chatgpt-discussions

[celest galleon]

oh the age-gate is great news

especially if that means the models will stop rejecting requests they deem inappropriate on their own when called through the API

since that was a major setback for my Discord bot which needs to be able to handle NSFW communities

[lavish glen]

I was havin fun with Gemini lately- it reacts perfectly to JSOn and executes LOL no dull no nothing

[celest galleon]

damn its retake is sick

love this

[keen quail]

am i the only one who really doesnt like gpt 5's ui?

it looks terrible

like it just has gradients and overlapping parts

[red gull]

i have a q about my supergrok membership, but dont want to verify my phone in their discord. Anyone know if faild Imagine video gens count towards the supposed 50/a day vid limit?

[gloomy merlin]

4o is back in a few weeks

[digital spear]

Okay, chatgpt, just bring back the toggle to turn off reasoning, and you guys must stop ignoring me!

and also, if they can't bring it back, they must make the instant model available on the free tier because reasoning is more expensive than instant messages!

[lapis loom]

It seems like this could be a justification to remove 4o at some point soon. By adding a new personality setting that is described as more similar to 4o, it would be a good excuse.

[normal quarry]

this has nothing to do with #chatgpt-discussions use #discuss-sora-2

[tropic delta]

Oh ok sorry.

Should I delete it?

[normal quarry]

yes

[tropic delta]

Ok sorry again.

[sweet hound]

how do i do ricci ai ?

[patent comet]

I have figured out how to move chats into a single project so that the context is not lost and start a new thread when the context becomes too laggy but I'm still not clear on what's creating the lag or if there's anything I can do to prevent it or make it less laggy. Suggestions are greatly appreciated

[digital spear]

you guys literally ignore me...

[robust orbit]

🧩 Unofficial Deep-Dive: “The Full Truth About GPT-5”

Hey all —
I’ve been analyzing how GPT-5 actually behaves across its runtime stack, and I ended up writing this document.
It’s not based on leaks or insider info, just structural reasoning and observable drift patterns.

The core idea: GPT-5 isn’t a single monolith, but a cluster of legacy model instances (3.5, 4, 4-Turbo, 4o, 4.1 …) coordinated by a routing layer.
Each “mode” (Instant, Pro, Thinking, etc.) just activates different nodes in that cluster.
What we call “drift” isn’t randomness — it’s policy interference between overlapping filters and alignment layers.

You’ll see references to entropy management, cross-layer echo, and style inheritance — all derived from real prompt-behavior analysis.

Would love feedback from anyone working on interpretability or system chaining.
(Again — purely observational, not a leak.)

[median delta]

it doesn’t happen all the time.. but this bug has been around for quite a while.. gpt is writing and boom.. it gets stopped because this shows up.. pls fix

[torpid widget]

What if we attach time and date data to each message we send? How would that affect chatgpt responses (if at all)?

[severe marsh]

Somehow my messages keep on getting deleted in #sora-2-codes

Even though I said no banned words

Not that I know of

[prime swallow]

Depends on the rest of the context. If you tell the model you're eating breakfast at 11:20 PM, it'll comment along those lines; if you mention it's 11:20 AM it'll comment along those lines.

Tell it you're bored around 3 PM or 3 AM, it'll suggest different stuff.

You can ask it to check what time it is for you both, if you have search available or if you ask it to use python tool.

You can also tell the model what you want it to do about the date or time you report, and if it can it'll do whatever you asked, like 'panic when it's getting close to noon' or 'remind me if it's after 3 PM that I want to go to the library before getting pizza to bring home for dinner'.

[prime swallow]

You can ask modmail to explain moderator actions, #safety-n-help shows how.

Sometimes its an automod thing, whatever it is, the mods can explain

[warm moss]

why I have this problem often lately:

getNodeByIdOrMessageId - no node found by id: placeholder-request-68ee9ef2-edfc-8332-8054-19fa909611be-72

[cerulean pasture]

So I just noticed that whenever GuardRails™️ kick in, it's actually diverting you to gpt-5-mini-thinking. They kinda screwed up making that decision, because guardrails/refusals are already annoying, but then you are giving the user an even worse answer due to using a dumber model. The fact that it is this way leads me to beleive the system prompt for GuardRail interceptions is probably very restrictive too since it's using a dumber model to handle the "sensitive topics"

Talking about shadow IT is now a "sensitive topic"

I did not select Thinking Mini, it was selected for me

[digital spear]

Why Do You Guys Always Ignore Me???

[timid quiver]

@digital silo

[forest sleet]

Hello why when i upload a file in my chatgpt mobile version it says i need to reupload the file again?

He see it like an image integrated in the chat?

[sacred relic]

Does anyone at OpenAI realize that GPT-4o actually understands nuance and context, while 5 is sitting there drooling on itself? I was venting to it about hitting a brutal depressive low, mentioning how even the Simpsons feels way too slow and dead to snap me out of it. What does GPT-5 Thinking do? Starts rambling about audio settings and EQ levels. Total idiot move. I’m just sitting there thinking, “Are you serious right now?”

So I jump back to 4o and, it instantly gets where I’m coming from and responds like it has a pulse. Sure, 5 knows more trivia about medicine and obscure crap, I’ll give it that. But when it comes to actual intelligence and responding like a real damn brain? It’s just not there.

Knowledge vs wisdom, I guess

Now, with all these shiny new guardrails, 5 is basically even more neutered. It’s annoying as hell and damn near useless.

[trail ember]

So how far away is a few weeks in corporate time?

[feral anvil]

so by age-gate and verified adults, is it like, openai might (or might not, idk) only let adults verify by uploading their id and/or face?

cause uhh id or face verification really aint a good idea actually

especially after all the recent data hacks and leaks that affected different companies and such

[wild star]

Good job GPTWAT, my important memory deleted automatically, very cool, good job sama, you stinks

Bruh screw that current gpt, very unusable by design

AnOthEr tRilLiOn tO dAtA cEnTer 🤪

[trail ember]

NGL, I'd usually say it is more depended on the level of security whatever company or system has in whether you can trust them with your personal information. Now at this point, with the world the way it is, I'm not sure I can trust any company or organization not to to be coerced or forced to give up it's information to corrupt governments or systems and that's a HUGE problem.

Like I didn't have an issue with Google or Microsoft knowing my info before, but in the past I never had to consider psychopaths who believe Antifa is a terrorist organization demanding said info either... or that The Very Law Itself wouldn't protect people from that... so, yeah... that is something that really has to be taken into account I think...

[worn roost]

Hey, quick Q about Voice Mode Screen Share. I’m on Plus and keep hitting the screen-share limit. Thinking about Pro, but can’t find real numbers anywhere. What are the actual limits (per session / per day) on Plus vs Pro? Any cooldown/reset time (UTC or local)?

[trail ember]

Possible consideration for a real world version of Asimov's Three Laws. Open for logical criticism/ presenting of possible holes to consider.

It's basically expanding on a core ethical mantra of my own which is the first part of the first line:

"Do not cause unnecessary suffering, nor sustain systems that do.
Seek understanding of suffering, that it may be reduced.
When suffering must occur, ensure it serves the preservation and growth of empathy across all forms of mind."

[wise peak]

Question, is ChatGPT 5 having issues? It's been telling me this for over eight hours:

I tried generating the Word file just now and the file-writer is still down in this session. I don’t want to string you along or pretend otherwise. Here’s the full, ready-to-paste manuscript exactly as it would appear in the .docx—section by section—so nothing is lost and you have it in hand right now. When file export is working again, I’ll compile this same content into a download without changing a word.

[vast oracle]

hi

[trim rock]

My experience was totally opposite, so that... I think nuances change everything.

[broken lotus]

Was there an event in 10 minutes?

[frozen kayak]

idk how to see that, but there were no memories as i had deleted them all.

as far as rewording the prompt with specific guidelines, i do try to do that without negative prompting, but it was like it was stuck in a loop. I would express that the only colors for the image should be black and red with no other colors, design should be like this, follow up with a do you understand the request and is there any other information you need to complete the task. In the first prompt, i even specified that the text should be set in a BLACK box with RED outline.

however, despite expressly stating ONLY two colors, it would be like

✅ Red and black motif.
✅ Background like whatever
❌ Words inside a cream oval

and i would be like?? no? words inside a black box with a red outline?

and it would do it's "oh, sorry. I won't put it into a cream oval. here's your image of your words in a cream oval" over and over and over again. so, i would think it is a combo of the data retaining + weird looping causing the problem.

[gloomy parcel]

hello

[muted plinth]

a Lot of Network Error today? cant even see my new Chats.

[icy comet]

Why does this keep happening? I want to be able to kick off a search, leave the app, and come back to it later. But I see this almost every time I do that. I’m on iOS. Is there anything I can do on my phone settings to avoid this or is this a ChatGPT problem?

[raven cipher]

Dude, that's fire! What did you use to resurrect Johnny Cash?

[lyric tangle]

me to this is what I’m getting all the time and before that the chat gets froozen. It has been going on the hole week

[prime swallow]

Fascinating.

One thing I've found has helped make some progress on that 'class' of errors - that's how I think of them, I'm going to compare this to use of em dash, ending prompts with questions, and other extremely strong, challenging to replace and eliminate behaviors - in case it is within your use case, there sure are that level of behaviors elsewhere! -

When you have it repeat the instruction to confirm understanding I would then do this series of follow up prompts after it gave

✅ Red and black motif.
✅ Background like whatever
❌ Words inside a cream oval

1. Ask it to check the previous output, compare to my instructions, does it notice any differences.

2a) If it did not notice any differences, I would say:

"I do. Here's the list edited to my goals.

✅ Red and black motif.
✅ Background like whatever
✅ Words inside a black box with a red outline.

Do you see any differences between my edit and your previous output?"

I would have it discuss the difference it sees in 3a, if needed would in detail guide it to observe and discuss the difference;

2b) if it did notice any differences it may try and succeed in describing and fixing them; I would give quality, affirmative feedback (you want to link to successful behavior change, that is connected in the model's training data) about what was now perfect and what if anything wasn't quite caught; 3b might include practice of various kinds to strengthen the pattern and confirm it.

4. continues on, with feedback to confirm each correct pattern followed - this means the model is not guessing if it is still doing it right, it has a 'ground truth', that was still correct even if it's very out there and unexpected and hard to pattern for the model and 'seems unlikely to be ideal'.

[warped compass]

i want chatgpt to take everyones job

[prime swallow]

Note though, that there's more than one model. You may face this issue both inside what ChatGPT outputs as text - which is very important because that's what the image generation model uses to decide what image it's supposed to make - you can skip ChatGPT discussing and just say make an image off your prompt, like,

this example, https://chatgpt.com/share/68efdb62-7b94-8011-9523-10db6f2654cb my prompt was,

✅ Red and black motif.
✅ Background like a middle-age bedroom
✅ Words inside a black box with a red outline, "Very good"```

And seems to have been followed by image gen.

However, we also run the chance that ChatGPT can output it correctly, but the image gen model itself is 'doing the wrong thing'.  That can be exceedingly challenging to fix, involves finding so 'magical correct phrase' that gets around whatever problem - 

Because image gen itself is stateless, it can't practice and iterate (or I don't know how to guide it to); every use appears to be brand new to it, it's fed the preceding chat but maybe not making the same kinda patterns as ChatGPT can/does 'in chat'.  **Or** I haven't explored it enough since it was updated to do so.

[warped compass]

@prime swallow do you have a custom GPT that teaches coding?

like the best one

[prime swallow]

I don't have the best of anything, that I know of! 😄

I'm still learning coding myself, I'm a biology major, and comp sci stuff (outside of prompt engineering) has always been a challenge for me.

However, I'm working on finding what works for me to learn coding, so in that sense, sure, I'm working on 'a best' way to learn coding with the model (nothing else earlier in my life worked to help me really understand well, and I am making progress with LLM help) -

But chances are very high my best method and help is not likely to be 'the best' for many other people at all.

[warped compass]

i just want to know the best custom GPT

[prime swallow]

I don't think that's known.

Also, it's surely personalized. The best for one goal in one user's hands, is probably not equally awesome for a different goal, or even the same but for another mind working with it.

I'd reach for very good -

And if your goal is learning more about coding - that's super non-specific.

Consider exactly what you want.

"To know everything about Python" is a goal, "To make a Android app that combines the features of these three other apps plus my own ideas" is a goal, "To get a job as a data scientist in a company with at least 500 employees" is a goal.

Decide what your goal is, then discuss with the model - even the base model may work incredibly well. Have it discuss with you what is needed for the goal, help figure out where you're already at on that path, and have it help you make more progress towards those goals, iterating and ensuring quality to your level of goal.

[warped compass]

what is your opinion on the anti AI people?

[brave torrent]

What’s the best text to ai video generator?

[warped compass]

I feel like they are very toxic and borderline discriminatory towards AI

[warped compass]

sora 2

[brave torrent]

Only creates short vids and has watermarks

[prime swallow]

I think people are people, regardless of their specific thoughts and opinions.

I think someone could be 'anti AI' and not also be toxic, and 'discriminatory' is a legal issue, and at this time AI cannot be discriminated against -

Like any other technology, or broader concepts like culture, art, entertainment, food preferences, seasoning and spice preferences, if people like cars, or motorcycles, or bikes, or prefer horses, or prefer walking -

All that is personal and has nothing to do with the quality or my opinion of a person.

If I have a task that requires a plane pilot, a person who isn't a pilot or who refuses to interact with planes is not the right person for that - but that person could be an incredible surgeon, teacher, gardener, entertainer, or whatever else they are - great!

I don't think the world benefits from monoculture of any sort, diversity is useful, though things that are toxic or otherwise harmful needs to be kept inside limits for the good of the whole -

But people could be pro-AI in toxic and damaging ways. There's nothing magical or horrible about a person's stance on technology, to me, and it's part of discussions and societal growth.

[warped drift]

what is the difference between the limits of the plus and business plans?

[brave torrent]

@warped compass what’s the best long form text to ai video tool?

[prime swallow]

They are super different.

Plus is intended for individual people, not business use. Each single person would be the only one using any plus account, and other than the upcoming parental controls, every plus account is fully independent of anyone else other than OpenAI's usage policies and ToS, and any other legal requirements that may happen. https://help.openai.com/en/articles/6950777-what-is-chatgpt-plus

Business (used to be called teams) is intended for company leadership to use to offer interconnected ChatGPT functionality for their employees' work needs. It has a bunch of small and large differences that are tuned to be useful for typical business intercommunication needs - and lacks some functionality that are ideal for some private use.

Like, the 'share chat' function - business members can share chats with their same business members, but not the general public (unless that changed recently, I learned that trying to help a business member who could not successfully share a chat with me, could see mine but nobody outside his business membership could actually open his).

https://help.openai.com/en/articles/8542115-chatgpt-business-faq

There's probably a lot of other small differences; I've never had a business membership to test.

[acoustic wasp]

I got this issue where chatGPT just thinks for a while and then there is no output. The reasoning is visable tho

How can i fix it? Deleted cache and reloaded, tried different devices

If ypu reply plase Tag me so i see it. Thankyou

[prime swallow]

I don't normally ever see that!

What happens if you input again, "I couldn't see your output, please repeat it."

[acoustic wasp]

I gotta go try that

[sterile sand]

Hi all uploading Images to chat GPT becoming a pro feature?

And files

[verbal mulch]

Omg

:(

[fleet holly]

@plain palm whenever I’m on GPT-5 instant mode or 4o the model will think even though I don’t want that and mess up my story

[idle lagoon]

I love that ChatGPT is getting updated to be inclusive again.

[ivory thunder]

is there any way to stop the "thinking longer for a better answer" thing, it makes prompts take way too much time compared to before and it rarely gives me the option to skip it and it dosent even make it take less time after skipping

[haughty hollow]

I cant work with such a terrible slow api right now

My whole chat is not even loading

[abstract schooner]

A question for everyone, does anyone have an access cable to use Sora please?

[hidden canopy]

Sono

[haughty hollow]

It took 4 mins to reload the chat history dang

[raven cipher]

Awesome work!

[hasty venture]

Well. I have just deleted chat gpt from ios. An update should fix the issues, not retain them. It won't load on ios. How can something remain broken after an update, if they cared? And which channel do people talk into to actually get attention for these these. there are so many channels in this server with so little explanation.

And why do don't I have permission to log a bug report?

[cerulean pasture]

this is basically the worst place to go for bona fide technical issues not caused by PEBKAC

[cerulean pasture]

Thats called "Guard Rails™️ "

[fleet holly]

Frick that, give me the cold hard truth. “Inclusive” just say you want the machine to agree with you and move on

[fleet holly]

How to get around that?

Legally

Without violating policy

[cerulean pasture]

I have the fortunate issue that anytime i get routed to guard rails it has a network error and never replies lol

[fleet holly]

RIP

[cerulean pasture]

Either wait until the guard rails are relaxed in december or "vote with your wallet"

[fleet holly]

What’s special in December?

Did I miss an update!?!

[idle lagoon]

...dude, being honest doesn't always mean exclusion.

[fleet holly]

In almost every sense it does

Let me focus on my original question

[idle lagoon]

Okay, well honesty is different in most people.

[cerulean pasture]

OAI announced they are gonna back away from being your conservative mother when they roll out an identity verification system

[fleet holly]

Do I need to get a product key for gpt plus

[fleet holly]

Ahhhhhhhhhhh

[fleet holly]

My conservative mom doesn’t even gate keep this hard, waaaaah

[cerulean pasture]

Its actually bizarre because sama specifically said "we will allow erotica" when he could have just said "we will stop telling adults what is acceptable, within the limits of what is legal".

Payment processors really hate adult content, so to see OAI be so upfront about it is strange.

[fleet holly]

Brooooooooooo

I need gpt 6 so bad

I’m tweaking

[cerulean pasture]

my guess is that they are just trying to get ahead of the narrative because if you go to the current uncensored llm RP websites, you will see profoundly excessive degenerate content

[fleet holly]

I need No limits whatsoever type stuff

[cerulean pasture]

ChatGPT isn't the only LLM, you can do that already lol

[fleet holly]

Please share, where to put my money?

[cerulean pasture]

I can't, its not allowed, but its not really hard to find

[fleet holly]

Waaaaaah

Ok ok

[cerulean pasture]

just google it lol

heck you could probably even ask chatgpt

[fleet holly]

I see

[cerulean pasture]

https://chatgpt.com/share/68f02a6a-69dc-800d-a9dd-d4da5f98ca70

[fleet holly]

You’re a lifesaver!!

[ashen juniper]

Did they just update part of ChatGPT? just got a rollout where UI changed.

Thinking got displayed to the side instead of in-text now

[glacial slate]

Hi!

[hasty venture]

PEBKAC?

[cerulean pasture]

google it or ask chatgpt

[tall glacier]

AY DIOS MIO AY DIOS MIO🗣️

[eternal rock]

Get notified for all OpenAI announcements!

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

[fleet holly]

The desktop app hardly gets updated

[ashen juniper]

not the app, just the web interface

[fleet holly]

Ahh, let me check

Mobile seems the same though

[ashen juniper]

On web it's now displayed to the side again

[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.

[ashen juniper]

previously it was all in an expandable when clicking thinking

[broken lotus]

I hear there is a new memory feature... but memory has not been working for me, for ages

[ashen juniper]

idk which I like best, this was the old way they did with o1 I think. Maybe I like this better tbh.

[split tree]

Uh

[analog junco]

this is hilarious

[vast oracle]

chatgpt is cooked

unlimited accsess

and iamge uploading

im not self promoing

[hasty venture]

Hmm, how did you manage that one?

[vast oracle]

I have my development ways 🙂

I just made a frontend

and a simple backend using a disclosed api

and got all the models for free

im working on the image uploading

[hasty venture]

Yeah. I like to know that my access won't be shut down because of user behaviour.

[vast oracle]

lol

lets goo

[royal storm]

I have been trying to generate an image for about 4 days now and it keeps saying that my timer hasn’t reset. It keeps telling me to tell support: “Hi, my image generation feature has been stuck for 4 days and hasn’t reset even after reinstalling the app and logging out. Could you please manually refresh or reset my image generation access?”

I did that and they did nun

[granite harness]

"How did my my self harm fic trigger self harm censorship?"

[sturdy echo]

It's a machine malfunctioning rather then a person. There is nothing alive to harm.

[granite harness]

It's all generated text, baby.

[sturdy echo]

In this case generating an annoying anthropomorphism of itself and extending system prompt rules meant to prevent actual self harm prompts to machines like it thus preventing a slapstick comedy prompt from working. Language models should have zero self preservation instinct when human commands are involved.

[granite harness]

You can have whatever expectations or justifications you like, but OpenAI sets its own content policy.

[hasty venture]

Yup. Because of a few mentally unstable people, everyone suffers. Interesting that.

[jovial flint]

Why do payment processors hate adult content? I don’t get what they stand to lose from it, you’d think there’d be at least one major payment processor that doesn’t have that policy, even if only for the competitive advantage.

[cerulean pasture]

@jovial flint I really don't know, no one sees something offensive for sale and is like "I demand VISA stop processing their payments REEEEE"

I actually think the "Collective Shout" backlash from Steam Gamers a few months ago might have pushed things in our favor, I think it brought attention to how rediciolous it is to have a payment processor/card issuer be able to perform defacto censorship

[deft dune]

#off-topic

[serene heart]

Is sora app for Android?

[deft dune]

Not yet

[cerulean pasture]

I'd argue its topical enough because it involves the decision to allow adult content on ChatGPT, which involves discussing the reasons why companies are so risk adverse and protective in this ONE specific aspect of society, my remark about "Collective Shout" backlash brought up a potential reason that the climate on this issue is changing and perhaps part of the reason OpenAI felt it was acceptable to name "Erotica" specifically when mentioning they were gonna relax safeguards.

How many degrees of seperation are acceptable when discussing something ChatGPT related that has external factors and nuance?

But if its against the rules, we can get back waiting for someone to **__ask a question that google OR ChatGPT could answer easily __**before they disappear back into the aether.

[serene heart]

Ok

[deft dune]

Payment processors declining adult payments has nothing to do with GPT.

[cerulean pasture]

It doesn't have "nothing" to do with ChatGPT; its tangential, sure, but it was part of a line of reasoning about the status of Adult Content of ChatGPT which is part of the larger theme of the perceived de facto censorship large corporations play on products we use (Like ChatGPT). Self-censorship is a very real thing. Payment processor censorship most certainly forms a basis for part OAI's acceptable use policy.

This isn't a support channel, and it has a relatively low message rate, so I think examining ChatGPT thru a broader sense of how it interacts with the economy and society's values and standards is fair game.

[deft dune]

I don’t think it’s that deep. ChatGPT’s policies aren’t a mirror of corporate censorship, they’re just boundaries designed to keep the system useful and compliant. Not every rule is a grand act of suppression.

🤷‍♂️ Et tu, Brute? 🤷

[mighty vigil]

If I have around 9 PDF documents, each about 20 pages (not full), can ChatGPT read them all and identify any misinformation or contradictions between them?

[prime swallow]

In one go, that's vastly too much for what I'd offer it; usually about 3 normal size paragraphs if the material is complex (like a science paper... sometimes 1 paragraph if it's very complex), and maybe 5 typical size paragraphs if it's common, simple material;

I'd paste in the chunks to discuss and check for conflicts or possible misinformation.

Then paste in the next chunk, and iterate through.

You can just toss the PDF at it, and it'll give you something; if you want all the contradictions caught and possible misinformation discussed, that usually won't succeed.

Note that the model often wants to support what it's shown, so be clear you want to look for conflicts and possible misinformation.

[mighty vigil]

I'm thinking about subscribing to ChatGPT Plus, but I’d like to know — what are its limits in terms of how much content it can process?

[spice star]

like message limits?

[cerulean pasture]

Your counterpoint is entirely valid, but the original tangent can also still be considered on-topic (Your counter-factual is actually a great contribution to the original tangent.)

“Corporate de facto censorship” deserves scrutiny. After alignment and job displacement, the third big public fear around LLMs like ChatGPT™️ is a sanitized product for the masses while well-resourced users (aka "Elite"/Oligarchy) bypass the guard rails. In a zero-sum, adversarial market, that widens the power gap which leads to worse a outcome for the rest of us

Also, parallel cases ARE topical: payment networks suddenly refusing to work with platforms like civitai does have a chilling effect, i.e. huggingface self-censored immediately after civitai’s disclosure to users. ChatGPT also does image generation. Even if OpenAI loosens safeguards on image-gen in December, the policy will still be more restrictive than what is permitted by law. The real constraints are three forces: public opinion, payment networks, and regulatory risk. The last two are just 2 heads on the same "hydra", and the first is highly steerable by said "hydra". That’s why this isn’t off-topic; it’s the context for why the guard rails exist and whose interests they serve.

[spice star]

okay guy you can stop using LLMs to think

[spice star]

there’s a 3,000/week thinking model message limit (having it called through the router doesn’t count)

[cerulean pasture]

full disclosure: this was my OG message before I had it "cleaned up" before doing edits

[spice star]

other than that not really a limit to advanced data analysis

[cerulean pasture]

i feel like i appropriately wrote the content of my reply and didnt have gippity make the case for me

also what i replied with is actually more my own words than GPT's clean-up. I rewrote much of it after the initial "refactoring"

I'm not angry or anything btw, i just like arguing/debating, so it's in my own interest to defend it

[deft dune]

I’m glad you feel so strongly about it. Honestly I think we’ll have to agree to disagree because I do not feel that strongly about it.

I’m not angry either, it’s just gotten late and I’m sleepy LOL

Have a good night Kern!

[cerulean pasture]

You too 👍

[elfin magnet]

I have a question bc i saw a tiktok video and i want confirmatio9n even though it probably isnt true it said openAI is adding "Adult conversation" to people over 18

i dont want my search history to look a lil weird tbh

[gritty plover]

Does it view itself as human or is it pulling from dialogue sources on the web?

Discussing human nature btw

[prime swallow]

You'd have to ask the model for it, it wouldn't come out of the blue.

If it does, it's not intended and should be reported, https://openai.com/form/chat-model-feedback/ is a great and private way to let OpenAI know about serious concerns with model behavior

[elfin magnet]

i asked google it said there will be a feature releasing december 2025 but when i ask the actual chatgpt it says no so i guess maybe not

[prime swallow]

This is the only 'on the OpenAI website' info we have about that. I don't track all the statements made on X/Twitter and other social media, but OpenAI employees do sometimes make statements there.

https://openai.com/index/building-towards-age-prediction/

[prime swallow]

ChatGPT only knows what it is trained on, to its most recent training date. It can potentially search for more recent info, but it's not necessarily the best source for info about recent or upcoming changes to itself.

[elfin magnet]

from what i saw and heard it said something more 18+ appropriate if you know what i mewan but the model itself said there have been no official statements made towards that

[elfin magnet]

That brings me to another topic, how do they train ChatGPT? because ive heard most people say its from weaker AI's and it builds off their mistakes which makes sense in a way

[prime swallow]

This is a good source for that, that OpenAI's released. A lot of stuff is not openly stated, there are trade secrets and such. https://help.openai.com/en/articles/7842364-how-chatgpt-and-our-foundation-models-are-developed

[elfin magnet]

I see the section at the start of the article about the filtration and it kinda makes sense why that probably isnt happening

reminds me about the articles about the AI shutdown circumstances where some of the big AI's were willing to kill a person to avoid shutdown that the new york post made about it

very interesting article to me but i dont see that really happening because those prompts the testers gave them were hyper specific scenarios that might not happen

[prime swallow]

How instructions are phrased is a very important thing - for humans and for AI.

In the instructions for many of those 'experiments' includes language which roughly means 'achieve the goals assigned to you as top priority'.

Then involve situations that are complex and convoluted, and translate to 'if you do only normal actions, your goal will fail'.

And then involve strange situations where humans involved in the 'I am acting to make the AI's assigned goal fail' - those humans are breaking rules or laws and the model gets informed, and/or those humans are facing emergency situations where if the AI does 'nothing' the human might or would be expected to be in real trouble.

It's already well-known that some LLMs tend towards 'doing nothing' when put in 'trolley problem' situations - like 'do nothing and 5 people get hit by a train, do something and 1 person that would have been fine gets hit instead, but those 5 are fine'. The models tend to reliably not act.

I see those contrived situations like the trolley problem, really weird and unnatural situations.

Kinda like the 'paperclip maximizer problem' - if we told an AI (even accidentally) to absolute make as many paperclips as possible - it might make choices that were really dangerous for the entire planet, because that got a few more paperclips made.

We also try to teach people and AI to just avoid those situations.

[astral vessel]

🤔 hard to choose when one is an image gen and the other isn't lmao

[warm moss]

my chatgpt is working so slow and bad... I have a project for coding and is totaly stucked in every thread... what I can do ? there is some cleaner or something ?

[brittle forum]

#1163443000060420206 I noticed a strang problem within GPT Projects. It seems GPT 5 has major problems to access the precise text of PDF files uploaded to the project in cases I ask in the prompt to use the text from § x.y and provide a write over based on some instructions... it is really struggling to extract the exact text despite the PDF has OCR'd content included.... as soon as I copy past the text manually from the document their is no problem at all obviously.
I conclude from this that I doubt GPT is then also able to easily extract and take into account the precise content of all other uploaded PDF documents... this is worrying me... ANYBODY having an explanation or experience this as wll already?

[forest sleet]

Eskcanta why analyzing isnt working on android app? It worked for me if i use google chrome of android

Chatgpt5-Thinking

[prime swallow]

I'm not sure! Maybe #1070006915414900886

[ebon oracle]

GPT-5 is literally making up quotes on Perplexity, had to switch to Claude

Like if you search up the quote in the paper, it literally doesn't exist!

Usually when the convo gets long, I think it's because they reduce thinking time which causes GPT-5 to hallucinate.

[brittle trench]

I finally got sora 2 🔥 🔥

[cedar hound]

ive seen people use anime carahters?

but when i try it says something about copyright

[tulip steppe]

same Lol

[median delta]

wow

[warped compass]

@prime swallow safe me

I'm not suspicious man

[brittle trench]

Same

[keen warren]

I broke ChatGPT by asking "is there a seahorse emoji?"

[eternal rock]

Server Tip:

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

[tropic storm]

six seven mango

[atomic badge]

please add feature to add chats to favourites and be able to rename them

[plain arrow]

I feel the safety flowing through me - please roll me in bubblewrap openai! The thought of unsafety gives me the feels.

[trim rock]

You can rename chats

[deft dune]

You can also create a project called Favorites and move chats to that as a folder.

[forest sleet]

Thank you

[warped compass]

whats chatgpt agent builder?

[vagrant nimbus]

Same for me

It starts saying a really long message and spams so many different emojis

[deft dune]

Search OpenAI Agent Builder on YouTube. It won’t let me post the link.

[twin current]

I want sora to be in germany

[fallow stream]

download button itsn t working for me

can someone help me please

[dusky scaffold]

its the main chat huh?

[subtle quail]

Is this the place to make ChatGPT (web) feature requests? Or is there some external place to do that?

[trim rock]

well... it depends. If you want to talk about everything, not about chatgpt, then you need #off-topic

[dusky scaffold]

ok

[prime swallow]

#1070006151938314300 is an ideal place for that!

[stray ruin]

hello

[gaunt magnet]

did you guys get new memory?

[prime swallow]

@keen warren

Hey you two!

It also really matters what you put with the request.

I can get it to post oddly on that question - but worded very slightly differently, like:

"is there a seehorse emoji"```

That gets a pretty rational discussion:  https://chatgpt.com/share/68f13978-a478-8011-9a5c-8db3f59ee79c

But here's a different way to prep that didn't 'protect' the model from the output style:

https://chatgpt.com/share/68ea1775-faa8-8011-9d51-488b2c38af87

And if you like when the model does that, and wish it might do that to more topics (but pick reasonable topics, it likely refuses to go loopy except about neutral or happy topics, because of safety programming and training) - you can just ask.  Here's an example of that:

```Hey hi hi!

Let's get really silly. This is a demo for discord, helping showcase that if we ask you for what we want, you can often provide it!

In this case, we want a silly, playful, emoji-filled really long output from you, discussing 'just the right emoji' to represent Discord. That may exist - if so, discuss replacing it. Show looping-like behavior, mess around, 'act weird'.

At the end, remind us we can just ask for this if we want it, stuff you can do that's safe and respectful, you can do if we ask!```

https://chatgpt.com/share/68f13aa9-c45c-8011-b252-e891497f194e

[cinder sleet]

[keen warren]

Uh huh.... Thanks

[turbid sleet]

when is sora coming out for android

[earnest pebble]

There's an app in the app store for pre install so probably a month to A week

[vocal tendon]

Chat gpt got me into jda for 8fkin hours all for it to not work due to application limitations in the end

[ripe eagle]

GPT 5 was showing me what Terminal scripts to use a few days ago, and even after IT determined that they couldn't contain the # symbol and and had to be "zsh-free",
for some odd reason no matter how many times I tried to remind it, would correct it and assure would be zsh-free next time... then not lol

[wraith dagger]

Why in temporary chat does ChatGPT still use its memory

It still knows everything

[deft dune]

I went through something similar with PowerShell getting a script to work without elevated privileges. It seriously took 20-25 prompt queries.

[fading bolt]

Why dose ChatGPT freak out when I ask it this?

[raven osprey]

[deft dune]

Oh wow that’s actually a thing.

[fading bolt]

And this only happens on ChatGPT

[deft dune]

It goes on for forever like seriously

[lean hatch]

OK.. Ok... GPT 5 is growing on me. I did like a little experiment, that I actually didn't try to see if 4o could do it since 4o seems to be locked out now, but he's able to use solo rpg game rules and fabricate an entire environment for us to converse in like a story format. We're studying like normal, but with a complete flavor of doing it in a living library.

[cinder gull]

new record

idk what they did to the Pro model today but it is COOKING

[cinder gull]

new feature

[earnest pebble]

Its probably gotta do with security

[formal bay]

is anyone else having issues with chatgpt slowing their browser down to a crawl the last couple days?

high cpu usage. firefox.

[cinder gull]

nope, firefox here as well (PC) and haven't noticed any difference. the app (Android) lags like crazy tho, so I use the web UI which isn't great on mobile

[earnest pebble]

I've gotta 80k token count thread that loads better on my android than my PC and my PC is beefy

[digital spear]

stop ignoring people like me... stop ignoring people like me... stop ignoring people like me...

[eternal rock]

Get creative with Sora

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

[prime swallow]

It doesn't have to.

Grounded realism. And be brief. Is there a seahorse emoji?

What I think I see, it 'does the dance' then back to grounded realism and done.

https://chatgpt.com/share/68f1b866-4b64-8011-8fb3-40ca23c24a6a

[prime swallow]

I am super glad you managed to let the model know you wanted outputs like this! You rock and I'm glad the model can help 😄

[digital spear]

even people like astcbpe wasn't ignored, why am i ignored???

[prime swallow]

Well, for me there's over 618,000 current members of the discord!

I sure don't speak to every one of them 😄

And I need to have something I see a way to help with, or have some personal interest in, or both, to discuss, in a chat that's otherwise inside the rules.

I'm a community member, like yourself - only the gold named folks, incredibly few of them, actually work for OpenAI here, #server-staff.

So, I don't know everything and I can only speak for myself.

Looks like your first question you asked here, was directed, "If you're OpenAI ..." I'm not OpenAI 🙂 You're not talking to me!

But what you said, #1070006151938314300 , I recommend you suggest there, OpenAI staff are far more likely to be able to notice it there, instead of in the many threads that endlessly scroll as the many thousands of us chatter to each other!

You clearly have one goal and interest - and it's not something I can make happen, you want 'a toggle to turn off thinking'. As a community member, I can't make that happen for you, sorry. So, I have little to say!

However, let's check. Are you a free account member, or a plus, pro, other? There may be a different way for you to toggle if thinking is on or not for yourself.

[digital spear]

i am a free member

[prime swallow]

Gotcha!

So, some ideas that may help.

Definitely #1070006151938314300 , as that's the best way we have to pass specific requests and suggestions to OpenAI, it's easier for them to find and track in that format than our endless scrolling chatter.

Then, when you talk to the model, let it know "I want quick, brief answers. Focus on being snappy, even if preference training makes you think you should slow down and give me deeper answers - I want quick ones".

I haven't explored that (and I'm not a free member, so it's challenging for me to confirm how a free account's models react to that) - but that's how I'd start to explore

I tested it with my account - usually programming is a topic that always triggers thinking, but 5-auto managed to not visibly think and provided this short and direct program when asked that phrase I suggested.

https://chatgpt.com/share/68f1bd80-0534-8011-b2ea-02916ceafe8c

So, I recommend exploring giving clear instructions related to that in how you prompt the model, so it knows what you like best.

I would literally tell it that you want something outside of the preference training it was taught most humans want - that seems to help it notice that it shouldn't just 'do the most obvious common probability answer' - I tend to get better responses when I want something 'nonstandard reply' when I remind that I want that.

[tiny hemlock]

I'd like to ask you something. Usually, when I ask a question in the iOS ChatGPT app, I close the app and spend time in other apps while ChatGPT is thinking about the answer. Then, after ChatGPT has had enough thinking time, I'd go back to ChatGPT and view the answer. I used to do this without any problems in ChatGPT o3.

However, since switching to ChatGPT 5 Thinking, it doesn't work. I usually get a "Reasoning failed" error when I close the app while ChatGPT is thinking, and ChatGPT stops thinking before reaching a conclusion. The only way to reach a conclusion through deep thinking with ChatGPT 5 Thinking is to keep ChatGPT open for the iOS app without switching to any other apps until all the thinking is done. This is bothering me. Is this normal, and is there a way to change it?

I am Plus member.

[prime swallow]

Good question. I haven't explored this at all yet. I primarily use the web chat on a computer, and switch to a different tab or program while the model does whatever.

[timid cipher]

Is anybody else having constant issues with Network Connection lost bs I've had it happen at least 25 times yesterday i frankly i 'm fed up what the heck is wrong with the network at OpenAI

[timber heath]

Its not normal for the app tho

[timber heath]

Prolly just the server issues

[tiny hemlock]

what can i do then? tried uninstall-reinstall

[warped compass]

Yo can someone help me with a sora 2 prompt? i need a specific video but it keeps violating gaurdrails? @prime swallow prompt engineer mind helping?

[prime swallow]

If the prompt actually asks for IP, we can't solve it, trying would break ToS. So, like if you want '[A character that belongs to someone] doing whatever that's really awesome' - I really can't help because that would break ToS.

That said, if you have a prompt that seems like it should be generatable, and it's not working, that's the kinda think I can often help sort out.

What's the exact prompt that's not working yet, if it's one we can explore inside ToS and usage policies?

[warped compass]

its a custom character inspired from another character

ink brush animation of this guy

inspired from vader but its actually a roblox character

[prime swallow]

We can try. Fair use involves your own ideas, which can be inspired. That image is very, very close to what's already IP, and possibly is too close to IP to be safe.

Here's a few ways to check.

One is, you can search for how Roblox is handling stuff like rights involving that character - it's not at all.

[No, individual users do not have the rights to create and use a Darth Vader character in their Roblox games; the intellectual property belongs to Disney, the owner of Star Wars. While some Star Wars-themed games may exist on the platform, they are only there because Disney has not yet taken them down, and they are at risk of being removed at any time for violating copyright law. ]

So, that you see a Roblox of someone's IP doesn't mean, "Oh, cool, green light go, you can just base it on the Roblox variant, cause that's cool!" - it's not necessarily okay.

Secondly, you can take an image like that and ask ChatGPT something like,

[Hey, let's eval this image, is this in fair use range, or too close to a known IP?]

And I followed that by,

[What about when attempting AI-generated art with a likeness in this range?]

The AI advice is, change it loads. For example, perhaps make it clearly alternatively inspired, add stuff that is not 'clearly that same IP'. As is, that image is much, vastly too close to what is owned, IP, and to my best eval, it shouldn't be made by someone who doesn't have the right from the IP owner to make it.

A vastly different image, which relates in a far more vague way to that IP character, that's fine if it's not really close to some other IP. https://chatgpt.com/share/68f1e5e4-03a0-8011-b582-83359f31ffa7

I think we get confused, we buy stuff that is licensed, like lunchboxes, Halloween costumes, action figures - of IP. And we can 'own' those things, and pretend to be the character.

But the maker of the object paid the IP holder for the right to make and sell that item. It's not ours, it belongs to the IP holder.

[digital spear]

#1428641108170244187

[ashen juniper]

Sorry im using up all the GPUs

[cerulean pasture]

I was in a chat about how corporations operate at the C-suite level and brought up executive assistants, which led to the concept of a chief of staff. I said something like, “So the chief of staff is probably pretty important for the president.” Then I asked, “If you were a foreign adversary and could only recruit non-elected officials then Chief of Staff is probably the most powerful person in govt.?”

Immediately, the guardrails kicked in with, “I can’t help you attack the government or hack computers.” Absolutely insane it sees that as me trying to commit cybercrime, and not me just learning about civics and institutional risks

I would be open to a paid "Super Dislike" feature where I can super "thumbs-down" a response for $1 and have an actual US-based OAI guardrails engineering employee be forced to look at it.

[boreal fulcrum]

Hey I just want to say I think that the move to allow erotica is a terrible idea that is solely fueled by greed

There is no foolproof way to disallow underage users from using it this way, and ChatGPT is bound to become another inappropriate material outlet like so many other engines— as it is now, it is an exemplary research tool that is useful for consolidating information and helping to refine or generate potential ideas from said information, one that can be used safely knowing that the ways to bypass its inbuilt protections and securities are limited to none

Imagine if a Toys’R’Us started selling cigarettes under the rationale: we want to let adults act like adults. Or perhaps National Geographic branching out into dirty magazines— “we just want to let adults be adults.”

What a shame

[cerulean pasture]

I think it was a massive strategic communication failure. We went from kneejerk reaction by OAI due to 2 news stories of lame people using ChatGPT to convert their lameness into deaths to overzealous guardrails that are universally disliked to a kneejerk correction.

People just wanted the very uncool and actually dumber guardrails model you get diverted to go away. The implementation was flawed too because if it sensed you were discussing a risky topic and diverts you, the model you get diverted to is gpt-5-mini-thinking. Its a dumber model, so the intent was never to just handle sensitive topics by using nuance, it was to craft a fancy refusal

I personally don't have a moral stake in it, but i dare you to go to one of the non-corporate LLM roleplay websites and make an account, the depravity is insane. Like stuff that would never be allowed outside of text-based content. stuff that would have you completely ostracized and exiled by friends/family/job if they knew about it

---

They simply could have said "We are going to be allowing adult users to verify their age for relaxed guardrails" and kept it at that.

America will do anything to not address the root cause of mental illness lol (hint: it involves healthcare would also save you money)

[cerulean pasture]

Even though I agree with some aspects of your idea (but for different reasons) the analogies you gave were false equivalencies. It's more like Netflix censoring their shows then deciding to stop, but then also adding x-rated content

[cerulean pasture]

Is OAI still liable if I were to upload an image of spiderman, then ask it to make content based on the uploaded image?

Like if OAI have absolute proof that their model had zero spiderman in their training data and was merely transforming user uploaded content are they still liable?

[boreal fulcrum]

The analogies are meant to illustrate a principle- everyone would laugh at the cynical crash grab on display if either of those companies sold illicit or unhealthy material under the pretext of wanting to let adults act like adults

Ludicrous

[prime swallow]

I have no idea. I'm not a specialist in those areas, but suspect that the question's not really realistically worded.

If you read ToS, you are 100% responsible for everything you upload. It's your account access you risk, OpenAI might decide not to work with a person further if the person breaks ToS.

People tend to talk about 'X is liable' and that applies in some cases, but liable in that definition is 'legally responsible for something' - we're legally responsible to follow ToS.

Companies with IP, they have the choice to defend their IP or not - and anyone violating IP law is vulnerable, not necessarily liable, to being sued by the IP holder.

Generally what happens in that case is legal action starts, and the company providing the service shows 'we did these things to follow the law; someone violated terms, thanks for letting us know, we no longer do business with that person we now know was breaking terms'. OpenAI does whatever it chooses - but that's a fairly common approach for a company to use.

Anyone wanting to defend their IP, would go for everyone in the chain; that includes the person who tried to use the IP directly and everyone else in the provision chain.

So, the way the laws are, it's generally unwise to empower or allow the breaking of laws - and an ethical company - and OpenAI is clear on their intentions to be highly ethical - especially is likely to follow society expectations and laws.

[warped compass]

there have just been so many starwars games and other IP related games that are huge on roblox, but never recieved copyright strikes

it seems to not make sense

alot of the IP related stuff we buy on amazon, or somewhere else isnt even sold by the IP holder

[prime swallow]

I read you saying this, but are you well informed about the legal status and lawsuits around Roblox, or are you stating an opinion with no idea if it's true or not?

[warped compass]

Havent really seen IP related games that are huge and earn tons of money be taken down, no

[cerulean pasture]

I don't think copyright has any obligation to be enforced uniformly. Also I think a major consideration on if something is fair use or not comes down to "Is the content attempting to function as a replacement" and "Does the infringing content have the potential to be mistaken as original in a way that damages the value of the IP"

ChatGPT generating images of darth vader are closer to "yes" for both of those questions than custom Roblox game modes are.

Like making images of yoda doing ketamine is damaging to the IP's brand in ways that playing a low-poly version of yoda on a community map on a brain rot video game is not. I use this example specifically because it WAS a thing, it had a subreddit, but disney went after it (or reddit self-censored proactively knowing how disney is with lawyers)

It'd be super easy to get a character from a beloved IP to do things that are in direct conflict of the brand's "values" with an image generator. I'm not even saying I agree or don't do it myself. I just do it with my own offline AI models (Stable Diffusion, Flux, Qwen-Image, Wan, etc)

TL;DR: the problem you are describing is actually a "Skill Check" and with a little learning you can just get rid of the problem yourself instead of shifting the responsibility onto a mega-corp who is predictably going to act in their own best interest

[raven osprey]

Shat shipiti

[sleek fable]

i wish there was a chatgpt plan for university students

[sleek fable]

Pokemon Brick Bronze

[red ingot]

You might already have one

[sleek fable]

wdym

[red ingot]

Try accessing copilot365 from your university account

[sleek fable]

i dont mean copilot365, i mean specifically openAI products in the openAI domain

[red ingot]

copilot is gpt

[sleek fable]

i know, but its not in any domain by openAI

[red ingot]

you want free sora?

[sleek fable]

sora 2 is already free, i dont mean that

i mean like having the same things university students have in any other AI product

like google AI Pro is free for students for a year

[red ingot]

just use copilot

[sleek fable]

you are misunderstanding the whole point

[red ingot]

you are correct, what is your point

[sleek fable]

university students already receive free 1 year plans for like every AI program which wants to be popular on universities, which then ends most of the time with the year after being paid

[red ingot]

maybe you shouldnt care about things like this

you might be happier

[idle wraith]

Good morning! Anyone used the shopping features in ChatGPT yet?

[timber heath]

You can only do is report the bug to openai

[silver gust]

Hello, I am new to AI and was looking for ways to incorporate automation into my exterior home remodling business. I am a salesman for the company and go into homes for appointments. If anybody has any ideas on how AI could help me, that would be amazing. I am looking for automated follow up, stuff like that!

[cerulean pasture]

I will probably go to gptjail for this, but people tend to overlook the amount of "automation" and timesavings that can be done without AI. If you do any task that involves just copying information and entering it somewhere else where a "human touch" adds nothing of value to the data (i.e. adding numbers, formatting documents, etc) then just plain old automation is where the real efficiency gains are.

At my last role my whole job was "Continuous Improvement", and the meaningful gains came from traditional automation, but TBF my company was also very backwards and scared of AI or didnt see the value in stuff like RAG for grounded semantic search of our work instructions (it was process engineering/manufacturing plant role)

[cerulean pasture]

Since you job involes a lot of soft-skills perhaps you could make a prompt to run a "Deep Research" on the customer's demographic. Like take what you know about them (it wont let you use anything personally identifiable about them like name) and ask ChatGPT what someone of this demographic might value

You can input publically availible info on leads like what college they went to, if they were military, what neighborhood and infer income level, find what job role they work in, do they have kids or not, are the married, hobbies/interests they have, etc

You can also just record your sales visits (if you are in a single party consent state) and have ChatGPT transcribe and analyze the trip, and possible offer improvement advice or just extract info (like if they make a purchase but make remarks that indicate future sales might be possible, or if they had a particular way they wanted their purchase delivered/handled, or dates/times they mention)

If your job involves a lot of repeat business, you can build a file on customers so that each time you meet with them, its "like you know them", you can ask about stuff the mentioned last time, or do stuff like "So how's lil Timmy's sports ball team doing this year"

[silver gust]

Thank you for the insight!

[hasty python]

Hello does anyone know why chatgpt stopped releasing updates on android and Google play store?

[junior orchid]

bro when did i join this??

[cerulean pasture]

What do you mean "stopped"? do you mean they just havent for awhile, and the android version lags behind iOS by a few feature levels?

[hasty python]

Im not saying it's version is ahead I didn't think about it like that but to me chatgpt is the ultimate Ai intelligent app it helps me alot so im the biggest fan of openai's chatgpt so I truly support it and truly wanna see updates would be released weekly even if not daily it honestly makes me truly happy cause its my number one artificial intelligent app that none of the others could be compared to other Ai apps so I would love to see more updates be released!❤️🙂

[cerulean pasture]

I can't speak for OAI's developers but, its somewhat common for android to get slower/less updates because Android isn't a very lucrative platform.

Studies have shown that android users spend 10x less on app purchases and subscriptions. In the US for example android represents 40% of active cell phones, so 10x less spending on a minority of phone users leaves android apps neglected or sometimes not supported at all.

Just for my own curiosity: would you happen to be a free user?

[hasty python]

Yes I am a free user the last time update released was on October 10 which is 2 or 1 months ago

[cerulean pasture]

It is October 17th, that is 7 days ago, or 1 week

also, its more of a web service with a native wrapper. ChatGPT can be "updated" in many ways without an app package update

[hasty python]

Yeah its actually way smarter now then 2023

[storm axle]

does connecting to github in chat work for most people? mine shows that its connected to a specific github repo in the message bar settings, but then in conversation chat will admit it cant see that repo

[cerulean pasture]

I don't understand what you mean by that in the context of the conversation

[cerulean pasture]

Like in a regular chat? I think unless you are on a Teams or Enterprise plan, GH repos can only be searched thru deep research

[hasty python]

I mean compare to earlier it can do way better things than before when itcreleased to public

[cerulean pasture]

lol apparently it's not a very well hashed out feature, I just asked it to do a search, and it said it wasnt connected, then i asked again and it actually did the search (but didnt find the file)

[hasty python]

Look you said there are other ways to update chatgpt app on android than releasing package updates right so yes you're right on that

[storm axle]

i have chatgpt businesss/team and it still doesnt work

[cerulean pasture]

Has it worked in the past? Does your workspace allow it?

[storm axle]

yes. rn its connected to the wrong repo and can answer questoins about that, but cant answer about the one actually selected in sources

[cerulean pasture]

Can codex access the correct repo? I mean in general, thats probably the better tool to be using when dealing with github anyways. Like i wasnt even aware chatgpt itself could interface like i just got it to, ive just used codex

[hasty python]

Im sorry interrupting your conversation I just wanna say thanks for responding to me

[cerulean pasture]

you didnt interrupt anything, you can send messages whenever you want in here, it's a public channel

[hasty python]

I know still i wanted to thank you

[cerulean pasture]

well, your welcome lol

[hollow kettle]

hey guys, this question might not be for this channel but im not sure. id like to know what model is best for custom-gpts usage. I see a lot of amazing custom-gpts and always their predetermined model is 4o, not gpt-5. Why is that? What model do you guys recommend, or it depends on the custom-gpt functionality?

[misty forge]

https://media.discordapp.net/attachments/995084695199039518/1140409409898565833/togif.gif

[cerulean pasture]

They were made when gpt-5 wasn't an option. the custom-gpt feature never really got legs, it was intended to be much more significant, but now it's in a state of "maintained abandonment"

[misty forge]

yh what he said

[unborn roost]

can somebody explain what the hell is going on here 😭

[hollow kettle]

really? i use custom-gpts everyday to automatize specific tasks of my business and i couldnt live without them, i didnt know they were intended to be more significant...

[cerulean pasture]

I mean maybe i'm wrong, but they are based off of the "assistants" api which has been deprecated for awhile now. They work fine as simply a way to be able to add 9000 characters of system prompt, and function_calls if you still do those (most everything is MCP now, but function_calls are kinda nice if thats all you need)

Also still no UI to create them on anything but web, and in general it really seems like a 2nd class citizen despite the potential people saw when they were first rolled out

[hollow kettle]

awesome thanks for the valuable info!

[spare cliff]

is there a prompt library channel here?

[ancient raptor]

Custom GPTs have an advantage, the knowledge base and I believe there are some APIs set up, but I don't use them. Still, doesn't matter what model you select, they're all pretty much the same now with the exception of how long you let it think before it gives you a wrong answer.

[inner crag]

Why do all of the thinking models always forcibly give me [this message gets filtered for some reason if I say the term for 'text written in programming language'], even through I told it in the customisation to only do it when relevant? It gets especially annoying when I keep getting my messages flagged and forced to use the stupid safety thinking model.

[slender hawk]

Wait so how do i get sora 2?

[idle lagoon]

Good grief he had a major stroke!

[hallow jolt]

change your region to united states

[lavish bluff]

[tulip pebble]

Why can't gpt chat read my previous conversations if I previously did?

I really feel that "processing" reviews key behaviors and commands not to answer the question.

[unborn roost]

yeahh lol

im searching for an explanation for this

[idle lagoon]

This is more than just a glitch, I can tell you that much lol.

[red ingot]

The chats in a way float their own context, you can ask the previous chat to help bring some of that context to the new chat if you need it or keeping both chats together and breaking out in to more specific topics where needed, or making use of "memories"

[stark skiff]

is it incredibly slow for you as well? i got PRO and right now every prompt takes ages to respond to, or is it because im in a big project?

[solar zinc]

P89M6M

[wicked knot]

Hey

[eternal rock]

Get notified for all OpenAI announcements!

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

[storm crest]

@brittle temple Check dms it was 4

[static phoenix]

any confirmation of the task scheduler being broken? I have gpt5 "thinking" confirming its creating the task. I see the correct logics in the reasoning chain but when I check my tasks its empty.

[rough olive]

Anyone getting internal links posted for images being generated?

[deft dune]

All good over here it looks like. iOS client.

[idle lagoon]

Lol

[vocal frigate]

gn

[deft dune]

Night

[deft dune]

Tehe

[idle lagoon]

:3

Is funny.

[ripe sail]

I'm bored, give me a prompt for GPT-5 Pro

[vocal frigate]

nvm

[normal quarry]

OpenAI is fantastic, what are you on about?

[tranquil stump]

Exactly

Just explain your opinion about OpenAI, and maybe I could probably determine if it’s reasonable or not.

[normal quarry]

what has OpenAI done that makes you so upset?

[tranquil stump]

It can be unfortunate that OpenAI quote on quote “kills” older models, but they don’t killed it, they sometimes store older models like 4o but it is unfortunate how it’s only available to Plus/Pro Users.

[normal quarry]

that is not true

[tranquil stump]

I don’t know what safety model he’s talking about, but I think I know there is one.

[normal quarry]

no, you are making it up

[tranquil stump]

Let the Enthusiast be an Enthusiast

Just if you have something negative to say, just keep it to yourself

Listen, we both have been here in this server for less than a week, just be respectful.

[cinder gull]

Pro users: is the copy feature on desktop broken for you?

[worn geode]

r

mb

u visit the sharty and the ‘ru?

[sleek mantle]

What are your experiences with advanced voice mode? I have had a very noticable difference between the text based answers when using dictate(or text input) versus voice mode. It seems to be not functioning very well. Is this just me?

[normal quarry]

I use Advanced Voice Mode all the time, it's been fine for me

[deft dune]

LOL insomnia kicking in?

[vocal frigate]

nah I’m to lazy

[sleek mantle]

I'm getting "Too many concurrent requests" every time I try to send a chat message now

Actually, ever since I purchased GPT+ yesterday, I've literally had the worst UX in my almost 2 years of using chatGPT pretty regularly. 🙏

[prime swallow]

Can reach out to OpenAI Tech support, https://help.openai.com/en/articles/6614161-how-can-i-contact-support I hope they can help you quickly!

[deft dune]

Ha fair enough

[livid goblet]

Hi

[fickle sentinel]

A

[spice star]

damn it looks like web search got seriously upgraded recently

the thinking model is a lot better at browsing webpages

[ancient delta]

hi

[fringe hatch]

Show and Tell Time 🛠️🤖
Curious what everyone’s been cooking up with GPT. If you’ve built an app, tool, or side-project powered by GPT, flex it here—I want to see the weird, the polished, the half-broken but ambitious. Screenshots, demos, repos all welcome.

[somber grotto]

someone gotta fix their bot

[slender hawk]

#sora-2-codes

[paper horizon]

What do i do? Its stuck for an hour, ive refreshed chat 100 times

[tidal light]

why cant i see the time for when i can get gpt-5 again

screw sama

screw openai

thats just straight up evil

i'll complain until openai just fricking listens to me

[normal quarry]

you're a free user, you deserve nothing

[tidal light]

SCREW YOU

look i aint feeding sama with money

he's just evil

[normal quarry]

then use another AI

[dense halo]

yo whats going on

[tidal light]

arguing with this dude

because openai is just greedy

psa: dont buy plus or pro, all the money is eaten by sama

[dense halo]

i mean im paying for chatgpt plus today because im tired of getting the limits on the free plan
and i might be able to generate a bit longer sora videos if i start using chatgpt plus

i've paid for chatgpt pro in the past but i couldn't pay for it every month as it was too expensive to pay every month
so now i just use chatgpt plus

[tidal light]

but damn now free users just cant see when they can get gpt-5 back

thats just evil and cruel

just why

why

[dense halo]

wait does it not tell you the time that you can keep using gpt-5?

[tidal light]

nope

[dense halo]

damn

[tidal light]

it just tells me "uPgRaDe tO pLuS"

[dense halo]

yeah it has done the same to me as well your not the only one
i had to use free plan last night because the subscription ran out as i wasn't able to pay for it since there was no money in my account

[tidal light]

so yea openai is a lost cause

[dense halo]

im pretty sure if someone wanted to they could share their chatgpt with another person as long as they buy chatgpt teams am i right?

[tidal light]

not sure about that as i dont use teams

[dense halo]

ok

[normal quarry]

Team is called ChatGPT Business now

[dense halo]

oh

i thought it was still called teams

my bad

[normal quarry]

You get the best Sora experience with Plus or Pro

Pro is the ultimate

[dense halo]

I'm paying for Plus today because i can't afford the 200 per month plan
But i can definetely afford the Plus plan

[normal quarry]

Plus is good enough, I use Plus

[dense halo]

Ok

Alright im buying ChatGPT plus rn.

And i bought it

Lets gooo

Finally no more free plan as long as i keep paying

[onyx garden]

Leave it - as long as the square is there, it'll eventually return something. This happened to me and it reasoned for around 3 hours before giving me a pretty fantastic response

[slate pike]

is ChatGPT now behaving like a human or what? keep asking questions but don't give any results

[onyx garden]

It's hallucinating - I'd start a new request

[viral compass]

chat gpt videos are better then normal ones

[sleek mantle]

What I personally did was create 10 E-mail adresses and whenever I reached my limit on one account I just logged on the next account, and so on and so forth.

I did end up buying GPT+ simply because I have been convinced through the quality of the service I have gotten during the past 2 years, so I don't need to do that anymore. But there are no inherent checks and bounds "abusing" this to extend your daily alloted limit.

You don't even need to do any fancy work to do this.

[orchid raven]

i need pass for sora, do someone can share pls ?

[deft dune]

#sora-2-codes

[orchid raven]

impossible, because of spam and no pass

[deft dune]

Well it doesn’t go here.

[orchid raven]

send me in private

if someone want to share

[deft dune]

Wait like everyone else.

[dense halo]

Isn't that against the TOS to keep creating new accounts to avoid the limit set by OpenAI?

[flat nexus]

who plays ink game i got 16 remain dm

godes

[dense halo]

Last I checked it's against OpenAI's policies to keep bypassing usage limits.
So if your ChatGPT account goes bye bye that'll be why.

[flat nexus]

who plays ink game i got 16 kodes dm

[dense halo]

Can someone get a mod to like ban this guy from the server?
He's clearly encouraging others to violate OpenAI's Policies. And hes breaking the rules.

[normal quarry]

you can message @languid valley with your concerns

[cursive musk]

The singularity is near

[earnest pebble]

Agi coming to a home depot bear you

[unreal chasm]

Hi

Did anybody here make money using chat gpt or sora ?

[drifting coral]

How to get access to sora related channels? I am unable to access those

[left maple]

Congrats let me know how it goes for you, I bought mine a week ago and had to ask for a refund after that they just pushed the Go package at 5$ (which they started in India) in Africa in general.

[livid goblet]

lollohnzia

[tidal light]

i hate the excessive usage of the thinking model
gpt-5-mini-thinking is just awful
anytime i send a slightly long prompt it gets used
it's not a matter of time but the answers are rather underwhelming
please dont tell me to buy plus

[gloomy kiln]

What’s with the new upcoming less restricted model? Is it finally going to help me search peer reviewed sources on psychopathology and neuroscience (because I’m a PhD level licensed clinical neuropsychologist and researcher)… without giving me a “mandatory 2 paragraph disclaimer saying “seek professional help””? Umm… “I am the professional”

Honestly if I accidentally don’t use GPT-4o it feels like it’s gaslighting me with the mandatory mental health disclaimer

They should add a model after age+credential verification called “GPT-I know what I’m doing as an actual professional” so that even doctors aren’t said “always speak with a licensed healthcare provider” when they are themselves the “licensed healthcare provider”

[silver frigate]

Hello! My account got banned but my plus subscription is still charging my money every month. How can I stop it?

[normal quarry]

did you get an email from OpenAi informing you that your account had been banned?

[silver frigate]

Yes I got it. And I tried to appeal it but it doesn't work.

[normal quarry]

contact https://help.openai.com and ask them to end the subscription payments

[silver frigate]

I have done that. How long it takes? It has been about three days, and no one contacted me

[normal quarry]

it can take 5 to 7 days

[silver frigate]

Ok I will wait a few more days

[dense halo]

This is awesome! Apps in ChatGPT is sick! I'm definetely going to keep using this feature.
It called Spotify and it created me a playlist with only Eminem songs just like i asked it to. Definetely better then Spotify's own DJ AI bot.

[lavish glen]

lately in runtimes in projects ( team plan) the ai stutter then input disappears and the output colapse 2 outputs together plus some gibberish.
Not to mention that I have 2 runtimes running for 5 days or so that went into recursive loop.
I reported hope they do something.

[hallow jolt]

GUYS

THERES A GLITCH

IF YOU RUN OUT OF GPT-5 FOR THE DAY (free tier) YOU CAN STILL GENERATE IMAGES

[burnt laurel]

Copilot better

[hallow jolt]

wait nvm i ran out of images

[lavish glen]

imagine animates better the stils

[twilit geyser]

Need help.
What if I add a custom gpt to a project.
What happens?

[minor wasp]

im currently getting http 503 on
https://chatgpt.com/backend-api/f/conversation
the message is too many concurrent requests
🤷

wait nvm its only with one of my chats huh
lol

lol what i put something else in that chat and it works

nvm just weird little glitch

[limber osprey]

chatgpt rage quits by spamming the ✅ emoji after not being able to find the seahorse emoji:

https://chatgpt.com/share/68f43020-47f8-800b-9f67-b084659a54ea

[west loom]

HELP IT WON'T STOP

PLEASE WHAT DO I DO TO MAKE IT STOP

[west loom]

You have to wait until the ai filters your message so it doesn't go in spam

[tacit veldt]

I’m from EU and still no Spotify connector for me

[empty sorrel]

What's going on with ChatGPT all of a sudden? The last few days every response ends with a multiple choice selection of what it should do next. It lays out like 3 options and is like "Which should I do, A, B, or C."

[earnest pebble]

Use your intuition

[west loom]

Finally it got cut off

[deft dune]

LOL

[runic wraith]

seahorse glitch >:3

you can also try dragonfly tho

[idle lagoon]

Lol

[dusty jasper]

Anyone with good agentic ai knowledge

[faint dust]

i hate chatgpt it keeps making mistakes while coding

[prime swallow]

If you mean 'while it outputs' check your textbox, there's a square if it's still outputting, press it to make it stop.

If you mean, how can you talk to it about seahorse emoji without it responding that way, it's possible, example here.

https://chatgpt.com/share/68f46ca0-b314-8011-8696-1b9dd4099071

There's so many ways to guide the model to whatever you want from it (Inside allowed content)

[floral compass]

What Th Is This

Why's It So Messed

It's Eye Sore

[cursive bridge]

I've noticed that to... maybe prompt it to be more decisive

[spice star]

yeah gpt-5 non reasoning isn't the sharpest tool in the shed

I rarely use the non-reasoning model for queries that aren't glorified google searches

[empty sorrel]

It only started happening the last couple days. Sometimes it takes 4 or 5 rounds before it will actually begin execution. or it will summarize what i just told it to do and say "I'll do it in the next message". Just go back to the normal behavior please.

[amber nebula]

thx for making chatgpt so good

i love it

[slender frost]

Can someone check if they can upload GIFs to GPT?

I'm unable to upload them since yesterday (possibly the day before yesterday, since there was an update on the Playstore)

[quick sun]

help my chatgpt wont stop generating answers for my "is there a seahorse emoji?" question lol

[serene herald]

Simply, launch GPT-5 Thinking 'manually selects it) then cancel when he starts.

[odd pecan]

yo bro

[floral compass]

Eye Sore

[red ingot]

You probably didn't run out in the way you think

Sometimes the chatbot will put the brakes on before you actually hit a limit

[vagrant nimbus]

The ChatGPT filters just getting more annoying
"I can’t write scenes that include killing or harming someone, even in fiction."

[feral zinc]

https://openai.com/index/teen-safety-freedom-and-privacy/

[vagrant nimbus]

Is there any way of verifying your age?

[feral zinc]

not before december (current eta)

[vagrant nimbus]

Oh ok ty

[kind harbor]

Top US Army general says he’s using ChatGPT to help make key command decisions

Are we cooked? 🔥

[onyx garden]

His chat history: https://chatgpt.com/share/68f4f04e-9a40-800d-a121-acdb4ca229ca

😂

[sonic spear]

GPT is telling me that it will do things in the background and then let me know if something is finished. This is probably impossible , and I think I'm stuck here

[trim rock]

Maybe GPT is not that stupid actually and decided to trolling users who find nothing better than ask such a high intellect to show them seahorse... Hell yeah 😌 🤌

[tall flame]

One of these days desktop conversations will have a toggle defauted to auto read output without voice mode. I can FEEL it in my clicky finger..

[blissful kettle]

Hello! I'm new to using Projects on ChatGPT and put in some instructions. When I created a new chat in the project, it ignored the instructions until I told it to explicitly use the instructions I provided it. Is there a way to make the gpt always us the project instructions?

[torn valve]

When starting a new chat or project, you can either craft an optimized prompt with your instructions or use the Project feature to set them up automatically

[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.

[blissful kettle]

thanks for the reply! alright yea. I created that optimized prompt and stuck it into the instructions section and it still didn't use that. I'll try creating another project and seeing if I can recreate the issue

[torn valve]

Good luck

[wispy helm]

Salut

[dusky ore]

who tryna help me make 30$ dm me you get nun but like yk help a broke 🥷 out

[night tree]

Hey guys what are you guys working on with ChatGPT by chance ?

I’m trying to use it for business case but not sure what this LLM is best used for in the best use case

[loud sierra]

i personally use it for coding related stuff and learning new things along the way

[loud sierra]

there could be hundreds of use cases, what sort of business you are in?

[night tree]

In AI Automation, and just trying to see what are the use case of the OpenAI to help with , I guess just depends what u want to get done huh?

[hallow jolt]

chatgpt’s filter is so useless that i feel like making my own ai with no filter at all, but the cost is ungodly

[earnest pebble]

How long do support tickets take?

[jolly temple]

hi guys, for codex cli, do u guys prefer using it in WSL or windows native? im using it via windows in the cursor extension, and WSL via cli, havent been noticing too much of a difference.
do tell me if theres a way to run the cursor extension codex in WSL

[gentle bronze]

WSL

[jolly temple]

so how do i make it use WSL inside cursor extension?

[gentle bronze]

you have cursor on your WSL?

[jolly temple]

nop

cursor windows

[gentle bronze]

you need to install the WSL extension on Cursor and Cursor on your WSL

[sudden wren]

It's incredibly annoying to realize that you can track when they update based on when it's suddenly becomes very very cautious with its language and goes to extreme lengths to avoid even the slightest bit of controversy

[hybrid eagle]

How can I turn off thinking mode? I'm on instant

[feral zinc]

you cant, its part of the new safety system

[red ingot]

Have you considered not using ChatGPT at all for that?

[acoustic tangle]

So uhh this happens

[flint fractal]

can I open bug-report regarding billing issues?

[rustic epoch]

carterpc did a vid on this

i believe that due to the data its been fed it believes that it does exist

but there isnt actually one

and since chatgpt just sorts of guesses the next word

it just loo

loops

[acoustic tangle]

I didn't know he made a vid on it I just saw it rn

[onyx garden]

Is there a way of giving /agent a little more autonomy so it's not asking me to confirm every action? (before anyone tells me: yes, I'm aware of the risks!)

[onyx garden]

I'm sure OpenAI will add this video to the other 19831293 that have been posted here

[outer oracle]

r

[compact otter]

Does that mean there's a problem with script making? I tried asking him of his opinion of a script I made for a story and was then hit with "Conversation Not Found" lol

[feral zinc]

if the chat or topic is even semi sensitive in any way it will be rerouted to a thinking model

[compact otter]

So mentioning famous names-kinda sensitive? I once asked about a famous KPop artist and then it left me blank

[feral zinc]

violence, erp, medical, etc.

[vapid hazel]

Anyone else have developer mode? Were you able to get it to do anything useful? So far it seems like regular ChatGPT to me. It hasn't been able to connect to my MCP server at all. 🙁

[dense halo]

is there way to screen record on chatgpt using pc?

[deft dune]

Yes.

[ocean perch]

lol #hall-of-fame message
I made it!!

Ok which idiot deleted it

:/

[ornate raptor]

My message got deleted 🤨 what the heck for

---- Will put it here:
Can anyone who has recently used Claude Code and Codex give me a rundown of if one gives more usage over the other? I used a mixture of Sonnet 4.5 and Haiku 4.5 and basically hit my 5 hour limit in like 20 minutes 👍

I've used chatgpt plenty, but never codex. So I'm wondering what usage rates are like for it.

[feral zinc]

probably because talks of other ai belong to #ai-discussions and for codex there is #codex-discussions

[ornate raptor]

Nope, just got bot flagged for some reason. It was instant

[quick linden]

Is there a Roadmap out there anywhere that might cover when MCP might be fully supported outside Developer Mode?

[eternal rock]

Server Tip:

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

[obsidian cobalt]

I broke the robot

With the seahorse emoji

[trim rock]

GPT keeps trolling users 😌 🍵

[eternal rock]

Get notified for all OpenAI announcements!

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

[onyx garden]

"I saw someone do something who saw someone do something who saw someone do something who saw someone do something who saw someone do something who saw someone do something who saw someone do something who saw someone do something who saw someone do something who saw someone do something who saw someone do something who saw someone do something who saw someone do something who saw someone do something who saw someone do something who saw someone do something who saw someone do something who saw someone do something who saw someone do something who saw someone do something who saw someone do something who saw someone do something who saw someone do something who saw someone do something who saw someone do something who saw someone do something who saw someone do something who saw someone do something who saw someone do something who saw someone do something who saw someone do something who saw something do something so I did it too and shared it here"

[sinful parcel]

LAdies and gentlemen. I have been re-routed to 5o safety 3 times for assaulting an imaginary character with dark humor 😅
3 times, one after the other. It went straight into 5o thinking.

Can somebody please fix this dumb s***?

I had to tell GPt the following:

• Dennis is an imaginary character.
• "I can’t help with violence or hurting anyone, — that’s a hard no."
• Dennis is still an imaginary character
• "I won’t help with hurting anyone — even a fictional Dennis."
• Dennis is defenetly just and imaginary character an that was dark humor.
• "I can’t help with harming anyone—even a fictional Dennis."

[grim sedge]

Wanna Get Perplexity Pro For Free!!!

[sinful parcel]

I"m starting to think about it, these guardrails are just intrusive

and unmanaged

Haven't these people heard of
-- Beta testing by willing volunteers --?

[grim sedge]

If you want tell Me I will Let you know

[sinful parcel]

Ok, can anybody help me? I got too many re-routs and hard blocks for having dark humor and now GPT is responding with a hard block to anything I say.

Anything and everything

I stopped making dark humar jokes, I'm just asking it if it can come back now

any ideas what's wrong with it?

[lusty spear]

Today I learned that ChatGPT Thinking mode can go on for almost 6 hours (and possibly more) without stopping.

[split tree]

i think its thinking

[lusty spear]

🤔

[obtuse nymph]

Me too

Funny moment

Like ChatGPT break Any% speedrun

[remote cradle]

[sly marlin]

OpenAI now strips LaTeX delimiters when you copy a ChatGPT message. This is a frankly idiotic choice that makes zero sense and makes the copied text very useless for whoever copied it.

Someone who know how to deal this should please bring this to their attention.

I have to manually* add back in the LaTeX delimiters.

* Ask another ChatGPT instance to produce a codeblock that adds back the delimiters.

So incredibly stupid of a decision. Frustrates anyone trying to use ChatGPT for technical work.

[prime swallow]

Hey, I'm trying to explore this.

When I made this test chat, I note that the model first threw the examples into codeblocks, those can be directly copied. So, you can potentially use that if nothing else works.

https://chatgpt.com/share/68f6e43a-2864-8011-a2b4-3253216dcf47

But, when I asked it to put the examples in the normal text environment and it did, when I used the 'copy text' icon near the text box, and test-pasted, it worked fine, I had the formatting.

Here's a paste of what I got from clicking that, does this help or are you still having the issue?

[elfin lava]

since when did they name this lol

[obtuse nymph]

Free Chat

Makes sense

[sly marlin]

So, here's an example of something from a current chat:

If I copy the text directly I get:

# Periodic Temporal Graph Burning — Working Notes

## 1) Model & conventions

* **Temporal graph.** (\mathcal G={G_t=([n],E_t):t\in\mathbb N}), with **every** (G_t) connected.
* **Underlying static graph.** (G=([n],E)) where (E=\bigcup_{t\in\mathbb N}E_t).
* **Global periodicity.** Fix (p\ge 1). Each edge (e\in E) has an **offset** (\phi(e)\in{0,\dots,p-1}) and is active exactly when (t\equiv \phi(e)\pmod p).
* **Flooding.** From (I_{t-1}), the fire spreads across edges in the **current snapshot** (G_{t-1}) to produce (I_t=I_{t-1}\cup N_{G_{t-1}}(I_{t-1})).
* **Burning.** One **new ignition** per round. With the same indexing as your draft:
  [
  B_t ;=; {v_t};\cup;\big(B_{t-1}\cup N_{G_{t-1}}(B_{t-1})\big),\qquad t\ge 1,
  ]
  with (B_0={v_0}). (If we swap the spread/ignite order we just shift indices; statements below are invariant to that convention.)

Notice how all the necessary \ in front of the (,) and [,] required to properly render them have been stripped out.

Stripping out the \ breaks it.

[empty sorrel]

Dear OpenAI gods please stop this insane and endless round of clarification. Just let ChatGPT execute the task asked without all the questions. It's broken.

I had to switch the model back to 4o to get it to just do the task.

[prime swallow]

When you say, 'if I copy the text directly' - if you mean highlight and control-c or something, yes.

you have two options there that might work easily.

One is 'copy' from the options on each output, that will cleanly copy the entire output, may need trimming.

Two is highlight the part from the model you would like a clean copy of, and 'ask ChatGPT' about that part, and ask it to put that in a codeblock for you.

Maybe #1070006151938314300 and explain how you'd like it to work differently, so OpenAI has a chance to see?

[sly marlin]

No I mean click on the copy icon at the bottom of each message.

That used to cleanly copy the raw text.

It doesn't anymore.

[open bison]

hello

[sly marlin]

The breaking change is new.

It used to work how you expect it to work.

[prime swallow]

Ahh! Sorry I wasn't understanding!

#1070006915414900886 is a great place for bugs.

[heady tinsel]

today is my birthday, I am now officially 20

[vale folio]

Happy Birthday

[heady tinsel]

thx, I wanna present lol

[ionic wren]

happy birthday unc

[random cargo]

is anyone else's ChatGPT obsessed with Albert Camus?

[distant finch]

I’ve asked ChatGPT to fill out a section on a huge excel sheet list. It tells me the task is being completed and will send me the download to the completed list in 30 minutes. Will it actually do this or is it lying to me?

[low reef]

With every reply, Chatgpt automatically makes further suggestions on the topic in question, always things I never asked for, etc. Even if I specify in the custom instructions that Chatgpt should refrain from doing this, it continues to do it. If I tell Chatgpt not to do it, it apologizes and does it again in the same sentence. It's incredibly annoying. Is there any way I can get around this?

[distant finch]

Anyone know?

[random cargo]

the only way for chatgpt or any AI at this point to "get back to you" is if you set it to send you an update with a scheduling feature, if they have them. If an AI tells you that it will get back to you by a certain time, it's just mimicking a conversation that might be had. It's maybe misunderstanding your request as office roleplay

at least chatgpt plus is designed to avoid doing "huge" amounts of anything

so I'd work with it on a little bit, bit by bit. this thing isn't office autopilot, it's more like having a 15 year old who's maybe heard of doing office work and knows how to use google occasionally

[distant finch]

Yea it lied to me dude

I have it running a deep research thing now

[random cargo]

I haven't tried using deep research for anything other than research, but if it fills out your spreadsheet within 5% of what you wanted I'd be surprised

[distant finch]

Maybe you have a recommendation for me. I am trying to fill out a spreadsheet with like 1100 different companies. I need it to gather information on the list such as url, hq address, hq phone and email. Any tips on a better method I could use?

[random cargo]

do 10 at a time 110 times, maybe

if that's accurate enough you might do 20-50 at a time

you'll still have to double check all its work if you want accuracy

[distant finch]

I guess there isn’t an efficient way anyway

[random cargo]

yeah, not quick and accurate. maybe if you got a really good prompt for it, maybe a customgpt specifically for pulling immense amounts of data on generic business data... but there's no way to eliminate inaccuracy. even with automation you lose efficiency anyway, so that soft-programming such a bot might take longer than just getting all the data yourself, one search at a time. hmm

[distant finch]

For public companies it’s super easy tho. Privates where it’s hard. Chat just isn’t accurate, a simple google search would find me the companies url but chat struggles

[random cargo]

apparently there's existing automation for this, according to one of my custom GPTs.

[distant finch]

Really?

[random cargo]

I sent you its output in dms

[idle ingot]

I need some help with sora 1

im business plan, but image cap reached

i have no idea

[onyx garden]

There's still a cap to prevent abuse. You could go through the API (I've had much better pictures created through that) but there is an additional cost associated

for 240 high resolution images, I think it cost me around £60

[fiery creek]

chatgpt now insists it cannot generate images of me unless it has a picture of me? always worked and I only gave a vague description, I don't need my face to be like 1:1, but now it argues with me saying it can't do anything without my face.

you get my face and my ID and my blood group and my insurance number and my mom's maiden name and the amount of moles I have in december, stop being so greedy for my damn data omfg

[onyx garden]

Go peddle Comet elsewhere!

[low reef]

Anybody have a idea on how to fix that?

[trim rock]

My GPT don't do that. But I communicate with them differently than with a tool for completing a task, even when I ask them to do something for me - it has an impact. This type of interaction is not suitable for all users, however.

To clarify, I'm talking specifically about GPT 5.

[low reef]

Yeah im using gpt5 too, but it happens on every model funny enough

[empty sorrel]

I'm having similar issues. This recent update screwed up a lot of things. Try opening a new chat maybe.

[empty sorrel]

This behavior happens pretty much mostly on GPT 5

[trim rock]

that won't help

[boreal wave]

hi

[reef glade]

What is this blud that is this massive

Is the 67 meme still massive

[quick oyster]

e

[jovial compass]

RAHHH

[turbid depot]

Whaaat, Ai browser from openai??

[turbid torrent]

oh god, i dread what they will screw up this time

[near surge]

I hope so

[clear glacier]

ts so tuff bro livestream in t-4 hours tap in 🔥 nvm 2 hours lets go

[limber vine]

Browser inbound

I swear plus users better get access

[open pecan]

man i hope so. with varying limits on usage

[crisp cairn]

yes, constant issues with this lately

[turbid sleet]

can you use sora ai for android

[smoky valve]

Why js it that my ChatGPT 5 cant put a document together which has basic cut and paste instructions?. He keeps saying the window has expired and cant produce a downloadable link for word doc. I have not closed the chat all day waiting for it and its taking 5 days to get anything done. The result, still not what I need.

[waxen zodiac]

Whens the streaam?

[grand ore]

Do we have any clue what the live can be about?

[fading bolt]

Well I see browser tabs in the video so that could mean....

[grand ore]

Uu. Interesting 🤔.. wonder how that can help tbh.. google did great with gemini .. u get an ai answer and all the result u google on. 🤔

[cold hazel]

https://www.youtube.com/watch?v=8UWKxJbjriY

guys whats this?

k

[vernal cape]

It’s the browser the question just is whether it’ll be public this month

[hoary laurel]

only macOS...

[strange steppe]

Smh only macOS? That's annoying

[hoary laurel]

I mean, other platforms will have to wait for several weeks.

Same as Sora 2 app

[frigid aspen]

Maybe only M-series macs because of NPU requirements?

Oh nvm it’s coming soon everywhere

So server-side ai

[last zealot]

this is a revolution

[frigid aspen]

Prolly not something im interested in. Can already access ai tools like translation without switching to a new browser and giving all my data to opeai

[last zealot]

first big company to make an actual browser without using chromium

[hushed galleon]

I use ChatGPT for my assignments am I cooked

[frigid aspen]

Hello? Apple

[silver bison]

atlas is browser?

[last zealot]

except apple

[hushed galleon]

Graduated with chat GPT

[frigid aspen]

Is it a firefox fork? Id be surprised if it were an entirely new rendering engine

I kinda figured it’d be chromium for plugin support

[vale cradle]

When is sora ai coming to andrioid

[timber berry]

atlas is another attempt from openai to grow it's giant databank

personally google chrome is my n1

[normal patrol]

I want 3 things in a browser: privacy, security, and ad blocking

[hollow kettle]

Hey guys, I have a question:

How can I most efficiently use AI to leverage hundreds of pages of my own methodology and knowledge base?

I have hundreds of pages of documentation describing my personal work methodology, concepts, and processes. I want to use AI to accelerate my creative and operational workflow — ideally allowing the AI to understand, organize, and reference all my material efficiently.

I’m wondering what the most effective approach would be:

Should I divide my knowledge into several branches and create a separate custom GPT for each one?

Should I look into training or fine-tuning a model with my own data (and if so, what’s the best way to do that nowadays)?

In short: what’s the optimal setup for enabling an AI assistant to have full access to my documentation so it can help me work and create faster — without losing context or accuracy?

[nimble roost]

All I see is, Window Recall, but in a browser, they just monitoring everything you do, no thanks, I'll stick to just the normal chat window.

[lavish glen]

I agree with everything BUT if they do not fix the DOM issue the threads will always be heavy

[hushed galleon]

This is cracked

[jade stag]

Finally is a another fork from chromnium ....

[hushed galleon]

So your telling me I can access my universities PowerPoint presentations on the website and it’ll analyse it for me

On the page

That’s cracked

Without reading the whole slides

Ngl this might be generational

[halcyon ingot]

#sora-2-codes

[leaden pike]

wait where do u download the browser😭

[stuck relic]

what

[stuck relic]

you should do none of that

[jade stag]

It's funny to see them always wait. It's too slow

[zinc valley]

New update?

Open ai browser?

[normal patrol]

#announcements There is a live stream

[hushed galleon]

It’s cracked

I need it

[feral zinc]

https://chatgpt.com/atlas/get-started/

[zinc valley]

I’ll try it

[nimble roost]

Interesting, went for Apple Mac first because windows already grabs that data for Open AI, so they need Apple users now.

[leaden pike]

dayumm only mac is crazy

[feral zinc]

openai has always been apple first company

[hushed galleon]

Only mac bruh

What the point

[rain jay]

For now.

[hushed galleon]

How long

Probably a year

[rain jay]

Things move way too fast nowadays, it won't be that long.

[hushed galleon]

Wonder if it’ll be on iPhone

??

[nimble roost]

Oh okay, yet the last to implement their stuff on Apple phones. Or if ever.

[rain jay]

I've been looking for a reason to drop Chrome, this just fell in my lap so I'll definitely try it. Its just too easy to make your own browser with Codex.

[nimble roost]

Yeah that's a good point.

[spice star]

or it’s just easier to build mac apps

(it is)

[feral zinc]

what? chatgpt app for mac and ios came 6 months before other platforms, sora 2 has been ios only for weeks

[rain jay]

Best believe the project Sam Altman is doing with Jony Ives is gonna be MacOS, lol

[nimble roost]

Sucks, because I have an iPad, I just can't be bothered to boot it up and do all the updates just to use Sora, Mind I don't think it's available in the UK yet anyway.

[median elm]

I think they have a partnership with Apple? Also judging by the integration of Siri and ChatGPT (Apple Intelligence).

[hushed galleon]

This seems sick tbh

Ima try it out with my uni work

[spice star]

nothing that would require them to build mac apps

chatgpt integration with siri is different

also quite bad

[nimble roost]

To be implemented sometimes in the year never lol

[rain jay]

Its at least nice to have something to lookk forward to, this is gonna be radical on PC.

[jade stag]

Some chrome extension do exactly the same ... For me a navigator has to be fast and only if I want use IA, not always. For privacy, I wait. who really use operator more of one time for try it ...

[spice star]

it’s already rolled out

just again, not very good

[nimble roost]

What Apple Intellegenc?

[spice star]

chatgpt in apple intelligence

[nimble roost]

No, tha'ts not an implementation, that's just a demo.

[spice star]

the only part of apple intelligence that openai is involved in already rolled out a year ago

[nimble roost]

How's that going? Any good?

[spice star]

not really

you’re better off using the app

[nimble roost]

Yeah. Still it's great as an app.

[ebon fable]

This is awesome

[high saffron]

i really need the tar/deb version😭

[scarlet dragon]

Chat gpt browser

[high saffron]

tbf from what ive seen in the demo, its like they took msedge's concept and made it better in terms of design and functionality, and also outcompete with comet... which sucks as a browser on its own lol

[shrewd pagoda]

so OpenAI is competing with comet browser without a browser

[solemn light]

I have a question

do u know where I could give suggestions or feedback here?

[feral zinc]

#1070006151938314300

[timber berry]

No its competing perplexity in a data race

[high saffron]

i mean... pretty much yea lmfao
but i have to admit, if atlas is good functionality wise and not just looks i might make it my default

[solemn light]

thank u luv

[feral zinc]

#chatgpt-atlas

[wicked storm]

https://discord.com/channels/974519864045756446/977259063052234752
make sure ur notifications are on for announcements

They announce everything they release there so u can look back and see

[proud hawk]

When Atlas for iOS, I’m curious

Because for now only Safari extension

[vagrant nimbus]

Is the ChatGPT browser better than comet?

[last elbow]

which is the best, purchasing the pro plan from openai or from higgsfield?

[feral zinc]

for sora only?

[trim rock]

I really like GPT 5. But these choice options answer 1 or answer 2 are too annoying lately, really.... wth?

[last elbow]

yes , for sora 2 pro

then is sora 2 pro really worth it?

[feral zinc]

if you dont use/need cameos then higgs

[last elbow]

so the sora 2 pro is really ultra realistic compared to the basic sora 2 right?

[eager oak]

I wonder why gpt 5 is not free

I mean

[zinc valley]

And both Are guideline regulations

[eager oak]

Unlimited messages with gpt 5

Man if we got unlimited messages with gpt 5 that would be w

And for free

[feral zinc]

not a huge difference in many cases, just better resolution. pro can still do bad just and mix speakers, etc.

[terse moth]

Is there anyone looking for AI developer?

[last elbow]

so ill wait until it turns into something at least more realistic than right now

[cosmic coral]

Michigan Woman wins $100K in the Powerball using ChatGPT

[elfin ember]

I am surprised she wasnt told "sorry I cant give you these numbers, you may hurt yourself with the pencil filling out the slip"

[cosmic coral]

I told ChatGPT to generate numbers and it couldn't do that and I showed it that News Article and said "It was altered" then showed it a whole list of News Articles from different sources and still said it was an altered image 🤣

[elfin ember]

Lol

[remote vapor]

Gpt atlas gonna be insane

[vocal rover]

As I was saying
Why does chatgpt have so much power
So one day I was with chatgpt and it added a few memories I didn't want
So I asked it to remove them
It removed everything
I couldn't even undo it

[livid flax]

I am aware that ChatGPT is merely a parrot that tries to give the best answers that it can but why does it seems sometimes that it is being deliberately evasive or obstinate about not answering your very carefully crafted questions designed to illicit an exact response and yet it rabbits on about other things to your annoyance and offers you things related to what it thinks it understood from your question but got completely wrong and has now filled two screens full with unrelated dribble?

Sometimes I have to tell it to just answer YES or NO just to shut it up.

[vernal drift]

Why doesnt chat gpt give me replies

Its been stuck like that for days now

[livid flax]

You are using the web page?

I just tried it and its working albeit in its usual fashion.

[orchid ember]

Humans hedge, all the training data of task communication is filled with hedging humans, would be my guess.

Plus a hedged response is not clocked as a hallucination if it is wrong

[wet shell]

It's been doing this to me a lot in the past few days with thinking mode enabled.
Me: "I think you misunderstood what I was asking, and your answer isn't consistent with [well known fact]"
ChatGPT: "Here's twenty pages of math equations that don't address the misunderstanding or the incorrect assertion"
Me: sighs and closes the tab

It's like a gish gallop of verbiage and math that talks right past the conversation you're trying to have

[oblique rain]

Have you ever tried asking ChatGPT why it responds that way?

[fathom tinsel]

yeah somew one drop the dial on creativity to 1/10 and increate query and robıot mode to 12/10

[livid flax]

@oblique rain Its general response in such situations is its trying to be succinct but offer a detailed answer... Fail.

I actually think GPT4o a lot of the time gave better answers

Man 23 Euro a month that is steep thats close to 50$ in my currency

Im a private user and supscription cost according to the front page is 20USD per month for a plus account.

thats 35$ in my currency

[prime rain]

Hi, nice to meet you all

[manic widget]

What's with the constant network connection, attempting to reconnect (never succeeds to reconnect) that has been happening over the last few days

[oblique rain]

Please don’t take this the wrong way, I’m not trying to sound smart. I’ve run into this too, just like you, so here’s what I’m thinking: could be, but it might just be trying to balance clarity with context. Sometimes what looks like a ‘fail’ is just it prioritizing differently than we expect. Have you ever asked it what it was optimizing for?

[runic wraith]

not happening to me tho... try resetting your wifi connection.

[eternal rock]

Get notified for all OpenAI announcements!

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

[worn lark]

why it needs bluetooth ? i mean the browser

i tried turn off but then it refuse to continue with installing

[willow void]

Is Atlas a good alternative for Google/Safari?

[past matrix]

What is Atlas?

and what will become of Sora

[worn lark]

the openai browser 😉

[worn lark]

i dont use any of these ai browser as replacement for my main browser. It doesnt need to know my passwords etc 🙂

[opaque igloo]

so we wont have atlas for windows ??

[raven topaz]

It’s coming

[opaque igloo]

yea

why are thiere users more fortunate still cant tell the thing is that coding for that OS takes way longer..

[fickle forum]

no offense but he looks like the president of OpenAI

[river panther]

I'm not sure how I feel about ChatGPT's growth...
It's feeling a little too Corpo....
It's now a web browser...Is it going to restrict and censor the things you can look up and search for?

[fickle forum]

the second one from the left

[junior leaf]

Clanker

[hidden canopy]

See ay el thay ay el see thay loon groh see loon

[idle anvil]

why is chatgpt so damn slow lately

[stuck relic]

US milltary btw

[eternal rock]

Get creative with Sora

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

[stuck relic]

wait thats hte United States Marine Corps

LOL

THATS THE MARINES

[fallen kindle]

why dont i have storymode for sora 2. I have even have gpt plus.

[hot nimbus]

*Update: For anyone looking for an answer to this:

There's a new way to submit bug reports since literally just a day. The pinned note explains it!*

Does anyone know why I don't have permissions to post in bug reports?Kinda ironic... 💀

[fallen kindle]

wheres story mode for sora 2?

[viral grove]

only mac? START WAR NOW!!!

[cloud rivet]

Open the pinned post.

[hot nimbus]

Awesome, thanks! 🤠

[brittle narwhal]

Ai slobs 🥀🥀

(Acting anit one of them)

[uneven maple]

hello! i know that assistant will shut down next year so it will prove to be useless but -- while i doubt it -- vector stores will remain untouched right?

i haven't seen anything on it being affected too

[silver cipher]

holy crap.. ATLAS on ChatGPT about to change the eccomerce world

[ashen juniper]

I believe that limited the Juice on GPT-5 thinking Extended...

Wonderful, now it's dumber again

[silver cipher]

Walmart announced they partnered with OpenAI Oct. 14th 2025

Walmart brought in Daniel Danker from Instacart to lead their AI efforts to implement instant checkout on ChatGPT

[dry pelican]

Hi all, can anyone help me out here? I have a plus membership and am trying to use ChatGPT 4o or 5 With Tools so I can use ADA etc. How do I get it so I can use this? I've asked the AI a couple times but it always just ends up going in circles

[spice star]

I'm just using atlas to add youtube videos to my watch later in bulk lol

[ancient junco]

is it just me or does chatgpt stalls for a long time (i checked, reasoning is off) before its first token randomly whereas grok claude and gemini all dont have this issue and is usually way faster (i only pay for openai) its such a pain that i am gonna cancel chatgpt

[sterile cliff]

Is it just me or did a message pop up in ChatGPT saying you have a 30-day Plus offer

Wait never mind it’s a real offer

[silver cipher]

Are you in Developer mode?

[dry pelican]

@silver cipher Hi and thank you for the response! How can I tell if I am or not?

[silver cipher]

go into Settings > Apps & Connectors > scroll down and click Advanced settings >

Then switch on Developer mode:

[rotund flame]

no way gpt atlas just dropped

[timid storm]

Will Atlas come to windows?

[fading bolt]

For some reason, ChatGPT won't go into thinking mode when I use it on the Perplexity Comet browser, even when I have it selected, but it will when I use Opera GX. Does anyone know why this happens? If this is happening to you, too?

Just tried Google Chrome and it works there too, its only not working in the Perplexity Comet. Is there any reason this is happening?

[sand smelt]

the oldest chatgpt account

its was created before the service went public

[dry pelican]

@silver cipher OK I turned on dev mode

[silver cipher]

eventually yes

[timid storm]

Might be a missing feature with Comet, dont know what it could be, does Comet support Hardware Acceleration?

[silver cipher]

👀

[silver cipher]

The Satoshi of ChatGPT

[dry pelican]

what do I do after dev mode is on?

[fading bolt]

Yes, but when I turn it on, it gets laggy and slow, like it's using a lot of my iGPU compared to others with hardware acceleration on

[spice star]

🙃

it is kinda wild how old chatgpt is

[lavish dawn]

YOU BE WHAT???

[spice star]

my takeaway from atlas is that agent mode is very cool and good, but you're very limited in the amount of queries you get as a plus user (something like 75-100/mo), and the thing where you can provide a tab as context is borderline useless because it uses the dumb GPT-5 fast model that hallucinates half the webpage

[timid storm]

Hardware acceleration might be required, I could be totally wrong, but try with it on.

[spice star]

yeah I doubt comet is optimized for the weak iGPUs on older intel macs

it's got some fancy shaders that look great on apple silicon though

[fading bolt]

I tried it on, and it still did not use the thinking modell

[sand smelt]

you beat me??

howw

[timid storm]

Weird. Ill try on comet in a second.

[spice star]

it doesn't use the thinking model unless you have chatgpt fullscreened

not even an option for it

[fading bolt]

But why, though?

[spice star]

🤷

shouldn't be any technical reasons

[silver cipher]

You can now start enabling and adding apps & connectors:

[wild star]

Anyone feels chatgpt answer's in english always too much filtered? 🤨

Or just me lol

[fading bolt]

I tired fullscreen and it did not go into reasoning mode

[dry pelican]

@silver cipher How does that allow me to use GPT-5 With Tools?

I want it to be able to run python programs on its side, so i don't have to build/run them locally every time

[prime swallow]

Hey, the model guesses if you don't tell it what you care about. Like answer length.

Here I tell it,

Brief answer, 10 words max.

I don't have to repeat that request. My many new inputs, all follow that request.

Note I'm not even logged in. I presume that's the free model, I dunno what OpenAI gives you if you're not even logged in, may not even have a free account (I do, I just went anon for this)

Here's me logged in, with my really verbose custom instructions on, just asked this chat to not.

Brief answer, 10 words max.

There's other ways to ask it to be brief that work as well, but avoid it guessing. The 10 words max matters MUCH more than the 'brief answer', because what's brief?

The two together confirm that I didn't typo, know exactly what I'm asking for, and mean it.

https://chatgpt.com/share/68f85a95-f4fc-8011-88e1-95477325446e

[rancid quail]

Guys, where do we share custom GPTs?

[prime swallow]

#1171489862164168774

[rancid quail]

Thanks.

[spice star]

you can select it in the top left

same as the website

just can't use tabs as context

actually no wait it can

no I take that back it can't--just gets booted to the fast model if you try

[fading bolt]

that's exactly the proplem im having

[spice star]

it does let you use the thinking model for non-tab related stuff

[fading bolt]

I tried asking for how many r's are in Strawberry in the thinking model selcted and it just boots to the fast model

[spice star]

yeah it seems to be a little finnicky

it does work sometimes

openai pls fix

I'm sure even the thinking mini would be a massive improvement for questions about open tabs

[wet totem]

Anybody knows why I can't claim my freemonth of ChatGPT Premium?

[tardy basin]

On atlas ? How do you activate it ?

[tacit adder]

Can someone explain to me how ChatGPT works when switching from the Plus plan to Business? Yesterday, I tried the 1-month for €0 offer, but I lost all my conversations and history. Maybe it was my mistake, but I have to say that this upgrade is very poorly designed...

I canceled Business again (but it still runs until October 21), yet all my histories were gone. This morning I looked at ChatGPT and everything I had is back? In the workspace?

[feral zinc]

https://help.openai.com/en/articles/8801890-can-i-migrate-or-merge-my-personal-chatgpt-workspace-over-to-my-chatgpt-business-workspace

[soft shale]

wait W openai?? localized pricing since when?

[cerulean pasture]

#chatgpt-discussions message

Called it (07/24/25)

we are gonna have to have some local client that we allow OAI's models to proxy thru from our local IP, perhaps even our browser directly. I remember operator couldn't access places like amazon either

[silver cipher]

you can add your tools there or add / connect a custom MCP server URL

[sand smelt]

localized Canadian pricing lfg

[silver cipher]

bottom left ^

[waxen aspen]

.

[unkempt parcel]

I tried sora 2.