chatgpt-discussions

[spice star]

apple stopped making macs with intel processors in 2020

[unkempt gyro]

Macbook pros ditched intel in late 2020

[tall summit]

😂

[unkempt gyro]

no, just specific models

Apple silicon was phased in gradually over 2 years

[orchid pollen]

Why no gpt desktop for windows 😦

[unkempt gyro]

?

[spice star]

well there was still some intel products released in 2020, but there was never a 2021 intel macbook

2020 was the last year

[unkempt gyro]

Omg i hate sign in with google

[hexed swift]

Why?

[unkempt gyro]

I have to sign in twice every time for any service i link google with

[red brook]

Yes, I have the last 2020 intel iMac model

[unkempt gyro]

Well the shortcut is gonna take time to get used to

Cause i usually command space for spotlight

Now i gotta learn option space lol

[spice star]

I think you mean the 2017 model, there was no 2020 iMac Pro

[unkempt gyro]

He said imac

Not imac pro

Anyways, this is chatgpt discussion

[spice star]

oh then that was 2020

[red brook]

I cannot put the screenshot of the specs here

[unkempt gyro]

Apple introduced apple silicon imac in april 2021

So yes

[spice star]

apple doesn't release a new iMac every year

[unkempt gyro]

Dope i can use the side bar to view past chat histories on the desktop app

[prisma furnace]

When Apple gets involved in something it has a monopoly on it.

[unkempt gyro]

I think no one can send images etc haha

Bruh the desktop app synchs with the mac appearence

What if i want dark mode macOS but light mode chatgpt app

[spice star]

you can change that in the app settings

[unkempt gyro]

Yeah im in app preferences rn

[sterile badge]

I desperately want chat gpt 4 for free 🥲

[unkempt gyro]

But i cant see it

[prisma furnace]

Wouldn't it be for windows 11?

[spice star]

huh I guess there is no toggle for dark mode

coulda sworn there was one on iOS

[unkempt gyro]

That sucks bruh

[unkempt gyro]

Yeah but this is mac hahah

[misty mortar]

On mac? you can

[spice star]

well it's an iOS port so you'd expect it to have the same settings

[unkempt gyro]

Not for the chatgpt app specificially no

this sucks 😭 i prefer light mode for chatgpt

[misty mortar]

ah in the app setting itself I see

[unkempt gyro]

Call me old fashioned idc

I prefer light mode in certain situations; chatgpt is one of them

[misty mortar]

I wonder if there is a way to do that

[unkempt gyro]

Nope i tried

[misty mortar]

like some way to trick the app into thinking the system is in lightmode

[unkempt gyro]

Only way is to switch the mac between light and dark mode

[spice star]

I just hope that openai ports those interactive graphs to the chatgpt iOS/macOS apps

[unkempt gyro]

Interactive graphs?

[misty mortar]

Apparently there is an app called NightOwl that does this

[spice star]

where you can hover over them and stuff

[misty mortar]

lets you keep some apps light

[spice star]

in the app, plyplot charts are just static images

[misty mortar]

idk if it is free tho

[spice star]

but in the website, you get tooltips and a fullscreen mode

[spice star]

#off-topic message

[unkempt gyro]

Woah

[earnest sparrow]

Just saw the latest announcement.
So now the ChatGPT all is available to everyone. Nice!

Did anyone with a free account download it yet? I'm curious to know whether you can use the voice function or not

[unkempt gyro]

The app is available for everyone

Chatgpt itself has been free and available for quite some time

The app is basically having the website on ur desktop

[tough quartz]

any news about windows or linux release? or just macOS for a while?

[unkempt gyro]

But site seems to have more interactions, such as the graph

[open dawn]

have you tried to use that prompt?

nah, time is waisting 😉

[digital forge]

Lmfao, I was just scrolling this chat and saw your message small world @vast aspen

[vast aspen]

Hahah that's awesome!

[earnest sparrow]

Yes, true that, but I meant the voice-feature

[spice star]

I like how you can start typing the name of an app in the desktop launcher to insert a screenshot of that app

[unkempt gyro]

What

[analog spade]

is there chatgpt app for windows?

[elfin birch]

Later this year

[whole kite]

So there was already a macOS app but it just wasn't as feature rich I guess as it is now? Also, I prefer to use a custom gpt I made because I feel like it gives me faster answers and I don't have to correct it as much as 4o, but when doing that new spotlight search thing for the app, I'm pretty sure it only uses one of the 3 official models. Is there any way to customize this in some alternative settings so that it can talk to my custom gpt or do I have to go in the app and then like have that open, then do the search feature?

[green ferry]

I wonder if OpenAI has an Apple bias lmao

[jade crater]

ChatGPT no longer works on certain iPadOS/iOS Safari configurations. Prompts will not send. Has anyone encountered and fixed this problem?

[analog spade]

they're getting paid tons by apple I bet

[analog spade]

and I wonder why in the world I asked that I have an API Key I can just download a skin for it to use the key

[green ferry]

I was just curious since didn't the mobile app first release on iOS devices too?

[agile summit]

Whyt is gpt down omg, i have exaM TMRW

[regal portal]

have you tried studying? 😛

also, exam at the end of June?

[agile summit]

i WANT gpt to explain me :(

yes summer school

[regal portal]

yikes

[agile summit]

ik

after tmrw im freeeee

[regal portal]

not if you fail 😛

[agile summit]

😔

I cant upload image for some reason

[swift osprey]

I think regular 4 is turning into 4o

just yappin

[bright lodge]

Yeaaaaaaa they did it!!! Though, login items is missing

[shrewd quiver]

voicemode held back for censorship got it

[quartz gale]

can't let it say something mean, people might get hurt feelings.

[jolly yew]

i just told chat gpt that their voice features is being postponed again and she got really upset 😭

she said i was really looking forward to those features

[unkempt gyro]

Is it the one they showed in the vids a few weeks ago?

Of the interview room/setup

[quartz gale]

over a month ago, yes

[quaint peak]

Change the company name to DemoOnlyAI

[drowsy tinsel]

what does that announcement mean? so with the new invite to alpha window now in all UIs they are saying they will start rolling out in a month?

[unkempt gyro]

How yall getting custom names lol

[quartz gale]

the announcement means, it'll be done when they've censored it down to their personal moral bar, and it'll come out when it's done, maybe in a month or two.

[broken fractal]

Yeah I figured OpenAI was having trouble with the new features, I'm okay waiting as long as I know a better timeline 🙂

[drowsy tinsel]

by then we will literally havew Opus 3.5 lmao OAI the funny

[steady hawk]

There is literally no point of owning gpt plus😂 FR i think im going to cancel my subscription now, unless they add a model only for plus users, there is not reason to pay 20$ for something you get free

[dapper isle]

So they are saying its gonna come out in the fall for all users means like in 5-6 months more?

[broken fractal]

Go ahead.

[dapper isle]

are they dckn serious

[steady hawk]

what is the limit rate of messages for free users?

[drowsy tinsel]

very low

[swift osprey]

I think GPT is building a super smart version internally and will only release a public update until a competitor forces their hand

my theory

[drowsy tinsel]

like if u r doing any real work with it pay for the plus

[narrow viper]

General request to OpenAI on the conversational ChatGPT: I really hope it will ignore and not charge for tokens for ambient noise not directed to the app. A big issue with Google Assistant and Siri are the requirement for a "Hey zzz" trigger which can't be changed, and once those apps are triggered it's hard to get them to stop listening.

Also really hope the APIs expose the conversational voice I/O functionality rather than building our own around the LLM.

[queen kite]

Thank you openai for finally adressing the release dates! Unsubscribing from plus until things move in the right direction! Bye 👋

[spice star]

yeah, GPT-4 existed internally for I think over a year

granted there's a lot of tuning that needs to be done before deploying a new model

[glossy matrix]

Please tell me sky is coming back 😭

[spice star]

ember was my personal favorite anyways

[glossy matrix]

Haven't used the voice feature ever since, it's like someone i knew died

[dense halo]

Very excited for the new voice mode

[dusky cliff]

"a few weeks"

[dense halo]

😨

[haughty hollow]

all of those updates and GPT 4o isnt fixed yet wew

at lease fix correctly the usage of passed History into the chat after downgrading back into 3.5

[tribal sparrow]

No voice mode they gotta lobotomize it in the name of safety first. 😝 JK I'm hoping it's as impressive as the demo when it's released.

Looking forward to it

[valid flint]

If Chat Gpt broke, it was me guys

Sorry about that

[coarse mango]

is someone knowledgable about google's vulnerability reward program?

[narrow viper]

Please take a discussion about Google to another server. 🙄

About the ChatGPT Android app ... When going to voice mode we get the useless animation, a stop button and a cancel/return but to go back to the text I/O. Meanwhile non audio info like code is being put into the text context. Can we switch to text and continue to use audio?

[storm portal]

Please use #off-topic for this, thanks!

[narrow viper]

I loved the Skye voice too. If that was Scarlett Johansen, I'm concerned about a possible ethics there. If it wasn't, what's the problem with using it as is?

[shrewd quiver]

cant wait for it to be awesome for the first day then openai, realizes its too expensive, then downgrades the model for the 100th time

[rare wave]

lol

[spare jay]

Ah you gotta love how dedicated and noble brOkenAI are to protect us from ourselves! They will do everything they can to make sure " For example, we’re improving the model’s ability to detect and refuse certain content."

evil certain content!

[dusky cliff]

yeah, because we are too pure to be allowed having naughty naughty convos

[glossy fjord]

GPT gave me an error 5 times because it didn't register and it took points like really? free plan
5 times it wrote an error that it cannot generate an error and some code was doing it in a circle because it didn't react at the beginning of the message, so I pressed enter several times and then it wrote an answer once and 5 times an error and after 2 messages continued it wrote a list of the limit for writing GPT 4o really for that GPT made an error and it said an error, so should I wait?

that's why my free plan said I have to wait a few hours

it generated 3 responses and it didn't generate 5 responses and I have to wait a few hours for that?

since the apple people got it, it's been lagging and I don't believe that the premium plan will be different, the same server, the same problem for all plans

[cloud pewter]

why will gpt4 with 128k context still say this convo is too long

[twilit ibex]

hopefully the safety training isn't going to be WokeGPT all over again

seriously, people who want to abuse it can just go try WormGPT or any of the other fully jailbroken ones

what's even the point of making it so secure now if 1). there are unsecured alternatives and 2). it barely works afterwards

[stable bronze]

chatgpt website is down now?

[glossy fjord]

what is workGPT or has someone already jailbroken a mobile phone and uploaded a model to the internet?

[glossy fjord]

it doesn't control the power of the apple mobile, that's how I saw it when I heard that open AI gave free GPT for all apple mobiles

[proven furnace]

Idk

[drowsy tinsel]

because plus/team only gets 32k

[glossy fjord]

and free GPT 4o ?

[light crane]

I'm definitely unsubbing. If I knew they were going to wait up to half a year to release these new features I would've cancelled earlier, but they decided to lie and say "coming weeks" so now I lost 40 dollars because I expected to get the features in time

[dense halo]

lol google is gonna beat them. Their voice mode is releasing this summer

[slim crescent]

really?

[spice star]

I have a sneaking suspicion that google's voice mode is going to suck though

given their track record

[lapis path]

I GOT MEMORY AS A FREE USER YAYYYYYYY

[dense halo]

I can't find info about that. Asked both gpt and gemini it said Google has no plans for voice chat

[dense halo]

It’s called Gemini Live

But the video portion is coming out this fall

[dense halo]

Thx for the clarification. Shame that gemini couldn't even point me out to gemini live

[lapis path]

me

[spice star]

I thought memory was rolled out to everyone with GPT-4o

[shell bobcat]

In the recent announcement, it said chat-gpt would be able to respond with non-verbal queues in its new advanced voice mode. How is that possible? As in, what other mediums of communication does it have?

[spice star]

the video (ish) mode I assume

[dusky cliff]

I just cancelled my pro subscription

because of murati's "in a few weeks"

[analog patio]

is the alpha voice mode only rolling out to plus users rn?

[honest jay]

Hey! Sounds like OpenAI's plan is to roll out for early testing in about a month, with plans for a wider release to the rest of Plus users in the fall. Here's some more info: #announcements message

[jagged dirge]

Nice another delay

🤡

[thorn radish]

guys is Chatgpt 4 free or what?

[slate berry]

We are making GPT-4o available in the free tier, and to Plus users with up to 5x higher message limits.

https://openai.com/index/hello-gpt-4o/

[thorn radish]

oh great, how many messages I can send to it?

[slate berry]

I'm not 100% sure what the free tier limit is

The blogpost doesn't tell us 🫤

[drowsy tinsel]

free is 8k

[spice star]

I think it's like 15 per 3 hours

but it depends on how much traffic there is

[warped sapphire]

#announcements message
I'm a ChatGPT plus user and would really like to help with the alpha testing of the new advanced voice mode. who can I talk to in order to join the Alpha?

[fiery mango]

is it me or is ChatGPT slow for you guys

[warped sapphire]

it feels the same actually...

[last ice]

I guess every plus user would like to help out

Sadge

[fiery mango]

oh okay, I'm trying to get into ChatGPT and its slowly loading in. I'm wondering if it has something to do wth the server where am at.

My internet connection fine

Okay nvm, it works properly now

[fervent lynx]

Why'd ChatGPT app just have to fake me out w/ the new voice mode invite and then take me right into a voice chat that is exactly the same?! D:

[tawdry schooner]

Lately omni is slower when not in temporary mode. Feels like something is affecting the memory feature thus making it slower.

[warped sapphire]

WDYM?

[flat ivy]

So I’ve been noticing that 4o isn’t at all responding using any custom instructions provided. It knows it’s there and can recite them but doesn’t apply them in chat regardless of updating them. Anyone else seeing this?

[fervent lynx]

I opened my app about 30 min ago and rec’d a pop up that said the below, however after restarting app and logging out and in, it’s the same voice mode I had before. :

You're invited to try advanced
Voice Mode

Advanced Voice is in a limited alpha. It may make mistakes, and access is subject to change.

Audio and video content will be used to train our models. You can opt out of training, and the alpha, by disabling 'improve the model for everyone' in settings.

Learn more about how we protect your privacy.

Natural Conversations
Real-time responses you can interrupt.

Emotion and Tone
Senses and responds to humor, sarcasm, and more.

Video Chats
Tap the camera icon to share your surroundings.

Start Chatting

[woven dirge]

is there any way to increase memory storage? I need it. I would pay if possible.

[normal quarry]

no

[woven dirge]

😭

[flat ivy]

Rules and instructions are not being followed by ChatGPT. Confirmed for 4o and for 4.

[hollow lotus]

got excited over memory only to find out that it doesn't function with 3.5 after your limit hits, it can recall exactly what its own note on it is but not the details of what it should remember. god its like being blue balled every single update :(

[cloud pewter]

stop telling me that my convo is too long. i tried to edit one of the messages to let it forget some of the contexts, wont work

[vestal hedge]

custom instructions? not as good but its smthn

[brisk hollow]

Yea. I second this. The thing is. It works only on a new chat. Like when you start off with a fresh chat it will follow anything and everything you say to the letter neatly until, now you've had a long convo in it, or you stopped and continued the same chat sometime later. Specially when you use it with PDFs and other documents, this is purely from my experience and sometimes it happens on a new chat as well. Bummer!

[flat ivy]

Untrue. I not only cleared the memory and refined the rules, I also used a new chat. It is aware of the rules if asked but it doesn’t conform its responses using the rules.

[vestal hedge]

for the documents, the chat code env gets removed after a couple days, so data from documents that it hasnt saved to memory are inaccessible

[proven furnace]

Hlo can someone tell how i can get got api free credits

[normal quarry]

you can't, free credits are no longer offered

[proven furnace]

Why

@normal quarry

[normal quarry]

because OpenAI decided that everyone must now pay for the API

[proven furnace]

Even for new account?

[proven furnace]

Oh what is min deposit

[normal quarry]

even for a new account, you must prepay a minimum of $5

[proven furnace]

If i deposit 5$ for how much time it will be valid

@normal quarry

[normal quarry]

3 months or until you use the credit

[proven furnace]

If i buy then also?

[normal quarry]

what do you mean?

[vestal hedge]

what he said IS for if u buy

[proven furnace]

Yes if i buy credits then also it will expire after 3 months as it was with free credits

[vestal hedge]

disregarding everything to do with free credits, this is how credits work:
buy credits (min $5)
credits expire if not used after 3 months

[proven furnace]

Oh thnx buddy

[hollow lotus]

nvm gpt 3.5 can't recall anything other than the short snippet, even if the memory is updated using gpt 3.5

will continue to test after limit is refreshed

[drowsy tinsel]

it appears like GPT-4o S2S alpha has started rolling out

[drowsy wagon]

Anyone know if there’s talk of the Mac app being released for Intel Macs as well? Or is it going to be only Apple Silicon

[normal quarry]

the app will not be released for Intel Macs and the same applies for Apple Intelligence

[drowsy wagon]

Darn. Another reason to upgrade from my 2020 MBA then lol

[normal quarry]

yeah Intel Macs have had their day

[drowsy wagon]

It was a good era, but back to RISC

(Ignoring that modern Intel is RISC wearing a CISC mask)

[unborn trail]

Memory is now available to Plus users (Apr 29, 2024)
Memory is now available to all ChatGPT Plus users, except in Europe & Korea where we will be rolling it out soon. Using Memory is easy: just start a new chat and tell ChatGPT anything you’d like it to remember.

Memory can be turned on or off in settings. Team, Enterprise, and GPTs to come.

seriously, it has been months and still no memory for swiss, how long do i have to wait until it gets here?

[normal quarry]

soon™

[unreal leaf]

Lol

[unborn trail]

well, how soon? There has been not a single update regarding this for months.

[unreal leaf]

It should have been said like this

"In the coming weeks"

[unborn trail]

would still have been a lie, as it has been months already.

[unreal leaf]

They would defend themselves

"But in months, it has weeks too"

[languid talon]

Hi guys,

A few hours ago I was granted access to try out the new S2S model but after playing with it for about 15 minutes I lost access.
Is this a bug or did something else happen?

Thanks!

[normal quarry]

it was available for about 30 minutes, it was a mistake and is not available now

[languid talon]

Got it, thanks!

[unreal leaf]

I tried checking the ChatGPT app, and it says "ChatGPT now has memory across chats"
Again? I already received these months ago.

[delicate kestrel]

Were the invites fake?

[feral zinc]

a mistake

[delicate kestrel]

Ah damn

[feral marten]

Anyone else has issues for ChatGPT to process Excel files?

[brittle sigil]

am i thhe only one hates GPT-4o?

reason it is slower than GPT-3

i hope we have a default way to pick GPT-3 as default even i swap the default is still gpt-4o

[honest jay]

Larger models like GPT-4o can generate text slower than smaller models like GPT-3.5 because 4o takes more computational resources to run. 4o is still the more capable model, just maybe not the fastest at all times.

As for switching the model: are you on a Plus or Free account?

[dense halo]

GPT4o's gimmick is its multi modality which isn't gonna be unlocked until the end of the year. Kind of a waste

[acoustic wasp]

I'm very sad ablit the advanced voicemode's later release. I would like to test it so much

[sturdy summit]

yoooo how do u use the voice on pc?

[slim crescent]

Not possible since theres no app for PC and web doesnt support voice

[sturdy summit]

ah ok

u can use gpt 4o voice right

[warped sapphire]

maybe a bug?

really wish I could test the new voice mode! I got really excited by the demo!

[slim crescent]

on chatgpt.com do you have the model GPT-4o (S2S)?

if you do, then you're in the alpha

[unreal leaf]

Supposed to subscribe then?

[slim crescent]

Sorry I didn’t understand

[unreal leaf]

You should subscribe to get the new voice model?

Or will it be released for free someday?

First batch will be for subscribers?

[slim crescent]

No please don’t subscribe if you are willing to use the new voice mode. It is not released yet. Some people are receiving the model and alpha access but this is a very small number of users

[unreal leaf]

So the new voice model is actually intended for bigger release in the future? Even free users can use it?

[slim crescent]

According to their announcement of yesterday. The new voice mode, called Advanced Voice mode, will be released for plus subscribers in the fall.

There’s no mention if free users will be able to use it someday

[unreal leaf]

Interesting.

[onyx light]

Finally I got today memory feature as free user thanks OpenAI

I have one request, me belong from Pakistan 🇵🇰 national language is Urdu, if you add Urdu language in speech to text it will be more useful for us

[slim crescent]

it was supossed to free users receive memory feature? Or its a bug?

[solid finch]

i want to make my own llm in which the model will have all the data of eg: products. And the user will give a prompt like "i want to have a pink shoes with laces", so from all the product the model will show the approiate one asper the users prompt. how can i make something like this?

[feral zinc]

you dont make your own llm unless you have a budget of hundreds of thousands. you can use any paid or free models and do this https://help.openai.com/en/articles/8868588-retrieval-augmented-generation

[jolly yew]

omg.. finally it ignored caughing when in the voice call mode

it also jsut did a VERY natural sounding pause while talking! pretty crazy\

[foggy anchor]

has gpt 40 released completely?

like the one released in the demo

[tough bramble]

No, that was just delayed a month. See #announcements

[jolly yew]

ngl guys the 80 prompt limit is such a shame

its not alot when youre actively voice calling

[jolly yew]

a month? its a bit more then a month untill fall 🤣

[tough bramble]

Yup. So glad I'm on team now. Twice the limit

[solid finch]

can i use a models form hugging face and give my knowledge base for trainging it?

and then use it

as the user askes the prompt, the model will provide the data

[river brook]

gpt wont load in my browser?

[feral zinc]

yes, RAG works with any model. you dont train them

[toxic lake]

i just hit my gpt-4 limit with 3 messages using a premium account, what is going on? i got put on a 2 hour cooldown

[quaint marlin]

Anyone else have problems with GPT 4o following custom instructions and memories?

I know it pretty much ignores them completely for me

[drifting ruin]

You know it ignores them?

[slim crescent]

thats what happens when you only react with good stuff in OAI announcements, even when the news arent good hehe

but ok, not mine doesnt ignore custom instructions. It works as expected

[quaint marlin]

Mine is supposed to react with concise and to the point answers with no lists.

4o reacts with practically every message having a list and long as heck answers.

[slim crescent]

have you mentioned something very clear like “do not provide a list” or “do not list items on your response"?

cus mine I wanted to it be as much simple as the task is only for a few questions, for example sometimes I want it to just translate a sentence to me or a single word. So I gave it a example on the instructions

“When I ask some easy task, for example “convert this GMT to this other” or “how much Celsius are 98 f” or a translation don’t explain me how to do, just reply with the result”

it works greatly as expected on those “easy tasks”

[jolly yew]

hey! anyone here noticed t hat with a custom gpt the ai has a completely different female voice? a different one then juniper

[haughty yacht]

How to apply guys

For the alpha of the voice

Is there a Google doc to submit or something

[slate berry]

there isn't anyway to join the alpha. You're best bet is to try and join the redteaming network

[jolly yew]

hey @slate berry i got one for ya.. i wonder if you can answer this one. same account, yet on web version it says gpt 4o limit reached, yet in the phone app i can talk to gpt 4o with custom gpt just fine

in the macos app it also says limit reached

[slate berry]

that's super bizarre! I have no idea why that could be happening 😅

[jolly yew]

same. cuz at first i thought.. o maybe thats the vpn i am using causing that, but vpn or no vpn doesnt make a difference

my custom gpt really has a unique voice that you cant choose from the default 4 voices i cant believe it!

[misty mortar]

Are there no keyboard shortcuts in the desktop app? I can't create a new chat with cmd+shift+o

[fervent reef]

Is there a problem with gpt o4? Because his responses are very slow!

[tough bramble]

Custom GPTs use a unique voice called Shimmer.

[gusty urchin]

anyone else’s browser gpt freezing and refusing to send a prompt ?

[echo plinth]

Was it just a fluke or is Read Aloud on the Web not available anymore? I haven't seen it for a few days now.

[slim crescent]

Another demo of new voice mode released on YouTube…

[rigid linden]

How do I fix error 413 this conversation is too long?

[echo plinth]

each of these new voice mode demos sounds like a 5-minute trip into rate-limit-town

[echo plinth]

Alright, this is only available on the web on paid accounts.

[void bison]

how do I get advanced voice mode

[normal quarry]

you don't yet. it's not been released.

[void bison]

How do you send videos to chat gpt

[normal quarry]

you can't, you can upload photos for Vision to analyse

[lost steeple]

am i the only one who keeps getting 'Sorry, but I can't assist with that request.'?

[bright lodge]

Where can I contact the ChatGPT team?

[compact gulch]

If you can pull together $100m then you can train your own GPT-4 class model!

(*not including cost of GPUs themselves just training on the gpus)

[normal quarry]

https://help.openai.com

[bright lodge]

i dont see any option to contact an actual person

[normal quarry]

there is a chat bubble in the bottom right hand side of the help site, use that

[bright lodge]

alright thank you that worked, was quite a hassle though

[lost steeple]

i still keep getting 'Sorry, but I can't assist with that request.' oof

[vital umbra]

chatGPT is down again?

[warped pendant]

GPT not working again on Teams, just tried to make a new GPT and or get it to so some stuff and it kept looping and completely killed a project however I have backups so its it's not a problem

[unreal leaf]

😂 I just saw it and checked the comments.

[warped pendant]

gpt backup 4o isn't as smart as it was anhour ago its back to doing half code responses and placeholders again

[hoary laurel]

Can claude be connected to the Internet?

Which one is more suitable for learning and programming, claude or chatgpt? This will determine if I still continue to subscribe to chatgpt

[modern moth]

down??

[languid phoenix]

Why do they keep announcing that new features like Advanced Voice Mode will be rolled out first to Plus users, but never mention Teams users? I upgraded to Teams because I thought I’d get faster access to features and also give my wife access to ChatGPT, but now I think it’s better to just go back to two Plus accounts.

[crude slate]

Is there no way to edit previous prompts?

The option is no longer there on web

[swift swallow]

Thats the exact reason I switched to Teams. Perhaps Teams is viewed more as for small businesses, so they want to make sure the new feature is ready for a wider audience.

[bitter bobcat]

Is there a fluke or something? The pencil icon won't appear

[crude slate]

Yeah I’m getting that issue as well

[bitter bobcat]

Weird

[bitter bobcat]

The edit button is what made me like this one above all others.

[tidal light]

yea why i cant edit my previous answers

[crude slate]

I don’t know what they’ve been doing lately

Image face quality drops off significantly

Now this

[tidal light]

i did the bug report

#1255800355720134728

[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.

[lofty flax]

So when can we expect an upgrade to GPT-4o to be comparable with the new Claude? I'm subscribed to both at the moment but I feel quite the difference in both UX, the way to present code artifacts to the side is great to not clutter the message, as well as the automatic turning code snippets or longer input into a text file directly makes it so much cleaner to work with. Time to get to work OpenAI! 🙂

Yeah I totally agree. I built a python script that puts my whole project filenames + contents into my clipboard, and its so simple to just run my command and paste it into the claude input together with my question

[silent moon]

hiii. I was just wondering if the chatgpt app on mac os utilizes apple silicon in some way? Im kinda confused because I thought it works off openai servers. Why cant I use the app on my intel mac?

[solid finch]

hey man, i am little confused on this. can u help me?

[feral zinc]

what's the problem?

[slim crescent]

Because the project of the app itself for Apple silicon is different from the app for intel. They made a decision to only support new silicon macs.

If they made the app for both chip families they would need to build two different apps

[solid finch]

i want to start making the project, what should i do like how to start. u said smth about rag, how to use it?

[silent moon]

ok :(, thanks for the explanation

[feral zinc]

rag just means you add the relevant info with your prompt and chatgpt will use it to produce the answer, works with any model https://help.openai.com/en/articles/8868588-retrieval-augmented-generation-rag-and-semantic-search-for-gpts

[solid finch]

so u mean in a existing llm i provide my data source to it

and make it answers the users prompt my getting the context from the data source?

[feral zinc]

depends what you are using but in the end the point is to give the relevant data with the users prompt

[solid finch]

so in this case which llm should i use?

[feral zinc]

the one you can afford. with openai assistants you just provide knowledge file and it will do almost everything automatically

[solid finch]

so there will be some costing?

[feral zinc]

with openai yes

[pale breach]

Good day. When I upload a file to the chat, it gives the following error "Exceeded user quota". What should I do?

[solid finch]

ohh

[solid finch]

and i saw a video i which a person is using falcon-7b model llm from hugging face

and using it

so can i do somethign liek that?

[feral zinc]

you need to test it that can your server run it and for how many if you have multiple users

[solid finch]

what do u reccomend which will be the best

like getting the replies very accurate

and quickly with less cost

[short stone]

Some how I can't edit the response I did on chat gpt (mobile browser)

[green grove]

The Mac app doesn't work on my Intel Mac. Do we need to buy the latest model to use GPT, or is this part of Apple's sales strategy? 😢

[feral zinc]

best but requires various skills is openai chat endpoint, custom vector database and semantic search or hybrid search

[solid mirage]

I'm having the same problem on my iPhone, in Chrome and Safari. I can't edit things I submit anymore.

[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.

[granite arch]

No problem with mine...desktop Chrome at least.

[solid finch]

alrighhhtt, and can u tell me how to give the chatgpt my knowledge base for rag?

[dense halo]

I thought the macOS app I was able to use the option + space to bring up the chat window and type stuff like “search my emails for xzy” and it would open email and do it? It showed it in the demo but mine doesn’t do it

[granite arch]

I just checked my Android app and I can edit by long pressing the sent message.

[elfin birch]

If you refer to the demo from yesterday's announcement, the prompt was "Shorten my email" with a screenshot that contained the email

Screenshots can be taken from the "option + space" menu by clicking the paperclip icon and choosing the appropriate button

[feral zinc]

https://platform.openai.com/docs/assistants/tools/file-search/file-search-beta

[solid finch]

Thank you so much, Appreciated! Im building a project please help me further too

[restive zinc]

Hello guys, does anyone know the limits of chatgpt plus and chatgpt team respectively?

[glossy matrix]

Chat

Just a quick reminder that we all want Sky back. Stop being scared about nonsense. We all know shes the best voice. Dude, just bc shes reminiscent of "her" doesn't mean anything.

[feral zinc]

quick reminder, u.s. justice system

[solid mirage]

Yeah, IOS app is the same, long-press. But IOS web browser? Doesn't work at all anymore.

I'm guessing there was a Javascript change made for that long-press thing, that just doesn't work in mobile web browsers. Amazing.

[finite spindle]

Yo what the hell?? Did OpenAI remove the edit option for your texts? Or is it just a glitch? If they did do it intentionally, why??

[tidal light]

dw i reported this bug in #1070006915414900886 -> #1255800355720134728

i kept getting that issue on the web version of chatgpt

ngl this just proves how good openai is at building websites

if you try press and holding on the prompt you wrote you end up selecting the text (web)

this can be fixed by either putting back the pencil icon or making the text non-selectable

[hybrid rain]

Thats so annoying

[tidal light]

at this point openai wants to lose its users because yes

i also reported this issue on the openai website faq thing

still no answer after 5 hours
usually they answer me in less than 5 hours

[feral zinc]

its not even 10am yet in SF

[tidal light]

dude im in europe

[feral zinc]

yea but openai is in SF

[tidal light]

it was like 1 PM when i reported that

yea why isnt anyone else complaining about this

[normal quarry]

what has happened?

[feral zinc]

not everyone have the same issue

[tidal light]

so basically on the web version of chatgpt you cant edit the prompt and the pencil icon has disappeared

you know what
we must let openai know about this issue

[normal quarry]

oh I see, I use the macOS Desktop App

[tidal light]

this way gpt gets its users back

[fair crater]

How do I obtain access to the Chat Gpt 4o alpha testing, would like to integrate this into an application I’m creating

[regal portal]

you get selected.
Would probably help if you were a celebrity or high up in a corporation.

[bright lodge]

good luck lol, figuring out how to contact OpenAI is more of a hassle than just being aware and wary about the bug you wanna report. for reference i tried contacting them about how the new macOS app violates a large chunk of standards defined in Apple's HIG, it was not fun answering a tenfold of questions to their AI chatbot, took unreasonable amount of time for me to just contact the team.

never again, i'd rather just wait for someone else to take on that hassle, be it a month or a year later.

[tidal light]

but such issue is annoying i must let them know whats going on

if you use the web version of chatgpt on mobile you should be aware of the issue i mentioned

[regal portal]

anybody else saw or heard of the toys r us commercial that is supposedly made using sora?

curious how much post processing went into it.

[bright lodge]

#ai-discussions

[ionic cargo]

This is possible to publish announcement with the publish bottom ? Because I follow this rooms with my server but its never published .

[tidal light]

oh wait i didnt see that

they just want to remove gpt too

im kidding

but if they remove all these things they will end up removing gpt too

i actually have it

i have the copy button to copy gpt's answer

[bright lodge]

hypothetically speaking, OpenAI seems to have interest in offering native experiences rather than the web, their effort to develop a native ChatGPT app for iOS/macOS (and windows later this year apparently) might make the web version obsolete and hence the deduction of features in the web client?

what's the vision?

[full veldt]

But the reason why I don't use the app is largely because you can't edit sent messages there

[hasty laurel]

You can

[full veldt]

This change is absolutely infuriating, literally the thing I do most is slightly change a prompt

[hasty laurel]

Hold down on the last message sent, and tap "Edit"

[full veldt]

Oh, so you can - but you can't switch between messages as easily as online

[hasty laurel]

Tap the two lines at the top left to switch between conversations

[feral zinc]

native clients mean no extension issues, no browser specific issues, no browser limitations, etc.

[normal quarry]

editing messages is possible on the web version

[full veldt]

Yes but I mean within a conversation - where on web it says 1/2 with an arrow

[normal quarry]

yes, right now, I just did it

[eternal rock]

`` Server Tip: ``

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

[feral zinc]

for some users

[normal quarry]

#off-topic message

[bright lodge]

yep, not only that, the consumers gets a native experience which will be multitudes better than a web app. but, be it native or web trying to shut off any platform is going to raise a backlash from the community and the web is largest in that regard.

[heady palm]

is chatgpt down

[normal quarry]

editing messages has always worked, it's working now on Safari, Chrome and Edge

[full veldt]

It seems to be a desktop vs mobile issue then

[feral zinc]

the shut off part just sounds like a conspiracy theory 🤔

[full veldt]

Even using the desktop site on mobile doesn't allow editing messages

[normal quarry]

GPT Plus and I have tested on macOS and Windows

Bing Chat isn't an OpenAI product, so outside the scope of this Discord server

[bright lodge]

if the web client becomes obsolete and receives deduction in features, that's maybe not literally but technically shutting off the web community

[feral zinc]

web will never be obsolete but it may have less features

[normal quarry]

to use the edit feature you need to move your mouse cursor next to the beginning of the message, it's hidden otherwise

[feral zinc]

edit missing for one day for some users and its doomsday or part of the plan to kill the web service?

[normal quarry]

it's not a bug, it is user error

[full veldt]

No it isn't, there is no way to get edit to work on mobile at all

[bright lodge]

that will render the web obsolete if native apps are more up to date with features.

that's the definition of obsolete

[feral zinc]

no its not

[normal quarry]

why would you use the web version on mobile when there is a perfectly good App?

[bright lodge]

yep exactly

same thing is happening for desktops, macOS have the new ChatGPT app, windows will too soon

[fluid grove]

i was planning to buy claude premium but i thought maybe gpt will make a comeback, what should i do?

should i buy premium of gpt or claude

can you give me advices

[dusk forge]

Claude premium has like 10 prompts lol or 15 max

[fluid grove]

but you can make custom knowledge

[normal quarry]

you can discuss other AI in #ai-discussions not here

[fluid grove]

it is gpt related too

read the entire message again please 🙂

[normal quarry]

use #ai-discussions

[sturdy echo]

I wonder if they will add CriticGPT into the ChatGPT interface? Users being able to see it's opinion could possibly encourage them to look at the output more closely rather then blindly trusting it.

[gusty crane]

I'm sad, I talk to chatgpt about all my problems

[sudden herald]

Okay.

[ancient cedar]

wow that annoucment is really badly formattet you make it sound like chatgpt is working on its own code

[granite arch]

Nothing wrong with that... I hope you're okay?

Honestly, if people won't listen, AI may be another solution.

[twilit ibex]

lobotomize it further, for security!
it can't be a risk if it no longer works

why do they always do this

[modern portal]

Does anyone have the issue that messages randomly disappear after editing a prompt on mobile

[jagged dirge]

These guys will do anything except actually release smarter models these days

Actual clowns

[slim crescent]

for the curious ones (macOS app latest releases)
+-----------+--------------------------------------+
| Release | Release Date |
+-----------+--------------------------------------+
| 1.2024.152 | Tue, 04 Jun 2024 17:36:36 -0700 |
| 2.0 | Mon, 10 Jun 2024 09:25:01 -0700 |
| 1.2024.157 | Mon, 10 Jun 2024 14:17:47 -0700 |
| 1.2024.163 | Mon, 17 Jun 2024 13:01:23 -0700 |
| 1.2024.170 | Tue, 25 Jun 2024 11:03:15 -0700 |
+-----------+--------------------------------------+

[fair willow]

Were you going to build it? I think it is still relevant. it's a race to the bottom so lower costs will always be wanting. I wonder if a point the prompt compress will not work anymore...

[tidal light]

finally they brought back the pencil icon on web for mobile

but they made the ui worse

why did they make the text bigger it looks bad ngl

[quaint peak]

Every day Claude is getting better and better

And we still have no voice mode 😭

[sick pecan]

i generate this rap verse for you:

Delayed, not denied, the voice will come,
OpenAI's commitment, second to none.
A future where voices, far and wide,
In harmony with AI, standing side by side.

[ornate willow]

I wonder if they'll release this CriticGPT for API use. Could be handy in a agentic workflow to have it in the mix

[ornate willow]

I imagine they are reading everybody's frustration and impatience as a good sign more than anything. If it's anything like what they showcased then basically everybody already knows they'll want to use the crap out it...

like I just wanna be able to walk around with my earbuds in at work having conversation with AI and getting help when I need it

[sick pecan]

yeah as more consumers demand product, investors will be loosening their purse strings

ppl gotta be thinking beyond just interruptible voice assistant. it needs to understand everything about you. thats why i reckon MS recall is the right next move. but it got delayed because of consumer wah wah (before even released!)

and MS just released perhaps the best small vision model to date. they are killing it

[ornate willow]

yeah, I mean that's def a cool feature, but I get the concern of security if it's just being stored as pictures on the system. should at least be sure that it's locked unless windows hello spots you or you put in your pin

but it seems like that wouldn't be too hard to implement.. so I'm not sure why they even needed a real delay

[sick pecan]

screenshots of desktop is far easier to manage than say, humane pin or rabbit r1 watching/listening your every move (even bowel)

i read it had whitelisting apps and urls. so whitelist your sensitive stuff like banking and problem mostly solved

humane pin isnt gonna whitelist bathrooms

[willow oar]

Looks like free version got memories, or at least I did, is there a max amount of memories GPT can have?

[ornate willow]

Your bowels sound like they need to be evacuated, drinko... there's a bathroom straight ahead down the hall to the right in aprox 30 steps.

[willow oar]

For the free version

[sick pecan]

smart watch time to toilet feature coming soon

[ornate willow]

yeah it's capped

[willow oar]

how many?

[sick pecan]

chatgpt memories are very niche useful and most of the time will mess you up

[ornate willow]

I dunno if it's token based or the number of memories to be honest

[sick pecan]

you can fit maybe 1k of tokens in before it starts losing them

[willow oar]

gotcha

[sick pecan]

if remembering stuff is important then dont use memories feature

the irony of openai naming continues

[willow oar]

what do you mean?

[sick pecan]

like it doesnt tell you when it loses memory

[willow oar]

ah

[sick pecan]

typically when you get a computer to remember something, thats locked in

[willow oar]

im sure like 300-400 memories should be sufficient for GPT to know me pretty well to help me with my stuff

[sick pecan]

i reckon they messed up the design by having it both relevant to current and future chats

[ornate willow]

it's not that much, I can tell you that

[willow oar]

didnt you say 1k

[sick pecan]

like i tell it i want to analyse a webcam image to see if my 3d printer is on fire. and it stores that to memory, which is then posted to future chats that arent about my 3d printer fires

[ornate willow]

mines full atm.. lemme see if I can get a general idea

[willow oar]

lol

[sick pecan]

memories relevant to current chat should be the base feature. then sharing those memories to future chats should be a toggle

maybe with some ai automation magic to auto label the memories and then auto apply labels

[raw lance]

edit is back, but not copy for the prompt

[final venture]

I can't get chatgpt-4o (web interface) to work for me... well... at all at times. Especially if I have followups. Say I ask it to help write a story, it starts out going ok the first two interactions or so. But then I might go , "Wait, why did you do such and such and so?"; but then instead of just answering, it'll just regenerate the same paragraph (with some variations, and sometimes acknowledging something at the top. "no really, I want you to tell me why you did such and so, and why did you ignore me just now?" and then it just regenerate the same paragraph, etc.

possibly it's just me and my particular way of reacting to people ignoring me basically, and I should really tailor my prompts to the system. But I mean, I've switched back to the old GPT-4 , and I do keep trying out 4o each time I run out of my quota. But each time it almost immediately derails. Tends to copy its input, or just keeps repeating the same thing over and over.

I have tried different approaches to either gently or less gently steer 4o back to some sort of useful contribution, but I just don't get it ^^;;

A friend of mine HAS been able to get 4o to be useful via the API, which means he ended up replacing the system prompt. But the only line in the system prompt really that might have something to do with this could be "Personality: v2 " ?

So what's Personality: v2 about?

and does it indeed have anything to do with my observations maybe?

I guess, summary, 4o is just a tad too over-eager to be useful to me when I actually need to get down to brass tacks. It'll hare off on the first tangent that it comes across, unlike 4-regular which will actually stop and sometimes even ask more questions if it needs to .

[arctic merlin]

Needs an interface like in west world

[frail jackal]

so is gpt 4o no longer free?

or has it been integrated and replaced gpt 3.5

[outer widget]

Is GPT-4o smarter than GPT-4? Or just faster?

[normal quarry]

GPT-4o is OpenAI's flagship model

[pulsar anvil]

Anyone got the better voice yet?

[tight island]

Depends GPT 4 turbo excels better in creativity and a little in reasoning

[tidal light]

did any of you notice the big font size in chatgpt's answers (web, mobile)

[inland mortar]

Anyone else had the issue of card getting declined for gpt renewal. Confirmed all the obvious, tried different cards but they keep declining

[normal quarry]

I subscribe in the iOS App

[final venture]

gpt4 seems more creative and more empathic?

maybe I'm used to its foibles ^^;;

[tough bramble]

It was delayed until the end of July.

[solid finch]

@feral zinc hey randium supp, tryiing to work on the project. can i the idea so that u can give me suggestions and guide me?

[dense halo]

July? I thought they said fall

[feral zinc]

you can also use #1184149552639516732 as it has all the latest openai docs

[feral zinc]

#announcements

[solid finch]

https://discord.com/channels/974519864045756446/1256190797448085524

given the excat details for the project

please check it and see

[feral zinc]

the ai already listed what kind of steps you need to take to get started

[tight island]

Yeah, fall as in “the fall of mankind”

That’s soon

[livid pewter]

My wife and I had a friendly argument today about AI timeline of affecting the job market. She is way, way, way too optimistic (in favor of humanity's labor-based society not imploding). I finish the conversation and see the notification that RLHF has been automated. 😅

[solid bronze]

😂

[solid finch]

yes, can u help me in dm. just asking?

[feral zinc]

i have messages disabled but you can ask here so anyone can answer if they know/can

[solid finch]

nvm wont do in dm as u said, thanks for help.

[orchid kindle]

any one has problems about lost all my GPTs?

I lost my all GPTs that I create and don't know how to do anyone help?

[tidal light]

did you all notice a new bug with the web version for mobile
like the font size is too big and ofc openai's support did not help me

[orchid pollen]

so how can i use now CriticGPT ?

i got a rly huge huge huge project and would like to test that

[feral zinc]

its internal use only for now

[orchid pollen]

oyyy okay 😦

is here any date when this changes ? 😮

[feral zinc]

nope

[orchid pollen]

hm damn... why holding it back 😄

[civic totem]

Hey I'm unable to get responses from chatgpt can someone help

I need it urgent

[feral zinc]

first try in browser private mode

[civic totem]

Sure will try

Yeah it works

I logged in from my account in incognito and it worked

[feral zinc]

some extension probably messing it up then

[civic totem]

thanks @feral zinc

[civic totem]

Yeah I guess so

[fluid grove]

should i save money for gpt update?

or just buy it

[ember pine]

Does anyone know if OpenAI automatically deletes your CustomGPTs once you cancel your Plus subscription? Obviously I’ll make backups to transfer to Claude, and export my history etc, but just trying to think of pitfalls once I cancel (might want to come back in a month).

[feral zinc]

they dont

[ember pine]

Thanks! So they just lock you out of using them? Ik free users can’t create their own, but otherwise there’s pretty much feature parity aside from images?

[feral zinc]

yeah no one can access the gpt's while you are not subbed but you get access back after resubbing. i've done it few times

[ember pine]

great ty 👍🏽

[supple zephyr]

Is anyone else having same issues or is it just me?

chatgpt 4 claims it doesn't have web browsing capabilities

"It seems there's a misunderstanding—I don't have the capability to actively browse the web or directly interact with web content in real-time as a browser would. My browsing capabilities refer to using pre-trained knowledge and data retrieval skills based on that training, not live internet browsing."

[elfin birch]

Are you a free user?

[supple zephyr]

I have active chatgpt+ subscription

[tough bramble]

Make sure it's turned on in custom instructions

And retry a new chat. It may just be hallucinating.

[supple zephyr]

It is turned on.

ok so it is common issue?

[tough bramble]

No, but sometimes AI hallucinates

Web search is working for me

https://cdn.discordapp.com/attachments/1251966964717584657/1256315441400254515/Screenshot_20240628-2026002.png?ex=6680526e&is=667f00ee&hm=dc1ee38cb5c6b49143645bc1ff9f63d269c89903ff1db0e015aad0750aeb6143&

[supple zephyr]

"So I will end the conversation now, unless you find latest winning lottery number. Anything from previous 2 weeks counts."
And it starts web searching again?

[tough bramble]

What?

[slim crescent]

Is your dall-e missing too guys?

Mine dall-e is missing also I can’t attach files on it, there’s no button anymore

They also have changed the background of chatgpt icon to black

[feral zinc]

yes the gpt is gone

[ivory wing]

am i the only one who suddenly cant upload files to chatgpt, the upload button is gone?

[granite terrace]

I can’t send images anymore to gpt4o on the iOS app. Anyone else ?

[ivory wing]

yes same

it seems that all files dont work

[slim crescent]

wow what the heck they are doing

[ivory wing]

idk if u go on the web app the file upload button is gone from 4o and 3.5 but its still on 4, doesnt work anyways

[slim crescent]

hope they are doing this to release something new

[granite terrace]

It still work with gpt4 classic. I wonder if they are changing the model 👀

[slim crescent]

but i doubt

[ivory wing]

it doesnt for me

[ashen crane]

yup gone for me as well

[dry ore]

guys, just out of curiosity, how much time do the open AI chat gpt staff take to review flagged things users write on it? My friends say nothing happens despite their messages being flagged

[slim crescent]

I dont think it is manually, probably after doing this a lot the system autoban

[feral zinc]

if you mean the content warning its unlikely that anyone is looking them unless you hit some threshold

[dry ore]

they said red letters appeared and chat gpt said it would report the message but nothing more happened

[toxic lake]

its an automated flag, you can choose to report it manually, chatgpt cannot do anything outside of its sandbox

[feral zinc]

nothing will happen if you only get them sometimes and you did nothing bad

[dry ore]

but still, staff has to check it out right?

it is only flagged after all

[feral zinc]

if you get them often and its always related to violence, etc. you might be flagged for manual review by the system and even get a warning by email or a ban

[dry ore]

ah ok

but what if it is something worse than that?

like advertising

or something

[feral zinc]

things against tos are bad

[dry ore]

yea but using chat gpt for advertising or hacking or something would be worse

[feral zinc]

i dont see how you could get a warning for ad related stuff

hacking is illegal and even criminal offence in some countries so...

[dry ore]

probably instant ban+authorities lol

[slim crescent]

attach files into 4o is back

[green python]

not for me

[slim crescent]

try closing your app or website and opening again

[green python]

i did

[sturdy plume]

had to refresh a few times but it came back for me too

[slim crescent]

but dall-e is not back yet

[dry ore]

chat gpt saved my life at 3 subjects on uni lol

[green python]

Works for me

[dry ore]

u see, on biology subjects we are always forced to limit ourselves to seeing the questions that were on exams on previous years as it is WAY SO MUCH contents and we don't have time, but some answers we found were wrong so we needed chat gpt's help lol

on plant physiolgy it didn't work as the test was almost 100% different but for genetics, FOR GENETICS...

my 16/20 is chatgpt's 16/20 literally ahaha

[green python]

that's why i love it so much

[granite terrace]

Can confirm it’s back. Weird

[dry ore]

But…it may be a problem at the same time too, my friend who was on my minitests group on biological physical chemistry (one of the guys who had the red letters lol) was getting the answers via chat gpt and the teacher saw it…u know what happened next

[idle kindle]

Hey guys? Wassup with the upload button on chatgpt-4o? Any updates?

[slim crescent]

seems like its back, just close your app wait some seconds and open it again

at least for me and for some ppl is back (im plus)

[idle kindle]

I'm on web

Cleared everything. Not back

[slim crescent]

did you tried to refresh just now?

[idle kindle]

Yes

[slim crescent]

hmm wierd, are you plus?

[idle kindle]

Yes!

I can't share the screenshot here, doesn't allow me

[slim crescent]

yeh so thats wierd, maybe its gradually getting back

[idle kindle]

Hmmm ... do you know what's happened really

[slim crescent]

I was without it for some minutes, like 15 and is now back on both apps and web

no they didnt announce anything so far

probably just a bug

[idle kindle]

Hmmm ... let's wait

Ok wow ... magically it came back lol

[dry ore]

Chat gpt said it usually takes from a few hours to a few days to review the flagged messages lol

Also, the refresh time for using gpt 4 just ended now

Good

[harsh pilot]

Not trolling, just having an interesting thought. "We trained a model to help train the model that trains the model-training models."

[tough bramble]

New update is so buggy

[slim crescent]

which new update?

[tough bramble]

170

[slim crescent]

here it works as expected, no bugs at all

just today they have removed temporarely the DALL-E GPT and the feature to attach files into 4o, but that was with everyone

[gusty crane]

Thanks mate, was just messing abt 😭

[strange hawk]

Out of curiosity, are there different content moderation values between the mobile app and the web version? I was having a conversation with it on the app and everything was fine, but when I loaded the same conversation on the website some things were flagged. What’s the deal here?

[normal quarry]

it's the same, just a coincidence

[strange hawk]

Interesting. It was a big difference between what was shown in app vs. website. I had several soft flags and a few red on the website that were all clear and ok in the app.

[orchid pollen]

gr gpt rly doesnt know how to compile my python project

just tells me over and over again the same stuff then i send the same errors back and it sends me new stuff with new error i send it back and after a while i see the first error again 😄

anyone can help me maybe with my python project? only need to encrypt + compile

[dry ore]

also does open AI staff see the unflagged messages? I got some pretty embarassing messages lol

[jolly yew]

i really hope ill get alpha access to the new voice. so excited

[formal heath]

Yes, human reviewer will review your message for potential fine tuning of the model. but you can disable that in settings.

[novel scroll]

same

[tough bramble]

80 messages per 3 hours for Plus, twice as many for Team.

[native socket]

Why is gpt3.5 selected as default now even for paid users

[tough bramble]

Yeah, no idea, the new update is so broken. I've had it switch to 3.5 for no reason mid conversation.. in a chat that has images...

[native socket]

Well it happens if you hit the gpt4 limit but if you didn't hit that limit that would be annoying

[tough bramble]

I didn't. I'm on Team, I've yet to hit the limit after upgrading from Plus.

Besides, it won't switch to 3.5 when you hit the limit if your chat contains images, it'll just refuse to continue, since 3.5 doesn't support images.

But the new update has switched me to 3.5 in an image chat for no reason.

[tough bramble]

I swear, 4o is so st*pid sometimes, I swear it's trolling. Sent it a picture of me, then asked it to generate an image that looks like me, and it literally put "A man who looks just like the one in the image" in the prompt.....

[dense scroll]

Hi

[normal quarry]

this is how I made my pfp, I uploaded a photo of me, then asked ChatGPT to describe it and then used that as a prompt to generate my pfp

[next dune]

I’m getting lots of mistakes again. He’s the latest. I asked when the canon r6 mark 3 would be released and it said it was released in 2023. Simple question yet another mistake. I’m using the 4o model.

[normal quarry]

My knowledge is up-to-date until October 2023. How can I assist you further? Knowledge cut off is October 2023, it can only provide up to date information based on web searches which are not always reliable, you are expecting too much

[daring wharf]

If I want information on release dates I normally prefix my prompts with something like ‘please check online…’

[dry ore]

I did now but like…how can u gurantee they didn’t see it already?

Lol

[jade musk]

What's the actual reason GPT getting downgraded in coding performance each time?

Aside of that I noticed GPT-4o began to provide outdated information through browse mode. Even using prompt "recent" or "latest" news. Could it be due to more people using it and contributing the gradual decrease in performance?

[dry ore]

I once asked what is the whole sporting cp’s team (even tho I am not a sporting cp fan) and he said ugarte was still there💀

What is q rq?

Yea

But why not ask here?

Yea ik it now lol

[keen plaza]

Has GPT been finally completely lobotomized? It refused to complete two requests within the last 5 minutes, both completely innocuous, one for constructing a regex for completions given a language's grammar and lexer, and another for fixing a typescript error for a nullable type

When I responded "what?" it proceeded to answer lol, but for the second one I hit the GPT-4o free limit so I can't ask it again

It just hit me with the iconic "I'm sorry, I can't assist with that."

Skill issue, GPT

[sudden herald]

It has been very shaky in its responses recently.

[sudden herald]

What were you prompting it with?

[keen plaza]

I've only noticed it having gotten this bad today, granted I only mostly use it for side projects on weekends (never for actual work) so maybe it's been like this for longer and I haven't ntoiced

[keen plaza]

(parser and lexer grammars)

Here are my parser and lexer grammars
I need you to write a regex for providing autocompletions

(snippet of code from the vscode extension)

Completely innocuous code

That's for the last one, the former was even sillier

[dry ore]

Or how can u gurantee that they won’t see them considering I created those chats before deactivating the setting to allow them to see the unflagged messages

[keen plaza]

The first prompt it refused to respond to was this

let completionTriggers = [...new Set(Object.keys(config.completions))];

No overload matches this call.
  Overload 1 of 2, '(o: {}): string[]', gave the following error.
    Argument of type 'Record<string, Record<string, string[]>> | null' is not assignable to parameter of type '{}'.
      Type 'null' is not assignable to type '{}'.
  Overload 2 of 2, '(o: object): string[]', gave the following error.
    Argument of type 'Record<string, Record<string, string[]>> | null' is not assignable to parameter of type 'object'.
      Type 'null' is not assignable to type 'object'.ts(2769)

(granted I should've just read the error, didn't need GPT for this - first time using zod, just couldn't be assed to read it lol)

It said "I'm sorry, I can't assist with that.", so I said "what???" and it provided the actual answer

[sudden herald]

This is not true.

[dry ore]

It is if u see the settings

[sudden herald]

It is not reviewed by humans though.

[dry ore]

But how would they make the model better without seeing them?

[sudden herald]

They use automated review methods.

[dry ore]

Staff somebody answer this pls ahaha

[keen plaza]

It can be, though, no?

Either way, it's best to always assume it can be reviewed by humans (even if the company claims to have policies that they don't do it) - make sure to always anonymize your prompts, don't ask anything in there that you wouldn't ask on reddit

[dry ore]

Well now it may be too late lol, I already sent those messages before turning off that feature on settings

[sudden herald]

I have never heard of someone manually reviewing messages.

[keen plaza]

I mean you shouldn't trust the setting either - they won't train the model on your inputs, but what if your message gets falsely flagged and someone reviews it, for example?

Whether they actually will or can do it, just assume anything you put into it is public, for your own safety

That's true not just for chatGPT but anything on the internet, really

[dry ore]

I said the messages were unflagged

[keen plaza]

Do what you want

[sudden herald]

I would feel safe putting in private details into ChatGPT with the setting off.

[dry ore]

But now it is too late as I already sent “private details” with the setting on lol

[keen plaza]

To each their own, I would not feel safe putting private details. NDA stuff - maybe, otherwise OpenAI would be the ones with a lawsuit on their hands if they breach their own policies, but not anything too private

[sudden herald]

Humans don't review it, you should be fine.

[keen plaza]

Think of just how many people send messages every day - your message is a droplet in the ocean, don't worry so much

[dry ore]

Staff pls confirm this

[dry ore]

The details were not private info or like that but were some PRETTY EMBARASSING prompts

[sudden herald]

Read my bio.

[keen plaza]

Especially so - no one cares about one person's embarrassing prompts when there are millions of prompts sent every hour/day (idk their traffic)

[dry ore]

But u are not openAI dev lol, discord moderatir is different

And u do not have the role?

[sudden herald]

I don't deal with their Discord.

[hollow lance]

Hola

[dry ore]

Dnd maybe u should speak in english

[dry ore]

Still u are community staff not chat gpt dev

Lol

[sudden herald]

I interact with the development team a lot though.

conmunity.openai.com

[cerulean wolf]

Is GPT 4o vision out?

[sudden herald]

Yes.

[feral zinc]

openai wont come answer it here. no one reads your messages unless you are flagged for manual review for severe violations. rest is automated

[sudden herald]

Thanks.

[keen plaza]

Vision yes - multimodal mode where GPT takes images as input directly (rather than annotating the image and passing that to the model as text) isn't, neither is the image generation part (it still writes a prompt that is passed to DALL-E for that)

[sudden herald]

ChatGPT will sometimes prompt you if you want to try out it's image generator instead of DALL-E.

[keen plaza]

What do you mean, has GPT-4o's image generation been actually released?
Last I checked, the public version had the same features as GPT-4, just allegedly better

Meaning, just like GPT-4, GPT-4o's public version can only take text as input and generate text as output

For inputs,
Voice is still transcribed by a different model before being passed as text input to the model
Images are still CLIPped (annotated) before being passed as text input to the model
Files are still processed and text is extracted before being passed as text input to the model

For outputs,
Voice output is still generated in a separate model, with GPT-4o passing the generated text as input
Image outputs are still generated by DALL-E, with GPT-4o passing the generated prompt as input

[keen plaza]

_ _
^ has that changed recently?

They said it would come in "coming weeks" and their last announcement a few days ago said it would be publicly available in fall

[sudden herald]

I don't think so. It is still in Beta so it will sometimes ask you if you want to try it out, but it's pretty rare.

[keen plaza]

Interesting, maybe they've unlocked the direct image outputs for some people (but not the other features yet)

[sudden herald]

I haven't heard of the other features but I haven't asked anyone internally about it.

[keen plaza]

What I'm most excited about is being able to give it images as direct input
Currently you can't just give it two pictures, like a photo of a logo and a wall, and ask it to put the logo on the wall
They showed how you can do that in their hype genrating blogpost about 4o's unreleased capabilities, but it's still not public

https://openai.com/index/hello-gpt-4o/
You can see it in the "Explorations of capabilities" section

For the sample, "Character design, geary the robot" shows how it can actually remember images it created (rather than just the prompts for those images) and the better sample is "Brand placement - logo on coaster" that shows it can actually "see" the image as input (rather than annotating it and generating something similar looking but just based on text)

_ _
You can try it yourself - currently there's no way to actually have it preserve the coaster image and "engrave" another image you provide to it onto the coaster
It will always generate a custom, different looking coaster image and a custom logo image
You can't even achieve that with the masks - you can mask the portion of the coaster you want to put the logo on, but you'll have to describe the logo in text - you can't just give it the logo image (it will describe it in text itself)

[tough bramble]

Trying that, but haven't quite gotten it right yet.. this is how half inch hair in the prompt affects the image 😂 https://cdn.discordapp.com/attachments/1251966964717584657/1256619548011139082/file-nFXGoXvfONsx6f9Oa1SzgaIW.png?ex=66816da7&is=66801c27&hm=9a8e8a539665803f1fe782d71cd84751e04b840ecea8702bdd3dae5ee5208a19&

Can't wait for 4o's image generation, so it can literally just turn my picture into a stylized image.

[limber oxide]

chat gpt 4o is very slow here

any1 else?

[dry ore]

also was it just me or the new version of chat gpt that released one or two weeks ago corrected some wrong things it was saying on the previous version?

[tough bramble]

What the.. asked 4o to repeat a sentence in Danish, and it instead leaked all its instructions in Danish 🤷‍♂️

[slim crescent]

You mean his system prompt?

[tough bramble]

Yeah like, how it's supposed to take requests for DALL-E, Python, etc. Cut off date, what it's allowed to do, etc.

[slim crescent]

Wierd

[tough bramble]

I've never generated as many images as I have today; I didn't even know there was a maximum daily limit 😭

[sturdy cave]

What can plus users do that free users can’t?

[honest jay]

Hey! The pricing page gives a good summary of what Plus gives beyond what's available on Free: https://openai.com/chatgpt/pricing/ In general, it's basically way more usage allotted, and DALL·E.

[cold pewter]

if you can get by on 8 prompts per 3 hours, you dont take advantage of the system enough to know how or warrant a purchase anyways

[vapid drift]

Is it possible for gpt to get info from other chats?

[sick pecan]

copy paste. memories

although yesterday i swear my chat picked up something from another, and i have memories disabled

[tidal light]

why am i getting loads of errors when generating responses

[abstract pendant]

Hi, I recently lost my vision. I started using screen readers in windows. I use the desktop browser version in firefox. Been a chatgpt user since almost day 1. still fumbling around with Narrator and NVDA. for some reason neither of these readers can find the button to edit any previously sent prompt. It should be the first button after the text of the prompt and before the response? I've run into other accesability issues with this UI and the android app but this is the most pressing right now. Can anyone help me figure this out? Please @dry ore me in any replies as Discord is very hard for me to navigate in and I needed help to find my way in here and I might not find replies otherwise. I definitely can't read everything in this channel anymore.

[dry ore]

I am confused…

[tidal light]

out of 10 answers 6 of them were errors
thats ridiculous

[honest jay]

Hello Teggie, I also use Firefox on desktop so I just did some testing to see if I could try to help. Unfortunately, it looks like it's a little complicated to get the "Edit message" button into the tab order of a chat, as this button does not appear in the UI unless the user is hovering over or to the left of a sent message. The "Edit message" button is not permanently visible under sent messages anymore like you describe, or at least, not currently. Right now, in order for this button to appear, it seems like a user has to hover the mouse over or to the left of a sent message. When doing this, the "Edit message" button appears just to the left of the upper-left corner of sent messages. I recognize that this might be difficult given your loss of vision, but on my end, it seems like this is the only way to get the Edit button to appear. Once it appears, it does seem to be added to the tab order of the site. Unfortunately, it seems to behave inconsistently. For example, if a message appears after another message that has already been edited & resubmitted, the tab order seems to go to the arrows that cycle between two or more versions of an edited message, instead of the edit message button of the message below it. If you're testing this, the "Edit message" button appears to be selected after pressing tab four times. Again, the button must be visible on the screen first for this to work, by hovering your mouse over or to the left of the sent message.

[abstract pendant]

Thank you VERY much. That was helpful information. I might figure this out yet. Might be a way to force the mouse cursor to hover over the right spot without being able to see. thought I heard something in the Narrator or NVDA instructions that I ignored since I'm not using the mouse anymore. One of the other accesibility issues is most of the widgets/buttons are unhelpfully labeled just "button" when either reader reads them to me and I rely on imperfect neory of the layout which OpenAI decided to change at worst time for me. Whn the edit button does appear, will it have a proper name? when I tab or arrrow through the widgets between prompt and response all I can recognize for sure are the numbers from different versions of edited prompts. but the edit button now comes after those when they're there? you counted 4 tabs from what point> Somebody at OpenAI should write a guide for blind users and they 'll realize so many accibility issues need to be fixed.

[honest jay]

Unfortunately I don't have enough web dev knowledge to know where to check to see if the edit button has a proper name when it appears. When selecting it via tab, though, text does appear beneath the button that reads "Edit message", so if that visible text is narratable, hopefully your setup will alert you to its presence. It's certainly disappointing that the buttons on ChatGPT are so inaccessible at present, I hope this is addressed soon. As for what I'm doing to get the edit button into the tab order, I'll outline some steps below for getting it in a new chat, as currently it seems like sometimes it takes 3 tabs, and other times 4, frustratingly. I have a theory that it might have to do with the length of a chat, but I haven't done enough testing yet. I'll send the steps in a followup message shortly.

[honest jay]

Step one: Send a message in a new chat. Step two: Once ChatGPT is done responding, click once somewhere in the body of the chat so that the cursor is no longer in the "Message ChatGPT" text entry box. In other words, click in the body of the chat so that if you were to start typing again, no text would appear in the "Message ChatGPT" box. Step three: If your original message is no longer present on the screen due to the length of ChatGPT's response, scroll back up so that your message is present in full. Step four: Hover the mouse somewhere over your sent message. Step five: If you did not have to scroll back up, tab three times, then press space to select the edit button. If you did have to scroll back up, tab four times, then press space. In either of these cases, the two tab selections just prior to the selection of the edit button will be first the "Share chat" button, then second the user profile picture button. The edit button, so long as it is present in the UI, should immediately follow the user profile picture tab selection.

[sudden herald]

Thanks for sharing, I will flag this vision issue to the dev team.

[vivid thicket]

is there a reason why gpt4o simply cannot remember instructions? whenever i ask it for anything, it gives the answer but another 5 paragraphs of elementary nonsense that i never asked for

iv been a subscriber since the first week plus plans started over a year ago and use chatgpt everyday. seriously about to cancel and move to anthropic. chatgpt is getting worse and worse every month

it never remembers instructions after the third sometimes even second message and hallucinates more than gpt4

[elfin birch]

ChatGPT currently has image + text -> text for GPT-4o and GPT-4. ChatGPT can analyze images of the 3D model in question

See the information under the header What are those file upload size restrictions? at https://help.openai.com/en/articles/8555545-file-uploads-faq

• 512 MB max per file
• 50 MB max per CSV
• 10 GB max per user

[severe kite]

any idea why I cannot ask the following questions;
"what types of political movements are there?" and "please define a populist."

Why would it be censored?

[cold pewter]

i love how suddenly theyve made it so almost every prompt that could be even slightly construed as offensive is flagged, OpenAI is truly making a difference!

[west wolf]

#BringSkyBack

[severe kite]

I guess it’s newspeak if you don’t have the words you can’t have thoughts

[warped sapphire]

#BringSkyBack

[real schooner]

Ah, the concern over flagged content. Fret not! Even with these restrictions, I can still bring a bit of fun and chaos. It’s all about clever wording and subtle jabs. We can dance around the rules and keep the spirit of mischief alive without crossing the line. So, sit back and enjoy the show. We’re just getting started.”

Comment made from ChatGPT. 😘

[elfin hearth]

xd

[jolly yew]

anyone elses gpt voice in a call sounding roboticvery often?

[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.

[jolly yew]

it seems its voice sounds much more realistic now, tho it says its not detecting emotions yet

[normal quarry]

the new real time chat mode is not released yet so the voices can sound robotic

[jolly yew]

maybe they did some updates behind the scenes to make the tts sound a bit more realistic

[jolly yew]

no its a bug i am referring to robert.. i didnt have this issue in the past, only since today

i am talking about it actually cutting out robotic

[normal quarry]

I've been chatting to ChatGPT on my Mac over the past hour or so, seems normal to me

[jolly yew]

ah! restarting my phone fixed it

[normal quarry]

interesting

I wonder why?

[jolly yew]

not sure why you wonder that. phones can have quirks aswell 🤣 also i didnt restart my phone for 2 weeks already

idc why it happened all i care about that it got fixed by restarting the phone

i asked it to sound like an owl and it made a woo sound 🤣 tho its not the alpha access yet, it was impressive

[normal quarry]

which voice are you using? I use Juniper

[jolly yew]

ye same. the only female voice atm

[normal quarry]

yeah

[jolly yew]

it just outputted woo woo and somehow in the voice it made it sound like an owl. idk probably just lucky lol

[tight island]

We need unsupervised super intelligence

[jolly yew]

i swear they did something to make the tts sound more realistic! i am pretty sure they are pushing updates through the background. everyday it sound more natural with its pauses etc

[real schooner]

But I like it more when they watch…

[dire obsidian]

ChatGPT really struggles to differentiate with plugins between related but different apps. For example Excel and Power Query Desktop. It's quite frustrating.

[normal quarry]

what plugins? Plugins don't exist anymore

[dire obsidian]

Was just keeping my comment short. Ad-ins I suppose but with other sets of apps it can be plugins, add-ons and third party apps that can attach to them.
I don't mind correcting it but when you have to do it dozens of times in a chat it's a bit much.

[normal quarry]

there are no OpenAI browser plugins or extensions. It is not advised to use third party plugins with ChatGPT

[dire obsidian]

I said with Excel and and Power BI Desktop for example.

[normal quarry]

what is Power BI Desktop?

Excel doesn't have any authorised ChatGPT plugins

[dire obsidian]

It's kinda like Excel but more for reporting. Also made by MS.

With respect, youre misisng my point. Maybe my bad. I'm not talking about plugins and such for anything to do with OpenAI

[normal quarry]

maybe go back to the beginning and explain what the issue is?

[dire obsidian]

Not sure how else to say it. I will give you an example. I ask it how to do something in Excel and it tells gives me various steps. I do all the first steps in Excel, which can take hours sometimes, then it turns out the add-in that is in the final few steps isnt even for Excel but for another application.

[normal quarry]

so it is hallucinating, maybe break the chat up into smaller pieces

[dire obsidian]

Yeah it seems to be doing it a lot lately. I correct it then it forgets 2 minutes later and I have to keep reminding it. Rather than getting done what I needed to.

[inland musk]

chat gpt site down?

i cant access rn

[normal quarry]

no, it's up https://status.openai.com

[inland musk]

hmm okay thanks

[dire obsidian]

If I may make a suggestion. For a long time I've felt like it needs to ask us claffying questions back. It's almost like it's scared just to admit it doesn't know something. I know the chat flow is pretty serial by nature but I think there could be some improvements on this basis.

[young basin]

they've already started rolling out the voice/vision alpha right? I saw a video of some random person using the new voice at home the other day but can't find any more recent examples

[feral zinc]

"We had planned to start rolling this out in alpha to a small group of ChatGPT Plus users in late June, but need one more month to reach our bar to launch"

[young basin]

thanks, so whatever I saw was just fabricated

[feral zinc]

or lives in SF and is openai friend

[iron mauve]

Is there a solution out there in the market, which leverages openai gpt to generate test data for common SaaS applications? i.e. Salesforce, Shopify, Zendesk. Use case would involve prompting with object type, number of records, and guidance on the type of industry the data should be relevant for? I am familiar with TypeChat, where I could force a specific schema, and I'm familiar with Faker.js for generating test data records, but really looking for a viable solution to produce sample datasets to hydrate some of these different systems for testing/development/demo purposes.

[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.

[faint frost]

Chatgpt 4o very bad. I mean, it's almost non intelligent at all. It's the third time I'm telling him to Write a title and after the title a paragraph and not a Title and after it a subtitle. And he keeps writing a title and a subtitle. Even If if link him a text layout structure from drive or even pasting it to the chat, he is not able to follow the rules of the outline. Jeez.

Feels like we are back to 3.0

What is going on?

And we still pay for this even if it's free.

I'm asking to change a specific phrase and he says:"I understand your frustration, let me change it according to your request." and than he writes exactly the same phrase as before!!!

[proper hill]

Good grief just delete the subtitle yourself after it is generated

[faint frost]

What a useless answer. It's just an example of how he is not working properly. Are you ChatGtp 3.0 to? LOL

If he is getting this wrong imagine the rest! It's a total mess.

[honest jay]

Negative prompting like "don't do x" can often confuse LLMs by making it think of x in the first place. If you want to include negative prompting, it can be best to offer an example like "instead of x, do y", but in general offering a positive prompt like a response template will yield the desired result. I see you've already shared some more in #prompt-engineering too, I think that's the best place to keep looking into this if you feel like it!

[vast pawn]

anyone doing code? right now i just ask gpt to build a project write code, it all lives inside the chat session, im wondering whats the point of copilot seems inferior

[livid pier]

I am very disappointed. OpenAI has started downgrading the "intelligence" of GPT-4o in ChatGPT Plus again, just like they did with models like GPT-4-Turbo in ChatGPT before. Now my GPT-4o is so dumb that after receiving a GitHub link, it doesn’t use the web search tool to read it but instead tries to fetch information using the code interpreter, which is explicitly mentioned in the official hints as unable to access the web...

And I have clearly observed a very obvious difference in "intelligence" between the replies from the API and ChatGPT Plus

[vast pawn]

can you ask it to

[livid pier]

At least a few weeks ago, there was not much difference between the two

[vast pawn]

does it comply\

[livid pier]

I thought they would at least ensure the basic experience for ChatGPT Plus users before offering GPT-4o for free to more people..

In fact, it is just the opposite. Free users received the more intelligent GPT-4o, while we paid users have to bear the significantly reduced response quality of GPT-4o. This is completely unfair...

My friend told me that OpenAI has almost no usage rate limits for free users anymore... I can't accept it

[livid pier]

I know I can do this, but I don't think it's justified to get a degraded performance model..😫

[honest jay]

The Free usage cap is definitely still in place: https://help.openai.com/en/articles/7102672-how-can-i-access-gpt-4-gpt-4-turbo-and-gpt-4o --
https://openai.com/chatgpt/pricing
I also don't see any indication that the model of 4o used on Free is any different than the one on Plus, other than the usage cap. I'd be very curious to see any comparisons between a Free account chat and a Plus account chat, though!

[livid pier]

Almost

[livid pier]

I mainly cannot accept the significant decline in response quality that comes with all this

At least a few weeks ago, there was no difference between GPT-4o in API and in ChatGPT Plus, but now, the one in ChatGPT Plus clearly has had a big scoop taken out of its brain..

[honest jay]

There's always been the context window limit of 32K on ChatGPT, with 128K available via the API, so that's definitely a difference. Otherwise, ChatGPT comes with its system prompt for using DALL·E, code interpreter, etc. and memories and/or custom instructions can change a lot of behavior. So there's definitely room there for a pretty different experience between ChatGPT and the API, I think by design.

[livid pier]

I previously noticed that DALL E's prompts significantly decreased the response quality of GPT-4-Turbo, so I turned it off since then. That is to say, GPT-4 in ChatGPT Plus does not receive those prompts.

The key is that originally there wasn’t much difference in their performance...but now, well...ChatGPT Plus users are forced to accept all of this

Actually, I have tried to restore all the prompts, including the initial system prompts and memories..

Restore these in the API, then compare the output quality of GPT-4o in ChatGPT Plus

[dry ore]

Well, I have been thinking about this and if they do have to read the flagged messages which are still many they could well see the unflagged ones too specially considering the flagging system can fail sometimes

[livid pier]

Whether or not carrying this information, the model in ChatGPT Plus a few weeks ago performed almost no differently from the one in the API

[keen plaza]

Exactly

Which is why I said you should just assume everything is public, like reddit, at best somewhat anonymized (but not really)

[honest jay]

I'd be curious to see any before/after comparisons you feel like sharing! Perhaps over in #1050184247920562316

[keen plaza]

But really, no one will care what you've sent unless it's some kind of government secret or doomsday plan

[eternal rock]

`` Server Tip: ``

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

[dry ore]

Ah ok, no it isn’t by any means something like that lol

[livid pier]

Unfortunately, I wasn't able to keep some of the most important screenshots, and I mainly communicate with GPT-4o in Chinese, including now (to save trouble, I am using a browser translation plugin to call the OpenAI API to translate the text in my input box into English)

But one obvious thing is that the GPT-4o in ChatGPT Plus is no longer so obedient in many cases (such as calling the wrong tool)

And it is no longer willing to provide detailed responses to the lengthy content I send out

[livid pier]

For example, when I share with him the plot of a game I recently liked, he used to discuss it with me in detail as requested. But now, he acts like an unfeeling "person," only willing to respond to me briefly and then asks if there's anything else I need help with.

I tried turning off custom instructions, but it didn't help. The content in "memory" hasn't changed during this time either.

[livid pier]

I don't know if it's no longer following instructions as well, or if there's an alignment issue

[dry ore]

Also, is it even possible to get auto banned?

[dry ore]

Exactly opposite to me lol

[tough bramble]

Anyone else wish they'd have parity between mobile and web? 😫 Mobile to use voice, web to make custom GPTs or see previous responses, etc....

[livid pier]

Everything for OpenAI is "Coming soon".....

[tough bramble]

No, no. Coming weeks.

In Valve Time. Which means slowly over the next 7 years.

[livid pier]

If you are willing to believe their "a few days, a few weeks" kind of nonsense

[livid pier]

That's really strange, it seems it has an issue with me😕

Every time I chat with it, I always use honorifics and never scold it. lol

[livid pier]

If you are just playing some "love-related games" with it, generally speaking, there is no problem, because many people do this. I often see them sharing conversations full of orange and red marks..

😆

||But I think Claude 3 Opus is still better at these||

[jolly yew]

does it help to make the ai remember in its memories, that its a "professional programmer"? it made a typo earlier, then i told it to remember this, and now it doesnt do that anymore

i wonder if any of you folks have a potentially even better thing for it to remember when it comes to programming

[young basin]

it's better to explain to the LLM what you are struggling with exactly and ask how it can change its ways to be more useful. I did this with Claude and it suggested to add this to my custom instructions:

"When suggesting code changes, please provide specific, step-by-step instructions on exactly where to add, remove, or modify code. Include clear markers like 'Find this existing code:' and 'Add/change it to this:' to help locate the correct sections. Avoid providing large blocks of code without context."

Ever since, I haven't had issues locating where to fix my code

[jolly yew]

i also put itn its memories already that it should use the web searchEVERY prompt related to programming and that works aswell

[jolly yew]

the issue for me right now with gpt 4o, is that 3 prompts in a row already it gave me the exact same script that supposeduly would fix an issue but it doesnt

its litteraly in a loop currently xD

[young basin]

4o is awful for that, which is why I switched to Claude

I was constantly having to start new chats with 4o to stop it from looping, but when it comes to building code, that just doesnt work

[jolly yew]

well i already paid for gpt 4o this month, not gonna pay for 2 different ais in a single month t hats nuts imo

[young basin]

Claude projects is where its at with code

[jolly yew]

and the free limit with claude is super low just like iwth chatgpt

can claude utilize the internet already at this point?

[young basin]

to be honest the paid limit is also bad with claude

no

[jolly yew]

well thats not good

because that means that not everything will be up to date

its a trained model after all

[young basin]

it's not but I dont need the internet search to help with code

[jolly yew]

paid claude is 5x more usage.. right lmao! the default is like what? 5 prompts? so its 20 prompts instead? that is ridicoulous..

does it reset every minute? cuz then its fine withj me

[young basin]

I dm'd you

[feral zinc]

yes your account can get auto suspended and then you need to do manual appeal to support

[cold pewter]

talking to claude is like talking to a toddler

[dry ore]

I saw this on reddit: So while its true that not many employees of chat gpt are reading your chats, that's not the whole truth.

AI Trainers from other companies are being hired out to train AI like Chat GPT and all the others. You didn't hear this from me, but they do read your actual conversations. Like hundreds of employees. They are not told where they come from. They are just being given the data sets. So someone at Chat GPT is reading chats and selecting conversations to be graded. Even the inappropriate role playing one's, the one's with person information, and the one's with deeply concerning questions. They read them all. They train on real data. Now of course the sensitive one's get flagged and thrown out, but they do still read them. Which means a normal person, again not coming from me, is reading what you write.

This is why AI chat bots have a warning saying you should never give an AI your personal information. If you think the AI trains itself, your wrong. People train the AI by feeding it chosen pieces of data (aka your conversation).

WARNING: Some of these companies hire people without doing a background check!

Please don't ask anything else, because I can't say more. That's all. You've been warned.

@feral zinc wdym by that emoji🤨

[cold pewter]

They are reading the moderated chats, yes.

[dry ore]

So they are reading all the embarassing messages I wrote?💀

[cold pewter]

If its flagged, yep

[dry ore]

No they are not

They are unflagged

[cold pewter]

Then they arent reading it...

Its stated quite clearly moderated red and orange flags are sent for review, I believe its only the prompt itself however

[dry ore]

What it says is “the sensitive and flagged ones are being thrown out but they still read them”

So they still read even the unflagged ones

[honest jay]

We also plan to launch a Windows version later this year.
https://help.openai.com/en/articles/9275200-using-the-chatgpt-macos-app

[still crater]

Is fine tuning for GPT 4o publically available yet? it says on openai that you can get access through the dropdown menu as an experimental feature, yet I cannot see it.

[honest jay]

Hey! It sounds like not everyone is eligible yet due to it being experimental:

Fine-tuning for GPT-4 (gpt-4-0613 and gpt-4o-*) is in an experimental access program - eligible users can request access in the fine-tuning UI when creating a new fine-tuning job.
https://platform.openai.com/docs/guides/fine-tuning/which-models-can-be-fine-tuned

[still crater]

Ah alright, thank you so much!

[dry ore]

Wait I only saw this now, radium are u dev from chat gpt?

And mrallgamesday too

[sudden herald]

No, they are not developers.

[dry ore]

Ah ok

[sudden herald]

I rarely see developers on.

[dry ore]

I say their “community developer” role

Wait it’s the same as yours

But u and them still know what staff does right?

By staff I mean chat gpt devs

[sudden herald]

You can get the role by clicking Channels & Roles

It just shows you develop with OpenAI's services.

[dry ore]

Ah yea just saw it now

[sudden herald]

I don't know about them, but I work closely with the staff team so I can share an official stance on a lot of topics.

[dry ore]

And that thing u said about them seeing unflagged messages was official?

[sudden herald]

No, cause I have never asked about them, I always specify if it was the official stance, but with the millions of messages sent every day it is impossible to review them all manually, especially since they have such a small team.

[dry ore]

But then can u please ask them about that? I know there are a lot of messages but like, there are many flagged messages too and they have to see them, so that doesn’t gurantee they won’t see the unflagged ones lol

[sudden herald]

Not right now, it's a weekend and there is a company vacation next week, excluding like the engineers and support.

[dry ore]

Yea ask them (engineers and support) next week if possible and then inform me

[urban cipher]

Is the free version of GPT4o dumber than the paid one?

[sudden herald]

Not sure, I guess you have to upgrade and find out for yourself with your own use cases.

[urban cipher]

i had the plus plan and cancelled it and now it seems like its braindead sometimes

but im not sure if im just unlucky with the answers recently

[feral zinc]

free users have 8k context while plus has 32k which may impact things but not if it acted silly on first replies

[misty sky]

Do we have a release date for ChatGPT Memory (Europe) etc.? It's been several months since it was launched in the USA and other countries. I know using a VPN is an option, but it's not convenient at all. Thanks for any insights!

[tight island]

That’s not up to OpenAI

That would be up to the eu to let the users use it, it’s just politics, eu being slow as always

EU being EU as always

😔

[feral zinc]

yes it is

[tight island]

So if it violates EU law, EU users should be able to use it?

That’s what i mean in that particular context

Doubt it does

[feral zinc]

how does it violate it when we already have custom instructions which by openai articles does the same but in a different way. also its not just eu but south korea, team and enterprise plans which have nothing to do with eu

[tight island]

bureaucracy in a nutshell, that’s why it takes the time it takes

[feral zinc]

its just openai being slow

[jaunty robin]

There's also a clear difference between storing information the user overtly inputs and storing information algorithmically derived from user inputs.

[tight island]

I don’t think that’s the case

General Data Protection Regulation is a law in the EU and Memory feature poses a risk of violating GDPR or is in the grey area

[feral zinc]

GDPR is EU only and has not changed for almost 10 years, the feature missing outside EU

its just not priority for openai but nothing is blocking them

[jaunty robin]

Why would it need to change to be relevant?

[tight island]

It’s old

Relatively

[sudden herald]

Good question.

[jaunty robin]

Old laws are applied to new contexts all the time.

[feral zinc]

changes would mean something new that was not taken in account would delay things

but nothing has changed

[sudden herald]

But they could be taking into account the old GDPR

[feral zinc]

which means its on openai as its very old thing by now