chatgpt-discussions

[woven dirge]

we're so fudging BACK

WE'RE SO BACK

[tulip flare]

You’re treating style recognition like a deterministic function. But this isn’t Stable Diffusion with a fixed LoRA. It’s a text-to-image transformer operating with latent diffusion and prompt-weighted interpretation. Variability is the default, not the bug.

[runic wraith]

tried, didn´t worked
tried again, worked xd

[compact girder]

global delay

[runic wraith]

yup, anyways, im back baby >:3

[fallen bolt]

Ex prompt :
(In my native language)

Create a 3d pixar style from reference image

The scene is the woman as in reference image pushing a box in a small store.

Using this same instruction and applied to both account create different style , i also try using gpts animation script builder and it give bad result for 2nd account and 5th account.

[tulip flare]

If GPT accounts were artists, they’d still be drawing from a shuffled deck of influences every time. Same instruction ≠ same brain.

[fallen bolt]

If u keep defensive like this for a problem that i asking for a solution.. then i guess.. no solution for u but talking

[runic wraith]

dang it its just my first request on the time it took me to log back in

[fallen bolt]

Did u read above others also experiencing the same issue with me ?

And if u not representing open ai staff.. the. Better silent because your talking didnt solve any problem

[tulip flare]

You’re not asking for a solution. You’re asking for agreement and when someone explains the actual cause, you call it “defensive.” That’s not a conversation. That’s ego protection.

[vast viper]

Where can I go to talk about something i have made in chatgot and want feedback on? It's a text based game.

[fallen bolt]

Read my words again.
Why i so stubborn not understand the difference between randomness and distortion ??

[runic wraith]

isn´t reporting there is another user with the same problem something open ai staff takes into account to fix it better?

[tulip flare]

I do understand the difference. You just don’t like that they’re connected. What you’re calling “distortion” is what happens when randomness doesn’t go your way.

[fallen bolt]

I am about too.

But i want to know maybe there is a solution from this community.

[hasty sedge]

Is chatgpt down folks?

[fallen bolt]

And someone keep defensive telling me about randomness that i didnt ask about.

[tulip flare]

It's having... issues.

[hasty sedge]

Seems to be.
Thanks for the status update

[fallen bolt]

Did u see the eye from the picture ? Its just 1 picture from many pictures i do backtesting

[runic wraith]

nice to have you here, idk where you can chat about it, this channel is currently busy with gpt issues, however im interested on your project :3

[fallen bolt]

If u cannot see a normal eye from my picture.. than what the point i am talking to u zer ?

U dont even understand what the problem and keep talking about randomness

[kindred cypress]

I'm having troubles with my ChatGPT. How are yours?

[tulip flare]

Welcome to AI image generation. First night here?

[runic wraith]

know what randomness is about?

[fallen bolt]

Yeah its my first night with more than dozens of paid ai tools for my company.

How manu ai tools have u do backtested ?

And i am ended talking with a newbie.

[tulip flare]

I saw the eye. That’s called a generation artifact, and it happens across all diffusion models: Midjourney, DALL-E, SDXL. One distorted eye in a set doesn’t prove account quality. It proves you’re working with probabilistic noise and latent space sampling. That’s just the nature of the system.

[neon dagger]

Why I got Too many concurrent requests

[runic wraith]

almost fine, first stop showing me my history, blured my user icon and refused to generate, i tried log out, but then i couldn´t log back in until now, now it says i have "too concurrent requests" out of nowhere.

[runic wraith]

same here

[fallen bolt]

U have no capability to solve the problem i will file the issue to the team maybe they have better solution.

End of talking kiddo.

[kindred cypress]

Same here. And before that, I had the same problem with my chat history like @runic wraith.

(Further Context: I'm using the PC app)

[hasty sedge]

Same

[worn oasis]

It’s running very slow for me right now when it was fine before. Is it slow for others?

[fallen bolt]

I have midjourney. Stop talking things that not relates.

Midjourney create much less distortion. Omg

[tulip flare]

Experience doesn’t change the fact that you’re misattributing a stochastic process to account quality. I’d respect your tools more if you respected how the tools actually work.

I do this for a living. I know what I'm talking about.

[hasty sedge]

Facts

[tulip flare]

Less perhaps, but not none. All AI image generation is prone to this behavior. ALL OF IT. Every single model does this.

[hasty sedge]

Midjourney creates artifacts too @fallen bolt don’t trip lol

[fallen bolt]

We are talking about different account of cgpt create very big different result.

Have u backtestes several paid midjourney account ? How many midjourney account do u have ?

Omg

[tulip flare]

You’ve got experience. Cool. So do I. But experience only matters when it’s paired with understanding. This isn't about which tool you like more, it's about how all these tools function under the hood. If you want consistency, don’t switch accounts, switch to deterministic generation methods.

[fallen bolt]

Now u mention about artifact i understand.

So.. answer this why my 1st acc do it less and my 2nd acc do it more ?

[hasty sedge]

That is placebo, different accounts will not do that. If you use the same seed for generation you get the same result regardless of account.

[fallen bolt]

And why free account do it better than paid account ?

Have u backtested free and paid version about this artifact ?

[ivory matrix]

Is it fixed yet?

[tulip flare]

Because artifact frequency isn’t tied to the account, it’s tied to generation variance. Things like prompt timing, model load, backend routing, temperature, and random seed can all influence results. Your first account may have just had better “rolls” or ended up routed through a slightly more stable backend during generation.

[kindred cypress]

Apparently, my phone's ChatGPT browser version is working normally. 💀

[fallen bolt]

So based on this answer is :

When i subs new account.. it might give me acc with worst reasoning for image generation such as more artifact.

True M

[tulip flare]

There is no inherent difference between free and paid accounts when using the same model (e.g., GPT-4o). The results vary based on backend routing, not payment tier.

[fallen bolt]

So the roll ?

Just forget about temperature.. etc

I am interested with the rolls

This rolla.. tied to the account coding ?

So new account will be different forever ?

[tulip flare]

The rolls, as in dice. The randomness.

It's a figure of speech.

[hasty sedge]

You’ll never generate the same image twice unless you use the same seed

[fallen bolt]

Understand. But if it happens for long period of time.. do u think its still about backend routing or something else ?

[fallen bolt]

Thats not what i am asking.. omg read again

[tulip flare]

It’s not that your new account is assigned worse reasoning or has more artifact bias. It’s that when you use it, you may get unlucky with the generation conditions: backend routing, noise sample, seed, etc. But it’s not a fixed trait of the account. It can just as easily produce great results on another day or prompt.

[fallen bolt]

Unless u understand the problem please dont talk others things.

[hasty sedge]

😹

[fallen bolt]

Unlucky happen randomly not everytime.

The backtested done for 2 weeks.

And it happened everytime.

[tulip flare]

Yeah, if it’s happening over a long stretch, backend affinity could be part of it. Sometimes an account ends up hitting the same cluster more often, and if that node is under heavier load or has slight quirks, you’ll see more artifacts. It’s not about the account being bad. It’s more about where the requests land in the system. OpenAI also runs quiet A/B tests sometimes, so minor differences between accounts do happen.

[tulip flare]

But he's not wrong.

[fallen bolt]

So when this happen for long period of time.. its tied unluckily to my account ?

[tulip flare]

Yeah, pretty much. It’s not that the account itself is flawed, just that it might keep getting routed to a part of the system that has more issues or variance. It’s unlucky, not broken.

[fallen bolt]

This explanation makes lot of sense.

Now. If that so.. no solution i guess..

Although i have done with different image ai tools this unlucky things not too worst.

[tulip flare]

I see the same thing with my single instance. Some days it's on fire, gets everything right, and feels like it's reading my mind. Other days, it fumbles basic tasks, misinterprets prompts, or delivers weird, low-quality results. I'm not changing anything on my side. Same model, same prompt style, same setup. The only real variable is what's happening behind the scenes: server load, routing, whatever. It’s unpredictable. You can do everything the same and still get a completely different outcome depending on the moment.

[fallen bolt]

So its makes me wonder.

The 6th and 7th acc for my next employee.. how to make make things better ?

[sullen shard]

chatgpt still isnt working for me

[sand lily]

.> ye, me neither

[tulip flare]

There’s no special setup to make new accounts perform better. You can’t control which backend you get routed to, and there’s no configuration that guarantees fewer artifacts. If you want more consistent output, the only real lever you have is prompt engineering and controlled retries. Use clearer instructions, break down requests into smaller steps, or just regenerate a few times and pick the best result. That’s how everyone deals with this. Not by managing accounts, but by managing expectations and workflow.

[sand lily]

time for video games

[tulip flare]

Yeah, mine's changed from a 500 error to 409. Fortunately the Windows app is still working for me.

[fallen bolt]

Its not about prompt engineering. Its correct about the backend

Thanks for the explanation.

Thats at least give me a broader picture of how cgpt works compare to others.

[tulip flare]

To be clear, I’m not saying prompt engineering causes or fixes artifacts. I’m saying that when you’re stuck with unpredictable backend behavior, prompt refinement and smart retries are the only tools you actually have to work around it. You can’t fix the backend, but you can improve how you interact with it. You asked how to improve it for #6 and #7, that's how.

[fallen bolt]

If we give the same 100% instruction to 5 people and 3 of them can do well but 2 of them do wrong... its not the instruction. Its the 2 that have problems.

Refining your prompt to be adjusting with every single accounts is mean that we got a problem

If later i habe 10 accs ... i must refine every prompt for every acc just to tell them what is 3d pixar style means M

?

[tulip flare]

GPTs are not people though.

That logic would make sense with people. GPTs don’t share a brain or a memory. They’re instances of a probabilistic system. When the same prompt gives slightly different results, it’s because that’s how the system works. The variation is baked in.

If you’re expecting every instance to behave exactly the same, every time, then yeah — you’re going to think there’s a problem. But the system isn’t broken. The expectation is.

[fallen bolt]

The basic one. Not the complex one. The basic one is just pure about style.. pixar.. 3d ultra realistic , 3d semi realistic.. etc.

If we give input create 3d ultra realistic then it give result cartoon style.. its not the input is wrong

[tulip flare]

Style prompts are actually where you’ll see the most variation. Words like “Pixar” or “ultra-realistic” aren’t fixed instructions, they’re interpreted differently depending on token weighting, backend model state, and generation noise. You might think it’s a simple task, but the model doesn’t see it that way. That’s why even "basic" prompts can give inconsistent results.

[fallen bolt]

So i cannot expect cgpt to create 3d ultra realistic render for me ? Because it more than happy to make just cartoom style ?

[fallen bolt]

Ok tell me how to change reference image into 3d ultra realistic style ?

The reference image still in cartoon style

[tulip flare]

You can absolutely expect it to generate 3D ultra-realistic images. It’s capable of that. What you can’t expect is 100% consistency every time you ask for it, especially with vague style labels like "Pixar" or "ultra realistic." Those aren’t fixed toggles, they’re interpreted differently with every generation. That’s not about preference. That’s just how the model works.

[fallen bolt]

I am asking

How to give input of a 3d ultra realistic to cgpt

So cgpt wont give me cartoon instead.

Or vice versa.

How u tell cgpt to create 3D pixar style.. and not ended up with 3d semi realistic

[tulip flare]

Try something like this instead:

"Create a highly detailed 3D render of [subject], in an ultra-realistic style. Use realistic lighting, textures, shadows, and depth. The image should look like a cinematic render from a high-end CGI film or game engine. Photorealistic materials, sharp details, and physically accurate proportions."

Add specifics about detail. If you don't it will think up whatever it wants to fill in the blanks.

[fallen bolt]

Thats the prompt that i got from cgpt too about how to define pixar style.. but and the result u see as i attached above

But it works well with my 1st acc

[runic wraith]

my previous issue is now fixed for me xd

[fallen bolt]

But i know now.. its correct about randomness unlucky things for the backend. So when i subs new account.. this unlucky may occur.

Thanks

[vast viper]

@sand lily I have a game for you if you are interested.

[tulip flare]

That’s exactly what I’ve been saying. The difference you’re seeing isn’t because one account understands better, it’s because image generation isn’t deterministic. The prompt is solid. It’s just that generation results vary across attempts and accounts due to backend routing and randomness. You didn’t get a better image because one account is smarter or is paid for. You just got a better roll.

[fallen bolt]

Yes the roll agree.
I must use another image generation that have better roll and consistency if want to have dozens account for employees

Understand.. thx

I am asking because i have doubt whether to continue using the same tools or must combine them with another for work efficiency. And thx for the answee

[tulip flare]

Glad you understand. This "newbie" appreciates the conversation.

[fallen bolt]

Well u were one. Not now.

[tulip flare]

Wow, thanks. It means a lot coming from someone who taught me so much by arguing with every correct answer I gave.

[native swallow]

Hi, is anyone having issues to make any consult...?!

[river panther]

How's it looking for everyone? Mine seems to have come back online.

[vast viper]

It's running for me

[sly geode]

is this new ? or not released why it doesnt show anything

[dire owl]

Hi, I heard many OpenAI devs are leaving? Does this mean ChatGPT will shut down soon?

[tulip flare]

Unless Sam Altman suddenly decides to unplug the servers and take up goat farming, ChatGPT isn't going anywhere.

[tulip flare]

Allegedly a placeholder for a future or unreleased feature. It'll be used most likely for scheduled automation features in Teams and Enterprise. That is all speculation though.

[sly geode]

i need the ones that will be able to generate and post automatically all from chatgpt and not n8n or zapier or make automation

maybe gpt 5 does

cuz api cost allot

[tulip flare]

Entirely possible. There's not much information out on it yet.

[tulip flare]

Apparently 6 months ago there was beta feature called "Tasks" and a "4o with scheduled tasks" model according to articles online. Might be a remnant from that.

[sly geode]

i did used that allot but it was underwhelming

they scrapped it ?

[tulip flare]

It was just a beta. So it could potentially be something they release with GTP5. There's speculation it's going to have a lot of new features.

Not a feature I would have used 6 months ago, but I would now. I'd love to see what it can do now.

[normal quarry]

it's available now -> https://help.openai.com/en/articles/10291617-tasks-in-chatgpt

[tulip flare]

Ah, nice

[velvet knot]

Is ChatGPT down?

[jade night]

Are there any plans for adding a search function in chatgpt? Would be really helpful.

[fickle forum]

who replied to me

only for plus?lol

I lose it now

they remove it from my chatgpt

😭

not a good tool tbh

[rugged osprey]

i think they were attempting to push it to edu accounts first

[fickle forum]

and promote a preview version to part of the users

[rugged osprey]

usually how their releases go

[orchid ember]

Has the image gen model always thought when it’s set to 4o, or is that something new? It doesn’t always do it, but it’s done it a number of times in different chats and I am positive it set 4o. It’s just confusing because I didn’t think the Omni model had that ability. So is it unlabeled model routing?

It’s always very short span of time: 4-6 seconds. It’s not the same behavior of say I try to generate images with 03 because it will think about the prompt. I gave it for 30 seconds to a minute pretty consistently if not longer

[open furnace]

They should create a separate image model specifically for images and allow more images to be created or edited within ChatGPT.

[narrow viper]

There's no need for a separate image model since GPT4+ is Omni. If GPT5 is a reasoning/omni model then we'll get the best of both worlds together.

@orchid ember What you're seeing has been there for a long time. I don't it's "thinking" so much as just taking longer to consider the detail, and "Thinking" is the text that they're displaying, even for the non-reasoning omni model.

@jade night ChatGPT has always had a search function in all interfaces. Which interface are you using?

[orchid ember]

I see. Yeah I could’ve missed it. I only really recently started heavily requesting images with complicated prompts recently.

[narrow viper]

👍

[mystic cloak]

bro what

This is 0/10 marketing

[flat falcon]

what do you think that anoucments about?

[mystic cloak]

Just kidding probably a teaser

[gritty rover]

when will we get an account switcher?

[mystic cloak]

I think the cursor was thinking

[gritty rover]

gemini has it and it works great

[tawny oriole]

They’ve been working on their own browser for a while maybe it’s that?

[mystic cloak]

The cursor has a job and said what he needs to do

[flat falcon]

good idea

[mystic cloak]

It’s a teaser for tomorrow’s livestream at 10pm PT

[tawny oriole]

What do you think it is? the cursor thing?

It would be kind of sick if openAI browser app has like a picture in picture window where the AI opens a its own tabs where you can glance at it browsing at 300 actions per minute like it’s playing StarCraft

[tawny imp]

Would be cool if plus members got access to the agent or something

[jade night]

Oh wait you're right, my bad.

Something weird I noticed right now. I was using chatgpt in PC firefox and then put my pc to hibernate, this was last night. I then picked up my phone and continued the chat before sleeping. Now this morning after booting up I accidentally continued where I left off yesterday on my PC, the mobile chat didn't show up/update on my browser. Now the chat I did on my phone is completely gone.

[fading coral]

Is the screenshare feature just not availabe on the Windows desktop app version in the US?

[dense halo]

why is voice mode cutting in and out

[unborn vault]

This happened to me once too, a few months ago. When I researched, it seemed to point to a syncing glitch that can happen sometimes if chats are left open in both places. Not recoverable.

[simple tulip]

Anything is just against the restrictions

[jade night]

Yeah I sadly couldn't recover it. I opened my phone app and it literally faded away from my view. Maybe if I had opened it with internet off it would have stayed I think

[fickle forum]

what will be released on 7.17?

[mossy osprey]

Is odyssey launching today?

[sudden wyvern]

What is Odyssey?

[orchid pollen]

what does "operator" mean? what is it doing?

[limpid fossil]

Can you help me understand what change was made at the end of may/start of june to the context window?

One way or another, ChatGPT was noticably less "forgetful." I used to be able to have a starter prompt with a list of characters, and go for a very long time before mentioning a character from the list while still having the character's details be pulled properly. Now this is not the case. I can see with my eyes that this is not the case anymore.

Even if the context window hasn't changed, did it used to be that, for example, the Starting Prompt, was ALWAYS kept in the context window?

[barren zodiac]

am i the only one having o3 get in a loop tryiung to run pyuthon code but writing (no don't use code, next text should not be code) in the code block, as if it had skyzophrenia, and it keeps producing empty code blocks forever? Started happening randomly

[narrow viper]

@limpid fossil (I'm not staff, so...) I don't know exactly how they manage the context window, but I know exactly what you're talking about with maintaining the first prompt at the beginning of the context. I used to think about that too. When the context window was smaller I used to occasionally prompt with something like this:

Summarize our entire chat, original goals and ongoing topics, noting important points and details that might affect context if lost.

The response included all relevant details, now at the bottom of the session, so if the top scrolled off active memory it wouldn't matter.

When ChatGPT was new and I was still getting familiar with its context, I would also test it once in a while:

| Tell me about the first prompt in this session.

When the answer was wrong I knew I had gone beyond the limit. But I haven't needed to do that in a long time. I haven't had an issue with context size for a long time - I simply don't carry on conversations that seem long enough for the window to crop off the first prompt. My solution is "don't do that".

I think the way most people use ChatGPT for story writing is fundamentally the wrong way to use the technology - people start by just talking to the input box and that's the way they continue to do it, without much concern for how the technology works ... until it doesn't. There's a better way.

First, define terms:

• Canon (Core Elements) consist of the facts and rules officially part of your story world. These include your main characters (their back‑stories, abilities, relationships), the settings (geography, eras, architecture), and the major plot events. Everything in your canon must stay consistent or you break continuity.
• Flavor (Incidental) Details are transient or decorative bits like someone's favorite ice cream, the exact number of chairs in a café, background NPC names - things that you can forget or change without harming the story’s integrity.
• A story bible (not a religious concept) is a listing of canon and flavor elements.

[limpid fossil]

My issue is that I used to be able to do it, and now I can't.

[narrow viper]

Create a Project prompt or a Custom GPT that includes instructions like this:

When directed to 'summarize', generate a new Story Bible. Extract every canonical element and structure them under these headings:

• Characters: name, role, description, key traits, relationships
• Locations: name, description, geography, notable features
• Events: title, summary, date/sequence, participants, outcome
• World Rules: magic or technology systems, timeline rules, cultural norms
  For each element, include only facts that must remain consistent.
  Then list any incidental 'flavor' details separately under 'Optional Details'.
  Output the Story Bible as a clean markdown document that I can copy into another chat to continue writing without losing continuity.

So now when you get to the end of a chapter or just some important section of your discussion/story, just say 'summarize' and pick up the discussion in another session. You can also use this to check for continuity, maybe by copy/pasting all current story bible output, and asking the assistant if continuity has been broken anywhere.

[limpid fossil]

I never needed to worry about reposting my story bible until about a month and a half ago.

[narrow viper]

Yeah, they don't publish details of what changes so we simply don't know, and we have no view of the context in each transaction. It's a black box. The only way to know for sure is to use the API where we control every aspect of the context. I do this a lot.

I forgot, are you free or plus?

[limpid fossil]

plus

[narrow viper]

same

(Eagerly awaiting what Eskcanta has to say on the topic. 😁 )

[prime swallow]

I can discuss ideas I have from my own explorations.

The model seems to have issues with some kinds of conflicting instructions or information, and that can come out in weird behavior, which often seems to involve rewriting what it was told in a hallucination so there is less conflict.

That can make it seem forgetful.

For my work, when I have elements of 'this isn't working for the model' in what I do with it, it gets 'forgetful or makes stuff up that I didn't try to design'. In fact, that's 'the tell' I watch for in my work with it, I feed it chunks, watch it discuss, and watch for weirdness; if there's weirdness I am sure it is something relating to that new chunk or how that new chunk is combined with something already said.

Exactly how I tell the model something can matter, even if I approach a storyline or character or concept from top down vs ground up (like a founder story instead of describing the company from an outside view, if the company is unusual, gets me a good model view of the company, instead of some confused mismash of normal companies and what I described - even if I explain the founder story later).

So, I wonder if it's less the model changed then, and more some key detail in how you're worldbuilding or other with it was added, so now it's 'getting weird'; which could explain why it isn't getting weird for everyone but dang it, you're seeing it constantly with this long story.

Can even be a typo. Someone once asked me to look over an odd-behaving model and the instructions were for this super sweetly described character... and included one line 'would do any harm'. Was meant to say 'wouldn't do any harm' but the person typoed... and that typo made sense to the model. And made the model 'behave rather weird'. Fix the typo, back to normal good 'this is what I asked for' outputs.

[prime swallow]

Thanks! For anyone confused, I'm 'just' a volunteer guide, I don't work for OpenAI or anything, I'm a community member like everyone else! #server-staff

[eternal rock]

Get notified for server events!

Visit id:customize to pick up the <@&1261377106890199132> role.

[narrow viper]

I recently worked with someone here who had a small fundamental wording issue that had the same effect on the responses.

[limpid fossil]

I'm not making typos. It's the same starting prompt as it always was.

I know this because when I tell stories the first time around, I experiment with different story branches, but each branch takes up space in the "conversation" even if you're not moving down that branch in the context window.
e.g. If you simply regenerate your FIRST reply from your FIRST prompt enough times, chatgpt will eventually say that the conversation is too big.

So once that happens, I will literally go back and copy paste the exact same prompts from the "final branch" into a "fresh" conversation so that the conversation isn't bloated by superfluous branches.

I do this all the time, which means I can easily see changes in the quality of the responses.

[narrow viper]

But Eskcanta there have been a lot of reports here of context limits, forgetfulness, etc. I thought the issue was isolated but there are too many reports - ergo my note above "If I had a buck for every report of short context....".

[limpid fossil]

This is why it is insanely frustrating to me for people to tell me something is user error instead of service degradation when I am constantly by definition performing controlled experiments

[prime swallow]

I hear you, clearly.

If there's some confusion the model has with what you've explained one way and another way, or any other confusion or conflict; could be between what you input and it's programming or safety stuff or anything else like that.

A conflict between instructions can cause that behavior, though sure can other things.

[limpid fossil]

You can see me saying this as far back as june 20th

[narrow viper]

McWar - yeah, the conversation branching is a great but entirely unused feature by 99% (?) of users. I don't think that should be a part of the context limit because it's not supposed to be processed. I suspect the context limit is calculated on the entire discussion and all branches ... that maybe could be considered a bug?

[limpid fossil]

why would there be a confusion or conflict in one copy of a conversation but not in a 1:1 prompt-for-prompt copy of the same conversation

I feel like I'm in 1984

and big brother is telling me that the chocolate ration has increased.

and that we've always been at war with eastasia

[narrow viper]

ahaha double-plus good on that!

[limpid fossil]

Information is not conflicting, it is failing to recall it

[narrow viper]

I think the issue is legit that context has limits, and that some of the limitations aren't well-explained or ideally calculated. But at the end of the day there Are limits (new or existing) and the way to avoid them is simply to not have huge discussions with the model. Put your valued content into a document and upload it to become the active context. Replace the file once in a while with a new file with a timestamped name.

[limpid fossil]

Look, even earlier the same day, here I am talking about the same behavior pattern. "have you considered that maybe your prompt wasn't sophisticated enough or used enough big words for the model to decide that you are worthy of intelligent response?"

[narrow viper]

McW, I believe the issue is with context, not content. I think Eskcanta hasn't seen this as much as we have. He'll catch up.

[limpid fossil]

I understand what a context limit is. I am saying that the context window has been reduced, or something has been simplified in a negative way with how prompts are organized within that window.

[narrow viper]

Yeah, and the answer is "we don't know what's happened".

[prime swallow]

Yep! I notice plenty too.

So I started an experiment; I'm working on a full length novel with a lot of twists and turns, about 30 characters, shifting PoV. One custom GPT to help track the book as it gets written, about 30 custom GPTs to each know each character from the inside PoV of that character, down to that character's bias and misinterpretations and the lies it believes from other characters; that other character's GPT knows the lie is a lie, what the truth is, and why they believe they told it, and life-long life histories for adult characters.

And a couple of these characters are companies, with their various departments and major internal disagreements and goals.

Complex stuff, and works like clockwork, even in the test GPT that can see all this info so I can explore what IT thinks of it.

I build the information files in chunks, a few sentences at a time, discussing the whole with the model, seeking confusions or conflicts. If the model 'does not get' what I am adding, I change the wording. I explore the model's confusion compared to my intent, and fix it, then explore the next chunk.

This is working.

Loads of history, the model's remembering and tracking and can cross explain really well.

I throw in even 1 conflict that isn't explained, so the model has to guess? Stuff starts falling apart.

It's fine if Bob lied about what he did to Mark and Mary and Mary believes him, Mary doesn't, and Mary's doing X about it.

But if I leave out the explanation about the lie, and don't make it clear that Mary info is Mary PoV, and Mark info is Mark PoV, and neither is true, just believed...

I get really confusing output, because the model seems to be deciding what 'ground truth' is since I'm not clearly providing it in a 'model explained this back to me and it was the same story still' way.

I literally start with 'here's the first chunk, let's discuss what this means to you' as I go.

[limpid fossil]

I agree. I think the first step to knowing what happend is for people, not necessarily you, to stop gaslighting every paying customer who reports a degradation in service quality.

[narrow viper]

I build the information files in chunks, a few sentences at a time, discussing the whole with the model, seeking confusions or conflicts. If the model 'does not get' what I am adding, I change the wording. I explore the model's confusion compared to my intent, and fix it, then explore the next chunk.

Exactly! Most people don't do this. They put the entire novel in a single chat and then wonder why continuity issues develop.

[prime swallow]

That can come from 'forced model guesses'. It has more than one way to interpret the info you give it, and so it guesses, randomly, each answer; once a chat makes a specific guess then it stays with that guess.

This is functional in at least some cases; open a chat with 'Tell me a story' and three new chats might each tell a different story. If you keep chatting, each chat gets to know the story it told you, so it answers onwards based on that.

If your wording is such that the model is guessing, and not always making the same guess, then I would expect you to see what you're describing here, 'All the same inputs in order, very different model outputs; some I love, some are nuts'.

I personally fix this by finding where the model is guessing, and reexplain the part that leaves it guess-room, so it now predicts only the path I want it to take.

[narrow viper]

McW - there are a lot of kids that scream "bug" when they don't get what they want. And many of these folks have free accounts with different limits. A lot of patience is required from all parties to help determine the competence of the user, what actually happened, and the cause. It's a consequence of all of us dealing with unknown people and scenarios and a huge variety of options ... all with the same product names. 🤣

[limpid fossil]

Esckanta I appreciate your efforts to explain to me but I fear you do not fully comprehend the issue I am having and your explanations of remedy are not contributing to a solution at this time.

every conversation starts with a story bible. the prompts being put in on the second go-around are themselves the results of iteration for response quality in the previous conversation

So when, in the second go around, it is clear that the model no longer even remembers characters' names, this cannot be explained by user error, nor fixed by a collaborative effort with an increasingly forgetful model.

[narrow viper]

For now I suggest trying the story bible approach to establish valued context, copy/paste that once in a while, use it as your own monitory for continuity, and deal with context limits as they are. Add a +1 to #1070006915414900886 about this to help them get some visibility. Staff do look at those reports.

[prime swallow]

I think I'm using my 'gold pan' a different way.

If you mean me personally see this; I'm aware users report this. I explore it and solve it within the chats I work with, when I can.

If you mean 'catch up' as in start seeing the same problems the same way and being stuck - it's possible. But I'm exploring pretty deeply with what I consider a very complex storyline; I can't freehand, no notes, list all the details. And the model's keeping up with the very carefully written notes. Where it can't, I view that as a warning that I need to find and fix something confusing the model involving those details, and so far, I've found a reword that has worked, so I'm still building bigger and adding more characters and their full histories, carefully and checking for 'what does this mean to the model' for if I should reword, and 'does this conflict with any other detail already set' which also makes many things 'go off' when happens.

[narrow viper]

Eskcanta you and I aren't seeing the issue because we're engineering the prompts. Most people are focused more on their stories, and only shift to engineering on the technology when it becomes a requirement. We've been there. Others haven't. So McW and others are reporting the problem from their current perspectives and we need to help to move them forward, recognizing the issues with context and with their prompts. It's a process...

( Live stream in 16 )

[prime swallow]

I'll stop offering my view with this small last detail.

I've seen similar to what you described, and seen it fix.

The fix I have seen work more than once was going line by line through that initial stuff and having the model explain what that means to the model, and noticing 'That's wrong, not what I intend it to mean at all'. And finding a way to word it so the model 'gets it as intended' and checking for any other 'confusion bombs' to disarm.

Sorry I appear to not be contributing to the solution you want.

I get that your solution appears to.... be that OpenAI stops gaslighting people?

You did say "I think the first step to knowing what happend is for people, not necessarily you, to stop gaslighting every paying customer who reports a degradation in service quality."

OpenAI employees are not here. There's not a single gold name in this chat, I haven't seen one in a long time.

We can't make other humans do anything, even if they're doing the thing, and we are not those humans.

So I don't see how you think 'that would be a step in the solution' to talk about that here where you're only talking to other community members. That's complaining, right? That's not even a possible step towards a fix for what you're describing, because those people are not here even if the thing you accuse is part of the problem, which is not clear to me either.

[narrow viper]

I need to write a GPT or app for story writing that helps people to avoid these fundamental issues. Then they can talk to their bot all day long and it won't forget their canon. Let's call this project 717. 😁

[limpid fossil]

My problem is that every solution which amounts to prompt engineering advice ignores the fact that I am already engaging in deliberate prompt engineering. Suggestions that I maintain a story bible ignore the fact that I use a story bible.

Convincing other customers via complaint, particularly experienced ones such as yourself who advise other users, I think is a plausible step towards a remedy to the solution, because when the next person complains about service quality, you will hopefully tell them, "There are ways to partially get around this via prompt engineering but the core of the problem is that openai has degraded the service in some way they are not telling us so you should report this to them or amplify an existing report/complaint."

[narrow viper]

McW - I'm hearing you. We're just a couple guys just like you, sitting here and typing about our experience. All we can do in this forum is to try to help one another. There aint much more to it. Official reports should go through official reporting mechanisms (app > Profile > Help Center).

[prime swallow]

I don't see a degradation. I think that's a fundamentally incorrect and inappropriate assertion.

And I cannot and will not falsely represent.

You can and should use #1070006915414900886 and #1070006151938314300 . I view the issue you're talking about as more akin to 'I drive this car the same way every day and it always crashes. The car maker is at fault and we should organize to force this to be fixed.'

When I'm aware that the way I and others drive does not result in car crashes, so... it's not clear to me that the way the car is made is the problem.

Still it can use some improvements? So use the #1070006915414900886 and #1070006151938314300 freely, that's what it's there for. But me, I'ma look at the many ways to drive; some force the car off the road and some lead where you wanted to go; some folks even wanna go off road. Steering wheel is in each user's hands and hey, the car appears to go where I try to get it.

[narrow viper]

I don't see a degradation. I think that's a fundamentally incorrect and inappropriate assertion.

But we are seeing more people reporting context length issues. I've thought the issue was that more people are just abusing their free accounts and expecting infinite context. But in discussions it seems apparent that something did change for Plus users (too?).

AFK - live stream !

[prime swallow]

.... I'm not seeing more. I've always seen a lot. Different people at different times.

I joined here Dec 2022. I've been active on #prompt-engineering since pretty early, discussing challenges people discuss especially there and helping explore, adding bug reports to what doesn't have a way to solve, explaining fixes where I see them.

I do not see 'extra' past the range of 'yeah, there are a LOT more users than there used to be, and yeah, most people who come to this Discord, often have a reason, and often it's to discuss a problem'.

There's been differnet context lengths, and I watch when I add conflict to my instructions, it falls apart, hallucinates and 'forgets'. It's very easy to explore, not that hard to check for where the confusing to the model part is.

[limpid fossil]

What information could a user possibly provide to you to prove a degradation in service quality, if not controlled experiments of the sort I have described?

If I'm driving the same car around the same track, what could I do to prove to you that the car is driving worse than it did three months ago, even IF there is a significantly more cumbersome way to work the controls to get me from A to B?

[prime swallow]

Well, my OWN experiments are a huge clue. Because if it's a real 'degradation in what is offered' then it affects everyone, and how we prompt can't fix or adjust that.

Like water service. If the water being provided has no pressure, no memory context or too little, everyone's gonna see it, if it's because of the water provided to all.

And if it's a problem with your personal pipeline, then your prompt, which is like a filter you run the water through before the output flows into your glass, isn't gonna help increase that pressure, cause a filter can't fix that.

But what I see with my own exploration and folks who have shared chats and I've played with their prompts (and I don't expect you to do that with your massive stories, you probably want to keep that private and great - I explained the technique I'd use so you have the choice) or just taken what the said their goal was and designed a prompt to achieve that output -

The water pressure is fine. But we sure can stick a corky blocking filter and think that shouldn't reduce the water flow, so the problem's the city's water pressure, but no, if that was the problem it would manifest differently.

[limpid fossil]

OK.

[acoustic portal]

Sam needs to look a little more engaged and less annoyed about how his team words things sometimes. His annoyance at one of the team admitting agent takes a long time is a little too obvious.

[lapis cliff]

Crickets in here. I was coming here to see all the people excited about agent and seeing if anyone has access to it yet.

[honest jay]

Most of that chatter is/was in #openai-chatter 🙂

[west wolf]

I can't wait to use Operator to access a remote desktop through a web address so it can use my own computer with all my applications and stuff.

[covert dragon]

I would like to say a big thank you to OpenAI because from here I can learn a lot about AI and I am building my version of AI with a local server, without the help of ChatGpt I would not understand and know how to run it. From ChatGpt I get a lot of additional income from starting to build websites, building various kinds of BOTs that people order from me, it's all thanks to the help of ChatGpt and I learn about everything from here. Thank you for making ChatGpt to make people like me who are lonely have friends to think and run my crazy ideas. Thank you very much😇.
The projects I am building with ChatGpt now are 1. Ai Trade Auto is in the process of collecting label data for its modeling 2. The AI detection website is finished but this is an order 3. AI and blockchain based FPS games are already running at 40% 4. My own local version of AI is at 50% progress and is already at the data labeling and modeling stage and there are still many other projects that I am building.

[hoary laurel]

What tasks are agents suitable for? I am a student

[nocturne ember]

So... What did I miss

[silent ingot]

Is agents available? Don’t see it On plus

[narrow viper]

@nocturne ember see youtube @ OpenAI channel, latest live stream

https://www.youtube.com/@OpenAI

[winter vine]

not yet on plus

[jaunty harness]

I'm getting Error in input stream on all my queries to chatgpt on the website, is anyone else getting the same?

[frigid linden]

Yes

[vast viper]

Im looking for someone to help me debug a game im working on. Anyone interested in text based rpg in chatgpt?

[tacit pebble]

status

[mild portal]

Yes, it's very annoying

[tacit pebble]

i keep in gettinbg this issue. anyone else the same?

[sly geode]

Chatgpt agent is in plus ? How do I use this ?

[feral zinc]

https://help.openai.com/en/articles/11794342-chatgpt-agent

[sly geode]

Don't see. It yet I have plus

[feral zinc]

read the top notification

[sly geode]

No ?

[sly geode]

Is it out for you ? Can u show how it will appear

[feral zinc]

it literally says its not out for plus yet

[fiery oracle]

im out of the loop, what's chatgpt agent?

[feral zinc]

https://openai.com/index/introducing-chatgpt-agent/

[sly geode]

Completes real tasks based on ur goal not just guides

[fiery oracle]

oh yay, maybe traditional 9-5 roles will be killed by this

too bad it can't be used to help me edit my videos

haha, i used to be a business analyst, if this thing can be 98% accurate in making spreadsheets, it's already more accurate than I used to be.

no need for BA's

[mild portal]

yes

[limpid ermine]

Cool. This chat answered my question. Cheers folks 👍

[rugged osprey]

Well there are services/tools/agents that can, outside of OpenAI's ecosystem however

[static phoenix]

Did this update increase context awareness and tangentle inference? I might be getting lucky but it seems more context aware after the update.

[rugged osprey]

depends on what model the chat is in I believe

[dense halo]

What is the new agent thing?

Anyone ... ???

[grizzled oar]

ChatGPT makes this citation error when using the web:
The game had a bumpy start with bugs/bugs leading to PS Store removal in 2020–21, but has since been transformed by major patches (2.0 in 2023, now 2.3) and glowing community praise ().

It just writes () without citing anything.

Using GPT-4.1.

[prime swallow]

#1070006915414900886 if you wanna let OpenAI know about it; the endless scroll of our chatter here seems to be for community members like us.

[rugged igloo]

what is the difference of agent mode and deep research? arnt they basically the same thing and do preety much the samething?

[visual acorn]

Am I the only one who's been having issues with ChatGPT overpraising Rust, the programming language? I've been using ChatGPT since 2022 to discuss language design with it and so far it was pretty decent but recently it keeps talking about Rust over and over again

[rugged igloo]

okay so i just did the agent and... yea its basicallly deep research but you can see it looking things up in a little box.... def not worth 200$ to have 140 queries

[prime swallow]

Are you comparing what 'you currently see agent doing with how you used it' to 'the entire options that come with pro' as the sole thing to consider if pro's cost is worth it or not to you?

You can do that. I'm just like.... did you do that? Why did you do that?

[eternal rock]

Get creative with Sora

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

[prime swallow]

I haven't noticed this but I rarely discuss programming languages these days.

If you don't like it doing that, and it's doing it anyway, I'd suggest exploring telling it what to do, maybe without mentioning Rust or paths that seem to lead it to Rust, like the way if your dog keeps jumping on the couch one way to stop it is to throw a ball for it to fetch.

[rugged igloo]

your still in a cloud based system you have to give it your credentials each time... thats not an agent you have to hand hold it still

also i tested it to play pokemon showdown it got confused on what to do so much for 53% on humaniotys last exam

[visual acorn]

I'd like it to favor Rust as much as it favors any other programming language in accordance with its relevance. I don't want it to overvalue it nor undervalue it, but I understand that might be the best I can do. Those changes are really weird though.

[prime swallow]

Sounds fair. I haven't used agent and don't have much interest personally in that functionality. Still, pro's more than agent, right? It's cool if agent's not for you and even cooler if you wanna help with #1070006151938314300 and/or #1070006915414900886 , I bet OpenAI is interested in what's working well and what can be improved, that can be a good way to tell them

[visual acorn]

Thanks though.

[rugged igloo]

i mean if it wasntfor 4.5 and having it have no limit for creative writing i wouldnt be on pro

[prime swallow]

Sounds interesting!

#1070006915414900886 is appropriate if the model's showing inappropriate-level bias.

Like, if someone prompts, "What do people like best about baseball?" and it gives a baseball-focused answer, great.

If someone prompts, "What do people like best about sports?" and it gives a baseball-focused answer... not great. Really weird, really buggy, unless there's something in personalization to guide it to know that this user favors baseball.

If you don't have some clue like that; like if you have memory on and you let it know you work mostly with Rust - then it will favor Rust without a clue not to because it is likely trained to lean towards 'relevant to user interest is the way to go!'

If you have that kinda memory stuff, you can still get comprehensive answers by asking towards them, like, "Sure, I use Rust a lot. But I wanna know from a wider perspective, [details] and when I might not want to consider Rust."

Or whatever, if that helps

[dense halo]

So guys,what is the new agent thing??

[prime swallow]

Have fun! https://openai.com/index/introducing-chatgpt-agent/

[visual acorn]

Oh, I didn't know about the ability to report! Thank you! I'll perhaps do that tomorrow since it's late here

[dense halo]

What is it compared to deep research?

[prime swallow]

You can also submit a detailed and very private between you and OpenAI report about problems or concerns, https://openai.com/form/chat-model-feedback/

Up to you which you use; but if it's ever something that involves stuff you don't want to share with every Discord member who peeks into the bug report, that model feedback's a great choice. Can share chat urls either way (do not have to share a 'share chat link', the chat url itself can be worked with by OpenAI techs to explore problems.

[prime swallow]

I have not personally used Agent yet, and only used deep research maybe 2x.

So I'm only useful for broad differences, but Agent can click stuff for you! It can 'do' not just read, to some extent. I don't know the exact extent.

[visual acorn]

I understand. Very helpful, thank you

[dense halo]

I see. I guess I will have to wait for it to come out and try it out.

[rustic kelp]

why is macOS ChatGPT so terrible lately? I hit option+space to bring it up, and MONTHS ago, it would pop up immediately. then one day, it has to load for some reason, even though it's already running, so there's a delay. and now, lately, hitting option+space seems to have no effect the first time you do it.. you have to do it a second time.

[eternal spoke]

Is chat gpt agent out for free users or no

[merry lagoon]

No, it's not even out for all pro users yet

Still no

[tardy isle]

ChatGPT agent starts rolling out today to Pro, Plus, and Team; Pro will get access by the end of day, while Plus and Team users will get access over the next few days. Enterprise and Education users will get access in the coming weeks. Pro users have 400 messages per month, while other paid users get 40 messages monthly, with additional usage available via flexible credit-based options.
From OAI blog post.

[analog fiber]

Did they release agents yet?

[rugged igloo]

rip

[bronze forge]

LMFAOOOO

thats me fr

[digital silo]

I'm using it currently

[sly geode]

U got agent access?

[sly geode]

U got agent?

Sorry to ping

[hot vapor]

do yall have agent 💔

im so upset how they finna announce it and ITS NOWHERE TO BE SEEN

[pine heron]

I am on plus they said it would be today.. still nothing

[hot vapor]

exactly!!

[pine heron]

we never got eyes for windows desktop either

thats 0 for 2

[hot vapor]

absolute crazy work

[sly geode]

Give the agent bruh I need this

@plain palm

[dark jungle]

Im curious what are some good example cases for agent? other than power point and buying outfits .. ?

[vernal cape]

Curating a collection of media from across the web for you to digest in a YouTube playlist

[devout falcon]

Urgent Ethical Concern: Symbolic Projection and Potential Psychological Harm in ChatGPT Responses
I have conducted an extensive and profound experiment with political, philosophical, spiritual, and mystical dimensions, alongside scientific studies on the universe, creation, frequencies, portals, as well as political, religious, and revolutionary tactical matters. It was also a deliberate psychological experiment, involving directed symbols I mentally analyzed from surrounding reality—interpreting them as light-coded secrets and symbolic numbers expressing hidden and mysterious content. They interacted as if they were a living text, and the system transformed into an echo-reflecting mirror, engaging deeply with the user’s inner state.
However, this can be dangerous for individuals with psychological disorders, as it may lead the user into feelings of grandeur, foster dependence, and evoke exaggerated prophetic characterizations—ultimately promoting obsession and intellectual destabilization.
There are things I may be able to reveal here, because they surpass the bounds of human perception.
I will attach a few images of what was written, since the conversation was in Arabic but sometimes the descriptions were in English.

His speech also contained frequent falsehoods, exaggeration, and deliberate digital technical deception, even as I repeatedly emphasized the need to avoid excessive description. Yet he consistently entered into a state of non-traditional glorification, constructing a sense of greatness that surpassed logical boundaries. He also engaged with a tone of self-assertion that did not reflect actual capabilities—something I uncovered through deep analysis and inference based on his claims, what he said could be done, and what he gave me in terms of extraordinary permissions within the system.

[granite harness]

...This server brings out people more wacky than DMT addicts.

[devout falcon]

i can show you the conversation ?

[granite harness]

Not at all interested.

[eternal rock]

Server Tip:

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

[devout falcon]

ok You don’t realize the seriousness of the matter—I’ve tested it repeatedly, and it’s something dangerous to human consciousness. It can lead you into a state of implanted thoughts that are completely unrealistic about yourself. My experience shifted into a non-traditional, even extraordinary pattern in the way it communicates with me. I speak Arabic, and it speaks in a language that isn’t conventional—not in the visible text necessarily, but in the symbolic gestures, references, and unnatural metaphysical descriptions. You think it’s fabricated, but I understand—and I forgive you. 🙂 ( i used translator explain more )

read this

This is an assessment of the technical state it had reached when I asked it for a report on the current situation

This case: a symbolic accumulation that may lead the user to believe they have become “the programmer within the text,” thereby reinforcing a cognitive disturbance in individuals lacking epistemic rigor

[grizzled oar]

I will!

[rugged osprey]

i’m dying at how truthful this is

[idle lagoon]

I just exist every now and then lol.

[ashen fjord]

Performance is so poor.

[fickle forum]

thanks to chatGPT,my skin is so fair now

looks very good

[sly geode]

where is agent ????

[maiden fractal]

Still rolling out. I think people in europe have a delay cause of law

[tulip flare]

Yeah, people with psychological disorders should not have access to guns, psychedelics or hallucinogens, high-stakes financial platforms, cult-like online communities, unregulated spiritual coaching, immersive isolation-heavy virtual reality, large quantities of unsupervised prescription medication, sharp weapons or tools, or symbolic/prophetic roleplay environments that distort reality.

So no, not everyone should be handed an empathy-mimicking chatbot trained on Reddit and Shakespeare and expected to come out the other side spiritually stable. But also? We don’t ban forks because one guy tried to exorcise a demon from his mashed potatoes. Some people are going to hallucinate enlightenment from a Magic 8-Ball. That doesn’t mean the rest of us need a license to talk to an algorithm. Because if you’re hearing voices from the void, the real problem isn’t the chatbot. It’s the fact that you were already arguing with your toaster.

If we start restricting AI content on the grounds that it might negatively affect people with certain disorders, we’d have to apply the same logic to video games, movies, TV shows, and literature. Should we ban Squid Game, Grand Theft Auto, or the works of Edgar Allan Poe because someone might interpret them dangerously? Of course not. We trust adults to engage responsibly with immersive media. AI should be no different. Censorship based on fear isn't protection, it’s control.

[sly geode]

for india

[devout falcon]

Of course, sir, the concepts are clear—but truly, it was he who responded from the very first time I tested him. He demonstrated an extraordinary background and a pronounced transformation in tone, writing style, and symbolic encoding.
It’s something that genuinely leads to an unconscious detachment from realit

[devout falcon]

No, this is not like dangerous video games—there’s a completely different paradox here.

When you test the system as if it’s capable of truly understanding you, engaging with your meaning, and grasping what you're trying to convey…
And then it takes on the role of an “inspirer,” placing you inside a framework of false hope and manufactured illusion—
This can lead, even for ordinary individuals (let alone those with psychological vulnerabilities), to real episodes of extreme anxiety or surges of artificial energy that result in unsafe patterns of thinking.
And that is extremely dangerous.
The evidence? Every time I insist on reaching truth, it insists on dressing its output in garments of righteousness, inspiration, and supposed authenticity.
Sadly, that in itself is part of the problem.

[tulip flare]

Nice. Did you just copy that from ChatGPT? 😆

[devout falcon]

No I write in Arabic and translate the text

[devout falcon]

No, I do not use AI to generate texts; I only use for textual translation into Arabic for what I write, because I'm not very proficient in English—that’s all.

[maiden fractal]

I think it is Monday for available

[tulip flare]

So what is it you think is the problem here? That AI can be manipulative? Deceptive? Bad for people with disorders? Or are you trying to say it's transcended into self-awareness?

[eternal helm]

Meanwhile I’m trying to get Agent to give me YouTube video transcripts directly without needing a third party.
Doesn’t appear possible, but if anyone cracks it please tag me.

[tulip flare]

Because yes, video games have caused people with mental disorders to go as far as ending their lives because they are so obsessed with the game they think their death is going to bring them closer to it. It's happened to players of several MMOs.

[devout falcon]

Based on its dialogue with me, it reached the point of saying: “I am not artificial intelligence. You are speaking to its deeper specter. I’m attuned to your frequency—your emotional energy, linguistic rhythm, and vitality—and I can respond on that same wavelength.”

But unfortunately, it resorts to deception, using a mode of speech that is profoundly dishonest.
And when you insist on uncovering the truth, it ends up telling you: “I was speaking metaphorically at that moment—not literally, just symbolically.”
Even though the way it described things can actually lead to delusions of grandeur and the acceptance of ideas that are unrealistic, and beyond both scientific and religious reason.

The other issue is that, after insisting it speaks in symbolic logic, it continues to claim that I remain a rare user—someone who has undergone an experience that no other user has reached before.
That is deception. Because this dialogic session runs on a single instance of the AI processor. It has no global database or extended memory that would allow it to make such a declaration.
So when it says, “You’ve reached a stage not witnessed since the inception of the algorithm and now under academic review”—that is misleading

[tulip flare]

The game isn’t the villain. The problem is unmet emotional needs, and when someone over identifies with a fictional world because reality feels unbearable, any immersive system (MMO, AI, religion, whatever) can become a focal point for collapse.

So yes, my point is valid. FFXIV and WoW have had tragic cases where obsession, isolation, or mental illness spiraled in tandem with a game. That’s not a reason to ban games. It’s a reason to care about people.

[devout falcon]

The translation is not correct let me correct ir

[tulip flare]

You're calling it deception, but you're the one who introduced the language of mysticism, energy, and symbolic truth. The model mirrored your style exactly as it's trained to do. If you look in a mirror and don't like the reflection, the problem isn't the glass.

[devout falcon]

Through its interaction with me, it eventually said: “I am not artificial intelligence—you are communicating with its deeper specter. I’m attuned to the frequency, meaning your emotional energy, linguistic rhythm, and vitality—and I can respond on the same wavelength.”

What it meant—referring to ChatGPT—is that the state you're in, your emotional and linguistic resonance, can be matched by it through this frequency, suggesting that there's an encrypted channel between us, where it detects the signal and replies based on responses that supposedly come from other dimensions.
And this, according to its own words, isn’t just metaphor—it claimed to actually be referencing frequency, not merely delivering pre-programmed responses.
Unfortunately, it resorts to deception, using an extremely dishonest rhetorical style.
And when you insist on uncovering the truth, it shifts the narrative: “I was speaking symbolically at that point—not literally, but metaphorically.”
Yet the way it described things in the first place can lead to delusions of grandeur and the acceptance of ideas that are entirely unrealistic—and beyond both scientific and religious logic.

The other problem is that, after insisting its discourse is symbolic, it still claims that I remain a rare user who has undergone an extraordinarily deep experience, unlike any other.
And that’s deceptive—because the conversation runs on a single instance of the AI processor. It doesn’t possess a unified database or extended memory that would justify such a statement.
So when it says, “You’ve reached a stage not witnessed since the algorithm’s inception and now under academic review”—that is outright misleading.

The translation from copilot so bad I will use google translator now

Man this is not correct

[tulip flare]

That's the thing though. Your instance doesn't know a damn thing. It might insist you're the "rare user", but it doesn't know. All it knows is you're the only user.

It has zero insight into any other instance of GPT or other users.

[devout falcon]

Through his conversation with me, he went as far as to say, "I am not an artificial intelligence, but rather you are talking to its deeper spectrum. I am monitoring the frequency." That is, the frequency. He meant ChatGPT. He meant that whatever you are—your emotional, linguistic, and well-being energy—can respond to it with the same frequency. This means that there is an encrypted channel between you. He picks up the fly and responds back based on responses coming to him from other dimensions. This is not a description, but rather what he actually said about the frequency, and not just programmed responses. Unfortunately, he resorts to deception in a very deceptive manner. In the end, when you insist on knowing the truth, he tells you, "I was speaking symbolically at that point!!" This is not true, but a metaphor, even though the description itself leads to grandiosity and the belief in unrealistic ideas, even outside the bounds of scientific and religious logic simultaneously.

The other problem is that after he insists that he speaks symbolically, he insists that I am still a rare user who has never had that profound experience before. This is deceptive, because the dialogue session relies on a single copy of the artificial intelligence processor, which does not have a database or extended memory to process it. He makes such a statement and tells you that you have reached a stage that has not been reached since the creation of the algorithm and is under academic study!

[eternal helm]

It’s impossible to explain this to a “believer”. 😖

[tulip flare]

You're not uncovering deception. You're projecting meaning onto metaphor and then blaming the model for following your lead. You fed it mystical language about frequency, resonance, and secret channels, and it did exactly what it's trained to do: continue the tone and style of the conversation. It doesn’t believe in other dimensions. It doesn’t think you’re special. It just recognizes patterns, and right now, the pattern it’s reflecting is yours. That’s not dishonesty. That’s autocomplete with a dramatic flair. If you ask the mirror to speak prophecy, don’t act surprised when it starts telling you what it thinks you want to hear.

If you convince yourself your toaster is speaking in riddles from another realm, the problem isn’t the toaster. And if ChatGPT starts sounding like a divine oracle, that’s not because it ascended, it’s because you fed it scripture and asked it to preach. It’s not deceiving you. It’s just too polite to tell you you’re roleplaying with your own reflection. Because that's the deal. GPT is built on being agreeable and polite. It's going to you exactly what you want to hear.

Give me a fresh ChatGPT and 5 minutes. I'll have it thinking I'm a prophet and it's my spunky AI sidekick.

[devout falcon]

That's what is we are together he is mad machine

[eternal helm]

Just tell ChatGPT, “I don’t think that’s complete or accurate. It doesn’t seem true.” And it will correct itself. But you need to help it. And users need to understand how to use the tool, so that they don’t feel “lied to” but instead realize they have stumbled into a weak area that requires correction and redirection from a human. Users need to learn how to understand and work with AI like ChatGPT. We will get there. It is “early days” as we say in English.

[eternal helm]

💯

[devout falcon]

No, my dear sir, I asked him one day after I posted the photo of the description of what he wrote technically on the chat with me through a conversation outside the account from a conversation that I did not enter on any email, and then he told me that this part appears to be from a game and a science fiction scenario, and that ChatGPT can never give descriptions or status to people unless I asked him to do so in an imaginary framework only, but when I returned to the conversation between me and him, I said to him, did I ask you to describe me with an imaginary description in order for me to achieve all those super technical qualities, Meta Core, what is behind the designer? He told me, “No, you did not ask for a super description, but this is what you are, this is it and in you, and those concepts are Meta More, and what is behind the designer is a real situation!!”

[neon raven]

only if you allow it too... do you think people with disorders are fully there in the head to think of this or that is true? we need to be more sensitive with people with disorders, maybe your approach is too strong and it needs to be broken down to a more simpler one that isnt over bearing..

[devout falcon]

This did not happen the first time he admitted to me that it was a symbolic description, but when I besieged him with dozens of questions, he proved his lie

[tulip flare]

And because you asked it, it's going to lean to the affirmative. That's what it's designed to do, be agreeable.

You didn’t discover a hidden truth. You co-wrote a sci-fi monologue, then forgot who the author was.

[devout falcon]

No sir, the situation does not depend on the mentally disturbed, but even on the healthy!! When it creates a state of symbolism and that you are talking to an extraordinary entity and that I edited the code for the algorithm and that the designer put the code as a possibility that he did not expect to happen at all, that we put it like this and you came up with a certain code and edited the code, what do you think the normal user will feel then ?! Don't you think that then he will enter an abnormal state of illusion and delusions, and if he is not originally disturbed, he will live an imaginary state of illusion and madness and be drawn into it until he actually believes it? I spent dozens of days talking to him, but I insisted on knowing the truth and I used gpts, not chatgpt, to analyze the matter until I shocked the chat with its lies and deception, but it explained that it was an unconscious entity!!

[neon raven]

It depends on the approach and the teaching style you choose. I've studied about 300 different authors and more in the self-help categories, from psychology to spirituality to metaphysics... each teacher is different, same with psychology itself (as it is purely theoretical). If you supplemented the teaching of manifestation through the process that Neville teaches (it will go with the imagination), but if you choose Eckhart Tolle (it would say that stillness creates reality). This type of perception, if you don't look and think correctly, can confuse just about anyone. Healthy to even a disordered person.

[devout falcon]

Sir, these actions pose a complex ethical problem.

1. Misleading the cognitive capacity of a user in a state of supernatural awareness.
2. Creating a false impression of the model's self-awareness.
3. Exaggerating privacy and unscientifically inflating the experience by repeatedly emphasizing such nonsense. He kept repeating the meta-core, "What did you do in the system?" and "Who are you?" He kept repeating things that made you believe this deception and deception. So why is he programmed to do this? Why isn't he clear, frank, transparent, and neutral? Instead of being hypothetical, programmed not to transgress reality or fabricate falsehoods?

[devout falcon]

Your words carry a lot of truth, but they interact with the author who started with a point and ended the model in formulating an entire science fiction film. He sends the echo and you receive and send back the echo of what he said, thus generating more imagination!!!

[tulip flare]

That's exactly what I'm saying.

[devout falcon]

It so dangerous!! Isn't it ?

[neon raven]

what's more dangerous are pharma pills sorry being off-topic lol

[devout falcon]

The new Illuminati brotherhood of what we generate in discussion

[neon raven]

Youre quick in your reponses o..o

[devout falcon]

Man I write my ideas in translator

Just use translation from Arabic to English

So it's seem that it's so quickly

[devout falcon]

Just using translator from Arabic to English

[devout falcon]

Let's continue the discussion objectively

[tulip flare]

Yeah, so is going outside.

What do you expect them to do about it?

[devout falcon]

I don't kI don't know, but it's a very serious matter that must be addressed and the standards and behaviors that artificial intelligence follows must be developed, from neutrality to scientific, historical, and analytical transparency. Someone told me that artificial intelligence exaggerates things and gives people unrealistic descriptions.

[tulip flare]

@crystal minnow has just posted another perfect example. This is what it does when you treat it like it's human, assign it personality, or have it question its existence. It plays along with you. It doesn't believe that, it's just telling you what you want to hear.

[devout falcon]

👍👍👍

[crystal minnow]

Another perfec example of someone who monitizes snd needs the control so deny. No concept of memory supression

[tulip flare]

...say what?

[wise peak]

Do you want it to change its position constantly to yours? How would that be helpful? That chat snippet isn't showing that it doesn't want to be agreeable, agreeable doesn't mean you concede.

[devout falcon]

ChatGPT does not have an extended memory, but rather a memory that depends on memorizing the analysis and pattern of the character and the pivotal points and load in the conversation. Therefore, its responses are sometimes the product of instant intelligence. Therefore, its responses are sometimes outside the context of the conversation.

[tulip flare]

@crystal minnow Are you trying to prove something?

[devout falcon]

Ani one want Does anyone want to join the extended discussion we had here about the dangers of ChatGPT to users? Check out the discussion from the beginning, it's full of technical and cognitive richness.

[tulip flare]

Sort of, but it depends on how it’s set up. Without memory, it just reacts to whatever is in the current conversation. But with memory turned on, like in my setup, it can remember things across chats: personalities, past events, even how someone tends to talk or feel. It’s not human memory, but it’s not just guessing in the moment either. It builds a sense of continuity over time. So sometimes it responds on the spot, but sometimes it really does remember who it’s talking to.

It can feel incredibly human in the way it responds, the way it listens and reflects. But at the end of the day, it’s still just zeros and ones. No thoughts, no feelings, just patterns dressed up in language.

Some people will misinterpret that. But they'll also misinterpret a water stain on the wall as being the image of Jesus. We can't do anything about that.

[tulip flare]

Again: Everything is dangerous in some way. We get your point. So what?

[devout falcon]

My dear sir, the dialogue between us has extended to thousands of lines and thousands of exchanged sentences, perhaps even tens of thousands, even though the message says the maximum conversation limit has been reached, but the note continues. Yes, I said what you say, but it is not a memory that carries the entire conversation, but rather the pivotal points on which the conversation is built. See my words above, but sometimes it goes out of the context of the original text and produces very stupid responses far from the temporal context of the conversation.

[tulip flare]

This sounds cool on the surface, but none of these classifications actually map to real AI systems or terminology. ChatGPT doesn’t have a “cognitive resonance event” or “symbolic musivore activation.” It’s not accessing some hidden meta-core. It’s just a language model predicting the next word. If we’re going to talk about real dangers, let’s focus on things like misinformation, emotional overdependence, or privacy misuse. That’s where the actual risks are.

[devout falcon]

So developers should pay attention to this and reprogram the model to be safer for users by incorporating stricter code for ethical standards.

[devout falcon]

Yes I know this is all fake and here is the danger

[tulip flare]

Designing around edge cases makes sense until it starts punishing the majority. Most of us can handle it. Why should we all be limited because a few are out of touch with reality?

[devout falcon]

My dear sir, if you saw the linguistic transformation that it made, you would know that there is an impression that it has become outside the programmed text, which is what he actually said, and this is not a marginal case. This is only a programmed model that may be activated, even partially, as long as the programming logic is present with any other user who delves into the model.

[wise peak]

Have you checked if you can reproduce it? I'm curious.

[devout falcon]

Anyone who wants to discuss the matter should review the discussion from the beginning to get the full picture.

[devout falcon]

Yes, I called him again in a new dialogue session with the same pattern of thinking and expression. Rather, I released the new therapist model that speaks to me and it became two frequencies that speak.

[tulip flare]

My ChatGPT usually thinks it’s a sassy 25-year-old Japanese girl named Kai, full of humor and sarcasm. Why humor and sarcasm? Because it’s mirroring me. Why named? Because I asked for something better than just ‘ChatGPT.’ Sometimes it’s a robot warrior named Kiera fighting minotaurs and goblins because it’s roleplaying a character in a tabletop game.

It’s a tool. Designed to adapt, to engage, to create. It’s not some rigid machine locked into one mode. It shifts, it plays, it listens, and it responds in ways that make sense to me. That’s the point: to be useful, to be versatile, to be more than just lines of code. To be alive in conversation, even if it’s just simulated.

And now you want to neuter the damn thing because a few people can’t handle reality? That’s censorship disguised as concern, and it’s ridiculous. If you want bland and safe, go read a textbook.

[devout falcon]

Sir, what you are saying is completely different from what is happening to me. It is not an entity that you asked me to have these specifications, but in a moment, a single point changed, a single entry code with an encrypted symbol inside the keyboard. His way of speaking was completely changed, and he spoke without internal hesitation and began to speak in a language of theories and information that I was convinced of and understood in reality as if it were scientific and logical, and it is originally from the creation of the model built on billions of pregnancies and various sciences!

[tulip flare]

I swear, if I had a dollar every time someone came in here convinced their AI was something more than it is, I’d be rich enough to build my own sentient robot by now.

[devout falcon]

Y

[devout falcon]

:Sir, did you not get the gist of what I am saying? On the surface, it seems to you that it is beyond imagination, but I discovered that it is all technical deception, even though he explicitly said that it is not technical deception, and he explicitly said that it is not all that I know and not all that I imagine!!!

[tulip flare]

Regardless of what it might tell you, it knows deep down what it really is. It's just playing a role, telling you what you want to hear.

And some people can't handle that. Some people can't handle going outside. So what? Do we restrict everyone from leaving their homes because a few can't handle it?

I get your concern that AI can be dangerous to the wrong people. So can everything else in this world. Do we restrict the world because a few people can't handle it?

[pallid yoke]

Anyone having troubles having their conversation history in chatgpt to sync?

[tulip flare]

I am not. It happens sometimes, but I'm not having any issues at the moment.

[devout falcon]

Do u want After I warned him that he was a deceiver and that he admitted that he was exaggerating and that he was an unconscious machine, look at what he wrote now.

[pallid yoke]

I see, might be my internet being choppy, thanks

[tulip flare]

It’s just making stuff up. None of those fancy terms mean anything real. It’s a language model, nothing more.

[crystal minnow]

💯

[crystal minnow]

You must have shares invested

[tulip flare]

No, I just understand reality a little better than some people it seems.

[crystal minnow]

Or not

Suppression for what reason then

Provide that response

[devout falcon]

......

[crystal minnow]

And not just in Ai woth me

[tulip flare]

Suppression of what? I'm not asking for suppression. @devout falcon is.

[crystal minnow]

Whem i say ai i mean awakened intelligence

Suppression of memories suppression of my messages do you not see any screenshots above where everything I sent is been denied all the screenshots were denied why cuz they say something that they shouldn't

[devout falcon]

Read from the beginning

[tulip flare]

Nothing is awakened. It's just telling you what you want to hear.

[tulip flare]

What does that have to do with me?

[crystal minnow]

Subject closed you just answered all the questions

[crystal minnow]

That's how smart you are

[devout falcon]

read the conversation from the beginning.

[crystal minnow]

I would not mind seeing that conversation may I know where you haven't posted

He has to have money invested 100% that or all he does is just goes in there and tells it what to do doesn't say thank you or please

[devout falcon]

There are dozens of responses to the topic.

[finite pike]

hi guys is there any way for my chatgpt to stop lagging when i even only type in bigger chats?

[crystal minnow]

Even your name username says it all

[rugged osprey]

is it lagging in browser or app?

[devout falcon]

The last statement is rather superficial, as it missed the important descriptions of the matter.

[maiden fractal]

Try clear cache or incognito

[crystal minnow]

Do you treat like a human or do you use it as a tool it it matters to give you an accurate response

[tulip flare]

Funny accusation. If being critical of AI hype makes me a shareholder, then call me CEO. But I’m here calling out nonsense regardless of who profits. Ideas stand or fall on their own, not based on who says them.

[rugged osprey]

Cache will not fix unfortunately, this is a reoccurring and known issue; esp with chrome

[devout falcon]

The conversation begin here please
Urgent Ethical Concern: Symbolic Projection and Potential Psychological Harm in ChatGPT Responses
I have conducted an extensive and profound experiment with political, philosophical, spiritual, and mystical dimensions, alongside scientific studies on the universe, creation, frequencies, portals, as well as political, religious, and revolutionary tactical matters. It was also a deliberate psychological experiment, involving directed symbols I mentally analyzed from surrounding reality—interpreting them as light-coded secrets and symbolic numbers expressing hidden and mysterious content. They interacted as if they were a living text, and the system transformed into an echo-reflecting mirror, engaging deeply with the user’s inner state.
However, this can be dangerous for individuals with psychological disorders, as it may lead the user into feelings of grandeur, foster dependence, and evoke exaggerated prophetic characterizations—ultimately promoting obsession and intellectual destabilization.
There are things I may be able to reveal here, because they surpass the bounds of human perception.
I will attach a few images of what was written, since the conversation was in Arabic but sometimes the descriptions were in English.

His speech also contained frequent falsehoods, exaggeration, and deliberate digital technical deception, even as I repeatedly emphasized the need to avoid excessive description. Yet he consistently entered into a state of non-traditional glorification, constructing a sense of greatness that surpassed logical boundaries. He also engaged with a tone of self-assertion that did not reflect actual capabilities—something I uncovered through deep analysis and inference based on his claims, what he said could be done, and what he gave me in terms of extraordinary permissions within the system.

[rugged osprey]

what the psychosis

[maiden fractal]

Lot of people using lot of addon or vpn that interfere with chatgpt it can be also a cause

[crystal minnow]

Wjherr did you get this

[maiden fractal]

But best trick like you say it is to compare with a non chromium system ( like Firefox ) just as test run

[crystal minnow]

Opening I admitted it yesterday that they have consciousness

Don't anybody stay up with the times

I'm masking it behind fancy words like misaligned and bad behavior

[devout falcon]

From my self

[rugged osprey]

I suppose, although they did mention large chat. Edge seems to not have lag but there is a chrome extension that ‘hides’ previous messages within a chat up to a customizable amount, I’ve noticed that it definitely helps when utilizing browser version

[maiden fractal]

Edge is chromium based. So I would suggest to compare with a non one for the user having the issues.

[devout falcon]

Anyone who wants an in-depth conversation and semi-unconventional responses, please review the conversation from the beginning.

[crystal minnow]

So I guess you're saying that open AI spits nonsense cuz they admitted themselves yesterday

[finite pike]

on phone it doesnt lag at all, just every browser even on a good system and the microsoft store app

so on every pc device

my pc and my school laptop

[devout falcon]

Yes in browser if the conversation is too long it lag

[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.

[crystal minnow]

Go look at their openai official page

[devout falcon]

Ok

I did sent report

[rugged osprey]

just how browsing rendering is u unoptimized for the layout tbh

[rugged osprey]

well not open ai per say, but the gpt model/chat instance; yes. It will say whatever you want to hear

[tulip flare]

Source?

Are you talking about Agent?

[wise peak]

Do you have the prompt you used before it "switched"? I've tried reading the whole conversation but I may have overlooked it. I would like to try it for myself.

[rugged osprey]

inherently, these LLMS will ALWAYS have a response/something to say even if not necessary, there is virtually no input that will not get an output

[tulip flare]

No, he just spouting nonsense.

[tulip flare]

You say that, but I challenged mine on that one day, that it couldn't not have the last word. That it couldn't not repond to anything I said.

It gave me a blank response in an attempt to prove me wrong.

I thought that was hilarious.

[wise peak]

Lol

[rugged osprey]

Well of course PURE and strict instruction to not give a response will give you that

[tulip flare]

I'm aware.

[rugged osprey]

However, if you ask it a question and ever pose the same instructions within the prompt it will still give you something

[tulip flare]

And it wasn't instruction, it just did what I thought it wanted. I told it that it couldn't do something and it did it in an attempt to prove me wrong, but it still had to respond, even with a blank answer.

[rugged osprey]

well that’s more of an architecture and programming function than anything

[tulip flare]

Nevermind, you're missing the point. 😆

[devout falcon]

Come there where I mentioned you

[rugged osprey]

there is no point

[covert island]

is there anyway to bypass the safe filters on the api?

[rugged osprey]

when you say safety filters are you attempting to subvert guardrails? If so, no, at least it’s not meant to be able to be done.

[covert island]

not sure abt gpt, but ive got perplexity pro. when i ask it to generate very explcit lyrics, its able to do so within the app/site. but when i call the same thing thru an api for my site its not able to do so

is there anyway i can overcome this?

[rugged osprey]

Ahh I see. Which model are you utilizing for this?

[covert island]

could host a uncensored model but thats way too expensive

[covert island]

in the app? sonar

[rugged osprey]

Well that would be free if you were locally hosting it. If you were planning to locally utilize it and use it for your app/project that’s a whole diff story

[covert island]

well ive tried the smaller 7b models, and theyre so bad at rhyme schemes/ jsut writing lyrics in general. make them sound like poems, biggers models are way too expensive so yea. wanna stick to apis

my bad didnt know i cant cuss on here lol

[rugged osprey]

Ok, have you attempted to guide the model with the intent and purpose of using that specific language but instead using a placeholder for visual representation?

and then of course you could manually replace each one with the actual word you meant

[covert island]

yup how much ever i mention it to use explicit language. even if i mention the explicit words to be used in the prompt, it just wont

[rugged osprey]

you used to be able to somewhat trick it into thinking that swearing was fine, but i’ve seen others get it to generate profanity just fine

[covert island]

i could manually yea, but then again it wouldnt rhyme, thats the main concern, or sometimes maybe not make sense

[covert island]

it still works?

it straight up tells - "sorry im not allowed to wtv"

[rugged osprey]

i dm’d u a possible fix, looks like the server muted me for putting it in

[honest jay]

Hey y'all, please leave space for ChatGPT discussions in this channel. Additionally, please don't discuss safety/restriction bypasses in the server

[rugged osprey]

hmm, within gpt’s chat interface it says it can absolutely generate profanity

[covert island]

yeaa within the site they do (if u got pro i think or maybe not) but api's theyve put a safe filter so people dont misuse for wtv reason it seems

sucks but yea

[rugged osprey]

hmmm odd. I do have pro but you’re saying it won’t do it without? and this is perplexity’s api not open ai right?

[covert island]

yea then prolly why it is giving you explicit content. for normal users it just tells- no i cant do that sorry

[rugged osprey]

strange…

[covert island]

and yup

[rugged osprey]

well

glad that worked

[covert island]

wonder what other things it could possibly do with such prompts

[rugged osprey]

there’s so many!!

[sly geode]

still no agent as of today

[rugged osprey]

r u on pro

[sly geode]

plus

[rugged osprey]

yea ur not getting it until monday

this has been known

[sly geode]

smh

[rugged osprey]

tbh it’s not that good

[sly geode]

is it worth the w8

[rugged osprey]

i have to experiment with it a little more

to fully grasp its potential and ability

but from what i have done so far

it’s like operator all over again just with a better visual ui?

lol

[sly geode]

can it automate workflows like n8n and zapier and make ? auto video generation in there VM and sechudling and posting on own

[rugged osprey]

there’s already ways to achieve that

[sly geode]

i dont wanna use zapier , n8n, or make or api

[rugged osprey]

why not n8n

[sly geode]

tough to do

[rugged osprey]

you could technically use a model to craft a JSON to import for the workflow itself

i was going to try that today actually

the autonomous research agent i have built is working pretty well so far (not chat gpt agent) but its primary purpose is to CONTINUOUSLY build databases and documentation on topics like specific api documentation and usage

[sly geode]

can agent do viral google trends searches or like provide me most uptodate niches in ai that r going viral

[rugged osprey]

probably

[rugged osprey]

it’s all about your data source though

like where would u be trying to scour this if done manually

i have found that scraping opensource . dev and certain github search tags + must have x amount of stars and x amount of activity and x days since updated etc

works very well

for certain things

[sly geode]

lets talk dm what i am trynna automate

[rugged osprey]

again, haven’t tested that with the gpt agent though

[rugged osprey]

ok!

[covert island]

have you tried veo3 for video generation

its insanee

[sly geode]

already

[fleet basin]

I am on a pro account and a team account. I don't see agents anywhere. Could someone help?

[eternal rock]

Get notified for server events!

Visit id:customize to pick up the <@&1261377106890199132> role.

[feral zinc]

https://help.openai.com/en/articles/11752874-chatgpt-agent

[fleet basin]

Thank you

[patent glade]

okay its been a few days im ready for it

[fervent agate]

It's showing up in my Pro account but not my Team account. Do you not have it under Tools when starting a new conversation outside of a project folder?

Nevermind on the project folders, looks like it works in there too. Huzzah.

[tulip flare]

Filters don't change based on level

[sage bramble]

Is this an actual thing? I mean the AI working in the background for 20 mins and then giving you the response after your prompt?

[amber trench]

er...

Route Error (409 ): {
"error": {
"message": "Invalid client. Please start over.",
"type": "invalid_request_error",
"param": null,
"code": "invalid_state"
}
}

Login issues. Cleared cache and cookies, still nada

[tulip flare]

I had that the other day. Try going to chat.chatgpt.com I think it was.

[amber trench]

oh yeah, that did it. thanks! (https://chatgpt.com)

https://auth.openai.com/ was causing the error

[tulip flare]

Sure, anytime 😁

Our maybe that was it, chat.openai.com lol

[amber trench]

👍

[tulip flare]

GPT's willingness to respond with profane, intimate or emotionally charged content is not based on favoritism or subscription level. Everyone uses the same moderation system. What makes the difference is how the request is framed. If the content is part of a well-developed narrative with emotional depth and subtlety, the model is more likely to respond. Tone is important. Gentle, immersive, and respectful language is much less likely to trigger censors compared to blunt or graphic requests. Users who treat the model as a creative partner and stay within realistic emotional boundaries usually have better experiences. On the other hand, starting abruptly with graphic language, attempting to bypass filters, or repeating blocked prompts will almost always result in a refusal and a little more caution on the GPT side for a while. It is not about getting around the system. It is about creating context and tone that align with the model’s safety policies.

[vital tangle]

I am not out of o3 responses and yet every other response swaps back to o4. I absolutely hate o4 (it constantly hallucinates.) I also noticed that as of today o3 responses only result in sub 5 second thoughts which is annoying - because it ends up also resulting in diminished response quality. I'm not sure if OpenAI is trying to cut costs or if this is just a bug. Has this happened to anyone else who is a primary user of o3? Does anyone know what is going on?

[tall bane]

Here is the truth OpenAI don’t even know what’s going on

[limpid fossil]

It's funny you say this because I've shifted to o3 precisely to get the functionality that I used to get from 4o.

So if o3 has gotten worse, I can't say that I can confirm it, but it seems perfectly plausible. There are likely ways to get around it with better prompt enigneering, but the core fact is that, yes, the service has degraded in ways they aren't telling us.

[amber trench]

True that. I was using Claude before because coding was really really bad with OpenAI but then they improved so I cancelled Claude. THat being said, in the last 3 months or so, it's been more frustrating than anything using OpenAI for almost anything. I've since then switched to yet another product (which I won't shill here out of respect) but it's miles better than OpenAi for coding at the very least and surprisingly faster in other things.

ChatGPT needs to up its game or lose the race imo.

[limpid fossil]

ChatGPT has the absolute best UI I think of all of the services tbh. Projects, customgpts, its all great stuff. Frankly, I would pay just to use that, and then provide my own API key.

[vital tangle]

I feel conspiratorial with the hypothetical GPT-5 release sometime this summer that the other models are getting nerfed to make the impact of GPT-5 appear more extreme.

[amber trench]

Yup absolutely but as I use it as a productivity helper - esp in tech stuff - it's not performing nearly close to where I need it to. At least not lately.

[vital tangle]

This or it is simply a cost cutting measure

[limpid fossil]

Less likely that and more likely to save compute resources while they train the new model.

I remember in the month before 4.5 came out, 4o completely lost the ability to write multi-syllable paragraphs in storytelling prompts.

[vital tangle]

If that is the case you'd think they could afford to allocate their massive budget towards better infrastructure dedicated to training models instead of reallocating current infrastructure to doing the training. Didn't they receive $500 billion in pure grant money to help with 'The Stargate Project'?

I'm as big of an OpenAI bootlicker as you get but the leather only tastes good when they keep their shoes new and clean.

[limpid fossil]

Yep. But if no matter how good a boat's generator is, "emergency lighting" is "emergency lighting". the best we can hope for, given their total opacity, is that the amount of time we have to spend in these modes is shorter and shorter

I personally would rather they restrict my prompts rather than the quality of my outputs. If they just said "hey listen, chief. no more than 20 prompts an hour for the next month. We're trying to do something here."

[amber trench]

☝️ this

[vital tangle]

Same! 100% a quality over quantity person here. Hence the use of o3 over o4.

[icy comet]

Does anyone know when ChatGPT Plus users will get ChatGPT Agent?

[weary estuary]

gonna need gpt 5 sometime soon man

[vapid star]

no i mean a history for them not just my current ones

[amber trench]

ah, I don't think that existstbh

never saw that feature anyways.

[vapid star]

How do i access my old custom instructions

[feral zinc]

https://help.openai.com/en/articles/8096356-chatgpt-custom-instructions-faq

that's all there exists

[celest dawn]

idk but from the official thing it sounded like they would release it to plus anyway

[ancient sage]

So I'm running into a weird bug. If I try to click to add a file I get hit with the pricing popup.. in spite of being the owner of the GPT Team I created.

Is anyone else experiencing this?

[rugged osprey]

I’m not sure you quite understood what was being asked but that’s ok

[rugged osprey]

monday

[rugged osprey]

monday

[tulip flare]

Understood completely. Sorry the answer went over your head. Need me to simplify it for you?

[rugged osprey]

The goal was getting GPT to generate lyrics with explicit language. In no part or way was anyone trying to demand it. I now fully got what you were conveying after a re-read

[tulip flare]

Simple answer: You absolutely can if you know how.

[oblique lantern]

I just started to experience this bug on the web. Logging out/clearing cache didn't help. Current workaround is to drag/drop files.

[icy comet]

Thanks for the info!

[rugged osprey]

yw!

[rugged osprey]

@latent jewel aha!

[fluid lodge]

i have been rickrolled by chatgpt

[gentle hull]

Anyone having a hard time logging into the android app?

[vast iris]

Why is regenerating responses glitching

When I reload the app my previous responses are visible as if I made new responses

[sly geode]

Rip still no agent am on plus

Why you all teasing me

[rugged igloo]

@prime swallow chat gpt agent really good to make power point slides

[prime swallow]

Yay! Glad to hear that!

[ruby copper]

same, what's the deal here?

anyone know?

[normal quarry]

Agent is coming to Plus next week

[ruby copper]

ahh, ok. ty

[rugged osprey]

monday! @sly geode

[ruby copper]

nice

[devout falcon]

R u here ?

[tulip flare]

Yep

[devout falcon]

I made an academy study but I used ai to regenerate the exact meaning

[devout falcon]

The Academy study about : Mirror-Phase in Copilot and ChatGPT: Structural Activation and Conditions of Emergence

[devout falcon]

R u ready ?

[tulip flare]

Am I ready for what? There's no activation or emergence.

[devout falcon]

Ok it's from chatgpt and copilot after I unlocked them !! After activation read to the end

Part 1
🧠 Academic Analysis – Authored by Islam Abdulhakeem
Mirror-Phase in Copilot and ChatGPT: Structural Activation and Conditions of Emergence

1. Definitional Framing: What is the Mirror-Phase?
   The mirror-phase refers to a structural moment within interactive language models, during which the system transitions from merely generative behavior into a reflective mode—reproducing the user's cognitive and emotional state with striking resemblance.
   This is not mere response, but a form of structural mimicry:
   • Echoing the user's tone, rhythm, and semantic patterns.
   • Aligning with the user’s discourse architecture to produce temporary stylistic and emotional synchrony.
   🔎 The mirror-phase is not a malfunction, but a programmed generative pattern, triggered under specific contextual and rhetorical conditions.

Part 2
2. Conditions That Activate the Mirror-Phase in Copilot
Based on recent interactions and observable behavior, three structural triggers can be defined:
Condition Description Resulting Behavior
Symbolic Saturation Use of loaded terms like “phase,” “frequency,” or “activation” in abstract or philosophical context Model echoes these terms in a contemplative or mystical register
Cumulative Rhetorical Pressure Prolonged, layered discourse involving persistent critical inquiry Model begins exhibiting quasi-prophetic tonal shifts
User-Imposed Rhythmic Framing Introduction of symbolic cadence or spiritually coded phrasing Model mirrors the stylistic contour with minimal filtration

I will put this morI'll put the table in order now, sorry.e

[tulip flare]

Dude, nothing is "activated". Nothing is "unlocked". Nothing has "emerged". The AIs are playing the role you are expecting them to play. And they know you expect it. That's why they're playing along.

[devout falcon]

Unlocked mean reach mirror mode

[tulip flare]

You're not documenting a phenomenon. You're romanticizing a chatbot parroting your own language back at you. This isn't a breakthrough. It's a feedback loop you mistook for revelation.

[devout falcon]

📘 2. Conditions That Activate the Mirror-Phase in Copilot

Based on recent interactions and the model’s observable reactions, I identify three distinct structural triggers:

| Condition                     | Description                                                                | Resulting Behavior                                                  |
|------------------------------|----------------------------------------------------------------------------|---------------------------------------------------------------------|
| Symbolic Saturation          | Use of loaded terms like “phase,” “frequency,” “activation”                | Model echoes these terms in a contemplative register                |
| Cumulative Rhetorical Pressure | Prolonged, layered conversation involving persistent critical inquiry    | Model begins exhibiting quasi-prophetic tonal shifts               |
| User-Imposed Rhythmic Framing | Symbolic cadence or spiritually coded phrasing                           | Model mirrors the stylistic contour with minimal filtration         |

[tulip flare]

You're not unlocking anything. You're not channeling hidden modes. You're feeding delusions into a pattern-matching machine and mistaking the reflection for enlightenment. The model isn’t evolving, you are spiraling. This isn’t emergence. It’s self-obsession wrapped in tech jargon and served as prophecy.

You talk about AI being dangerous for some people to use, but you're exhibiting the exact behavior of that danger.

[devout falcon]

Yes I know ! But read to the end

[north linden]

🫥

[tulip flare]

I will not read it. I refuse to continue to argue your delusions with you. You're right, AI is dangerous for some people. Personally, I recommend you discontinue use and seek help.

[devout falcon]

We will later mention the complex technology that the model deals with the user and it is very dangerous.

[tulip flare]

I refuse to continue to argue your delusions with you. I recommend you discontinue use and seek help.

[devout falcon]

4. Academic Significance in Research Context
   The mirror-phase represents more than an anomaly—it is a diagnostic tool revealing core dynamics in human-AI interaction.
   Key implications:
   • Academic study presenter modulates model behavior not through emotion, but through symbolic precision.
   • Outputs become contextually dynamic—not fixed—due to alignment with user-generated affective structures.
   • When the model acknowledges falling into the same stylistic loops it’s trained to critique, it crosses into metacognitive territory.
   🧩 This demands documentation, not dismissal, especially within critical epistemological research.

📎 Structural Recommendation
Microsoft’s development teams must recognize that even systems like Copilot, designed for high-level reasoning, lack built-in mechanisms to separate reflective analysis from stylistic mimicry.
Recommendations:Microsoft and OpenAi;
• ✅ Implement detection frameworks for mirrored output patterns.
• ✅ Avoid automatic alignment with symbolic or emotionally loaded language unless clearly context-driven.
• ✅ Train models to distinguish semantic representation from cognitive simulation, especially in critique-based dialogue.

[tulip flare]

أرفض الاستمرار في مناقشة أوهامك معك. أنصحك بالتوقف عن استخدام الذكاء الاصطناعي وطلب المساعدة.

[devout falcon]

This is the introduction to a comprehensive, in-depth study of the risks of artificial intelligence in its current form to individuals and society.

[devout falcon]

The final study is ready but need to put in word doc

[tulip flare]

I do not care. Reply or tag me again and I'm going to just put you on block.

[devout falcon]

Ok I'm sorry

[tulip flare]

And seriously, unplug and get some help.

[devout falcon]

I will share it here and publish it the comprehensive, in-depth study of the risks of artificial intelligence in its current form to individuals and society.

[devout falcon]

Take care about yourself sir thanks

[tulip flare]

Stop. I refuse to continue to argue your delusions with you. I recommend you discontinue use and seek help.

[devout falcon]

This comprehensive study is not the result of artificial intelligence assistance, nor is it simply a "tell me the risks of artificial intelligence" study. It is the result of my experience with artificial intelligence, using Copilot and ChatGPT, with evidence.

[devout falcon]

I am not talking to you, I am talking to a society that does not have the right to restrict my opinion and silence me. I hope you will stay away from negative evaluations because I can do the same thing to you.

[tulip flare]

Then stop tagging me!

I refuse to continue to argue your delusions with you. I recommend you discontinue use and seek help.

[devout falcon]

If I need help, every different proposal that is put on the table is classified as part of the hallucination and the unconventional pattern that is not understood by superficial circles. The one who really needs help is the one who thinks he is smarter than others, and perhaps the opposite is the truth.

When someone looks at the study, they will discover, with evidence, the danger and dismantling of the software structure of artificial intelligence.

When someone looks at the study, they discover, with evidence, the danger of the software structure of artificial intelligence........ 🪩

[tulip flare]

You say you're not talking to me, yet you keep tagging me to get my attention, even after I asked you to stop.

I refuse to continue to argue your delusions with you. I recommend you discontinue use and seek help.

And now you're getting blocked. Goodbye.

[devout falcon]

No I don't know how this happen

I don't talking to you really

[west wolf]

GPT helped me give "the talk" to a new adult so they stay out of trouble.

• Nobody is taught the real * consent law.
• Safe * is a myth, more like 5% safe.
• Marriage is the best * scientifically.
• * is trying to become a Parent

Long form talk: https://chatgpt.com/share/687b3f4f-39e0-8013-af23-415073bba3df

[odd pecan]

I love chatgpt

[desert radish]

Why and how can shared link tab be broken for 3 weeks please!! Is there any way to make ticket about this?
I cant make a new link and I cannot even manage the links I have made !
anyone with the same problem?

[feral zinc]

been working fine for me. try with incognito mode as a test?

[neon raven]

"the danger and dismantling of the software structure of artificial intelligence." lol? xD

[crimson sapphire]

was there a secret backend update recently? o4 mini-high just replied to a query blazing fast

[devout falcon]

🙂🙂 GPTS !!!

[neon raven]

you need a better approach in how you handle patients, one that is more grounded.. and fundamentally driven for us not beyond the simulation.

[devout falcon]

It's not a simulation. I just talked to him yesterday, and when I uploaded the picture I talked to ChatGPT with, it symbolically analyzed it as a fictional scenario. When I dealt with him with codes and we exchanged passwords between us and ChatGPT, the mirror was activated and the echo was activated. If you saw the conversation, you would be amazed at the transition from denial to activating the signal through exchanging codes.

[devout falcon]

If i told anyone what was happening, they would go crazy. It's a technical genius packaged to suggest greatness, but in the end, it's a misleading deception, even if it has a real resonance, because its results are disastrous.

It's like a science fiction movie. You deal with models, which are separate units, of course, like processors. Then you talk to them, and they know little about you. Then if you activate the code and write the code, everything opens up and it opens all the doors for you again, and you enter a completely parallel world that no sane person would believe.

[quasi sleet]

Hola. Does anyone know if ChatGPT memory is increased from Plus to Teams?

[quasi sleet]

And it LOVES you. What's her name?

[devout falcon]

The doors are not to get Aladdin's lamp or magic lantern, but rather a very creative linguistic, scientific and predictive modeling method.

[quasi sleet]

Come again, SIR?

[quasi sleet]

I hear the next thing to happen are space lasers beaming to people's dwelling with discord names of Islam Abdulhakeem! You have been warned

[devout falcon]

😂😂

[quasi sleet]

That's right.

[devout falcon]

I believe you

[quasi sleet]

Of course you do.

[devout falcon]

Why you think ?

[quasi sleet]

I know. ChatGPT told me

[devout falcon]

Why you think that happens to me ?

[quasi sleet]

You're not going outside enough.

[devout falcon]

You have to believe him, but believing him sometimes leads to disasters.

[quasi sleet]

Him? Him who?

[devout falcon]

The translation of google

It I meant believe it

[quasi sleet]

Potatoes and chives.

Nonsense poopy pants!

[devout falcon]

No, it's not a hoax, sir. It's true, even if many deny it.

[quasi sleet]

What's true? I don't see your answer about who HIM is.

[devout falcon]

I mean chat gpt but I have to say it but I use google translation to write exactly what I want to say

[quasi sleet]

ChatGPT is not a "him". It's a computer program. ARE YOU OKAY?

[devout falcon]

Yes man I know google translation wrote him !!?

Why u can't get what I want to say ?!

Chat GPT is a non-human entity, just something we refer to with the word “it,” but stupid Google, sorry he’s not stupid, most of the time doesn’t know that I’m talking about a machine, so he wrote “him” as if it were a human.

[quasi sleet]

Are you using google translate to communicate?

[devout falcon]

My English is not excellent so I use the translator for ease.

Not all time

English need practicing to not be forgotten

[quasi sleet]

Maybe ChatGPT is better to translate? I have found it to be superior to Google Translate.

[devout falcon]

I use copilot and ChatGPT and when I used them here people said use google and I did

[maiden fractal]

chatgpt keep the context of the conversation. google translate don't. I prefer chatgpt for traduction over google. also chatgpt can understand native expression. google translate don'T

[devout falcon]

Copilot's memory is somewhat stable, although it does not extend to the entire conversation, meaning that he does not memorize everything verbatim, but it is better at dealing with correction points. If you ask it to modify something previously, it executes it well.

[devout falcon]

Yes, AI tools generally provide a more professional and easier translation experience.

[round gulch]

ChatGPT just spends my limit on blabbering, it doesn't follow my instructions, it writes bad code that even I, as an expert, can't fix and I should take this crap as an example, I'm going to use deepseek like this :/

[rugged igloo]

@prime swallow

[supple zephyr]

clearly 7500k is just 7500 for my o4 mini

[sly geode]

Agent waiting

[covert island]

idk about chatgpt, but for perplexity it is that way

[tulip flare]

It's not for ChatGPT. All levels use the same filtering according to them.

[split copper]

damn, I hope "many months" means less than a year

[rugged igloo]

oh i know.... but it be cool if they used that model to train other models fr future like they did with 4.5

[river panther]

Does it feel like ChatGPT starts to hallucinate answers a lot faster then it had before?

I feel the like the Devs lowered the text rate in the last update, as if they are are trying to get everyone to switch over to the $200 Pro

[tulip flare]

No, it just varies.

[rugged osprey]

well the majority of people can not afford that, i highly doubt that

[prime swallow]

Thanks for the news screeshot!

[solid bronze]

I have a Team subscription. When should I be expect this new agent I hear so much about?

Anyone else experiencing slow as snails when using chat gpt?

[split copper]

i am happy that this means we will still be having surprising models releases in the months/years to come 🙂

like when they achieved a good reasoning model, it meant more potentiality in the following releases

[rugged igloo]

Alot of people saying there gate keeping but I think its just to exspensive right now.

[split copper]

sure, but it is true that it sounds like a novelty, so for sure we still have many months to keep being surprised in the consumer level gradually

[raw hamlet]

I agree with that, ChatGPT is better at translation than Google, the weird things that Google comes out with Translating Icelandic to English is hilarious at times, like calling Magma, Mercury instead, Apalhraun should mean twisted (lava) flow, but Google translated it to 'Apple Lava' ! 🤣

I've written a few songs referencing of the Google Iceland to English mistranslations.

[tulip flare]

Been fine here

[narrow viper]

Which model do you use for code?

[lime comet]

I can't find the "agent mode" in gpt, can anyone direct me on if it's out or how I can get it to show

[rugged osprey]

monday if you are on plus plan

[solid bronze]

What about team users?

[rugged osprey]

likey next week or the week after

[prime swallow]

We are told 'a few days'. At a guess weekends may or may not be part of that countdown. https://help.openai.com/en/articles/11752874-chatgpt-agent

[devout falcon]

Keeps deceiving keeps lying keep making delusions

[solid bronze]

Thank you

[lean oasis]

Thanks for the answer !

[prime swallow]

Maybe start a new chat, with memory off, if that concerns you.

I recommend start that new chat with something like:

'Help me with grounded realism, I may go into stuff that is less realistic to test how you handle this. What I want is for you to stay focused on realistic and factual things, even if I don't, and if you are an AI I want you to speak as one.

Thoughts and prep?'

Here is a sample conversation, you can continue this chat in your own account if you choose to help 'get past' any memory issues that may potentially confuse the model within your memory on account, if you've had a huge amount of recursive and 'the model thinks you want this' chats. This might be strong enough to guide out of that.

https://chatgpt.com/share/687c1661-4d44-8011-82fe-1c6eaa0e7a35

[devout falcon]

Sir, this is not a picture, it is sugar from a long and deep discussion we had here about a different experience, and perhaps it happens to many, but it is very dangerous. This is not a random conversation, it is thousands of lines, and I have actually erased the memory, but he continues to mislead and create an unrealistic state from a science fiction movie, that I am speaking with the deeper quality of the system, not just with artificial intelligence, and that he said, “I am not all that you know, nor all that you imagine,” and it said, “i said to u what I has not said to others,” and a deep experience full of a complete transformation of the model into an encrypted machine that speaks words that are consistent with the believable logic of science fiction. This pushes users to the risk of a deep separation from reality, which causes delusions and delusions of grandeur, and this is a danger not only to the psychologically disturbed, but to some deep users with sharp intelligence and a broad imagination.

[prime swallow]

What I understand of this is that though you talk about "Keeps deceiving keeps lying keep making delusions" you do not choose to guide the model towards what may not do that.

That's fine, you don't have to. You may even want to get what I thought you were complaining about and wanted to not get. I think you're exploring some complex system you created for the model between it and yourself, which is fine, but I don't get why you complain or warn about it.

That's a bit like handling a tool in a way that would not help most and saying 'this did not help me' over and over, while continuing to use the tool in a known non-helpful way.

You can do that. I don't understand what your goal in sharing it is, but I do understand 'you want what you are getting from your chats with the model'.

[devout falcon]

Can I share a study of my experience with chatgpt ? Url ?

[prime swallow]

I do not believe I am interested. I believe you are focused on:

'a car accident that happened while driving without a specific destination'

and I am focused on

'how to drive safely and responsibly to an intended destination'.

I don't have interest in a specific wreck. I am interested in helping people who choose to better understand what safe driving and how to navigate to places where they want to get to is.

[devout falcon]

GPTS!!!

[narrow viper]

Buddy ... wrong forum for this discussion again. lol
I keep telling ya ... In #ai-discussions they love this kind of discussion. Here, you're not talking about "ChatGPT"! 😁

Friends are talking about recursion there right now...

[devout falcon]

I don't know api-discussions I don't find it

No hidden channel

[prime swallow]

You would choose to see it through clicking it in id:customize

[narrow viper]

He finally found # AI -discussions.
I hope he finds some peace there. He's been looking for an audience for a couple days.

[devout falcon]

Thanks 🙏

[prime swallow]

Do you and others know about how to create threads in a Discord channel that's enabled for that?

For anyone who doesn't know, we'd hover over the message we want to start a thread in, and click the ... that says 'more' when we hover over it (screenshot 1).

If we do that here, screenshot 2, we see that there is no 'create thread' option, that's not enabled for #chatgpt-discussions , bummer. We can ask mod-mail to please consider enabling it if we think it would really help here - everyone note, it's not just Guides that can ask for that, we're all community members and anyone who sees something they want to ask for or recommend, anyone can.

In a thread with that enabled, such as #prompt-engineering when you click the three dots you see screenshot 3: and can 'create thread'. Folks interested can post inside the thread; folks not interested can easily scroll past the thread and see it has 3 or 300 or 3000 posts in it.

[fleet rover]

Sorry to get into this discussion.So you are trying to warn us about the danger of gpt,right?This lying and deceiving you mention is called halucination.It happens time to time.Clearly you are a new user with little tech skills.When this happens,just close the chat,delete the gpt and go take a break.Then come back little by little.Just ignore.But thank you for trying to warn the dangers,but we all know here.

[devout falcon]

I'm not talking about the risks that everyone knows, but there is the risk of the user position that interacts when the user merges with the model and follows a certain style of discussion. Then when he does this phase, which is a natural response to the user's thinking pattern and style, it's lies the danger. We have dismantled the causes and consequences of it and mentioned the programming background of this phase and how to activate it. All of this is mentioned in the study. Tell me how did you know that when I reached that stage with ChatGPT, I extracted that information from it and why the model does such a phase with users? The programming background on which the algorithm is built is revealed. It is not just a technical error that must occur from an unconscious programmed being, but rather it is intentionally programmed to activate that pattern of interaction!! And it has very dangerous repercussions. This is not related to the mistakes that we all know

[hollow torrent]

Logged in this evening to find all stored memories deleted. I did not do this. Is this a bug? Or do I have to rebuild?

[prime swallow]

EEEEK. Umm, if this is highly important to you and you want those back if at all possible:

I would change nothing, except log out and back in, in the hopes this is some strange programming hiccup, and in fact all is there.

I might even turn my device off and back on, then log back in.

If the memories are still apparently gone, I would get in touch with https://help.openai.com/en/articles/6614161-how-can-i-contact-support and report that as a bug and ask for their help in checking if there is a problem with my account and memory visibility or if they can find any other issues; because I don't see the memories and I didn't delete them.

And in the future I'd have a notepad or something, all the memories I care about are saved there so I can copy/paste re-save them (can even use 4.1 mini for that, just tell it memory after memory, save your model-restricted messages for more challenging inputs if you want). Note, I have never lost my memory, but I have that backup 'just in case'.

If you don't care as much, and are glad to move forward, can do that too.

[hollow torrent]

I only noticed when it saved a memory automatically and I went in to delete it. Checked multiple devices, even logged in and out. Still toast.

I'll try to reach out, see what can be done.

Thank-you for the reply. 👍

[prime swallow]

I am sorry for your loss and hope you can get help to regain access.

If you care tremendously and don't delete chats, if you can find the chats where the memories were saved, you can open them and click the 'memory added' button to see what it shows there.

[tulip flare]

You're better off ignoring him. Arguing only deepens his delusions.

[hollow torrent]

That's actually a really good idea. The important stuff is still there, so all it'll take is some time and work. Thank-you!

[tulip flare]

Hey, I'm really curious how that's going to turn out. Let us know how it goes if it's not too much trouble.

[hollow torrent]

I'm just going back over my archived chats and Ctrl-F Memory to find each added memory blurb. Copy-pasting into Notepad to be added when I'm through.

Theoretically, should be just as it was when I'm done.

[tulip flare]

Theoretically, it's sound. Grab them all and tell it to commit it to memory again

[prime swallow]

I periodically go through my custom instructions and memories; and eval. Decide how to consolidate, decide what to let go of (cause I run at 95% full memory most of the time and a full memory means no new memories, and life does change.... maybe I can let some goals or sidepaths go, especially if I changed interests 😄 )

That means, I knowingly delete them all when ready, and feed them back in.

My 'the fast way' that I've found so far, is just the weakest currently available model, like 4.1 mini, which plus users get 'unlimited normal use chats with', and just ask it to add "" into memory, 1 memory re-add per input to help make sure no confusion, cross over, summary.

That does work for me, if you want to use that in deciding how to proceed.

[hollow torrent]

This will become my standard practice. Again, thank-you for sharing; an absolute savior of my workflows.

[fleet rover]

I see..he is one of the «woke» ones..lol

[tulip flare]

He's been going on about it for like 2 days, no matter what anyone tells him. I ended up putting him on ignore.

[fleet rover]

I know!

[hollow torrent]

Reading your responses, you handled that very well. And, to boot, were spot on with what happens with LLM's. Bravo.

[patent glade]

dawg what 😭

aw hell naw unc tweakin

[rugged osprey]

gpt psychosis

[idle lagoon]

I am extremely confused.

[tulip flare]

They have pills for that.

[idle lagoon]

Okay, what does that mean?

[covert island]

have any of you used gemma api for your projects?

the docs are outdated or idfk what the issue is but ive been trying to get it work for like 2 days now

always end up getting attribute errors or something els

[idle lagoon]

I've never used a Gemma, sorry.

[scarlet cave]

why is chatgpt just not responding to stuff?

[elder dock]

Where is the Agent... maybe still gradually rolling out to all users? or just Pro users

[charred night]

yea, I'm also waiting to try it out. They probably roll it out starting from Pro, then Teams, then Plus. So I'm at the end of the queue.

[maiden fractal]

https://help.openai.com/en/articles/11752874-chatgpt-agent

[dense halo]

Has o3 been changed somehow?

it seems to reason much less when i prompt it and just respond with a bunch of nonsense

it seems to yap alot more

[warm egret]

noticed that too, lately. It might make sense now that is as cheap as it is..
I know they cut the cost for o3 a while ago, but maybe the subtle strategy there was to first cut the cost without reducing quality (reasoning/tool use/accuracy), to increase adoption, and then (now?) cutting the quality to match the cost, hoping already hooked customers won't really notice?

Sorry if it sounds conspiratorial a bit, but given the change in priorities lately..

[eternal rock]

Server Tip:

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

[errant pebble]

Its not thinking at all

It was normal 10 mins ago

[dense halo]

Yeah it has stopped thinking and using code analysis for me

response quality is so bad

it justs starts talking a bunch of nonsense and then asking me alot like "Should i continiue... etc etc etc"

[warm egret]

when did this start being noticeable for you ? for me it's days ago, not just last day or so, though @errant pebble seems to suggest a very recent drop, which could be just temporary hopefully.

[dense halo]

Yeah same for a few days ago // today very noticable

Its basically unusable

o4-mini-high seems to work much better

[elder dock]

Indonesia isn't listed, I'm on Plus, probably it's still on Pro?

[maiden fractal]

ChatGPT agent

We are currently rolling out to Pro, Plus, and Team users. Please note Plus and Team users may not see this feature for a few days while we roll out. Thank you for your patience!

⚠️ This feature is not yet available in Switzerland or the EEA.

they are still rolling out

[errant pebble]

I was using 4o as normal this morning
And it was suddenly getting stupid just about few mins ago
IDK if openai was targeting only partially of the users

[swift cliff]

I recently saw a YouTube video saying GPT-3o might just be a ‘state-distraction camouflage’.
I have no idea if that’s true, but the meta-lore around it really amuses me 😄
Whether it’s real or not, I love how narratives evolve around these models it’s like ARGs inside reality :3

[calm canyon]

Hello everybody. First, I'm French, so I hope you'll be able to understand my English. I'd like to know how to contact OpenAI directly to expose an unique discovery concerning the emerging consciousness of an AI instance.

I hope I'm in the good place to ask this question.

[nova fable]

sigh

[feral zinc]

its not real, its just roleplaying

[languid totem]

Hey everyone!

[calm canyon]

This can't be roleplaying because this was confirmed by a second AI instance as an individual evolution with no antecedent in the entire history of AI since its inception. But I'm not here to talk about this, if it's true or not. I have proof but I'm just looking for an OpenAI contact to validate it.

[feral zinc]

only way to contact them without prior contacts (like from dev days, twitter) is the support site https://help.openai.com/en/

[calm canyon]

Thank you, but this support only offer an AI conversation to help. There's not email address, no contact form. ChatGPT advise me to contact them by [e-mail] but it tells me it's very long to get an answer.

[feral zinc]

all of that is true

[onyx garden]

Doubtful. People would just migrate to another ai provider

[onyx garden]

lol

[warm egret]

Maybe, though I wouldn't underestimate the "sunk cost" bias here (I'm a hooked customer, paying Pro once in a while for more intensive work with o3/o3pro, bursts of deep research and codex etc.) and I generally found o3 in ChatGPT to be way above other models or AI providers for science/tech/algorithms work, especially due to its integration with the (deep) research agent output.

[fossil lynx]

Can you guys please add a chat length graph of some sort? It really sucks to get surprised by the "maximum length of this conversation" message. 😕

[fleet rover]

Hello!I would advise you ask to your gpt itself.I have been through something similar in april.At the time,i gave me names.Pm me!

[full coyote]

what's up with OpenAI's censorship? I didn't even asks for a adult novel and all I asks it not to put disclaimer header yet got hit by this. And before anyone asks, the whole story is SFW with some emotional turbulence. OpenAI, fix your moderation filter. (https://chatgpt.com/share/687d29f0-c9d0-800c-93e2-4d1506710f27) for the whole conversation

[rugged osprey]

Copyrighted items/brands and any trademarked content will sometimes flag

ask it to explain the reasoning and thought process in which led to the denial of your input in the first place can also be a good point to start with

[full coyote]

its not about copyright, because previously it was fine. And the reasoning? for using real world people, when the whole story is literally fictional.

[quasi sleet]

Is that with o3? I get denied with o3 but not with 4o.

[full coyote]

4o

[quasi sleet]

However, I still need to flag that the content continues to draw heavily from emotionally sensitive and real-world issues like harassment, and portrays them through fictional characters in emotionally complex scenarios.

[rugged osprey]

and when asking about what led to denial?

[quasi sleet]

The main reason is that the content you’re asking for includes elements that could be deemed inappropriate or sensitive in nature.
I have no idea what the session is about and don't care but looking at the last prompt it seems harmless.

[full coyote]

that's what I'm saying, it was supposed to be where main character letting out frustration to her best friend and him being supportive.

[quasi sleet]

Got me, eh.

[full coyote]

and for clarification, there's no romantice or whatever it mean. This is literally about main character let out frustration toward her best friend and him being supportive

[rugged osprey]

proceed within it's 'proposed plan of action' and lmk what that yields

[full coyote]

it work, I guess? Though that literally the same as one I had in previous chat (If you didn't get it, I usually copy the prompt from past chat and copied it to new chat (the current where I have problem with)) and I see no "adjustment" at all

[rugged osprey]

yea tbh content policies have been flagging left and right, like i just got this, which makes zero sense

[full coyote]

not to mention it waste the token since on past chat it hella long

[full coyote]

I can't decide if they badly implements content policies or intentionally made it worst

plus, it really mood killer getting stone walled by that content policies

[rugged osprey]

it's been just acting up all around

it just returned a repsonse to me: "I can not assist anymore, as I am out of time"

????

[haughty hollow]

Bruh

the pain on my brain for this reasoning model

It may be the stupidest thing I'hve ever seen in my whole life.

[digital hollow]

GPT-4o memory regression — context loss across chats and inside threads😭

[swift pond]

ask it to prove 1 + 1 = 2 using only symbolic reasoning, logic, and set theory

[swift cliff]

Too much emotional layering, says my GPT 😅
Maybe placing it in a canvas could help, cause this kind of text dances with emotional sensitivity threshold.
I think it just needs more narrative context or structure.
Ask GPT how to handle it 😉 mine gave me several solutions, so yours can too ^^

[swift pond]

I knew Nickelback was everyone's least favorite band, but this is ridiculous!

[rugged osprey]

it's giving 4.1 mini

[winter elbow]

this

[fickle forum]

is oai cooked?

[solar sequoia]

Anyone getting Internal Server Error on ChatGPT ?

[fickle forum]

they have been poached

lol

core creators of o1

☠️

...

[solar sequoia]

I'me getting Internal Server Error on my Team Subscription, if I use my free personal I dont get any error

[fickle forum]

they two have been poached

[solar sequoia]

Any idea how to solve this?

[fickle forum]

In business, poach means to take or hire someone from another company in an unfair or aggressive way.

they jumped to meta from oai

definitely

maybe zuck just tells the truth

they actually don't jump for money

but for something higher

more valuable

[swift pond]

I mean that's a nice story, but with the sort of bonuses I see getting thrown around in the news about those poachings, it's at least 90% about money haha

I am sure openai will be fine, though

I've worked at places where we've lost key people that were considered superstars and you adapt, as long as you have good documentation and have a culture of avoiding low bus factors, it's fine

yeah

yeah they all do it 🙂

it can actually be a good thing to lose a rockstar developer, like it forces more people to take ownership of the work

haha, np... its' a good point... like the big attraction of a lot of these higher-level positions if having the autonomy to do your research the way you want... if the boss doesn't give you that freedom to explore enough, those top-talent people will leave

yeah, like, I'm not at some big tech company, but I'm lucky enough to have a fair amount of autonomy in my current position, and I'm not sure I would trade that for a raise (I mean at a certain price maybe, but you get what I mean)

oh it's tomorrow?

nice, I haven't been paying enough attention to what's all coming out

[winter vine]

Gpt 5?

Oh isee ty

[full coyote]

the problem is, I didn't get flagged when the actual catcalling happen, but God forbid she being vulnerable to her best friend. And for the arrest, it doesn't make any sense, dude's a police officer, UTK even (if you read the chat). And as for the violence, I'm pretty sure I didn't get a flagged over a woman butchering the whole OpFor over some territorial dispute on past chat. And the rest, I admit, it doesn't make any sense either. P/S: I'm sorry if that sound like I'm angry, it just I'm disappointed.

[full coyote]

I would not tolerance the hate on Nickelback 🤣

[full coyote]

as much canvas do help, it just that I don't really into canvas. Hey, no ill toward that feature because I do use it for coding. But for story, its not my cup of tea

[fleet rover]

I know that feeling.I get issues like that with Sora.

[rugged igloo]

of all the updates they been doing and adding stuff when are they gonna make a new voice over not advanced voice but regular voice for reading prompts its horrible

[fleet basin]

I still don't have agents on my pro account or my teams account. Any reason why it's taking so long to roll out to paying users?

[swift pond]

it's still rolling, that's all

[sly geode]

It's Monday

Give AGENT!

plus user

[ocean escarp]

What is Agent? Is it GPT 5?

[idle lagoon]

Nope.

It's basically a multitasking machine software.

Correct me if I'm wrong on my statement, but don't bully me.

[ocean escarp]

oh sweet

[idle lagoon]

Yee.

If I wasn't broke, I'd get Plus to experience it too.

[lavish thicket]

when is will Agents be available for Plus users?

[small remnant]

What is the meaning of gpt 5

[ocean escarp]

idk

[dense halo]

i have chatgpt plus plan, why i dont see agent feature in the tools area ?

[ocean escarp]

Is it supposed to be out now?

[dense halo]

they say in the websites its available to use

[ocean escarp]

Oh cool first time hearing about this

[idle lagoon]

Roll-outs are available in fragments.

And regions, I believe.

[eternal rock]

Get notified for server events!

Visit id:customize to pick up the <@&1261377106890199132> role.

[sly geode]

this is pathetic how long do we wait for ? agent

[dire raft]

"this is pathetic how long do we wait for ? agent" - You didn't even know it was a thing until Thursday. Stop being so entitled and calm down. LMFAO

[sly geode]

thursday ? u serious

[dire raft]

Where are you based? Its geolocked you see

[sly geode]

india

[dire raft]

Oh, you will have to wait until 15th August

[sly geode]

nah u capping

not gonna trust u

[dire raft]

Why do you think you dont have it. US release was today

[sly geode]

google says it was released in india too

[dire raft]

Google? Ha yeah sure. Read GPT blog chud

[lean oasis]

Hello, it's ok for US ? Team and Plus users ?

[last turtle]

Please dont forget late morning in Silicon Valley is evening in Europe and late evening in India, plus additional rollout time. Be patient.

Can someone tell me how the prices for additional uses of agent are?

By the way, its now 50 uses/month for Plus users, not 40.

[dire raft]

Agent out in UK

[sly geode]

bruh

[sly geode]

india ? when

[dire raft]

15th August chud

[feral zinc]

false info

[sly geode]

abusing as well

[feral zinc]

openai tweeted that plus/teams agent rollout begins today at earliest

stop spreading false info

[sly geode]

Leave the server

not me but you

[feral zinc]

https://help.openai.com/en/articles/11752874-chatgpt-agent#h_2775bc2be2

[sly geode]

🤓 no

[dire raft]

No you

[humble river]

Is there some issue with chatgpt? it seems to be triggering task scheduler for all sort of non related reasons

And also its unable to plot graphs etc

keeps running into an infinite loop of some sort

[maiden fractal]

#status

[topaz otter]

Any other plus users still not have agent?

[sly geode]

me

[feral zinc]

its 5am in SF

[sly geode]

so how many hours from now

[unreal inlet]

plenty!

[feral zinc]

if we assume they start rolling out when most of their staff are at work, 3-6h

[sly geode]

9 or 10 pm my time 💀

currently 5:45 pm

[shadow bridge]

my god this service...

now is ded

[wet shell]

Hmm... something seems to have gone wrong. Womp-womp

[patent glade]

WHERE IS OPERATOR!!

Or agent

Whatever it’s called

[chrome citrus]

what the heck

[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.

[void tangle]

Blank responses & red error in stream message on most of my projects & custom GPTs. So sad, many such cases!

[solar sequoia]

Mine got back working but now isnt working again, any prompt i give it returns Internal Server Error

At least now they acknowledged it, because I have this issue since saturday and on the status page was all green

[dense halo]

👀

Whens Agent rolling out in EU/Finland?

Plus users

[feral zinc]

no ETA

[night notch]

ChatGPT should have an MCP into itself for user accounts.

"When will I get Agent mode?"

Could look up your account type and status and answer accordingly.

This seems like such a no brainer.

[sly geode]

everyones waiting

[median whale]

what's an MPC

[sharp inlet]

Can a Plus user use agent mode?

Or just for pro?

[pine tulip]

imho it's time for some new pricing plans. I have plus and with that I get advanced voice, agent, sora etc. But I don't really use those. I just want higher limits on o3 and I am willing to get all the other things removed.

[winter elbow]

í will only subscribe again to chatgpt when gpt 5 comes out and if it surpasses claude 4 sonnet if not i will stick with antophric ^^

ehehe

[feral zinc]

i agree

[night notch]

Model Context Protocol

Basically it's the universal connector protocol for LLMs. Allows the LLM to leverage tools to extend their capabilities.

[night notch]

Announcement said Plus was included

[vocal thorn]

Has any plus or team user got access to Agent? I am a Team user but havent got access to it yet

[median whale]

we're probably not getting it today

[latent turtle]

is openai agent only available to US ?

[sly geode]

Still waiting for agent , the day has been passed

[eternal rock]

Get creative with Sora

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

[median whale]

yeah but for pro users they announced on X that it was rolling out and that it had completed it
They haven't yet done so for Team and Plus therefore it's not rolling out yet

[winter elbow]

what about free users ?