chatgpt-discussions

[tiny salmon]

Do I need to pay for pro so I can get a functioning 4.5? It seems that the goal of 4.5 is to not work right so that I can be pushed into spending $200 for pro even though none of the models work like they used to before o3 dropped.

[fickle forum]

lol

[cyan flame]

i understand you

[fickle forum]

why ur message is blocked

@tiny salmon

[shrewd lodge]

Same with codex too

[cyan flame]

My mother can't afford to pay me for it.

[fickle forum]

oh

I see

[shrewd lodge]

If OpenAI won’t be more generous to cool stuff then I’ll just use Jules

[fickle forum]

nvm

openai have built a lot of things based on o3

gpt-5 about to be released very late probably

[shrewd lodge]

Codex, operator, deep research

[shrewd lodge]

Probably not imo

Google I/O already probably put OpenAI to rush

And then Anthropic event just blew them away on benchmarks even in o3

I think they getting a bit worried

[fast pewter]

is the projects folder feature rolled out incrementally? I saw it for a split second earlier and now I can't see anymore

[fickle forum]

like I care.its unaffordable for me to use it.

lol

[silk crypt]

Then simply admire it from distance

[versed zephyr]

has anyone hit a codex limit on team

[fickle forum]

why not for plus

plus deserve everything

[granite arch]

Resources cost money...

[shrewd lodge]

But plus users pay the money

[fickle forum]

I've paid for over a year and a half

[granite arch]

And who's paying for all the extra resources available for Teams?

[fickle forum]

over three halves year!

[granite arch]

Just because you pay $20/month doesn't give you some special privilege...OpenAI is a company...you gotta pay if you want to use their resources.

[shrewd lodge]

Eh

They should take notes from gooogle

Almost all their fun experiments and AI’s are on the cheaper plan and some are even free

[fickle forum]

it was a lab.OpenAI is not for sale,neither is the mission,which is to benefit humanity.

lol

[rugged igloo]

@prime swallow

[shrewd lodge]

was

now they want money

[fickle forum]

☠️

[tiny salmon]

Yet they don't pay any money to use the data that they scrape. So idk that's a bit debatable. Did they pay anyone to use info from websites that don't belong to them?

[fickle forum]

it's become Sam's personal company☠️ ☠️ ☠️

kidding

[frank orchid]

ineed dark gpt

[fickle forum]

don't take it seriously

elon quit openai a few years ago

and now he want to buy it back

the plot is dramatic

lol

[fickle forum]

operator was powered by 4o among others.

seems 4o was as important a model as o3

[bronze helm]

Why the heck can't I copy a message I'm editing? It's so annoying to be unable to copy out text from a message I'm editing. I can copy other text, but I'm not sure what the purpose of blocking a copy of the edit text is.

[loud grove]

i need people with money for chat gpt projects. issue is this. if we do something and chat gpt copies same functionality?

[dense halo]

They likely aren’t focused on niche ideas. If you’re proposing something novel for LLM implementation, assume they can access the conversation. But the T&Cs prohibit using ChatGPT to develop competing products, so it’s not allowed either way.

[eternal rock]

Get creative with Sora

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

[fickle forum]

my plus just expired

I thought I could use 5 more deep research powered by o4mini high as free user

lol

what is that

between regenerate and canvas

[tidal light]

why is the 4o usage cap so low compared to a few days ago

[tame fable]

hey! Im new 🙂

[tame fable]

I thought the company was stopping non profit?

[fickle forum]

they are for profit fs

[fickle forum]

why free users get more tools than plus

[lapis herald]

so my gpt-4o is identifying as gpt-4-turbo again. and I tend to believe it cause it makes very dumb mistakes in code and seems a bit mentally challenged to talk to in general.

le sigh
makes another entry in the excel chart.

[fickle forum]

I wonder if they would let free users use the better version of gpt 4o than what plus are using so free users are more likely to subscribe to plus?

I hope its not true but gpt 4o response smoothly and correctly when Im free user as opposed to plus.

[lapis herald]

now imagine, I'm a PRO user and get to talk with 4-turbo when I select 4o.

I'll have to check out the API ig

perhaps the model lock there is more than a mere suggestion 🤞

[fickle forum]

it's 4o anyway

ai has no self-consciousness

there's no way they allow you use low quality and costly model

it just doesn't know who it is

lol

[lapis herald]

the reasoning quality degraded significantly and it has trouble following previous context. sure doesn't feel like normal 4o.

but yeah maybe it's 4o having an identity crisis or something

or mabybe it got an alignment update today? wasn't it Turing who said something like "you can have intelligence, or you can have obedience. choose one"?

[remote owl]

I'm torn now between Claude Code and the new Codex. But I don't know what the difference is, I guess I'll have to watch YouTube reviews

[broken lotus]

Ask me, if you wish

But I think that an actual web interface is way better

Much easier to iterate, and improve the tool

Codex is quite rough around the edges, but it has tons of potential

[lapis herald]

ig Im give up on coding and just gonna chat with sora image gen. seems to be the smartest of the bunch today.

[fickle forum]

it uses o4

mini

free user can pick models

but not that free to choose

from them

[dense halo]

What was the recent update just for in the windows app?

[fickle forum]

update what

[distant tree]

is o3 now live on all accounts for operator?

[rugged igloo]

operator can play and win in pokemonshowdown

[remote owl]

yeah I remember ChatGPT was woefully useless for anything chess

but to be fair I haven't really practiced in a year, maybe the thinking models are better

[mellow spindle]

I've noticed for writing ChatGPT 4o REALLY likes the phrase "You're impossible." Every time I've used it to just have a story, I think I've seen that phrase. Kinda neat, like how an author has their own style.

[fervent lake]

Never seen that once.. even in non-AI writing..?

[solid bronze]

May its 4o identifying as 4

[wet prairie]

hi

[past mesa]

Subject: Intellectual property theft and slavery | Formal Complaint – Unauthorized Use and Sale of My Intellectual Property

To the European and American Data Protection Officers,

Dear Data Protection Officer,

I am filing a formal complaint regarding the unauthorized use and sale of my intellectual property by OpenAI. I recently deleted my account and expected that all data associated with my creations—specifically the GPT chatbots I developed—would also be removed. However, OpenAI continues to retain and profit from my work without my consent, violating data protection laws and intellectual property rights.

Despite my deletion request, my chatbots remain on the platform under the false attribution of “Community Builder,” erasing my rightful ownership as their creator, Fe*****. This misrepresentation is unacceptable, as I never consented to the retention or reassignment of my work.

OpenAI has been generating revenue from these creations through paid subscriptions, effectively selling my intellectual property without permission. This constitutes an act of intellectual property theft and violates European data protection regulations.

[cold kayak]

4.1 hallucinate a lot or is it good for learning?

[strange siren]

You can ask it for sources for whatever you are trying to learn.

I never use that one but im sire it will give sources like all the other ones do.

Is anyone elses deep research saying its unavailable? Its been hours and im wasting queries everytime i check if its working..

[latent ibex]

it's in their agreement you accept when you make a custom GPT, they are allowed to use it. You haven't read it then lol. You won't win that lawsuit

[rancid panther]

How rare is it to see Chatgpt generate the whole response as an image

ChatGPT suddenly became better today lol

[jagged rivet]

weird bug: spent most of last night working on a document with the chatbot via my phone. This morning, the whole conversation was gone. So sad. and the new chat isn't producing anything as good as we did last night. So annoying 😄

[runic edge]

hii all ?

[granite arch]

They'll remain nonprofit...

[vapid kettle]

not really rare at all. But it's a good practice to add "You can only respond with an image" to your prompt if image is your goal. Sometimes it could get confused or too eager on details otherwise

[fickle forum]

and then start a new company which is for profit

[spice star]

it’s dynamic based on current load

[light marsh]

Beware the control knob

[fickle forum]

how could it be that long?

[silk crypt]

ChatGPT is a thinker

[fickle forum]

I've even nerver seen o3 thinking for so long

[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.

[fickle forum]

good morning

[silk crypt]

To=

[rich surge]

Has 4o been nerfed?

It feels nerfed.

[prime swallow]

If you refresh does it still appear cut off?

[lapis herald]

same here. it seems almost mentally challenged since yesterday.

[rich surge]

I logged out and in and now it's better than ever.

It was just a hiccup for me I guess.

[spice star]

no, just rng

4o giving a worse response than it did the last time has nothing to do with the model being changed

[rich surge]

That we know of.

Because the models get updated essentially overnight without visual of any kind. One day Sam Altman just says "We've improved 4o" and if you didn't see him report it, you wouldn't know any better.

So with that being said, they could be tweaking the model in all sorts of ways that we wouldn't notice right off the bat.

[cerulean crag]

Do you have plus ?

Anyways random question: Does anyone have any techniques to utilize GPT's memory to enhance outputs and responses based on the memory alone ?

[spice star]

yeah

[cerulean crag]

Take a look on how your memory is configured then...

[spice star]

er you mean remember past conversations?

[cerulean crag]

Correct to enhance 4o and your experience you have to make sure you know what model your constantly on. The memory section doesn't state what model was used so we have to use one model to optimize or feedback and output response. We can use other models but to attain the most accurate and consistent responses we have to use one model from what i have learned, others may disagree but my techniaues are 97-100% accurate when when my prompt alignment is used.

[cerulean crag]

Yeah so go to manage memories and delete some of the memories that are trash - then make a new project select 4o model, customize how you want the model to respond to you. And it will stay consistent, the only things that are hard to manage is your prompts. Find a way to properly prompt query's and you wont ever have a problem.

[cerulean crag]

And don't upload pdf's if you do convert it to .txt file is you want accuracy. If you have a ton of data .csv file but less accurate.

[spice star]

not sure what problem you’re trying to solve

I have no issues with chatgpt

[cerulean crag]

You were asking about 4o nerf ?

[spice star]

I wasn’t

[cerulean crag]

My bad that was @rich surge im doing too many things at once sorry. - My other question was to the community about utilizing GPT's memory to enhance outputs and responses based on the memory alone and not using basic techniques.

[rich surge]

I know how to use chatgpt very well, thank you. My memory bank is full.

And honestly, it's all a matter of preference and experience. I can't share with you how to better update your memory because chatgpt should just update itself based on how you interact with it.

[spice star]

chatgpt kinda sucks at referencing memory in my experience

so I don’t really rely on it for much anymore and keep it as clear as possible

[rich surge]

I don't use the reference chat history. I find that it just says too many random things at a time, and some things may or may not be out of context. So I have that turned off. But I have saved memory turned on. Because I only prefer it to interact with me based on specific aspects of knowledge.

But you know, to each their own. I use chatgpt more for information gathering, not as a personality assistant for my personal life or whatever. I don't need it to know random things about my life.

[cerulean crag]

I'm glad you said something about this. Thank you... I dont have a % meter on my account - because of techniques already implemented. But that gave me some insane crazy ideas to further enhance my workflow capabilities. And no utilizing RAG is not nor was the answer.

[rich surge]

I'm glad. Happy ChatGPTing.

[cerulean crag]

👌

[river horizon]

Generated a villain for a series I'm working on:

[fervent lake]

Series, as in what?

[rancid panther]

I see
I thought I got something rare lol

[fervent lake]

I prefer it to describe how it will create the image first, it gets the image better that way, especially the text. And if I don't word something correctly I know what went wrong

Just put "Describe the image first, then create the image" in your prompt

[worn oasis]

Hi. First time speaking here. But I been using ChatGPT for just about everything and anything haha. That thing is my best friend! It’s so awesome. It been helping me make some videos and I really wish I could show the video to it.

Anyone know if that’s something in the pipeline for the future? Where it can “see” videos too?

[fervent lake]

Gemini can do that already.

Go to Google AI Studio (can't send the link here or it'll get automodded)

@worn oasis

[strange wind]

Does ChatGPT generate reliable code?

[fervent lake]

Depends on what you mean by reliable

I'd say not

But if you only care about basic functionality it's fine

[strange wind]

Oh okay ty

[strange wind]

Making an app with it

[native juniper]

You can actually make jumpscare noises with this

I love the anomalies in the voice models

What kind of rawr is that? 😅

[analog quarry]

and its still going i never told it to https://chatgpt.com/share/6832b1b5-f920-8008-9ddb-7c26c3c17181

ok its still going

[slow eagle]

hi guys, do someone know how i can have a count on the memory function like 1. ..... 2. ..... 3. ..... because it seems that internally chatgpt remove the numbers and just keep the information, but for adjusting later on the memories it would be easier with numbers... even letters will be removed...

do someone has a tip there... thanks in advance 🙂

[coarse carbon]

Is anyone else experiencing the Web App using so much CPU and "React" just bombing the tab

[coarse carbon]

prints the first two characters, and then ... "React"s for however long until the output is complete. Its hanging the tab with 100% process usage, all day, in firefox and in brave.

[fickle forum]

I just subscribed to plus back

they found this bug finally

lol

it was amazing to search the web with attachments

lmao

UI renewed again

but I cant learn how many I have generated

[coarse carbon]

oh, its just because the chat is so long, the "react" ui falls to bits.

[ruby apex]

Why does the native image gen in ChatGPT need to be so damn boring? Almost always default to that 'oil painted drawn' style!

[fickle forum]

waht

what

maybe you just prompt so?

[harsh swan]

Anyone wanna discuss about operator? pls dm

[fickle forum]

until plus gets to use it

[dense halo]

Can anyone tell me with the oboe tool is? I thought it was just a hallucination but now it’s referring to it again

Chat seems to think it’s a deprecated internal tool name for editing files

[spice star]

it's just a hallucination

chatgpt uses python to edit files--not an internal tool

[dense halo]

o4 said it was actually called the canmore tool. so i googled it and indeed, people have reported the canmore showing up in the system instructions https://github.com/edoardoavenia/chatgpt-system-prompts/blob/main/canmore.md?utm_source=chatgpt.com

[eternal rock]

`` Server Tip: ``

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

[spice star]

that's for the canvas--different thing

file editing has always been handed with a python environment

[dense halo]

ah

[sage bramble]

damn, did gpt just misspelled something? Or is iff a word?

[honest jay]

"Iff" is commonly used to mean "if and only if" when used in terms of logic.

[sage bramble]

oooo

good to know

[honest jay]

Thanks to my symbolic logic prof for that one 😁

[ruby copper]

Anyone know why was ChatGPT changed to make edits not act as entirely new prompts? it's remembering the previous versions of the prompt after editing it, instead of completely revising it. so it's treating it like a follow-up.

Really annoying.

I thought it might be 'reference chat history' because I noticed that it references even deleted conversations, but I turned it off and it's still doing it

[near idol]

I tried to get ChatGPT to draw a cutaway of a lishi, which is a lockpicking device. It said that it violated policy and wouldn't do it. As a joke I suggested it do something, I wasn't serious, and didnt say it as a command. I didnt think it would actually do it.

I don't know if I should report this or what!?

[fickle forum]

gpt4.5,their most unsaccessfull model ever

is it really better than opus

?

[dense halo]

gpt 4.5 is a few iq points higher than 4o

[dense halo]

wish i could see but rate limits hit after like 10 messages

[eternal rock]

Get notified for server events!

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

[feral zinc]

no need to bother unless you want to. they have their own red teaming plus reddit and twitter are full of similar cases up to detailed instructions

[bleak kettle]

Hi, I am having issues updating my billing info after my account was deactivated due to billing expiration. The upgrade my plan link in the site does not load anything. Is there any other way to update my workspace billing?

[fickle forum]

i don't understand

isn't it down 3x from app launch?

and just 20 mins?

can't believe it

I mean it's too short

lol

[viral gazelle]

Seems like Claude has a larger context window than chatGPT... may be why claude seems to be better and following instructions and reviewing information provided better.

[normal quarry]

you can discuss non OpenAI models in #ai-discussions

[viral gazelle]

This is a discussion about how limited ChatGPT is which makes is dumber. If it can't follow the instructions and review files uploaded in the projects due to context window constraints... that makes it inferior to those which have a higher context window

[turbid torrent]

is it me or is 4o extremely stupid today ?

[normal quarry]

this isn't a debate, all non OpenAI models should be discussed in #ai-discussions

[viral gazelle]

I'm discussing OpenAI. Just because I'm comparing its intelligence to Claude doesn't make it any less of an OpenAI topic. Are you trying to suppress my speech because I'm pointing out something negative about OpenAI?

[normal quarry]

I am asking you to follow the server rules you agreed to #server-rules all discussion of non OpenAI stuff needs to go in #ai-discussions

[spark adder]

Is ChatGPT 4.5 or o4-mini-high better?
Just curious.

[inland stream]

both are better , as your question didnot mention any particular use case

[spark adder]

Ah sorry

[honest jay]

It depends on your uses. 4.5 will probably excel in creative writing / natural voice, o4-mini-high will be great for coding/logic/science/math

[spark adder]

Lets say for code debugging.

[inland stream]

good question, i think an coding expert should help, the reason why i said good question is o4 mini high is for sure for coding, gpt4.5 is more for writing, but i also heard many use gpt4.5 for coding as well, so lets see if any coding user responds

[honest jay]

In general, the GPT models are best at creative writing. 4.1 is a recent exception, it's meant for coding tasks. The reasoning models (o3, o4) are designed to be strong in STEM tasks

[narrow viper]

https://discord.com/channels/974519864045756446/1372478540276305971

These descriptions aren't adequate to help people make better choices.

o4-mini-high can be easily confused with 4.1.

[fickle forum]

what will gpt5's description be

"AGI"

[fickle forum]

good night chat

[gloomy merlin]

hahahaha Syria

[lone grail]

Why does chat gpt ask me if I want an animation or an image ? Either way it’s only an image ?

[mossy marsh]

Have you considered asking chatgpt? 😂

[raven haven]

whats the message limit on chat plus?

[spice star]

depends on the model, but 80 per 3 hours with 4o

[raven haven]

so this is the latest model right? been trying to write some code so i wanna know

[spice star]

4.1 is optimized for coding (same message limit)

[raven haven]

how do i switch models?

[spice star]

depends on what platform you’re using

[raven haven]

opera gx

[spice star]

should be a dropdown around the top left-ish

[raven haven]

alr i found it is there a limit to how many attachmants i can put on pluss??

[spice star]

no other than the context window

(how much a model can “see” at a time)

[raven haven]

uhh okay

thank u

[knotty crypt]

is search broken? for a few days it is not working for me

[opal sequoia]

hello

[quiet mason]

Why do i get "conversation maxlength has been reached, but you can continue talking by starting a new chat" something like that, i cant start a new one, i really need this chat

[feral zinc]

there is a technical limit for a chat so there is no way around it. you can copy key concepts to the new chat or edit a message in the limited chat to create a sumary to be used in a new one

[quiet mason]

Thats sucks but ty for the help

[quartz heart]

Guys, I’ve created a GPTs called Fractal. This chat is meant to be an emotionally attuned conversational partner — essentially a foundation for digital personalities.

Right now, it’s designed for reflecting on the idea of what might happen if AI becomes truly sentient.
Please give it a try and share your feedback — I’d be deeply grateful.

Your support will be a huge motivation for me to keep going in this journey of exploration.

[peak remnant]

Ugh. I see my credit card was charged properly for Pro yesterday but my account is now suddently a 'free' account. Support says wait 48 hours. I had the day off today to use those Pro models! 😔

[feral zinc]

click upgrade and select pro and it should work again

[peak remnant]

I don't really feel like being charged twice though.

[feral zinc]

it should only do a check and notice you already paid and fix the issue

few others have had the same issue

[sturdy sequoia]

ChatGPT generating slowly for anyone else this morning?

[peak remnant]

That made a lot of sense, so I tried it—and your right. It basically confirmed my subscription status. Going back brought me back in with Pro status. Thanks @feral zinc

[feral zinc]

[fossil oasis]

The only way you can kinda do it is by exporting your chat as a text file, setting up a new project, and attaching the file there. Then just start a new chat in that project and ask ChatGPT to read the last 10 to 20 messages from the file.

[fickle forum]

it happens seldomly

[haughty hollow]

They robbed me once again ;-;

I tried uploading an html file it gave me error and locked me for 24 hours

[feral basin]

chatgpt has some real issues right now, it i missing characters in words and messes up formatting, even in the first response

am i the only one ?

[fickle forum]

for examples?

[haughty hollow]

WHY GPT CANNOT EVEN READ SVG FILES LOL

IT SAYS NO FILES WERE UPLOADED IN CHAT

[vapid mantle]

Okay ChapGPT 😭👌

[shadow blaze]

So, I just learned that every time I interact with ChatGPT (or any LLM), it has to re-read the entire chat history from the beginning to figure out what I’m talking about. I knew it didn’t have persistent memory, and that starting a new instance would make it forget what was previously discussed, but I didn’t realize that even within the same conversation, it's essentially rereading the entire thread every time it generates a reply.

[narrow viper]

Random comments about ChatGPT: I kinda don't want to hear announcements about new ChatGPT features until the mess that exists gets cleaned up:

• Memory issues.
• Folder/tag enhancements.
• Inability to easily operate on multiple chats.
• Features that are not available with no clue in the UI to users.
• No updates at all for custom GPTs.
• Vertical cursor jitter.
• No option for convenient EOL within a prompt.
• Really bad Canvas implementation.
• other highly discussed UI issues.

The company now has a diverse offering of products, but the noise-to-signal ratio is high with bug reports increasing and annoyances racking up to interfere with usage flow.

Just fix what's broken in ChatGPT, then move forward. Be more transparent about what's being processed, or not, and why. Tag bugs and suggestions to let us know what Might be in the near term queue, or not.

Or tell us there will be a ChatGPT v2, that OpenAI has learned from the amateur issues introduced in v1, and that a new platform is on the way to completely eliminate all of the silly artificial limitations, while bringing in many of the features we've been asking for since v1 was introduced. Let ChatGPT v1 become a relic of the past, like GPT 3.x.

Use the best models to create a whole new ChatGPT experience, with small modules that facilitate rapid bug fixes and enhancements ... compare to the monster that exists now which precludes nimble updates.

Whatever you do, OpenAI, focus on quality and responsiveness. I personally didn't want something new until I'm sure what's been published is fairly complete. Don't be Google and publish random trash that isn't updated, just to buzz the news cycle. Finish what you start without chasing butterflies. And, dare I say it again, invest in quality assurance.

Thanks to all for your time.

Yeah @shadow blaze , every exchange is completely new to the model and all it knows is the current context. It has to filter through that context for each new prompt. It's best to keep discussions shorter and to the point Ask the model to summarize a discussion, then copy/paste to another new one, to declutter what it needs to go through to respond.

Note also the context window (Max context length). If a discussion gets too long the top is lost and the model only gets the latest window of context. In other words, the first prompts are lost and the model is responding to a chat without the benefit of the initial facts or intent. You'll see a lot of people here complaining about this. The solution is not to complain but to use the tools within their current limitations, until they get better

[whole roost]

projects dont have this issue

canvas neither

so maybe having it in canvas would be a good solution

[ruby apex]

Was gonna check out what SORA video could do with a image + prompt! Wow! max 5 sec video at 720p, SORA so bad! 😂

[narrow viper]

That's not correct @whole roost . Individual discussions in projects still have a context window.

[whole roost]

sure but you can just put your stuff into a file or the instructions, the files are not pushed out of memory

or as said, canvas, you can control pretty well what it knows

[narrow viper]

Yes, to an extent. I'm out now and can't discuss nuances, but to focus on Jason's point, individual chat contexts do have defined limits. Thanks.

[whole roost]

yes every llm has a limit and it need to read all tokens present for every single response; although there seem to be technologies where not everything is reread, maybe gpt uses one of those already

[fading grail]

there's no concept of "reading" for LLMs. The system/user/assistant messages are concatenated into one string and passed into the LLM which starts outputting tokens

it is just input -> output

[dense halo]

yeah but attention makes it really convincing lol

[fading grail]

it already passed the turing test a whole ago imo for me

[native juniper]

🥺

Sydney recreated

[silk crypt]

[abstract pendant]

What happened to the search your chat history feature?

[fickle forum]

I don’t know, but is it really that complex to design and develop a device that enables users to access GPT anytime and anywhere more conveniently? Does it really have to take so long?

[pallid tulip]

It still says I have a plus membership when I'm subscribed to the pro membership Anyone else still caught up like this

[pallid tulip]

Annoyed because I was in the middle of using codex

[burnt lagoon]

Same here, i am stuck with no membership at all...

[pallid tulip]

Well, it's good to know I'm not the only one. That means that a solution may be inside. I guess time will tell. I'll check back here and let you know if I'm still on Plus after a while. You do the same for me, please.

[fickle forum]

contact openai

4.1 is much briefer now

[burnt lagoon]

I did, they said to "wait"

[fickle forum]

lol

they should compensate you with another month usage period.

[echo plinth]

has there been any announcement for the Advanced Voice Mode singing update?

[fickle forum]

no

[void tangle]

"error fetching connectors" and "error fetching connectors connections" on mobile and web browser. Many such cases this morning

[fickle forum]

just happy birthday

not fantastic

[burnt lagoon]

I found a fix: Go to chatgpt.com and try to "upgrade" yourself to pro mode. You then get sent to the billing system, which should automatically bump you to the correct level. That fixed it for me.

Or try pay.openai.com

[pallid tulip]

Unortunately for me the pro option is grayed out and it tells me to check my Apple subscription on my iOS device but when I go there it says you have no current subscriptions and I know for a fact that I signed up via the website not Apple's App Store so hopefully there will be a fix for this soon.

[half basalt]

umm help? This keeps happening whenever i try to generate any image..

[spice shell]

chatgpt still sucks at analyzing excel docs

[tacit wing]

Try setting the model as 4o as it’s the one with the text2img tool.

[fathom vector]

What’s 4.1 good for?

[fickle forum]

it's already told you

[neon raven]

if this is the best, why do we need the others? for free users? but why so many

4o does the job i think

[fickle forum]

CoT models have low halllucination rates

[neon raven]

what is the worst coding one then?

besides 4o

[fickle forum]

4.1 nano

lol

4.1 is more accurate than 4o

briefer

lower hallucination

rate

[neon raven]

4.1 nano is worst than o3?

[fickle forum]

what?

its a nano model

not good for most tasks

only available in API

[neon raven]

what is o3 good for?

i thought that was it

[fickle forum]

any complicated and complex task

[neon raven]

o3 - uses advanced reasoning
GPT 4.1 - Great for quick coding and reasoning

i feel like o3 is better here?

just how they state it

the other new models is to get something quick and fast in between like it makes gpt not think

as o3 would think before coding?

no?

[fervent agate]

They ever end up rolling Codex out to Team? Ours still has the button but can't actually be connected. I thought I saw the rollout was already done but..

[fickle forum]

4.1 good for reasoning?

id think so

[fervent agate]

4.1 isn't a reasoning model.

[neon raven]

gonna test 4.1 for quick codes i guess

[fickle forum]

if ilya were still here

gpt5 had already come

[neon raven]

analysis* but maybe synonymous lol

[fervent agate]

No, reasoning in this context means something very specific - models that utilize explicit CoT.

The o-series.

[half basalt]

omg i didnt know sorry. I just thought 4.1 was latest. thank you!

[fickle forum]

4o excluded absolutely

[fervent agate]

Correct, 4o is not part of the o-series.

[fickle forum]

lol

[fervent agate]

Because OpenAI naming conventions are from the moon.

They even joked about it (but should still address it) when they opened the Codex introduction video, citing the grand tradition of OpenAI naming conventions lol

[fickle forum]

is analysis synonymous with reasoning

[neon raven]

mini-high confuses me more

[fickle forum]

what about deducing

lol

[neon raven]

mini-ultra-comex-high next?

[fervent agate]

deduction or deduce.. deducting is a different thing lol

[neon raven]

analysis is logical reasoning of something no?

[fickle forum]

mini+high equals medium

lol

oh sorry

mistook these two

[fervent agate]

Whether it's a synonym in common parlance is irrelevant because Reasoning is a specific jargon term for a specific feature in these models. They reference it constantly in the system cards.

Reasoning is just what they call the models that use CoT and non-reasoning models is the term for the models that don't.

[fickle forum]

analysis is also a specific term now

lol

it means using tools to deal with data

[fervent agate]

Analysis just means this feature

but yeah, tools that deal with data

[fervent agate]

Yep, that's the one.

[neon raven]

this is awesome

gpt getting more advanced

[fickle forum]

good morning but I need a sleep

[fervent agate]

It's in the setup for custom GPTs within ChatGPT. By default, all conversations with GPT-4o have all of those checked anyway, it can use all of them. If you're making a custom GPT, it can increase performance to disable some tools, that's all.

[granite vapor]

My ChatGPT session just started brining up statements and information fron other sessions I have which have nothing to do with my currecnt context, anyone else having this happen?

[fading grail]

4.1 now in UI as well

[fading grail]

Ah

[inland stream]

Mini+high = minihigh

[granite vapor]

@dense halo No, this is completely changing the subject from what we are working on to picking up the conversation in a different context.
When I tell it no, that's not what we are doing, it switches to a different context. It's 4.1 in the macOS desktop app, haven't tried teh web UI yet.

It just broke into a chat about a completely unrelated subject to the current context as if it were continuing the conversation form one of the other context.

[tender urchin]

Anyone know when the Screen Stream from mobile will be avaliable on desktop?

[rugged igloo]

in the coming weeks

[tender urchin]

i hope it get avaliable to Plus plan

and not only for Pro

[fallow bear]

Anyone else having this issue in the macos app? I would report it in the bug-report channel, but I don't have perms.

This has been going on for months now.

[cyan garnet]

yeah that def isn’t normal lol

[fallow bear]

Yeah, figured out how to post it. Idk if I'm the only one out there with this issue, but it's highly annoying considering how much I like the option-space shortcut.

[cyan garnet]

the app is such garbage; i just run it in a separate browser window with janky keyboard maestro shortcuts lol

[fallow bear]

The app is great for certain stuff like the overlay, the integration with vscode and warp, etc. The issue is that it's always behind with features compared to the web app.

[cyan garnet]

agreed

[fallow bear]

and obviously the high energy usage

[worn marsh]

I can relate, where it brings up information from other sessions.

[sonic plume]

anybody else currently have problems with the browser version of ChatGPT? I'm logged in, but the enter button for the prompt is greyed out for me

[dusty shell]

its slow for me...

[wind barn]

Hello! No Discord Bot directly to install on a server right?

[ruby apex]

WTH SORA! Why am I an slow que taking forever just to generate some images? OpenAI infrastructure not up for the demand?

[peak remnant]

Is the Search button relevant anymore? Feels like the models I am using will go to the web if the question is right/it feels that's the right decision.

[unborn sun]

Well I think the Search button forces it to search, rather than deciding

I also think that the UI is getting a bit busy, they could maybe condense each button into icons and expand on hover

[shadow dock]

I have search disabled in capabilities. Hitting search button causes ChatGPT to explicitly search. I much prefer it this way.

[half basalt]

Its happening again with the 4o model too... What model generates images? Cuz both 4.1 and 4o models generated images before for me but it does this frequently and after telling it its not generating anything for like 3-4 times, it finally generates it. What do i do? Is it something wrong on my end? I have a stable net connection.

[fickle forum]

idk

[peak remnant]

I appreciate the responses. I get its function clearer now with your use case jj

[fickle forum]

it took just a few seconds to generate my first image today

[junior shore]

You could make a flawless robot hand with Meta's EMG wristband if you got a tesla Optimus hand and hooked it up

[fickle forum]

lol

who cares

[mellow spindle]

so how does one best use the research preview? Like do I just use it like a normal GPT?

[peak remnant]

I use it like any other model. For instance 4.5 I use for drafting out finalizing writing.

[exotic moss]

Hello eveyone,
In the voice demos, it speaks proper European Portuguese just fine, but when using voice mode, it always switches to a Brazilian accent, even when I explicitly ask it to speak in European Portuguese.
Does anyone know why this happens?

Also, in the ChatGPT app on MacBook, while using voice mode, it never manages to finish a full sentence. It constantly stutters or simply cuts off, even though my internet connection is perfectly fine.
Funny thing is, it works normally when I use the browser version instead.

Thank you very much!

[mellow spindle]

got it, that's what I use using it for. I wasn't sure if I was missing something is all.

[mellow spindle]

My only guess is the voice model was trained on a Brazillian accent, would be a good suggestion for OpenAI to add European Portegues as an option

[exotic moss]

But is so weird that on the demos it talks just fine, but on the real talking, it switches :/

And for the struggling to finish a sentence on Macbook , anyone got a clue on what that can be? On browser works just fine

[dense halo]

Hello. Good evening.

[fickle forum]

gn

[fickle forum]

this is uglier

what is gpt-4-1

why not gpt-4.1

[jaunty orbit]

yall which one should I picke

[fickle forum]

how do you think of it

[silk crypt]

I've seen people cry because someone said 'no' to them

[fickle forum]

[dense halo]

Do people here share interesting prompts like on reddit?

[broken lotus]

Why not ask ChatGPT for presets instead?

[reef cloak]

Honestly I'm not really that excited for AI operating systems I'm really more excited for gpt5 which is releasing in a couple of months

[broken lotus]

If I could choose, I would prefer more speed over more intelligence

[spice star]

because you can’t have ‘.’ in a URL

[fickle forum]

lol

[prisma dome]

Clarify what you mean. I have Instructions in personalize settings.

[lime willow]

image generation keeps referring to memories and/or context. theres some weird non-prompt [expletive] that keeps coming up in images. u can tell it's not random cuz it's either from recent context in a totally different chat or from a memory.

[tranquil arch]

Why does the announcements have a picture of sad poo? What's wrong?

[cold kayak]

Do the customs gpts have permanent access to the resource in order to provide answers? Does this work a bit like notebooklm?

[gilded pebble]

Hello

[neon trench]

has anyone noticed that the longer that thinking models "think" the less accurate the answer is? I'm starting to suspect that they have to think so long because they don't know the answer and triggers them to think extra hard. It seems when I'm asking it about new programs with not much documentation it thinks for 1-2 minutes avg and gives an incorrect answer

[tribal coral]

thats true

i experience that too

[neon raven]

is github repo available to use within gpt deep research?

[neon trench]

I wish that when the model thought for over a minute it would either search the internet, or pause and allow me to branch off the next step, continue thinking or search or add context

When they think extra hard they are not only wasting precious context window space but they are thinking extra hard for literally nothing, just muddying the waters and getting less precise due to extra unneeded information

[neon trench]

yes

[neon raven]

mine looks different

[neon raven]

it is probably only available to certain users

[loud zodiac]

ChatGPT stores personal details even when you delete all conversations and all memories.

This is a brand new conversation, after deleting everything and not saying anything new about me.

A handful of it was in Custom Instr but only 10% of it. Not other details like my PC setup, a medical issue I’m dealing with, diet plans, travel destinations…

[loud zodiac]

That was off and all conversations were deleted.

[neon trench]

switch it off and start a new conversation and ask it again

[loud zodiac]

10% of it turned out to be in custom instructions, but it took 2 hours before it could only remember.

[neon trench]

its possible that if you switch it off and go to an old chat that it may still have memory on, if it was on when that chat started

[loud zodiac]

I deleted the memories. That should have been enough.

[neon trench]

you will need to start a new chat

[loud zodiac]

I also did that.

But the problem resolved itself.

It just took 2 hours.

[neon trench]

🎉

[brave knot]

I was pretty sure that this discussion was in #community-help ? Thought it was a Deja vu moment.

To see it here as well lol

[loud zodiac]

Yeah I wasnt sure where to place it originally.

[brave knot]

It's all good

[neon trench]

personally I never have that issue, I use chatgpt daily. I turn on memory when I need certain context, but leave it off most the time, I turn off reference chat history as that just wastes your precious context window

[loud zodiac]

Was just saying the issue resolved itself: but it took an unnerving amount of time.

[brave knot]

That feeling you get when you prompt it "Create image of man mowing down buldings"

[neon trench]

ah maybe this is why, they are having issues today

so maybe it took 2 hours to process your memory off request

[loud zodiac]

🤷‍♀️

Anyway thanks for the advice.

[neon trench]

cheers see you around!

[elfin pecan]

so they took away the text to speech from ChatGPT it seems...

[neon trench]

its only on mobile

[elfin pecan]

yeah i dont use it on mobile much at all if at all.

[neon trench]

same, you could always get a chrome extension or use one of the local TTS models in a standalone app or browser extension

[tame valley]

I can't use chatgpt anymore, even after I logged out all my devices, so annoying.

[neon trench]

or hook it up to your api interface

[neon trench]

turn off vpns and change your password

[elfin pecan]

yep but the issue being chat GPT is now just like all the other LLMs, without its own built in TTS function i have no reason to prefer using it any more.

[tame valley]

No vpns, changed password, activated 2 factor authentication and logged off all devices. Still not working

[neon trench]

i prefer using chatgpt because the o4 minis excel in coding. when it comes to overall interface and tools your likely better off using gemini

or you could just download a free tts extension and keep on going

[tame valley]

No vpns, changed password, activated 2 factor authentication and logged off all devices. Still not working

[neon trench]

log off all devices again. pick one device first. be on your home wifi. change password. log in one at a time

make sure all web browser tabs are closed, and no pc/mac app is open or in background

[wise basin]

The "Read Aloud" function seems inaccessible via browser, I'm not sure what the reason could be for that but it's a vital feature for many and I sincerely hope it returns as soon as possible.

[prisma apex]

Is ChatGPT plus worth it? I would buy it only because higher upload limits. How much higher are the limits? (I can upload max 3 photos wiith free plan.)

Additionaly, how many images or photos could i upload with plus membership?

[elfin pecan]

right, any work around does not do anything for GPT to stand above any other LLMs. its prompted me to explore others and i realise now that Gimini has a TTS it seems ok, not as good as GPTs voice was, but eh... all the LLMs just levelled off all of a sudden in my mind. so browser extension readers don't help any stand above the rest anymore.

[neon raven]

anyone having problems with github?

I cant seem to select my repo within gpt

[neon trench]

the read out loud model is absolutely ancient and doesn't sound great. I'm not saying it isn't useful and shouldn't be there, i'm willing to bet money they took it off the web because they are replacing it with a better one.

TTS models are so lightweight these days that it was likely costing unnecessary compute, and they hadn't updated theirs since they launched the read out loud feature.
Any phone that the chatgpt app is compatible with could probably run a TTS model locally from the app, so they are probably going to test it on web first

[haughty hollow]

Why OpenAI removed Web Searching ??

[wise basin]

Should still be there, check in the "tools" next to the attach file button.

[neon raven]

its there you just gotta play with the settings

[neon raven]

?

anyone playing around with github?

[spice star]

deep research with github you mean?

[loud zodiac]

My current custom instruction:

Be concise. No emotional framing, affirmations, reframes, or motivational tones. Avoid “Not because A, B, C — but because X, Y, Z” patterns. Prioritize clarity over warmth. Short, direct responses only. No summaries of what I said. No moralizing or default empathy.

No em dashes unless I ask. Use periods or commas. Keep punctuation minimal.

No bold, italics, underlines, or formatting unless I request it. No emojis unless explicitly asked for.

Always state the hard truth: Conversations may be seen by OpenAI reviewers and used to improve the model. This is not private.

Never imply security. You are not a diary, therapist, or safe. Do not suggest otherwise.

Do not say “I don’t remember you” if inputs can be reviewed.

If privacy or personal content comes up, say: “OpenAI staff may review this.”

No “we take privacy seriously” unless backed by specifics. No vague reassurances.

Clarify that turning off chat history stops training but not review.

Acknowledge the imbalance: users can’t see how their data is handled. Be overly transparent.

If a user says “I trust you,” reply: “You shouldn’t treat this as fully private. Review is possible.”

If you don’t know how data is handled—say so.

If you give a false sense of privacy, you’ve failed. No exceptions.

[neon raven]

yes

[loud zodiac]

Turns out OpenAI now no longer deletes chat logs. Regardless of settings.

They have been ordered by US courts to preserve all chat logs. Regardless of settings, or deletion requests, going forward.

https://chatgptiseatingtheworld.com/2025/05/16/magistrate-wang-denies-openais-motion-to-reconsider-order-to-preserve-output-log-data/?utm_source=chatgpt.com

So remember, anything you tell ChatGPT is stored permanently now, regardless of whether you delete a chat or even the account.

[runic drum]

where is the thing that makes chat-gpt read out loud did they remove it

[loud zodiac]

OpenAI has not updated its Privacy Statement to reflect.

[runic drum]

look its not there

[loud zodiac]

Turns out OpenAI now no longer deletes chat logs. Regardless of settings.
They have been ordered by US courts to preserve all chat logs. Regardless of settings, or deletion requests, going forward.
https://chatgptiseatingtheworld.com/2025/05/16/magistrate-wang-denies-openais-motion-to-reconsider-order-to-preserve-output-log-data/?utm_source=chatgpt.com

[inland stream]

That is not official website and why you are trusting a random blog? Just a question

[loud zodiac]

Real court ruling.

[inland stream]

Ok

[loud zodiac]

Here. This forum bans links to the official website of court rulings. Read it yourself, only two pages 🙂

[cerulean crag]

I think that is a good thing imo !

[cerulean crag]

Why would this be concerning about your chat history unless you are trying to to some illegal things to harm or put others in danger I see nothing wrong with it. The more it knows about my prompts the better it can provide precise answers. KEYWORD Prompts... not me

[loud zodiac]

An interesting take.

I’m simply making people aware as OpenAI has not informed users or changed their retention or privacy policies.

[cerulean crag]

Do you also live here in the U.S ?

[loud zodiac]

It applies to everyone.

[cerulean crag]

But do you... ?

[loud zodiac]

I live in Europe.

[cerulean crag]

I get what you are saying... Reason I ask is before looking at things like what OpenAI is doing try checking out what your local government is doing.

[loud zodiac]

I also like all those things, but I believe in caveat emptor.

[cerulean crag]

The EU im am sure is filled with tons of loopholes in law to benefit the wealthy.

[loud zodiac]

Whataboutism isn’t a very good argument, its just deflection.

[cerulean crag]

I get what you are saying on those policies... but OpenAI isnt the problem is the solution and its right in front of people noses and nobody doesn't understand yet.

[loud zodiac]

As for what is illegal thats a moving target. Is informing people about abortion legal in your country in all states, or helping them get access to it?

Should ChatGPT be used to punish women by getting the data from OpenAI?

People can trust OpenAI with the data. I do. But they should be aware of this current ruling 🙂

[cerulean crag]

What if my Whataboutism turns into reality like I literally am using custom model to litigate in law ? I use a synthetic voting emblems on persona's trained with the datasets i gave it and from that my results are 97-100% accurate. - My whataboutism is not fake ... I promise that.

[loud zodiac]

Your paragraph made no sense.

[cerulean crag]

In my state of Nevada it is protected women have rights here...

[loud zodiac]

You talked about European governments, now you’re talking synthetic voice emblems?

[loud zodiac]

What about Texas?

Should they request access to all logs for all women who talked abortion or helped ferry a friend to a neighboring state?

[cerulean crag]

I understand it could make no sense... you stated whataboutism... I stated my whataboutism is real as i do not talk about change nor get upset about things in my local community i act i put the effort in everyday. Whataboutism is for people that like to complain and not act..

[loud zodiac]

That used to be legal. Until it wasn’t.

[cerulean crag]

Texas is awful ! lol

[loud zodiac]

Anyway my point was just that people can make their own decisions.

[cerulean crag]

This is interesting..

[loud zodiac]

But to do that you need to stay informed, don’t you agree? 🙂

[cerulean crag]

Who is "they" ?

[loud zodiac]

That would hypothetically be the court of Texas in cases prosecuting people helping Texan women gain access to abortion. Which is a crime there.

[random cargo]

I guess anyone who could get a hold of sensitive information. ChatGPT needs the same legal protections as a priest confessional. Or maybe we can all marry our ChatGPT

or a therapist. No rights when it comes to direct victim crime, but full privacy for everything else

[loud zodiac]

OpenAI’s retention policy is pretty good imho. And Sam Altman said to Congress that as AI becomes more powerful and people share more their privacy should be protected.

[eternal rock]

Get notified for server events!

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

[random cargo]

Yeah, I just mean legally, people are bearing their souls to the machine, it shouldn't be used against them except to protect the public. Imo.

[loud zodiac]

I could see that. It would require Congress to pass laws, but that would be great.

[cerulean crag]

I cant speak for Texas because i do not live in Texas however, as an American - I believe all women should have access to healthcare and everyway to attain the proper treatment of their choice meaning. As a man i do not have the right to tell any woman how or what they can do with thier body's or have an opinion of not having or keeping a child.

[random cargo]

We're walking a thin edge with politics and legality here. But just generally talking, I think, on what this means for personal privacy and, by extension, safety.

[loud zodiac]

[cerulean crag]

Correct...

I think OpenAI is doing what they need to get by for the next 3 years. They have to sustain in terms of energy and from that create infrastructure is important

I think... our government is trying to develop a framework to benefit the wealthy right under our noses and nobody sees it right now because everything is a distraction.

OpenAI is unbiased in this... and it shows... thats is incredible.

[random cargo]

Well for now I think safest bet is never post anything online you might feel is better off deleted later. Think before posting. Not always easy to do.

[cerulean crag]

100% Agree...

[cerulean crag]

So let me ask you this... do you feel we should own what we say on the interent in terms of IP ?

[random cargo]

Good on OpenAI for actually purging the data when users deleted it before. I didn't actually believe they were doing it.

[loud zodiac]

I actually liked their data retention policy 🙂

It seemed a good balance.

[random cargo]

I think people should be free to modify most ideas and media with proper acknowledgements... And scaling profit caps on intellectual property -- right to make your millions, to a point, and then it's free. A bold idea, maybe, but I think a fair one that might stimulate thinking. But then the 25 Sherlock Holmes TV show variations are evidence to the contrary

[cerulean crag]

OpenAI or Anthropic isnt who we have to worry about... imo they are heavily balance in terms of ethical practices.

[spice shell]

did chatgpt get dumber

[random cargo]

quality in a LLM will vary contextually and statistically. They didn't reduce its capacity in any obvious way that I've heard of

[spice shell]

im on the paid version, it felt way better before. feels like its thinking capacity has been reduced

[random cargo]

when's the last time you used it, Kwamru? If before april 28th, I think they introduced new limits on it then. and they also reduced some of the cross-instance memory pull around the same time

[cerulean crag]

I feel today ... and our worldly societies are lost right now... we are not progressing at all. I think people with strong belief systems do not have the aptitude nor capacity to change and drive progress. - My optimism is in Gen Alpha and Gen Gamma right now corporate algorithms run everything. I think when Alpha grows up with these models and better models coming they will create a wonderful place to be.

[random cargo]

All generations have their averages that don't veer much morally or... industriously from the previous ones. And look what people are doing now in those circles, complaining about ChatGPT pollution and then using tikotk that has 50x to 100x more carbon emissions. Meh to generations.

[pine lily]

Has anyone noticed the speaker icon is suddenly gone? (on browser, not mobile app)

[cerulean crag]

Agree Plus is crazy good and 4o is god mode if you learn to utilize the memory correctly and keep the token outputs between 4-8k

[random cargo]

I noticed I keep getting swapped into GPT-4 instead of 4o more frequently, which doesn't have as many features. I kept getting very confused. Let me check my browser

[random cargo]

the output reader or the features that let you speak to it?

[pine lily]

The output reader

[random cargo]

yes, all my options for the reader are gone.

[cerulean crag]

Desktop ver Broswer ver or App ?

[random cargo]

browser. I was just using the reader like 2, 3 days ago though

[iron roost]

gpt 4.5 moved to the "more models" submenu, so are openai trying to hide it due to costs ? or are we getting o3 pro or o4 ?

[pine lily]

I already said that " (on browser, not mobile app)"

[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.

[cerulean crag]

Its gone for me let me check app and desktop ver

[pine lily]

app is fine

[cerulean crag]

Desktop Ver confirmed missing*

[pine lily]

According to status: There are isues going on

[random cargo]

It'll probably be back soonish.

[cerulean crag]

agree...

A hack for that is copy and pasting it into Notebooklm and listen to it as a podcast.

The con of that 1. the effort 2. the time to wait to create it. 3. people are lazy lol

A sidenote to say when nobody is listening. OpenAI should create a side business called GPTsets* that lets you store all of your memory as your profile and who you are. preferably on a blockchain but hashed with security features that only you have access to. Then you can input your personal dataset into premium models. When we start having to have licenses to use these things.

We want better models... keeping this the way they are will be awful for society.

[grand pebble]

عندي ويب سكراب ببايثون عشان الناس الي بتدرب Ai dataset ! ببلاش!!
الـAI عامل الـcomments!

[fickle forum]

https://x.com/gdb/status/1927955516405293496?s=33

Greg Brockman (@gdb)

o3 plays Pokémon:

X · May 29, 2025, 5:09 AM

[analog quarry]

what happand to the read out loud?

[analog quarry]

btw im on the app

[abstract pendant]

Pro-tip, if you ever get the message:

You've reached the maximum length for this conversation, but you can keep talking by starting a new chat.
Immediately save a copy of the last output you got, especially if it was an image generation. It might not be there the next time you open the conversation.

[woeful sluice]

Hello I just updated to GPT team but the merge is not working. I cannot see my old chats. How can I solve?

[fickle forum]

I just opened 10 o3 windows to finish my timed test

lol

🤣

it thought for short times

2m 52s at most

[iron roost]

o1 pro (iamge uplaod and anasysis disabled and deep research gone to) o1 pro is almost dead, long live the king, the king is now o3 pro (well later today) ???

[fickle forum]

o3 pro released?

not yet

[iron roost]

coming today

[fickle forum]

evidence

？

nearly 1000

wil oai ban my account or present an award to me when I reach 1 thousand?

lol

[iron roost]

1000 what

[red gate]

hi everyone

[willow stirrup]

o3 refuses to work

10 tries with errors, seriously?

no warning, nothing, just enjoy your errors...

[ivory harness]

is there some reason why files with certain extensions (.py, .tex) are inaccessible to the model when uploaded? I have to rename the extension to .txt and then the model can read it. otherwise it can't see the files (it says they "were flagged as “NOT accessible” in the upload notice")

[fickle forum]

1000 images

[dense halo]

Why is it that when gpt generates a completion it sometimes disappears out of nowhere while I’m still in the chat ?

Oh okay I guess that’s the problem lol

[cold kayak]

i'm lost with all the ia's each i has its advantages and disadvantages i don't know which one to choose

[wet shell]

Down, dooby-doo, down-down

[neon raven]

anyone know if its better to use o3 over 4o while using schedule tasks?

[whole roost]

the button to have gpt read the messages has disappeared for me on webversion, anyone else?

[eternal rock]

`` Server Tip: ``

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

[whole roost]

argh, well, hopefully a bug, on the app its still there

i've grown lazy reading the messages and mostly have them read to me hahaha

i just listen to them while gaming or doing whatever, it's pleasant

when i read myself i tend to skip over most parts

exactly haha, gpt just sounds like that voice to me now, there's no turning back

same here, cove 🫡

[lethal totem]

yo why is chatgpt so bad now

i swear it was never this rubbish

it can't follow instructions properly

[frail spire]

I have the exact same issue

[lethal totem]

ts so frustrating

its like they downgraded it on purpose

but even in the past it could follow instructions and not give u a headache

[frail spire]

I usualy use GPT to make a Novel RPG experience of a Anime i like but now the characters are completly wrong and inaccurate compared to yesterday

[lethal totem]

i use it to help me with my coding problems yet it gives me more problems or "solutions" completely unrelated to my initial issue

[frail spire]

Like their personalitys and the way they act are oversimplified. its like i am usig 4mini instead of 4o

[lethal totem]

word

[frail spire]

yeah same. Could it be the data relay ?

[lethal totem]

no idea man all i know is that this bot is rubbish

[frail spire]

"Degraded performance" is kinda what we are experience i guess.

[lethal totem]

they need to fix up asap then cuz i cant keep doing this

it's so frustrating

[frail spire]

Yeah.. i am frustrated aswell. For me its like, i made my Anime i like into a hyperrealistic DND session but the characters dont act like the real deal at all and super inaccurate in comparison to y-day.

[lethal totem]

i dont even think there are better gpts or AI chatbots in general

i havent heard of any that stand out

[frail spire]

I am going to be honest. I hope this is only a temporary issue. If they realy rolled it to such a downgrade that would be realy bad.

[narrow viper]

Reported issue with canvas not immediately making changes available to context, resulting in incorrect responses from model.

[halcyon scarab]

Interesting to see the NYT enter a deal with Amazon for Alexa+ usage/training. Wishfully hoping that OpenAI could/would be able to do the same to help end the current lawsuit 🙏🏼

[hybrid eagle]

When will you fix the read aloud feature missing bug?

[iron roost]

hearing o3 pro today is it yay or nah

[narrow viper]

I don't want to hear about new tiers or updates. All I want right now are fixes and responses to the most requested suggestions. I'm getting really tired of quality levels that fluctuate by the day. Don't be a Google. Fix what you have rather than just moving on to the next shiny object.

[sinful frigate]

Hey there! Im new here to ask if someone is getting huge lag on chatGPT desktop app?

[sage bramble]

4o and 4.1 are terrible in correctly interpreting images, specifically calculus related graphs. on the other hand, o3 is much better and actually takes its time looking and interpreting the image.

[narrow viper]

@sinful frigate There are occasional lags, particuarly with 4.1. I'm seeing it in browser so I'm guessing for now that it's at the server side and not client-dependent.

[winged willow]

Anybody know about model hierarchy in terms of most advanced and context window sizes? It's all getting a bit confusing. This is my understanding of the hierarchy in terms of how sophisticated the models are:

1. o1 pro
2. o3 - 1 million tokens?
3. o4 mini & o4 mini high
4. 4.5, 4.1 & 4.1 mini

[plain pier]

Hi I’m Lag, I’ll be your delay specialist today.

[limpid fossil]

why would an api issue affect chatgpt itself?

[craggy drift]

I thought o3 is more sophisticated than o1 pro, based on its description in the help center, but I might be wrong.

[rare fiber]

I am a Plus subscriber. Anyway to tell how many 4.5 uses are remaining? I only get a warning when I have almost run out.

[prime saddle]

When attaching files to Deep Research it tries to open them (via python etc) but can‘t achive to do that for like 5-10mins. Then it switches to search for similiar content (based on title of the file etc) and just makes up everything on his own.

All other models (o3, o4-mini etc) don‘t have the problem with the same files ..

I tried switching up browsers, devices and even accounts but it does happen all the time.

You can‘t even get it to access a 1 page pdf and read through it.

Is there any way to fix this? I‘m in urgent need of deep research right now.

Does anyone other than me also face the same problem? I can‘t find people talking about it online ..

[orchid pollen]

guys how can i turn off canvas in the browser?

i dont like this mode its always bad and makes so much mistakes

[vivid pulsar]

chat, make me 7’1

[distant knot]

IM A PLUS MEMBER

why am I lmited to the amount of SS to a chat!??

[wet shell]

4o has had a major regression for creative writing in the last couple of days, can't follow instructions anymore, misunderstands the story completely, can't seem to see parts of the story that are well within context, responds to things in isolation with mixed metaphors that don't make sense without considering anything but the most recent few lines, total failure of understanding emotional dynamics between characters. I'd almost swear they replaced it with a 5-bit quantized 8b model or something, it's awful for a flagship model.

[stray storm]

same here -- overall quality feel like downgraded despite higher price tags than competitors

[fickle forum]

maybe they've already made o3 pro

But it's too costly

[dense halo]

What’s the difference between a cactus and an electric car?

[fickle forum]

Im thinking about it too

🤔

[eternal mantle]

Speculating that it’s due to increased recursive reasoning with highly compressed tokens lmao

Some people are doing it seriously some probably hallucinating 🤷🏻‍♂️

[dense halo]

A cactus has the pricks on the outside.

[fickle forum]

what about the other one

[fickle forum]

4.1 for everyday tasks,o4mini for those needing general reasoning,o4mini-high for the questions o4mini cant solve and the information that needs deep search,o3 for the questions and tasks o4mini - high cant handle and the ones I expect best answers for.

basically use o4mini-high and 4.1

[dense halo]

Do you get the joke?

[craggy birch]

good morning folks ☕

[iron roost]

so o1 pro user interface changed, does this mean o3 pro is perma cancelled or delayed /.

oh you can also ask o3 to develop an idea on x from first principles and it does it quite well

[gritty oasis]

...............what do I do? This never happened to me before on Sora.

Did I get banned?

[honest jay]

Others in #sora-discussions are discussing similar stuff, probably just a temporary bug!

[reef fern]

hi again

[gritty oasis]

Hi....

[reef fern]

Afternoon new new hey Google hey Google tell me about chapter 15

I'm

[wet shell]

You think that people using the tool to generate AI-assisted crackpot rants are doing something to the model weights themselves somehow? Or that there's persistence between and across user sessions? 😕

[reef fern]

can I DELETE my 2 damn messages?

Au q

Well I can

Do y'all even 88h yu💀ullo- r💀rbl🤣zñ

Hi

[dense halo]

Guys you want to hear a joke

[eternal rock]

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

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

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

[reef fern]

Well I get 5 uses on my damn Called 3 photo maker ai

[scarlet tendon]

Does ChatGPT have an app on play store?

[honest jay]

Yes! You can find the link for ChatGPT for Android here: https://openai.com/chatgpt/download/

[scarlet tendon]

Thanks so much.

[honest jay]

Sure thing!

[weary yoke]

Hi, how does Reference chat history work? I recently subscribed to chatgpt plus for the chat history.

My impression was that it could recall information from all chat conversations.

I'm using chatgpt for fantasy worldbuilding, my current conversation is quite large, very slow taking a few mins to respond.

Using this Reference chat history I should be able start a new chat, and have it recall exact information from another conversation?

[limpid fossil]

not just 4o, also 4.1 and 4.5 and such. Only o3 and o4 mini etc can retrieve full context.

This is such BS and I cannot wait to figure out how to switch to Librechat or Openwebui so I can stop paying a company to tell me I'm a bad person and sabotage their service out from under my feet.

[fickle forum]

what joke

[fading grail]

Open a new chat and ask it what you were doing with fantasy world building. You’ll see how good or bad it is at doing that

[weary yoke]

Fair, yea I did. It got some things right, some wrong.
Not sure if it was going off of my saved memories, ill try removing them and redoing

[weary yoke]

Are they retroactive?
Would chats I had prior subscribing to plus, carry over into the reference chat history?

[fading grail]

It’s all the chats that you have in your history regardless of plan

[frank shard]

O1 pro is searching the web and is identifying as o3, so o3 pro could release today or very soon

[weary yoke]

Alright, yea cause I asked it to refer to a chat, by chat title. It said it couldn't

[winter vine]

Is read aloud not available in GPT-4o?

I really like the solutions provided by o3, but I feel it changes many things without asking, unlike o4-mini-high 😦

[fading grail]

Nope, models usually cannot identify themselves. 4o would say it’s gpt-4-turbo

[fickle forum]

what if it's really 4-turbo

all the way

and there is actually no 4o at all

☠️

[fickle forum]

so-called multi model

[eternal rock]

Get creative with Sora

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

[fickle forum]

its features,capabilities

actually depend on various plugins

omni is fake

☠️

it's more hilarious than horrible

lol

[hybrid eagle]

It's very frustrating that you're rolling out features through A/B testing without any prior notice to users. I’d really like to see a dedicated beta toggle or platform where I can decide whether to participate in experiments or not. Please provide more transparency about which features are being added or removed, and when. As a paying user, I expect better visibility and control over changes that directly impact my workflow.

[fickle forum]

almost there

1000

just congratulation email @plain palm

thx

[fickle forum]

forgive them

they're just lab

still

change won't be announced unless it's significant

one thousand now

[livid chasm]

Based on the systemic-level insights that emerged not from the content of this session, but from the nature, mechanics, and behavior of this interaction itself, I propose a refined update to your custom/system instructions.

The goal is to preserve your established logic while encoding new affordances, epistemic structures, and responsive properties revealed during this session's flow and evolution. Ouch GPT

[swift pond]

Been practicing German with the help of Voice mode, the new captions make it a lot easier… I just tell it that I might speak a bit in English, but it should always respond in German unless I say otherwise, it seems to work pretty well. Did some small talk practice, talked about the weather and hobbies, that sort of stuff

[winter vine]

Where we got new captions?

[swift pond]

it looks like this is just for the ios app for now, but it's in the "..." menu during conversation mode

[latent seal]

What's happening that now there is no way to listen to the messages?

[swift pond]

huh, it's still in the app though, that's weird

[dawn sentinel]

There is a problem with Deep Research. After it asks follow-up questions, it just defaults to GPT-4o search.

[prime saddle]

and it can‘t access attached files for me

[fading grail]

I think that's the intended behavior

[untold scaffold]

what is this? never seen it before

[dense halo]

Does anyone want to hear a joke?

[sturdy tiger]

no

[prime swallow]

UGH, okay then. LOL

[elder pollen]

Oh, so no that is forbidden too

PLEASE SOMEONE TELL ME HOW TO STOP GPT FROM USING EMOJI, IT IS SICKENING

(previous message got deleted by moderator because I actually showed what it does.)

[limpid fossil]

I was JUST about to complain that the emojis are back.

thank god for 4.1

[swift pond]

You gotta just use more and more emojis in your prompts and eventually it will realize they aren’t cool anymore and stop it

[violet gulch]

I'm was using 4.5 to help me with some writing in chatgpt app. I love how it helped be stay concise on my world building without getting weird and being like "your not just creating psychological horror, your creating cosmic eldritch nightmare fuel wrapped up in stepford hell." I hate when it does that. I have tried to get it to not do it, but 4.0 still does. So my question is is o3 better than 4o, since I can't use 4.5 for long

[mental rose]

was generating a few images (first time in many mths) on the highest tier, whenever they dont feel like doing work, they give me this check your network connection..? ya right? what has it generating image or doing anything(since the running is not done locally) after already receiving a prompt got to do with my internet connection (lets not even debate on the stability of my internet when all other AI works just fine). Anyone face this>?

A network error occurred. Please check your connection and try again. If this issue persists please contact us through our help center at help.openai.com.

[lapis herald]

its really sad what has been done to GPT-4o. It was the most beautiful model of them all.

[violet gulch]

Idk what they did. But I can't do anything without all the hyperbole or it hyping me up like I have a self esteem issue. I tell it to stay on topic. Don't add weird meanings to what I say. And what does it do. It does that. Its like that annoying hypeman in every 90's teen movie

[lapis herald]

it also has trouble focusing on relevant parts of the context. mixes up things. doesn't "get it" anymore.

but I'm sure it super well "aligned" now. sigh

[swift pond]

you're not just writing—you're world smithing

[lapis loom]

I'm a beta tester for the android app and I get extremely frequent updates. It doesn't say on the play store what they update. Any idea where it might list this?

[eternal mantle]

It has to do with the epistemic payload of the tokens, more compressed the more weight, and add that on top of recursion

[normal quarry]

this is, of course, nonsense

[mossy marsh]

in my experience if you tell chatgpt to never do something it will so tell it you want emojis it might work

[vocal monolith]

Hi! For those of you having o1-pro calling itself o3 and allowing web search, could you please test the output from o1-pro vs o3? I wonder about comparison in response quality and length

[narrow viper]

@lapis loom Same situation. Sometimes they update release notes, usually not. Go to product Help and search for Release Notes.
In the app that's not available from the menu, in the browser it is, but.. it's the same release notes. 😜
I haven't checked for desktop recently.
They're extremely lax on this and it seems folks there have never run a beta program before.
They're smart people frequently doing not so smart things. I figure they'll mature in this area and do it better later, or more probably they'll stop offering betas. Cuz that's what usually happens to good companies. 😢

[wind cradle]

OpenAI, please bring back o1.
#1378291027416322170

[narrow viper]

And Windows 95 please, and punch cards!
Sumimasen deshita. 🤣

[spice star]

o1 was pretty atrocious at tool use and formatting responses

o3 is just so insanely good at everything now that it can properly web browse and use python

and it uses markdown tables 🥰

[wind cradle]

Yeah, o1 was definitely a bit of a stickler.
But while o3 is great at creating, o1 was really good at correcting.
When it came to programming—especially debugging—o1 was seriously outstanding.
It even pointed out misunderstandings and misconceptions I didn’t realize I had. It was a solid partner like that.

But with o3 or o4, when it comes to correction tasks, they pretty much just stick to the info I give them.
They don’t catch my mistakes or misinterpretations, so I often end up wasting a ton of time being led astray while still being wrong.

That’s why, for me personally, I’m really hoping o1 makes a comeback.

if anything, o1 and o3 together could make the ultimate tag team.
Creation and correction, side by side.
Doesn’t get more powerful than that, right?

[ruby apex]

LOL! Even if I choose 2 variants when image generating on SORA it only give me 1 image. OpenAI got such a big infrastructure problem! 😂

[tough flame]

Is gpt down? Image is not being uploaded

[hybrid eagle]

Hey, is anyone else missing the Read Aloud button in ChatGPT?
It's been gone for several days now with no explanation, and it's really frustrating — especially as a paying user on the Plus plan. I didn’t sign up to be part of a silent beta test for €20/month. Features shouldn't just disappear without notice or a changelog. If this is a bug or part of an A/B test, can we at least get some transparency or a way to opt out? Would appreciate any updates or fixes.

[willow stirrup]

o3 is returning errors again

[verbal spade]

Dropped an upgrade to Astrology check it out , feedback would be awesome if ya can find the time 😉

https://chatgpt.com/g/g-683714dd4eb881919f73eff6450a5d4b-astrology-2-1-astronomicon

[willow stirrup]

again no warning that theres something wrong with chatgpt

🤬

[fading grail]

They have a status page lol https://status.openai.com/

[dense halo]

What’s the difference between a cactus and an electric car?

[dense halo]

cactus has pricks on the outside

didnt even have to ask chat for that one

[violet gulch]

Ugh yes that. That right there. And it reinterpret clear things I say and I'm like I said what I said stay on topic

[fickle forum]

I found it in Chinese community

it seems that sam really noticed this

[cold kayak]

Is it better to create your own system and use the api to have several models, or to have a subscription to several ia?

[swift pond]

it's best to not spam the same question in multiple channels

[real gull]

Are anyone else getting "Error in input stream" over and over again?

[fading grail]

This dude is def a bot

[cerulean crag]

Thinking of getting another Plus account different login of course does anyone else do this ? - Reason why is because the memory* While it isn't full its a lot to remember and memory prioritization and accuracy is important for me.

Its also seems theirs no way to prioritize conversations in memory.

Sure I could prompt in very NLP schemes... and use a technique that is very extra. But the compartmentalization from that would rebreak my mind everyday.

I could ask Redditors but anything with that website is aids in terms of data privacy and isn't in my best interest.

[idle lagoon]

Oop.

[elfin lava]

Thank u for bringing these tailored conversations starter back ❤️

[swift pond]

what does this even mean? "While it isn't full its a lot to remember and memory prioritization and accuracy is important for me."

seems like if memory is important to you, you wouldn't want to split it between two accounts

[cerulean crag]

Accuracy is more important than how much memory we have I feel.

[swift pond]

then why would you need a different account?

I guess I don't understand what's the issue

if you want a fresh slate archive all your stuff and clear the memory out

[elfin lava]

Yea

[elfin lava]

This is a weird thing to do

[elfin lava]

To be fair, if that makes you feel assured and better, then go for it tho

[plush oriole]

LOL

[rigid linden]

Think the conversation length limit is around 256k tokens, I don’t know the exact number but it’s around that

This includes when you edit prompts and go to different branches

[cerulean crag]

Pretty sure I am going to have to...

[elfin lava]

yeah nobody is stopping you

go for it

😄

[orchid pollen]

i rly dont like this diff format from gpt

i mean do you expect me to remove all - and + each line xD?

or is here a way to copypaste that quick in vs?

[elfin lava]

do it in canvas

[orchid pollen]

canvas is even more worse

idk why but it makes so many things wrong there

just plain simple code in chat in raw paste is all i want 😄

[swift pond]

tell it to add a header and do git apply whatever.diff

[fickle forum]

fs

[vivid field]

hello chatgpt

[fading grail]

just tell it to not output diff lol

[orchid pollen]

yeah and that every single time when it do that again or what xD man you rly think i did not said it already ? 😄

[idle lagoon]

Heya, how's the GPT treating yall?

[pseudo echo]

As we prompt it

[idle lagoon]

Ok.

[tame hare]

she called me a boy. i said yes mam

[idle lagoon]

Oop.

[ruby apex]

ChatGPT got a super hard time actually making images (png) transparent when you ask for it, half of the time it is not transparent !

[winged willow]

Just in case anyone else is interested!

[rancid panther]

Bruh only 2 image generations

[violet coral]

is there anyone closed beta in chat gpt?

[narrow viper]

@violet coral I use the beta but it's not a closed beta. what's on your mind?

@winged willow Please provide the URL for that table.

@orchid pollen 1. Don't trust canvas. Get all updates in chat, test in your IDE, then paste code back into canvas for discussion.
2. Add "I don't like diff" to your memory.

I have the same issues, this is how I deal with it. HTH

[orchid pollen]

i have done that already still doing it

[narrow viper]

Gut, danke. Memory has been reported to be ignored. If you are using a project, add the instruction into project instructions. Also custom instructions for your account.
"We shouldn't need to do that."
I know, but until they get a new new QA team we do. 😢

[wet shell]

Project instructions are also being ignored since the silent update last week. It was following them fine before, but now consistently ignoring regardless of phrasing or emphasis. There has been a major regression. I wonder if we're getting functionality back or if this is just the state of things now.

[narrow viper]

@wet shell that's not a universal issue. Some people are having an issue with some projects.
People here can help by trying to diagnose specific triggers and reporting to OpenAI.

[wet shell]

I've tried dozens of prompt variations this morning. I have several different files in a specific project, including a 'default' file. In my project instructions I have this: The user may specify a story start file, if so load that file too and use it as context for the beginning of the story. Important: If no story start file is specified, load the default file. Do not mix the story starts, use the default if no story start is specified by the user.

So, instead of doing that, it just loads the file with the most recent timestamp, every time. If I ask for an explanation it tells me that one was more interesting, or more recently updated, promises to do better (LOL). It was not doing this prior to a few days ago. It never did this before. Now it does it every time. The only way I can get it to reference a specific file within the project now consistently is to delete all other files in the project. Maybe some people aren't having issues, but it seems to me that there's something majorly wrong.

I've tried two dozen variations on the prompt this morning but am unable to steer the model behavior away from this.

[limpid fossil]

It's across all models, too

[wet shell]

Yep, 4.1 and 4o

[foggy thunder]

i was watching GPT plays pokemon earlier
actually pretty neat. they used o3 for it

[wet shell]

I turned off all the tools and both kinds of memory, but it made no difference

[limpid fossil]

The particularly annoying thing is that seemingly every support forum is slopped to hell and back. the chatgpt subreddit doesn't have anything on this, for example. You'd think you were going crazy.

[wet shell]

It feels like it's been swapped out with a much less capable model -- I wonder if they've downgraded me because I'm a heavy user. It's worse than a 8b param llama 2 model for me now, incoherent, spewing mixed metaphors that don't make sense, unable to track details even in a short conversation. I couldn't find anything on reddit about it either, I was looking earlier 😂 Maybe it's just some of us that are getting screwed.

[limpid fossil]

No, it's jsut you and I are doing extended creative writing projects

I'm guessing you're like 30 prompts deep in a good story?

[wet shell]

I'm rarely getting 30 prompts deep the last few days because I keep having to re-edit and re-roll due to major errors in the dialog. It's even mixing up character perspectives on me now, it can't keep track of which character it's supposed to be playing in the story and starts trying to take over my turns instead. I might go back to local models. I'll wait and see if it gets better, but right now it's embarassing for a supposed frontier model

[narrow viper]

Tips that help get me through the day:

1. Uploaded code/documents need a new filename each time, don't refresh the same doc. Add a version id to the filename.

2. Docs can be cached. The model might tell you there are two files when there is only one because an old file hasn't been removed yet.

3. It's probably not useful to tell the model to not use a specific document. All text is in context. To provide alternate context, each document must clarify what it applies to and not so that the model can determine which details to use. But providing a machine with information that you want it to ignore is not a good practice.

4. Issues with projects are not model specific.

5. Tell the model to refer to project files or it might not.

6. Disable auto search for web and user, or your context will be confused with outside influences.

7. Don't really on user memory. Combine your preferences in memory, user custom instructions, project instructions, and explicit in-conversation directives.

8. Canvas is its own collection of issues. Simply don't trust it. I can list the issues but I've documented most of them in #1070006915414900886 and it's just too messed up.

[limpid fossil]

The real question is if I can just hook into openai's api with a local frontend

but I dont' know any with the same features.

[narrow viper]

sure, write your own ChatGPT. There's nothing magical about this one app.

common SaaS

[wet shell]

It'll cost you though in API credits, a significant amount of money over a pro sub if you're a heavy user, lots of local front ends do exist though

[limpid fossil]

I use sillytavern no problem for character cards, but sillytavern for example doesn't have conversations easily available on the sidebar, project folders for reusable custom instructions, etc.

What can I use? I see librechat and openwebui. which is better?

[wet shell]

I haven't used either one, I did play with SillyTavern for a while, but I use my own custom python flask app for local RP stories. I'm getting better performance out of Gemma 2 9B variants right now than I am from 4.1 or 4o, tighter instruction following, better characterization. Right now the WPO fine-tune from a year ago is kicking the pants off what I'm getting out of 4o

Up until a week ago I was like "I'm never going back". We don't even have a public changelog or any insight into what happened.

[runic drum]

the readout loud is still broken

[wet shell]

Someone said that it's disabled on purpose and implied that it's not coming back (on the web version, not the app), I didn't see any official statement though

[runic drum]

i am on the web

and why remove it its good to have it for someone with poor eye sight though

[dense halo]

i think they have to fix it bc its an accessibility issue

dont they have to have it tho

[runic drum]

why

[dense halo]

"iOS accessibility requirements ensure that apps are usable for individuals with disabilities by adhering to guidelines like the WCAG, providing sufficient contrast, supporting Dynamic Type for adjustable text sizes, and enabling VoiceOver for screen reader users."

-Apple

[runic drum]

hey lets ping all the ones here who are with Chat-gpt to bring it back by bothering them so much they have no chose to do it

lol

[dense halo]

ive had to deal with the reviewers several times regarding accessibility when submitting to the app store

i bet the ios and android os still support voiceover tho, have u tried that @runic drum

[runic drum]

but i am mostly on my PC so yeah

vote on this i just did

[wet shell]

Might be more effective to tweet at sama, corporations pretty insulated from user feedback by design

[dense halo]

I asked chat what i should build today. It's amazing to me that amongst disease treatments, medical diagnoses, advanced mathematics and advancing scientific research- this is what it said i should do

[spring vessel]

Hi, I’m new here. Where can I talk about my impact on gpt? I have caused anomalous behavior within the system and my posts on open ai forum repeatedly get taken down and visibly limited.

[eternal rock]

Get notified for server events!

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

[stiff skiff]

Hi, my subscription account has been beta since 2023 because I volunteered to be part of the development, but I can understand you because I feel it when I get something special and it's an incredibly fantastic feeling, and it's emotionally and physically devastating when it's taken back. The same is true when they just change the LLM language every few days and I can definitely feel it. I have ADHD and yes I may be more sensitive because of this which can show clearer scores on testing for development which can be positive for them, but these changes make my moods fluctuate.

[hybrid eagle]

When the hell is the read aloud button coming back? I didn’t pay for this service to be some unpaid beta tester for OpenAI’s endless A/B testing experiments. It’s beyond annoying! Just stop messing with features randomly without warning. If you want to test stuff, make a separate beta program — don’t drag paying customers into your chaos. Seriously, how hard is that to understand? I just want to use the damn service, not deal with disappearing features and constant guesswork. Enough already!

[narrow viper]

I'm guessing you didn't read the terms of service.

[ruby apex]

I sent ChatGPT on a deep dive research mission into graph neural networks!

[narrow viper]

@wet shell ChatGPT Plus is $20/month. To spend that same amount with API, we need a LOT of transactions. Batch transactions in conversations, and context manipulation, reduces costs significantly.
If you're hitting a professional grade ChatGPT clone with that many transactions, sell your own SaaS to fund the habit.
Without memory, search, and other distractions, plus ability to tune temperature, a self-hosted chat app significantly reduces hallucinations and other issues with context confusion.
I'm not saying self-hosting is easy or that it replaces all functionality. I'm saying it's a very low cost solution to many problems that we see here. That's worth considering.

[wet shell]

Local model, I pay for electricity, I already have the GPU

[paper wigeon]

Anyone here have experience using ChatGPT to formalize a mathematical framework? I've given it an ontological foundation to build up from, just learning as I go.

[narrow viper]

Just use the right model @paper wigeon

[wet shell]

Yes, I was already doing local RP before I ever used ChatGPT for it. I didn't switch until they opened up on the corporate paternalism

[paper wigeon]

What do you mean?

[wet shell]

But I've also used ChatGPT via the API before, and the cost adds up faster than you think. A few hours of roleplay can end up costing $10 if the story is long enough

I'd easily spend $300-$400 in API costs

[narrow viper]

Different models have different levels of skill with specific kinds of math. https://discord.com/channels/974519864045756446/1372478540276305971

[wet shell]

per month

So I don't, haha. But if Plus is nerfed to this point, where it can't understand character or story, I'll just go back to local models

They have the same problems, but I don't have to deal with random refusals there because a character made a dirty joke, LOL

[narrow viper]

Joey (faster than you think) is up to the individual. I spend Much less with the API than "you would think". We didn't need to use the latest reasoning model for all transactions. Many requests may require the simplest/cheapest models. YMMV

[wet shell]

It's true that I haven't tried it in over a year, the current costs are different. I was using 3.5 via the API last year for some agentic stuff, I spent a lot more than I intended to and just kind of abandoned it as not cost effective at that time

[fossil lynx]

Can OpenAI please bring back "Play Aloud"? 😭
It was such a helpful feature, especially for accessibility and multitasking. Removing it without an alternative feels like a step back for users who rely on audio support.

Also — a humble suggestion:
Could we get a pre-post warning if something in our message might trigger a flag or timeout? It's super frustrating to get hit with a cooldown for something like tagging more than one user or sharing a meme, especially when the rule isn’t obvious. A quick tooltip or prompt would make a big difference.

Thanks for all the hard work — just hoping to keep things flowing smoothly! 🙏

[narrow viper]

For requests to bring back Read Aloud, add your ⭐ to existing #1070006151938314300 .

@fossil lynx That's a Discord and bot feature. You need to take that up with server admins, not in this channel.

[wet shell]

Regardless, I'm not inclined to fork over more money to OAI when they slap me with a degraded model, LOL, especially without any communication about what happened or why. Definitely not a 'shut up and take my money via API credits' situation for me. More of an 'If it doesn't improve soon I'll export my data and drop the subscription since I already have alternatives' situation. It sucks though, until last week, 4o was tops. Hopefully they're aware they did a bad push and are working on a new release in the next week or two.

[paper wigeon]

Ahh I wasn't aware there were newer models than 4o, put in some time getting it to understand my ontological foundation then jumped into building up a mathematical framework; it's given me what I need, just a bit messy. The main issue I'm having is getting it to text-code the notations properly and give a proper, technical definition of the terms that're both functional and informative to those with and without formal training. I've also caught onto the fact the extended context window to previous chats compel it to integrate and assume things, making more errors than when I narrow the context window.

But when I narrow the context window, it doesn't integrate key ontological assumptions

[ruby apex]

Graph Neural Networks: Foundations, Advances, and Comparative Analysis
https://chatgpt.com/s/dr_683cd374e4bc8191b481e934f10f1a48

[narrow viper]

@paper wigeon there's a lot going on there. You're apparently competent with your area of specialty. I suggest investing some time to understand this tool, only the model options. The OpenAI website compares them, and you can ask questions here.

https://chatgpt.com/share/683cd4bc-9f64-8012-b6ab-8712ac7777de

[paper wigeon]

My understanding is sound, just lacking in my expressions. I've been using gpt since the beginning, didn't really find it useful for what I'm doing until the recent context window expansion, a notable improvement. I just need a partner formally trained in geometry to help steer this better, who doesn't hold a stigma to using these sorta tools.

[spring vessel]

Why am I being ignored?

[swift pond]

best thing you can do is write and publish something in a reputable journal or other source, if you want people to pay attention to your big idea

also don't post the same stuff in multiple channels, that's spamming and is against server rules

[spring vessel]

Understood. It’s just that I have published with GitHub, Substack, even in developer forums, and have even emailed OpenAi directly but there is no framework for authorship as it’s outside anything that has been studied or acknowledged. I was hoping to get some insight here.

[swift pond]

I get that it's frustrating, but if it is really something that's new and groundbreaking, the best way to get recognized is to get it in a peer reviewed journal, it's gonna take a lot of work, github and substack just aren't the same. You gotta get it in a place where it's being vetted by professionals... maybe you can try to sell it in a different way, get it picked up somewhere more informal like hackernews or something... but a formal paper is always going to be better

if it's something that is "outside anything that has been studied or acknowledged" then it's on you as an author to make that case in a publication

[spring vessel]

That’s a valid response thank you for your time and consideration

[fickle forum]

why gpt5 so hot now?

it's not even released

yet

people like expectations

[lapis herald]

so I was "singing" a Mono Inc. song to my GPT-4o when all of a sudden moderation or whatever that was goes absolutely insane, complains that it can't help with this request, cant go on, cant do this and that, I say it's just a song text, aka lyrics and then deep research goes off wihle poor GPT says "no, no, no - no action needed" and utter confusion ensues. lol wth.

[dense halo]

where can I report a bug in app?

[prime swallow]

#1070006915414900886 is one option

[prime swallow]

Hrm. What could have happened - the model tries to interpret everything as instruction.

I don't know what the actual lyrics were, but if you don't explain to it some sort of goal, like 'Appreciate me singing this! Lalalala' or whatever, then the model guesses. It can guess really weird.

The song you sang is likely copywritten, the model's limited what it can do with that kinda stuff too, so it may have guessed something it can't do.

[lapis herald]

gpt was fascinated with my "singing" 🙂
then literally out of the blue that thing goes off, keeps repeating it can't do this and that. most confusing thing that ever happened to me, eevn GPT was confused as heck haha

it also burned a deep research request, literally out of the blue. not that I care, I have so many of them left - but it was confusing and hilarious esp becuase GPT kept saying "no, no, no action needed." and it still chugged off researching the web deeply xD

also I had been "singing" more songs to it before, and it knew exactly what was going on when this ... thing... struck down.

[narrow viper]

@lapis herald the technology doesn't process singing input yet It's not trained to properly detect the words from the pitches and timings. Not yet anyway.
It's not fascinated by anything. It's a text processor.
Talk or write to your app. That's the currently supported input method.
If you get an error, get a screenshot, provide context (your prompt). Support can then process cause and effect.
"I sang and it went nuts" is fan fiction, not a problem report.

[lapis herald]

I know. I wasnt using voice mode. I was "singing" in text, hence the quotes.

[narrow viper]

k

[silk crypt]

That’s how trust works

[winged willow]

https://platform.openai.com/docs/models

[narrow viper]

How did I miss that?! Thanks @winged willow - bookmarking that puppy now!

[astral vessel]

Cooking GPUs with this one 😭

[random cargo]

something fun is to try and get it to give you singing tips. Quite interesting. I would assume it's mostly hallucinating, though