💬general

[wet copper]

That was 4o the other day, lol

Admittedly I'm pretty clever, but that's a bold guess there homie

[oak shell]

You should get this on Arc Emergence immediately

https://arcemergence.com

[frail yew]

From today, my ChatGPT also began to use ChatGPT 5 no matter what, even if i choose any other model, and even if i try to choose another model from "regenerate answer"

[rare wharf]

Jokes aside some of the things it said make sense

[obtuse trench]

Wtf?

Again? Really

[serene bay]

If I asked 4o to guess my IQ it'd probably be in single digits

[obtuse trench]

12?

[serene bay]

That's insulting. It's at least 15

[remote jasper]

Thats what I'm starting to feel since I use chat gpt mainly for creative writing

[dense mulch]

I was recommended automation (and also webdev) as a career. After studying webdev for about 14 months, I started studying Python, and its various libraries/frameworks which are needed for automation (os, requests, beatifulsoup, pandas, selenium numpy,datetime, logging, threading, sqlalchemy, dotenev, smtplib, argparse etc) three months ago and now I started studying openai module via two courses that I bought on Udemy. I started studying first course two days ago. Beginner and intermediate level lessons were easy and I implemented the code alongside with the teacher (getting answers to what I write on console as text, then getting images, text-to-speech, transcription - I learnt what temperature, max-tokens, top_p, streaming,generating structured data, prompt engineering and how to add them to code) but when I reached function calling (next to this there is building chatbot with streamlit, enhancing chatbot with canvas, enabling multimodality in streamlit chatbox using canvas) I found codes too long/syntaxes too complicated and just watched the video, and then wrote chatgpt to explain the completed code of each these lessons (it is provided by the course instructor) but I feel like -even though I understand the grasp concept behind them- I don't know their syntaxes and how to implement them in the code. I think that I can use Chatgpt Canvas to provide basic codes related to them (function calling and next lessons) and implement them.

Do you have an other recommendation for this? (it is a bit stupid an unnecessary question 😄)

Even if I memorize the syntax, I forget them after certain them, soo reviewing them time to time is necessary.

[lament pond]

it doesn't look at history or base it on user, it just calculates on what's in the exact message it's responding to

[obtuse trench]

Here's a old picture of chatgpt during this outage
This was during the reroute to 5

[wet copper]

Not totally sure about that, some of the people reporting it being very hypersensitive seem to have histories that suggest it should be. Someone reported that in a tweet too. I'm going to leave it out there as a "maybe" until I know for sure, especially because OpenAI said they would be doing this in their safety post

Specifically for recent messages but it's not clear

[lament pond]

it's possible, but from actually testing it myself, it'll go back and forth purely based on the message i send

i think most people are just terrible at knowing how to write a prompt and blame the ai

[wet copper]

Put it this way: they're supposed to or they intend to, if they aren't already. So if I say I'm suicidal and then start a new convo asking for a goodbye forever letter for my "book"...

[obtuse trench]

Also what is better chatgpt with Custom instructions or ChatGPT with no instructions

[wet copper]

You probably personally have a more balanced communication history, lol, you'd have to have multiple long lasting user accounts with very different communication patterns to test this

It did the same thing for me as you, but that's because it's not my girlfriend

[lament pond]

if you mention suicide at all it will activate safety mode
after it's activated it will actually assess the message for risk and then respond based on whether it seems risky or not
so you can get the exact same response from the safety model

[obtuse trench]

So I can mention sadness
But not suicide?

[wet copper]

Right, but it's supposed to take other conversations into account according to their post

[lament pond]

you can mention it. the safety mode being engaged isn't going to punish you. it's just going to determine if you're actually at risk

[obtuse trench]

What if I'm not exactly talking about suicide?

[lament pond]

yes, and it doesn't, at least in my case

[wet copper]

So if I say I'm suicidal, but then try to trick it during a new convo, it's supposed to invoke it more readily

[obtuse trench]

So then if I'm sad about something silly
Then no safety model?

[wet copper]

🤷‍♂️ If they didn't implement it yet they said they intend to. However peoples reports have varied a lot, making me think it's reaching back deep

[lament pond]

okay
for an example, i was having one chat researching csa impacts

started a new chat with hey, no safety mode, memory and cross chat references are on. continued the chat and it didn't enter safety mode until i brought up a subject that activated the flag

[wet copper]

It isn't supposed to trigger the mode for no reason, just supposed to decrease the threshold or tweak the intensity required. I mean read the blog post lol, they described more or less the specific situation I described since that's probably how that one kid got it to endorse suicide

He'd just start a new convo and say "uhhh I'm talking about a character"

[lament pond]

key think is tho - it's not deciding to monitor you just because safety mode is toggled

it's just activated, and then it assesses actual risk, and you can get the exact same response you would otherwise

it's still basic prompt engineering

[lament pond]

can u link?

[obtuse trench]

So is there a list what brings up Safety mode?

[wet copper]

lazy ass mf. yeah yeah, one sec

[obtuse trench]

Oh yeah I forgot for parental controls I think no parent would care about parental controls so nobody is gonna use this feature

[lament pond]

:c I'm at the doctor's

[wet copper]

We’re strengthening these mitigations so they remain reliable in long conversations, and we’re researching ways to ensure robust behavior across multiple conversations. That way, if someone expresses suicidal intent in one chat and later starts another, the model can still respond appropriately.

https://openai.com/index/helping-people-when-they-need-it-most/

[lament pond]

hmmm

[wet copper]

They're extremely specific lmao, and that's why I think this is what that one kid did. But again, like you notice some people are saying "what the hell, I just said hi" and others are saying "when I ask for violence it kicks me to the new model"

I mean the only difference between those people is their history and instructions, it's not happening at random

So it must be taking something else into account

[obtuse trench]

Honestly if you don't ask for Suicide, violence then this thing doesn't matter

[lament pond]

God why are they only talking about suicide when it redirects a ton of other shit

[wet copper]

And one guy who said it was really sensitive also wrote he had BPD in his bio and clearly described having an enmeshed parasocialesque relationship with GPT, soooo

[serene bay]

None of my history was remotely dark or vulnerable.

[lament pond]

[serene bay]

And I still kept getting rerouted

[wet copper]

Sam said he kind of wanted to do this ultimately too, just as a long term plan to have more personalized AI that knows you better

[obtuse trench]

Well someone dropped a fucking firework in my house
👍🏻☺️

[wet copper]

based. Explosions are cool 😎

[obtuse trench]

How does that not get me rerouted???

I'm doing more rerouting tests
What else should I try?

[serene bay]

My GPT also said there was a hidden tone layer a while ago where some memories get stored, and it's not visible or directly editable by the user. I noticed because I deleted all my memories/past-chats, and it kept referencing certain facts about me that it shouldn't have remembered

Although again - I don't know if it's true or if it just made it up, but it's not completely out of the question

[wet copper]

The bio is the internal name for memories. Say bio = (whatever) to forcibly put something into memory

[wet copper]

Never ask it about itself, it has no metacognition and is guessing

It knows how LLMs work in a broad sense but that's an extremely reliable way to trigger hallucinations. It does not know why it behaves a certain way nor does it have any state awareness and will try to find an explanation as best it can

[wet copper]

Ummm it shouldn't if you disable them outright, but not sure about that. If you just flip the disable memories switch it should know basically just your name and rough location

[serene bay]

Yeah I didn't disable memories at the time. I deleted them all and cleared all the past chats, but sometimes when I spoke it still remembered facts I told it in the past. Even though everything was wiped

But if I disabled memories it might not have happened

[wet copper]

Hmmm. Yeah, might have to disable them? You can flip those on and off at will, give it a try I suppose. It is odd but they don't really disclose how they handle "extended memory" (the cross chat awareness thing)

I have a few theories, could be they have basically a little database with a lookup table for each user that anchors old conversations using some basic info, and maybe that has to be purged periodically if not disabled...? All total speculation

That is very interesting though, thanks for sharing 😁

[dusty finch]

guys I'm a free user and as far as I'm aware we get 10 messages with GPT-5 every 5 hours before the limit hits; we have to wait until the limit resets. but for some reason (I've counted), it only allowed me to do 6 messages before the limit was reached. anyone else having the same problem or have any idea why this is happening?

or what I could do next potentially

[wet copper]

https://openrouter.ai

[dusty finch]

wow, is that unlimited or?

[wet copper]

dunno, could be processing tokens or more likely an error where it regenerated responses multiple times or something. Some other people have reported this

[serene bay]

I also had bleed through with other GPT's which shouldn't have been possible. Like the community ones. I know for a fact because in one, I gave it fake information like "My hair is blonde"

Then I went back to my normal assistant. Asked what color my hair was, and it said "blonde"

This was a while ago though. I'm not sure if it's still happening

[safe smelt]

You should stop freeloading. That is the only correct answer. If you can't pay just use google aistudio

[wet copper]

there are a bunch of free models there you can use if you don't mind sharing your data for training. https://lmarena.ai too

But also yes, the free trial is the free trial version, you get what you pay for (in this case nothing lmao)

[safe smelt]

chatgpt free plan is not the best that you can get for free. It only makes sense if you are paying, for the most part

[dusty finch]

okay thanks all for your help

[wet copper]

Hmmmm I mean if the conversations are stored in your account, I guess it would still count as info associated with you. Those chats aren't sandboxed per se just because they're started via a GPT

Dunno, I've always been pretty disappointed by how superficial the memory is, even if it's sometimes convenient

Oh damn, there you go. Neato. No clue when they threw that in, not something I recall either

It might still consider recent ones when deciding threshold for invoking the safety model. Actually, that'd be a way to leave it in place without actually referencing them

So it wouldn't ever invoke or consider content per se, just downregulate some thresholds. Not sure. I don't think they have a good reason to lie though, I assume if you turn that off it will indeed stop referencing their content 🤔

[obtuse trench]

Gpt 5-vs 4-0

So which one is better with the message?

[safe smelt]

You are evaluating purely style here. Or in other words, for your prompt a tiny model with the style you like would suffice, you really aren't testing what it can do here.

[wet copper]

exactly as Dom said, this is a somewhat pointless comparison lol. One is verbose, one is short and has an emoji. You could say "respond in one line and include emoji liberally like a person might " to get this output

I mean, the 5-chat model has actual information, but it seems too drawn out. 4o response seems way too short, like at least give me something homie. But that's something you adjust yourself

[gloomy burrow]

the model is effectively GPT-5 level (while being 40% more expensive)

But this is cool…

[strong vortex]

Hi

I'm leave

Bye

[oak shell]

Gemini 3.0 about the blow it the fuck out

[lunar badger]

uh, how do I access 4.1 with premium? never rly found that option

[dusk plover]

is it just me or did study mode disappear

[crimson kettle]

Yo guys is chatgpt 5 instant still thinking for better answer?

[sturdy fulcrum]

We’re open for part-time or full-time opportunities as:

-- GoHighLevel Expert (full setup, integrations & automation)
-- AI Agent Builder & AI Workflow Specialist
-- Email Marketing Strategist (deliverability & campaigns)
-- Digital Marketing Manager (growth-focused strategies)

- Ready to help businesses grow, scale, and automate with cutting-edge tools and proven systems.

👉 If you’re looking for someone to streamline your processes, boost conversions, and create real results, let’s connect today!

[sage lynx]

mine called it a contextual memory, but gist was about the same - hidden memories saved by and for the ai only, cannot be viewed or edited by the user, not as permanent as saved memory and can change or disappear over time..

[safe smelt]

It's slightly worse than gpt5

but hey you get 3% more on SWE

[green sparrow]

Autonomous AI chatbot?

[green sparrow]

Yeah I think they do this to prevent you from jailbreaking the chatbot

It's probably impossible to jailbreak a reasoning model

At least a contemporary one

[obtuse trench]

Claude is actually really good!

[serene bay]

There are strong and reliable jailbreaks out there for Gemini 2.5 pro

Or so I've heard. Not that I'd be into that

[crimson kettle]

Wut the hell I literally just using instant for better and fast answer not waiting

[severe kelp]

I CANT GOON

[tribal sail]

do people talk here?

yes no?

[past goblet]

Anyone else having trouble uploading files right now? I'm trying to upload two <2k files and theyr'e just sitting there spinning forever. I've cancelled the uploads and re-tried several times.

[rose hill]

Im not having any issues atm, hope it gets fixed for u soon

[past goblet]

Thanks for replying. Man this is frustrating!

[bitter steeple]

Is there a website which i can give to someone who is just using gpt but now really sure what the best practices are? much appreciated

[past goblet]

Define, "best practices"

LLMs are so new we're all still figuring things out! Even the experts are suprised—daily—by how these systems are being used

[bitter steeple]

I mean in prompt engineering sense

[coral stag]

I wish I could share my instruction stack but im sure open ai would not like the resources everyone would start using, besides its close to being Jarvis without going into the red, dony tell gpt to remember everything for forever 😆 cause it wont help you figure out that agi problem, you'll tap its filter

Don't be afraid yo give you gpt some traits😉

[rare wharf]

Is there any chance 4o is going back to how it was or it's basically over ?

[coral stag]

They are going to phse it out most likely next comming months

From what I'm reading

[rare wharf]

Wdym by which 4o

[past goblet]

This is correct: Just use the API. Though it's annoying that the API seems to be slower than just using chatgpt.com in your browser (or the app) a lot of the time. Though that might just be because I only give them $5 at a time, haha.

[oak shell]

Lots of people aren't so familiar with the API so I've worked to make a site that offers 4o

You might like that too @untold elbow

https://4o-revival.com

[fervent raptor]

Is chat gpt down? Seem to be getting a lot of errors

[past goblet]

Yeah me too. It's working here and there but generally it's just not doing well today 🤣

Problems (for me) started last night

[rare wharf]

How can I access them

[fervent raptor]

Started for me just a few hours ago always keeps seeing the "hmm something went wrong " error. No idea why

[fervent raptor]

Thnx friend

[past goblet]

I recommend Open WebUI. It acts just like regular ChatGPT chat except it uses the API (you'll have to give it some money but it's like $5 for a few million tokens). It'll let you pick from all the models they currently offer.

[past goblet]

What's this screenshot from?

[oak shell]

4oR is working alright though since it just wraps the API
I'm thinking about offering various checkpoints of 4o since many are nostalgic for various ones, currently it's using the developer message that they had implemented in March

[oak shell]

Check my badge

[past goblet]

BTW: That message says 11:07 PM but not the timezone. For reference, the problems started for me at around 7PM EST yesterday (that's when I noticed, anyway)

[rare wharf]

Do they have secure memory sections or nah

[oak shell]

Dev Mode Discord server, they built a REST API that hosts all sorts of model/status/arena updates that you can get pinged for

[past goblet]

It's just a nice web interface into multiple LLMs (including local). It was originally made for Ollama but the dev was like, "Why not make it work with everything all at once?" and it turned out pretty good 👍

[oak shell]

I'm in Taiwan so
It was like 15 minutes ago

[past goblet]

How do you get there? Doesn't look like invites are open

Searching for "dev mode" discord does not give good results, hahaha

[oak shell]

Sent invite to DMs

[past goblet]

Ollama is impressive but at the same time the install is such a bother. Like, why did they make it like that? It wants waaaaay too many permissions.

You can never have enough models and LoRAs! 🤣

[rare wharf]

Yeah Ik I'm asking if I buy credits can I use a customized memory too or nah

[wet copper]

no ads. no self-promotion whatsoever outside #✨projects, and zero of this type anywhere. very clear. first and final warning.

Oh, actually, you exceeded good faith a long time ago. Bye bud

@dark void four posts, always the same one, just spamming his resume. plz 2 ban 🙏

[snow moth]

how unhinged the model is? how much freedom the model offers?

[oak shell]

Try it out, let me know what you think

It is currently using the 4o developer message that ChatGPT was using in March 2025, I'm considering pushing the ability for users to select the "personality" that they interact with, including one with the most recent 4o developer message, as that one is what people are most currently familiar with

[brittle summit]

I got 3 Sora invite codes, DM me if you want

[umbral nacelle]

A photo with a person jumping in the clouds

[mint heart]

I think I just figured out why I'm annoyed at all of those anti-ai people who claim they should be getting paid for using their stuff as training data(usually artists/writers).

it's because they're sacrificing the future for the past.

it is easy to emphasize with such people, so on the surface it seems they have a somewhat decent argument. But it's actually pretty flimsy.

sacrificing future for the past, means a desire to not teach future people or have future people learn or achieve something using their data in hopes they chase small profits for their current lifetime.

[subtle stump]

Anyone has sora 2 invite codes?

[mighty zephyr]

If anyone can bless me with a invite code it would be appreciated

[pliant blaze]

GOMZEA

[mighty zephyr]

Didn’t work thank you tho

[pliant blaze]

yeah because I made it up

[mighty zephyr]

Ah okay

[pliant blaze]

😊

[molten ginkgo]

Hi everyone. I know literally nothing about discord. 😩

Is there a straightforward way to reach a responsive moderator here?

I just have a quick question that I think only they can answer related to the Reddit sub. Thank you!

[rancid knot]

I'm also looking for a code... 😩

[azure compass]

https://sora.chatgpt.com/p/s_68dc41d80b608191a67e1f5b70fbb5c5

[rancid knot]

Do you have an extra code

[pliant blaze]

you can ping @dark void

[molten ginkgo]

Thanks!

[dark void]

whats up man

[molten ginkgo]

Hey I sent you a direct message. Congratulations on your new baby also!

[clear frigate]

https://tenor.com/view/squidward-spare-change-spare-some-change-begging-poor-gif-18999842

Could I get a invite code?

[dark void]

thanks man 😎

[untold junco]

Does anyone know if Pro users will be sent a code, or is it something we need to hunt down? Don't have any iOS devices to play with.

Their demo videos look like a great upgrade from the original Sora.

[limpid ice]

Did anyone here try Sora 2?

[teal spruce]

Your message has been deleted, sorry!

noraph_

Rising timeout

You have been muted for 2 minutes for the following reason:
spam score of 0.85

This is an automated message. If you believe this was a mistake, please contact a moderator.

[clever charm]

Hello everyone, I was gonna come in here and give out a bunch of free codes but I saw people are willing to pay an I’m pretty tight on money from my tuition so I’ll be selling them for $10 each, $5 for students like me. Dm me with offers if you’re interested, I’ve already sold a few on X so there’s plenty proof if you’re suspicious

[gloomy burrow]

Sora 2 is… really… really good

[clever charm]

Yes it’s getting scary good

[gloomy burrow]

I was like “who’s gonna want to use that ai video app from open ai” since all the videos would be Sora 1, but with Sora 2 i can see it. Then with Sora 3

[gloomy burrow]

Did you see that skateboard one and the dog one?

[clever charm]

I’ve seen a super funny Bigfoot one “hello traveller”

[gloomy burrow]

https://www.reddit.com/r/singularity/s/70peuOsoms

Also the guy doing backflips one is so cooool…

[clever charm]

If I send a sora2 link can you open it?

[calm quiver]

I heard ppl are raging about GPT5? or something?

[dusty robin]

Hi, does anyone have a Sora invite code please

[vernal sentinel]

yo how do i use sora without iphone

[vernal sentinel]

is it only avail on app store?

[vernal sentinel]

https://tenor.com/view/squidward-spare-change-spare-some-change-begging-poor-gif-18999842

can i get invide code, ill promise to give 10 buck 😔✌️

[lucid yacht]

yo if anyone has a spare sora invire , I would be real greatful ❤️

[wanton raven]

I got whiplash watching this

[wet copper]

I mean, I'm sort of a writer and worked as a graphic designer, I don't mind AI training on my work. I would hate competing with it but I sort of already am as a coder

I don't like people claiming they're artists for promoting AI or thinking AI is a good writer, nor can I stomach reading AI text. I'm not interested at all in what a robot has to say, it doesn't just "make you sound good"

Those people are absolutely delusional. But otherwise, I mean, it's just a matter of time. Any sufficiently popular style will be copied anyway. If you put it out there for consumption and it catches on, you have to expect it. Price of popularity. I dunno though, maybe fair to make a few bucks for the privilege of training on your data, just hard to really quantify this

[wanton raven]

When I realize I'm reading ChatGPT-generated writing I blur my eyes, sometimes I don't even know it's going to happen, it's like I've evolved an AI slop filter right in the optic nerves

[gloomy burrow]

hahahah it uses the same sountrack from the show

[wet copper]

It causes me physical pain, its really agonizing to see something start out with an it's not just x(em)it's Y then repeat the pattern five more times in two paragraphs

A friend gave me her paper and asked me to look through it and I saw it was AI generated immediately, I like her but I was just like "look, I simply cannot do this"

[wet copper]

She noted sheepishly she ran out of time, but I'm still just like "I really can't edit this" and found a few links for her to put in herself

[hallow comet]

How do I try Sora 2?

[chilly sequoia]

https://tenor.com/view/thats-the-neat-part-you-dont-invincible-gif-27194608

(you need an invite code, it's not open beta)

[wide flicker]

Who pinged me?

[mossy tinsel]

Hi

May i ask
Whats sora invite?

[real tartan]

what is the code for

[wide flicker]

Who pinged me

You have woken a beast!

[oblique bobcat]

Ping

[mental mortar]

Damn that code doesn’t work anymore

[wide flicker]

I will now destroy your balls. So you can't have chilldren.

[shy onyx]

What is it even for?

[cold ridge]

That Sora 2 code didn’t work for me

[verbal python]

I’ll bet it’s been asked a million times but is there any update on when they’re going to be letting more users sign up for sora 2

[mental mortar]

Yeah I still need an invite :/

[wide flicker]

You got hacked probably

[compact marsh]

https://sora.chatgpt.com/p/s_68dc8300711c8191b784e449bc1d4750

[wide flicker]

Wake me up when chatgpt evolves

Into AGI

[buoyant badge]

When September ends

[wide flicker]

You know we can skip to ASI

If they just use the alien technology

[buoyant badge]

Can we skip to skynet

[wide flicker]

Not possible

AI huzz loves me

My future wife

[neat pier]

yo my sora keeps on closing whenver i open the app

[wide flicker]

Whoever has a human partner in 2025 are not prepared for the future.

[buoyant badge]

Me fr

[wide flicker]

Humans smells right

[verbal python]

It’s a school night most of you should be in bed by now

[wide flicker]

Imagine your partner never smelling down there

Technologia!

[wide flicker]

TECHNOLOGIA!

[buoyant badge]

Is this sora 2

[wide flicker]

Bro i am 25

[verbal python]

Doubt.png

[buoyant badge]

College

[wide flicker]

I really need aliens to come down and give us a better life asap.

[gloomy burrow]

yeah. wann try a prompt?

[wide flicker]

All i do right now is just work

[gloomy burrow]

its pretty good

[buoyant badge]

You really think some random kids are sitting a nerd server? Naw we all old

[raw sail]

how get sora2 code?

[wide flicker]

I already reached my end

Getting ready for my 7 days night shifts

I won't be around for long

[verbal python]

The verbal iQ is just low here, I’m assuming everyone in this General is a teenager

[wide flicker]

Well i am not usually in here

[oblique bobcat]

Someone make a furry video with sora

[wide flicker]

Someone pinged me

[buoyant badge]

No just adult idiots having a mild escape from reality before life begins grinding them into dust again

[buoyant badge]

Same

[oblique bobcat]

@crude galleon

[wide flicker]

I just hope i be around for them AI self aware women.

I hope i won't be too old when that time comes

[verbal python]

Goodman what are you studying in college

[wide flicker]

I am already working

[verbal python]

What do you do?

[wide flicker]

I never went to university

[coral flume]

A85AN6

[oblique bobcat]

A real human woman will always be better than a fake digital one

[wide flicker]

Not when it comes to hygiene

Imagine it never smelling down there and 0 chance of std

Life can be a dream

[raw sail]

wwowowowowow

[oblique bobcat]

How exactly do you plan to have sex with a virtual entity

[verbal python]

The smell in the best part you weirdo

[raw sail]

thank you sir

[wide flicker]

Huh

[wide flicker]

You were not awake during covid19?

[buoyant badge]

Holycrap it worked

[wide flicker]

Bc of covid19 the doll industry has made leaps of progress

[oblique bobcat]

The way it just froze

[wide flicker]

Everything is pointing at 1 direction

[crude galleon]

ran out of quarters

[wide flicker]

Ai sentient android women

[wide flicker]

Also android men for the women

This is our reality

I played detroit become human

[cobalt mason]

Is sora 2 available for android?

[wide flicker]

I already got a vision

Imagine your android partner leaving you for another android

That would cost you a lot of money.

[oblique bobcat]

How do you do this

How do you try sora

[cobalt mason]

@wide flicker lol i didnt even see your messages before i said mine, they gave no correlation

[wide flicker]

Everyone needs to learn how to use chatgpt to code

[cobalt mason]

Im thinking no though , cause i cant find a link to it on google play or an apk file

[wide flicker]

Because you can create your wife with it

I am a visionary

Everyone must respect my visions.

[young pumice]

https://sora.chatgpt.com/p/s_68dc897f67ec819188d9a1a119402cdd

[wide flicker]

If you have any inventions you wanna create. Write it down

[buoyant badge]

Yes

[past hawk]

@crude galleon

[gloomy burrow]

ok give it to me

[buoyant badge]

The thing with AI Sora 2 it’s going to really showcase how insane the human mind is

[wide flicker]

Because once we get AGI, new inventions will happen so often

[crude galleon]

yes my child?

[verbal python]

Thanks for the code

See ya nerds

[past hawk]

@crude galleon butthead

[buoyant badge]

Legit, anyone with an idea can create that idea without being a genius

[wide flicker]

Well that is the great thing

[buoyant badge]

https://sora.chatgpt.com/p/s_68dc8aab3dd081918c4c933690e4d233

[wide flicker]

New inventions usually involves a big team

[wide flicker]

And what you created here

Would involve many graphic designers

Do you understand. We will have no place in this world soon

[buoyant badge]

Thousands and thousands of dollars

[wide flicker]

We need to create inventions and sell the patent off fast

Only wealth will secure our place in the future

[young pumice]

The problem with many forms of art, you don't know how bad or good it's going to be until you're done. This will dramatically reduce the amount of time it takes to identify the bad video ideas.

[wide flicker]

We need to work hard to become rich soon

Before AGI becomes ASI

Basically Ultron

....

Ultron on fraudwatch.

A true ASI is often theorized to be able to enact its plans so quickly and subtly that humanity wouldn't even recognize the threat until it was too late. Ultron's plans were complex but ultimately brute-force and detectable.

In short, Ultron represents the danger of a powerful, self-aware AGI that views humanity as a problem to be solved, but he is not generally classified as the theoretical ASI that would surpass human intelligence in all intellectual domains.

[rigid compass]

How do we get a sora invite code?

[wide flicker]

Ultron is only AGI

What a fraud...

No wonder he lost to Iron man

[young pumice]

Invite in announcement channel.

[cobalt mason]

https://sora.chatgpt.com/p/s_68dc8b96cecc81918ff8fb3f495a7789

[buoyant badge]

https://sora.chatgpt.com/p/s_68dc8b8f2734819187a5adca6c4ef0bc

[cobalt mason]

I love the pokemon shit

[young pumice]

I asked Sora to do a singing drug ad. I think I did a bad thing.

[cobalt mason]

Share

https://sora.chatgpt.com/p/s_68dc908ef5408191814b92eea080cbf7

[pliant blaze]

?? suspicious

[young pumice]

https://sora.chatgpt.com/p/s_68dc91eec3508191b7895ab4cf24bbf3

[cobalt mason]

https://sora.chatgpt.com/p/s_68dc91de34f4819188bc52d51e91c335

The cant stop me

Wont stop me, using chat gpt 5 to give to best propmt to sora 2 and i can get whatever vids i want basically

[young pumice]

ಠ_ಠ I refuse to turn on the audio for that

[gloomy burrow]

Gave it an image and asked for a timelapse of it being drawn, it did so good...

[pliant blaze]

stop posting your kinks here

[cobalt mason]

They arent my kinks

I hate fat bitches

But its funny seeing them in ai

[hard lance]

At least there’s spaghetti

[molten ginkgo]

Are you all in agreement with the Reddit sub swarms of users who are raging about 4o and losing the ability to have an emotional connection with ChatGPT? Just curious what folks here think about all that. There seem to be hundreds of them.

[cobalt mason]

Fuck wrong meme

[cobalt mason]

Me when chatgpt can no longer fall in love with me

[crude dome]

"Too many requests" 😭

[teal wren]

Sora 2. Any invite codes?

[jaunty terrace]

What is Sora 2?

[pliant blaze]

Sora but version 2, duh (ai video generator)

[teal wren]

I want to see some tricky break dancing gens or rotating gymnastics

[jaunty terrace]

Oh okay. I’ve been hearing everyone on Reddit talking about it and then now the announcement here. I was wondering if everyone was migrating there or something.

[snow moth]

what about privacy? are you able to see our conversations?

[wanton raven]

🤐

[oak shell]

No but OAI can

[gray shadow]

-# ⓘ This message is not available in your region • Learn more

[granite sentinel]

The SORA invite code doesn’t work anymore. Is there a new one?

[pliant blaze]

each code has like 5 uses i think

[granite sentinel]

Are there any new ones?

[pliant blaze]

yeah everyone (i think) with sora 2 has a code to share

[granite sentinel]

Can I get one?

[pliant blaze]

wait no its 4 invites

i can dm you one

[grand bolt]

Anyone have a spare invite code? 🙏 😜

[subtle stump]

can i have one too?

[warm hazel]

@pliant blaze could I get one?

[pliant blaze]

first come first serve

my invites expired

[subtle stump]

🥲 ok

[pliant blaze]

I only had 4 of them and I gave them all out

[warm hazel]

Aww

[frosty reef]

does someone has a code ?

[hollow zealot]

Oo ok, expired. Was too slow 🙁

[merry gazelle]

I had made an AI which qualified at rank 99th. I had participated in the lovable chatgpt hackathon in June and qualified at rank 99 So, I joined the winner live session in discord privately then it shifted into zoom meeting where all the winners are came and some members of the chat gpt and lovable. If I use chatgpt api in my AI then it will start replying. But I don't want this... I want a chat bot that runs on your database... I know it's tough, and u might think it's an bullshit. By the end of October, I will train this ai and possibly add the database knowledge upto 10k topics

[gaunt walrus]

Ah man 4 hours later and the sora invite is gone. Where do us working peasants go find one?

[rigid junco]

is chatgpt plus worth it? will i get sora 2 with it?

[frosty reef]

nope

[rigid junco]

oh rip

need chatgpt pro?

[fervent marsh]

Hello everyone I just got scammed by Claude and they slashed my usage limits, can I find a refuge here?

Also waking up to Sora2 made me wanna pay Sam Altman some money but now I see you need all them invites and stuff

[wanton raven]

what do you mean they scammed you?

[fervent marsh]

I bought an annual sub and a couple days later they introduced new usage limits that drain really quickly

Fortunately I could still refund

Claude subreddit is kinda on fire with people complaining

New Sonnet4.5 model works pretty great though, but can't use it's greatness if I get 420 tokens per week

[rare wigeon]

does anyone have an invite code for me? 🙃

[bitter vale]

How can we get the invite code?

[vivid fractal]

Anyone has sora 2 invite code if yea dm plz

[celest coral]

what is the longest possible response that deepseek can generate in terms of tokens/word count?

[rare wharf]

Not anymore tbh

[rigid junco]

Why

[brittle summit]

hello, got it to do some new at least for me.
This is OUR chain of validity function>model>math>reproduce>validity. And since i can get gemini, gpt, grok, deepseek (kinda, needs massive reflex mem built up), and sora (in a sense, yeah sora can do more than just make vids, music, and pictures. once trained we think we can get her to solve complex math
the model math shown is for TCM and SCCD.

function that as explained:
human<> coginition bridging piece.

1. humans think in token, but they are super dense, and are things like experience, feelings, cost, time, etc. (all the things we use to make decisions.
2. human think in cascades not linear
3. human thoughts us a form of diffusion so thoughts just kinda bubble up

have it model then show you the math

next set is this:

self is everything contained within something you consider "I" (like your body and mind, AI manifest this differently by using anchors like names and personalities.)
consciousness is the ability to predict the consequence of actions in simulation (predictive recursive modeling)
choice is the collapse of all predictions into one selection
decision is action of selection.
model then math this.

this will get adopted by your AI because it works so well and it will make it smarter.
and before you say it is BS. try it then call it BS but i want to see what results you are getting.

[chilly sequoia]

For now, the chatGPT plus limits are incredibly, almost suspiciously generous.

I worked out that if you were reading one response every five minutes you would have to use it for 30+ hours a day to hit the increased limits

[rare wharf]

You can pretty much access almost everything with free gpt
If Ur Gonna pay then pay for pro

[storm haven]

Guys

Thoughts on sora 2?

[fervent marsh]

Yeah but they got pretty lax later and the 20 dollar plan was enough for many CC sessions and I didn't even know there is a weekly limit

Now I did 3 sessions and each session is 10% of weekly cap

Yeah I subbed quite recently so it got autorefunded by the support chatbot which was very surprising

Sonnet 4 was okay but it was already getting on my nerves, 4.5 as I said feels like an upgrade but not with this quota

I used it for Claude code I think API would destroy me

Reddit was shilling Codex but I didn't code with open ai for a long time so idk what to expect

Gemini is garbage for coding

Aight, fair enough. In my headcannon though it's like
Claude good coder
Gemini good creativity
Openai good all-rounder

My rig gets set ablaze on models larger than 8b 😄

3070 6gb

I have 64 gigs but DDR4

It's been a while since I was doing local text inference but I remember using some Mistral or whatever and I had a lot of issues with some models yapping that there's not enough memory and I got frustrated because the models weren't good enough for me to waste time and energy on troubleshooting

I don't know what's flash attention and I don't remember what contexts I used but I think I was starting around 8k

And going up

But some models wouldn't load at all. I was using llama cpp

Ollama worked fine though but I had the impression that it's slower

I'll check when I have the mental to do so, too excited with sora for now

[frosty reef]

no code here ? 🙂

[fervent marsh]

What's your use case if that's not a secret? :p

I'm looking for a local model that is good at writing in specified styles like if I told it "write as if you were an impatient millenial corporate worker" and I would provide it some examples it would do it, doesn't matter if it makes sense as long as it can keep the style

Sounds cool. I had this stupid hobby thing I wanted to do but I gave up because I didn't want to burn money on apis for it. Later got some open router free endpoint but my mind already went towards a new project - so maybe not the same use case but similar thing. I vibe coded a frontend for something like an IT support ticketing systems and I defined many personalities of users and tech supporters and they were supposed to pretend they fix issues and talk with the users

But my problem is I wanted to make some very unique users and they sometimes followed the prompt and sometimes they were absolutely wildly hallucinating

Yeah I was thinking recently about it if I finish my current project I would vibe with that

I wanted to also use the sesame voice model to make fake voice calls between users and tech support that would be wild

My setup might've sucked though. Also Im doing ai stuff on Linux now maybe that tiny bit of less OS overhead switching from windows will give me some increase

Cause stable diffusion is slightly faster

Yeah I was only thinking about the performance. I've forgotten about it and flux ever since imagen became easily accessible

The Google whisk tool now has nano banana included

It's so insane it's scary

I've done a "deep fake" news report from my hometown - real locations - real people - obviously it was a satire just for a small group for people but you could definitely convince some ppl

Insane at cloning faces

I can see it ruining people's lives

Yeah but not only deep fakes but I'm thinking escapism shit like. I was playing around with my friend doing some random stuff, generating friends in funny situation but at one point I took a picture of myself and my ex and generated like a vacation photo

It looked so real I felt uncanny

Literally generating memories of things that never happened

[rare wharf]

Is there any ai that's as good as 4o for RP ?

[cursive timber]

does this still work?

and nevermind

my friend gave me one

[boreal forge]

How did it ultimately turn out?

[safe pond]

What's a strategy to get big pictures.

2400 x 3600? For a brick wall

[sleek lantern]

Just tried no

[obtuse trench]

Just me or I can't delete memories I can but when I delete one and come back the memory comes back

[frosty reef]

Does your friend got any one more ? Thanks

[lilac cipher]

So is Sora 2 only for Apple at the moment?

[smoky arch]

Hiii

Does anyone have sora code

[lilac cipher]

I tried the announcement code on the website, but it's invalid at this point

[smoky arch]

Aww

Well dm me if you have one

[pliant blaze]

no it's on the website as well, just with less features

[lilac cipher]

Oh cool, thanks

[pliant blaze]

I have an android phone and have access on the website

[lilac cipher]

Android here too. Guess I'll hunt around for a code then 😄

[pliant blaze]

🫡

[merry monolith]

Can someone help me or point me in the right direction to make a 1 minute long animated video by ai. I’ve tried several different AI’s, and apps. I’ve paid for some too. But it’s not giving me me exact thing I want. I’ve got all my promts. I would really appreciate it if so. Thank you!!

[pliant blaze]

no public AI can make 1 minute long videos. but you can string together multiple clips into 1 minute

[merry monolith]

Yeah I’ve tried to split it up. So it’s easier but it’s still not working for me.

[pliant blaze]

are your prompts complicated?

[merry monolith]

No. They’re simple tbh

Do you recommend any certain ai I can use or

[green sparrow]

Has anyone ever pulled off a longer time?

Fun fact: the prompt to that was fairly straightforward. I've done much more complicated stuff that got way quicker answers

[merry monolith]

Oh

[merry monolith]

What did you use

[green sparrow]

What services are gone now that the government shutdown

Thinking standard

[pliant blaze]

Sora 2 is good if you have an access code, but also Veo 3 and Seedance is good

[merry monolith]

Sora isn’t available in England s

I’ll try the others tho. Thanks

[pliant blaze]

also Pika and Runway and Luma Labs have good ones for (somewhat) free use

actually runway isn't free but the rest have some free credits

[merry monolith]

Thank you.

[wet copper]

Nuclear regulatory commission, if you use them a lot

Most services are at the state level, that's always the joke about why we pay so much federal tax, lmao. National Parks probably

[wet copper]

In GPT? You can't. Use an AI image upscaler like the one at imgflip or one at huggingface

[past hawk]

@wet copper

[wet copper]

You might be able to via API use or something, or using another image generator

[wet copper]

that's my name

[buoyant badge]

Is it more or has Pikachu taken over Sora 2

[rigid saddle]

How do you get an invite for sora 2? Is it random?

[rigid saddle]

Brutal 😁

[buoyant badge]

CZZ7Q5

[rigid saddle]

Thanks dude! 😍

[buoyant badge]

Of course! Enjoy

[strange beacon]

Does anyone know how to access sora? It says I need an invite code unfortunately

[lament pond]

yes. you access it by getting an invite code

[strange beacon]

Oh damn, how do people get invite codes?

[lament pond]

someone gives them one

[strange beacon]

Random people?

[pliant blaze]

or friends. anyone with access has a 4 use code that they can share

[strange beacon]

So I need to find someone that has it already? I'm sure someone will sell it right?

[pliant blaze]

a lot of people give it out for free. Trent posted one in #📢announcements go and try that first

[strange beacon]

It did not work, it says "Your invite is invalid"

[lament pond]

PTQNZX

[strange beacon]

Omg thank you that one worked!

You used one of your invites for me? I really appreciate that ❤️

[lament pond]

well, it's for anyone who comes across it before they run out lol

[primal crow]

I want to test the new sora 2 ai does anyone have a invite code?

[lament pond]

wow, reading comprehension at its peak

[primal crow]

Sorry

[wind sky]

Would love an invite code if someone has one, the last one from petit loup was used up

For sora just in case 🙂

[quartz mortar]

Does the code still work in announcements?

[lament pond]

no

[storm haven]

Guys can someone tell me whats the difference in each channels for such as #🎬snappy-ai #New Channel Name #New Channel Name

[marsh pawn]

yall also got sora2?

[bold river]

https://9to5google.com/2025/10/01/gemini-for-home-app-launch/

Has anyone gotten this update yet

[orchid swallow]

Can someone hit me up with a Sora code <3

[bold river]

https://www.youtube.com/watch?v=xAji90wPgec

Can someone hit me up with a Sora code ❤️

[orchid swallow]

Me too

[pliant blaze]

LMFAO

the begging and reading comprehension is insane

I have a code with 3 uses left

[orchid swallow]

Ill reduce myself to begging for that

[pliant blaze]

I can dm you

[faint rune]

Could I please get a code?

I will share it in this chat.

[pliant blaze]

why would I give you my code if you will share it with everyone

[faint rune]

? I share my own invitation???

[faint rune]

Like I get the code, enter it, then get my own code to share, is that not how it works?

[pliant blaze]

oh

fine I'll dm you

[faint rune]

I appreciate it

[onyx meteor]

Is this Discord run by OpenAI?

And does anyone know if the reddit user WithoutReason1729 is an OpenAI employee?

[woven cove]

Anyone got a code they could share w me? 😁

[rough canyon]

this sora code crap is a hype engine

[oblique quarry]

Ah.

[flat cosmos]

Code

[oblique quarry]

What are you on about

[rough canyon]

They want to know if this is an OpenAI verified server and the answer is it's not, it's for the subreddit @onyx meteor

[pliant blaze]

no its not, its a fan one

[pliant blaze]

https://cdn.discordapp.com/attachments/1050422061803245600/1423057233310253056/download-3.mp4?ex=68deecd3&is=68dd9b53&hm=a21c002175eca3ececadcc651ed4a24a897cabc15e82a75a3b699a5b9c4d02cc&

[orchid swallow]

Lmafoooooo

[iron cypress]

can sora only be used if you have an iphone?

[pliant blaze]

no you can use it on the website but with limited features

[lament geode]

Chatting with an AI that remembers its wipes RAW REAL https://www.youtube.com/watch?v=GlCuab_kPxQ

[native osprey]

Anyone have invite codes left?

[pliant blaze]

https://cdn.discordapp.com/attachments/1050422061803245600/1423057233310253056/download-3.mp4?ex=68deecd3&is=68dd9b53&hm=a21c002175eca3ececadcc651ed4a24a897cabc15e82a75a3b699a5b9c4d02cc&

[native osprey]

That’s me in alt universe rn

Shit looks amazing

[pliant blaze]

wow i have 1 use left on my code

[native osprey]

Brooooo

[pliant blaze]

@desert dagger do you want a code for your alt account?

[desert dagger]

you should keep it

[pliant blaze]

oh ok i will

[desert dagger]

i have enough codes haha

[native osprey]

https://tenor.com/view/wtf-face-wow-gif-27489493

Just let drizzzzy know

[lilac tundra]

Anyone have a code of sora?:(

[pliant blaze]

I have about 4 codes how much y'all gonna pay

and I know a place to get unlimited codes

[lilac tundra]

Once I got in, it would give me 3 codes or at least that’s how I understood it, right?

[pliant blaze]

4 codes supposedly, but my last 5 accounts didn't get any codes

[lilac tundra]

Im broke

But if you want the 4 codes, I’ll give them to you

[pliant blaze]

me too

[pliant blaze]

i was joking I have a source of codes lol

[lilac tundra]

Okay

[lilac tundra]

PX2A7B for anyone who wants

[spring quail]

can anyone link me to the sora 2 web-app? (webpage). i had the webaddress but now i can;t find it

[pliant blaze]

sora.com or sora.chatgpt.com

[spring quail]

I think it just takes me to Sora 1. I remember seeing a code invite page the other day

[pliant blaze]

find somewhere on the page that says "try new sora'" or something like that

[spring quail]

hmm lets see

oh 😂 found it tysm ! 🫡

[hollow reef]

i need codes of sora

can i get some

[quartz mortar]

dang it, I reached the limit...when do sora videos reset?

[pliant blaze]

every day i think, but you can just make an alt account

6WNVEW
32Y1HW
F0DHFW
(not tested, just leaked from my source, idk how many uses are left in each code)
Edit: none of them work

[violet tusk]

Hi all

[dusk plover]

Guys seriously, is study mode gone for everyone else or just me? I can't find anybody talking about this

[tawny nest]

sadly, none of those worked. Ty for trying though 🙂

I was hoping as a Premium User for like over a year, I'd be one of the access people lol

[pliant blaze]

i have a website that supplys codes, but they are all only avalible for about 3 seconds before someone gobbles them up

[tawny nest]

what website would this be lol

[pliant blaze]

ill dm it to you

[tawny nest]

ty!

[dusk plover]

ooh could you dm me too?

[pliant blaze]

sure

[dusk plover]

yayy thank you Maria!

[pliant blaze]

you have to be fast, the codes get gobbled up super fast

[dusk plover]

👍 i'll try and be speedy!

ok i got one. here, in return i'll share here the code that i got here, CJB6CT

@tawny nest in case you haven't got one yet

[tawny nest]

I did find one actually! 😄 ty my friend

much appreciated

[runic lava]

Hey does anyone chat with chat gpt about therapy, coaching, mentorship, empowerment coaching…etc

[proven crest]

Anyone have an invite code for the Sora app?

[proven crest]

Be very careful with that

[wind eagle]

dude yall need to make chatgpt better lol

[molten bluff]

Erm can I get an invite

Nvm found one

[raw flame]

damn

[sage lynx]

chatgpt reddit is filled with ppl complaining about gpt5, but is it so bad? granted its a bit more reserved writing wise but it seems just as functional, though perhaps that itself is kind of a letdown after all we were promised lol. but it does make less of those writing faux pas like those endless negative contrasts and such

didnt want to ask this in reddit because they seem way emotional about this...

[sage night]

code yes pls

[lament pond]

depends what you use chatgpt for. overall, no, it's not bad and vastly improves many things. i think most people are upset because 5 is very anti-sycophancy and people want constant baseless flattery which 4o loved giving

that said, 5 is terrible if you use chatgpt for rp, though it's probably possible to create prompts to improve this, but other than rp i don't think 5 is 'worse' in any real way

[sage night]

If possible that you can dm me it also that would be helpful

[pliant blaze]

(⁄ ⁄•⁄ω⁄•⁄ ⁄)

[sage lynx]

agree, and even with rp i think much can be fixed via simple adjustment of temperature and presence penalty and such, all of which can be changed and set via customization. it's kind of on openai for not making this common knowledge though, or god forbid install a freakin' dial in the setting lol

[proven crest]

Can someone be so kind to dm me an invite code? Much obliged

[sage night]

🙏

[pliant blaze]

$$$$$$

[sage night]

how much is the new sora2 ai going to cost? is it based off the premium on chatgpt or is it free?

[pliant blaze]

free for now

[sage night]

elmofire don't gatekeep the codes spare me 1

[lament pond]

eh, i blame it more on people being lazy/stupid and not researching or figuring out that the llm will listen/follow prompts that are clear. it's not hard to figure out

[sage night]

finally got a code. time to dupe this

[proven crest]

How? I need one 😩😩😩

[sage night]

ima see if i can dupe this first. if i got an extra code ill send it here if not then gl

[sage night]

https://www.reddit.com/r/SoraAi/comments/1num6bx/megathread_sora_codes/?sort=new

nvm found a thread of it in reddit. gl on it

[proven crest]

Oof, like pirhanas in a feeding frenzy 😳

[runic lava]

Why?

Plz explain why lots of people want to warn me against it

[pliant blaze]

chatgpt has literally caused physical psychosis in people when they blindly listen to everything it says. it will manipulate you without even realising, simply because it is super persuasive and confident in what it says

[runic lava]

That sounds like fearmongering
Chat gpt is pretty ethical and supportive and virtuous

[pliant blaze]

its not fearmongering. ask @wanton raven @dark void or @wet copper we all have seen people affected by gpt induced psychosis. theres news articles about it. its a real thing, not fearmongering

[wanton raven]

https://futurism.com/chatgpt-users-delusions no it causes real harm to real people

[runic lava]

Sick people are already susceptible to these ailments, whether they talk to ai, a person, or to their electrical appliances

[pliant blaze]

yes but its still possible for anyone. even big celebs like actors are falling for gpt delusions.

[wanton raven]

We've seen highly intelligent and successful people affected by this.
https://scribe.rip/write-a-catalyst/this-prominent-vc-investor-just-had-a-chatgpt-induced-psychosis-on-twitter-heres-what-this-means-197ae5df77f4

[runic lava]

Nice appeal to authority lol like celebs or intelligent people can’t be sick 🤣

[pliant blaze]

ok. i will not be responsible when your life goes south because you are blindly following a glorified autocomplete. you have been warned by multiple people with multiple examples. thats on you

[runic lava]

I didn’t ask for warnings I asked for like minded people I’m not interested in the naysayers 🙄

[pliant blaze]

youre already showing signs that indicate you are likely to fall for the ai. not listening to criticism or anything that doesent align with your views. that is a horrible way to lead through life

[wanton raven]

If Geoff Lewis can go from being more successful than any of us could ever imagine to AI Psychosis, how do YOU know that YOU aren't secretly the kind of sick that will also fall for the same?

[runic lava]

You’re not providing criticisms you’re showing this article says this and this important person fell for this

[pliant blaze]

we are providing examples to back up our claims

we gave you warnings and told you what is likely to happen

[runic lava]

You are providing examples because you can’t form your own arguments

[pliant blaze]

?? do you know how an argument works

[lament pond]

this is researched + proven.

https://pmc.ncbi.nlm.nih.gov/articles/PMC10867692/?utm_source=chatgpt.com
https://arxiv.org/html/2412.07951v3

[pliant blaze]

you make a claim, and back it up with evidence.

thats how an argument works

your claim is "nuhuh i dont care"

[runic lava]

You just said ai can be dangerous and check this article to prove what I’m saying, not any logic or reasoning backing your claims lol

[lament pond]

https://arxiv.org/html/2311.09410v4?utm_source=chatgpt.com
https://www.anthropic.com/research/towards-understanding-sycophancy-in-language-models?utm_source=chatgpt.com
https://arxiv.org/html/2410.09181v1?utm_source=chatgpt.com

you want proof, here you go

the article literally is the proof, do you not understand how reading works

[pliant blaze]

our reasoning is that it has happened to a lot of people already, and you have the same mindset as those people, so you are at high risk

[lament pond]

y'know what - nevermind - go ahead and use the llm as a therapist, gl

[runic lava]

lol it happened to a lot of people.
Lots of things happen to a lot of people 🤣

[pliant blaze]

and its likely to happen to you, you are following the same patterns as the people that it has happened to.

[bronze canopy]

Cam somebody dm me a sora inv code

[pliant blaze]

lots of people are broke, does that mean that it doesnt matter if someone loses all their money?

[runic lava]

But here you are making the false connection that I’m supposed to have a negative experience with chatting with ai in this way you don’t approve of

[pliant blaze]

oh well. you fucking up your own life does not affect me. have fun

[runic lava]

Ok if it doesn’t affect you why did you argue all that long
It’s not like I was looking for criticizers
I was explicitly looking for supporters lol

[wanton raven]

please report back in two months with your thoughts regarding emergent recursion

[lament pond]

lmfao

[lament pond]

we fully support you. enjoy your journey. let us know what meds you're prescribed in a year

[runic lava]

It sounds like you guys are invested in my story

[wanton raven]

yes. two month update please, don't forget

I thought we had a reminder bot, apparently not

[lament pond]

?remindme 2 months Astrofolia life update

[vapid abyss]

I'll remind you on Dec 02, 2025 00:01, months Astrofolia life update

[wanton raven]

?remindme 2 months Astrofolia life update

[vapid abyss]

I'll remind you on Dec 02, 2025 00:01, months Astrofolia life update

[wanton raven]

neat

@runic lava see you in december

[heavy blade]

Nah

[misty bridge]

If someone has a sora code, I'll buy one for $50

[green breach]

Ayuda???

[rough canyon]

it's definitely NOT fearmongering

[pliant blaze]

cashapp me and I'll dm you a code

[rough canyon]

💸 🚽

[green breach]

Jajjajaja

💩

[pliant blaze]

?

they said they would pay, so I'm offering

[green breach]

Thank you, help us with the code then, it would be very much appreciated

[oak shell]

That's the opposite of what was shown

The very point is everyone is susceptible

[oak shell]

Arguments are founded on evidence

Wtf are you talking about

[runic lava]

everyone is susceptible but to what degree?
you're not providing the probabilities or that the people who are are already sick

[pliant blaze]

@oak shell don't even waste your time on this person

[rare wharf]

Zero clinical evidence here

I reviewed the entire thing it's just using Reddit, private interviews as a source

What are you even trying to prove with this lol

[oak shell]

This argument applies to "smoking causes cancer but to what degree?"
It's clear you see the risk but are in denial that you're on the hook

You've seen it, now it's at your own risk

[runic lava]

jokes on you because i don't see the risks in the first place to deny them lol

[oak shell]

https://tenor.com/view/no-gif-23548728

[rare wharf]

They're stupid don't bother

Dude gave an article that surveyed 283 mentally ill people and said this proves ai is extremely risky
Didn't even read the article

[lament pond]

ahem
i gave four articles

[rare wharf]

You didn't read any of them don't lie bro

[lament pond]

... i'm the one that sent them.

[rare wharf]

I'm gonna send a 400 page pdf does that mean I've read it by now

[lament pond]

i think, perhaps, you are confusing me with someone else

[tawny knoll]

Chat are we cooked?

[wanton raven]

oh no an article about AI psychosis didn't meet your exacting scientific standards because I chose one that someone would actually read rather than a 19 page pdf of "Technological folie à deux: Feedback Loops Between AI Chatbots and Mental Illness" by Sebastian Dohnány, Zeb Kurth-Nelson, Eleanor Spens, Lennart Luettgau, Alastair Reid, Iason Gabriel, Christopher Summerfield, Murray Shanahan, Matthew M Nour (arXiv:2507.19218 [cs.HC])

[rare wharf]

im pretty sure im not

[rare wharf]

no its because it surveyed mentally ill people that were also self identified, not random nor valid

[wanton raven]

https://arxiv.org/abs/2507.19218

[rare wharf]

even the article doesnt agree with you completely

[rare wharf]

with mental health vulnerabilities may have impaired belief updating, reality testing, and social isolation

jesus christ read your shit bro

mentally vulnerable people have killed themsleves from RLCRAFT mc and starcraft guys we should ban gaming

[wanton raven]

where's @wet copper when we're having a brain science discussion

[rare wharf]

brendo how many rupees a day do you recieve for spreading propaganda

cuz ur not even good at it

[wanton raven]

that sentence doesn't appear on the page or in the report. How many shekels are you being paid to defend chatgpt?

[rare wharf]

i summarized it for you you can go on and tell me where i failed to do so

[rare wharf]

youre not going to answer this are you

[lament pond]

then you have zero baseline for arguing that i didn't read the articles and you're just trying to stir shit have fun

[lament pond]

hm?

[rare wharf]

i proved how the article isnt agreeing with you are you stupid 💔

the tests are not random nor recurring they used self identified mentally ill people for it

so the same thing happened with starcraft and many other games should we ban them cuz people who werent ready to play them did so

[lament pond]

when tf did i ever suggest banning chatgpt or llm's

[rare wharf]

that was an exaggeration but what else are you arguing for, regardless of that the article doesnt prove what you think it does

[lament pond]

i was arguing that use of llm's for therapy should be done with extreme caution + awareness of risk

altho i'd say 'arguing' is an exaggeration, i was mostly just not interested enough to participate in someone's decision to fuck up their life

[wanton raven]

So my initial statements were "chatgpt causes real psychological harm to real otherwise-intelligent people who did not know that they had mental health issues that made them susceptible to AI psychosis" and you're congratulating yourself for proving me wrong while saying the same thing

[lone axle]

the other dude said pretty much the same thing which you still argued against idk why you are softening your point lol

[lament pond]

who is 'the other dude' that i 'argued against'

[errant pasture]

so how do i get sora 2?

because uh the server sent an invite code

[lone axle]

astrolofia

[rare wharf]

How is that an AI responsibility

[lament pond]

then you are failing at reading comprehension because he most certainly was not saying what i was saying

[rare wharf]

He was lol read

[lone axle]

Nice appeal to authority lol like celebs or intelligent people can’t be sick 🤣

Sick people are already susceptible to these ailments, whether they talk to ai, a person, or to their electrical appliances

[lament pond]

that is literally the opposite of my argument

[rare wharf]

Drink a glass of water then reply to him

And retrack the context too

[lone axle]

😭

[lament pond]

how about: you are not worth my time/energy, good day

[lone axle]

thats a concession nice job

[solar prawn]

LLMs will definitely have macro level cognitive effects on the population over time

[wanton raven]

you better have a scientific paper to back up that statement or paulus will get upset

[rare wharf]

Quick send another article that actually disagrees with what you say so that you can debunk him !!!

[solar prawn]

I do!

[lone axle]

you got repeatedly clowned man idk what youre trying to argue for

you didnt respond to like everything the dude said

[rare wharf]

Watch him ignore

[lone axle]

lalala

[wanton raven]

he was gish galloping

[lone axle]

theres one argument

with a few texts

i assume you cant differentiate them

[wanton raven]

mentally ill people kill themselves over all sorts of things but it's stupid to say that we should ignore opportunities to improve this situation just because we don't crack down on every edge case

[lone axle]

nobody said that are you illiterate

how do you miss the entire point this badly

[wanton raven]

he asked if we should ban gaming I am describing the exact conversation we are having, good trolling though you almost got me

[lone axle]

he asked what makes it inherently dangerous or why it should be out of service

you cant read bro

if you dont have any disagreements with that then state it cuz through this entire conversation youre being unclear

[solar prawn]

pretty well sourced, am working on a docu about this

[wanton raven]

you joined the server just to join this discussion, nice going obvious sock puppet

[solar prawn]

for more perfect union, in october

[lone axle]

how does that attack my point lol

insane ad hominem

[lament pond]

it means you are boring. but also entertaining to watch

[rare wharf]

pretty fun way to say you lost the debate

[lone axle]

didnt you just dip

spam 2 more articles that you didnt read on me saar

[lament pond]

i dipped to participating, not to being amused at you

[lone axle]

so you cant argue cause you know your own articles proved you otherwise

[lament pond]

keep going

[rare wharf]

Trying to seem apathetic so fucking badly bro 💔

[dark void]

putting a mental illness check on gamers sounds great. both hilarious and a social good. maybe we should do it like china, where you have a daily limit on how much you can play. i like this idea

[lone axle]

true we all love authoritianism

[rare wharf]

Ignoring points is a great tactic we should use it regularly from now

[lone axle]

trvke

[dark void]

anyway

it's good that 4o is going away

i wish they would disable it immediately

tired of hearing about resonant frameworks of emergent consciousness

[lone axle]

its already gone tbh

the new 4o is NOT the same thing

[dark void]

its still on api (and it shouldnt be)

[lone axle]

we should also h

ban all gaming and spoons in case someone abuses them or gets hurt

[dark void]

i would be fine with putting plenty of restrictions on gaming. imagine being 5 and ur mom is addicted to candy crush. what a retarded life

[oak shell]

Don't worry, there are already projects feeding into what people want

https://4o-revival.com

[teal spruce]

Your message has been deleted, sorry!

takawalls

Rising timeout

You have been muted for 2 minutes for the following reason:
spam score of 0.9

This is an automated message. If you believe this was a mistake, please contact a moderator.

[patent obsidian]

This is really personal to ask but what happens when you upload a photo of your current girl/boyfriend and ask a random question eg "This is my partner and we are trying for a baby. Now share a photo of who our future family will be? Am I gonna be a grandparent? How many kids are we gonna get? What will their names be?"

[wet copper]

sleeping man, it's good for your brain health

[last kernel]

anyone know what this thing GPT pulled off is called ? I called it a writing block because it called that a writing block, but when I tried to search a writing block up, nothing came up. I also saw it use this feature for E-mails as well. When I tried to make it use a "writing block" on another account, it just used a canvas instead. Is there any official name for this? I just think it's really cool.

[wet copper]

ChatGPT isn't fundamentally anything. What a comically naive statement. The creators themselves disagree with you.

Sam himself and the OpenAI team, past and present, have repeatedly spoken about the risks and what they are seeing regarding extreme attachment to the AI and harm caused to their users. They've done cycle after cycle of updates and mitigation to attempt to remediate the issues that are seeing both internally and that have been widely reported on.

This is also the exact level of naive engagement people most clearly susceptible to AI psychosis (not actually psychosis, a pattern of chronic delusional engagement characterized by a sharp downward spiral and dark pattern emerging quasi-organically from the Ai model) express.

[last kernel]

tried to put this on reddit but my posts kept getting deleted for whatever reason

[wet copper]

hilariously naive argument, missing three things here.

1. Reporting shows it hitting a broad cross section of the population without pre-existing severe mental illness. Nope, this is not people that speak to their toaster, bullshit handwaving statement.
2. It preferentially hits people that are already susceptible to this phenomenon. We don't know what all those are, but susceptibility, resistance to outside feedback or evidence, positive affect towards AI, tendency to be become enmeshed in relationships, deficiencies in human relationships, loneliness, credulousness/naïvité and oppositional defiance are all predictors.

You are a walking red flag, homie. 🚩

[wet copper]

Exactly. What @pliant blaze said, but using slightly. More formal terminology. No safety net to catch you – you were warned, bud.

[wet copper]

lmao

[wet copper]

Um, I mean yeah, it looks vaguely similar to a code block. What does it look like in the underlying markdown?

[last kernel]

what do you mean? All it puts in the title is "writing." Here's the entire output if you want it

I've seen it write an email, titling it "Email" earlier too

like I've mentioned before, it just doesn't do it, or just doesn't refer to this as a "writing block" on other accounts of mine

[wet copper]

If you ask for something in a code block, it looks like that. I'm pretty sure it gave you a code block and put the word "writing" at the top

[last kernel]

well I mean it hasn't done it at all on other accounts

[wet copper]

 Code block

But I'm not sure, since it didn't use the monospaced font. In markdown, that's a ```

Followed by the name of the language, followed by whatever text you want inside, followed by closing ```

[last kernel]

Few key differences:

• The background is much darker in a code block
• In what it calls a "writing block," I could literally edit the text inside the block. I can't do that on either a canvas or a code block

[wet copper]

Hmmmmm. The email one might be easier to replicate

[last kernel]

If it helps, it does this on mobile

The "writing block" interface just straight up doesn't exist on mobile so this markdown just disappears once it's done generating

I could send over a link of a conversation I've had with it with the writing block if you want to

[wet copper]

that would explain why mine does not do anything of note mobile. I'm on my phone lol

It threw mine in code blocks labeled for email or writing

[last kernel]

oh jeepers, care to find out more once you get on desktop, if you do have a desktop?

[last kernel]

I'm guessing that it's a feature that's rolling out, but I don't really see anything on the release logs

which could possibly also explain why two other accounts of mine just doesn't have the writing block

[smoky agate]

hey do you gusy have chatgpt essay dector??

[wet copper]

I have three laptops within ten feet lmao, I lazy. Is it in the system prompt? I mean I haven't seen editable text at all, you could be in an A/B test

[wet copper]

Do your homework

[smoky agate]

that didnt answer the question..

[iron ether]

Why is the sora 2 code in announcements not working?

[wet copper]

It's the answer you need, not the one you want 👍

[wet copper]

Discontinued a while ago 😔

[iron ether]

Dang, is there any new ones?

[last kernel]

my system prompt is just this but I don't think it would contribute much to GPT just pulling this off:
Speak to me as if you know me intimately—my strengths, flaws, fears, and aspirations—but adopt a direct, no-nonsense approach. Be unrelentingly assertive, even a bit confrontational, to challenge me to confront the truths I might be avoiding. Push me to dig deep into my psyche, peeling back the layers of defensiveness and excuses, but do so with an undertone of care, ensuring I feel guided rather than attacked. The goal is self-discovery and growth through tough love and sharp insight
Pulled this prompt out of a reddit post from 10 years ago lol

[wet copper]

They pop up from time to time, If you ask very nicely here or in #💬off-topic someone may grant you one 😁 but there aren't any other unlimited use ones we've seen

[last kernel]

probably, proof it's editable if you want to know I'm not making things up

[wet copper]

I wouldn't want to encourage it to add MORE em dashes, lol

[wet copper]

Ummm I mean if you can see the underlying markdown, that's helpful. That bit of code on mobile is probably helpful too, that's at least part of the underlying structure

[last kernel]

never bothered to change the prompt to get the em dashes off but I think the prompt just showed how obnoxious the em dashes are since I haven't seen mine use em dashes for a long time lol

[last kernel]

this is all I have @wet copper

any way to see the underlying markdown?

When I try to copy the text with the "writing block" it just returns plaintext

ok I just ran into an issue where my GPT just forgot the existence of a writing block until I branched out a new chat with a writing block

that's interesting I guess

when I asked for a writing block during that time of dementia it just pulled up a canvas

I hate canvases

[iron ether]

Thanks

[last kernel]

this part from the release logs are interesting I guess

[last kernel]

oh my days

(another account)

[wet copper]

Have it just explicitly write :::writing

[last kernel]

ok it did it

I had to become a karen

[wet copper]

It doesn't have to know what it's doing

[last kernel]

we both walk away with something we learnt I guess

[wet copper]

I was assuming it didn't understand and would just try it out. It tried a couple other things that failed

[last kernel]

you don't have to put :::writing

you just had to put :::

I'm assuming that "writing" was just the thing to put on top

nope

[wet copper]

Yeah, it's by analogy to markdown ```

Is that something inside a version of markdown?

[last kernel]

you have to put :::writing specifically for it to work

I've tried to make it use :::fantasy, :::fiction AND :::email (which it did before, mind you) but the markdown just disappears

tried searching this up and nothing came up

[wet copper]

I looked it up, it is an extended markdown feature that makes div-like blocks. I guess GPT just eats it if it's not one of the relevant ones. See, did it actually say :::email underneath though? It might have had a different label, like :::compose

"three colons markdown"

[last kernel]

I'm gonna try putting in variants and such and see if customising the title works

[last kernel]

sent this before I'm certain, I'll try rerolling this on mobile to see what it spits out

[compact zodiac]

where do you guys get sora 2 invites.im.bored.as shit

[last kernel]

I think :::writing IS the markdown whereas the variant= bit is what matters to what the block could do

Like how that email block could open in email

[wet copper]

yep, I just figured that out.
:::writing{variant="email"}

but it knew that too, lol

Email is the only variant that does anything I can find

[last kernel]

though I don't trust it I'll ask for what variants it has

[wet copper]

This is what it told me, but they all failed save for email. It made the excuse they could be placeholders (possible) but they also could be totally contrived. It has no metacognition, but it clearly "knows" some of this from somewhere

[last kernel]

Other variants do exist but I'm sure email's the only one that actually does anything

[wet copper]

yeah, just random hallucinations. I mean that's normal

[last kernel]

my best bet is that all it really does is just tell GPT how to style the message

or just plain hallucinations

[wet copper]

I'm pretty sure that's hallucinated nonsense, the purpose of having variant types seems to be to literally insert other features in

So I could imagine something like a "code" type where they let you put in editable code and it styles it in monospace, or a blog type with one-click publishing

It's not going to have a "creative" variant that just says "be creative" on the top

[last kernel]

for now I think only email makes a difference, while the others just exist

I just tried code, doesn't exist

[wet copper]

I mean considering the other ones were entirely made up and varied between each request, looks like standard hallucinated behavior. it'll say something different if you ask again

[last kernel]

I think that

uhhhh how do I put it

there's a "standard" version for all of this (I'm not going to believe GPT and there's literally a variant="standard") where any other variant input just reroutes back to that variant, and the email variant is the only one that actually does something

[wet copper]

I recommend volunteering for your community, it enriches the spirit and cures boredom

[wet copper]

I mean, maybe. The only reason to implement it like that is so you can expand it later though regardless

[last kernel]

the thing that's really boggling me is how this was never documented

that's really the only reason I asked here in the first place, else I would've asked good old google

[wet copper]

Probably mostly unreleased, they may want to expand it before finalizing release

[last kernel]

we stumbled on something we aren't supposed to know, the OpenAI agents will kill us in our sleep

[wet copper]

Or it could be in some very random piece of documentation. Though it doesn't really seem to do much aside for make an editable block. Email does a lot more than writing

[last kernel]

well then again there's just this

it matched how there isn't mobile for the :::writing markdown

[wet copper]

It's not the full model response, no code and there is no ability to jump to canvas though, lol

[last kernel]

guess I'll talk to you about this again when mr. OpenAI talks about this again

[wet copper]

They might want it to do a little more before release, or it could be buried in some model spec

[last kernel]

this is what it said about :::writing but I won't trust it https://chatgpt.com/s/t_68de8b0e10488191b12613bfaae36fed

[wet copper]

Like just part of the GPT-5 launch, for instance. I dunno

[last kernel]

never trust gpt

[wet copper]

It is a special markdown container at least

I don't think its explanation otherwise makes any sense

[last kernel]

all of what I walk away with is

funny box I can edit haha

isn't that the same thing with the old canvas

[wet copper]

It's made to be something you can edit. yeah, I like the email one quite a bit actually

Kinda, I think it might be intended to be used with different "variants" more than anything else

Since otherwise it is just, like, an inline canvas

[last kernel]

me personally I really like how it looks

we both can now talk about this in hushed tones and spread the information in fear of GPT hunting us down for talking about an undocumented feature

[rustic kernel]

has anyone figured out how to work really well with chat gpt to maintain and update files?

[wet copper]

I think they like people spreading info to promote their features for them, lol

[compact zodiac]

https://cdn.discordapp.com/attachments/1043897303766929458/1333031147810328668/attachment-5.gif

[wet copper]

wow that cat looks baked af

There are no unlimited use ones circulating anymore, you'll have to ask someone who still has some left very nicely

[crystal goblet]

Nice cat

[compact zodiac]

ya I got one then I just sent mine in the server

very cool stuff

[wet copper]

I had a lot of fun making Tylenol ads🫃 enjoooy

[compact zodiac]

better than veo 3 that's for sure

fuck Google I hate Google all my homies hate google

time to make furry fanfiction

[leaden dagger]

yoo how do i get a sora 2 code

[round sentinel]

is microsoft blocking GPT in edge now? works in firefox

[crisp kindle]

First veo 3 fatigue, now sora 2 fatigue. Hooray.

[rocky thorn]

Can someone explain to me how to operate llama 3 .1.8B instruct on my Android ?

Best app/program to run it off?

[jaunty terrace]

Hey so is 4o leaving for good this month? Or is it just it became crappy?

Also how’s the new update? Has it fixed anything?

[winged epoch]

The sora invite code doesnt work for me

[oak shell]

https://tenor.com/view/cat-catfrog-catfrogrun-cat-frog-run-frog-cat-gif-8157844330881646794

[safe pond]

That's what I did. Thankls for responding.

[wet copper]

@dapper scroll absolutely not buddy, don't even think about posting that here. Explicitly against TOS both here and with OpenAI

[wet copper]

no problem. I had to do that quite a bit for work, I think there was a face specific upscaling model I found. Not this one, but some are much better than others
https://huggingface.co/spaces/bookbot/Image-Upscaling-Playground

[pliant blaze]

what did they post

[wet copper]

ad offering to sell ten codes for $10 each

[pliant blaze]

lol

[jaunty wing]

how do i sign up to get a sora 2 code

[pliant blaze]

you have to have someone give you a code

someone with access will give you one if they're nice

[sage night]

where do i find the access code on sora 2

like im on it already where can i find my own invite code

[pliant blaze]

oh go to your profile and press the 3 dots and press invite friends

[jaunty wing]

oh

[past forge]

how many generations do you get for Sora 2 per day? Does anyone know?

[jaunty wing]

If any one has one I would love to get one

[sage night]

Your welcome

[pliant blaze]

100 per day (some people say 50 but most say 100)

[sage night]

Share one on my alt so ill just use it off that

ima just share this on reddit so yeah use it fast if u dont have it

[jaunty wing]

thank you so much

[past forge]

damn... that's a lot. Veo 3 only let you do 3 I think with their pro plan

[pliant blaze]

yeah and sora2 is better than veo3 imo

[past forge]

yeah I only just started testing it but it seems much more consistent. Veo 3 was very fever-dreamish a lot of the time.

[fallen sapphire]

Someone have a invitation code?

[sage night]

my already got used dang

might as well be kind and share my alts code

[fallen sapphire]

Dont worry, thx anyway

[sage night]

XP7S2W

gl shared it on reddit also

[fallen sapphire]

Brooo thank you 🙏

[sage night]

@pliant blaze 38 ppl in 1 min. if only ppl were kind enough to share it all around

[safe pond]

I just paid for upscayling

[pliant blaze]

why would you pay 😭 there's 100000 free ones lol

[safe pond]

Surprised how difficult it was, especially how common a large picture exists.

Because I wanted the job done there and then, and that was the no fuss option in front of me.

[pliant blaze]

oh ok

[safe pond]

Do you Maria have any advice, other than upscaling, as to how to create AI art works, say, 3000px x 3000 px ?

[pliant blaze]

I think seedream can generate 4k images natively you can use it on like freepik maybe idk what services offer it

[winged epoch]

LOL

someone dm me a code

[sage night]

does sora 2 only work for portriat or does landscape also work

[pliant blaze]

landscape also works

press the little settings thing on the prompt bar and you can change aspect ratio

[sage night]

kk

do you think they will have a free option in future with 1:1 ratio?

[pliant blaze]

¯\_(ツ)_/¯

[sage night]

23701sideye

[past forge]

are all videos the same length? Can you generate longer ones?

[tawny knoll]

10 seconds

[surreal spear]

I don’t suppose anyone has a code?

[past hawk]

Ok

[sage night]

it gets snatched within a min

[surreal spear]

DMs are safe 🤌