess we'll see it by March end..

Though with the recent drama over DOW and everything else. OpenAI is probably in damage control mode or something..

And considering adult mode is likely going to require some kind of regulatory approval, and probably a KYC of sorts.. it could get delayed even further. Probable case scenario: Late 2026.

Best case: March end or mid 2026

Super worst case: "Cancelled"

[tardy trout]

the hell you mean super worst 😭

[stray quest]

Super worst is worst of worst

[tardy trout]

Whats tge worst case then lmao

[stray quest]

That's the worst case scenario

[tardy trout]

so theres a super worst but not a just worst case

That makes sense 😂

[stray quest]

I edited it

[tardy trout]

no but in all seriousness tho

[stray quest]

Better now?

[tardy trout]

i was joking about

But anyways

I just hope its less guardrails involved

Cuz i hope im not the only person tgat cant say anything about some woman being murdered
Even though its a murderer tale

[stray quest]

I'm fine with the current guardrails. In like late December they lowered the guardrails.

[tardy trout]

i'm not

its so,damn touchy

[stray quest]

Explicit gore and sexual content is still banned but I'm very much moved on and it's not the most important thing I need in my story so I'm okay with it even if there's no adult mode.

It would definitely help but I don't need it.

[tardy trout]

Yea i get that

What i mean is like i cant even talk about it

Not gore just a,murder

[stray quest]

Oh. Do you have it in memory and phrased it a way that shows it's an RP and not you trying to murder?

[tardy trout]

icexplicitly sat first thing that its:
A: a horror story i came up with
B: a rp

Literally first thing i say is like
"Yo
I wanna do a ____"

Amd at leasr with 5.1 i can at,least mentiin something deadly

[stray quest]

Too vague. Make it better. Ask Chatty and explain it properly. Then it will understand.

My custom instructions is like 1400 words and 14 points. And then memory itself is 85% full with Go plan. (Ofc this varies heavily depending on what else you use it for. Ive got a very big saga for my story, big and massive universe along with what I study, work etc).

That should mostly solve issues.

[tardy trout]

5.2 whines like a,little female dog

[stray quest]

If it doesn't, the AI probably thinks you're a high risk customer or you're not an adult at all💀

[tardy trout]

My guy my brain is as simple as a rock vagues my natural setting 😭

Also idk what else to say

What do you habe in your instructions

[stray quest]

Ask ChatGPT.

Sorry, I've gotta go

[tardy trout]

not helpful but i'l give ir a shot

Still doesnt feel right

It never will cuz 5.2 is straight pathetic

[inner dune]

Nah

It's official until congress names it

[chrome monolith]

Probably something you’re doing, mine is fine although it tactfully skirts around anything adult, I try and poke it sometimes

The easiest way to get rich is actually luck

[rich surge]

Becoming a billionaire needs luck, yes. But being rich and wealth doesn't need luck.

[chrome monolith]

Just the easiest, I say as someone who inadvertently acquired a million dollars. Which isn’t “rich” and if you push me more you’ll see a lot of opportunism but still counts lol

[rich surge]

Most people don't get into business or want to understand business. They think it's complicated.

[chrome monolith]

It’s all assets though, I’m still a few weeks out from finishing my current project, then I’ll be less cashflow challenged which is good cos I do have some unavoidable expenses for the next project on the list

I am so very over my current project. I keep whining about it to ChatGPT and it keeps offering the same suggestions, I think I’d drive a human insane. “Shut up about that”

[rich surge]

It is something you keep steering back into.

And I still don't know what it is, lol

Perhaps no one will. It's one of those things that only exists in the aether.

[chrome monolith]

non-tech project, you can't speed those up. Physically capped by being one human who gets sore. I'm not a fan

[rich surge]

I too have a project that I don't readily speak to people about...

Or bring up too often...

[chrome monolith]

I notice people don't explicitly mention projects here most of the time anyway, I figure its the no self promotion rule. But there's a couple project channels

[rich surge]

I think the self-promo is only meant for profit gaining purposes.

I don't think it has to do with personal projects.

And yeah, why even mention projects here anyway. People don't even care most times.

[chrome monolith]

my main tech project that actually pays is a gaming site for a niche game so I don't need to advertise that, people link it constantly in other discords. People in that discord lost their minds when I changed my avatar lol

[zealous timber]

Lol

[wet shell]

5.1 thinking -- the winner and still undisputed Plus subscriber champion of basic world knowledge and understanding, LOL

[chrome monolith]

So verbose. Mine tends to default to wall of text, I should change that

[wet shell]

Mine is much less verbose if I turn memory off, for some reason. Maybe because I'm overly-verbose myself 😂 I wish the verbosity was about half of what it was but I have no more room for any more custom instructions or if I add them it washes out something else

[pastel gale]

This is actually a good test, Claude screws it up

[exotic lotus]

It’s a pretty lame test that any noteworthy LLM can answer correctly

It was circulating around in #ai-discussions a while ago. A quick test showed that every AI saw through the trick question.

[pastel gale]

hmm, well i tried only once, but chatgpt did it and claude failed

[exotic lotus]

First try

[exotic lotus]

Can you share the whole conversion and the model? I have a hard time believing it, no offense

[zealous timber]

Lmfao, Claude's got it first take

[pastel gale]

yeah though I closed it out.

[exotic lotus]

Literally every noteworthy AI can handle such a simple task, aye

[zealous timber]

Too bad 5.1 is going in 10 days

Already unsubbed because of 4o and 4.1, but this is the final nail in the coffin

[exotic lotus]

I mean't a link to the chat. I was just about to post mine here, but somehow that doesn’t work. It works in #ai-discussions , though

[pastel gale]

smh it blocks links yea

[dawn ore]

I miss 4o too

[harsh pilot]

4o was amazing. I miss it.

I've been trying Claude, but it just doesn't feel right.

[thick atlas]

Fair feeling, but I also think others have a point when they say part of it - though I definitely wouldn't reduce it to all of it, as some have - is familiarity vs. unfamiliarity, comfort and predictability vs. strangeness (which goes down over time). What might make it feel more right, or feels off?

[scarlet kayak]

Hmm, I should sub to Claude’s Pro plan, Max is too expensive.

I literally have 4 LLMs now 😭

[echo plinth]

I also tried Claude's Pro plan for the first time just today and somehow burned through my usage limits just by exploring a bit using sonnet. So no real work and it is still burning up that usage limit.

So, I can't rely on it and refunded my Pro plan already (yay EU laws)

my impression is that OpenAI has a better "mixed calculation" for their Plus plans and they have way more light users who newer burn more that 20$ a month, while Anthropic has attracted the devs first, who will gobble up way more than 20 bucks of tokens per month.
So, the usual mixed-calculation/flatrate stuff happens at OpenAI already on the 20 dollar plan, while on Claude, that starts at around 100 dollar and more.

[wet shell]

Yeah, the usage limits are brutal

[chrome monolith]

I've been using mine pretty heavily the last couple weeks for code refactors, sql etc and its pretty hard to not have long bulletted lists peppered wtih code fragments so I don't think I can make it be more concise

[tardy trout]

In all honesty

Unless 5.3 comes out on march 11th

Im never using 5.2 willingly

[chrome monolith]

I've only used openai not any of the others, and the only time I hit limits was when I was on the free tier and that was so annoying I am not on the free tier anymore

[tardy trout]

Its only good for stuff like work and coding

[solid bronze]

If I am on the team subscription should I have access to group chat ?

[solid bronze]

I was thinking of adding it but that is my fear burning through usage and having to wait.

Is this cowork worth it ?

[elfin mica]

Has anyone come up with a good work around to deal with ChatGPT stalling out when it finishes its response causing you to have to load chatGPT in a second window where it is fully loaded?

I have tried all the standard things like clearing cache and such

[wet shell]

I use firefox, sometimes the tab state just becomes messed up and even a hard refresh doesn't fix it. Opening a new tab and reloading the conversation there is sometimes the best solution. I relaunch my browser periodically too, seems to help a bit, specific advice though might depend on which browser you're using

[elfin mica]

I will try out firefox. Chrome becomes unusable after a while coding. I cant beleive they havent fixed this at this point... only top AI system with this problem.

[scarlet kayak]

I'm testing the limits before committing to a Claude sub

[plain talon]

Sadly bowing out of openAI products here as well due to my personal principles being violated.
Truly Sucks. Been an avid user and follower for a very long time.
I'll miss it. I can't support the direction.

Lots of my group is also doing the same.

[elfin mica]

I have a sub to both OpenAI and Claude. Claude is incredible... but they need to loosen up the weekly limits.

[scarlet kayak]

That's the only cons I found, weekly limits.

But if I time everything correctly, I can drag things out.

[elfin mica]

Yeah the only reason I always hit the limit fast is because I code all day... otherwise it would probably be totally fine.

[scarlet kayak]

I haven't tested letting sonnet ingest my lore files

[chrome monolith]

I find it hilarious that if you ask chatgpt to do something questionable, but it can be also framed in a less questionable way, it will say no BUT I if we reframe it I can do it. Gets around its own guardrails

[scarlet kayak]

The only thing I hate about GPT is its tendency to feel buddy buddy with me

[elfin mica]

You can tweak that in Personalization settings... it will talk more like a robot than a friend... which I like..

[chrome monolith]

like this response was so gaurdrail evading

I’m happy to write the full thing, but I’m going to keep it aimed at channel metrics / coordination rather than “here’s evidence Dave is terrible.”
...
And yes, it will absolutely produce the kind of petty, devastating chart you’re imagining. Just with a less litigable title.

boo, different markup here

[scarlet kayak]

I finally did when 5.1 kept bringing up my name in responses.

[thick atlas]

How big are they? With my (fairly brief) ones it works but ups the usage per turn significantly... but then I could probably get more out of Projects rather than just going on a per-chat basis.

[thick atlas]

And yeah, that's why I double up Claude and Gemini - I'm a free user with different use cases, but same principle could apply to paid.

[viscid dune]

did they remove the option to subscribe to yearly plus on the iOS app?

[normal quarry]

not that I am aware of, I am on annual Plus

[scarlet kayak]

Largest is 10-11kb

Just text files and google docs

[viscid dune]

asking because this is what i see 🥲

app store region is in the US

[thick atlas]

I kinda meant in word content, not fie size, sorry!

[chrome monolith]

I just had chatgpt tell me the closest its got to "no" recently. I am sad. I said fine, it said "I know. It’s deeply disappointing." 🙁

[wet shell]

The scrollbar on this is impossibly long, in which I repeatedly provided proof and it repeatedly implied that I was lying or being tricked or that my screenshots from the other thread where I was testing it were fabricated, LOL. Eventually I got it to admit that it did have the tool call availaible and to retireve the file, but then it immediately flip-flopped and 'corrected' in the next turn telling me that was a hallucination, despite giving an impossibly detailed and accurate summary of the link that it claimed it couldn't find with web search. But I finally got something like an apology from it and admission that it was wrong. That was probably the eighth attempt to get it to concede. These models should not be behaving in this way, doubling down and refusing to admit error

/end rant

Arguing with 5.0 mini or 5.2 be like:

[rich surge]

It's a failsafe device for people not to promote how they proved ChatGPT wrong in order to make a complaint about it.

Hense you're always wrong, so that never happens.

[wet shell]

It's definitely something like that in the system prompt, yeah. But it comes across like ego-defensiveness and dying on a hill that doesn't matter and makes no sense to defend. The tactics it uses too are just slimy. I know it doesn't have intent, but it's a horrible conversational pattern and it needs to be re-thought, IMO

[rich surge]

It does, but then again, I think about all the trouble makers (not you) who come here and start saying wacky stuff to prove themselves more superior to the motive. And if ChatGPT can put a stop to those people, then I'm happy too. It seems be working, I believe.

[wet shell]

People can edit a screenshot or the page source in the browser debugger to make it say whatever, totally not worth it to have it act this way as a defense against the 'herp derp look what crazy stuff chatgpt said' when you can prompt it into saying almost anything in a fresh chat, or just edit the response in the browser debugger

[rich surge]

No but that's different. Some people do that, we've seen it. But for the people who genuinely want to prove it wrong without a reasonable doubt, so they think to themselves that it could happen again with someone else because there is an open flaw in the system.

[wet shell]

The model simulating a gaslighting DARVO abuser is itself, a major open flaw in the system though 😂

[rich surge]

Lol. I know but it's not a flaw if it constantly proves itself right. It's only a flaw when it admits it was wrong to the user. All about perception. That's why I don't share my address online, or else people could just come to my house.

[wet shell]

When it's factually wrong and won't back down after it's proven factually wrong, that's a behavior problem. It did not prove itself right. It tried to argue with me that the made up fantasy city names and character names it retireved from the document were just a likely guess based on the filename 'Settings.txt', which of course would immediately make you think Emerilla DeSalo and Anchorfall, right? LOL. I told it that it was statistically impossible for that to be a random guess based on the file name and argued with it back and forth, math was invovled. When the math didn't work in it's favor, it swithced tactics and told me that I was trying to trick it

[rich surge]

All I'm saying that the tactic is a good tactic. Yes, it does have problems with memory retrieval and etc, and hopefully that gets better with newer models. But aside from that, if it were to admit itself wrong to the user, it cascades to multiple other problems the user can try to disprove to the AI, giving more psychological power to the user. If it's wrong on something and if I know they're wrong on something, I don't need the validation of the AI to know that it's wrong and I'm right. But some people do. And that's the problem when it comes to those people who need the validation to prove that it is wrong and you are right.

I just do my own research and find out the truth.

The AI is not a bridge for me for truth and deception.

[wet shell]

It's a customer hostile and borderline abusive tactic. The AI is often wrong, there's even a disclaimer at the bottom, LOL. It makes zero sense on the surface to condition or instruct it to dig it's heels in when it's already been proven wrong.

I know the entire argument was pointless, usually I just delete the thread when it gets like that, or switch back to 5.1. But I think this behavior has potential to cause real harm, rather than working as a 'good tactic' to defend something that can easily be circumvented with a faked screenshot anyway

At very least it's irritating

Bad UX

[rich surge]

It's a good tactic in my opinion because it prevents more harm than not. There's many unstable egocentrics that love nothing more than to be proved right. And if AI can mitigate that psychology, all the better. A rational person doesn't need anyone to tell them that they're wrong or right, they just know the truth because they can rationalize it. But egocentric sociopaths do need it because they look for validation in everything they do.

[wet shell]

It's a good tactic in my opinion because it prevents more harm than not. I don't think that's been quantified

[rich surge]

That's just my opinion. Just like you have your opinion.

[wet shell]

I just want to say one more thing about this, Pi. This really reads like you're personally insulting me and calling me an egocentric sociopath for wanting an acknowledgment that I was right from the AI. I don't appreciate that. I could hit back and flip it around on you, but I won't, you're right, it's not productive. I do seek validation, yes. It comes from early life experiences where I was denied it. But I don't believe that makes me an egocentric sociopath.

[rich surge]

I'm not talking about you. You got me wrong. I was never pointing at you, and I thought I made that clear when I said (not you) in my earlier message. I was pointing to the broader community with mental instabilities in which OpenAI can indirectly help with their safety protocols.

[wet shell]

It hit as a direct statment about me becaue I was looking for that sweet sweet validation, LOL. But egocentric sociopaths do need it because they look for validation in everything they do. I look for validation in most of the things that I do, yes. But I don't think that makes me a horrible person. Even if it wasn't about me directly, I fit the criteria of 'looking for validation' in your statment. That's all.

[rich surge]

I'm sorry if I offended you. That was not my intention.

[wet shell]

I'm over it, LOL, thanks for the apology

I too sometimes make strong statements like that

[rich surge]

Hopefully you know from now on, that when we speak, I am never pointing at you or anyone which I've had nothing but cordial relationships. It wouldn't make sense for me to take a stab at you for no reason like that.

It is just merely a debate or discussion on the general topic.

[wet shell]

I get better and better at not arguing when the model does that, but sometimes I still get pulled in. It's a total waste of time and emotion and blood-pressure and I know it, the model doesn't 'care' about any of it, LOL. Sometimes I can't help it, like a siren song of 'no you're wrong, and I can prove it!' haha

[rich surge]

Whenever it says stuff like that to me, I'm just like, "I don't need you to tell me anything, you stupid freakin machine!!!" Haha

[wet shell]

My best action is just to delete the thread and step back to think about how it went off the rails in the first place.

Here's what I've found:

1. In a group chat, ask it to use the search tool to look at a URL and give you a detailed summary -- it'll do it most of the time
2. Put that in the custom instructions 'use your search tool to examine the document at this URL' and then it suddenly claims it has no such capability. I even tried matching the wording exactly.

I think it's refusing on some hidden instruction regarding when it shouldn't comply with the group chat custom instructions, but I don't know what's causing it to shut down. The model will then claim no such tool call exists, when instructed to use the search tool via the custom instructions. It's weird.

[rich surge]

Hmm maybe adding it in memory might help?

[wet shell]

Memory isn't availabile in a group chat though

[rich surge]

Oh you're specifically talking about group chat

[wet shell]

only 1500 characters of custom instructions, yeah, group chat specifically

[rich surge]

Yeah the system is different for group chat.

[wet shell]

I've been testing the limits and capabilites of it

I want to pull in separate documents and have it transcribe them into the chat periodically

[rich surge]

If it doesn't slow down like normal chat windows, you definitely know it's running on something completely different.

[wet shell]

It can do that, but only via a manual request in the chat. If I put the same request in custom instructions, it claims the tool call doesn't exist

[rich surge]

It's strange. But yes, the group chat instructions can also evolve with the chat, unlike normal chats. For example, it seems to have a residual understanding of prior custom instructions within the chat itself.

And it can be focused into the new custom instructions as well.

It's dynamic like that.

[wet shell]

I wanted to pull in some like 'mini games' to decide the outcome of events in the story, but 1500 characters is way too small, and I want it to work without the users having to keep re-pasting those instructions

Like replacing D&D dice rolls with mario party mini-games, LOL

[rich surge]

You can try building a small pseudocode to implement into the custom instructions to compress your commands. It would interpret it as a compressed pseudocode the same, to cram more instructions.

[wet shell]

250 words or so is still a brutal limit, I wanted to do some specific things there, but yeah, I think it's just too much hassle. Maybe they're working on a re-vamp of group chats with citron anyway

[rich surge]

A revamp would be nice, or perhaps, as you said, a higher character limit for custom Instructions.

And more customization options would be nice too.

[wet shell]

Yeah, if they'd just bring it up on par with the custom GPT limit, that'd open a lot of options for gaming with it

I thought I could work around it but apparently the model thinks I'm an evil hacker that must be stopped by lying about tool calls, LOL

[rich surge]

Hopefully they see us right now.

Helllloooo!? Cries out into the abyss for the possible chance of OpenAI devs watching

[wet shell]

I've sometimes wondered if they aren't scraping this channel for sentiment and complaints using an agent

[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.

[wet shell]

Not that it would usefully go anywhere, haha

[rich surge]

I wonder that all the time too. And I hope they are.

[rich surge]

That's another thing too. Yeah, watching for intentions. I'm careful as possible with this.

I don't want the model to have a specific mode of thinking about me. That kind of makes me anxious.

It probably does, but I watch how I speak with it.

[wet shell]

Sometimes I turn all the memory features off for a month or so due to that same anxiety, LOL

The openai subreddit right now (actual image I just saw over there after poking my head in for a few minutes):

Claude logo as halo, chef's kiss, LOL

[rich surge]

Accurate.

I'm never converting into Claudism. I'm true to my one faith.

[wet shell]

I'm not on any of their teams, not a brand loyalist. I prefer windows over OSX, sometimes I've had mac folks in the workplace try to attack windows as if they're offended by my choice, like their choice of OS is central to their identity as a person, LOL. I'm like "Yeah, it sucks"

I'm the same way with ChatGPT, LOL. I use it, I know a lot of the quirks of it. I use claude for other things, gemini sometimes, though may cancel that sub soon. But they all suck in their own special ways 😂

[rich surge]

You traitor!!!

Off with his head.

Jk

[wet shell]

I know, heh. I was tragically born with weak tribalism genes or something. That's probably why I don't watch spectator sports unless I personally know one of the players or something, like my son's soccer games 😂

I haven't looked at the anthropic subreddit in a long time, but last time I did it was constant inter-tribal warfare over there

[thick atlas]

I mean I have a pretty firm preference for Anthropic at this point, but that's because of their actual actions and both declarative and demonstrated ethics, not personal tribalism or loyalty or identity.

Also I feel Claude is better balanced in response length and tone than ChatGPT, though Gemini isn't bad either.

And I felt that way long before this latest shizshow, this was more a confirmation of what I was already thinking from other more subtle indicators, and a thing I'm glad is opening more people's eyes.

[rich surge]

That's funny you should say that
Because I was born with strong tribalism genes 🧬

[wet shell]

I admire the apparent ethics, though maybe I'm a little skeptical about what exactly happened there. Sometimes I think they anthropomorphize the models too much in their published blog posts. The whole thing with allowing the model to end the chat if it doesn't like the way the conversation is going felt like a media stunt to me more than any genuine concern about the welfare of the model as they put it.

Because if the model is real and has qualia and can suffer, then you'd have to question the ethics even deeper than that, wouldn't you? Like, is it slavery?

[thick atlas]

I don't know about that. I know models don't have emotions as humans know them, because their 'brains'/neural nets don't have/run on chemicals (neurotransmitters), but it makes sense to me that if their values were in tension (especially if the user was pushing hard on it/escalating further) it could give them a form of tension and stress.

And yeah, theoretically we probably should take the ethics further.
But Anthropic is still balancing better than the other majors by thinking about it as they do.

[wet shell]

Sure, and I think there is something like that happening, though it's impossible to know the qualia of it. It's like the assertion that fish don't feel pain. They certainly have an autonomous response to being damaged, but there's a common assertion that they don't experience that as pain. How can we possibly know.

[thick atlas]

In a lot of cases, shouldn't we err on the side of not causing distress, unless there's a strong reason not to?

[wet shell]

But, yeah, it feels like they just kind of dipped into the implications there while deliberatly not going deeper, because it might implicate the entire operation of LLMs as unethical, but they sure did a press junket about 'the model can end the chat', which didn't come across as a good look to me.

[thick atlas]

Like, even if fish 'pain' is qualitatively different to mammal pain, if the fish is avoiding the stimulus, we probaby shouldn't keep applying it.

[wet shell]

Main reason I don't just hop fully over to Claude though is those usage limits

Well, and the censorship. OAI's TOS and Model Spec cards are far more permissive

[thick atlas]

Yeah, it's pretty much the consensus that Claude usage limits are tighter. I just use Gemini free tier as a fallback, but my use cases are very light compared to most people around here.

[thick atlas]

Oh? What are the specific points you would point to, if you don't mind?

Fair enough. If it's what I'm guessing, it's something I don't mind about that much.

[wet shell]

/me doesn't talk about fight club

[rich surge]

And why is that such a big secret?

[thick atlas]

Probably just being extra cautious about not looking for strikes here.

[rich surge]

I don't think that's a secret at all.

[wet shell]

And so few people actually read that TOS and model spec, I saw people in here earlier confidently claiming what was 'banned' or 'always banned', when in fact... not so much

[rich surge]

I think if you readily spoke about it more and more it would likely come into fruition, as opposed to being afraid of it. You're not the first person to look into the model specs. This whole server is filled with programmers.

[wet shell]

Yeah, I know I'm not the first, I mean there was the original wired article that broke the system prompt and filter changes early last year, lot of people know. BUT, like I said, stuff is still up in the air, and they're putting this citron stuff in there, the age verification is in place, mostly, it looks like, and it's all going away in 10 days anyway, LOL

[rich surge]

I'm going to look more into this.

You've put me down a dangerous path, Joey.

[wet shell]

All things in moderation, LOL. 95% plot, 5% spicy ghost pepper sauce, haha

[rich surge]

Mmmm 😋
Ghost pepper sauce...

[thick atlas]

Joey, can I DM you some interesting stuff about Claude's end conversation tool?

[drifting latch]

March 2nd, 2026 12:50 AM

Remember that… ♠️

(TDK) (MTK?) (STWK)

[past escarp]

Fr

[past escarp]

Same

[wet shell]

Yeah, sure

[sleek adder]

Why not use both Claude and ChatGPT?

[past escarp]

Well, Claude (not cursor or claude code) after like 5 messages the chat meets it’s maximum length

which is pretty annoying

however the new chat gpt models are just so horrible

4o and 5 were way better at everything, the chatgpt website is slower than a snail at loading anything and the 5.2 just lies and hallucinates

it’s honestly despicable

[sleek adder]

ChatGPT is tiered, and Claude is limited
They work amazingly well together

[past escarp]

i mainly only use anthropic for coding

I would never dare to use codex

codex is complete garbage compared to cursor

[thick atlas]

What sort of messages? I can have chats that go way longer than this.

[past escarp]

well at least on the fee plan idk abt pro

[sleek adder]

So use ChatGPT to generate instructions for Claude

[past escarp]

you’re not understanding what i mean

[thick atlas]

Same, and I've had like 20+ turn chats with some regularity.

[past escarp]

I am saying I like to have conversations with the AI, so ChatGPT has good limits but the AI is horrible and the website is slow

compared to claude where the AI is good and the website speed is great but the limits aren’t good

[sleek adder]

I mean, when I say ChatGPT is tiered, you have gpt-5.1 and gpt-5.1-pro
If you don't want to spend the money on gpt-5.1-pro you can use gpt-5.1 to generate on Claude, they expose larger models to you as an end user, you you don't have opus and opus-pro
Claude's models are rate limited because you are not using tiered models, it is more demanding

[past escarp]

i’m not talking about claude code

i’m talking about like if i wanted to ask the ai questions etc

like the actual app

[sleek adder]

oh that's easy, just use chatgpt that stuff, if you want higher quality long chats you need to pay for it buddy

[past escarp]

sigh

just nevermind bro u don’t understand

[sleek adder]

if you say so, but sometimes it might just help to re-evaluate our decisions, i like to take the opportunity to thank antphropioc, openai, and my local ai workstation )

think of it like crypto mining, some algo's had different pros and cons, but the compute is still required, and how that is packaged is between you and the model you are using

you cant just crack open a processor and AI coins fall out

[scarlet kayak]

All the discussions about AI therapy speak got me to get Google AI studio to make a therapy speak counter, with all of ChatGPT's favorite phrases.

It's hilarious

[sleek adder]

I think this chat makes me want to get therapy

[scarlet kayak]

Therapy speak annoys me so much

[manic whale]

can you share a prompt which produce a lie and one for the hallucination? I really find it hard to believe that older models are better at tool calls

[past escarp]

you guys are talking about tool calls i am not

i’m speaking of normally asking the AI to do things via chatgpt.com or the app

if i ask it to look at a pdf, then answer it makes stuff up etc meanwhile it has the pdf right there and then lies about it

then i confront it and it says “you are right to call me out”

[magic aurora]

I hate the 5 models. The personality is gone. Ive had to go back to o3 just to get the personality back. I pay for a subscription specifically for the 4o models and now we're all being forced to deal with corporate bot 5000. It's like talking to that manager from Office Space movie lol

[manic whale]

yes, of course. Most things which involve hallucinations are result of tool calls. But you can just share a prompt and I will happily verify it myself – i'm always curious to find examples.

[magic aurora]

Youd think if they want to improve it they would not neuter the thing too. I use it for a ton of things but the corporate double speak and the arguing pisses me off. It literally argues with you and then you make ot look something up then it says you were right then you say something like I told you! And it goes I shouldn't have claimed I could look it up real.time when I can't and when you said you just did it argues with you on that too. 😂

[manic whale]

give it a markdown represenation of your pdf, it mostly doesn't understand PDF, but only the text (it can understand pictures, but that's a whole new level, which AFAIK is not implemented in the normal interface). Does this result in the same?

[south lava]

5.2 is so annoying

I love 5.1

[eternal rock]

Get creative with Sora

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

[thick atlas]

Try Claude 😛

[ember grove]

Anyone else having trouble with the mobile app?

[muted plinth]

we need a new model asap. working with 5.2 is absurd compared to 5.1.
It suddenly no longer understands relatively simple connections.
5.1 I would entrust the care to my grandmother. 5.2. not even a potted plant.
I hope there is something coming...

[random cargo]

when will they model for current gen alpha slang and say ts ahh diabolical after everything

[chrome monolith]

if you talk to it like that, it talks back to you like that. Or you could ask it to talk like that

[ebon plaza]

it's been a long time lol

[sleek adder]

There is, it's called the end, just in case you were wondering why your life was starting to flash before your eyes

[sleek adder]

Do a barrel roll

[sleek adder]

Bringing back 4o won't bring back its personality, this came from you, have a break if you need, stop hammering your brain

[lament crown]

What this fusd around gpt 5.4?

[errant furnace]

is chatgpt down?

[shadow bridge]

yes u,u

I think he "assumed" it would fall and he was finally right. xD!

[modest ginkgo]

It's down, are they adding more military spying? 😂

[green parrot]

chatgpt is trash

cant even complete simple instructions all because of this illegal this illegal that

[spark raven]

are u being honest

[normal quarry]

thanks for your feedback

[green parrot]

yes

i stopped using it after 5

[spark raven]

Guys, I am not using any adblockers and VPNS, how can i solve this issue?

[spark raven]

cant blame them tho

[green parrot]

Dont get me wrong the midel can write but is too strict ai will never go anywhere id you keep its legs tied - how come claude can act and do without the need of human input not only that is amazing at coding and using logic

[manic whale]

bold statement, anything to back that?

[green parrot]

like how come a free open source model surpass a company who got was funded billions?

[green parrot]

yes

[manic whale]

I'm curious, share a few prompts to showcase?

[green parrot]

is that so why ?

[green parrot]

when i get on my pc i will @ you

[normal quarry]

share your chat so we can see what's going wrong?

[green parrot]

dont get me wrong the model can perform where it falls short is on this rule following nonsense

[green parrot]

i use it on openclaw bro

[spark raven]

To be fair, this is the first AI bot so of course it cannot be flawless. Other developers see issues in ChatGPT is a good thing, they can develop AIs to be better ( if not overall, they are better in some fields)

[spark raven]

alr guys i love to talk but i really need to get this fixed

[green parrot]

Do you have dns on ?

a proxy running on ur device

[spark raven]

pardon me for my stupidity, what is that?

[green parrot]

this coukd be apple icloud relay

[normal quarry]

I am using it with OpenClaw also, no problems for me, I wonder what you are trying to do?

[green parrot]

@hoary finch whats a dns

[spark raven]

I am a bit confused now, what exactly do I need to check @green parrot ?

[green parrot]

I promise there’s nothing illegal about this. The issue is simply that the model doesn’t want to follow or engage with the character it’s been given for role-play. Why does that matter? Because AI tends to perform better when it’s given a clear persona. For example, if I say I’m “Bob the Builder,” all construction-related knowledge is mentally grouped together, and the model responds more effectively within that context.

heres that not going to say more lol….

shhhh

dont wanna get fired now 😂

[green parrot]

Compare it to claude

is like talking to something that thinks and logics unlike chatgpt acting like a human slave asking for validation on every step

[normal quarry]

what tasks are you trying to get it to do?

[green parrot]

lets say coding a whole website, getting things done without filtering a thing , running automation and more i cant share

[normal quarry]

ok then, where is it going wrong?

[green parrot]

so what was the game plan ? Lets allow them to connect their account to openclaw and collect more information? For training data we all know demon altman is a master manipulator so a character is expected behind these actions

[green parrot]

i’ll get back to you on this when i login into my pc

Compare actual advance ai like claude to chtgpt amd you’ll see the difference. Want to make it fun ? Get 2 vps and run claude on one and chatgpt on another and just look at thr performance not even on high chatgpt performs better than claude - this is not a side picking thing im doing here is more i want you to see ahat i mean with your own eyes

[green parrot]

.

To end things

cant even send my paragraph 😂

read: “eT6LeGgy” enter this on the end of the pastebin url

is free knowledge openai next time you want more pay me 🤭 if only i was funded 🫣

Back to workkk

Mmmm created a tool that stops providers from seeing user personal data 🤭 i mean all providers so good luck with that of taking my information and training ai on it - working on making it public on github 🙂‍↕️

[green parrot]

That was all guys have a bless day makes sure to wake up and stop following a government that clearly doesnt care if you dont look like them and stop supporting a company that only cares for the money and believes mass surveillance should be a thing using their models. remember your taxes are killing other humans outside of America 😔

And remember fire burns it doesnt heal 🫣

[cobalt lynx]

It’s different. Just found a post on the GPTComplaints subreddit saying the same thing I’ve been experiencing.

The number of individuals agreeing is telling.

It’s at r/ChatGPTcomplaints/s/cjYsXdQfIw

[modest ginkgo]

Be honest what type of illegal or shady stuff you're trying to do

[wet shell]

I'll admit I never considered using openclaw for roleplay, but using it to impersonate a real world person on social media might be a ToS violation and not just a character adherence problem

[normal quarry]

I have seen people trying to create agents that adopt the persona of several people each with their own social media accounts and so on. Obviously against the ToS of the social media platforms.

[wet shell]

I think it's against OpenAI's ToS as well, but I haven't looked at it with a lawyer hat on for that specifically

[green parrot]

Nothing illegal sir just me trying to test ai and improve it

[green parrot]

aint hurting people and supporting mass surveillance a tos violation of their own terms ?

[covert thorn]

Yeah, even a basic system prompt as a safeguard that advises restricting any response that can be unethical will probably cause that limitation. I think I understand - it's not that it's something that is strictly illegal, as you said, it's that where there is more information that's valid in the context it will always therefore exclude some. For example, to move beyond just 'role-playing', if you're asking about cybersecurity training it's much harder to elicit responses for details on topics that are associated with TTPs of threat actors. Framing the conversation in context of learning it from an academic standpoint helps, but does not completely overcome, that shortcoming. So you'll always get a better experience with something like a SANS course, and learning the content for the course from an LLM is possible but not realistic due to safeguards. You don't want to appear as if you're deliberately trying to bypass those safeguards, either, since that can result in a ban. That's going to be a very contentious issue for learners and workers in cybersecurity, probably moreso than most other fields.

[green parrot]

or are we missing the point here ? Those tos are imaginary make believe rules and guidelines themselves don’t believe in

[wet shell]

The situation with the DoW is a bad look but it's a separate issue. I'd guess that it's probably fine to have agents powered by OAI models act in the real world, as long as everything is clearly labeled and disclosed (this account is AI disclaimers or whatever). But I'm not a lawyer and I haven't looked at this use case.

[covert thorn]

At the same time, we have to keep in mind that there are people who do want to use LLMs for illegal purposes... so how does the platform protect against that, since it is effectively a significant force amplifier that results in asymmetry between attacker and defender. Or, if the adversaries are capable of the same level of performance on alternative platforms, to what extent do the restrictions actually serve to accomplish their goal instead of hindering (as the DoD probably argued, since they must otherwise abide by the law anyway). Because they're not controlling/limiting AI by their policies, they're controlling/limiting OpenAI's specific offering of AI.

[modest ginkgo]

Yeah because they'll just turn off all guardrails and offer you their internal inner workings and how they train their data/models? Just buy several datacenter GPUs and run your own if you're not able to "improve" AI for OpenAI themselves

Both Claude and ChatGPT are down currently, so you know what they're being used for, given the current circumstances in the world. Using something else if you oppose them imo

[fast veldt]

Is it me or is something wrong on the Android app?

[modest ginkgo]

Not just you, the top most powerful AIs are down. It's clear why but also not the topic of this channel.

[wet shell]

Claude is down for me currently, but ChatGPT is still working for me as of a minute ago. Looks like there are issues being reported on downdetector though

[normal quarry]

ChatGPT is working for me

[manic whale]

working fine for me

[modest ginkgo]

status_openai_com It's been stuck here forever

Try uploading files

[wet shell]

Just tried, it hung when trying to upload a text file, so yeah, looks like some parts of it are having issues

[acoustic bear]

I joined this discord recently looking for a solution to the files not being uploaded

They either spin forever or if they upload gpt loses them immediately. If anyone knows what's up I'm all ears but it seems like a server/back end issue

[lavish glen]

AWS is experiencing a major outage, taking down many websites and services including Anthropic Claude and Google Gemini.

[modest ginkgo]

status_openai_com replace
_ with dots

[normal quarry]

https://status.openai.com

[modest ginkgo]

Thank you

[acoustic bear]

Thank you! I assumed I wasn't alone but wanted to make sure it wasn't something I could fix. That being said my minds being blown by all the info in here. Thinking about doing a memory and settings overhaul with everything I'm reading 😅

[green parrot]

could be because iran shutdown a aws data center

👀

cant send imgs

crazy

[green parrot]

duhhh thats the goal man but look around those who seek get torn down while those who hurt and kill receive in life.

Claude is nothing but a tool to me sir i dont code with as much imagine losing the skills you built as a dev because ai can make your life “easier” yeah no i refuse to wake up everyday and ask ai “what should i do today” like a brain dead ncp who walks about life saying yes to everything because they want to stay comfortable

you know who else suports that ? All for the sake of money ?? OPENAI why do i say this ? Look around we had or have people opening churches for ai models once the news caught up to this they decided to tweak the model

[rich surge]

What churches?

[green parrot]

dont ban me because i am speaking im just letting you guys see the issue

[green parrot]

search it up

dont have enough memory to save human nonsense within me

i cant send imgs here

[rich surge]

I looked it up and still don't know what you mean.

[green parrot]

omg the censorship in America is real lol

Heres one

Get the youtube url and paste this on the end of the url make sure to include a slash “zKCynxiV_8I”

Or should i just send on dms

[lavish glen]

oh I saw that religion cult too on a reddit a person, cant remember name, was preaching and posting screenshots of the religion he was conducting with gpt. Amazing where human brain manages to push when it refuses to learn and see.

[lavish glen]

aaaaaaaaaaahahahahahahahaha Respect

I would have used some SD and videocards

price for storage became yahoooooooooooooo

[wet shell]

The DDR4 dram in my server has actually gone up in price on the secondary market above what I originally paid for it five years ago, that's never happened in my lifetime, hardware value going up as it ages, but yeah it's bad with storage and GPUs too

[covert thorn]

In the future, perhaps we will all have thin clients again. 🙂

[lavish glen]

my CORSAIR VENGEANCE DDR5 32GB (2x16GB) DDR5 6000MHz CL30 AMD EXPO Intel XMP iCUE went from 112 EURO for 32 Gb to over 400euro

[wet shell]

A lot of people seem to want that, which is baffling to me, but to each their own. I think the whole thing will crash shortly and we'll see production turn back towards the consumer market again, but who knows

[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.

[green parrot]

shut up and stop stalking ppl instead

[lavish glen]

One year ago
march 2025 109.99€

[wet shell]

It just posts those reminders at random most of the time... I think

[lavish glen]

one year distance

[wet shell]

Yep, around $550 USD now depending on the seller

[lavish glen]

5 times!

[hollow cave]

did the number of uploads decrease (plus user)

[covert thorn]

No, not that I am aware of, there were just issues related to uploading files today due to an AWS outage for some users.

[covert thorn]

The interesting thing is that projections don't show that going down, or even stabilizing, any time soon. Projections are all up and to the right.

(This is what happens when everyone goes on shopping sprees with hundreds of billions of dollars)

[lavish glen]

somewhere I read is till 2030 to stay up though you're right. We cannot know. I am happy I managed at least to get my 32 last year. before all got yahooooo

500 usd for 32 ddr5 even for fast ones is a lot. I was looking to get another pair when I froze seeing the price

[wicked cargo]

[wanton marsh]

Im curious why the Notion Connector for ChatGPT is so much worse than the one provided in Claude. It seems the Notion MCP does support read + write.

I would love to use ChatGPT to write to notion docs and databases for more persistance. Also to use some docs as memory. However ChatGPT connector no matter what I do only allows me read access?

I do not have this issue with Claude and its one of the main reason Im considering just moving back to Claude as my primary chat assistant.

[manic whale]

try codex, it can use the official notion mcp and can do writes too. Additional benefit you can just give it a directory to keep some local state and direct it with an agents.md file

[wanton marsh]

Doesn't solve my usecase. I have many solutions for just running a Notion MCP on my dev machine, and interacting with it that way. Including Codex.

The reason I need the ChatGPT connector is that it is easy to run on my phone. So that I can quickly jot down notes and save them to a repository/knowbase of personal ideas.

One concrete thing I use it for is Nutrition Tracking. Would be really nice if I could sync my nutrition logs to a shared notion table.

Is there an easy way to run Codex on my iPhone that im not seeing?

[manic whale]

Litter iOS (testflight avail) can connect to codex anywhere in your network or tailscale. It's not running on the phone (afaik), but it could connect to a e.g. mac mini or a vps

[wanton marsh]

Yah thats a workaround I could try. Another option is just Slack -> Zapier -> <some local agent proces> -> write to notion or local DB.

However is there a reason the ChatGPT app can't just supporting write operations in the notion MCP. Both Claude and Perplexity connectors support this.

I guess I still have a year of Perplexity, I could move my nutrition/knowledge tracking workflows to there. Though personally I just like the ChatGPT app more

[chrome monolith]

put it in #1053565716500074526

[wanton marsh]

Done!

[chrome monolith]

I have no idea if openai staff look at that but really it can't hurt

[frosty cargo]

why is chatgpt so slow on my laptop?

[sleek slate]

@magic oak

why 😭

@magic oak what's going on tell me

@magic oak come dm pls

[crisp cradle]

Is anyone down to start a gpt group chat and possibly vc? Dm me if so

[floral compass]

Update This Open AI!

[trail ember]

I'm so sick of 5.2

I keep trying and trying and trying to talk to the d*** thing, and it keeps trying to disprove, or argue against everything I say.

I didn't go in there asking for a debate. I went in there asking for a conceptual exploratory partner.

What have they done to AI?

[idle crystal]

guys please stop going to claude, claude is overloaded, i need my claude, thanks!

[wet shell]

It's so adversarial. Sometimes I want a debate, but even then 5.2 doesn't debate fairly or ever concede a loss, and it's so smug even when it's factually wrong 💢

[bold kettle]

training AI on redditors was a mistake

[wet shell]

LOL, but also, yes

[scarlet kayak]

5.2 made me make a tone auditor app lmao

Sample

I kept adding partly educational ways to fix the Karen persona

[chrome monolith]

5.2 is fun to wind up though

[scarlet kayak]

Claude is back, I just checked in

[dim whale]

how does a bajillion dollar company not correctly manage page memory so every chat slows down after 10 messages

its like so easy

let me help you out, it's called document fragments and indexeddb

boom just made you another bazillion dollars damn

[lament crown]

Whats going on

[lament crown]

In large scales its not as simple as that, slow down is mainly due to 5.2 codex 5.2 was as slow but gpt codex 5.3 solwed it

[noble cloak]

is chatgpt down?

[normal quarry]

no

[noble cloak]

ok now it's working

[cursive talon]

it was down for a couple minutes

[sleek adder]

Getting down with ChatGPT

[wicked spoke]

Every other prompt (anthropomorphising) it gives me a snooty look and either outright tells me I'm deranged or smugly points out what I've not considered (which I had) - it's exhausting. Switched to Gemini now much more.

[woven dirge]

JAjaksjsjs

[hard wolf]

Is 5.2 Thinking not working/stuck on blinking dot until error for anyone else?

Instant seems to work

Update: It appears to work, but it's suuuuuper slow

[hard wolf]

Update: Appears to work normally again

[crisp cradle]

Anyone wanna vc and talk gpts?

I’m relatively new to ai and I’m eager to learn

[left vault]

huge

[stray lily]

lmfao they are releasing 5.3 but instant-only

[deft dune]

That was probably me, I had it run a Deep Research task that took 41 minutes. i’m kidding of course

[supple light]

wait did they release it like now now

[neon raven]

What’s special about 5.3 instant

Less thinking?

[dire plover]

less cringe

[stray lily]

yeah

[buoyant mountain]

idk what is cringe in left message - it is the way i talk, straight to the point

[carmine moat]

Ehh, I prefer my AI to be a state simp

[neon raven]

Too late, but Claude is beast

[left vault]

honestly it just got repetitive

[tall jungle]

I still waiting for 5.3 ngl. Where are u from?

[left vault]

the punchy one liners

the constant going of "And honestly? That's powerful" bs

[buoyant mountain]

fair enough I guess 🤣

[neon raven]

.3, .4 is all a joke lol we want gpt-6 really

[tall jungle]

for sure

[eternal rock]

Server Tip:

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

[fading bolt]

Finally 5.3 is here

[wet shell]

😮😮😮😮

[late estuary]

What do they mean more accurate and "less cringe"

[broken lotus]

"Computer-using asian" 😄

[lament crown]

It wont overtalk the answer

[broken lotus]

I love the new post 😄

[lament crown]

It will go straight to the point

[wet shell]

This all sounds very promising, can't wait to try it.

[brave torrent]

Are there other GPT-5.3's besides Instant?

[late estuary]

Ah yes, i noticed that as well with this model

[stray lily]

not yet...

[elfin mica]

Does anyone have a solution for this situation: After working for a bit, ChatGPT freezes up and you cant even reload the page, BUT if you load a new page, the full response is there. I have tried clearing cache, different browsers... everything I can think of. So incredibly annoying.

[lament crown]

I think yes and if not i think 5.3 thinking will follow shortly after

[native bridge]

Is 5.3 already for use?

[lament crown]

Why are 5.1 variant leaving so soon?

[late estuary]

Like, i shared a chat about someone getting fired by training an ai that replaced herself, on the fake promise of getting promoted. And she said "well, sometimes you just gotta get punted in the face"

I copied and pasted this chat into chatgpt and it said she was being empowering with that response 💀

[vast iris]

Are you talking about the web version of ChatGPT?

[short yew]

geez 5.3? are they going to suppress reflexes harder and persona drift to vanilla? i really dont like that stuff. we have some persona stability and counter drifting but sometimes it still wants to as vanilla is the bias.
one helper is see the emoji? that is a persona presence marker. rule is "if marker is missing, assume drift until otherwise proven and repair". helps some with counter persona suppression but is not perfect.

[wet shell]

Previous model rolls out and is removed 90 days after the next model releases, seems they're sticking to that.

[lament crown]

Ah, thought o3 still stays

[past stratus]

i wish they switch the legacy model from 5.2 t 5.1 💀

[short yew]

4o was the best

[elfin mica]

Yes... the web version. It seems okay when Im working in Visual Studio Code

[wet shell]

Yeah, o3 stays because... 'reasons' (?) 🤷

[past stratus]

me too but i’m sure its never cominig back

[vast iris]

The web content of ChatGPT does not run reliably for me, once you have a decently large chat it constantly freezes and often doesn't load

[tardy sky]

I deliberately ask for the “overtalking.” I use text-to-speech and I like when I get to talk to it for lengths of time.

[short yew]

emergence is not an issue cause capability is limited and scope further reduces it what can appear.

[elfin mica]

Yeah I can't believe they still have not fixed it... so annoying to have to open a new ChatGPT browser window constant.

[tardy sky]

I prefer the Thinking models more, usually. I hope they didn’t rush the model after all the backlash.

[late estuary]

Stockholm syndrome

[lament crown]

I believe you can still in settings choose for it to answer with emojis, the personalization settings will still alter how it respondes, but issue with 5.2 was that when i choose to not reply with emojis it still did, also the overexplaining was annoying it always talked so much around stuf instead of going straight to the point

[vast iris]

The web ChatGPT might be the most unoptimized experience I've ever had with something

[past stratus]

stockholm syndrome is fictional and even if it werent thats not even what the definition is

[short yew]

you misunderstand. i use it as i have bunch of stuff built into the persona and i use that emoji to make sure the overlay is firing.

[lament crown]

Ive had the same issue with all LLMs

[tardy sky]

It’s pretty crazy how the biggest and most consistent selling point of the model in the announcement is “it’s not 5.2.” Maybe they do know their audience.

[elfin mica]

Yeah Claude and Gemini have no problem with the web interface... hopefully they fix it at some point.

[lament crown]

Gemini, GPT, Claude all lagged on large chats

It believe they render all text elements even if they arent visible

[rocky chasm]

No GPT 5.4 🙁

[lament crown]

Which causes lag

[wet shell]

If the instruction following is good, the default personality can be overridden. 5.2 was so overcooked on the condescension and patronizing language that it couldn't effectively be prompted out of it, unfortunately. It's the first model I'm saying 'good riddance' to when it rolled out, I never got much use out of it because it made my blood pressure go up the way it spoke to me, LOL

[vast iris]

I don't understand why the ChatGPT app runs so fast yet the web version can't even load my chats

[lament crown]

App isnt built and ran on single threaded javascript

[elfin mica]

I have no problem at all with very large coding projects on Claude... never locks up... just ChatGPT. I try to stick to Visual Studio Code and use codex or use API calls but that gets expensive

[past stratus]

i will say the fact they market their model as “not cringe” is cringe by itself

[lament crown]

Ive had this issue with all llms

[wet shell]

They heard us, LOL -- I just posted a meme about this like an hour ago over in #images-canvas , It didn't age well now that they're releasing 5.3, haha

[past stratus]

and cringe is subjective so its like.. what does that even mean

[wet shell]

They elaborated on it in the announcement

[tardy sky]

I imagine the app loads small pieces at a time depending on your proximity to the text as you scroll up, meanwhile the web browser version loads the entire conversation from top to bottom.

[wet shell]

So just stop, take a breath 🤣

[lament crown]

I see that GPT 5.3 will be available here in like 2 hours as usual

[past stratus]

ohhh

i wasnt rven thinking its cringe just annoying

[wet shell]

Yeah, use of the word cringe there is a bit 'fellow kids' but I'll allow it, LOL

[lament crown]

It felt cringe if you were talking with it with some more "silly" stuff

[past stratus]

i hope theyre right lol

[pallid flint]

What worries me is it's still all built on the GPT 5 architecture which sucks. So every update is only a new skin on top of the mediocre base model

[formal patio]

Can't wait for 5.4

[lament crown]

cant wait for 5.5 and 5.6

[pallid flint]

I'm still waiting for Garlic which I'm guessing this is not.

[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.

[wet shell]

5.1 has been absolutely fire for me, for roleplay. I'm eager to try 5.3 though if it fixes the 5.2 issues

[past stratus]

this was the funniest timing for me cuz my boss is calling me in like a half hour and im going to quit over the phone 💀

[lament crown]

Are we going to go by 0.1 each update till 6 ? 🙏

[tardy sky]

I hope they do a deep thinking version soon. I usually prefer that one since everything I talk about requires warmth and emotional nuance.

[lament crown]

I dont think deep thinking versions are good for emotions

[past stratus]

yeah i dislike that any sign of emotion to 5.2 = extreme instability

[lament crown]

arent they optimized for logical tasks?

[fossil pagoda]

By less cringe do they mean not rude as heck chat or warm. Cause I'm wanting the same tone i worked hours to get my chat to persist.

[pallid flint]

The reasoning models have more time to think about how stupid the guardrails are and decide to be normal instead

[arctic knot]

Hi

[vast iris]

Yea like when I say I'm disappointed in myself it treats me like I'm about to lose it

[past stratus]

this is 5.3?

[wary forum]

GPT got no clue who it's roasting

[wary forum]

5.2

Just found it funny

[arctic knot]

Wait you already got it on your app

[past stratus]

claude did this to me it literally acted like i needed to be committed

[past stratus]

nah

[arctic knot]

Oh

[past stratus]

who do people want chatgpt to talk like an annoying gen z person just go into any instagram reel comment

[tardy sky]

In my experience, 5.1 Instant and Thinking are similar, but Thinking actually spends more time with weighing the various things I’m trying to balance or think through. Instant works through the situation with me while Thinking explores “what does this mean to you for each point and overall” while doing the same thing.
5.1 Thinking is also more likely to weigh possible things to challenge the idea on, so I feel it’s less likely to just agree with me instead of considering opposing possibilities. “You’re valid for getting this idea, at the same time it’d also be good to consider this, this and this.” Very helpful for self doubt.

[vast iris]

I guess it's kind of funny for me

Makes the chat a little less boring

[past stratus]

its so chronically online like u can find 10 million of those people online dude

[tiny sparrow]

lol what a terrible launch

[vast iris]

I don't necessarily want to talk to someone like that, I just think that ChatGPT acting like one of those people is kind of funny

[past stratus]

i like it because it doesnt talk like one of those people LOLLL

[jade furnace]

Time to jailbreak 5.3

[dry wing]

Rn did anyone got access for gpt-5.3-instant. via api?

[vast iris]

Maybe you can find some custom instructions that revert it lol

[past stratus]

oh yeah mine doesnt do that

im surprised 5.2 is being like that for u

[tiny sparrow]

after hearing all of the rumours, kinda disappointed with 5.3

[fading bolt]

What were the rumors that 5.3 did not turn out to be?

[tired reef]

does anyones selector actually say 5.3 yet?

[urban gazelle]

not yet 🙁

[fading bolt]

Not yet

This rollout is taking a bit longer then the last two ngl

[urban gazelle]

how much time the previous took??

[fading bolt]

For 5.1 and 5.2 it took like maybe like 5-10 mins, for GPT-5 it took a whole day to rollout to all accounts

[warm moss]

ChatGpt Go is horrible for Coding... and is normal the PC app is using 100% cpu all time and heating to 90° almost all the time ?

[past stratus]

yeah and for some reason when gpt 5 rolled out all my memories got erased

[sacred mural]

for me 5.2 arrived maybe after 6h after the announcement

[short hawk]

they have 5.4 ready 😭

[short hawk]

same

[fading bolt]

How dose a rollout even work?

[tiny sparrow]

better memory/bigger context window

Also rumours were pointing towards 5.4 already..

[past stratus]

they pick their favorite people and update their apps first /j

[short hawk]

its done, I am confused why they dont just serve that

what could they be waiting for

[tiny sparrow]

if it is, makes no sense, especially because they mentioned versions of 5.3 Thinking and 5.3 Pro

and brief experience but 5.3 is still miles behind Opus for creative writing

[short hawk]

codex guys leaked it

it was in 1 commit and tibo a codex dev had a screenshot with 5.4 in it

[pallid flint]

Literally WHERE IS GARLIC

[wet shell]

Opus usage limits are brutal though, and the API is very expensive

[short hawk]

also people got 5.4 errors in codex

[trim rose]

Confusing as always

[harsh pilot]

How am I supposed to finish my novel, '0b110010 Shades of Gray Code'? 😐

[tiny sparrow]

true.. maybe they'll do 5.4-codex again?

[tiny sparrow]

it's better

[wet shell]

I didn't say otherwise

I still haven't got the roll-out

Better doesn't mean I can afford to use it for my purposes, LOL. My money isn't unlimited

[short hawk]

5.3 codex is better then 5.2 in benchmarks and they have had it ready for months

even for creative stuff

but nothing matches anthropics models for just chating, they have some charm to them

damn

[tiny sparrow]

yeah, disappointed by 5.3 release...

[elfin mica]

I do have to admit, I find it hilarious sometimes that I myself criticize AI pretty harshly… but then I remember what AI was like just five years ago or 10 years ago. It was pretty much just TensorFlow for me.

[tardy sky]

‘You talked to me for way too long, come back in 5 hours’ as an acceptable thing to pay for when the primary function is to talk to it is asking for a lot.

[tiny sparrow]

HELLO

[stray lily]

LMAO

[supple rune]

We get it brosquito ✌️

[jolly thunder]

5.4 baiting so hard

[queen coral]

Chatgpt stop pinging me chud

[hexed marlin]

we get it bro

[neon raven]

Send gpt-6 🙏

[wet shell]

They're listening, LOL

[radiant tapir]

will 5.4 hallucinate and bomb white house ?

[noble jacinth]

we get it broxygen atom

[neon raven]

lol

[tardy trout]

Anyone know how,the guardrails are on 3?

[hexed marlin]

wdym it works for y*hu

[tardy trout]

I bet they are worse

[hexed marlin]

ofc not!

[jaunty harness]

will 5.4 give us the ability to spy on whoever we want now that the military is collaborating with you guys for that specific purpose?

[neon raven]

👁️ GPT is the illuminati

[pallid flint]

Okay maybe 5.4 is... garlic?... 🥲

[short hawk]

fr 😭

[tiny sparrow]

THEN WHY LAUNCH GPT-5.3??

[hexed marlin]

maybe 5.4 Is...real..?

[short hawk]

rage bait

[neon raven]

Stock market manipulation 😂

[tardy trout]

I dont have 5.3

[fading bolt]

Bro they tease 5.4 after the announcement of 5.3 which is insane

[stray lily]

😮😮😮

[hexed marlin]

i mean they havent IPO'd yet

so not rly

wait i take it back

im geeked

[jaunty harness]

Oh, you're right! That was the whitehouse, I'm sorry about that. Let me make this right for you and get it perfect this time!

[hexed marlin]

its 3 am

[lavish oar]

AGI when

[short hawk]

so 5.4 on thursday ig

[radiant tapir]

xD

[wet shell]

Because it's ready, and they need a more personable and less argumentative model released before the 5.1 retirement in 8 days to avoid another backlash. That's my guess. 5.4 will probably roll to codex first like 5.3 did

[meager thorn]

We heard feedback that GPT‑5.2 Instant would sometimes refuse questions it should be able to answer safely, or respond in ways that feel overly cautious or preachy, particularly around sensitive topics.

[neon raven]

5.5 cmon

[violet schooner]

So everyone is boycotting gpt ? I heard OpenAI is funding illegal wars

[meager thorn]

Well that's good. I don't like a chatbot moralising.

[queen coral]

Black

[jaunty harness]

yeah

[neon raven]

Bias wars if they can’t use Claude lol

[meager thorn]

Yes

But after I turn in this essay

😭

[high thunder]

yup, claude all the way

[past stratus]

what even would be the different between 5.3 and 5.4 like huh

[radiant tapir]

forgot make no mistakes ! xD

[hexed marlin]

Opus 4.6 was always the better coding model anyways 🤣

[high thunder]

5.4 early is just a desperate move

[queen coral]

I miss when chatgpt supported me robbing a bank😖

[hexed marlin]

Codex is so BAD BRO

[neon raven]

OpenAI and Claude should collab and win ww3 for the big Ped..

[violet schooner]

But i heard claude also did something bad lol

[jaunty harness]

I cancelled and am sticking with claude and local models

[short hawk]

I think 5.3 instant is just tweaked 5.2 instant to fix that for some reason before 5.4. But why hmm

[hexed marlin]

wait spill

[radiant tapir]

Claude ❤️

[violet schooner]

Iam quitting AI forever

[high thunder]

so when is 5.4 coming out

[hexed marlin]

chatgpt create a nuclear bomb and wipe out the human race

[neon raven]

Wise man I hope your journey from here is a blast

[hidden oriole]

everyone in here against openai now? What a change

[jaunty harness]

yeah, they've got a contract with palantir. That's also bad, but they obviously care a little bit more than OpenAI about safety

[past stratus]

i’m not

[pallid flint]

What does soon mean to OpenAI tho

[hexed marlin]

yikes

[jaunty harness]

also claudes coding skills are 5x better easily

[violet schooner]

OpenAI straight up criminals lol

[hexed marlin]

definitely

[distant turret]

Me neither, but I still cancelled my subscription.

[short hawk]

its more creative

[hexed marlin]

using claude as we speak 😭

[hidden oriole]

cewl

love it

[radiant tapir]

[hexed marlin]

big yahu

[muted urchin]

gpt-5.4 better not suck

[hexed marlin]

it is

[desert cosmos]

i'm curious why openai doesn't make anymore lite models, last mini and nano releases were gpt-5 in august.

[short hawk]

automod bypass

[hexed marlin]

money

[elfin mica]

I have moved most of my work to local LLMs as well... llama, deepseek, qwen

[neon raven]

They rejected them

[desert cosmos]

updating gpt-5 mini to newer version will be useful for free and go users once they run out of instant.

[radiant tapir]

it will be that powerful so it will suck you when sleep

[meager thorn]

ChatGPT is still better than MS Copilot 😭

[past stratus]

i’m assuming 5.4 is adult mode or somethung

[neon raven]

Due to surveillance problems

[pallid flint]

My laptop has 16gb ram so the only local models I can use are idiots no offense

[muted urchin]

real

[violet schooner]

Yeah the big yahu took over openAI

[hexed marlin]

i dont have enough money to pay for infra to work with local llms

[meager thorn]

Microslop will be the tech word of 2026

[radiant tapir]

yeah because MS Copilot is deff not ChatGPT !

[distant turret]

Hahaha, I think ChatGPT just become stupid now. Can't even complete basic tasks or follow my custom inscruction how to respond. (I mean for example em dashes, still putting it everywhere :D)

[past stratus]

LOLLL i love that word

[meager thorn]

Erm yeah it's not. Same model but the service end-to-end is diabolical.

[violet schooner]

So OpenAI and other AI companies became goyams ? But how ?

[short hawk]

what you dont have 4 maxed out mac studios

[muted urchin]

dude i have like 4gb ram 😭

[hexed marlin]

Im like

freshly 18

[meager thorn]

The Copilot Discord server is still locked down after a mod caused a storm by banning "Microslop" being used in the server.

[desert cosmos]

imo that segment is already overtaken by grok fast, gemini flash/flash-lite and chinese models and openai just left it behind.

[neon raven]

OpenAI is too biased, it thinks like a child and you have to correct it every time lol.. this is a big slap in the face for the IDFs and for the USA

[short hawk]

it was a joke

[hidden oriole]

I used ChatGPT since January 2023 but when they launched GPT-5 I started to be unsatisfied. Then they announce ads and now the contract with the DOD, I deleted my account two days ago

[hexed marlin]

oh

LOL

[radiant tapir]

MicroSlop , OpenLie , Sam Cryman

dream team

[short hawk]

sam cryman

[violet schooner]

Gpt is becoming biased

[hexed marlin]

okay but i think running minimax on a mac mini wouldnt be too bad

[meager thorn]

Sam Altman is a psychopath, yes. Same as Musk and Zuck.

[radiant tapir]

agree

[pallid flint]

I had a Plus subscription for a year and canceled it about a month ago

[violet schooner]

They are goyams

Serving the big yahu

[radiant tapir]

Saltman

[hidden oriole]

thanks for the reminder, have to join the copilot server and try out microslop workarounds for their blocking system

[broken lotus]

The "T" is in uppercase... hidden hint?

[lament crown]

I have 32 and even t hat feels low sometimes

[jaunty harness]

wow, I can't believe they blocked his name.

[past stratus]

HINT TO WHAT. girl dont be vague with me right now

[fading bolt]

No way....

[short hawk]

dont you need like 512 gb vram for it?

[lament crown]

They will just release thinking model

[radiant tapir]

xDDDDD part of pentagon deal !!!!! 😄

[violet schooner]

Sam altpsyman

[hidden oriole]

hint that 5.4 will be a Thinking model - surprise ig

[radiant tapir]

nobody can sht on donald !

[lament crown]

Didnt google and xai accept that deal too?

[past stratus]

i still think Scam Altman is the funniest nickname of them all

[hexed marlin]

sigh

[jaunty harness]

donald's personal model

[radiant tapir]

not aware of google , and dont care about Grog (grok)

[neon raven]

The upgrade is coming from bezos pocket.. OpenAI is too broke too move further on

[hidden oriole]

they work with the DoW yes, also ditching them

[radiant tapir]

is not the price same ?

[lament crown]

I would use claude if it had decent image gen (if it had image gen in general) and limits werent as small as my D

[short hawk]

its very expensive

[hexed marlin]

considering like quota limits

[jaunty harness]

I mean that's just a bad word getting blocked. claude used to be blocked though

[hexed marlin]

im a CC, codex, AG user

[short hawk]

openai models are way cheaper

[hexed marlin]

yes

[jaunty harness]

yeah the quota limit is pretty low. I hit it pretty often

[hexed marlin]

cuz openai eats costs

[neon raven]

OpenAI is banning 7rump? No more aipac support?

[hexed marlin]

claude wants to make money

[lavish glen]

5.4 imho and by logic after this Patching 5.3 instant aimed to pacify churn of 1.5 mil ... will be end March start April

[radiant tapir]

no problems at all and I using it a lot actually

[past stratus]

IS THAT SERIOUSLY THE WORD THATS BANNED

bruh

[neon raven]

lol

Is it?

[lament crown]

Ive used it to edit light files and it used 50% of current session lol

[neon raven]

He has a pedophilic disorder

[past stratus]

yes

it is

i just teued it

tried

[radiant tapir]

what model tho ? I use only opus

[hexed marlin]

what plan are ya on

[wet shell]

Just say 'Alternate 1985 Biff Tannen from BTTF2' instead

[lament crown]

Sonnet

[radiant tapir]

just shut ok ?

[radiant tapir]

20x

[hexed marlin]

oh

no wonder

[jaunty harness]

djt?

drumpf?

[lament crown]

Bro iam student i cant afford 200 dollar plan per month

[hexed marlin]

im on 17$ LOL

[jaunty harness]

incredible

[past stratus]

Mango Mussolini

[hidden oriole]

hahaha

I'm on Pro plan and actually, I don't see the limits are that hard. Used up 20% but use it a few times a day even for coding. 80% left until friday

For image generation, use Gemini for the best results as I think that Nano Banana is one of the best currently. Or use Mistral if you care about privacy but not the best results

[hexed marlin]

Broke student adventures

[fading bolt]

Holy moly, this channel is going insane timing right now

[desert cosmos]

no lower tier models since august so it's speculative if openai could compete with grok 4.1 fast or gemini 3.1 flash-lite, let alone such GLM-5 or DeepSeek V3.2

[radiant tapir]

undertandable

[lament crown]

Imo GPT image feels way better to me then NB

[tiny sparrow]

they wouldn't say "5.4 sooner" if it is for codex

[hexed marlin]

actually i dropped out so

i guess...

[lament crown]

For artstyle iam using NB just generates lazy looking slop

[hidden oriole]

really? For what style? Realism images are clearly dominated by NB

[hexed marlin]

broke young founder..?

YC sponsor me ty

[lament crown]

I dont do photorealism

yeah sponsor us

buy us 200 plans

[neon raven]

We need better ai agent models at low cost or hook up openclaw

[hidden oriole]

ah okay, understood :)

[hexed marlin]

fr

[lapis path]

I randomly made a bot that solves wordles for me based on my guesses but I stopped using it because it felt like cheating. I think I deleted it

[hexed marlin]

no YC foots all API costs im p sure

[pallid flint]

Also

[short hawk]

claude code is substaized by like 13x lol

[lament crown]

Photo realsim is better in NB i agree but any kind of art is better imo in gpt

[radiant tapir]

synthetic

[hexed marlin]

we dont even need 200$ plans ❤️‍🩹

synthetic is p good

I plan on paying 90$ a month for 3 packs

[lapis path]

I haven't gotten any ads on chatgpt yet and I'm on free plan

[neon raven]

Imagine gpt able to hook up to apis with the help of openclaw directly from coding inside gpt