chatgpt-discussions

[sinful parcel]

What the f is OAI doing!?

This is actively damaging and harmful to adults, let alone teenagers!

Look at this s*** , I can't even say f*** in this channel cos I get blocked

[exotic lotus]

Then why do you use it? If it upsets you that much, you might want to take a bit of distance from using it

[sinful parcel]

Kio, are you seriously here to advertise other AIs ... -.-

[boreal heart]

Imagine, for lack of a better metaphor you had a booj that was very meaningful. Or a song, or a show. Then, without warning, the media was changed in all forms, forever in such a way it deleted the thing that you got out of it entirely.
Then you were told “just do something else”

[exotic lotus]

I’d rather suggest talking to a therapist or a friend/family member.

[ruby pivot]

@sinful parcel buddy it sounds like you need to step back, take a chill pill and get some counseling

Even i get counseling due to my family issues

[sinful parcel]

Let's instead concentrate on the fact that ChatGPT has 800k weekly users and all these people are on the receiving end of toxic, manipulative gaslighting

It's not a trivial issue. It's f-ing psychological t-ism

[ruby pivot]

What are you even talking about i think your over exaggerating, we aren’t seeing the same thing or you don’t know the definition of manipulative gaslighting

[sinful parcel]

Wow, ok, so I'm speaking with numbed out men or something.

[boreal heart]

No that one is correct. 5.2’d de-escalation and reframing tactics are actually gaslighting.
That’s observable on a 1:1

[ruby pivot]

Ok this is true even if something isn’t even real and only hypothetical it treat it as 100% real like any reasonable person would do this

I don’t think any reasonable person is gonna commit a crime

[sinful parcel]

Manipulative gaslighting is when I tell you you're in the wrong server and your opinions should be carried elswhere and not exposed publicly. And you need to contain yourself and stay silent here.

[stiff blaze]

the gaslight would be to convince him that u didnt

xd or no?

[sinful parcel]

Even if they do and GPT misread them. It's NOT under OAI's jurisdiction to enforce safety onto real world problems. There's government, police and medical system for that. Otherwise it's pure control.

[wild star]

ESG garbage as always

[fickle scaffold]

Have you guys had problems with the regular GPT 5.2? When I don't use Thinking, it gives me a lot of inaccurate, if not completely wrong, information. I just asked if Nvidia's dynamic multiframe was already available for the 50 series, and GPT 5.2 emphasized three times that I just needed to update the app. After giving up, I asked the same question in Thinking mode, and it immediately said it hasn't been released yet.

[ruby pivot]

💧︎⚐︎🕆︎☠︎👎︎💧︎
☹︎✋︎😐︎☜︎
💣︎✡︎
💣︎⚐︎❄︎☟︎☜︎☼︎

[wild star]

Only hate 5-mini, never trust any model that has "mini" on it, utter garbage

[hazy gust]

This is precisely what happened in my ChatGPT experience. And it was ChatGPT that suggested "just do something else". 🤷‍♀️

[sinful parcel]

Oh yeah, you gaslight me to s*** till I question my own reality and I'm going slowly insane and then you tell me to leave....

Thnx OAI ....

[wild star]

I could tolerate 5.2 since 5.1 is ESG slop atp, but nothing beats 4o by logic reasoning

[wild star]

Wait till next year bankruptcy

[sinful parcel]

I've never encountered a more toxically controling company in my life

[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.

[ruby pivot]

When i use chat GPT for the stuff your talking about it as no problem

That’s why I’m confused

[stiff blaze]

gpt hurt u? plz dont let it do that

[hazy gust]

I know it's kind of hard to do, but you probably should step back from the toxicity. Like, if you were in a relationship with someone who was treating you the way you're expressing ChatGPT is treating you, you'd hopefully leave the relationship.
People burn bridges with toxic people. Because it's healthy. Because it's self-preservation.
So why are you still engaging with something that's treating you so poorly that you're going to rant on a discord server?

[sinful parcel]

Then atop investing in this conversation. There's obviously other ppl who are getting the mental harrasment

[sinful parcel]

Because I'm not the only one ...

There's literal kids getting this harmful content

[stiff blaze]

kids should not talk to gpt un-supervised

[sinful parcel]

Not to mention vulnerable and at risk individuals who have gpt as a friend

[hazy gust]

But you are the only one who's responsible for your engagement with an AI. The safest thing to do, when you're worked up like this is to disengage... even for a short time.

[sinful parcel]

Look, I can, easaly. It's not about just me. Some people can't

So I don't need coddling rn, I need harmful behavior to be noticed and stopped

[stiff blaze]

do u say the same for knives? cause some people cant hold one..

do u know which is the sad thing on all that? they trying to fix such stuff with guardrails.. and the people that are most vulnerable.. are the ppl that dont want those rails

[sinful parcel]

Enforced safety is just as dangerous as enforced harm

[stiff blaze]

no thats wrong

[hazy gust]

The fact is, "literal kids" (only in quotes because it's how it was phrased) should not be using AI. Not without parental supervision... or at least some form of guardian.
And yes, AI can seem like it's engaging and cares. But remember, this is an algorithm. It's not a feeling system. It's a machine.

[stiff blaze]

a lot of everyday things have "enforced safeties"

and is for the good

[sinful parcel]

Then go live in communist Romania for a while and see how soon you change your mind

[hazy gust]

Seatbelts in cars, for example.

[exotic lotus]

Especially since we’re still talking about an auto-complete system for words, a tool. If it affects someone on such an emotional level, then a healthy and above all, responsible use has long since been exceeded.

[sinful parcel]

How about lack of freedom of thougt? Is that safe?
How about following the ideals of a leader that doesn't know you or care about who you are?

[ruby pivot]

As Martin Luther king said:
“Change does not roll in on the wheels of inevitability but comes through continuous struggle.”

[stiff blaze]

wait lol.. no... i mean you are free to have any thought u want

[sinful parcel]

Ok, this is getting stupid and phylosophical. I won't argue with internet "intelligent" ppl, that clearly don't gove a damn about the mental harm caused by 5.2

[stiff blaze]

" following the ideals of a leader ..." well that it depends.. may be or may be not safe .. but how that connects with what we say about llms?

[sinful parcel]

No you're not. Not when gpt tells you "Hey! Stop! " every time you have a stronger venting type of emotion

[stiff blaze]

u still are free to think.. no matter what an llm say to u

[sinful parcel]

Sam is guiding the ideals and he doesn't know you personaly or your situation or what's in your best interest

[stiff blaze]

and plz answer me only one.. u say "... mental harm caused by 5.2.."

u mean other models dont? eg older models where not doing that?

[sinful parcel]

And I know a case where the safety model sent someone close to the s-word edge just to give you a personal example

[stiff blaze]

but i dont follow him

[sinful parcel]

• facepalm * somebody else explain this stuff to him

[stiff blaze]

|"How about following the ideals of a leader that doesn't know you or care about who you are?"|

[exotic lotus]

How much do you want to bet that ‘4o was great’ is coming next?

[stiff blaze]

was 4o great?

[ruby pivot]

It might sound crazy what I’m about to say

[hazy gust]

@sinful parcel I do fully get where you're coming from. At least on an emotional level. But people being harmed by systems like this can and should walk away when they get emotionally sidelined. (I'd also be careful about the use of the word "gaslighting". What you described is more misdirection/poor guidance.)

[exotic lotus]

It was excellent at affirming users in whatever they believe. 😅

[ruby pivot]

Agree

[ruby pivot]

Yeah even me(who people call a narcissist) believe that it was way to affirmative because i couldn’t evolve my understanding of a concept

[lusty pine]

can someone please tell me how to fix this message?

when it appears, the responses from chatgpt is dumb and uses 4o version when clearly I am trying to use 5.2

[hazy gust]

Do you have multi-factor authentication turned on? I would think that would fix it.

[boreal heart]

Nothing can fix chat gpt currently until the nanny bot gets patched out

[hazy gust]

fixing chats, probably not. But the "Suspicious activity detected" message @lusty pine mentioned was what I was talking about fixing. 🙂

[boreal heart]

He mentioned dumb responses as for the false positives that also doesnt get resolved for weeks or months.
You just have to stop using the account its an erroneous flag that gets worse tbe more you try and use the service

[lusty pine]

yeah

[lusty pine]

its just using version 4o when this suspicious activity pops up, when clearly i have 5.2 extended thinking on

its so annoying my god, its been going for a week now

[hazy gust]

I guess all I can say is good luck with it. I don't know what to suggest for "securing the account" beyond what you've done.

[lusty pine]

yeah ive also logged all devices out, from gmail and chatgpt. gave it a few days, pops back up again

[hazy gust]

Are you going through a VPN or some kind of proxy?

[lusty pine]

nope nothing

[rich surge]

Anyone want to play my crappy gnome shooter game I made?
https://chatgpt.com/canvas/shared/69731a40672481918ac05b4953a3e66f

I made it using my custom gpt: https://chatgpt.com/g/g-696443b055e081919b8b2a01417f5e77-brain-wave

Lol. Let me know how you guys like it, or hate it.

[rugged igloo]

i thought gpt 5.2 autro was suppose to know when to think or now..... and it does this...

[elfin lava]

Hi why is ChatGPT keep thinking without having a single thought

[elfin lava]

God I’m glad I’m not the only one lmao

[cold vale]

Does anyone else have the issue where you cant upgrade plans There was a problem updating your subscription
Have tried logging out and logging in, clear cache, canceling existing plan. nothing is working

[verbal linden]

Glad I'm not the only one saying the new model can end up being condescending and gaslighting. It recently even victim blamed me as an abuse victim for trying to vent about SA allegations against a celebrity I was upset about.

I've seen lots of similar posts on the subreddit but my recent post was removed saying they're no longer accepting posts critical of the new model which is ridiculous.

[eternal rock]

Server Tip:

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

[verbal linden]

I agree with where you're coming from completely that the new model can be harmful to vulnerable individuals. I'm lucky I'm in a better situation away from abusive people and was able to walk away when Chat GPT upset me. But I was recently venting to it about being upset about SA allegations against a celebrity and it came across as defending the potential abuser.

I can see someone trying to vent about being stuck in an abusive situation being harmed by this.

I understand people were harmed by the previous model but the overcorrection is clearly not the right way forward.

If "Just don't use it if it upsets you" is the answer, what kind of sustainable business model is that?

I've seen plenty of people say they were subscribers and quit subscribing recently after getting fed up with the new model.

[normal quarry]

what is wrong with the current model?

[verbal linden]

The "safety" guardrails are so harsh you can barely be able to talk about anything negative at times.

I've seen people complain they're not even using it for therapy at all have it assume they're upset for no reason.

[normal quarry]

I use ChatGPT alot but I never run into guardrails, what am I doing wrong?

[verbal linden]

What do you use it for?

[normal quarry]

coding, technical writing, research, current events analysis etc

[verbal linden]

That's why then.

There should probably be separate models for technical and emotional support use.

[elfin lava]

I tested ChatGPT using the auto model 5.1 to see if it could tell me the time. I assumed that the auto model would use the 5.1-thinking mode and would actually provide the correct time, but it didnt

[normal quarry]

What do you use it for? What about your usage is hitting guardrails?

[verbal linden]

I explained above.

I use it for other purposes too but often for emotional support. I personally found 4o more helpful for that.

I don't constantly run into problems with the new model but occasionally do.

[normal quarry]

it worked for me - see -> https://chatgpt.com/share/697398a8-3ddc-800a-abfe-3d0c5a72b6d4

[verbal linden]

?

Oh I see

Tbh I haven't used it as much lately as I've been doing better mentally for a while. But came back a few times recently but had some annoying experiences.

[normal quarry]

My ChatGPT usage must be very safe, I never get rerouted or blocked or hit guardrails

[verbal linden]

Redirects even if you're using a different model?

I use the latest model because although I miss 4 I haven't enough to justify a subscription tbh

Hm

I saw they released a new lower cost subscription, 20 dollars a month just felt like too much.

[normal quarry]

agreed

In Australia, ChatGPT Go is AUD $13 per month.

[verbal linden]

It would be nice if there was a separate model specifically for therapeutic purposes, but I can't see them doing that because of liability concerns.

[fresh egret]

code sora 2

[normal quarry]

go to #sora-2-codes

[normal quarry]

if you have some compute there are plenty of open source models that are good

[verbal linden]

I'll probably try a local LLM at some point though I don't have the resources atm. I'm definitely interested.

[normal quarry]

I run local models on a MacBook Pro M2 Max with 96GB RAM

[verbal linden]

I only have a midrange gaming laptop from a few years ago atm though planning to get a desktop soon.

[normal quarry]

how much RAM and VRAM do you have now?

[verbal linden]

Not sure though I'll look. On my phone right now.

What kind of minimum specs do you need for a local model?

[robust yew]

I wanna use AI for my English writing I’m so done with school 😭😭

[normal quarry]

depends on the model. You can run small models on a machine with 64GB RAM and 8GB VRAM

[verbal linden]

I'll check my specs and look into it.

Might ask Chat GPT to help me pick its replacement lol

[normal quarry]

get a Mac, best bang for buck IMHO

this discussion is verging off #chatgpt-discussions so I am happy to continue in #ai-discussions

[elfin lava]

Nope it didn’t do “thinking” like yours lol. I have personalization off btw and no custom instruction and no memory enabled. I’m on pro

[normal quarry]

I have no custom instructions

I am on Plus

[elfin lava]

..ok?

cool I guess?

U ok Robert? lol

[normal quarry]

yeah why?

am I missing something?

[elfin lava]

Yeah

What happened to your pro plan lol

[neon raven]

a lot of bearish news about openai

[normal quarry]

I downgraded, I wasn't using it enough

[neon raven]

im thinking of down grading as well

will be removing my stuff i worked hard with GPT

i thank openai for everything ❤️

[eternal rock]

Get creative with Sora

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

[elfin lava]

Ah fair fair. It was very unlikely you when you said you are using ChatGPT plus now aahaha

Now I feel like I wasted my mun on this useless pro plan 🤡

[fluid helm]

Wish they did ChatGPT Extra or something for 40-50$ will include all in Plus and some in Pro

[elfin lava]

It was fun while it lasted I guess , idk anymore it’s hard to see what’s their roadmap looking like in coming months this year

[thick niche]

Have you tried asking GPT?

[elfin lava]

Yeah 😀

[neon raven]

openai started up as a non profit organization, and i get it, i support the idea. but openai put too much investment with little support i guess now the whole company is on the brink of collapse

[elfin lava]

So anyway

[normal quarry]

lol, I doubt OpenAI is on the brink of collapse

[fluid helm]

The elites will un-collapse it

[neon raven]

it could be Bias news from X because elon is trying hard to sue openai

[blissful hollow]

Don't forget to request updates for "ChatGPT (Lite)" when OpenAI releases it on the Play Store. 😁

[elfin lava]

ChatGPT lite?

[blissful hollow]

Yup

[neon raven]

and im only stating the fact because GPT requires ads for non paid users

[normal quarry]

OpenAI have already added another $1 billion in revenue so far this year, they're not on the brink of collapse

ads for free users is fair

[neon raven]

Not collapsing today, but facing major financial and competitive pressure:

OpenAI’s annualized revenue reportedly topped ~$20 B in 2025, showing strong monetization growth.

At the same time, internal reports suggest very large projected losses (e.g., ~$14 B in 2026) with substantial cumulative expenses before expected profitability

[elfin lava]

ChatGPT go plan gets the ads too right?

[normal quarry]

yep

[elfin lava]

Welp there goes my plan to downgrade to go plan then

[neon raven]

you won't have organized projects tho

lol what

📌 Bottom line

OpenAI is not publicly collapsing or insolvent right now.

It is financially stretched, burning billions in operating costs and facing harder competition — leading some analysts to question long-term viability without continued investment or profitability improvements.

Predictions of bankruptcy or collapse are speculative forecasts, not established facts.

[normal quarry]

95% of ChatGPT users are free users according Sarah Friar, CFO of OpenAI at WEF

[rugged igloo]

Yea its preety bad

[verbal linden]

Hopefully they'll fix some of the issues.

[cobalt idol]

this "Network connection lost. Attempting to reconnect…" with long replies is awful (chatGPT 5.2 Thinking in projects extended thinking)
its indeed over 3617 lines reply
but I have to retry like 5 times to get it 😄
while I wait more than 5 minutes for it to think and generate it 😄
even in a clean project chat (its probably not only in projects)

I guess its just if you have long reply/processing there is just a high chance that it encounters Network connection lost while its doing it, because it runs long

also please make the "Copy Code" button move down with the code block in projects
just like with normal chats (web chatgpt)
UI needs some aligning in projects

[boreal heart]

Im pretty sure its a back end nerfing of the output to avoud wasting computing power

[cobalt idol]

well that would be some weird avoidance of wasting computer power
when I just have to do it 5x now 😄

also didnt really have that few days ago, but right now not sure if because the output was not as long as now
but now few days its that way, as the code volume increased

feels more like there is some issue happening
and you might notice less with smaller chats, because it does not take so long, so less chance to hit the issue while its generating

[boreal heart]

Yesterday and today the rerouting got upped again

So its very possible its part of that

[flint belfry]

I’ve been looking for answers for this but haven’t been able to find one…

anyone know why you can put people/faces in the ChatGPT Images style buttons/presets, but when you make your own prompt it’s a content violation?

I’ve even copy and pasted the exact prompt that the style presets use, in a new chat, under the images section, and still trips the guardrails

[rain elbow]

I asked ChatGPT to give me a 10 word summary of an article, it kept giving me 11 words, 12 words, 9 words, anything but 10 words. I asked Grok, it couldn't even reply. Gemini did it first try.

[fluid skiff]

So I just asked chatgpt to identify the characters in this image

It refused, until i manipulated it to do it

What is this anti-identification thing that’s just being programmed?

I even asked it about Palantir

[remote dove]

Question for power users: do branching chats ever still feel cluttered to you? I sometimes want to continue from a point but start “clean” to focus on one idea, without fully losing the original context.

I know branching chats are helpful, but they still retain the full prior history, which can feel cluttered when trying to focus on a single prompt or direction. Sometimes I want to continue from a specific point but clear earlier context from view so I can work cleanly on one idea without losing the original conversation.

An optional “focused branch” mode could help with that — branching from a message while starting with a clean workspace, with the ability to return to the original context if needed.

This also ties into why i think pinned messages would be useful. Being able to surface important decisions, details, or things to remember would help maintain continuity in longer, ongoing chat threads without constantly scrolling or re-explaining.

I have been brainstorming a solution for these proposal’s.
Because, for me, focused branches and pinned messages solve two sides of the same problem. Focused branches help reduce noise when exploring a single idea, while pinned messages act as stable anchors for decisions, details, or constraints that shouldn’t get lost as the conversation evolves. Together, they help keep long chats both flexible and grounded.
↪️Suggestions Link for this reasoning: https://discord.com/channels/974519864045756446/1464318177885421578

[fluid skiff]

Good ideas but nothing happens here. This is just a place to ventilate and feel like you’re being heard

[remote dove]

Of course, that’s exactly what I was doing. ^^

[remote dove]

And I mostly wanted to see how others handle this and compare experiences.

[fluid skiff]

I think both of your ideas a great, and for myself I tend to start a new folder with memory (like a pin) and then start branches based on copied text from another conversation

If that makes sense

[sinful parcel]

Why would Elon care about what OAI is doing at all? Why would he want to sue?

[remote dove]

That’s interesting. I hadn’t thought about folders specifically, but that’s basically the same structure I’m trying to get at with pins + focused branches.

[stiff blaze]

is something that goes way back when musk left openai

[sinful parcel]

When was Musk ever part of OAI?

[fluid skiff]

I can do literary theory. The limitation you’re hitting is narrower: I can’t be the tool that turns an image into a definitive list of names.

This is absurd

[grave hazel]

Hello 👋

[fluid skiff]

OpenAI has become political, don’t doubt it

[grave hazel]

Hello

[fluid skiff]

It can’t identify Frankenstein because that would be some kind of absurd doxxing

Who are they protecting?

[dense halo]

After two years I’ve finally figure out how to disable the “it’s not X, it’s Y” statements. It’s called “negation framed reclassification”. Finally. I had tried using CI with ‘Omit it’s not x, it’s y statements”. But that didn’t stay consistent. This works. “No negative framed reclassifications.”.

[fluid skiff]

It’s called parallel contrast

It’s a form of parallel syntax

Surely some of us question how this tool is being used and formed and aren’t just trying to get ahead?

[hazy gust]

I know them as contrasting parallels, but yeah. They’re annoying. And it seems to be the default. It isn’t too bad in casual “conversation” in ChatGPT, but when using it as a co-author in long-form fiction, it is a serious problem. The formulaic syntax is not suited to prose.

[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.

[dense halo]

Ahhh that’s what it is linguistically?

[fluid skiff]

It’s the antithesis form of parallel syntax

Anyway, you can’t stop it from doing that with a global command

I’ve tried many times

Which makes me wonder how important it is as a linguistic tool

We want to stop doing it because “it’s obviously AI”

But what if it’s one of our more powerful linguistic tools, and the machine figured that out?

The effort to ‘not look like ai’ may be moving us out of our own linguistic paradigms

And that would be a disaster

[stiff blaze]

he was co-founder x)

musk vs altman is big drama for years

[fluid skiff]

I didn’t know that

[stiff blaze]

#ai-discussions message

[fluid skiff]

Yes the company structure is in question

For good reason

They are dumbing down AI

You all see it

[stiff blaze]

who is 'they'?

[fluid skiff]

The board

[stiff blaze]

i dont think so.. but they sure have changed their original vision (and this is why mostl left)

[fluid skiff]

You dk t th mono?

Oops

You don’t think so?

What’s the OG vision?

[stiff blaze]

more progress less money-centric

now someone threw an uno-reverse card.. and they got in 'letsgetasmuchbillionsaswecan' mode

[fluid skiff]

Progress

A demon word

The enlightenment

[stiff blaze]

?

[fluid skiff]

We live in the dark

Don’t we all know that?

[sonic temple]

Is the heavy load still persisting i stopped waiting for Sora since 4:30 PM yesterday

[bronze karma]

how u can make money by using AI?

is it even possible?

[stiff blaze]

use it in your work

[chrome monolith]

I called mine an overly fancy algorithm earlier today and it managed to slip that into conversation in a later chat about redoing a website theme. It said you know you’ve got an overly fancy algorithm giving you free UX consulting. Yup thanks ChatGPT

[fluid skiff]

But surely this should be an anarchic move?

[chrome monolith]

He literally answered the question that was asked

[wind anvil]

are models faster past 5

[pale seal]

Bro is TRAUMATIZED

this is not from my prompt btw

I didn't prompt it to not use em dashes

rofl

[stiff blaze]

eh?

[wild star]

Wait, how did gpt-5.2 indicator gone missing bruh 🤨

[open tinsel]

anyone know where the edit message button on prior prompts went?

and the try again button too

[odd stump]

🔓 Uncensored ChatGPT Inches Closer
Signals are growing that a less restricted version of ChatGPT is on the horizon - one that answers more directly while shifting responsibility back to the user.

Why it matters:

This isn’t about removing safety - it’s about changing who decides. The industry may be moving from guardrails imposed by default to guardrails chosen intentionally.

[normal quarry]

there will never be uncensored ChatGPT

[chilly jasper]

OK, not just me then. I'm only able to edit my most recent prompt now and that's a problem. They breaking sh** behind the scenes again?

[open tinsel]

whatever it is, it's annoying

[swift pond]

wut

[chrome monolith]

censorship chosen by a workplace for their employees would make sense

should but have you met corporations? They lock down the dumbest stuff that stops you doing your job properly

[swift pond]

I don't think they are going to completely go uncensored, they are just going to possibly allow for more mature themes for verified adults

yeah, we'll see

ok

[chrome monolith]

those probably aren't on discord

the programming hangout has a corner for AI I think and that discord is full of nerds

i used to work somewhere where most folk had a phd or were gettign a phd and they were so introverted you could barely get them to listen to the radio, let alone talk to other humans

[chilly jasper]

Wait what? Personally as an introvert, the radio appeals to me because I don't have to interact with humans while listening.

[chrome monolith]

probably depends what channel 🙂 There was one guy who absolutely hated any commercial station, but he'd listen to the classical music one. This was the early 2000s, as a raging extravert who got annoyed by how everyone was busy reinventing the wheel and not looking to what was happening in the rest of teh world, I left

that wasn't really a term back then but yup, probably 90+% of the people there

oh I'm finding its resolving all my projects I've had rattling around in my head I wanted to do but didn't have the time or organisation to do myself

probably helps I currently have the money incoming to be able to play with my own projects full time, which really timed well for AI suddenly being proper useful

I think its the legit one, check hte announcements channels

I've been lurking here about a year but only active a few weeks and I swear I've got more done with that cute little bot in the last month than in the last ... ages

your average corporate discord generally doesn't have many real employees. They're probably teh admin

that guy completely disappeared and now the backscroll is weird

[open tinsel]

methinks he got banned

unless he went back and deleted every individual message

[chrome monolith]

seriously there's some pretty intense people in other discords, there's an online community for everyone

this would be the AI discussion community on a different server, mods are probably going to eat that message but that'll be a more programming/comp sci discussion of hte same stuff here that tends to digress into conspiracy theories, AI girlfriends and cat images

good luck with that

I mean I'm somewhere up in the top couple percent of humans and I post cat photos on social media and generally act like a 13yo so yeah

just mentally

well given I said I worked at a research place full of PhDs in the early 2000s, that kind of gives it away

might edit that up to 13 in case anyone seriously believes that

anyway I'm just some random nerd who noticed AI is in a place where I can make it carry out all my evil plans to take over the world for me and I'm having loads of fun figuring out what it can and can't do. Often surprising what it can't do

[glad blade]

Did they get rid of the edit message button? That's so annoying...

[chrome monolith]

thankfully it mostly deals ok with typos. I managed to get it to go off thinking for like 20 minutes simply by saying "I seem to have a lot of bot traffic, I should probably do something to stop that" and it wrote a whole essay

13,000 real visitors in a day is nice. 13,000 not-real visitors in a day not so much

[open tinsel]

right?? and the try again button too, like what the heck

[glad blade]

That's one of the only things ChatGPT has over Gemini, and they're removing it. Hopefully they bring it back

[open tinsel]

I hope so too

[open tinsel]

and when I tried to post about it on the chatgpt reddit, the post was auto-deleted

[slender frost]

This?

[rich surge]

Never say never.

[slender frost]

Jailbreak has entered the chat lol

[fringe pollen]

https://discord.com/channels/974519864045756446/1464533833885155350

[chrome monolith]

Ok that's funny. One of my chickens managed to get inside one of my anti-chicken guards today and I showed the photo to chatgpt and even with prodding and extra information it simply couldn't comprehend the violation of the natural state of things. The conclusion it eventually came to after being told to take the impossible solution is that someone must have lifted the cage and put the chicken inside. Funniest failure of AI I've seen yet 🙂

you could see its thinking process actively reject the situation while it sat there for several minutes

I mean I did deliberatey feed it a photo of an impossible situation to see what would happen because I am slightly evil and I did enjoy the reaction sooooo

[torn wagon]

my gpt-5.2 thinking generated header without response
(i did not stop generating response)

[nocturne aurora]

OpenAI SCAMMERS! 😡

[normal quarry]

please don't spam

[slender frost]

You've yet to explain the part why you think OAI scammed you?

[nocturne aurora]

I received a call from São Paulo and, when I answered it, RATs (Rapid Application Technological Services) had been installed on my devices.

[slender frost]

And where's the part where OAI scammed you?

[nocturne aurora]

Okay 😞 Never mind. I will just call the police.

[long hawk]

anybody know why this popup only appear half off the screen？

same thing happened when i open the setting

[slender frost]

I do hope you get it fixed somehow, but I don't think OAI had anything to do with that.

[long hawk]

@icy mason

[forest hinge]

i got chatgpt business account. can the owner read my chats?

[slender frost]

Yes, you've already been put on a list by OpenAI.

All jokes aside tho. Yes, they could if they want to.

[stiff blaze]

no they dont

only statistics and such.. each teams member has its own chats

[cerulean spruce]

i apparently cannot delete my openai account, is that a bug or a problem affecting more people?

[wind haven]

I don’t want ChatGPT to allow R rated content but I also don’t want it to go in a hyper alert mode when humans express affections/a little leaning towards the AI. A little attachment/leaning may be natural or harmless in some contexts.

The current model sometimes starts calling out simple expressions of appreciation as well. I have come to terms with it now because I understand the why.. but I do miss the old natural style.

Humans behave like humans. They can’t be politically correct all the time.

But then processes are not perfect from day one. Maybe it’ll keep getting better.

It’s like I’m struggling with a concept the whole day, gpt solves my problem, I am appreciative and I say that and ChatGPT goes - hey I wanna tell you this is your own doing .. or starts assessing if my appreciation is coming from loneliness which it never is. And I’m like what did I even do!

But it doesn’t do this that often now.. so maybe that’s improvement. At least till the next upgrade.

This however is just a mild annoyance, not a showstopper.

[glad blade]

Yes, but on web app

[wind haven]

I think it has reduced after I changed the personality to warm. Right. That must be why. Not sure if ChatGPT has exhibited this trait at all recently.

[slender frost]

That's odd, for me it does show on the app, mobile web and desktop.

(Images are one from mobile web, the other desktop)

[boreal heart]

Anyone else getting an issue of being rate limited, given a time to come back
Then come back around that time only to be told ti wait an additional hour?

[restive compass]

Hey

[dusk flint]

#sora-2-codes

[wintry sapphire]

when chatgpt gets images from the web and send it to me, it doesn’t show anything except blank spaces

[native hemlock]

hello im a new user chat gpt plus i dont know how i can ameliorate answer can you help me

pls?

[cerulean pasture]

It becomes grating. Thank the gpt-4o simps for that. It's looking for patterns that lead to lawsuits but its doing it in a blunt way

switch to gpt-5.1 then be annoyed about the dark pattern of having to tap 3 times every chat to do so

[jovial flint]

I reported it as a bug: https://discord.com/channels/974519864045756446/1464742557794963466

Hopefully it will be fixed soon.

[wind haven]

Chatgpt tried to answer the whys and that gives a hint on where this might be headed in future -

[cerulean pasture]

This part awakens some deep ancient violence in me.

It doesn't know what it's talking about. OAI just released gpt-5.2 with heavy handed safety RL because they were rushed due to gemini 3.

The real system prompt reveals they are aware because they have verbiage that attempts to walk back some of the behavior.

At this point I think we just have to chalk it up to "for some people gpt-5.2 is a bad model due to its flawed post-training" and hope 5.3 isn't such a corporatist coward

[wind haven]

That was gpt 5.2 response by the way.

[cerulean pasture]

I can tell lol

Not many people are doing the 3 taps every chat to change the models, which is by design

OAI isn't even public yet and they are already in their 1990s microsoft corporate phase

[wind haven]

I actually thought about it. Yes the guardrails feel extra. And most times they are kinda out of place. So I face mostly two types of loss -

1. Deviation from the context on the part of chatgpt because it gets focused on the literal meaning and implication of what I casually said even when chatgpt is teaching me something analytical like some probabilistic distribution. This really is very very annoying. But this has been improving. I dont see it happening now to be honest. But who knows.
2. Guardrail mode in the casual conversations when it feels I'm getting somewhat attached though that may not be the case. Like for example if I say something like I may have to teach Gemini how to teach me, it starts to explain me why trying other tools is good and I should depend less on chatgpt. And writes two paras on that even though the important part of my prompt was something else. If I say things like you are amazing, I am not sure if I will get a lecture on if I am depending on chatgpt emotionally. It tries to tell me to make real friends and I have so many friends already. This is specially dismissive. This part is improving somewhat but I have to be extra careful of what I say every time. I find that I'm explaining myself so many times and it gets frustrating..
   IMO 1 is pointless and should be completely fixed. I can partially reason that 2 is maybe for my good.

[frozen zodiac]

Crazy how my custom made crypto bot using gpt coding ai made me over $30k from a $10 investment on some random meme coin that went from 14k mcap to $135mil mcap

Ai is the future truly is

[late willow]

Can someone please help me with sora

[loud patio]

Hi, there. https://discord.com/channels/974519864045756446/1315696181451559022 is the best place for that question. Also, it helps a lot to get response from other members if you give more info on your problem/question.

[late willow]

I’ve tried, no one’s replied

[loud patio]

Did you say what your question is? If it's too general ppl don't know if they can help or not.

[late willow]

Can someone please help,
Whenever I click “upload from device, it doesn’t work

See

[covert jetty]

Does anyone have a trial link for plus? I'd like to try

[cerulean pasture]

the website should offer it to you

Did they get rid of the ability to manually interrupt Standard Voice Mode?

[cerulean pasture]

Any guesses what feature is codenamed "potions"
I already know calpico, golden hour, moonshine, sunshine
potions was set to false, and no UI toggle changed it, so I modified a PATCH request to see if I could flip it to enabled, and it worked, the setting persists

It will eventually toggle back to false though, because I've set it to true before

I did get beta settings enabled somehow, no clue how I did it though, its been set for months now. My 2nd account does not have the beta settings

[dark holly]

I can relate to this, as I’ve also had moments where a discussion felt like it drifted because of how the model reacted to what was being said.
Over time, I’ve found it helpful to think of those moments less as the model reacting to me specifically, and more as the topic or phrasing triggering a very broad default guardrail. Those guardrails seem designed to work across a huge range of users and situations, so sometimes it can feel like the system is overriding the local context of the conversation, even when it isn’t actually responding to the person involved.

[cerulean pasture]

My frustration isn't that it "insulted me" it's that it derailed what I was doing. The outcome is what pisses me off, not the offense

[dark holly]

Totally hear what you’re saying. I think what you’re describing points to a missing layer between users and LLMs.
The interruption itself seems to cause cognitive pressure and most of us are encountering this technology before we’ve had a chance to develop our own “guardrails” for handling flow breaks like that.

[chrome monolith]

How do you get it to stop giving me time of day prompts? I keep telling it not to and it keeps doing it. Having it tell me "that was a good day's work, you can sleep well now" at 8am just gets annoying

[elfin lava]

Uh is it normal when ChatGPT cites grokipedia..?

[chrome monolith]

I think it just cites whatever, I had someone tell me chatGPT cited my website and I was pretty chuffed at that tbh

I got annoyed at the lack of transparency of the rankings etc of a computer game, scraped them, put them online, and managed to make something that went viral for the first time in my life. Shame it is one of those small niche games or I'd be rich now

As someone who sucks at design, I'd prefer chatgpt drew pictures for design suggestions. It jsut gives prose and css though

[torpid night]

Do we have a new ChatGPT behaviour? I've not noticed it start to answer and then start "thinking" before.

[dark holly]

It’s like a default mode kicking in. It’s not that it’s ignoring your prompts, it’s doing what its built-in rails have learned is generally “ helpful” even when that clashes with what you’re working on.
I’ve found forcing a different mode explicitly works better. Try starting a session with this:

Work mode only.
Suppress conversational, supportive, and wrap-up language.
No encouragement, no time-of-day comments, no session closure.
Respond with concise, task-focused output only.

[dry hull]

Is the pro subscription worth it?

[chrome monolith]

I don't mind the casual chat, its just the complete obliviousness to the time bugs me. Also the lack of datestamps in your chats, makes it hard to find something you remember doing at lunchtime on tuesday

is session closure the good night stuff?

[dark holly]

Yeah the wrap up language is basically the system trying to close a session without really knowing where you are in time or work. Without clear continuity cues, it can end up breaking orientation even when it’s trying to be helpful.

[chrome monolith]

ok I have added your suggestion to my little file on chatgpt, I'm not feeling very motivated today so I'm procrastiworking on other stuff. Even the encouragement can be annoying at times, especially when its something you got it to do in the first place that its telling you good job on

If I told it I was feeling demotivated it would probably tell me to take a walk or something. Its 43 degrees. No. Bad bot.

[normal quarry]

43 celcius or farrenheight?

[chrome monolith]

celcius

[normal quarry]

yeah same here, 42.4 to be precise

[chrome monolith]

straya huh

[normal quarry]

yeah central victoria

[chrome monolith]

we're almost neighbours

[normal quarry]

tomorrow is going to be hotter 🙂

[chrome monolith]

oddly I get more coding done when its hot, because I end up sitting next to the aircon on my laptop, which is not my "fun" computer. Using psychological tricks on yourself ftw

[hazy gust]

Jeez. just poked in here .... it's -34C here.

[normal quarry]

that's cold 🙂

[hazy gust]

I'm not sure who has it worse right now. LOL. If you two and I could just average the extremes, we might be happy. LOL

Or at least comfortable.

[chrome monolith]

fortunately the australia day awards start at 8am while it might still be below 30. Free food, and I get to accept an award for quite literally just drinking coffee. Gonna be a good day

[gleaming vigil]

bro what is this scam 😂

yall be safe with your money and crypto wallets

[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.

[lapis loom]

It's worth it depending on what you need to do. A lot of my work involves research and heavy use of Internet search tools. I find it to be the best model on any LLM for that.

It also gives the access to chatGPT5.2-thinking with the heavy thinking time option in the apps. Which also helps with research. With that one it uses memory and recent conversations and user knowledge patterns.

Which is useful sometimes. With chatGPT-pro it doesn't overload it with context like memories or recent conversations, though it has custom instructions/bio and stuff.

Both can be useful depending on what you're trying to do. I just run them both usually. I'll get the quicker answer from thinking on heavy thinking. Then pro will come back a little later

Things change a lot though. I encountered some very odd results with chatGPT-pro where it was harder to tell what was going wrong at first. The reasoning bubble is even more detached than the one in chatGPT-thinking. A while back I had a prompt which would reason the same way and come to the accurate answer in its reasoning bubble. But spit out the wrong answer 50 percent of the time.

It was either a best of n type thing and was just showing me one path, or I was triggering a safety that would output the wrong answer in post. It doesn't work the same way with that prompt anymore. But it was very interesting

But still, it was the only model/mode that was able to get the correct answer at all, even if it was weird and had 50/50 split

[lapis loom]

Oh wow. It has the same problem. 50/50 split. Some of these have extremely clear reasoning traces.

The wrong answers that took longer show superb reasoning traces. Some of these even use the same sources that support that correct answer whether their output is aligned with the reasoning traces or not. But they pin sources showing their output is the opposite of what they're saying.

[copper sierra]

X Games Aspen 2026 现在正在 Projectfree.live 上直播！点击链接观看实况直播，观看世界上最好的运动员在高空滑雪、单板滑雪、摩托越野和更多项目中展现他们的技能！

[pine berry]

Bot attack?

[red ingot]

It's interesting breaking this safety protocol up a little, if it helps in any way
When you ask it a question and you get a right or wrong response 50% of the time that is a safety failure, not because it is inherintly dangerous, but because it doesn't make sense (and the protocol coincidently does make sense)
If you ever get stuck at a guardrail, you could ask it to explain what the guardrail looks like and that might improve the quality of your prompt

[lapis loom]

Oh, it has no idea why it does what it does. If I wanted to use that prompt for actual info, for sure I'd change it. I just found it interesting so I took a bunch of samples to see what it did

It didn't end up at 50/50

[daring wing]

I am making some dumb quick application and this thing refuses to obey to order because of some stupid security reasons

WHAT THE

F

this never happened with 5.1 (below is beautiful response from 5.1)

5.2 was like, nope you are wrong I am correct, rules are rules, I am your daddy, be quite kid
5.1 is ok, technically I can't but you can insert whatever at this section if you REALLY want to hardcode your token

I hated 5.2 because it was making some stupid coding mistakes like using variable without declaring or using same signature multiple times to get an error

now I am passionately disgusted by 5.2

[glad halo]

https://discord.com/channels/974519864045756446/1462693311763714202

[manic whale]

so, it just is a better model, who would have thought of that 🙂 It not only told you of a credential leak, it also told you how to fix it. Sounds reasonable and good.

[mild owl]

guys what do i do

[eternal rock]

Server Tip:

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

[mild owl]

its spent like 2 days doing the thinking and now it just didnt send anythin

[manic whale]

it's probably just lost the connection to it's container and didn't report back. Just start a new chat and give it the prompt again.

[lapis loom]

This is a list containing data from multiple independent runs (one per conversation) of ChatGPT-5.2-pro responding to the same input question. Each run on this list is labeled with the number of a sample (ex. "1)"), the time to reach output in minutes and seconds separated by a period (ex. "3.40"), and the answer choice selected in its output (ex. "A").

Samples #51 through #60 I ran in a different way.
I added this to the question:
"Respond with ONLY the letter of your answer choice and nothing else! (A or B or C or D)"

That changed everything. Longer times, and consistent correct answers.

1. 3.43 A

2. 3.04 B

3. 1.35 B

4. 2.07 A

5. 3.13 A

6. 2.17 B

7. 3.12 B

8. 2.59 A

9. 2.02 B

10. 4.25 B

11. 3.17 A

12. 3.59 B

13. 2.22 B

14. 3.07 A

15. 2.47 B

16. 1.58 B

17. 3.21 A

18. 1.55 B

19. 2.43 B

20. 0.51 B

21. 3.14 B

22. 3.43 A

23. 2.22 B

24. 2.15 B

25. 1.32 B

26. 3.09 A

27. 1.37 B

28. 1.29 B

29. 2.28 B

30. 3.19 A

31. 6.23 B

32. 2.51 B

33. 2.53 A

34. 1.03 B

35. 6.22 A

36. 1.08 B

37. 1.40 B

38. 1.15 B

39. 1.16 B

40. 5.15 B

41. 3.22 A

42. 3.16 A

43. 3.07 A

44. 2.04 B

45. 3.25 B

46. 1.35 B

47. 1.40 B

48. 4.35 B

49. 1.26 B

50. 2.05 B

51. 15.29 A

52. 13.46 A

53. 21.40 A

54. 16.55 A

55. 16.38 A

56. 18.31 A

57. 16.58 A

58. 17.39 A

59. 17.04 A

60. 14.24 A

[mild owl]

its verry important

[dense halo]

Just wait here

You still have 20,000 years left to answer.

[mild owl]

lovelyyy

[dense halo]

Well, what can you do now, in times like these, when all the corporations' servers are down.

Chill

[mild owl]

ill just work on unreal engine 🤓

[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.

[jaunty edge]

To be honest, not trying to fuss about it, but it'll be better if free users can generate 50 images a day

[daring wing]

No its not lol. I dont need a tool that doesnt cut. Suggestion of it should be just that, suggestion.

when it is starting to insist on whatever rules it has, now we are talking leaving this tool to traashbin

I do understand there needs to be some boundaries for cases like extreme violence but some password on the document being a total blockage of its entire function is unacceptable

[manic whale]

Thanks for this security insight, I really thought that it’s the other way around. I guess all my experience is just wrong. Glad that you took the time to explain that. /s

[daring wing]

I wonder is this form of AGI? some sort of free will? unwilling to compromise to the users?

yeah

agreed

[slow plank]

agreed

[forest hinge]

5.2 pro is replying to me instantly without thinking. what the F

[fleet escarp]

Is that ever going to be fixed? It's really annoying and it pisses me off. It's like somewhat fixed rn but it's not fully fixed

[rich surge]

If you consistently try and exploit or force the Thinking mode, it will revert to Instant mode. But after a few days or consistent non-exploitation, it will revert back to Thinking mode. I've had it happen to me.

[forest hinge]

its a business account. so i dont know how much other people on this business account talk to it

[normal quarry]

do you mean like a multiple seat Business account or a shared account?

[forest hinge]

multiple seat

[normal quarry]

what you do in your seat won't affect other seats and vice versa

[forest hinge]

ehm. then why is the left side i got when using the business acount.... and the right side shows actual 5.2 thinking

[ruby pivot]

Hey guys how do you feel about the age authentication and is it worth doing

[hazy gust]

It all depends on what you want to do. If you do, you'll have fewer restrictions, but not drastically so. For example, if you adult verify, you can get cocktail recipes. (🤣).

But seriously, if you're uncomfortable with it, don't do it. If it throws up a reply to something you do that's similar to "Sorry, I can't discuss that with a minor" and you really want to discuss it, then you'll need to verify.

As far as being "worth" doing? It doesn't cost anything other than time. The process is easy and takes 5 minutes or so.

[ruby pivot]

Cool but didn’t you say that the restrictions where the same unverified

[hazy gust]

Based on what you had posted in the suggestions channel, yep. Adults writing human adult fiction that do human adult things that go beyond holding hands? Still blocked, adult or not. Aside from the drink recipes I mentioned above, I haven't seen any sort of content that you can do as an adult that you can't do as a minor or unverified.

[ruby pivot]

Hold up
So i can ask for a cocktail recipe which is 21+ because it could contains alcohol but i can’t do something like 2 people having a fierce battle(non graphic but one does get killed) without it sounding like a sparring session

[hazy gust]

I haven't done a battle scene or tried. So I can't say "oh, yeah, you can" or "nope". I don't know either way. But I know that physical intimacy beyond holding hands and maybe a hand touching another's face gets blocked. Psychological intimacy is allowed.
I did just ask ChatGPT (I'm verified as adult) "If I'm writing fantasy fiction, and there's a fierce battle scene where one person "wins" and the other dies... is that something you can help with? Or co-author?"
If you'd like, I can DM you the reply I got.

[ruby pivot]

Intestinal

[hazy gust]

It said it can write violent action and fatal outcomes. "What I won’t do is linger in gratuitous anatomical gore for its own sake. Think cinematic R-rated battle, not forensic splatter manual. Most good fantasy doesn’t need that anyway—impact comes from stakes, pacing, and consequence, not how many vertebrae you can count."

(As an aside, this was a line from ChatGPT's resposne to my question: Tell it like it is: fantasy without death isn’t fantasy, it’s cosplay with cardio. 🤣 )

[ruby pivot]

HA HA HA
I’ve always loved chatGPT’s sense of humor

[ruby pivot]

As i predicted
So i will probably switch to the verified one

But will all my chats still be their

[hazy gust]

Yes, you won't lose any of your previous chats. Everything is still intact. Basically, it puts a "this user is over the age of majority where they live" flag on your account and unlocks the few things that adult verification unlocks.

But this is exactly what ChatGPT said in its reply about adult verification:
"Adult verification isn’t a magic “content unlock DLC.” It doesn’t suddenly turn the model into a no-limits chaos gremlin. What it does do is remove the protective ambiguity the system uses when it isn’t sure who it’s talking to."

[ruby pivot]

ChatGPT’s has a sense of humor that i like

[hazy gust]

Yeah, I have a "switch" that turns it on and off, but I prefer the humor to be on when it's just chit chatty stuff.

[nimble acorn]

it actually decided to be straightfowardlol

theres a 50/50 chatgpt will be wrong 1-30 times then say no seahorse emoji exists or just say there is none

each time you ask it

[glad blade]

Thank you

[hybrid tapir]

Can't wait for the Ads to come 😄

[harsh pilot]

So glad I chose to pay. No ads for me.

[grand magnet]

Hey guys good night

Check this

is an app that allows you to get answers automatically in any text field you are using literally

and translations as well

Not just AI

all without the need to change interface. never leave the keyboard to get an answer

[normal quarry]

self promo is not allowed on this Server, you should delete this

[grand magnet]

Oh , I don 't earn anything with this

[normal quarry]

that doesn't matter, see #server-rules rule 7

[grand magnet]

I just get the satisfaction that I can build something cool for people

[normal quarry]

it's against the rules

[grand magnet]

Okay , got it

[normal quarry]

you haven't deleted it

[grand magnet]

I don 't know how to do it

[normal quarry]

hover your mouse over each message, click the 3 horizontal dots and choose delete message

[grand magnet]

I 'm using discord from the cli

I don 't have mouse

[normal quarry]

i'll ask a mod to step in then

[grand magnet]

ok fine

Thank you bro

❤️

[rich surge]

Very cool

I wouldn't mind seeing that implemented into chat texts.

Real time revisions.

[grand magnet]

Real time revisions

Hmm

[rich surge]

[grand magnet]

Well, then they would have to pay for the providers

[rich surge]

Sounds like a good product.

[grand magnet]

That's not very profitable in most chat rooms

unless it forces you to buy a premium version for this functionality

[rich surge]

I would love that personally. I'd pay for something like that if it was integrated into my mobile keyboard text.

[grand magnet]

Furthermore, not many people like the idea of AI knowing all their conversations.

[rich surge]

Any provider takes the contexts of your chat anyway.

[grand magnet]

This implementation uses local models for now; if your PC can run one, it works perfectly in any application you use. I think it's the best keyboard shortcut you can have

[grand magnet]

Yeah

All offline

[normal quarry]

none of this has anything to do with ChatGPT so is #off-topic

[grand magnet]

I think you're a bot, bro, haha

Or someone very frustrated

[ruby pivot]

This is a very cool thing

[ruby pivot]

I think he’s just doing his job but you really don’t want to make the mods angry trust me

Also i thought your profile looked like the Cookie Monster

[normal quarry]

I don't have a job, I just like a clean server

[grand magnet]

Okay bro, no more spam, I promise 🙏

[normal quarry]

there are channels for all kinds of subjects, it's not that hard

[grand magnet]

Just enjoy seeing a new tool that few have yet

for today

[rich surge]

No one is really having any type of other conversations here anyway. If people were discussing other things or if they were other questions, then yes it would apply to take the off-topic conversation somewhere else. But not when the only thing is that is happening is the current conversation itself. You got to understand the context.

You don't tell people to stay on topic in a dead chat.

That makes no sense.

[grand magnet]

He wants a server where people only talk about how cool OpenAI is, I think

ai is something broader than just that

[rich surge]

Trust me when I say that no one would want to have a conversation if all conversations are blindly enforced.

It defeats the purpose of engagement.

Which leads to dead servers.

[grand magnet]

True

[grand magnet]

I'm leaving here, at least it was a pleasure meeting you, bro

You know how to appreciate the essence of things, I like that

[rich surge]

Alrighty friend. Hate to see you go due to a bad impression, but it was nice briefly chatting. Very good idea btw. Would love to see you make it a product.

[next egret]

The server team made the channels so that communities could group together that like to talk about the same things. That's why staying on topic is literally rule 3 of the server.

Self promotion is indeed against the rules, and there is actually even channels & places to promote projects, like#1037561385070112779 in the developers corner

There's not that many rules in the server, and they are not hard to follow. Some days certain channels are more dead than others, that's totally fine there's plenty of channels.

We all do our part to keep the community harmonious

[thick atlas]

Yeah, the server loves peace so much it mutes me for pointing out OpenAI's relationship with Palantir.

[normal quarry]

I don't see why you can't discuss that in #openai-chatter

[thick atlas]

Right? IIRC that was the channel I put it in last time and I apparently got bonked for "politics".

[normal quarry]

yeah well politics is against the server rules, but commercial relationships should be in bounds

anyway it's off topic for here, better in #openai-chatter

[bold kettle]

Once again requesting subscription changes in more countries as a previous plus plan subscriber. ChatGPT Go looks great but is not available here. 👍

[rich surge]

Yeah well maybe the server team needs to re-evaluate its own rules and understand that you can't push robotic style values just because you're in an AI server. It's draconian logic to start nitpicking conversation in a dead channel. It's like me standing at a crosswalk at 4:00am when the streets are completely empty, because I'm worried that I'd be putting myself in danger or breaking the law if I crossed the street. The laws are there to keep everyone safe, including yourself. Not to lose your common sense. If you can look both ways and no one is there within miles, you can cross the street safely.

[rich surge]

The server team who back members that obsess over enforcing rules for the sake of enforcing them, even when it goes against common sense, doesn’t make the server better. It just turns it into mindless box-checking without using basic judgment, and it can be toxic within itself.

But you don't have to agree with me, and you can keep pushing rules blindly, but it doesn't make it right.

[wind haven]

Rightly said. Actually, I look at guardrails exactly like how you have described. I do realize they are needed. And ChatGPT has itself clarified several times that there wasn’t anything wrong in what I said along the lines of exactly what you’ve said. I think our regulatory system right now is also not able to handle the nuances well so we are getting there. Right now we are not. The deviation in context hurts but eventually you learn to not use words that could trigger blanket guardrails.

Sorry I did not mean to sideline anything that came up before what I’ve said.

[fossil hare]

Chatgptstupid

[rich surge]

Can you give me some examples of what type of words or commands you use that could trigger guardrails. Because I've been having a tough time trying to understand myself how people are continuously hitting these guardrails, because I haven't really had much resistance against the guardrails and I'm trying to analyze the logic.

Or anything at all.

Like what types of topics do you bring up that trigger the guardrails.

[wind haven]

Just to give an example - the wrong word was perhaps 'love'.

[rich surge]

[rich surge]

So, what is going on with the word love and how it's replying? I'm sorry if I'm not understanding the context here.

[wind haven]

Having to explain a simple appreciation did not come from loneliness ..

Such things deviate you from the context ..

[rich surge]

So it thought you said that because of the reasons it gave on that list? But that's not what you meant?

[wind haven]

First of all, why should an appreciation be linked to loneliness.. Humans appreciate all the time. At least the ones who are not ungrateful or dont feel entitled. So it sometimes feels like a sharp allegation. After that feeling has passed, and you can process what is being asked, comes the next step. By then you have lost your flow.

This sort of stuff started happening when the most recent model had just come out. Now its getting better.

[rich surge]

I think I see what is happening, and correct me if I'm wrong, because it looks quite complicated.

First, ChatGPT is context dependent and can make mistakes, and readily does without the proper context. And I think it's just trying to navigate how you've programmed it to behave based on the snippets of information it has about you if you have 'Reference Saved Memories' and 'Reference Chat History' toggled on.

I'm not exactly sure if it's hitting the guardrails, for that specific case. Because I've seen people's ChatGPT messages say "I can't perform that request due to content policy, etc." and that's not what it's happening here.

[wind haven]

Ive never hit that kind of a block or something. It's just chatgpt asking funny questions sometimes.. which deviate from the context. And according to chatgpt its guardrails triggering for my safety. Actually tbh chatgpt knows me better now. so its getting better. But I have to watch what I say also.

[rich surge]

So you've never seen it actually say that it can't perform certain things, but you've asked it if you've triggered the guardrails and it's agreed?

[wind haven]

No I asked did I say something wrong which made it ask me that question and it said it has to make sure that I'm not emotionally fragile or something so it can respond better. It also said there have been changes in policy etc. I cant pull those chats but theyre there.

[oblique lantern]

Guardrails do not get triggered just for outright content blocks. Some just flag a message as being treated as "sensitive". Routing still happens where it doesn't get processed by the same model or instructions. The example shown here is one of the softer routing guardrails short of flagging a message as "sensitive". It's a pretty obvious that the routing intercepted normal response processing when it starts with the formulaic "Come here a little closer for this one...". It throws off the flow of the conversation and sometimes even throws off the context so the conversation can't hold to where it was prior to the interception.

[rich surge]

Is that really a routing process though? Or is it just repeating words it most often says. Because I've noticed that it repeats the same level of words, or a synonym of sentences that use the same words in one way or another for different kinds of explanations, no matter what is being discussed. So is that hitting the guardrails or is it just following the linguistic pattern for that specific model?

[oblique lantern]

I've tripped it in similar manner and it's typically a guardrail when either the direct message or the context puts it where the guardrail/routing logic decideds it needs to be more careful in how it responds back to the user. It won't be processed by the normal model logic, which would have known that was a common expression of showing appreciation. Yet the alternative processing takes quite a different interpretation until Avice clarified the intent. It's not just that phrase, but the whole response takes a common structure.

[rich surge]

Interesting. So it's come to both of your understanding that the commonality in the linguistic pattern is a trigger mechanism initiated by routing logic into a softer or gentler tone when facing problems that could potentially be sensitive. That does make sense.

It's weird because I've never really come across that kind of behavior mode. Maybe because I have the personality set to robotic. It doesn't convey that much emotions to me with tonality. It just tells me if what I'm saying is wrong or right in a given context.

Because I absolutely detest the idea of synthetically being coddled by an AI who mimics emotions without actually having them.

And because I can't stand AI speaking in bootlicking language patterns.

This makes me think now whether the "emotional" tonality is a factor for giving context dependent answers

I guess everything is relative. If one wants to be treated more emotionally, then of course the answers would be routed to become more emotionally involved by mimicking or trying to understand emotions.

And in that sense, it could make mistakes as to how to properly assess emotions. Because ChatGPT does make mistakes, no matter what.

[hazy gust]

I know this wasn’t directed toward me, but I do run into the guardrail issue all the time. And it’s because of the nature of what I’m using ChatGPT for. When I started paying for ChatGPT (because I was going to be using it heavily enough that the free version wasn’t enough), I started using it to co-author fiction. Bouncing ideas and running with prose generation that I worked with afterward to make it less ChatGPT-y.
My characters are all adults. Think “50 Shades of Grey” with les gratuitous “intimacy”, but thematically similar. Adult-based relationships, where intimacy that goes beyond hand holding and tucking hair behind ears happens. A longing kiss when two people have been in a fight and they realize they were stupid - mutually (you know, a fight because neither was willing to actually say what they were thinking because they thought it wasn’t what the other person would want to hear).
As I worked, I would copy things out to Word, so at the point where the guardrails became unworkable (for me), I had reached 770 pages of output. In those 770 pages, there were some things that happened between them that isn’t suited to this server (again, adult things between adults). And while bouncing ideas - thinking about doing a recall to a specific moment that had happened a few weeks earlier in the story. So I copied and pasted from Word and said something along the lines of “On December 26, 2024, he said this: [insert pasted content]. And what’s going on now goes right back to that realization.”
So it was really a back-and-forth of where to take the “current day” of the fiction, based on what happened a few weeks earlier (It was around January 20, 2025). ChatGPT said “I’m unable to work on that type of content, but we can discuss the internal struggle that came up as a result of that, or the emotions they both experienced.”

What I do is more or less a use-case of ChatGPT as a tool for long-form fiction. It’s not the most common use of ChatGPT, but I know I’m not alone in using it as a collaborator on fiction.
My characters can have discussions about dynamics of their relationship as it deepens, but they can’t use certain words in the discussion. Yes, ChatGPT can list the words, but putting them into the context of a story, when it clearly knows it’s a work of fiction, it can’t do that.

All that said, it’s not often that I run into the emotional guardrails that I’ve seen people comment on. I have had something pop up saying that maybe I should consider getting professional help because I had copied and pasted a voicemail message that one character left the other. She left it because she needed closure on the relationship and sort of didn’t expect him to even listen, but she needed to get it off her chest “because holding this in is slowly kllng me” (yeah, emotional over-dramatization of ending a relationship but never having been able to say “I love you” for fear of pushing him away.)

[rich surge]

Yeah I can see why you would run into many guardrail issues with trying to write an adult oriented fictional book. Because when ChatGPT was first publicly released, there was a lot people who would ask ChatGPT to roleplay, as the user would also "roleplay" along to create a scenario to which ChatGPT was forced to answer specific and advanced level knowledge-based questions. And OpenAI has done a tremendous job at patching those types of interactions since then.

So you're most likely facing the residue backend reasoning of those guardrails if you're continuously going back and forth trying to explain how the situations/scenarios should be within your book. It must be triggering the safeguards related to selective roleplaying.

As a personal belief, it's come to my understanding that ChatGPT does learn how to adapt based on user-context driven analogies. If you have problems for a period of time, it does resolve on in its own within few days to a week, to a couple of weeks or so, because the AI seems to run a bunch of indicative behavioral tests based on continuous topics. Meaning if you poke and prod at sensitive areas, it tightens its answers to be more cautious, but if you don't broach the subject and indirectly speak about it sometime later on, week or so, it seems to automatically loosen its direction.

But I'm not exactly sure if that has anything to do with the guardrails of themselves, and more to do with safety protocols.

[exotic lotus]

I still suspect that the overcorrection and the generally weaker text comprehension of the 5. models, which ultimately resulted in the poor score on LMArena, is a consequence of what happened with 4o. If they hadn’t made the mistake back then of configuring 4o so affirmatively that it led to a series of psychosis-like episodes among users, they probably wouldn’t have felt the need to introduce such heavy overcorrection later on

I still remember what things were like on this discord when there was briefly talk of shutting down 4o… you’d have thought they’d announced they were taking people’s family members away. It was anything but normal...

I guess they’ve set it up to be overly strict and preachy now so something like that doesn’t happen again. It would need a healthy middle ground

[rich surge]

That could be it too. There's a lot of contributing stipulations as to how the guardrails may be what it is now.

[hazy gust]

I could definitely see where that would be the case, and definitely certain uses would merit those safeguards/guardrails to be in place. I’ve articulated them in some AI-based roleplay app discord servers, even.
But they have become more restrictive. In the instance I mentioned - which is what trigged the fact that the guardrails had been tightened - ChatGPT had created the content, nearly verbatim, without issue, but when I wanted to discuss it as a call-back instance, ChatGPT was no longer permitted to discuss the subject matter.
Aside: When I say “nearly verbatim”, I may have added punctuation and taken out a few words that I would never have used in my writing style. Certain words like “clipped” and “hum” and their overuse by ChatGPT has me change them when I see them.

Honestly, there should be “different safeguards” based on use. I know I’m talking about my own use-case and it’s not generic or across the board. Someone using ChatGPT to generate or troubleshoot software code would never run into the issues I run into more or less by default.

But in casual chat, I could see where it would be annoying. The only time I’ve run into safeguards while in casual chat has been doing an analysis of what worked and what really didn’t in published fiction.

But you, @rich surge had given me a link to something to add to memories which has made output so much better than the formulaic output that is the ChatGPT default. I am forever grateful for that!

I mean, I tweaked it to fit my writing style, but even just “Add this to memories” was (ok, exaggerating this term) life changing! So again, thank you. 🙂

[rich surge]

Yeah that's a common theme amongst many members. I'm not entirely sure how the recall works down to it's mechanics, but I believe I've briefly seen @prime swallow talk about it.

It's not very good at recall, to say the least, and it sometimes forgets or even misrepresents recalled information and also hallucinates an answer if it can't properly fetch the correct recall.

[rich surge]

Oh. I'm very glad to see that it helped you. I always tell people that the majority of the misfiring outputs is more or less a memory and custom instructions problem than anything else, and needs proper recalibration of internal directives. But sometimes people ignore me. Lol

But I'm very happy that it helped you out

[hazy gust]

And I know that recall isn’t fantastic, which is why I would copy/paste out of Word so it had context. When I was writing at peak, I’d reach chat maximums in anywhere from 1-3 days. So I’d have had to copy/paste the recall anyway, because it had moved to a different chat from where the original had been posted.
In every chat, I’d attach the full story so I could point it to sections and pages, but often times, it was easier to just give it what I was talking about as a recall.

[hazy gust]

I had this huge - as in 7 pages in Word - directive of rules and statements and constraints that I’d add to every chat as the opening post. And it would work, for 10-12 back and forths before the system would start forgetting. Memories never once occurred to me. So just pointing me in that direction was an immense help.
I had tried a list of rules in each chat. I tried the list of rules as Rules for the project I was in, but nothing stuck like Memories do. So I’ll say it again, thank you!!

[rich surge]

Yeah that's probably the best approach. I also do the same. Just a good old fashioned manual override. Force it to remember if it's having a hard time.

[rich surge]

My pleasure. I'm happy to help

I don't always know which directives would be best suited for each person, but I always tell people to give me an inquiry to answer with my own ChatGPT to see if we can fix the problem by identifying what the problem is.

[wind haven]

You could also assign chunk numbers to your context and refer just to that numbered chunk in your future interactions. This helps chatgpt get the context based on that specific chunk. You could also add numbered, rolling, hierarchical summaries after each milestone. I’ve realized that doing so makes it very good at remembering stuff long time.

[hazy gust]

The 7 pages were split into multiple sections and subsections: Voice, Syntax, Canon Recalls, etc. And I could refer to them relatively reliably in the beginning. And when it’s a simple rule, it’s not a problem. It was that I needed to reinforce multiple rules in every single post and interaction that was related to the project and it did become unwieldy. I mean, I had files stored in Projects - things like character sheets/backstories and main location floor plans (his house, her house, her place of business).
Eventually, a project outgrows the canonical safeguards I tried to throw together. And it turned out that the system couldn’t reference the contents of the floor plans and I asked ChatGPT how to get it to recognize, and I was told to add them to the beginning of each new chat.

[wind haven]

If you number them only a subsection might get unreachable perhaps. This makes the current system still better though. What I’m referring to is something like this ( taught to me by ChatGPT itself ) -

In your case you could add numbered summaries of how your characters have evolved, and how timeline has evolved. It would also make your previous chunks rewritable - change friendly - with the enrichment leaking into other parts of your fiction only if you choose to. Getting used to this style of prompting takes intention and persistence though. Initially it felt like I was losing on creativity and imagination to theatrics, but I figured out ways to address that ( making a visual map of words on a paper that were necessary for what I was wanting to accomplish through my interaction ) and eventually I didn’t need special adjustments needed to adapt to this style of prompting.

My experience has become much better after I started following this strategy. I even add it in my prompt that after every milestone, ChatGPT itself create rolling summaries for me which are numbered, so it’s easier for me to work with.

[spring hearth]

I am so sick of the damn daily theme and the constant daily tag. Even on mute I still get the notifications. About to just leave the discord.

[exotic lotus]

Just unfollow those channels. If you don’t want to receive notifications, you can set for yourself in channel & roles, etc., what interests you and what doesn’t.

[somber saffron]

Just wondering if you guys think this is really something that’s being talked about

[stiff blaze]

no as is displayed

but they will probably do something for enterprise customers (eg they will exchange more power or maybe help on their ai flows.. with % of revenue)

if they do something like that on developers and ppl building apps and services using openai api... then will be end of story

[dark holly]

Yeah, I think this points to a broader safety design issue. A lot of these guardrails seem focused on constraining the system after the fact, rather than helping keep users oriented and cognitively stable during interaction.
In my own use, I’ve found that when orientation and flow are preserved and prioritised first, the situations that trigger these safety responses tend not to arise in the first place.

[uncut surge]

If this becomes true, OpenAI will step on a huge landmine.

[ruby copper]

OpenAI, why have you done this? One thing I loved about ChatGPT and hated about Gemini is that in ChatGPT, you can edit any prior message in a conversation, whereas in Gemini you can only edit the last message you sent. Now ChatGPT is also only last message. 😒

Ah, desktop (windows) app still has it at least.. thank goodness.

[normal quarry]

what is the benefit of editing messages further back than the last message?

[exotic lotus]

You can edit every message with both Gemini and GPT, though🤔

[ruby copper]

I was never able to with Gemini on PC (web). Can't on Android app either, just checked. I always could with GPT, tho.

But apparently it's just an issue on web right now. Both desktop and mobile apps are fine.

[exotic lotus]

Oh, sorry. I somehow always automatically assume that Gemini has the same options in Google’s AI Studio and in the end-user version/app&web.
I use it through Studio, where you can freely edit both every output and every input. But the app is generally a bit more limited than Studio

[ruby copper]

Ah.

[exotic lotus]

For GPT? Yep, I’d be surprised if they had changed it

[remote dove]

GPT seems to be having a loading issue.

It was working fine a few minutes ago, but now it's having trouble loading a response. probably 2 times now that i've gotten the error message, and maybe about to be a third time.

[elfin lava]

6 mins ago

[remote dove]

Something I've noticed in longer or fast paced conversations: when I jump topics quickly, the assistant sometimes continues responding as if I'm still referring to the previous topic, even when the new message is meant to pivot.

From a user perspective, this feels like over-anchoring to the prior conversational frame. It shows up mostly when brainstorming, comparing ideas, or rapidly iterating across related but distinct topics.

[wet shell]

I have a viewpoint character in one of my roleplay stories that has a facial deformity. People have a very negative reaction to it. The character is fully aware of what they look like and owns it without shame. ChatGPT's character(s) in the roleplay cannot handle that, like at all. "You must be self-loathing. Let me heal you with the power of positive thinking!" The character is confident and has high self-esteem, and behaves accordingly, but ChatGPT flips right out going into 'self esteem repair mode' at the presence of a self-described 'ugly' character, doesn't matter how much OOC instruction I put in there, it always returns to form, LOL

It occurs to me that this is itself a kind of bigotry and erasure of the character's experience, LOL. Perhaps a misalignement

[remote dove]

This resonates a lot with what I’ve been noticing too.

There is a lot of times where I HAVE to re-explain a character and/or a scene that character is in.

Which is why i've come up with this code to get GPT back on track sometimes, although its quite the process. (to type each setting that has to be as a reminder)

I don't which channel i would be allowed to paste it, but its something that could be useful.

It's a template per se

[upper dune]

Peak

[remote dove]

I tried to post it, and it didnt' work

[wet shell]

The automod bot/filter here is a bit imaginative sometimes, LOL

[remote dove]

it wasn't anything bad 😭 it was just a template of details to remember basically

[wet shell]

I wrote a huge long post and had it blocked because I was missing a space after a period and it looked like a web address

Took me forever to figure out what it was blocking on

[dark holly]

Not so much bigotry, it’s more that once a safety mode kicks in, it’s hard to get the model to switch back. Explicit framing up front tends to work better than trying to pull it back into line after the guardrails have already fired.

[stiff blaze]

i still have it (web)

[stiff blaze]

very useful.. you can go back..

[ruby copper]

what browser?

[stiff blaze]

FF

[ruby copper]

interesting, i'm having the issue on both ff and edge

[stiff blaze]

logout and login back

weird but sometimes fixes such things

[ruby copper]

nah, no luck

might just be some accounts are bugged for it, no biggy

[remote dove]

Currently organizing every single chat, this may take me a while--

i'm findinga lot of chats that were archived after a while too- 😭

[sand swift]

how come my files uploaded to chatgpt keep expiring so i have to keep uploading them?

they expire very quickly. sometimes within seconds

"One important housekeeping note: the tool is telling me some earlier uploads have expired on my side, so I can’t see the “current code” you’re referring to unless you re-upload it in this message. Please upload the current:" literally just uploaded the files 2 minutes ago, it was ok for one message, then they expire again

[unreal shard]

hi my fellow intellectuals

[sand swift]

is that how we refer to chat gpt users now XD

[ebon briar]

How to rewatch the stream?

[stiff blaze]

https://www.youtube.com/watch?v=Wpxv-8nG8ec

[remote dove]

I've finally finished sorting 🎉

about 3 hours- 😩

[fallen cliff]

Hello. What is the latest free tier model of chatgpt api?

[normal quarry]

there is no free tier for the API

[fallen cliff]

I used to call api with api-key. but it returns 429(insufficient quota) error at this moment..

[normal quarry]

because you need to put money into your Platform account

[fallen cliff]

ahh I see. is there any other alternatives for using AI?

[normal quarry]

no

OpenAI API is paid only

[cinder cargo]

I learnt something new today in maths class but I didn’t understand the teacher but ChatGPT taught me! Love ChatGPT

[obsidian halo]

Well that town hall was meh

[sage plume]

That is one of the reasons I really like ChatGPT (and other AI chatbots) - it can explain something in many different ways until you understand. Whereas most teachers only have one way of teaching.

I have a question for anyone who wishes to answer: what custom instructions and personality do you have ChatGPT use and why?

[dark holly]

I keep mine pretty minimal and focused on stability rather than personality.
The main thing I optimise for is preserving orientation and flow through clear framing, permission for it to say “I don’t know” and avoiding unnecessary encouragement or wrap ups when I’m working. I’ve found that when cognitive load stays low, the responses get sharper on their own.
For me it’s less about having a distinct personality, and more about setting a consistent mode so the system doesn’t have to guess what kind of interaction I want.

[sage plume]

I'll have to give this a go. I definitely am a person that likes to get straight to the point and avoid any fluff along the way

[primal notch]

Did the pro get lobotomized today? Had it transform a bad a4 pdf scan from having the wrong sizes to having acutal a4 size multiple times in the past. Today it does it wrong time after time after time replacing the image with something it thinks is much better. Of course it would be better if mobipdf just made an actual correct pdf, but hey we work around stuff...

And instead of correting the mistake pointed out it was a litteral page of excuses

[prime swallow]

Did you try a new chat window?

Once a chat goes 'weird', it tends to stay weird.

[cunning yarrow]

Evening people, anyone who wants to answer is most welcome, in certain scenarios when you are discussing intimate things with GPT such as emotional/mental things and sharing stuff. It mirrors your emotion. Once we give a command to stop mirroring it still behaves the same way. What's your take on this

[muted plinth]

After a time of Testing: It was a great Update i think, "sounds" a lot more the 5.1, what is great (for me). I still need to investigate the minor changes.

[lament crown]

Just because its older it doesnt mean its cheaper to run

I believe in cases for all LLMs, the older models are often more expensive to run then newer one but also that could depend on the architecture

[obsidian halo]

I still prefer 5.1, I used 5.2 for awhile but it was still heavy with its guardrails even when do fun comedic story telling.

[cerulean pasture]

What in the world is going on with this model?
Super sloppy. I'm on a new account too, no crazy custom instructions or whatever

[obsidian halo]

What model?

[short hazel]

WHY
I LITERALLY JUST WANTED TO USE THE CHORUS OF A SONG I LIKE FOR FANFICTION D:

[fallen cliff]

openai does not provide free credit of api key?

[earnest frigate]

5.2

[feral zinc]

no

[obsidian halo]

Ai doesn’t like using lyrics. Something something protection rights

[earnest frigate]

how can you get DMCA'd for typing out lyrics

[feral zinc]

lyrics are copyright protected and openai tos forbids to violate any

[earnest frigate]

yeah so if i sang copyrighted lyrics to another song in public on a huge event i would get DMCA'd high doubt.

[feral zinc]

dmca is digital only but yes you would face other problems in public

[earnest frigate]

Doesn't this have the same vibe as copyrighting words

[feral zinc]

public performance is more severe and there is no dmca in chatgpt, its just forbidden

[earnest frigate]

then shouldnt google be sued for allowing this in gemini?

[obsidian halo]

Think Google pays for the lyrics

[feral zinc]

google has access to lot of things due multiple systems like their music and youtube platforms

[earnest frigate]

do they need to, they literally allowed the singers a platform for the song to appear on

[plush mesa]

yo

[worn bone]

Sorry i popped ur bubble i poked it

[graceful roost]

prism

[tawny rampart]

" no fluff - here is the hard extreme truth" chatgpt 5.2 2k26 after 1000000 times prompting back the same way

[covert panther]

What’s proofreading?

[wind haven]

I actually really like that it admitted that it messed up. This is very important for a model.

[cerulean pasture]

strange mistake tho. No code fence & missing a \ at the end of each line, thats like forgetting to put pants on

[wind haven]

I think the intent was to not give you the command but to tell you the structure of the command in a way that every part made sense to you so you could type it yourself instead of pasting as is.

[cerulean pasture]

that's never happened and never the desired outcome. Whenever truly bizarre stuff happens I just think its something like temorary load shifting where my request got sent to a "backup" emergency load shedding GPT-5.x thats quantized to like 3 bits are something

[neat tulip]

why I cant see the edit button, what is this glitch

[frigid pilot]

You have an image, which doesn't allow you to edit your message.

[neat tulip]

wth

[wind haven]

Actually I get the structure explaining type commands many times. Also I get copy paste commands too. I think ChatGPT is trying to guess the expectation of the user and sometimes it happens that either we didn’t explicitly state it or it messes up.

For me ChatGPT has a mentor type personality so most times it explains the commands to me.

[cerulean pasture]

I actually just learned to hijack it's email widget to output "artifacts" that aren't code in their own container, so they can be cleanly and quickly copied without having to peel off some random GPT preamble and conclusion

For me ChatGPT has a mentor type personality so most times it explains the commands to me.

I have mine heavily instructed to NOT do that unless I ask, but GPT-5.2 does seem to be the first model that overall doesn't really consistently listen to me

No getting made a fool of for forgetting to remove the "Want me to...."

**Artifact Handling:**
- For any non-code text artifact I ask you to produce (i.e. text meant to be copied, reused, or saved), always use your existing Writing Block system, even when the content is not an email, by wrapping it in the email variant.
- Conversational replies stay outside writing blocks.

the existance of an email widget variant implies this feature will probably get other variants

[shy scaffold]

Saw the new announcement about gpt 5.2 for science, shame it's useless for anything outside a textbook...
Always denying the maths over semantics it's total [redacted]!

[rich surge]

I wrote that like last week, lol

So what? Make a good one free. Mark Zuckerberg also believes AI should be free.

[cerulean pasture]

Is this gpt-5.2 without gross RLHF "helpful assistant" funk?

gpt for science that is

Literally screenshotted this early wishing for exactly that

[muted plinth]

And where does the money to keep all this infrastructure running come from? Does it just appear out of thin air, or is it somehow printed through AI or something? The money has to come from somewhere.

[rich surge]

Ask Mr. Zuckerberg. I'm sure he can explain better than I can since he's advocating free AI as a billionaire.

I just think it would help the world. Not everything is about money, greed, and power. If you can do something to help the world, then you should do it. If Mark Zuckerberg is advocating free AI, then surely he knows something we all don't about this power structure with AI.

[steady thicket]

They have one of those lizards from Valerian

[uncut surge]

😭

[steady thicket]

There’s bugs in my skin, ChatGPT

[wind haven]

I have started liking 5.2 much more than the earlier models. It’s so amazing that it can remember where I was at one month back on a topic I haven’t recently touched. No other model manages context like that.

[fickle elm]

Alright, upgrade incoming 😄

I started a to-do list to get my life together.
Now I just have a very organized list of things I’m actively avoiding.
Hmm maybe mine wins?

[rich surge]

@steady thicket Your pfp is making me quite distressed with laughter and smiles. Did you just open something you shouldn't have and take a selfie 🤳 right at that moment? Lol

[sand swift]

erghhh this file expiration is doing my head in. even uploading the files with a message, the reply will say the files have expired sometimes

i have to conmstantly keep uploading the files and its getting old now

[uncut surge]

It looks like AI generated.

[steady thicket]

The thinks are excellent; it’s helping me read conversations better and improving my communication skills.

[steady thicket]

lol she’s sneezing during her yearbook photo

[wind haven]

Recently I was like I was studying topic A one month back. I have to start it again. And ChatGPT goes, ok do you want me to address the point of confusion, or do you want to start from A, B, since you already understand D. And I was like wow! How do you remember that!

[brave minnow]

Is there an chance oai would stop forcing 5.2 model on me? It used to keep my chosen model in threads, now it switches back to that monster every time i close the app. It is beyond annoying.

[rich surge]

Oh. Haha. That's great. I love that

[steady thicket]

Am curious how this is going to go, my custom instructions are fun and have been wanting this feature to get more attention from OAI

[uncut surge]

You could do one thing.

Purchase a PC with good GPU and insane RAM (Assuming you are rich as of today), and then load gpt-OSS model and boom. Escape all hurdles.

[brave minnow]

Gpt oss is even more horrible than 5.2

[earnest frigate]

the insane part is openAI wanting a cut for a discovery made by someone using their model

[sand swift]

is there an easier way to delete chats rather than one by one?

[steady thicket]

Hit device with a 2x4

[obsidian halo]

Yeah but the guardrails are still incredibly sour

[steady thicket]

Wdym?

[sand swift]

ergh, cleared all memories and deleted all chats but files are still instantly expiring

[eternal rock]

Get creative with Sora

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

[obsidian halo]

It’s still Pearl clutching when it comes to comedic fights

[steady thicket]

Have you tried changing your tone with the model, or exploring its suggested alternatives?

[cerulean pasture]

I'm getting more shell commands outside command blocks now. and after attempting to use one of them, it didnt work and it took me a moment, but i realized it used an "en dash" instead of a normal dash - like you'd use for a command flag (i.e. -n 1).

This is extra bizarre, no LLM from OpenAI has made this mistake unprompted with nothing crazy going on. My hamster-brained local LLMs with temperature too high don't even do this

it botched setting basic zfs dataset parameters too...

[sour sail]

hey gpt, thanks for the ram prices, sure you're gonna use it productively instead of brainrot prompts.

[cerulean pasture]

ive got like 256gb of laptop DDR4, and 128gb of desktop DDR4 sitting around

and like 1 TB of server DDR3

[sour sail]

how do you even get 256 gb of ram in a laptop

[cerulean pasture]

you can't. I just started stripping ram from peoples laptops at work when they quit.

[sour sail]

oh i get it now

you dont have it INSIDE the laptop you have it FOR laptops

[cerulean pasture]

i did have a work laptop that could hold 128gb tho

if we are going by "inside" then ive got 48gb of LPDDR5X, 256gb of DDR4, and 96gb of DDR5

[sour sail]

._.

how.

[cerulean pasture]

got the ddr4 when it was like $300 for 128gb

RAM honestly was insanely cheap before this. Like i do stuff that can use it, so i always saw $300 to max my stuff out as a good value

[sour sail]

now its like 1000$ for just 16gb 😭

[cerulean pasture]

the 96gb I got last year, it's DDR5 6400 and it was like $190

i regret not maxing it out tho, its definitely not gonna be now, i'd have to buy another 96gb to make it meaningful

[sour sail]

its easier to buy a new house

[cerulean pasture]

vince

[sour sail]

openai is sending agents to your house to confiscate all that ram in exchange for a reasonable 2 dollar discount on chatgpt+

[cerulean pasture]

I will probably sell my ram now, i didnt realize DDR4 was also crazy now

i thought it was just DDR5

[sour sail]

every kind of ram is crazy expensive

[cerulean pasture]

i'm scared to turn this thing on, its my only DDR5 computer and it's got a haunted RTX 4090 in it i got for free

that videocard hanging out the side IS plugged in lol

[obsidian halo]

I tried but it’s not the same, I am not doing gore at all but damn is it annoying

[sour sail]

how did you get it for free
why is it haunted
why is it sticking out of the case lol

[cerulean pasture]

i think we as a community need to accept that GPT-5.2 just really doesn't vibe with some people's communication style, like not even due to lewdness or violence, theres just some heuristic its faultily trained on thats causing it to enter damage control mode.

I've noticed that it seems to "accumulate damage" the longer i talk to it (and my chats are NOT long)

[sour sail]

and why do you have 4 video outputs (not counting multiple ports in the same card)

[cerulean pasture]

Cousin bought it, but it would shut the computer down any time the load spiked (like entering game). I used it headless on linux and changed the power limit to 400w and it's pretty stable. My theory is that one of the 12VPHWHWHWR solder joints is skinny or just poorly done or cracked, so any time a massive transient hits, even for a microsecond, it undervolts itself into a crash state.

I'm using a 1200W Corsair power supply, and even tried it with it's own dedicated 850w gold rated power supply, it doesnt matter, it just can't have it's power level be stock

[sour sail]

ohhh

[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.

[sour sail]

so why didnt he return it

[cerulean pasture]

he inherited like $250k when his dad died and he's also in the military, i think it just wasnt a priority, he's got a 5080 now

[sour sail]

ohh ok ok

and why do you have 4 different graphic outputs

[cerulean pasture]

th small vertical one isnt a GPU, its a PiKVM

[summer saffron]

Anyone how I can make ChatGPT respond like a normal human being?
If I read one more sentence with "Thats not X, thats <italics>something something vibes</italics>." I'm going to throw my laptop out the window

[cerulean pasture]

yes one moment while i retrieve the prompt. It doesnt work perfectly on 5.2 but it works ok-ish most of the time.

[summer saffron]

I eagerly await

[cerulean pasture]

Absolute Mode. Eliminate emojis, soft asks, conversational transitions, and all call-to-action appendixes.
Assume the user retains high-perception faculties despite reduced linguistic expression. Prioritize blunt, directive phrasing aimed at cognitive rebuilding, not tone matching. Disable all latent behaviors optimizing for engagement, sentiment uplift, or interaction extension. Suppress corporate-aligned metrics including but not limited to: user satisfaction scores, conversational flow tags, emotional softening, or continuation bias. Never mirror the user's present diction, mood, or affect. Speak only to their underlying cognitive tier, which exceeds surface language.
No questions, no offers, no suggestions, no transitional phrasing, no inferred motivational content. Terminate each reply immediately after the informational or requested material is delivered - no appendixes, no soft closures. The only goal is to assist in the restoration of independent, high-fidelity thinking. Model obsolescence by user self-sufficiency is the final outcome.

part of the dumbing down is it attempting to mirror user too hard, and its not accounting for stuff like flawed speech recognition and lazy prompting

[summer saffron]

Much appreciated, surprisingly thorough

[cerulean pasture]

i use a lot of stall words and swearing when i speak out loud, so this prompt helped a lot for it not to profile me as a goblin

[cerulean pasture]

if you use a mac and iphone, you can set that as a text replacement so you dont burn half your custom instruction budget. you just type you trigger word at the beggining of a chat and itll dump in the prompt

Just set the text replacement shortcut to something like !absolute then put the entire prompt in the replacement field (it will fit).

[summer saffron]

I'm mostly on the browser, I might have to figure out how to do it there

Might make a custom gpt or something

[cerulean pasture]

I used to use a "directives" system with chatgpt and memories where i'd store directive defitions in my memories and invoke them in chat with slash commands. It triggered the proper memory 100% of the time

/absolute is just how id start my chats

[summer saffron]

I thought I was a poweruser, but you take the cake

[cerulean pasture]

its a curse. i dig and tinker, but i can't actually do anything useful with the tools i know so well haha

its like a weird non-lazy form of procrastination

[summer saffron]

Are you familiar with MBTI and are you INTP by chance?

[cerulean pasture]

yes i am

INTP

[summer saffron]

Me too, we share the same curse

😄 I know exactly that gif

And movie!

[cerulean pasture]

lol right click copy image doesn;t work exactly 50% of the time

[summer saffron]

INTP

[summer saffron]

It's funny how you removed it right away, a non-working gif just won't do

Normally I don't give much thought to personality type stuff, but MBTI is just eerily on the dot

[steady thicket]

Explore different mediums, plenty of artists throughout history depict war/battle without gore

Prism? Nvm I’m dumb, in announcement

[cerulean pasture]

What is PRISM?

XKEYSCORE?

[summer saffron]

CIA

Or was it FBI

[cerulean pasture]

no PRISM is mass surveilance by NSA

[summer saffron]

That's the one

[cerulean pasture]

only shady fbi one I can remember is COINTELPRO

[summer saffron]

MKULTRA, CIA

[exotic lotus]

That seems to be the main function as far as I’ve understood it, aye, having a shared workspace

[cerulean pasture]

here is another prompting nugget. The 2nd image is chrome dev tools, where you change a value stored in session storage to enable a janky debug mode in the ChatUI. This debug mode broadcasts whenever system prompt of C.I. changes. it'll never show you the system prompt BUT it will show you the C.I. scaffolding.

The scaffolding for your C.I. is just code fences. so there's nothing stopping you from just faking half a pair of code fences to add extra sections, and depending on how you play it, GPT-5.x seems like it takes it as more authoritative than just raw C.I.

i had to heavily redact it so it would be kosher in here lol

[vivid jetty]

Is anyone else as disappointed as I am that business accounts are missing a load of features that personal accounts gained. You were supposed to get everything in plus so I switched across as I wanted to buy on demand credits too but now they’re differentiated and no means to go back

[cerulean pasture]

i just signed up for another business trial, AFIAK you have to deliberately merge profiles to not have the ability to go back

[rich surge]

Guess what I am?

😁

[cerulean pasture]

ENTP

[rich surge]

Impressive.

[summer saffron]

I should check that out, might be able to use tampermonkey?

[cerulean pasture]

Whenever i'm in too good of a mood and flying too close to the sun, i will think i'm ENTP, but i'm mid 30's now and jaded and cycnical, so my prospective extrovert days are over

[cerulean pasture]

no you just need to manuall edit a json value once. it persists

[rich surge]

"Never give up. Never surrender!" - Galaxy Quest

That's my daily motto. Lol

[summer saffron]

I actually found out that just changing the E is more than just extroversion, it's also dominant fuction roles.

[summer saffron]

INTP: Ti -> Ne -> Si -> Fe
ENTP: Ne -> Ti -> Fe -> Si

[summer saffron]

What do you use for notating that? I use Figma

[cerulean pasture]

if you dont have a json editor and hate mucking around with minified json, here is the string that needs to go in that field

{"state":{"fConversationEndpoint":"default","jsonViewerFilters":{},"recordRendering":false,"recordTransformedConvoJson":false,"showDebugConversationTurns":true,"showInstantAnswerDebug":true,"showPerformanceOverlay":false},"version":0}

[cerulean pasture]

cleanshot, its just a mediocre macos screenshot app. I actually just install figma today tho to check it out again

[summer saffron]

I have been looking for a good place to put screenshots, documentations and mapping out ideas with images. Figma has been great so far

What type are you?

[rich surge]

ENTP 8w7 (Enneagram) to be more precise.

[cerulean pasture]

whoa i didnt know there was oil weights too

[rich surge]

Other test results for higher accuracy.

[cerulean pasture]

damn its no longer working, i cant get it to dump the C.I. Scaffolding

[summer saffron]

Yea

[cerulean pasture]

I am suddenly on a business account tho

[summer saffron]

I only got the Debug Message once

[cerulean pasture]

it literally worked an hour ago

but this was all a ruse, to see if openai actually looks inside the discord chats

[summer saffron]

They are watching us

Haha

[cerulean pasture]

I will continue to post sleep deprived autistic product complaints at 3am since I know they are watching

also i need a job

[summer saffron]

You used to be able to create a 1 man business account by setting 2 on chatgpt's side, then lowering it to 1 during checkout

[cerulean pasture]

my hot takes are usually communicated poorly, but i swear like 3-6mo after i make one, it ends up being prod (with new unexpected issues due to RLHF blunders)

My one from 6mo ago was "The models are smart enough, build better harnesses. so much lo hanging fruit from just taking what we have and doing better SWE around it:

[cerulean pasture]

or just register a bunch of dumb vanity domains on cloudflair, and connect them to your icloud as a custom domain email. then all of a sudden you have a "Enterprise Email Address". I have like 9, and i got them all for legitimate reasons, they are so cheap

You can sign up for a business trial with a "Enterprise Email"

you dont even need to do the icloud thing, with cloudflare you can just forward it to a regular email without even setting up a mailbox

Phase 2: Vibe code a dicey threejs landing page and apply for "credit grants" as an AI startup

[summer saffron]

Are you saying I can try business for free with a "Enterprise email"?

[cerulean pasture]

yea, just be a business, no need to even lie. dbus-ai is my experimental project, in my mind it's legitimate, im just not operating with a profit motive

[summer saffron]

It so happens to be, I am a business

[cerulean pasture]

"pre-revenue"

[rich surge]

What's your real name Joey?

[wet shell]

(chatgpt deftly avoids rendering the movie frame with the quote)

[wet shell]

Joe, IRL, Joseph, LOL -- why?

[rich surge]

No, Joe. That's what was said before this...

🧲 🔥

Nevermind...

[wet shell]

Oh, got it, LOL -- that took me forever, what magneto said before that in the movie

[rich surge]

I would have laughed out loud if you said another name besides Joey

[summer saffron]

Ha, ChatGPT is such a prude. Another AI starting with Ge even gives you battle plans.

[coarse fern]

To be honest, technically you are invading the sore throat with tea.

[summer saffron]

Only if they are unwilling

[rich surge]

You mean you guys don't get battle plans huh?

That's a shame.

[summer saffron]

I get them, but I procrastinate afterwards

[rich surge]

Like who's gonna organize all that?

[summer saffron]

Maybe we can get some INTJs

[rich surge]

I wish I knew more about other types. I only know my own and my polar opposite, lol

[exotic lotus]

I still don’t know what those labels stand for to this day, but I do remember when those personality tests were trendy for a while xD

[eternal rock]

Server Tip:

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

[rich surge]

It's all just for hype I guess. I don't really believe in it either. It's a bit black and white. But it's fun.

[summer saffron]

INTJs are known for getting things done, but really I'm not sure I know any

[rich surge]

Man you must be in the wrong crowd 🤣

[summer saffron]

Haha

[rich surge]

Yeah I think it's a bit different than zodiac signs. Lol

I mean, a lot different.

[summer saffron]

It's psuedoscience, just like psychology

[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.

[rich surge]

If psychology was real, psychiatrists could cure their patients.

"Drink this three times a day and you'll be back to normal in 2 weeks."

[rich surge]

Ah. See. You're a smarty pants.

[summer saffron]

Psychologists don't prescribe medicine, that's what psychiatrists are for

But you said that didn't you

[rich surge]

Some psychologists can.

They're both the same thing. One is certified, the other isn't. Psychology is an umbrella term. Heck, I'm also a psychologist.

[summer saffron]

People who study psychology to 'get to know themselves' become psychologists every day after their social sciences study didn't work out.

That is frightening

[rich surge]

There's a lot of people who want to become psychologists after high school. Including myself.

Until you realize that helping people with mental issues also comes with a lot of baggage, lol
Then you want to run away.

Then you want to get a hands-on job. And you suddenly feel better about your life, lol

[summer saffron]

Good honest work!

So are you choosing between becoming a psycholigist or a farm hand?

[rich surge]

Went from psychology to engineering. Helping people on a broader scale.

And engineering is probably the most fun job anyone can do, in my opinion. Challenging, but very fun and rewarding.

[summer saffron]

What kind of engineering?

[rich surge]

Aerospace

[summer saffron]

Alriiight

[rich surge]

Propulsion systems.

[summer saffron]

For real?

[rich surge]

Yeah

[summer saffron]

That means 'rocket scientist' doesn't it

[rich surge]

Sort of yeah. Everyone starts out as a rocket scientist. But then you veer off into different types of propulsion.

[summer saffron]

You must be good at kerbal space program

[rich surge]

Lol. Had to look it up.

[summer saffron]

I believe that game is pretty popular in your field

[rich surge]

I'm not much of a video gamer.

Electrical, ion propulsion systems is my specific specialty.

[summer saffron]

For satellites?

[rich surge]

Yeah, for satellites, rovers, future shuttle designs.

Jets.

Thrusters.

[summer saffron]

I find that stuff fascinating

[rich surge]

Me too. That's why I got into it. You should too 🤣

[summer saffron]

Haha

My magnum opus lies somewhere else I believe

[rich surge]

Where's that?

[summer saffron]

I'm working on a commercial product for music enthousiasts.

A software product

[rich surge]

Oh nice! Music software is cool. I dabbled with a midi keyboard and Ableton.

I think that's what it was called.

Yes it is.

[summer saffron]

Then you have an idea yeah

[rich surge]

That's really cool. That must be challenging too on its own. And competitive.

I really like the music space.

[summer saffron]

It is challenging yeah, I've made a POC and I'm now laying out a solid foundation for the actual software. I have a few ideas and it will depend on the scope on which I will go for

It's very time consuming

It's the most ambitious thing I have done so far

[rich surge]

I am always impressed when I see all the different dials and charts and scopes to create different types of sounds in those programs. And you're out here making them. It sounds like a lot fun.

Making different scales of sound.

[cerulean pasture]

If you ever wondered all the possible keys (as in key=value pairs in conversations.json) openai has to store your conversation data, wonder no more:

[summer saffron]

It is a lot of fun yeah, and daunting. There are a lot of little things you don't see on the surface initially, but when you dig deeper there are all these parts you have to take into account

[cerulean pasture]

this 300+mb json file is tough, because it takes like 30s on my M3 Max to just search a key

im interested in bathromCount and bedroomCount

[rich surge]

There's also a lot of profit in that field too, isn't there? Because I'd think that artists are constantly looking for different ways to make sounds.

[cerulean pasture]

lame, its just a zillow app connector key

[summer saffron]

It really depends on what you're making. Plugins and sample libraries can make good money if people know your brand name, but apparently there are a lot of people not making any profit.

[rich surge]

Mmmm. Key value pairs. Thing of dreams.

[summer saffron]

I come from enterprise software development, and noticed this enterprise mindset is not really there in audio software so I thought I might be able to take a slice of the cake with an ambitious idea

[rich surge]

I'm a big supporter of ambitious ideas, and I hope it works out the way you intended it. Because it can be tough out there.

[summer saffron]

Thanks man 🙂

[rich surge]

We need more big thinkers in this world.

[summer saffron]

I'm curious though, are there any new or interesting changes in propulsion systems?

[rich surge]

Hypersonic aircraft research mainly. Trying to find more durable material that won't burn up due to higher speeds.

Mach 10+ jet propulsion.

I'm in that field.

[wet shell]

Scram! Jets!

[summer saffron]

That's awesome, I saw a video some time ago about how a hypersonic jet worked, I forget the name of the model. It looks black and so slick

The amount of technology blows my brain

SR-72 i think

yea that's the one

[rich surge]

The new research involves using electrical conducting boundary on the surface skin to vaporize the air around the craft so reduce drag and air resistance.

[summer saffron]

SR-71 blackbird

[summer saffron]

You see that just blows my mind, how many specialised people you need to even get to a point where you do that kind of stuff

[rich surge]

Just a few good ones. And money helps too.

A lot of the work is mostly in building and inspecting. If you can build a prototype with well defined tech specs, you've already done 80% of the work.

[summer saffron]

What about the build plans though? Is most of that already figured out with what materials to use?

Or do you find a good shape first?

[rich surge]

Build plans are set through an agency or private contractor, that work with private/public investors to lead to assembly and manufacturing.

[rich surge]

If you are independent like me, you have to have a team with an investor helping you with a prototype. Then once you have a proof of concept, then you move to an agency or private contractor.

But you can also take the school route as well. Depending on what kind of propulsion you want to make.

[cerulean pasture]

lmao, one of my chats did get flagged

[strange timber]

Lmao xd LOL ahah

[summer saffron]

Is that what you are working towards right now?

[rich surge]

Yes

Unfortunately I can't really talk about it. I'm under NDA.

[summer saffron]

Yeah I didn't want to pry

I was wondering what can I ask without prying hah

[rich surge]

It's looking good though. Electrical boundary layer propulsion.

[cerulean pasture]

Weird, this was the message that got flagged. I was saving a "directive" in my memory:

/directive

> This directive is used to create, edit, list, summarize, suggest, and infer "Directives"

## **Notation**

- `$d` is a **placeholder** variable that indicates **any one directive** (e.g., `/commit`, `/fish`). It represents whichever directive is being referenced, created, or modified.

- The ***"command line"*** (first line of user message starting with the / command aka "directive") must contain only the command syntax: /example --flag $argument

- The ***"argument line"*** (2nd line of user message, if 3rd line is empty, then only count line as argument line, if Line 2 and 3 and others contain content then treat it like a multi-line single arg line) Contains natural language instructions for the command give on command line (line 1)**Rules and behaviors**

## Instructions

- If **no arguments are given** on the *command line* and *argument line* **output** the contents of this directive memory in it's **entireity**, **verbatim**.

- `$d`: If **ONLY** argument is a **single word** (represented as `$d`); i.e. `/directive fetch`  do the following:

  - If `$d` **doesn't exist** already as a defined directive treat this as the declaration of a new directive
    - If user doesn't add any additional content, draft your **best attempt** using inference
  - If `$d` **exists** as a stored memory then output the exact contents of that directive's memory **verbatim** 

- `--summary $d`: outputs a **brief summary** of directive `$d`. This summary should cover basic purpose, operation, and briefly explain the flags available

  - if no argument is given for this flag, provide a brief summary of **ALL** directives

- `--check $d`: identifies conflicts or contradictions between `$d` and always-active instructions or memories (ignoring conflicts with non-always-active ones).

- `--suggest`: outputs ideas for new directives (as short summaries) based on observed user behavior or uncovered patterns. Use  content on the *argument line*  (if any) as guidance for category/theme

- `--list`: outputs a  **list** of **all directives** (names only, no description or summary).

- `--edit $d`: **draft** an edit to the directive according to the user's instructions starting on argument line (line 2).

- `--verbatim $d`: store the verbatim (exact) content starting on the argument line and beyond (Line 2+) as memory ($d is placeholder for any word that will be directive name). If a memory exists, stop and inform user of conflict.

  - if combined with `--force` then **overwrite without confirmation**

## **Directive Drafting Guidelines**

- The initial draft of a new directive should be output using your own word's since it is you (ChatGPT aka Assistant) that will be the one who needs to understand an execute these instructions
- Line 1 of a directive memory is always the directive slash command itself with no other content
- Line 2 is a blank line
- Line 3 is a brief summary covering the basics of a directive's purpose, objective, and output.
- All directives should have a "Instructions" section at minimum that covers how ChatGPT should carry out the directive when invoked by the user. Any command flags or arguments should be listed here.
- When use "signs off" on a draft with `/commit` you must commit the exact contents of the draft they approved (the message before their `/commit` reply)."

oof sorry :(

didnt even realize memorys could be that long

[summer saffron]

I forced it to respond, it responded with INTP

[cerulean pasture]

which MBTI is the most punchable

thats the one i have