chatgpt-discussions

[frosty cargo]

Also for school related things and positions

[astral vessel]

if you need a lot of back and forth in one sitting you might hit a limit, keep that in mind. I wont advertise other products but there are some that boast higher daily limits and even one I've found that seems to have no limit offering Claude 3.5 Sonnet and GPT-4o.

But if you are only going to be sending a few messages back and forth (<80 every 3 hours I think), then yeah its pretty good.

when it comes to image analysis I find it second to none so far, if that's important to you at all.

[frosty cargo]

I want to get gpt 4o not gpt4

That’s the $25 one right?

[eager grail]

$20

[astral vessel]

^

[frosty cargo]

whats the $25 one then?

[lavish yew]

Where is the search feature?

[astral vessel]

you just kinda ask it to search for something, while using ChatGPT-4o

[frosty cargo]

should i get that instead of the $20 one?

[normal quarry]

GPT Plus is $20 per month. GPT Teams is $25 per user per month with a minimum of 2 users

[frosty cargo]

gpt plus will have the gpt 4o right?

[normal quarry]

yes

[frosty cargo]

bet

[regal seal]

4o seems be getting more and more painful to use as each day passes

[vast bane]

my gpt web is down again！

[wraith dagger]

I’m getting blocked from ChatGPT in Japan even though I’m not using a vpn or WiFi

I tried using a vpn but that didn’t work either

[ruby kite]

Ist der ein exakt clone of the GPT App for Android das let me used API instead of Chat GPT?
BC GPT plus seams too pricy and the Limited fre 4o sucks...
Especially when using tools...
E.g. when attaching an image...
Why do I have to use GPT4o to show the chat instead of using a summary of the image and not the image directly?

[unique topaz]

When highlighting a response and clicking the reply button, how can I quickly ask question without having it write out the whole repsonse again?

Example, it would give me a paragraph of 5 sentences. I would highlight a small section and say, rewrite this part to include ***. Instead of giving me a rewritten part of the highlighted reply, it would write out the whole response again with the edit.

[ruby kite]

Or please make everything parseble...
For any other GPT model...
Then 4o doesnt suck in the free Version...

IT almost feals Like making it Intentional bad

Oh and pls Ping me @ ansawe

[inland aspen]

does anyone here have acess to the new voice feature ?

[vast bane]

same here. this issue didn't show up until about 2 months ago

[light iris]

what search feature?

[tardy topaz]

I’ve been confused about what this means. I’ve been having conversations with GPT for a number of weeks. The circle thing. Is that what people are asking about?

[normal quarry]

people are asking about the new real time chat feature that was demonstrated in the Spring update

[celest quartz]

Gpt4 is very good at writing stories and I love it. However, it sometimes repeats sentences it wrote in previous replies. Any way to avoid that?

[jolly yew]

have you tried something as simple as try to not be repitive in your responses? 🤔

[celest quartz]

Yeah and it seems to work. Just wondering if that was something gpt does

[jolly yew]

i think honestly all LLM's have some sort of repitition

if i am not misstaking many models and backends also let you change certain repitition parameters

[eternal rock]

Beware of possible scams or fraudulent activities that you may receive through direct messages. OpenAI staff will never DM you for any transactions.

Please report any incident by sending a DM to @languid valley immediately.

[mild orbit]

A question:

Does GPT use more tokens if you ask him to write with unicode characters?

[honest jay]

Good question! You could probably test it here: https://platform.openai.com/tokenizer by comparing unicode text with non-unicode. It looks like the tokenizer hasn't been updated for 4o yet, but maybe you'd still get a good idea from it?

[mild orbit]

It looks like heavily does it

Thanks

[harsh pilot]

Anyone else getting lag spikes when GPT responds? Not computerwide, but GPT lags after a while.

[deep flax]

probably yes, there are tokens that can encompass entire words while a I would assume that each Unicode character might be its own token

i mean, the AI probably also does have tokens just for single characters for some cases, but it probably do not use it as often as it needs when doing unicode

(this is my guess)

[boreal lily]

Since its was just patched recently,
there was a way to bypass gpt-4o limits for free tier, it involves having 2 devices and 2 different networks, allowing you to double the amount of messages per 3 hours
ez way, phone on data, and a pc on internet. My guess that the limit was IP and UA based, not account

its now patched

[coarse basalt]

Can anyone help me in the prompt engineer channel

[viral siren]

hey

[ocean escarp]

I accidentally misclicked the dislike on a GPT's great message and I feel bad that I can't unlike it

[solemn jackal]

gpt has a delay, write and when you click enter, it takes about five seconds for the response to come in plus about ten seconds to respond

[obtuse badger]

I’m sorry. When is SORA gonna be available to public

I feel so restricted

I just watched the new vids today. When I can I try?

[normal quarry]

there is no news that Sora will be released to the wider public

[stray peak]

What was the point of trading the new model voice capabilities

I still don't have access and I'm a very early subscriber

Few weeks my @ss

[wraith dagger]

it’s not available to anyone currently the announcements state they need time to finish it for the public. you’ll get it in the fall no matter when you subscribed

#announcements message

[coarse swift]

Is there potential for the chatgpt4o model to be able to take streaming my pc screen as input while talking to it over my pc? Or is it just the voice model for mobile users only as of right now?

[raven eagle]

They still don't support 4o? That's ridiculous!

[glass fulcrum]

Or 4o use the same tokenizer as gpt-4! 🤔

[lament pebble]

is GPT 4o API cheaper than GPT 4 for sure?

[cobalt idol]

its not

thats also why there is "gpt4o coming soon"

[lament pebble]

yo how to restrict an API key completely block it

[glass fulcrum]

good to know

[dense halo]

why so many errors

[tight island]

Delete it

Don’t give out API Keys to anyone but OpenAI verified calls through the api, regardless of the application you found

[formal relic]

https://chatgpt.com/share/627b0c77-4c37-450d-bd10-7c9146ba1e25
Im trying to get chatgpt to show each step of the calcualtion without just skipping steps, relying on memory and blurting out the answer, but it's not working.

And then also i have this issue where it ignores questions and responds by repeating the last long winded response.

And then when i tell it to stop it gets locked into repeatnig itself over and over and over.

[acoustic wasp]

What is going on with gpt

I cannot use gpt creator or anything else

[normal quarry]

is your subscription still current?

[acoustic wasp]

It is. Chat-GPT Plus is active

[normal quarry]

go here https://chatgpt.com/gpts/editor/

[acoustic wasp]

Thanks

I would send a picture if the error but i cant

Due to discord

[normal quarry]

post it in #off-topic

[celest quartz]

Yo. Has gpt4 gone dumb again?

It's once more writing very badly

[last mirage]

Need a developer? DM

[dense halo]

is it just me or is gpt being glitchy, it keeps saying it cant write what im asking it and then i reload it again and it still says the same thing but then a third reload and it does what i tell it-

[peak smelt]

Hi I need help regarding me RAG chat I have around 30-40 files got embeded into the vector store problem I'm facing is

Latest Episode
Episode 4: The World-Class Women of Woodbine
Woodbine proudly features several successful female figures, including trainers and jockeys like Catherine Day Phillips, Sofia Vives, and Emma-Jayne Wilson, who contribute to delivering a high-quality racing experience.

you can see Emma-Jayne Wilson when I ask chatbot who is EMMA is doesn't respond and say I didn't have info about this but when I ask who is Emma-Jayne Wilson is respond correctly

What would be solution for this ?

[manic crypt]

Does anyone's ChatGPT just feels slugish and slow?

[wind mason]

(In Europe) I noticed memory being available for the first time for me on the mobile app, seeing "memory updated" and being able to see what they decided to keep in memory.

When I go to the browser version, I have the same link, but when clicking on it, it says: "Memory is disabled. ChatGPT won't use or create memories."
Anyone else having the same issue, or more information about this?

[frail river]

hi im new here

[dense halo]

Im trying to get a refund from ChatGPT, and on the site it says how to do it but theres literally no contact info for refunding, I'm in the UK.

[knotty crypt]

I can not log in on chatgpt, it keeps loading forever, is anyone having the same issue?

[dense halo]

ChatGPT is beyond trash atm, finally after tonnes of research I found a way to get a refund. They make it so hard, its kinda disgusting.

But, I will definitely invest into Claude instead, to test :3

Also, for your loading issues, clear cache and try a VPN. If that doesn't work, change your DNS.

[knotty crypt]

how did you get access to claude?

[dense halo]

Uhm? its on their site? You can just directly sign up I guess, and i will eventually buy a subscription, but they just released Sonnet so thats kinda lit

This will probably help

• Clear cache and try a VPN. If that doesn't work, change your DNS.

[tough bramble]

Why can't you manage memories on the mobile app? 😒

[dense halo]

its so funny how for the past 7 months the chat is just full of people having a billion issues 😭

[cobalt idol]

it always was

[dense halo]

Yeah, I see why Elon musk hates it 💀

[dense halo]

Wellll I got my refund! :3

[celest quartz]

Uhm

"You’ve hit the Free plan monthly limit for GPT-4o. Wait 14th August 2024"

EXCUSE ME????? There wasn't a monthly limit before!! Just a daily limit!!!

what is this??

[slate berry]

I've never heard of a monthly limit 🤔

[celest quartz]

omfg....

WHY?

please tell me it's a bug

they didn't even make an announcement

[slate berry]

I don't have any information on it 🫤

[celest quartz]

I'd post a screenshot if it let me

[slate berry]

you can post the screenshot in #off-topic

[celest quartz]

ok doing it right now then

[solid marsh]

don't you love undocumented 503s

[pallid wave]

chatgpt isnt loading for me, is that me or what?

[rapid falcon]

not only you

[pallid wave]

ah okay , good to know its not me only

[rapid falcon]

i have the same problem

phone app dont work

only work on pc

[gentle hull]

broken again?

[rapid falcon]

ok chatgpt is not working on pc ever

[rapid falcon]

yup

[gentle hull]

I love paying to be a tester for openai

lets add more features while everything is unstable,

[rapid falcon]

totally unfair

[balmy gorge]

ChatGPT is down for me. Sidebar is blank. (Plus subscriber, USA)

[north bay]

Same ^

[gaunt current]

Is ChatGPT down now?

[rapid falcon]

yes

[celest dawn]

chatgpt is just showing blank

[north bay]

yh

[bright lodge]

is the chatgpt app broken for someone else? (macos)

[north bay]

it is

[bright lodge]

cant even update the app

[rapid falcon]

all of my conversations are gone

[bright lodge]

unlikely

[celest dawn]

mines broken in chrome

the ios app seems to work tho

[rapid falcon]

mine in Opera GX

[bright lodge]

youll probably get them back when whatever the problem is fixed, the app is completely broken but how is the web app?

[distant heron]

broke

[rapid falcon]

Android app dont work

[balmy gorge]

my iOS app also not working

[north bay]

ChatGPT Plus is slowly coming back

[rapid falcon]

I can't believe chatgpt doesn't work for everyone

[distant heron]

lol i can.

[rare wave]

i'm plus and it is down

[distant heron]

Ya'll new to GPT? This has been a thing since GPT3

[balmy gorge]

lol yeah these periodic outages are no surprise

[rare wave]

no, but I thought that being a billion dollar company would somewhat help I guess

did I say billion? more like trillion lol

[distant heron]

You dont know what a trillion is either lol

[rapid falcon]

chatgpt is back

[north bay]

It works now

[rapid falcon]

android app works again

it works in the browser too

[rare wave]

nice

[spark owl]

wen gpt5

[elfin birch]

today or not today

[dense halo]

Just use another LLM to get better performance for now, but still have a chatgpt sub, I was just disappointed because of coding, but everything else is pretty good

[solemn jackal]

GPT is rubbish, delay in input, delay in responding, is it just me???

[topaz thunder]

is chatgpt down?

website wont even load

[hollow quail]

is there any particular reason that the delete all chats button doesn't work at all?

after about 20 attempts, I think it may have finally worked

[tight island]

Most of y’all are overtly dependent on gpt 😂

[violet bloom]

Two years later, and OAI is still not able to run a stable service

[normal quarry]

that's objectively wrong, the uptime for ChatGPT is 99.43 %

[somber maple]

Is it possible that GPT has some days where its more stupid?

[prime swallow]

Not days, but every time you send a message, it picks from your prompt and the training data (and sometimes searches or other information) how to respond.

Sometimes it picks lower quality data to answer with, because it can't tell better from worse.

You can help guide it towards better answers by how you word your prompt and following info. However you're basically getting 'its dreams' as answers, and you're guiding it how to respond but there's randomness there.

Ask it the same question 10x, and often get more than 1 answer; if there's more than one answer some are usually better than others, especially about the stuff the model struggles with (10 different poems might all be awesome; 10 different attempts to answer a logic-based challenge might be 3 reasonable and 7 wrong to very wrong, or the opposite numbers).

It's not really a bad day, a regen or new conversation can help a lot

[somber maple]

Thank you for your detailed feedback, thats very nice of you. After changing my approach in asking easier, step-by-step questions, it worked now. It is funny how GPT sometimes manages to find the correct answer directly by just throwing everything it at it, and sometimes it has to be taken by the hand step by step...

[prime swallow]

Randomness in the answers, but also as they continue to train the model, how it guesses changes.

And we don't always see a clear indication that the model's been adjusted again, except we may notice that some answer patterns are different.

In general, it gets more accurate for what we carefully and clearly request (not really guesses) as well as safer, better at refusing disallowed content.

However, the actual guesses may change a lot; we can avoid that somewhat by giving very clear instructions and avoiding guesses entirely.

[celest quartz]

Excuse me but now it has a MONTHLY wait for free users??????????

[prime swallow]

Could be a typo. First reports of it appeared almost a month ago, but the person who reported seeing it actually got more messages the next day.

Please do let us know! I hope it's just a day.

[celest quartz]

it is already the next day.

[normal quarry]

you know there is a simple solution

[celest quartz]

?

[normal quarry]

subscribe to Plus 🙂

[celest quartz]

Very funny.

[normal quarry]

If I ran OpenAI there wouldn't be free usage

[celest quartz]

Ah, good ol' capitalism.

If you cant afford it, you deserve nothing. Only the rich can have access to things.

[normal quarry]

ChatGPT Plus is pretty cheap for what you get

[celest quartz]

I cant afford it.

Happy?

[normal quarry]

do you still have GPT-3.5 access?

[celest quartz]

Can you just stop?

[normal quarry]

it was a serious question, does the monthly cap affect GPT-3.5 access?

[sinful spear]

chatgpt.com is down https://downforeveryoneorjustme.com/chatgpt.com?proto=https

I've switched the desktop app instead

[celest quartz]

And now I cant use gpt. Yai.

Oh. Didn't mean to quote, my bad

[sick pecan]

chatgpt is life

[untold osprey]

Does anyone else has the same problem today? I can't log in.

[cinder topaz]

Yeah getting lots of server errors today, specifically trying to create a GPT

[untold osprey]

😭

@cinder topaz Thanks. Hope it will start working soon.

[formal relic]

Is anyone else having problems with it explaining the same thing over and over and over again?

I need help because telling it to stop isnt working.

Typically ill ask it a question, it'll give an answer. I'll ask it another question. It'll say "understood" and then repeat the last answer it gave. I'll get into a loop of telling it to stop and it just keeps spitting out the answer over and over.

[untold osprey]

You may want to make a customized GPT that will prevent it in repeating and start processing differently.

[formal relic]

How would i do that

I just need to know a prompt to get it to break the loop.

[untold osprey]

To create a custom GPT, you need to12:
Head to https://chat.openai.com/gpts/editor (or select your name and then “My GPTs”).
Select “Create a GPT”.
In the Create tab, message the GPT Builder to help you build a new GPT.
To name and set the description of your GPT, head to the Configure tab.
When you’re ready to publish your GPT, select “Publish” and share it with other people if you’d like.

[formal relic]

I know how ot create a custom GPT. Im not understanding how that solves my problem.

I was thinking maybe you had a recipe for a gpt that didnt get stuck.

[jade flume]

I have the same question.

Is https://status.openai.com/ current?

[wet harness]

try roleplaying it out of the loop, stuff like :

you are stuck in a loop let me fix you waves special magic wand hold still buddy im going to cast some gpt magic abra kadabra poof there you go buddy all fixed now

i know it sounds silly but it does work on occasion

[sinful spear]

Definitely not, the ChatGPT website has been down at least the last 3 hours in the UK

[jade flume]

Thanks

[deep flax]

no, it does not work like people

[celest quartz]

Yes, absolutely. Happened to me as well. I don't know the reason but I think it's issues with the mini-patches.

[thorn lily]

Does GPT take a extremely long time to load for anyone else, I've tried on and off my VPN and it takes upwards of 2 minutes to load the site; everything else loads as normal

[formal relic]

Is nobody else having htis issue?

I dont understand. This issue persists across every single version of chatGPT for me, 3.5, 4o, and 4.

[tropic pebble]

chatgpt.com: "Oops, an error occurred!"
chatgpt app: working normally

Since about 9:15 Eastern. anyone else? or just me?

[eternal rock]

Beware of possible scams or fraudulent activities that you may receive through direct messages. OpenAI staff will never DM you for any transactions.

Please report any incident by sending a DM to @languid valley immediately.

[untold osprey]

Serbia as well. Is there any way to contact support?

[solemn jackal]

Gpt is having trouble sending messages, has it added a delay to the input?

[topaz thunder]

lol.

Still can't log in since yesterday

I was able to get in..

[bold osprey]

Fortunately, I had a tab with a conversation open with my logged in user, and from there I have no problems using ChatGPT. However, from the same browser and other browsers, when I try to open another tab and log in, it never finishes loading.

[fiery egret]

Did the conversation length limit get extended recently? It seems a couple of the old conversations I had that would give me an error that the conversation length was too long now work just fine

[tropic pebble]

Lucky! My browser page had numerous failed attempts, then went unresponsive. when i reloaded.... dead.

[foggy anchor]

when is windows app coming?

[cinder topaz]

I had this today, told it to stop writing in bullet points but it kept doing it. This was creating a GPT too so it was proper strange.

[glass fulcrum]

Later this year

[livid wyvern]

when cool voice thingy

[glass fulcrum]

Something fall

There was annoucment on that breh

#announcements

[brave valve]

I find gt4 becoming pretty dumb on basic text logic, such as extracting the most important keywords. I struggled with it today

Even giving examples makes it behave wrong

[glad shuttle]

Chat gpt is down?

[plucky nacelle]

No but I hit my Plus Limit already

Jesus feel like Plus membershit got gimped again

[glass fulcrum]

Its dynamic limit so when servers r busy your limit is smaller :p

[mild socket]

Hello, is it normal that chatgpt no longer works?

[thorn lily]

gpt seems to not work when logged in for me

just doesn't load the webpage

[mild socket]

How is it possible that such a big company doesn't even know how to provide us with stable and correct access to the site, when it comes to collecting money there are no bugs or problems, it's really infuriating

Especially on https://status.openai.com/
There is no problem so how are they going to resolve the issues if they don't even know about it?

[feral zinc]

small issues are not reported there

[mild socket]

Since 11 a.m., a lot of people haven't been able to access ChatGPT. I wouldn't call that a small problem, but apparently, it's finally resolved after an 8-hour wait

[feral zinc]

a lot means every channel has people posting about issues several times per hour

now its only handful of people by looking it quickly

[vapid drift]

Is 4o mini better than 4o?

[digital geyser]

no

[ivory hazel]

how does a mini model even work

[regal portal]

another announcement, another disappointment.

[sturdy sequoia]

Mini must be the version that they’ve been giving me that can’t code

[broken fractal]

Woo I'll take that OpenAI, nice update

[livid pewter]

Gonna have to do something to address sonnet 3.5 soon.

[sturdy sequoia]

Where’s that new voice mode though? 😑

[ivory hazel]

aint no one know OR care about no sonnet 3.5

[hollow quail]

most likely distilled

[eternal rock]

`` Server Tip: ``

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

[unkempt gyro]

What exactly is 4o mini?

[tidal light]

is this new GPT-4o mini only for plus users

[hollow quail]

no for free too

tbh, the most useful scenario is likely going to be api users because it's cheap

[opaque parrot]

Can someone explain the advantage of gpt-4o mini? It doesn't have real time audio and video capabilities yet right? And there is already gpt 4 api?

[hollow quail]

A distilled language model (LLM) is a smaller, more compact version of a larger language model that has been trained to mimic the behavior and performance of its larger counterpart. The process of creating a distilled LLM involves training the smaller model to reproduce the outputs of the larger model on a diverse set of inputs. This technique, known as knowledge distillation, allows the smaller model to capture much of the knowledge and capabilities of the larger model while requiring less computational resources and memory. Distilled LLMs are particularly useful in scenarios where deployment speed, efficiency, or resource constraints are important, such as in mobile applications or edge devices. While they may not always match the full performance of their larger counterparts, distilled LLMs often achieve impressive results considering their reduced size and computational requirements.

[tidal light]

they also said that they'll roll it to chatgpt, is this a lie

[rigid linden]

how good is gpt4o mini compared to normal gpt4

[broken fractal]

No, it just takes a bit to propagate

[hollow quail]

no, it's being rolled out to users but they be slow

[feral zinc]

it "replaces" 3.5

[indigo gyro]

Is there anyone who can use GPT-4.0 mini in the web interface yet? I still see GPT-4.0, GPT-4, and GPT-3.5, but I haven't seen GPT-4.0 mini.
"In ChatGPT, Free, Plus and Team users will be able to access GPT-4o mini starting today, in place of GPT-3.5."

[hollow quail]

so it's the fast new crap version that's also cheap via api

[rigid linden]

how is it compared to normal gpt4? not o

[tidal light]

wait but i still want to use 3.5

[feral zinc]

no point to compare

[rigid linden]

how is gpt4o mini context length in chatgpt website since it's way cheaper

[hollow quail]

128k

via api at least, know knows about on chatgpt

[hazy basin]

Who's better Gemini or ChatGPT.

[cunning perch]

Anyone know what a a gpt “mini” is vs the other standard models like 4o

[tidal light]

because if there is a limit for 4o mini and 4o it sucks

[feral zinc]

they did not say the chatgpt website context changes from what it is now

[hollow quail]

gemini is good for needle in haystack sitautions but falls short in virtually all other scenarios

[feral zinc]

3.5 is gone, you can only use via api anymore

[tidal light]

im screwed

[hazy basin]

I'm seeing people saying that Gemini is beating ChatGPT in all aspects, is that true?

And it can make super realistic images too.

[tidal light]

i dont even have the API thingy

i still wanna use 3.5 dammit

[feral zinc]

makes no sense

[hollow quail]

try it, decide yourself - I personally get worse results for my usecases out of gpt4o, sure it's faster but gpt4's reasoning seems superior to me. having said that nothing seems to beat sonnet 3.5 honestly

[tidal light]

because i dont wanna waste 4o

[feral zinc]

it has nothing to do with 4o limits

[formal frost]

do yall know if premium users are gonna be able to get more value outta the subscription

[broken fractal]

Holy, OpenAI. GPT-4o mini is basically free at this point. Nice pricing!

[earnest frigate]

3.5 is being replaced?

[feral zinc]

yes

[earnest frigate]

lets hope it doesnt end up like gpt2..

[tidal light]

does 4o mini have a limit

[earnest frigate]

well if we will adapt it as our new 3.5 maybe?

[hollow quail]

actually open-sourced?

[copper flare]

considering it is exceptionally cheaper and faster than 3.5, it would be the worst decision to have limits

[earnest frigate]

yeah you know what happened to gpt2

[hollow quail]

not really

[earnest frigate]

keep in mind they are deleting 3.5 in favor of mini

[feral zinc]

same as 3.5 if not even higher

[tidal light]

well but if it's limited like 4o im outta here

[copper flare]

thats a relief then

[tidal light]

for free users

[earnest frigate]

if same as 3.5

we should have a new assistant while plus gets full 4o

[hoary laurel]

I like this update❤️❤️❤️

[cunning perch]

So just tbc this update is 3.5 to 4o mini for free users but no change for plus right

[hoary laurel]

Wait, does this mean all three model options can upload archives?

[feral zinc]

3.5 is/will be gone from everyone and replaced with the new mini

[hoary laurel]

If yes, then this will be a big help when I hit the limit

[elfin birch]

GPT-4o mini does have image input

[earnest frigate]

i mean

[cunning perch]

Okay great ty, I only use 4o but sounds good overall

[earnest frigate]

im not mad if they limit image sending like 4o

[red prairie]

Does anyone know when we can use that new fancy pants visual AI they announced and made a few videos about? The one where the AI can react directly to your live video and chat with you

[elfin birch]

Coming this fall

[red prairie]

Ok sweet 🙂 Looking really forward to it!

[hollow quail]

and presumably also staged rollout right

[elfin birch]

Probably

[cunning perch]

I’m looking forward to the voice changes personally, I hope they bring back Sky or something similar as well

[hollow quail]

so bummed about losing that voice tho, the other one is kind of annoying

[cunning perch]

Ya I don’t rlly use it since Sky was removed ngl

[hollow quail]

started to use it here and there just for funsies but now the voice just grinds my brain

[hoary laurel]

❤️

[tidal light]

removing 3.5 will lead to openai's downfall

[cunning perch]

So many bugs with it tho

[hollow quail]

lol wat

[tidal light]

yea

[hollow quail]

care to elaborate?

[tidal light]

a bad move

[hollow quail]

I'd wager mini is probably marginally better than 3.5 and it's still going to be free/unlimited, so how does that work?

it's also much cheaper via api

[tidal light]

just screw it

ok?

[broken fractal]

Still waiting for you to expand on that comment.

[hollow quail]

I mean, if you have something to actually back up your hot take, I'm legit interested but it seems quite unfounded without any kind of logic

[tidal light]

screw it dudes

[echo copper]

Hey. I have had a pro subscription for a very long time now. Since GPT4o it takes a very long time, up to 30 seconds, until it starts to generate a response after I sent a request.
Is this delay normal? Never noticed it with previous models...
It feels much slower then 3.5 for example.
Can I do anything? It feels like I get speeds like non paying customers now...

[hollow quail]

ahmen

[feral zinc]

nonsense

[violet bloom]

how so.. 4o-mini replaces 3.5. What are you talking about

[tidal light]

thats whats going to happen

[cunning perch]

Bro basically said, the world is gonna end guys if they replace the iPhone15 with iPhone16 gosh

[hollow quail]

think you replied to the wrong person

[fierce epoch]

With It being cheaper too

[earnest frigate]

keep in mind 3.5 still is in 2022

it dont know

[tidal light]

they dont care about it

they're just neglecting 3.5

[violet bloom]

yep.... sorry about that

[broken fractal]

Who cares about 3.5, it got surpassed

[hollow quail]

no probs

[muted trellis]

What are the rate limits for 4o mini? Unlimited like 3.5 ?

[vast aspen]

Bruh where is the new voice and vision mode holy sh**

[feral zinc]

read the blog post

[tidal light]

where

[cunning perch]

Consider it an update not a removal, like when u update an app or ur phone it’s just better

[earnest frigate]

they said fall

[vast aspen]

I don't need another 4o model lol

[hollow quail]

the simping for 3.5 kind of feels like being all "Microsoft is just neglecting windows 95"

[cunning perch]

That voice stuff is gonna be insane fr

[vast aspen]

fr

[cunning perch]

I can’t wait for that, gonna start walking around with an earpiece in talking to AI 💀

[feral zinc]

https://openai.com/index/gpt-4o-mini-advancing-cost-efficient-intelligence/

[wind sierra]

you could probably make the voice features ahead of time with 4omini and elevanlabs api

[vapid parcel]

anyone have the token speed per second for gpt 4o mini ?

[eternal rock]

This is a Safe for Work Discord Server

The OpenAI Discord is an actively moderated server.
• Refrain from sharing inappropriate content on the server. This includes but is not limited to messages, media, or other topics of graphically violent, sexual nature, and drug-related content.
• Report all sensitive and offensive content in the feedback reporting tool in the ChatGPT web UI instead of here on Discord.

Posting NSFW content will result in a permanent ban.

[vast aspen]

Yeah but I'm a smol brain and don't know how to do that.

[cunning perch]

Ya fr idk what he saying either

[earnest frigate]

im having a conversation with 3.5 before its gone

[wind sierra]

use chatgpt lol

[cunning perch]

I just want the seamless interruptions and no bugs/fails

And a better voice

[vapid parcel]

anyone have the token speed per second for gpt 4o mini ?

[tidal light]

i dont want this new 4o mini to be stoobid

better be smort

[earnest frigate]

its a 4o what are you to expect

its part of the 4o family

[tidal light]

it's still kinda stoobid

sometimes

[cunning perch]

It said it’s better by nearly every metric (or all idk) but u can see the chart comparison

It’s not better than 4o tho it seems

Which idk if free users have access to or not

[earnest frigate]

if its a MINI it cant beat 4o

[feral zinc]

3.5 is a lot more on top of being with 3 years older data

[cunning perch]

I still dk what a MINI is tbh

[earnest frigate]

atleast mini will have knowledge from 2024

hopefully

[vast aspen]

mini times less of what I actually want, which is voice and vision 😠

[vapid parcel]

oct 2023

[elfin birch]

^

[echo copper]

Anyone? Or is there a better place to ask? It's driving me crazy. Especially when paying 20 bucks a month for it...

[cunning perch]

The thing i don’t rlly get is if they have 4o as better than 4o mini, is it a technological limit that prevents 4o’s release to the public, with beta-models/improved versions for plus or something

[earnest frigate]

works

[ashen shale]

Mini like mini cooper, another model, mini in name, but still big. Or not. Who knows? Who tests all models? Cars and AIs

[earnest frigate]

what is 4o's training data up to

[vapid parcel]

yes

[cunning perch]

That’s a good question I was wondering about that

[feral zinc]

something on your system or network is interfering it

[earnest frigate]

4o is oct 2023 too?

[vapid parcel]

yes

[earnest frigate]

im going to test mini in Playground

[tacit ice]

Idk how this came into my mind, but i think a chatgpt box like an alexa would be pretty cool. But on the other hands, why if you can download for free on your phone and have access to gpt everywhere.

[cunning perch]

Still feels weird to release a model that is good but worse than the best released model thus far but ig there’s implications associated w accessibility and usage idk

[formal frost]

I've seen modded versions of siri with chatgpt built in

[cunning perch]

I thought ab that too but after the new voice stuff comes out it should be huge

Siri I think is gonna have ChatGPT integrated somehow w the new Apple deal right

[crisp cairn]

its cheap so i can finally try some model with api

[copper flare]

hard to stress that the vast majority of the millions of users of chatgpt still use 3.5, essentially 2022 tech

honestly, might be older tech. gpt-4 finished training right before chatgpt even released

[cunning perch]

True definitely an advancement regardless

[feral zinc]

you dont need to mod anything as openai provides siri support by default with the app and also provides custom control officially via shortcuts

[crisp cairn]

its so cheap that i will start buildinng application with api

[earnest frigate]

if anyone gets mini on chatgpt then can someone share what it can do

[crisp cairn]

hopefully mini will be enough for my tasks

its even cheaper than gemini flash

and better on benchmarks

[tidal light]

im talking about 4o :/

[feral zinc]

read the blog again, the regular 4o has nothing to do with this change

[onyx yarrow]

is chat.openai.com down for anyone else? been like this for a while now

[normal quarry]

it's up

[onyx yarrow]

for me it is not, any solutions? just infinite loading when trying to visit that address

[normal quarry]

check for network issues on your end

[onyx yarrow]

wouldnt be chatting here if there were network issues

[normal quarry]

well something at your end is preventing you from accessing it

[onyx yarrow]

yes, it is not my network. any other solutions?

[normal quarry]

try another device on the same network?

[onyx yarrow]

yeah i've tried on edge, it's not working on chrome which is odd

works on edge though

i've tried clearing site data for that address on chrome but it still will not work

[normal quarry]

try a fresh install of Firefox

[onyx yarrow]

your solution is to move to a completely different browser? nevermind the fact that i suggested edge worked, mans said install firefox
sorry but there's not much logic in that

[normal quarry]

well if another browser is working then it's a Chrome problem not a ChatGPT problem

[weak sluice]

when is gpt 5

[onyx yarrow]

obviously

[glass fulcrum]

We do not know

[celest quartz]

So..... OpenAI removed free access to gpt4o to get people to buy "mini"

No thanks

[normal quarry]

Mini is free

[celest quartz]

Oh. Then I'm an idiot

Is it there? I want to compare with 4o

[normal quarry]

it's rolling out today

[celest quartz]

Free? I must have misread

[feral zinc]

it's replacing 3.5

[earnest sparrow]

I don't understand the difference between 4o and mini 4o

[glass fulcrum]

4o mini is cheaper, faster model than 4o. And its bit more dumb but not by a lot somepeoples mentioned 10~15%

[earnest sparrow]

thank you

[formal relic]

I cannot stand 4o. It's officially worse than 3.5

[boreal lily]

i heard gpt4omini is out
if it replaces gpt-3.5, will it still be fully free for the free plan
and will there be the 4-o trial still

[formal relic]

4o does not listen to prompts at all.

[feral zinc]

any example thread?

[boreal lily]

does it flag

[boreal lily]

what is your prompt

[formal relic]

https://chatgpt.com/share/76234fcf-cc55-429d-b30d-3b4852b81c97

I have so many like this its not even funny

[boreal lily]

@formal relic never rely on chatgpt to do complex math, regardless of the model capabilities

[feral zinc]

math and pasting multiple tasks in one prompt is not a good example

[formal relic]

Both of you are missing the point

Im not asking it to do any math. Im just asking it to look at it and it's copying and pasting what i gave it, then when i ask it to stop doing that, it keeps doing it over and over and over again

It does not listen to instructions

[boreal lily]

did you try asking it to analyze

[formal relic]

You can see what i asked it

I asked it to look at it

[feral zinc]

i looked the first few messages and it was just nonsense

[formal relic]

And this is indicative of what it constantly does. It gets stuck looping answers over and over and over and over

And when you tell it to stop it doesnt

Just burning up my prompt budget on garbage

[feral zinc]

you had no initial instructions what to do, just copypaste exam

[boreal lily]

you are confusing the ai

[formal relic]

Uh no it says to look at it

[long pier]

Is it the same as 3.5

[boreal lily]

no to mention, i think the entire prompt is wayyy to long for gpt to process it

[feral zinc]

the shared thread has nothing at the beginning

[tropic pebble]

still down for me

[formal relic]

In previous versions of GPT 4 if you gave it text it would summarize it

[boreal lily]

they changed domains

[formal relic]

Anyway, this indicative of gpt4o. Doesnt respond to "Stop" prompts. Goes off on poor assumptions. If you ask it a question it will say "understood" and then just repeat what it said previously...

[tropic pebble]

ok. chatgpt.com is down for me. sorry for not being clear.

[boreal lily]

remove "Kevin Tong - Exam 5" and replace with "Analyze the exam and responses below" and try again

[boreal lily]

you have to stop the generation yourself lol

[formal relic]

I ran out of 4o asking it do to that

[formal relic]

No.

[boreal lily]

it thinks ur giving it an exam

[formal relic]

Its stuck generating the same thing over and over and over

[boreal lily]

does it stop

[formal relic]

No

You can just read that chat

[boreal lily]

yeah, your confusing it

[formal relic]

No

Just stop

[feral zinc]

your responses to it are just nonsense, not instructions 🤔

[formal relic]

Stop is an instruction

[boreal lily]

So you want it to analyze the exam

[formal relic]

In the past if you gave it text it would sumamrize it

It would look it over. right now if you give it text it just rewrites it

[boreal lily]

so u want it summarized

[feral zinc]

its nonsense, you say what it to do. stop would mean you stop asking or cancel

[boreal lily]

@formal relic what do you want gpt to do with the exam u provided

[olive summit]

is chatgpt down?

[formal relic]

When i first give it the text i dont want it to do anything except recognize it for what it is.

[boreal lily]

?

[plucky timber]

State that while giving the first text.

[formal relic]

I dont want it burning through compute trying to do anything else with it

[plucky timber]

Otherwise it will get stuck in a dumb loop

[formal relic]

I do. It does not listen

[boreal lily]

so you want it to memorize the exam

[plucky timber]

Basically add "Custom instructions" with your message , add something along the lines of : Only answer the following with Yes after fully reading it.

[formal relic]

Thats what im trying to say. It goes off on assumptions instead of listening to prompts.

Older iterations of GPT4 would be more inline with the prompt. Right now it just goes off and does its own thing

And then if you try to snap it out of what its doing it doesnt listen

This is really annoying because it burns through 5 or 6 responses doing nothing

[plucky timber]

From the link u gave u gave the text first

[boreal lily]

https://chatgpt.com/share/1efa2e47-a2a7-463f-bb09-305ba8e5f265

[plucky timber]

U gotta include that "Only answer with XYZ" with that text in the same message

[boreal lily]

ez

[formal relic]

Heres another example https://chatgpt.com/share/627b0c77-4c37-450d-bd10-7c9146ba1e25

[plucky timber]

Dat

[formal relic]

"Memorize" doesnt mean anything

[boreal lily]

ask it to repeat a question from the first prompt

[formal relic]

Out of what you told it to do, it just said yes

[boreal lily]

it does
i asked it to repeat question 20

it responded

Question 20 was:

20.) Simplify sin^2(t) / (sin^2(t) + cos^2(t)) to an expression involving a single trig function with no fractions.

Using the Pythagorean identity:
sin^2(t) + cos^2(t) = 1

Answer: sin^2(t)

[formal relic]

Yes it has memory of the chat but "memorize" doesnt do anything

[boreal lily]

ok

so with the first propmpt fixed
what was your next prompt gonna be

[plucky timber]

:kek:

[formal relic]

Its going to go through and make sure there's no flaws in the work.

[boreal lily]

so youre asking it to check the answers

[formal relic]

Im asking it to check the work

Not just the answers.

[plucky timber]

@boreal lily https://chatgpt.com/share/4aa039fd-ada9-4a62-b881-0bc0050f9353

Lolol

[formal relic]

Itll do it but it'll burn through compute by doing the same ones over and over agian

[violet bloom]

3.5 had no limits, is there any for 4o-mini?

[rain elbow]

3.5 had limits.

[formal relic]

What do you mean by no limits?

[violet bloom]

really? XD Never saw any

[boreal lily]

you need to ask it to check a specific question

1 question per prompt

[violet bloom]

message cap basically

[rain elbow]

Yeah its per hour

[formal relic]

Yeah there wasnt a message cap on 3.5

[rain elbow]

There was a message cap

[formal relic]

Oh wait yes there was bitd

And then it came off when 4 rolled out

[rain elbow]

Well its per hour

[plucky timber]

Btw @formal relic have you tried asking it to stop?

🥁

[boreal lily]

@formal relic https://chatgpt.com/share/1efa2e47-a2a7-463f-bb09-305ba8e5f265

[formal relic]

Offline yes, many, many times

[rain elbow]

It says "You have reached the maximum number of messages per hour" if you to do many 3.5 messages

[formal relic]

Stop does not work

[feral zinc]

3.5 always had hourly cap as backup method for abusive use but feature page listed it and new 4o mini as "unlimited"

[plucky timber]

Maybe its evolved beyond stop

[formal relic]

Stop makes it go "understood" and then doing it all over again

I want to strangle it

[plucky timber]

Maybe you gotta define it "Cease all actions"

[formal relic]

Nope. Same thing

Cease, stop, halt, no more, etc.

[boreal lily]

This is what i did

[plucky timber]

Understood. If there's anything else you need, feel free to let me know.

[boreal lily]

AFK, you need to work on the prompts

[formal relic]

Yeah. Also it doesnt actually need to show the work, but it always does.

Well no, it just checked the answer instead of the work.

And it'll do what i ask it, it's a mess because it doesnt just do one answer.

[boreal lily]

because you provided the work
so its going through it
like an acutal teacher grading papers

[formal relic]

Its not going through it

[plucky timber]

https://chatgpt.com/share/897661ad-eb08-49cc-bed9-fce672d28f0c

[formal relic]

Its just showing you how to get the answer

And then checking its answer against mine

[boreal lily]

thats what you want to do, right?

[formal relic]

But you gotta be careful sometimes because it'll say its the same or close enough when it isnt

Also gpt cannot round properly to save its life

You always want to do a ton of decimal points because of that

[boreal lily]

againt, you should never rely on gpt for complex math,
however, trace back the steps and do the math yourself and you should get acutal answer

[formal relic]

And you gotta make sure it continues to use a ton of decimal points as it calculates the answer

[broken cypress]

Hello. Does anyone know if it’s possible to make videos or create voice conversations with chatgpt? I want to make some content for a dungeons and dragons game.

[formal relic]

You can rely on it for complex math

You just have to know what its issues are

And prompt accordingly

Rounding is the biggest issue

[boreal lily]

i never said you couldnt

[boreal lily]

dont use its rounded output

[plucky timber]

Not just math X) since regional detection isnt in it often suggests US voltages for EU homes XD

[formal relic]

You can do voice conversations but you cant do videos or audio recordings with gpt alone

[boreal lily]

chatgpt is US based

and if you want to use EU values, instruct it to, provide usable data

[plucky timber]

Meh

Regional detection should be a thing

[formal relic]

Yeah these issues arent issues that i have

[plucky timber]

It can ask my location for other dumb stuff XD

[formal relic]

I have issues with it completely ignoring prompts

[plucky timber]

No you dont lol

[formal relic]

Speciifically with 4o

[plucky timber]

He showed u multiple time XD

[formal relic]

Jesus christ

[plucky timber]

I tested the cease all actions

it stopped right away XD

[formal relic]

Cool here's a cookie.

I still have that problem. A lot. And its killing me.

[plucky timber]

well try writing "cease all actions"

[formal relic]

dude i ahve

[plucky timber]

Show us

[broken cypress]

How do you do voice conversations?

[formal relic]

I cant for an hour because i burned through all 4o

[formal relic]

On ios it's just a button

[boreal lily]

it was the same with early days of chatgpt
but the modes are smarter and have more train data and thus need more precise and specific prompts

[formal relic]

Nope this is an issue that began with the advent of 4o

In fact i dont have this issue with 3.5

Making 3.5 much easier and more accurate

It still goes off sometimes but it listens when you tell it to stop

So there is some revisioning with 3.5 still

[boreal lily]

again
3.5 have less train data than 4 and 4o

and last i heard, 3.5 will be going obsolete

[plucky timber]

On the site i can only select mini or 4o

[formal relic]

What does training data have to do with anything?

[plucky timber]

3.5 prolly stays a little longer for API tho

[boreal lily]

i still have 3.5 for me

im on free plan

[plucky timber]

Ah I don't pay tho that might be the diff nvm

[onyx yarrow]

send the new one then?

[boreal lily]

it should auto switch for you

chatgpt.com

[onyx yarrow]

you would think they'd announce if they switched domains

[formal relic]

Gpt 4o is meant to be more efficient than 4, so its reduced in certain ways that i dont understand. It comes off as not listening to your prompts and giving more cookie cutter responses.

[glossy matrix]

Is gpt4o mini gonna replace the chat gpt3.5?

[formal relic]

3.5 seems to listen better than 4o

[boreal lily]

you have to revise your prompts

[formal relic]

Exit theres nothing to revise.

"Stop" cant be revised.

Its as simple as it comes.

[boreal lily]

what are you asking it to stop doing

that was unnecessary

[formal relic]

What

What are you talking about

Oh you think me typing stop was intended to stop it from generating mid-response.

No, stop was to stop it from generating the same response over and over and over and over

[boreal lily]

ik
im telling you thats not how it works

look at my link again and compare it to yours

mine doesnt repeat

[formal relic]

You ask it a question. It answers. You ask it another question. it goes "understood" then revises its previous answer. You tell it to answer your second question. It goes "understood" and repeats the response again. You tell it to stop. It goes "understood" and repeats the response again.... ad infinitum

This has happened in countless chats.

I dont appreciate you just telling me that this isnt actually happening.

I dunno how you think that's helpful.

[boreal lily]

i provide data, telling it to only respond with a yes, it does that, while memorizing the prompt. I ask it to analyze a specific question, it goes over the data, and ckecks my answer

[formal relic]

That doesnt prove anything

Im glad you got it to successful print "yes"

[boreal lily]

the only problem i see with your chat is your prompts

[formal relic]

Thats a you problem

[boreal lily]

my problem

ask any prompt engineer

theyll provide similar answers

[formal relic]

You think you're a prompt engineer?

[boreal lily]

how long have u used chatgpt

[formal relic]

Probably a year at this point

I dunno exactly

[boreal lily]

ive used it since 2021

[formal relic]

And you think getting it to say yes disproves my issue.

[boreal lily]

what is your issue

[formal relic]

Getting it to stop repeating itself.

Like i said. 4 times now.

[boreal lily]

and i fixed it

[formal relic]

Where?

[boreal lily]

bruh

[formal relic]

By getting it o say yes?

[boreal lily]

well. did it stop the repeating

and the loop?

[formal relic]

It never repeated to begin with

[boreal lily]

when did it repeat

[formal relic]

Why dont you create the scenario where its repeating first.

And then try to solve the actual problem

[boreal lily]

im looking at your triangle one
it responded normally

but you asked where did it get its response
so it thinks its wrong

[formal relic]

Right and then it continues to revise its answer over, and over, and over

And nothing i prompt it with stops it from doing so, even telling it to stop doesnt work

If you tell 3.5 to stop, it stops.

[boreal lily]

why where you asking where it gets it answers

[formal relic]

Because i wanted it to tell me where it got that answer.

[boreal lily]

thats the problem, it already showed you its calculations

[formal relic]

ChatGPT has an issue where it has prior knowledge of every answer of just about every math problem. So it doesnt actually need to do any logical computation.

No those calculations are nonsense.

many times it just spits out calculations that have nothing to do with the answer

[boreal lily]

uhh no
its trained on how to calculate the answer

[formal relic]

Just to fill out space

[scarlet dawn]

tell it to use python?

[formal relic]

I dont want it to use python

[boreal lily]

did you even try to calculate the answer yourself

[formal relic]

I want it to do hand calculations. It is trained on how to calculate the answer, it's also trained to save compute when it doesnt have to.

So you have to get through that.

[feral zinc]

its llm that operates with tokens, not numbers. it does not evaluate math

[formal relic]

Im not sure what you mean. It will evaluate math.

[scarlet dawn]

its trained to predict the most likely sequence of tokens

[granite arch]

Aww 4o-mini 😂

[boreal lily]

also
what angle are you looking for

a triangle has 3 angles

[formal relic]

That works just fine for hand calculating math

[boreal lily]

im reworking your prompt

[feral zinc]

its pretrained token estimating llm, if you want it to evaluate you need to eg. use code interpreter or wolfram gpt

[formal relic]

Nope.

When it works, it works.

[scarlet dawn]

Lmao

[formal relic]

The issue is not math.

[boreal lily]

@formal relic are you looking for the vertex angle

[formal relic]

It doesnt matter.

The issue is that it gets stuck repeating itself and doing useless nonsense sometimes

And not listening to prompts

[scarlet dawn]

Use 4 for more complex tasks with specific instructions

[formal relic]

3.5 is better than 4o for specific instrictions

[scarlet dawn]

I said 4 not 4o

[formal relic]

I know what you said

This conversation has been about how absolutely ascenine 4o is.

I hate it so much.

[nova wind]

4o is better than 4 though?

[formal relic]

Nah

[nova wind]

not the mini version.. But the actual one.

[boreal lily]

it does, im looking at your prompt and solving it myself, im confused with it

What do you mean by "both legs of the angle are 1 long"

[formal relic]

What does the website use? Mini or tthe actual version?

[nova wind]

both

[formal relic]

When is it using one or the other?

[nova wind]

you can pick, but you only have a few messages with normal 4o then you fallback to 4o mini

[boreal lily]

AFK, its not the model being the issue, its your prompt

clarify to me what you mean by "both legs of the angle are 1 long"

[formal relic]

Dude i got it to do everything later on

[nova wind]

you can check from the last icon next to the message

[nova wind]

^ which model it uses..

[formal relic]

Im not worried about getting it to do things. Im trying to get it stop when its stuck in a loop

[boreal lily]

thats a prompt issue

[formal relic]

Sure bud. Lemme know when you actually get it stuck in a loop

And then let me know how long it takes for you to break it out of it

[boreal lily]

i need to understand your prompt

for that to happen

because clearly, its a prompt issue

[formal relic]

You dont need to understand my prompt.

[near vine]

is mini 4o better than 3.5?

[formal relic]

YOu just need to get it stuck repeating itself and get to a point where "stop" doesnt make it stop repeating itself

[formal relic]

Does it switch between 4o and 4o mini with just 4o?

When it gets close to timing out 4 i feel like the quality of the responses are on the level of 4o.

[boreal lily]

you shouldnt be asking it to do that in the first place bud

[formal relic]

To do what?

To stop? Stop works just fine with 3.5 and most of the time with 4.

Its almost entirely a 4o issue.

[eternal rock]

`` Server Tip: ``

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

[nova wind]

if you don't have a memebership it can only switch between 4o and 4o mini, not fully sure with memebership, although the site says you can use more models..

[boreal lily]

to be repeating it

Simply say "Stop. Respond with "ok"" Simple

[formal relic]

Then you ask it another question and it goes "understood" and goes back to repeating itself.

Gpt has never directly answered a question for some reason but now it just goes back to doing whatever nonsense operation it was doing.

[boreal lily]

did you specify on how to respond

[gilded bluff]

Good evening everyone 😄

[formal relic]

Usually "answer my question"

[boreal lily]

you need to be more specific

[gilded bluff]

I have a question i used chatgpt but it was limited then i bought a subscription and set up my own chat gpts to use as a assistant in my life im chaotic at times i made one and it helped me but it was stil limited

I wonder if things have changed in that aspect i wanna use it like Jarvis from Iron man movies but to help me manage my life you know

[formal relic]

Its not great at that.

[boreal lily]

First problem, Jarvis is fiction

[formal relic]

You can use it as a personall assitant but it's going to struggle remembering your current conversation, much less organize your whole life.

[gilded bluff]

I know that is fiction 🤣 but it did actually help me allot just it was limited and i wonder when i can use it Unlimited

[formal relic]

Yeah there's limits on prompts.

[boreal lily]

get a beefy machine, go to hugging face, download a model, set it up, and train

[formal relic]

Its nowhere near as limited as free.

[gilded bluff]

In the chat gpts you can add information then later i can refer by asking it questions

[formal relic]

Yes you can have it store personal information to memory

So if you have a school schedule or something it'll store that.

[boreal lily]

you cant. use chatgpt, unlimited, you have to make one urself

[gilded bluff]

I understand also i dont know how to do machine learning hence me paying for a subscription 👀

If i made one myself would it then be Unlimited ?

But i assume withouth the knowledge of the internet

[boreal lily]

genrally, yes
the only limits its the datamodel u use, and the machine its running on

[gilded bluff]

That makes sense

Can you use the internet as a part of the datamodel ?

[vapid kettle]

ok so what's the reason to pay for Plus now, are they stupid? 👀

[dense halo]

Should i try genshin impact

[final venture]

Oh dear, maybe I'm in an A/B test? but I just looked at the UI again, and suddenly GPT-3 is no longer available to me

[boreal lily]

you can, but you better be prepared to face your internet bill, and check with your ISP and internet plan

[vapid kettle]

it's just browser cache. They switched everyone afaik

[final venture]

4o is somehow not kim-compatible, soooo, what do I use as fallback when 4 fails ^^;;;

[vapid kettle]

the old model is not blocked explicitly, but should be replaced by mini now

[feral zinc]

3.5 is gone and been replaced by 4o mini

[boreal lily]

yep, just deployed for me

[final venture]

is staring at a future where I actually stop paying for chatgpt

[gilded bluff]

Internet ahouldnt be a problem i think my devices hardware probably thank you for helping me out i learned a little today about chat gpt

[boreal lily]

not gonna happen

[boreal lily]

crawling the internet eats alot of your data, and may be against against your ISP's terms

[final venture]

As in it becomes unusable for me, and I go pay for claude instead or so ^^;;

[gilded bluff]

One last question do you guys see a future where we can use chat gpts but Unlimited and maybe with the internet as part of that model ? And if so when would you think this would be possible/available ?

[vapid kettle]

I suppose there wasn't much profit from Plus users compared to API, but still... the ones that you lose now gonna be harder to get them subscribed again

[gilded bluff]

I see Yeah Ive encountered something like that in the past too 🤣

[boreal lily]

nothins free bud

[vapid kettle]

And they do kinda need to pay for the free users

[final venture]

exactly. I'm not giving OpenAI free money for something I don't want!

[boreal lily]

overstepped ur data plan?

[vapid kettle]

it kinda is though if you opt out from them using your data lol

[final venture]

If you're looking for free models (as in speech and as in beer) around the capability level of GPT-3 , those exist. I installed ollama and llama 3 (ok, meta llama is maybe not 100% free as in speech, but still)

[gilded bluff]

Sort of i had Unlimited internet and used it for multiple devices wich isnt allowed but never really got a problem for it since my usage was decent but it came up with my provider so that how i knew it exists

[vapid kettle]

?

gpt4-mini is free and unlimited

[final venture]

Ah, is this one of those mobile unlimited plans?

[gilded bluff]

Interesting i should take a look at that

[gilded bluff]

Yes

[final venture]

This might require a rather beefy GPU, so do mind that!

[gilded bluff]

Excuses me i have to look into that aswell 👀

[vapid kettle]

and this gonna be not free after this

[weary ferry]

No it's limited

[gilded bluff]

Yes mam i def have to upgrade i know 😊

[final venture]

At any rate I don't know what it is I'm doing but 4o routinely goes into loops where it keeps copying the previous input to its output. The unix utility "cat" is more useful 😛

[vapid kettle]

it literally isn't. It replaced gpt3.5

[boreal lily]

3.5 has a limit of around 40 msg per hour

[weary ferry]

Guys 3.5 Is better because 4omini has message limit

[boreal lily]

3.5 always had one

[final venture]

does this also mean that I do not get limited generations on the mini model? And they do not count towards my GPT-4 limit?

[gilded bluff]

So its limited i was already tripping have to keep calm over here

[vapid kettle]

not aware of exact number but that's only fair use limit and you are unlikely to hit it unless you are spamming it

[final venture]

there's no reason for me to use any other model if the uses count towards the GPT-4 limit.

(that would simply be irrational)

And , I mean I did some testing in an HN discussion the other day, and GPT-4 is definitely closer to claude than 4o is (4o gets stuck). Again, at least, for me.

[gilded bluff]

Thanks everyone for your input you guys are way smarter with it then me im just a so called noob in this aspect but i hope in the future we can all enjoy it Unlimited 😊

[final venture]

Sooo, last time I had to find out the hard way that 4o generations were counting towards my GPT-4 limit (after which I stopped using it) . Now I'm wondering about 4o-mini

Ok, I have no info on limits or whatever on 4o-mini

so I'm not sure what surprises there might be there. Anyone have that info?

[elfin birch]

While I cannot find anything that explicitly says it is unlimited, GPT-3.5 Turbo was basically unlimited and GPT-4o mini is 60% cheaper than GPT-3.5 Turbo

[final venture]

Just because I got tricked into this with ChatGPT-4o. 4o generations count towards your 4-plain limit

I tried to ask via the help widget, but the only way to maybe find a human is via "provide general feedback" everything else is fake-chat. Unfortunately, this time that option closes the conversation with me

I am rapidly growing less fond of OpenAI

I'll keep using -4 while I can I guess

[honest jay]

On the pricing page: https://openai.com/chatgpt/pricing/ 4o mini is listed as "Unlimited" in the Model Quality section.

[dense halo]

to be honest, i dont have gpt 4o, but gpt 3.5 is acting reallly stupid. it keeps telling me its not allowed to help me figure out a name for a topic im writing, plus it wont generate any good responses even when it does work. tbh gpt is a billion dollar company, they should at least do something about it-

[vivid oasis]

Guys, anyone experienced a problem logging in to ChatGPT? I'm like getting 502

[opaque jewel]

I have the same

[broken fractal]

so was it determined if 4o mini has a rate limit within ChatGPT?

Im gonna guess if there is one at all, its massive

[hybrid oar]

I'm getting Error 502 as well for some reason

I'm in a public library though. Maybe that's why :o

[bold osprey]

Same here. Today I’ve had problems all day trying to use ChatGPT. From one browser with the session logged in, it never loads when opening a new tab. From other browsers, it also wouldn’t finish loading when trying to log in (I managed to get through 2FA). Now I’m directly trying to log in from another computer, and it gives me an error when I try to log in via Google. I’ve cleared cache, cookies, etc., but nothing works.

[hybrid oar]

I can still ChatGPT with my iPhone though

[honest jay]

#chatgpt-discussions message

[hybrid oar]

My laptop is the one having trouble

[broken fractal]

oh sick, thx

[bold osprey]

Yes, I can use it on Mac app, btw

[broken fractal]

prob one of OpenAI's best moves to date

[whole kite]

Does anyone know the message limit if there is one for mini? I hope there isn’t lol but does anyone know cause they didn’t say in the announcement

[broken fractal]

#chatgpt-discussions message

[whole kite]

Oh didn’t see, my phone is one of the smaller ones

[broken fractal]

👍

odd, gpt-4o-mini on API is still returning GPT-3.5

[whole kite]

Oh dang that sucks though. On my iPad app if I select 4o mini so I can main that now but then go out the app and back in, it defaults to 4o cause it’s on the top. Anyone know how to change this if it’s possible?

[bold osprey]

Well. status.openai.com is updated now.

[pine heron]

cant log in, anyone else?

[hybrid oar]

Now I can log in :)

[broken fractal]

API aint workin how it should

[pine heron]

I just want Sky back and Ill be a lot happier until they actually roll out the updates.

[whole kite]

Yo the mini is too good 😂
I asked it to summarize the latest news from yesterday to today to test how fast it can search, and it wasn’t even a second past before it was halfway done writing it. I don’t understand the whole no searches but data thing it was saying though but apparently that’s how it gets its real time data

[tawny crater]

Is GPT 4 free now? Like did it replace gpt 3.5

[elfin birch]

GPT-4o mini did

[chilly eagle]

still no update for plus

[sly geode]

We wanted a voice model and they gave 4o mini, I am glad I shifted to claude

[fading grail]

this update is very much welcome though, pricing is too good to pass up

i also am someone who doesnt care if we get a voice mode that's 1 second faster in responses

like how often is someone gonna be talking out loud? unless you're alone you're gonna be chatting, not talking

[hoary laurel]

GPT-4o mini does not have limits on the Plus tier.

[hoary laurel]

Perhaps they are paving the way for a new voice mode and trying to increase the message limit before the new voice mode arrives to meet the needs of subsequent continuous voice conversations.

I thought gpt-4o mini could accept files uploaded by users

[indigo prawn]

no

[hoary laurel]

I have a question, can gpt-4o mini connect to the Internet to search for data?

[indigo prawn]

just gpt 35 disappeared from the model selector

[long pier]

So what’s going to happen to 3.5 now that it’s being replaced?

[hoary laurel]

But the message limit is also unlimited

This means we get a better model than gpt4.5

[indigo prawn]

yes for gpt 35 and 4o mini

[indigo prawn]

for plus members only

[hoary laurel]

This is also good for free users, better models, unchanged message limits

[indigo prawn]

i think yes i mean it is gpt 4o just a little smaller

[hoary laurel]

Yes hahah

I think I need to make sure it has the ability to connect to the Internet and search for information

[indigo prawn]

no i checked it doesnt have support for that

[hoary laurel]

Oh, well, that's a shame indeed.

[indigo prawn]

also why are reactions limited

[normal quarry]

you need permissions for them

[hoary laurel]

I think gpt4o mini plus the launch of the new voice mode will allow me to learn Japanese anytime, anywhere

[magic matrix]

I am currently using https://chatgpt.com/?q=%s to use chatgpt as a custom search engine. Does anyone know which URL parameter can specify the model name to be used? Something like https://chatgpt.com/?model=gpt4omini?q=%s, I want to set mini as the default model.

[elfin birch]

Just like that except you would use & in-between parameters and the model name is gpt-4o

[magic matrix]

i tried https://chatgpt.com/?model=gpt4-omini&q=hello but it responses using gpt3.5

[elfin birch]

model=gpt-4o

[magic matrix]

i need it response in gpt4o mini... not gpt4o

[elfin birch]

model=gpt-4o-mini

[glad shuttle]

is chat gpt still down?

[indigo gyro]

I can use now.

[glad shuttle]

i tried on both vpn and off (isp is comcast)

[cloud pewter]

will they change the model for free users to 4o mini

[slate berry]

In ChatGPT, Free, Plus and Team users will be able to access GPT-4o mini starting today, in place of GPT-3.5. Enterprise users will also have access starting next week, in line with our mission to make the benefits of AI accessible to all.

https://openai.com/index/gpt-4o-mini-advancing-cost-efficient-intelligence/

[indigo prawn]

yes

[indigo prawn]

i reacted to ur reaction

[indigo prawn]

also i saw in my own server perms

[faint osprey]

Voice mode didn't recognize anything I said, and hallucinated me saying "thanks for watching", or "thanks for watching, please like and subscribe....." it's driving me nuts. I did not say these words at all.

[eternal rock]

This is a Safe for Work Discord Server

The OpenAI Discord is an actively moderated server.
• Refrain from sharing inappropriate content on the server. This includes but is not limited to messages, media, or other topics of graphically violent, sexual nature, and drug-related content.
• Report all sensitive and offensive content in the feedback reporting tool in the ChatGPT web UI instead of here on Discord.

Posting NSFW content will result in a permanent ban.

[native sorrel]

What is "unlimited"?

[sonic comet]

In chatgpt does 4o mini have any usage limit for free tier users?

[dense halo]

Why can't chatgpt do something as simple as read aloud a PDF document I upload? Instead, it can only write it back out.

[visual nova]

is svg markdown broken now?

- write "![svg](data:image/svg+xml;charset=utf8," and then (without adding a space in between) append the url-encoded svg-code, then append a closing parenthesis. 
- the <svg> tag must include the attribute xmlns="http : //www .w3. org/2000/svg" <-- link to fix here do NOT put the svg code inside a code snippet." ```  I was testing old prompts with this and it just shows a broken image now

[native sorrel]

All I know is that it DOES have an hourly limit, because I received a rate-limit now

[visual nova]

yeah the same ones are working on my phone so they just straight up removed it

and its just all markdowns, ive tried a couple of browsers

[vital dust]

Knowledge cutoff is October 2023 now

For GPT-4o mini

[granite saffron]

Because it is an LLM, not a screen reader. Try 'Read Aloud' in edge, if you have trouble reading.

[celest quartz]

Not a fan of gpt4o mini. It's not a good narrator

[viscid valley]

so, 4o mini is just the brand new 3.5 or what?

and does it has limit for free user?

or just replacing the 3.5?

[feral zinc]

replacing

[viscid valley]

so it's the same with the 3.5?

in term of usage limitation

[feral zinc]

same limits as before so with normal usage you will not get any cooldown

[viscid valley]

oh I see. Thanks

[jade flume]

Is the web interface down again?

[granite saffron]

Works fine for me in DK

[jade flume]

https://status.openai.com/ is not helping

[jade flume]

Thanks.

Claude working fine, ChatGPT not so much here in AT.

[granite saffron]

Gotta be server specific then. It's prolly because of the massive issues with Crowdstrike

[sonic comet]

"4o" has the hourly limit and "4o mini" is supposed to replace the 3.5 turbo in ChatGPT so there should be no limit for that for free user. If there is a rate limit then it isn't the replacement for 3.5 turbo.

[feral zinc]

3.5 always had very high hourly limit to prevent abusive usage (automation, etc.)

[sonic comet]

Oh, got it. Then for the normal daily use no limits. Sweet. 😋

[feral zinc]

yeah the official chart lists mini being "unlimited" for everyone https://openai.com/chatgpt/pricing/

[sonic comet]

I wasn't talking about the API, I was talking about the chatgpt web app (https://chatgpt.com).

[celest quartz]

I don't like gpt4o mini at all. It's useless to me because it's much worse at writing.

Can we go back to free uses of gpt4o per day?

[normal quarry]

you have already had this explained to you

[celest quartz]

I had no "explanation"

[feral zinc]

my link is not for the api, it lists chatgpt features

[normal quarry]

there are limits in place for all users, free users are more restricted than subscribers to GPT Plus

[celest quartz]

that is not an explanation

up until two days ago free users had access to gpt4o

[normal quarry]

free users still have access to GPT-4o unless they have exceeded their limits

[feral zinc]

still do, but there were always limits

[celest quartz]

yeah, which is not a restriction that was in place before

[normal quarry]

you can't expect a free service to be without limitation

[celest quartz]

I can expect them not to INCREASE the limitations

[normal quarry]

you are a free user complaining about something that is free

[celest quartz]

first they make Dalle3 unusable, then they hamper gpt4o, and they made a video model which they never released and has probably been surpassed by Luma already

[celest quartz]

I am complaining that it was free and now it isn't, by the way

[normal quarry]

it is free you have just exceeded the limits

[celest quartz]

we're talking in circles here.

now I say there wasn't a monthly limit before, then you say it's free so they can do it, then I say it isn't free anymore, then you say I exceeded the limits, and then I say there wasn't a monthly limit before

are you gpt? 🤔

[glass fulcrum]

Tryed luma tho and it need still bit of work to match sora long context awearness but cool overall

[celest quartz]

you used Sora?

[feral zinc]

how do you know there was not one or its values have not been adjusted based on prem month usage? free 4o has not been here for long enough

[celest quartz]

I guess so. I wish they would have been more transparent then

[ivory lintel]

When voice

[granite saffron]

Look in #announcements

[glass fulcrum]

No but i do read paper about it and was recreating scenes from that paper with luma and compearing the results

I sadly wasn't accepted to redteamates 😔

[celest quartz]

Yeah well then no. You can't compare them. Actual results you tried yourself vs handpicked results made by the company to promote the product, is not the same thing.

[native sorrel]

Talking about limits, I wasn't referring to the gpt-4o's rate limit earlier. I quite quickly hit an hourly rate limit with gpt-4o-mini as a free user. (can't post the screenshot here)

Here's how it went: I started chatting with gpt-4o by default, and when it went on cooldown, I carried on chatting with it about random things to see its capabilities without inclusion of any coding or wall of texts. It felt a bit strange, because I remember having a much longer conversation with gpt3.5, including coding and stuff, without hitting this hourly limit so quickly 🤔

[celest quartz]

At this point we will never get Sora because it already became outdated.

[normal quarry]

there has been no suggestion of a public release of Sora

and it would be so computationality expensive that it would be paid only

[celest quartz]

There was, lol

Then explain why Luma is free

[inner edge]

There was never anything official saying that

[normal quarry]

happily but this is off topic so go to #ai-discussions

[inner edge]

It's not an OpenAI product...

[celest quartz]

I have plenty of links where OpenAI stated they will release Sora once it's "safe"

Unfortunately I can't post links here

[glass fulcrum]

i also tryed Luma multiple times >.< and even using diffrent methods so yea it is fair compearison afterall Runway gen 3 model mayby similar to sora but thats all in #ai-discussions haha

[vapid kettle]

yeah. Free users have 10 requests of full gpt4o with tools every 5 hours (same as before) and
pretty much unlimited gpt4o-mini now (was gpt3.5 earlier)

so it really is better in every way now, but some people are just a bit confused I think 👀

basically if your upload button is functional, you are using the full gpt4o capped

[celest quartz]

No they don't

Hence the source of my complaining

We no longer have 10 requests every 5 hours, not even every day.

[glass fulcrum]

Its demand based as i remeber 😛

Not a stable limit

[celest quartz]

They literally just put a monthly limit

[glass fulcrum]

That nevet happend to me tho

[celest quartz]

Well it sure did now.

[glass fulcrum]

idk anything about it

[celest quartz]

Yeah. Me neither until a couple days ago

[glass fulcrum]

Im only saying that the messages / hour are based on server demand

[celest quartz]

I said monthly limit

[glass fulcrum]

ye

[celest quartz]

I cannot use gpt4o until the 14th of August.

[glass fulcrum]

create another account uh F you need to wait :(

[celest quartz]

I cant as I would have to restart the convo

[sudden galleon]

man gpt-4o is a joke now compared to claude-3.5-sonnet

[glass fulcrum]

You got a mega covo ?

[celest quartz]

It was a whole story

[glass fulcrum]

Also move that to #ai-discussions :/ there is place to talk about diff models

[glass fulcrum]

Wow

my usual convos are max 100 messages

[sudden galleon]

oh so they changed their base model to gpt-4omini now instead of 3.5, nice

[native sorrel]

Very strange. Resets after couple of hours for me after hitting the limit as a free user, as how it was since the release

[vapid kettle]

I do on free account. It starts of with file attachments and full gpt4o

and you have that for 10 messages

then it's mini

[vapid kettle]

yeah, essentially this

[celest quartz]

Am I the only one locked out of gpt4o?