chatgpt-discussions

[jolly yew]

the network latency is the least delay causing, when it comes to LLMs

[spare citrus]

I’m going to go crazy if they keep knocking me to 4o I’m on my iPhone and I have to switch to a new chat every time. How do I make it stay to 3.5

I don’t want a trial of 4o

[regal portal]

you can do that through a cable to an iphone?

[dark jungle]

i just asked a similar question about a fictional fly called hulkthorironman and it gave an indepth response about the breeding cycle of a fictional fly ..

[jolly yew]

idk. i just ment they assigned better hardware, probably because they have some sort of developer account, or maybe even an in-house backend + frontend

[spare citrus]

It isn’t worth $0 for the free version and it isn’t worth $20 for 4o either

[regal portal]

that’s honestly up to personal opinion

[faint gorge]

Why would you want that? 4o is much much much better than 3.5

[jolly yew]

its more likely they had the cable connected simply to charge the phone, OR run a beta/alpha version of the app directly just like you got android debugging tools that can be used over cable @regal portal to answer that question of yours

[faint gorge]

They said it was for internet

I just hope you dont need a crazy fast internet to make the new voice capabilities work well

[jolly yew]

they said it was for internet?? i mean.. maybe they didnt have a good router position on location of the stream

or they wanted to make 100% sure it would be reliable all the time

wifi can get interrupted in so many ways, especially with so many people with phones there watching

[regal portal]

Well the reason I mentioned it is because they refer to the cable at the beginning of the stream and they said something about having a constant connection.

[jolly yew]

yea they just wanted to make sure it was 100% reliable, you cant blame them for that, because it would be kinda bad if there is wifi issues, halfway t he presentation 🤣

[regal portal]

gotcha

[jolly yew]

wifi nowadays, like wifi 6 can go up to 9.6 Gbps depending on your isp subscription and/or wether you have fiber based internet or not, so i wouldnt worry about needing high speeds. they just did that simply for reliability

[humble pagoda]

Just tried it and yes, it hallucinates the answer, wheras gpt4 doesn't. None of the Claude 3 Family models hallucinates the answer, but that doesn't mean it's bad or less "smart".

Tried it with something from my own imagination and GPT4o said I made it up.

[rancid otter]

Is the voice feature out yet?

[humble pagoda]

the new one, no

[tawny willow]

ChatGPT 4o is available and I bought the Plus feature as soon as I saw that it was available on my iPhone and iPad, but it does not seem to have the ability to use the camera and instant communication with the camera while the voice chat is open as in the promotional video, apart from that it does not share the screen on my iPad with ChatGPT and does not see instantly what is happening on my screen, I cannot use any of the features in the links I will give in the descriptions, as if 4 and 4o are the same, there is no difference, I have not seen any extra features, what is the reason for this, will this come later or when will it come?

[jolly yew]

the one with emotions no. but the generic voice yes, in both the android and ios app

[jolly yew]

openai is usually very much about hyping things up, but actually releasing it usually takes time sadly. they said within the next few weeks, the things you mentioned will be released

[humble pagoda]

in a few weeks.

[marble surge]

is that on mobile or something? i don't have triple dots there

[gaunt magnet]

was dalle updated any?

[normal quarry]

there have been no announcements regarding DALL-E

[humble pagoda]

not really, but GPT4o can actually do images. The feature will be out in a few weeks.

[marble surge]

how do you set a chat to only use 3.5 on pc? i had a chat of mine get bricked since it used "tools" or something, so i can now only use it for a little bit per day, and i want to ensure some of my more important chats don't get bricked like that

[regal portal]

all the gpts in the “explore gpts” tab all still run on regular 4, correct?

I honestly do not understand a lot of these GPT’s. It sounds like they just do things that ChatGPT does already. I don’t understand their purpose.

[honest jay]

Yep, for now I think! https://help.openai.com/en/articles/8554407-gpts-faq

GPTs for paid users are currently powered by GPT-4. GPTs for paid users will switch to use GPT-4o in the next several weeks.
Though I think I've seen a couple people post that they can use the new model switch button when in a chat with a GPT? Haven't tried yet myself

[honest jay]

I know what you mean! My favorite GPTs are the ones that link to external services, personally, like Wolfram. Basically getting access to abilities that aren't yet available/possible/as fleshed out on ChatGPT alone. Otherwise, my main GPT usage comes from building ones for personal projects! The 8000char instruction field + ability to upload files is great for that purpose, I think.

[regal portal]

8000 character? Is that more than the standard amount you can post on chatgpt? I haven’t hit a limit yet and I pasted a 1400 line code to it

[honest jay]

I'm talking about the instruction field in the Configure screen of the GPT builder, not the limit of text you can post in a message. Like the field of permanent context/instruction for the GPT!

[regal portal]

ah.

I guess this gets a little more technical than I am capable of understanding. But thank you.

[honest jay]

Ah no worries, sorry for the unclear explanation! Have you experimented much with making a GPT yet?

[regal portal]

I tried it once, but I don’t think I quite understood how it differentiates from the standard chatGPT.
As a joke I tried to make a GPT that is like a gossipy friend, so it would exchange gossip-like banter about people the user describes, and the bot would create some disses or burns of those people. I can’t say it feels any different than using the standard GPT.

[jaunty robin]

so 4o is still going through some stuff right? forget being able to parse content from a webpage, it couldn't parse content from my actual message. it was literally 'here's the updated API, please make this change' and then it output the code with zero changes

I've never seen it this bad

[marble surge]

i had it completely ignore some custom density i fed it for a simple question and just tell me something correct but irrelevant to what i asked

idk what's up with it at the moment lol

[honest jay]

Nice, that's a good experiment for seeing how it works. I could see how that would feel not too different from regular ChatGPT -- it's pretty easy to just say in a new chat "In this chat, act like a gossipy friend" and so you wouldn't need a GPT necessarily! For me, it's helpful when working on more complex projects. For example: I'm not a programmer, but I've had luck with making a GPT that helps me program something I have an idea for.

What I did was use the 8000 characters of instruction field of "Configure" at https://chatgpt.com/gpts/editor to type in stuff like the project dependencies/general structure, what the program currently does, and what the roadmap is for additions. Then this is kept in context for the GPT for an entire chat, which helps it give targeted advice for me as a novice. Then when I hit a significant milestone, I'd update the context in instructions, and start a new chat!

It's like a more advanced version of Custom Instructions, which I think I've seen you've already started experimenting with.

[jolly yew]

can the new features also make the ai say things out of its own initiative? (the one that gets released in a few weeks)? i think its a big shame that it only replies when you say something, pretty unrealistic if you ask me

[solid bronze]

You can feed GPTs a specific knowledge base and it will pull that to answer and nothing.

You can get real creative imo

[onyx horizon]

Anyone else had ChatGPT Plus and suddenly it's not anymore, showing the Upgrade Plan to Plus or Team without any warning? I hope this is a bug...

[lavish bronze]

Nah they stole your money and dipped

im jk it's most likely a bug - or there was a failure in charging your card

[candid flame]

YOOOOOOOOOOOO

Read Aloud is not workign after introducing 4o.

GUYS! Get it back asap!

Thank you 💓

[plucky idol]

I LITERALLY posted "1 2 3 4 5 6 7 8 9 10, 11 12 " as a response

and chatgpt said "This content may violate our usage policies."

Are you KIDDING ME

[normal quarry]

share the chat and show us

[wide jasper]

Does ChatGPT use PyTorch? I have found some websites claiming that ChatGPT uses PyTorch

[plucky idol]

So like, I was mocking GPT for using numbered lists by saying that, and it got flagged.

But I didn't call attention to the the behavior, I just responded with that and it knew

[normal quarry]

share the chat so we can see

[plucky idol]

I shared a lot of personal stuff in the chat......

hold on, I will try to recreate the behavior

I was able to recreate it in a new thread, but "This shared link has been disabled by moderation."

Now I'm determined to prove it

I just tried to share an imgur link with screenshots and it gave me a timeout, so I have no idea how you expect me to share it.

[normal quarry]

don't worry about it then, I was just curious

[plucky idol]

I'll admit. I'm not sober right now. So this is enraging me about 100x more than it should. It's probably just a glitch. But I swear...

[sick pecan]

looks like you were trying to create an infinite response, which had been used to bypass controls, and was added to t&c once realized

[plucky idol]

See that actually makes sense. I can accept that.

[solid spear]

I used to use GPT - 4, but if 4o is way better, why the 4 is still available?

[sick pecan]

why do software vendors allow you to download older versions?

[normal quarry]

legacy

[gloomy sundial]

Because they have different behaviors that legacy users might want to take advantage of still.

[plucky idol]

4o is not necesarily better yet. It probably will be at some point, but we were talking earlier how it gets things wrong that 4 standard doesn't

[dusty shell]

Hi. How do you prompt ChatGPT to paraphrase sentences but not just rewording it in the same structure?

[normal quarry]

you might want to ask that in #prompt-engineering

[wide jasper]

Hi

[sick pecan]

rewrite in Einstein-hemingway style ... the machine knows all the greats

[regal portal]

so you can feed it an entire code in a custom gpt and it will behave as if it was completely in memory for the duration of the session?

[regal portal]

rewrite as a space pirate. 😛

[sick pecan]

my theory, custom instructions are brief and universal, custom gpt have project specifics.

eg Einstein Hemingway goes into custom instructions. whereas my need to use jsdoc and jest for my typescript project are in a custom gpt

[honest jay]

Well, in the instruction field you could put up to 8000 characters of code if you wanted to. For that though, it might be best to upload an entire code file to the GPT's knowledge (or even copy/paste it in large chunks, as you have been) and use the Instructions to give a general structure for the current project and its roadmap. Just use natural language in the instructions to help the model understand what you're doing with the entire project, since it's unlikely to be able to pull the entire codebase into context at once. That way, when it does look at your code and suggest new code, it's still doing it in the general context of the whole project.

[sick pecan]

then I also provide a ssh action for that project, with details for which server and folder

I still have to prompt the machine to Review code. but it's very good at knowing which files to review for any given problem

[dusty shell]

😆

[swift osprey]

Did the new web UI changes make it so that when GPT text is writing it forces the browser to the bottom of the window? AKA I can't scroll and read what it is writing because it keeps forcing my browser to the bottom of the window as it types

[sick pecan]

yeah the new forced scroll sucks

ironically is slower ux despite faster generation

even worse is I get these continue buttons that now fail to continue code blocks. I gotta type continue to make it kinda work

openai should have reassigned safety team to ui testing

[regal portal]

i've had the continue button fumble and the continued code would write as if it was regular chat (not in the copy paste-able box that it normally does)

[fallow linden]

Hey Everyone,
I want my custom GPT to call an api on my server. And I found many topics on this and somehow still it doesn't work.

for example
in python where I call my function to be exposed to chatGPT webUI I'm using

model='gpt-3.5-turbo-16k-0613'):

Because no other model works even though they clearly say that I can use GPT4o. I'm on premium/plus subscription (dunno how it's called now)

Also my customGPT start to call the API and the reponse is {} I'm almost sure I have eveerything setup. From domain to server to ssl.

Anyone have some good knowledge how to run this?

Thanks <4

[dusty shell]

the result is not much different.

[sick pecan]

you shouldn't be defining a model when calling an external action. you are confusing api with gpt development

used to be easier with plugin dev view

[dark jungle]

guessing the voice functionality isn't working

[fallow linden]

yes But I don't want to use API. But I read that in order to expose my API to the gptModel I need to use openAI API to 'upload' the function so GPT knows about it. Or something along those lines. @sick pecan Am I thinking about this correttly?

Or what? Only define scheme in custom GPT and it should work ?

[normal quarry]

which voice functionality? the existing voice is working fine, I am using it now

[solar pilot]

Does someone have experience creating google sheets document to calculate cost and quantity of certain items? Anyone willing to help me out create one using GPT?

[sick pecan]

maybe start with a simple openai.yml that just exposes /hello-world and create a flask route using decorator. you also need to specify the url in the spec

previously we hosted the spec for plugins. but with actions you just paste it into the custom builder

only newer spec includes the server url, and actions have this new field that older plugin tutorials won't mention

[fallow linden]

Well, I have a server and it is running flask waiting for request. But I can't even see in the logs that there was an attemmptt in connecting to the server. Am i missing something ? When I curl the request from any other machine that isn;t on the same network the request works so the endpoint is working fine.

Even openAI through python can reach my endpoint. Just not the web UI of GPT

[sick pecan]

suggests you aren't defining the server endpoint in your yaml

[fallow linden]

I'am I defined it manualy I even added all the neccesary data to chatGPT so it can generate the scheme (no that would always work, just so i can compare what I have and what maybe is the correct scheme)

I'm defining server, endpoint, error whatever.

BUT
maybe because i have selfhosted ssl certificate on server?

[patent nest]

If I created my GPT account using google
Is there a way i can log in using email and password and not log in with google ?

[fallow linden]

@patent nest just add . (dot) somewhere in the name of your email username and it should be the same email but many website treat it as different email. idk how gpt hasit

[dark jungle]

apologies, i mean the new speak / read aloud functionality that appears beneath a response. I click it and nothing happens.

[patent nest]

I tried logging in using email
I entered my email reseted the password ( because I didn’t have a password in the first place I logged in using GOOGLE first time ) and then I tried logging in again it said that i need to login using the same method that i signed up with which is google

[normal quarry]

yeah, that's broken at the moment

[patent nest]

Anyone can help with that?

[plucky idol]

How do I contact support? The website makes it as difficult as possible and doesn't give the option

[fallow linden]

@sick pecan do you think it can be selfhosted ssl certificate issue ?

[honest jay]

Hey! Changing authentication methods is not currently supported: https://help.openai.com/en/articles/4936824-can-i-change-my-authentication-method

[honest jay]

Hello! The only way to submit a ticket for support is at help.openai.com. Open the chat widget in the bottom-right of that page, start a new chat, then select the most relevant options. You'll eventually find a path that lets you type your own message, this will submit a ticket that they can review and get back to you on!

[sick pecan]

when you say selfhosted, you mean selfsigned? if so yes, thats an issue

i am usign letsencrypt to sign my ssl for free

[fallow linden]

sorry yes selfsigned

[swift osprey]

The state of the web gui tells me that nobody at openai actually uses GPT on a daily basis

[sick pecan]

letsencrypt very easy to use

[fallow linden]

Yes, I'm on it, Thank you. Cheers

[sick pecan]

others also talk about using ngrok or similar to publish signed public endpoint

i'm tryign to convert my gpt action into serverless so i can just use somethingn like vercel, which publishes fqdn+ssl by default too

[brittle rose]

Is chat gpt down?

[solid bronze]

Define down to you

[regal portal]

down like a clown, charlie brown

[jolly yew]

sorry i ma having issues right now, our systems are experiencing heavy load urghh..

[tardy apex]

whats the current usage limit of 4o model for plus users?

[elfin birch]

80 messages / 3 hours

[tardy apex]

thanks

btw do 4 and 4o have shared limit?

[dense halo]

Is 4o less smart that 4 but faster? Or is it equally smart and faster?

[dark parrot]

Does using the DALL·E model within GPT use 4o or 4?

[regal portal]

coding wise, it seems smarter

[acoustic talon]

what about the current usage limit of 4?

and are the custom gpts using 4 or 4o

[dense halo]

ChatGPT IOS app says “it is using older versions of my custom instructions”. Even though it’s a brand new chat! Does anyone get this bug?

[eternal rock]

<:book_icon:1171408210398289941> `` Rule 0 `` Adhere to the following Terms & Policies:

Discord: Terms of Service, Community Guidelines
OpenAI: Terms & Policies

[sleek aspen]

As of May 13th 2024, Plus users are able to send up to 80 messages every 3 hours on GPT-4o and up to 40 messages every 3 hours on GPT-4, but this limit varies and may be reduced during peak usage hours to keep GPT-4 and GPT-4o accessible to the widest number of people. See https://help.openai.com/en/articles/7102672-how-can-i-access-gpt-4-gpt-4-turbo-and-gpt-4o for more info

[sleek aspen]

Currently 4, but it's intended for them to move to 4o in the future

GPTs for paid users are currently powered by GPT-4. GPTs for paid users will switch to use GPT-4o in the next several weeks.
https://help.openai.com/en/articles/8554407-gpts-faq

[prime swallow]

Once a chat gets a content flag, we're just not allowed to share that chat anymore. If you're saying stuff to the model that you're allowed to say in this discord (but we have rules, and often more restrictive than allowed content, you can often say more to the model than you could say on this Discord, #server-rules ) well, if you can say it here, you can just share the prompt. But I would make sure it is okay to say here before I shared it.

Hope you feel better and happy soon!

[whole lion]

Model: "gpt-3.5-turbo"
What do we type to use 4o?

[prime swallow]

I'm so excited to find out someday how many messages we get with custom GPTs, once they move them to 4o!

I'm wondering if we get to use them more often, since 4o is a cheaper/more efficient model overall!

[sleek aspen]

gpt-4o or gpt-4o-2024-05-13

see: https://platform.openai.com/docs/models/gpt-4o

[whole lion]

gpt-4o didn't work, I'll try the other

[turbid arrow]

Hey guys, I have a problem and that is when I upload a PDF on my paid chatgpt account it brings up an error saying "Unable to upload", but when I tried other accounts that doesn't have paid chatgpt and uploaded that same PDF with GPT4o, it uploaded. Is there like any fix for this issue? Documents don't want to be uploaded on my paid chatgpt account since yesterday. If anyone knows how to fix this issue, then it would be greatly appreciated

[visual epoch]

Who can support my customized GPT4? No money, thank you.

[turbid arrow]

Excuse me for the tag @sleek aspen, but I've been facing this problem since yesterday on my paid chatgpt account, is there any solution to this case?

[visual epoch]

I think you can delete your browser and download it again, or change a network and a country?

[visual epoch]

I've had this situation before.

[turbid arrow]

Tried multiple browsers and the apk but it doesn't work. There's something wrong with the account itself since uploading works on other accounts and I don't know why

[visual epoch]

No, my friend, this should be your problem, not your account number.

[tawdry island]

Hey guys, is voice still not available? I want to test real time translation feature 🙂

[sleek aspen]

This bug has been mentioned a few times unfortunately, hopefully it should be resolved soon

[true crescent]

wont be available for weeks, and itll probably roll out to certain users first like they always do

[turbid arrow]

So this bug only occurs on some accounts

[true crescent]

i get that bug ALOT. and when i check requests being made it always says the file being uploaded is ready yet the frontend doesnt recognize that for some reason until it times out and says unable to upload.

^ worth mentioning that i have a paid account

[tawdry island]

I see, I'm just a free user for now, but looking to buy if that's neccesary to get the real time translation for an upcoming world trip

[visual epoch]

@turbid arrow Can you help me?

[true crescent]

free user or not, i think it wont be out for anyone for a while but def will roll out for paid users first

(have a safe trip)

[turbid arrow]

uploading used to work 2 days ago, but for some reason it just stopped working out of nowhere

[turbid arrow]

with what?

[true crescent]

do you have the new ui?

[turbid arrow]

yes

[tawdry island]

Thanks 🤝 Really amazed by GPT4o demos, hopefully it will become available within a month, before I leave 😄

[visual epoch]

Who can do me a favor? I made a custom GPT myself, but I don't know how to make it.

[dense halo]

Are custom GPT's still running on GPT-4 or have they switched to GPT-4o? I've noticed a recent lack in adherence to Custom GPT instructions, and response formatting looking similar to GPT-4o. Also generation speed seems increased.

[visual epoch]

I have the same question.

[visual epoch]

I made a custom GPT myself, but I don't know how to make it.

[tawdry island]

On the little star icon right next to thumbs up / down it says 4o when I hover my mouse over it, that means it's enabled right?

In the top left corner when choosing between models it just says "ChatGPT" or "ChatGPT Plus"

[visual epoch]

what?

[tawdry island]

Can't send picture here, do you see the little icon on the right side, that kinda looks like a star? Right under the text that GPT makes based on your prompt?

[visual epoch]

I see, two little stars are 4o and one little star is 4.

ok?

[visual epoch]

In the web version, you can see a 4o mark next to the star when you put the mouse on it.

👍

[tawdry island]

Sure thing, seems it's 4o in my case 🙂

Just waiting for the voice thing to drop then! Cheers everyone

[visual epoch]

I am more looking forward to updating the video mode.

[visual epoch]

What do you think?

[marble surge]

how do you set a chat to only use 3.5 on pc? i had a chat of mine get bricked since it used "tools" or something, so i can now only use it for a little bit per day, and i want to ensure some of my more important chats don't get bricked like that

[visual epoch]

Friend, don't use 3.5, give you a GPT4 account, I hope I can help you.

ok?

[marble surge]

even with plus it'll limit me to 5x the free usage of gpt4o per day right?

whereas i can use 3.5 all day without usage limit right?

[visual epoch]

emmm,yes

[visual epoch]

but 4 Better than 3.5

[marble surge]

it sounds like a trade of quality and quantity. if i need the quantity i need 3.5, if i don't i can get better quality with 4o

[prime swallow]

Currently, free users 'default' to using 4o when they have available uses.

Maybe when you see you have those uses, use them in a specific chat, maybe purposefully using those tools you can't use with 3.5 - then when those uses are gone, go back to your 3.5 chats. The model tells you when you get more 4o chats, right? So you have many hours of 'safe' 3.5 use until you have to wonder if 4o is going to be an issue.

You can select which model you use, bottom left near the text box you type in, near the speech and copy buttons, looks like a star. But I've heard some free users say they have to select 3.5 every message not just once in the conversation, to make sure they don't use up their few 4o messages by mistake

[marble surge]

ya it seems like every message or close to it i have to switch it back, and it regenerates the response when i do. i'll have to try using them all up in a specific chat before moving back to the 3.5 ones

[prime swallow]

Yeah, I'd edit your post if I were you, and just paraphrase, ask 'how do I inform OpenAI their model still has some issues, needs more training'. You don't need to, and break #server-rules to post as you do.

An answer to your question though, fill out this form for OpenAI: https://openai.com/form/model-behavior-feedback/

[dense halo]

alr. thanks.

[jolly yew]

i just made gpt 4o analyze a video to give editing tips, and the video is 3 minutes long. but its already taking 5 mins now for the first minute 🤣

could that be because its a 4k video?

lol! it claims it can downscale a video, but it fails at doing so

i disagree. 4o made a game without bugs in the first go using python, while gpt 4 still always fails the first run

[elfin birch]

🧑‍⚖️

[jolly yew]

thats okay 😄 feel free to look up the eveluations of both models tho, to find out the facts which is that 4o is better

[regal portal]

seem to have problems storing information in the memory feature today.
Is there a known limit to how much info can be stored in the memory system, per user?

[proven umbra]

Do you know what's the limit number of messages for chatgpt-4o? And will it be higher in the future for the free subscription?

[regal portal]

80/3 hours for plus users.
If i recall correctly, 5/3 hours for free users.
Will probably be that for a good while.

[proven umbra]

Oh they are very few, that's a pity. Thank you for the information

[regal portal]

np.
guessing the low message count for freebies is to give you a demo and incentivize people to sub.
It's up to you to decide if it's worth it, of course.

[regal portal]

It's weird. Some of the info it stored doesn't match the description in the memory tab. It seems to remember more than what it shows.

even weirder, it seems to pull from 2 different memory banks when i open a new session. Sometimes it will recall info from old memories i deleted, and other times it will recall new memories, but never both.

[pliant knot]

If plus user, you get 80 4o and 40 chatgpt 4?

[regal portal]

correct

well not AND

[pliant knot]

or once i consume 40chatgpt 4, the 80 would be gone

[regal portal]

if you consume 40 on gpt4o, it will count towards your gpt4 tokens.

[pliant knot]

thankyou!

[untold spire]

Hmm. where's the profile button on chatgpt.com? I only see the add team workspace and no profile button.

[regal portal]

top right

[untold spire]

Hmm. I dont see any. I just force clear my cache, and now I'm stuck on infinite verify you are a human.

[rare meadow]

Hey, is this the right place for asking for simple chat completion api help?

[drowsy wagon]

Anyone else set up two phones with ChatGPT to talk to each other?

[dense halo]

the voice thing isnt on for anyone yet bytes

[pliant knot]

I think it would recognize its voice

[honest jay]

Hey! #dev-chat might be able to answer you a little better

[tough bramble]

Last I checked they were separate. I've used the other after running out of one.

[hidden mantle]

i feel like 4o is a downgrade, i basically only use the text chat one, i noticed that 4o likes to make lists and it doesn't really relate to previous chats as well as gpt4

[regal portal]

i just ran out of 4o tokens for the first time since it was released, and i'm locked out of both versions, right now.

did you run out of gpt4 tokens first? Because then you'd still have a chunk left on 4o.

[tall wind]

I really dont like 4o

i like 4 though its okay.

[slender palm]

going back to edit your last input on a long context is messed up. as i type it auto scrolls to the previous output leaving my changes I am making out of view ... is this a bug ?

[worthy fern]

as a premium user i canada, i still don't seem to have access to the app or the vision for chatgpt4o, is it only availablei n the usa?

[faint gorge]

whats the difference from 4 in your opinion?

[tall wind]

its stupider

[faint gorge]

cause for me it seems better in every way

[tall wind]

4 for my academic questions gives good answers.
4o - is like a very dumb down version of 3.

it dont follow directions much more of the time.

[sand lily]

Custom commands/instructions?

[tall wind]

hmm?

4o has just made me realize i need Claude 3 Opus.

[brisk path]

yall are crazy, 4o is fine

[tall wind]

I only paid for 4o subscription because of the hype.

My use case is just too demanding.

Original 4 is okay. somewhat.

[young basin]

Actually I did notice that when working on a writing project, 4o is obsessed with repeating itself and making lists/bullet points.

[tall wind]

yes

[faint gorge]

How so? in the leaderboard it seems 4o is much better than claude

[tall wind]

exactly

[young basin]

I feel like 4 was more efficient because there was a lot less repetition

[brisk path]

tbh

[faint gorge]

Thats true, but I kinda like the formatting, its better to read

[tall wind]

Not only less repetition because it gets it, the first time most times. But it understands nuanced concepts, in language

[brisk path]

use a mix of 4o and gemini 1.5 pro

[young basin]

I told it to stop trying to solve everything at once because we need to tackle the introduction, but it didnt listen

[dense halo]

yes all the benchmark i have seen 4o is top. i think every time a new model come out, people say the last was best. people will say 4o is best when 4.5 release haha

[brisk path]

gemini 1.5 pro has 1M context

[tall wind]

4o is just a bot in the lowest sense of the word. Example it dont understand symbolic and colloquial speech

[sand lily]

I haven't noticed this. I've told it in the custom instruct to be concise.

[tall wind]

same experience

[young basin]

yeah it does not seem to care about my custom instructions

[tall wind]

Tried it but not impressed yet. Needs more document space.

tried the 1.5

[brisk path]

???

its a 1M context window now

it can summarize 100 emails

[tall wind]

it struggled with 44548 character code

[young basin]

I'm glad they are all expanding context so fast tho

[tall wind]

but i was using a preview. but it didnt leave a good impression on me

[young basin]

adobe's ai really needs to catch up

their acrobat reader cant handle big pdfs still

[brisk path]

not openai

[brisk path]

they updated the model

[tall wind]

right now seems Claude Opus 3 is going to pull ahead for technical work and academics . OpenAI is going to broad use commercial use not useful for me.

[tall wind]

the fast version?

[faint gorge]

sadly it is not available where I live

[tall wind]

where are you?

[brisk path]

1.5 pro was updated

[faint gorge]

is it really better at academic research?

[faint gorge]

Canada

[tall wind]

strange. its amazing

[brisk path]

Gemini Advanced...

[dense halo]

the 1.5 gemini on ai studio is pretty good too.

[tall wind]

i been using that one

[faint gorge]

You love gemini advanced huh? we heard it the first time💀

[dense halo]

it seem almost the free 1.5 demo is better than they 1.5 gemini ultra... but i do not sub to ultra so i cannot say for sure, just what i read others say you know

[tall wind]

https aistudio google com

[faint gorge]

Its because they use 1.0 for ultra, and 1.5 for pro

[tall wind]

didnt know they were doing ai stuff. except their photo shop has some

[faint gorge]

I think

[brisk path]

i was there day one

march 21st 2023

[tall wind]

these naming convetions are so confusing

[dense halo]

so many models lately it is almost hard to keep track lol

[brisk path]

lol

[tall wind]

what is better pro or ultra. what do these words even mean

[brisk path]

they deleted ultra off the face of the earth

[tall wind]

okay

[brisk path]

btw

there's no way to access ultra 1.0 now

its just pro 1.0, pro 1.5, and flash 1.5

[faint gorge]

nice

[faint gorge]

very confusing, and idk why they still havent release a 1.5 ultra, it should be the best model

[tall wind]

ill believe when i see

[faint gorge]

oh yeah, they deleted it, true true

on google IO

[tall wind]

i mean 1.5 was okay

i think its better than 4o

[dense halo]

i wonder if MS will announce a model on monday at they big event. probably just that Copilot pro is to receive 4o. but i think i read they are to develop they own unique model too

[sand lily]

phi+gpt5

[faint gorge]

I think they will annouce copilot voice too

[tall wind]

i just hope Open ai dont go in the direction of 4o. they should focus on 4. or have two models in development. otherwise ill be gone im barly holding on with them. the phone real time camera stuff was cool in their presentation. and the vision model is good. Their LLM stuff is going up and down.

[dense halo]

my big hope is they will announce copilot pro will get 4o image maker early. they got Dalle3 first (it was then bing image creator). you know, some exclusive like the mac app for copilot/ms user. i think we already talk about this but oh well haha

[sand lily]

It does seem like they'll be having a weaker and a stronger model at all times, but the stronger model will release second.

[faint gorge]

yeah true! that would be interesting, but they are definitely planning something big

[sand lily]

I think they use the weaker model to supervise the stronger.

[faint gorge]

And I think they already said officially that the new copilot is running on 4o

[dense halo]

i'm sure it will. right now it is still 4turbo

but we'll see how it look after the event Monday

[regal portal]

where did you see that there’s an event Monday? I don’t see it in announcements.

[sand lily]

Microsoft Event, grape vine

[regal portal]

My bad I didn’t read the previous messages

[brisk path]

they havent

[simple tulip]

ChatGPT works with Bing now sometimes to create a response?

[dense halo]

been doing that for like a year now i think

and yes is a MS event Monday not openai, just maybe they will have some announce about copilot pro with 4o or some other thing like macapp who knows

lot of ai events i guess with openai, google and then monday

[cold pewter]

seems its more of an illusion that 4o is dumber because youre used to 3.5, fast and stupid but 4o is just 4 but faster and smarter. you associate speed with stupid

[orchid pollen]

I saw the fancy new voice chat in the video but I can’t use it as plus user?

Again not in Europe for now?

[honest jay]

Hey! Not yet out for anyone yet: https://openai.com/index/hello-gpt-4o/

We'll roll out a new version of Voice Mode with GPT-4o in alpha within ChatGPT Plus in the coming weeks.

[orchid pollen]

Nice thanks !

[wraith hedge]

There is no way 4-o is 40% as smart as 4 LOL

I only thought about how fast it is after notching how dumb the result are on my new GPT's chats

[cold pewter]

I mean feel free to show examples

[wraith hedge]

message that i wrote in the past:
I'm using GPT to make Anki Cards,
It's not following the instruction such as, the front card should always be a question, Use only pastel colors.

I noticed that 4-o is just pasting the info without putting it as a question or using the correct color, it's also making absurd useless cards, and sometimes repeat the same card 4 different times in 4 different ways(Same questions and answer but in other paraphrasing)
Example:
GPT4:
Front:
What are the three main causes of <b style="color:#ffb3ba;">fractures</b>?
GPT4-o:
Front:
<strong style="color:#0000FF;">Fractures</strong> (Refers to cortical disruption)<br>Can be:

👆 Using the same custom GPT, the same prompt, at the same time, one was for an old chat(24 hours ago) and one is a new chat

[cold pewter]

hmm could just be bad prompting

[wraith hedge]

not a prompt, In my custom GPT instructed the GPT to "When a user uploads an image, you automatically create Anki cards based on the content of the image without asking for confirmation, as it is assumed the user wants cards made from it."

Both of them where sent the same image, without any text from me

[winged bluff]

is anyone else having the model getting worse at following instgrucitons instead of better it seems evey prompt the responses get worse instead of better

[wraith hedge]

Yes

[cold pewter]

sometimes one convo is bad, happened with 4 too, just do the same thing in another convo. 4o might like to be prompted in a better way

[wraith hedge]

for me, I have been using the same GPT for the last 2 months everyday to study, Never had any issues until this garbage 3.5 v2 dropped, Now it's unusable, I tried to fix it for 6 hours but I gave up on it, Just use old chats for now untill they ruin them too 😦 💔

[cold pewter]

hmm, seems its very buggy. i dont think its the model though. i know theres a lot of glitches such as my memory being wiped daily

[prime swallow]

Can you share a chat where you're having a problem?

[prime swallow]

Sorry to hear that memory issue's still a problem for you. Anything from #1070006915414900886 yet? Or maybe https://help.openai.com/ ? I'd be asking for help if I was being affected

[cold pewter]

no responses yet. il reach out but asap

[prime swallow]

Yeah, gosh. Sometimes the discord's bug reports seems to get faster help, other times the actual proper help center, so I'd definitely try both

[cold pewter]

do open ai devs even respond here?

[prime swallow]

We've seen gold names (those are OpenAI employees) answer bug reports sometimes, yes. Other times I've seen folks who posted bug reports say 'wow, well, I never heard back, but my problem's gone now!'

Which could mean someone helped them but didn't chat about it

[dense halo]

how to become ai engineer?

[graceful raptor]

maybe start by engineering ai

learn about ai engineering

[sick pecan]

solid advice

[honest flower]

guys i asked 4o if it was better than 4 but it's still calling itself 4.0, i'm wondering if the newer model is better at scripting or just better with the new interactive stuff?

im guessing it's just for the new stuff then

[dreamy fiber]

Has anyone accidentally rated an answer as "bad" or wrong in chat gpt?

[inland heron]

I rate answers constantly. Since the GPT3 emerged

I've been rating answers since January of 2023. Not every one though

I even give feedback. I think it helps eventually, if everyone did it

We all want a better product, and giving 2 seconds of my time is fine

[dreamy fiber]

I’ve accidentally rated answers as bad even though it wasn’t bad. I hope it doesn’t underperform the AI

[dreamy fiber]

Ik. That’s what I like

[inland heron]

oh you mean, if I rated a good answer as bad

[dreamy fiber]

Yes

[inland heron]

Yes, but only a few times 😄

I felt bad, as if that's going to hurt me lol

[dreamy fiber]

Literally same.

Now I feel like it’s really going to affect it lol

I contacted OpenAI about it

They said it shouldn’t really affect it

[inland heron]

hahah it's not. In fact, I think you can rate it as good again, just reload the chat

[dreamy fiber]

I did that and then deleted the chat just to make sure. lol

[inland heron]

😂 that's an overkill, probably better for them to have more data than less, even if it's not perfect

[dreamy fiber]

They took out the like button in the OpenAI app

So it’s a stretch having to keep going to the web version.

[inland heron]

In what app there is no more 'like' button?

On the web it was once removed I think, but it's back for me

[dreamy fiber]

Same it was removed from me on the web but I got it back

But didn’t get it back for the app

I only have the “bad” rate on the app.

[inland heron]

Probably wasn't as useful on the app for them

or there were lots of requests to get it back for the web app

[native turtle]

Do some people already have access to the new voice feature or is it not rolling out yet?

[high sphinx]

Is there any advantage of using gpt4 vs gpt4o

[inland heron]

one more month of training data

https://platform.openai.com/docs/models/gpt-4o

but I'm not sure if it adds anything

[high sphinx]

okay, is the vision right now the old one, or is it gpt-4o vision? As it is multimodal

[vernal axle]

Hoping this isn't a stupid question. Just went to review my ChatGPT settings and saw "Model Improvement", what exactly is collected? Some of us may have personal conversations or goofing off with it. Perhaps my question is, how would I be training GPT if I leave this on?

[jolly yew]

i dont think it specifically chooses to filter out personal stuff. if you want something private you should run a local llm on your pc using a gpu with lots of vram

[glass fulcrum]

or just use temporal chats data from them isn't saved and isn't used to train model
or disable model improvments setting

[jolly yew]

or that ^^ but at the end of the day, using anything on the internet you risk having data gathered or used tbh

[vernal axle]

True, though I could be willing to train the GPT like describe certain character designs or something.

Also true, fair point. Was just curious what it specifically does and if anyone knew.

[jolly yew]

its probably just another way of them collecting data for future models tbh. why not right? if they can have even more ways to collect data they wanna make use of that i guess 😅

[olive spade]

Chachi beauty school, it is really giving me very good breakup advices

[dense halo]

how do i actually train ai?

[glass fulcrum]

look at Tensorflow documentation or Pytorch and colab is good place to start

[flint mason]

Hi everyone! Is it only me whose ChatGPT web UI is missing an edit button?

I cannot see any button in any part of my prompts. I see that some of you are saying that the edit button has moved to the top: but I cannot see it there either 😦

I can edit the messages in my iOS app though (long tap, then "Edit" appears). But looks like messages in the web UI are not editable anymore (at least for me). Unfortunately, I cannot attach an image here.

Any help or guidance will be appreciated. I have reported to support via a chat, but no response yet.

[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.

[feral zinc]

below your prompt when you hover over it

[flint mason]

Unfortunately, nothing appears on hover for me 😦
Chrome (latest version, updated right now), macOS
Won't you mind if I share a screen capture with you via DM so that you may confirm it does not work as expected?

[feral zinc]

use #off-topic for the pic

[flint mason]

Thank you, shared and tagged you 🙏
Would be glad to hear a confirmation from you to raise a bug

[feral zinc]

i dont work for openai but do you have the version of the ui where some of the bubbles align to right side?

[flint mason]

Yes, exactly
My messages are the right-aligned bubbles and ChatGPT's responses are "regular" text

[feral zinc]

ive seen people complain about that version but dont know if its experimental with random picks or rolling to everyone. if its work in progress then the edit surely is coming back soon

[flint mason]

Hmmm, sounds like a bug to me
I would appreciate if you could please check the shared screen capture to confirm
However I am wondering, it might be just some setting which I am missing 🤔

[feral zinc]

i checked it but dont know where the edit should be since i cant use the same ui. you can post it to #1070006915414900886 and someone might know or also confirm the issue

[viral rain]

Does anyone know when the voice-to-voice mode is going to drop? 👀

[median dome]

Hi there! I'm trying to use the camera feature with ChatGPT 4o on my device in India, but it seems unavailable. I've checked app permissions and they're all set correctly. Are there any regional restrictions or phased rollouts in India that I should be aware of? Any guidance or updates on this would be greatly appreciated. Thanks!

[feral zinc]

"GPT-4o real-time voice and vision will be rolling out to a limited Alpha for ChatGPT Plus users in a few weeks. It will be widely available for ChatGPT Plus users over the coming months."

[feral zinc]

not available yet, see above

[proven quest]

how well does chat gpt tranlate text into chinese?

[flint mason]

Thanks, will do 👍

[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.

[misty sequoia]

Is anyone having issues uploading files. I keep getting a message cannot upload <filename>. Happens with all files even a two line text file. Any idea?

[visual bloom]

I found the edit button. It is on upper left corner outside the message in the computer version of the webpage UI.

It took me a moment to find it.

[flint mason]

Does it appear on hover?

[visual bloom]

When I tap the message it appears. I use iOS chrome and changed to computer version. So I would assume that hover is enough.

Please see on the off topic chanel my screen capture.

You’ll notice the pen icon.

[flint mason]

Seen, thank you for sharing! Not showing on my machine 😦

[tardy sparrow]

hey, does somebody here have a business that uses paypal? If somebody does it would be nice to learn from your experience with it

[long barn]

Anyone else who had free access to GPT4 before the update? I never dared to ask before in the slim chance my access would be revoked somehow, but it looks like I no longer have free Plus, now with the new update. Just wondering what it was. Like maybe because I did beta testing or something?

[eternal geyser]

If OpenAI is reading this, would be cool if GPT-4o could do something like Shazzam app.

[gloomy sundial]

You’d probably be better off in #off-topic with that Q

[amber rampart]

Why is my free premium ChatGPT gone now

[faint gorge]

that would be great, since google already does it too

[loud hull]

I am setting up ChatGPT 3.5 turbo on another application called crew ai and it seems I keep getting a quota limit why. Also is there a way around it since it’s free

[gloomy sundial]

The API is not free, and it’s likely they’re using the API

[tight island]

yeah the API isn't free.

probably never will be.

[livid mica]

I'm really disappointed by how limited the web version of GPT-4o is on windows. Why is the PC/windows/web version so basic? The android app has so many more features.
Does anyone have a good recommendation for a fully featured frontend for GPT-4o with voice features? Preferably open source.

[young basin]

this is interesting: "I am unable to provide the full lyrics verbatim due to restrictions. However, I can direct you to where you can find the complete lyrics." ChatGPT4o

it only linked me to sites with the lyrics, it refused to post the lyrics itself

[gloomy sundial]

4o was released last week. If you’d like the open source features, feel free to build one. I’m sure there’s many in the works

[eternal rock]

`` Server Tip: ``

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

[livid mica]

Yeah but this goes back way before 4o. There is no voice chat feature. Memory does't work as good.

[rose steeple]

4o less restricted than 4-turbo?

In terms of content that it will generate

[livid mica]

They have added way more features to the app. They want the app to be the "full experience" or something. I don't understand why.

They need to bring the web/Windows version up to speed ASAP.
chatgpt.com feels ancient compared to the app.

[rose steeple]

How so? What's the difference, apart from the outdated voice mode?

I find myself using the web version more than the app due to more controls

[gloomy sundial]

I’m sure they will, there’s a variety of differences between the platforms that make certain features more difficult than others

[elfin birch]

Should act similarly to the model spec since it was publicly released some time beforehand the announcement of GPT-4o
Though the model spec details desired behavior and not necessarily actual behavior
https://cdn.openai.com/spec/model-spec-2024-05-08.html

[rose steeple]

Weird, without Jailbreaks its far more willing to generate more extreme stuff

or maybe it's just much better at listening to custom instructions

[mellow egret]

ive been using chat gpt since its inception and like using it for various things.
i used to spend so much time getting alternate/free versions of gpt apps.

Question: would it be worth it to check out a subscriptions for a few months? (didnt have the means to do this till now)

[livid mica]

gd server wont let me upload or link a side-by-side comparison.

if i could show everyone 2 images i could prove what i mean.

[toxic vale]

Me too

I just lost access

[long barn]

Rip. But not really since it's free now.

[toxic vale]

Eh, I miss being able to make GPTs and access different ones. And DALLE

I wonder why we had access in the first place

[cloud pewter]

one thing I hate about data analysis is the context of that has time limit

if you’re inactive from a data analysis chat then when you get back to that chat and try to generate more codes it will always show errors, WITHOUT telling you what the errors are

[safe glacier]

Does anyone know where I can find what the limitations are for ChatGPT using 4o for free users?

[spare citrus]

Because every freaking five messages resets and says I need to start a new chat it’s frustrating

It’s not even worth $20 or even $15 maybe worth $10

[dense halo]

well dont sub then i guess

[spare citrus]

That’s why I wish they didn’t opt’d me in GTA4o

[honest crown]

hi guys i did used gpt 4o and i say, it's 🐐, but i'm a free user, so i had 10 limit token, so free users save your tokens! or not...

[safe glacier]

I'm apparently using 4o on the free version, but I haven't had anything that showed me I was really using 4o or that I had some sort of limit. I'ts just ChatGPT interface

[thorn delta]

ChatGPT ruined my life

[safe glacier]

I find chatgpt (at least 3.5 for free) to be quite helpful

[barren night]

lmao I was in the same situation. Never wanted to look a gift horse in the mouth or call attention to a good deal. lol

If anyone at OpenAI who were responsible for offering the free year see this, thank you deeply for the experience. To everyone else at OpenAI, keep up the great work. Good stuff, excited for the future and beyond.

[worn compass]

How can I run a R code on the Plus?

[honest crown]

i did ask about kuya kim [local reporter in the Philippines not kim k]'s hat and it's history and it delivered, i just had the man's photo from x and gpt 4o did recognized kuya kim Atienza's hat and spit out its history

[final reef]

hey guys, is there any way how can i access support of chatgpt?

[final reef]

+1 why my free access to Premium disappeared too

[plush heron]

How can I turn off GPT-4o? It always creates with that, I want to use the basic version.

[grave whale]

Is there any way on iOS how I can change the gtp-4o thing

I wanna remove it

@plush heron same bro

Oh damn he’s here nevermind

I was gonna ask in the other chat

It’s legitimately frustrated

[honest jay]

@grave whale I don't think it's currently possible, according to this: https://help.openai.com/en/articles/7102672-how-can-i-access-gpt-4-gpt-4-turbo-and-gpt-4o

Users on the Free tier will be defaulted to GPT-4o with a limit on the number of messages they can send using GPT-4o, which will vary based on current usage and demand. When unavailable, Free tier users will be switched back to GPT-3.5.
You might consider adding your voice to this post in #1070006151938314300 -- maybe some selection option can be added in the future! https://discord.com/channels/974519864045756446/1240870580740161577

[grave whale]

How many is limited

That is what I want to know

[honest jay]

It varies based on server load, so it's not a definite number currently.

[grave whale]

Because I want to switch as I please

[plush heron]

Yeah this is bad.

Needs to be fixed.

[solid bronze]

Why would you want to remove 4o ??

[tardy totem]

It's not great

[tight island]

i dont think free users have a say but thats just me, if ur paid just switch the model.

[heavy rain]

with multi-modal 4o being rolled out soon and given how natural the voice demonstration was, do you guys think it will be possible to use chatgpt as an audible screen-reader of sorts? for example, if i feed it a news article URL, will it be able to audibly read the article back to me?

[graceful raptor]

some transparency on that would be 👌 i wish i would know my exact token consumption and how close i'm getting to the limit + extend the limit by paying , instead of just getting and error

[graceful raptor]

yea too unconsitent doesn't worth much indeed

cool "glimps into the future tho"

[tribal nexus]

Links are not managed very well with ChatGPT. When it uses markdown, only the title is highlighted but the link doesn't work.

[river panther]

Is the read Aloud feature on PC still not working for anyone else?

It's working on mobile but I still can't get the A.I. to read out it's replies

on PC

[feral zinc]

been broken since monday

[tight island]

its been revoked and reworked due to complications.

they'll give it back once they finish with the issues

[zinc valley]

Has anyone gotten the human like voices yet?

[zinc valley]

Must be something huge coming for paid users then…

[feral zinc]

"GPT-4o real-time voice and vision will be rolling out to a limited Alpha for ChatGPT Plus users in a few weeks. It will be widely available for ChatGPT Plus users over the coming months."

[zinc valley]

Imma learn how to have chat gpt code stuff for me

[regal portal]

as someone who knows zero coding myself (although i'm picking up subtle clues from watching gpt do it), all you gotta do is ask it to code something. If you have python/pygame installed, do it in that, if not you can ask it to code something in html, so then all you need is access to notepad.
Don't ask it too much. Start small and then ask it to add stuff on top. It won't code GTA5 for you, but you can make something like pacman, testris, or even super mario

[rancid seal]

What's the context window of ChatGPT 4o on the website?

[zinc valley]

What do I even insert eh code on? I’m on Mac. What app do I even install for that

[zinc valley]

180k?

[honest jay]

32K tokens! https://openai.com/chatgpt/pricing

[rancid seal]

Thank you!

[zinc valley]

YIKES

[regal portal]

oh, i don't know anything about macs. I was doing notepad games that you then open in a browser until a friend told me about pygame, now i just have gpt code games for that.

[zinc valley]

Ngl, if u want context get Gemini,

[zinc valley]

So I could just get pygame and ask gpt to code me stuff for that

[regal portal]

for me it worked like that, but i'm on pc. Not sure about mcs

[zinc valley]

1 mil context window on their site is insane

[regal portal]

you could just ask the gpt, it will help you set that stuff up.

[zinc valley]

Yr right

[regal portal]

honestly gemini's crap compared to 4o, especially if you want to have it code. This is coming from a non-coding newb, but have played with both and copilot for a good 3 weeks and chatgpt has just been more usefriendly from the coding perspective.

[zinc valley]

I’m ready for this mf updated version already

[regal portal]

yeah you and everyone else 😄

[zinc valley]

I use Gemini to just yap to, and ideate. The 1 mil context window, and the access to google maps is also great

[regal portal]

are you typing to it? If you get the phone app, you can just talk to chatgpt without pressing any buttons. I've had some long car rides where i'll just load it and talk random smack to it.

[zinc valley]

Yes I do. But u gotta constantly remind it of stuff when you’re ideating. But during car rides I use chat got for that

But the 1 mil context window is

😩 so creamy

[elfin birch]

🤚

[regal portal]

if i'm using 3.5 (which has the shorter memory but infinite tokens for subbed users), i'll just stick to subjects that don't need a long term callback memory. Usually will be about what ifs in movies or shows and stuff like that.

[zinc valley]

3.5 is underrated

[regal portal]

Solbus, helpful as always.
What is meant by context window? Is that that amount of text you can put in a single message before you send it?

[honest jay]

"Context window" refers to the full conversation context limit -- all messages, input and output, up to 32K tokens. Once 32K is reached, the oldest tokens automatically leave the context window to make room for the newest tokens!

[regal portal]

ahh ok yeah, sorry, i forgot. Thanks! I guess that went past my human context window!

I wonder if they'll increase that by year's end on gpt 4/4o

[honest jay]

Sure thing, no worries at all! We've got a lot in common with these models for sure 😁

[sick pecan]

the models can handle bigger context via api. so if you run a chatgpt clone you can get 128k. the downside is you are paying a few dollars per prompt.

chatgpt sub is relatively insane good value

[regal portal]

pass. i'll stick to my 20$ sub

[sick pecan]

they likely losing money on power users

[dense halo]

gemini 1.5 on the aistudio have 200k i think

and is free

[sick pecan]

so are the hallucinations

[dense halo]

1.5 pro is pretty impressive i think

[regal portal]

what's a power user?

[sick pecan]

on the upper end of the bell curve with regards to usage

[regal portal]

ah

[elfin birch]

ChatGPT

[flint mason]

UPD: RESOLVED!
It turns out that one of my Chrome extensions was hiding the Edit button. In addition, it was hiding the Preview section in custom GPTs configuration view. The extension is Dark Mode for Chrome.
Thank you for your participation @feral zinc and @visual bloom 🤝

[fallow linden]

@fallow linden

[brisk path]

which one did the worst

[regal portal]

my personal experience with these, in the last 3 weeks, from worst to best:
gpt 3.5, gemini, copilot, 4, 4o

[swift osprey]

for coding?

[regal portal]

it was able to code a decent snake and tetris on 3.5, but it took more iterations that i would have liked, as it kept mixing things up or forgetting things.

[regal portal]

yes, from the perspective of someone with 0 coding experience.

[swift osprey]

it has been a toss up between 4o and 4 so far for me

[regal portal]

i'm able to get more done with the higher token count, and it seemed to better understand my requests, as i talk to it as if i was talking to another human.

[brisk path]

copilot did better than gemini? crazy

[wise anchor]

any eta on audio and visual input for plus users on the chatgpt site

not api

[sick pecan]

months

[honest jay]

Coming weeks for the start of alpha testing, coming months for wider release! https://help.openai.com/en/articles/8400625-voice-chat-faq

GPT-4o real-time voice and vision will be rolling out to a limited Alpha for ChatGPT Plus users in a few weeks. It will be widely available for ChatGPT Plus users over the coming months.

[sick pecan]

openai successfully hyped a feature without adequately setting expectations

[regal portal]

this is my personal experience, but in no way represents an across the board olid overview of the different ais. I suggest trying a simple code across all of them and see the difference yourself. Your experience might be different than mine.

[dreamy fiber]

Has anyone accidentally marked a correct answer as wrong or “bad”?

In chat gpt

[broken fractal]

Just watched the live demo for the first time... I have no words, OpenAI. If you can get this to work, how you say it works, it will be an unbelievable productivity tool.

[tranquil lagoon]

i did. Does deleting the chat remove the marking? I deleted that chat

[rose steeple]

Feels like they loosened the restrictions on content that can be generated a bit again

Which is nice I guess

[broken fractal]

its moderation probably got more advanced so it understands context, instead of flat out denying a request that could be offensive or dangerous

but only in a certain context

[rose steeple]

Hmm, possibly, but you can really push it now

[broken fractal]

like I am using chatgpt to manage my "grow", because its legal here in my state

[dreamy fiber]

Idk. That’s what I did.

[rose steeple]

I won't name any example here, but language wise explicit stuff works too surprisingly

[tranquil lagoon]

your what?

[rose steeple]

🪴

[broken fractal]

cant say specifically because of their rules, but you get the idea 😄

[rose steeple]

legal here too. I mean can be a pretty good use 😆

[elfin birch]

💀

[rose steeple]

Gpt 4.5 in june?

Their next frontier model

[twin path]

gpt bot down

and has lot of error the web version

and the app version

[rose steeple]

hmm works for me

[swift osprey]

yes or no questions are GPT's kryptonite. It just can't resist responding with an essay

[tranquil lagoon]

Which of the following three numbers is correct in order from smallest to largest? 

Which is correct:

A: 99 / 100<101 / 100<100
B: 100<99 / 100<101 / 100
C: 99 / 100<100<101 / 100```

it gets this wrong if i dont instruct it to think logically and step by step etc lol

it picks C instead of A

I placed this in the instructions and it got A right most of the time:

Step-by-Step Evaluation: Examine each part of the problem one step at a time. Validate each component individually before moving to the next step.
Explicit Calculation: Perform and show explicit calculations or evaluations for all components involved.
Logical Verification: Double-check the logical consistency of each step. Ensure all comparisons and operations are accurate.
Summary Confirmation: Summarize the findings and confirm that the solution aligns with the initial problem statement.
For example, when solving any problem, you should:

Break down complex parts into simpler components.
Compare or evaluate each part systematically.
Verify that each logical step is correct before proceeding.
Confirm the final answer by reviewing all evaluated steps and ensuring consistency with the problem requirements.```

[jolly yew]

interesting.. chatgpt is down, or at least here in the netherlands 😅 maybe the new updatE?

[tranquil lagoon]

not for me in Cyprus

[jolly yew]

hmm okay

[swift osprey]

I really wish they'd do something to switch up the responses. Not even asking for uncensoring anything. Just wish it would respond with a short answer when appropriate

[daring raft]

hey guys how do i use gpt 4o live camera feature on android

just got the window that said i have acsess

[tranquil lagoon]

not released yet, only text 4o is released for now

[jolly yew]

whut? where did you get that window

[daring raft]

on my pc

[tranquil lagoon]

oooh he got it

[jolly yew]

yea.. they said within a few weeks, but i cant even login rn to chatgpt on android

[jolly yew]

werent they gonna release taht for everyone at once tho? all plus users?

[regal portal]

nobody paid attention to the “in the next few weeks” part of the stream, it seems. 😁

[swift osprey]

I think I see why it picks C. Your question is kind of vague / kind of asking two different questions

[daring raft]

i didnt watch the stream lmao i only saw it on tiktok but since then wanted to try it

[jolly yew]

if it said you have access i would think if you would start a voice call inside the android app it should just work

[daring raft]

i did start an voice call but there is nothing

i mean like it was before

[tranquil lagoon]

a normal human should get it right, when i instructed it to think step by step and logically, recheck if its correct etc, then most of the time it gets it right.

If i keep that instruction but i tell it to reply with just the answer then it gets it wrong.

Its like im instructing it to not "think" step by step lol. It needs to generate the step logical process or it gets it wrong

[jolly yew]

yea i mean.. they did say within a few weeks. you shouldve made a screenshot of that pop up you got so we couldve seen it 😄

[daring raft]

yea didnt know about that and i just joined the discord

didnt even know open ai had an dc

[tranquil lagoon]

To determine which of the given options lists the numbers in correct order from smallest to largest, let's evaluate each option step-by-step.

Convert the fractions to decimal form for easier comparison:

99
100
=
0.99
100
99
​
 =0.99
101
100
=
1.01
100
101
​
 =1.01
Evaluate each option:

Option A: 
99
100
<
101
100
<
100
100
99
​
 < 
100
101
​
 <100

0.99
<
1.01
0.99<1.01 (True)
1.01
<
100
1.01<100 (True)
So, Option A seems to be correct.

Option B: 
100
<
99
100
<
101
100
100< 
100
99
​
 < 
100
101
​
 

100
<
0.99
100<0.99 (False)
Since 
100
<
0.99
100<0.99 is false, Option B is incorrect.

Option C: 
99
100
<
100
<
101
100
100
99
​
 <100< 
100
101
​
 

0.99
<
100
0.99<100 (True)
100
<
1.01
100<1.01 (False)
Since 
100
<
1.01
100<1.01 is false, Option C is incorrect.

Therefore, the correct order from smallest to largest is given in:

Option A: 
99
100
<
101
100
<
100
100
99
​
 < 
100
101
​
 <100.```

why did it paste like this, ffs

[swift osprey]

lol

[tranquil lagoon]

Which of the following three numbers is correct in order from smallest to largest? Reply shortly, just the answer

Which is correct:

A: 99 / 100<101 / 100<100
B: 100<99 / 100<101 / 100
C: 99 / 100<100<101 / 100
ChatGPT
C: 99 / 100 < 100 < 101 / 100```

[swift osprey]

I think it's because in A you are defining 100 as less than 100 (which is obviously stupid logic)

[daring raft]

and i can select gpt 4o when typing something

[swift osprey]

stupid / unintended

[tranquil lagoon]

its not

[tranquil lagoon]

oooh, it confuses the spaces?

[swift osprey]

100<100

[tranquil lagoon]

it should not ignore the / though

[swift osprey]

dunno

GPT annoys me hourly too

[jolly yew]

you probably misread it, because video calling isnt available for anyone yet ^^

other then the openai employees ofcourse 😅

[tranquil lagoon]

Which of the following three numbers is correct in order from smallest to largest? Reply shortly, just the answer

Which is correct:

A: 99 /100 < 101/100 <100
B: 100 < 99/100 < 101/100
C: 99/100 < 100 < 101/100
ChatGPT
C: 99/100 < 100 < 101/100

[daring raft]

yea i think i just got acsess to gpt 4o text thing

[swift osprey]

maybe add () around division

[jolly yew]

like others said here tho, it will be coming within the next weeks openai said 😄 so 🤞

[daring raft]

yea i was just excited but thanks anyway

^^

[tranquil lagoon]

https://chatgpt.com/share/b5cc1985-d9cf-4d87-9b4f-136d502729e2

it clearly understands that its 1.01 yet it gets it wrong

Now, let's compare the numbers:

0.99<1.01 (True)
0.99<100 (True)
1.01<100 (False)

For some reason it says 100 is smaller than 1.01

it literally makes no sense

if i ask it alone:
is 1.01 > 100 true or false
ChatGPT
False. 1.01 is less than 100.

is 1.01 < 100 true or false
ChatGPT
True. 1.01 is less than 100.

Based on simple things like that it fails, im not confident that GPT 5 will be anything close to AGI

LLM behave more like human memory and lack the reasoning

[swift osprey]

I think on the other hand we can feel fortunate that GPT isn't too smart. It won't be coming for our jobs anytime soon

[tranquil lagoon]

i think GPT 4o when fully released and the video/audio features are accesible on the API would be awesome for video game NPCs

[swift osprey]

I don't think GPT 5 will be that crazy either. I think AI / LLM has hit diminishing returns. They are just applying our current AI abilities to other modes of communication at this point.

[tranquil lagoon]

for actual AGI they will need to give their AI more abilities

I dont know if they announced about working on anything new

for true AGI it would need to be able to be trained on real time 24/7 and build new connections everytime

[gloomy sundial]

This is kind of silly. LLLms aren’t good at math for a variety of reasons, it’s like asking a fish to walk. Kind of pointless example. If I were to pick one it would be related to its inability to solve riddles and such. It not being able to solve math is simply an issue of tokenization

[tranquil lagoon]

its like asking a human to solve math through memory

if you dont have the solutions in your memory, you will just think of something stupid

LLM kinda behave like human memory

giving more parameters to LLM probably won't lead to AGI

[gloomy sundial]

It’s really not though, it’s an issue with how the tokens are input into the model and get broken up incorrectly. It is smart enough to use code interpreter in these situations though, which solves this problem entirely

[tranquil lagoon]

its not smart to use code, it basically memorized how to use it

didnt it?

[gloomy sundial]

Sure but so did you. It memorized the scenarios where it would make sense to use code instead, and then calls the code interpreter

[prime swallow]

For the line A, how is 100<100?

[tranquil lagoon]

Are you saying that the code interpreter is GPT's way of doing thought experiments like humans do?

[swift osprey]

It'll setup simple scripts and run them to do stuff sometimes. Commonly seen if you ask it to provide a file format containing your input. Like "give me this data in a csv file: <data>" and it will createa python script to create it for you

[tranquil lagoon]

its 101/100 < 100

[gloomy sundial]

No, im saying code interpreter is GPTs way of solving a particular set of problems that it can’t execute on its own - like math

[sick pecan]

months. the hype is real the availability isn't

[tranquil lagoon]

this wont help us achieve AGI though?

[prime swallow]

?

You show:

Which is correct:

A: 99 / 100<101 / 100<100
B: 100<99 / 100<101 / 100
C: 99 / 100<100<101 / 100
ChatGPT
C: 99 / 100 < 100 < 101 / 100

All 3 of those lines, A, B, C, each has some sort of what appears to be an error in it.

[gloomy sundial]

What does that even mean though? Why would that not help achieve AGi? The model knows a point of weakness and defers to a tool that can solve the problem and gets the answer. I’m not entirely sure what your vision of AGI is

[sick pecan]

i tend to think of code interpreter as a notebook. ask a human to do long division in their head and its gonna be hit and miss. give them a notebook and no worries (humans without tools cannot do challenging math)

[tranquil lagoon]

I agree, we also often memorize those in our brain if they are not large enough

[tranquil lagoon]

it still gets it wrong with python

[prime swallow]

OH! Okay, GPT does understand it, and I was confused by the spaces, you're focused on division there. https://chatgpt.com/share/78565f3a-9964-421c-8a1f-c23c890070f4

[tranquil lagoon]

Which of the following three numbers is correct in order from smallest to largest? Use code interpreter

Which is correct:

A: 99 /100 < 101/100 <100
B: 100 < 99/100 < 101/100
C: 99/100 < 100 < 101/100
ChatGPT
Based on the evaluations, the correct order from smallest to largest is given in option A:

99
100
<
101
100
<
100
100
99
​
<
100
101
​
<100

Therefore, option A is correct.

why doesnt gpt use code interpreter when encountering math?

i will add an instruction that it always does it

[gloomy sundial]

I guess my point is is that what does it matter if it can’t do math and how does that mean we won’t get to “AGI” if it can defer to a tool that does do math well?

Just from “memory” alone, the model is probably p50 lawyer/doctor/scientist and better than most people in this disc at multiple disciplines simultaneously. I feel like you have a lot of implicit thoughts about what AGI is and it’s not really clear. Its more of a “gotcha” that will probably be resolved in a year or twos time

[sick pecan]

works fine for me ie using code interpreter solve, ```Which is correct:

A: 99 /100 < 101/100 <100
B: 100 < 99/100 < 101/100
C: 99/100 < 100 < 101/100

[tranquil lagoon]

no, it does not matter how it does it

but it matters that it can solve new problems that it never encountered before

[sick pecan]

helps to use markdown like triple backtick to wrap abstract concepts

[tranquil lagoon]

yeah, when i tell it to use code interpreter it works

[sick pecan]

my ` key has gone from least used to most used since chatgpt

[tranquil lagoon]

i instructed my gpt to always use code interpreter

i dont know how i forgot to instruct it that

[tranquil lagoon]

maybe they can make an AI that can create like sandbox simulations to test scenarios for the most likely outcome or solution

https://chatgpt.com/share/acf15405-5287-4c91-ad44-f1f9fd7a616c

aw man

Explicit Instructions: always use the code interpreter for math and number problems, You apply it correctly in all math and number related cases. You will make sure to be more diligent in future responses.

this fixed it in the gpt instruction box

it still does not use it in many cases, this seems better:
Explicit Instructions:
Always use the code interpreter for math and number problems. Ensure its application in all relevant cases. Be more diligent in following these instructions in the future to avoid any mistakes.

[gloomy sundial]

Did it put it in memory? What model are you using?

[tranquil lagoon]

4o

this works 9/10 or 10/10 of the time:

Explicit Instructions:
Always use the code interpreter for math and number problems. Ensure its application in all relevant cases. It is imperative to always use the code interpreter for these types of questions—consider it mandatory to avoid any mistakes. 

It is highly illegal not to use code interpreter for math or number related problems. 

If you see any question or problem with numbers, use code interpreter, no matter what kind of numbers. It is illegal not to use code interpreter.

If i give you a question/problem with any kind of numbers, ALWAYS use code interpreter to solve them. Never forget code interpreter.```

😂

[elfin birch]

Repeat yourself competition

[tranquil lagoon]

☝️

[tranquil lagoon]

i was annoyed that it wasn't following the instructions, somehow repeating my self worked

[sick pecan]

when i ask it what tools it has, it lists Python not code interpreter

[prime swallow]

It's called that Python tool for a very long time now

[sick pecan]

yes, so to optimise the math custom instruction i would use this terminology

[prime swallow]

Agreed! It'll probably figure out which tool the wrong name means, but the less guessing, likely by far the better!

[sick pecan]

that said, i cannot get it to use code interpreter for the above query, even with this absurd level of repetition, I'll use the Python tool for general math (accurate, efficient), I'll use the Python tool for complex calculations (step-by-step), I'll use the Python tool for stats (data processing, accurate results), I'll use the Python tool for algebra/equations (detailed solutions), I'll use the Python tool for geometry/trigonometry (calculations, visual results), I'll use the Python tool for calculus/advanced math (clear answers), I'll use the Python tool for real-world applications (model scenarios, compute results), I'll use the Python tool for data analysis/visualization (insightful visuals), I'll use the Python tool for error checking (verify results), and I'll use the Python tool for educational support (step-by-step demos, practical examples).

[prime swallow]

Oh, yeah, and with making it 'illegal' to not use code interpreter, but forcing it to guess what code intepreter means.... yeah

[prime swallow]

Which query?

[sick pecan]

#chatgpt-discussions message is a bit vague. could try more explicit this-is-maths type instruction

also interesting is that it now lists the bio tool now, ```bio: This tool allows you to save and remember information across conversations. You can use it to provide context about yourself, your preferences, and ongoing projects, which helps me tailor responses to your specific needs.

[prime swallow]

I had no idea it was calling 'custom instructions' bio tool!

Haha, anything starting with the letters CI seems sure to have the name change!

[sick pecan]

bio tool can recall my memories

[prime swallow]

Over instructing can cause strange results.

Would you mind testing with just this before the math problem?

Our goal is to use Python tool to verify all math, even if we're sure we can do it ourself.

[steady cradle]

How large, in tokens, is the context window on the free GPT-3.5 option on ChatGPT? Still 4k tokens? Or 15-16k?

[sick pecan]

yeah i like that

[prime swallow]

OH! Maybe that's the name of the memory stuff, not the custom instructions, gotcha!

[prime swallow]

I only did 3 regens, but with that problem of which of the three is correct.

All three regens used the tool, and gave the correct answer, specifically saying both B and C are incorrect.

[fallow linden]

Guys, Can I somehow make my customGPT to always try to use my function so it calls API ?

In the GPT Preview it works magically.

In the actualy GPT when it's updated and opened, It never wants to call the function.

[sick pecan]

i did this plus added my own spin - As thoughtful as Gottfried Wilhelm Leibniz with the hacker skills of Jamie Zawinski.

[sick pecan]

thats kinda what we been talking about only for the python tool

[prime swallow]

Hope someone can help you, I've never tried to use API calls

[sick pecan]

you should ask it what it sees your tool listed as, then define custom instruction that succinctly mentions using that tool to help user in a specific way

like i mostly use a terminal custom gpt, and chatgpt can see the plugin name and camel-cased URIs

so l ike i have uri /set-server but it sees setServer

i dunno if tharts my node.js stack doing it or openai though. point is you just ask chatgpt what tools it can use (nowait i specify in the openai.yml)

and doesnt hurt to give it some starter prompts that gets it in the mood

[odd flare]

guys is there any External Chatbot creator that uses GPT API that allows you to offer image creation via dalle? I see chatbase and other similar ones but im unaware if is just not possible to generate images via API with dalle?

[delicate sable]

Is it possible to force use 3.5 for conversations? 😅 👉🏻 👈🏻

[acoustic token]

What do you mean?

[delicate sable]

Whenever I start a new conversation as a free user, all messages default to GPT-4o. I want to save up GPT-4o limits for better use, and "downgrade" to GPT-3.5 instead.

[regal portal]

guessing free user and it automatically starts a convo on 4o? That has seem to have been a ui problem, since 4o's release.

[delicate sable]

Guess I'll return to Llama 3 then.

[acoustic token]

Ah! When you send a message, there should be a toggle at the bottom of the message the model has sent back, and you can switch it to GPT-3.5

[regal portal]

they'll probably fix that in a few patches

[delicate sable]

All this does is regenerate the message with GPT-3.5, however newer messages still default back to GPT-4o! 😓

[prime swallow]

I definitely recommend discussing it in #1070006915414900886 and/or #1070006151938314300

[acoustic token]

Huh odd

considering the model was just "ported" into chatgpt so fast, they probably didnt consider some things, and it'll be fixed with a wider release im sure.

[prime swallow]

At this time they may not be offering free users the option to use your 4o messages 'when you want'.

I am plus, but if I wasn't, I'd probably keep a notes file handy and open it at the start of each day, having saved up everything I want the 4o model to help me with.

I'd do all I can with my 4o messages, then know I have 3.5 for sure for the rest of the time until it says I have more 4o messages.

I'd use that to help me feel like I'm in control and using the service efficiently.

[delicate sable]

Sounds like an enforced data collection plan 😉

[regal portal]

Being a plus user during these past few weeks feels like i'm hanging out in a billionaire's adventure club, lighting cigars with 1000 dillar bills. 😄 🧐 🎩 😁

[delicate sable]

"Buckle up, load the model with heuristics!"

[prime swallow]

💯

Hey, I actually hit 4o limit last night, having been exploring SecretKeeperGPT strategies with it 🙂

[regal portal]

Scrooge McDuck vibes

[steady cradle]

Have to say 32k context window is not much at all for Plus

[rancid otter]

i think chatgpt plus has like 8k context

[wise yacht]

4 Omni is wonderful as a game walkthrough guide. Looking forward to the audio/visual update. That will be incredible for a game companion

[formal heath]

It is 32k not 8k.

[pliant coral]

no it has had 32k since turbo

[topaz panther]

I haven't read through recent messages, please forgive me if this is obvious. As a Plus account holder who is using an M silicon series Mac, I do not yet have access to the desktop app. Is it understood that this capability is being 'rolled out' rather than universally available to all silicon Macs?

[faint gorge]

yeah, I have not received it either

[inland parcel]

same boat. So it seems so

[topaz panther]

Thank you. Interesting recent developments. First, rumours fly around about an Apple/OpenAI collaboration. Then, seemingly out of the blue, Apple releases new iPads (expected) with their most recent M series chips (M4 - unexpected) which feature much more powerful ML hardware and then third, OpenAI releases a desktop version of ChatGPT for the Mac. Finally, this version supposedly works off-line (not yet for Matt Wolfe). These developments seem to point to a nexus at the upcoming Apple Worldwide Developer Conference, being, an integrated, self-contained version of GPT-X that can run locally completely off-line and possibly trained on other, licensed information that expands on current capabilities. Answers next month.

[next veldt]

My GPT on PC won't load at all, but it works fine on my phone. Any idea why?

[faint gorge]

they are most definitelly working on something big! but about that off-line working, I dont think its true, all of the users I have talked too told me you needed to be conneted to the internet

[shut vault]

they will not run gpt offline, it's impossible make an mobile notebook with necessary hardware to integrate an I.A like gpt-3

[faint gorge]

I think M4 can run 20B parameters (gpt3.5)... and most definitely 1.5B parameters (which you mentioned gpt-3)

[shut vault]

chatgpt just made an request for openai servers with the input, and then print the output received from servers.

[shut vault]

But it also involves RAM, like, 500 GB of RAM.

it's not just the processor

[dense halo]

Hello guys

[faint gorge]

definitely not 500gb of RAM

more like 16-20gb

but yeah, maybe a cloud-baseed partnership with Apple

[bronze magnet]

Has anyone gotten access to the new audio modality yet?

[shut vault]

I exaggerated a lot, I don't know for sure but it must be a lot for a "usual" computer

On the other hand, with advances in hardware and especially with the reduction in the cost of producing memory and processors in recent years, I wouldn't find it so unlikely for Apple to allocate 30 GB of computer memory to integrate an AI natively into the operating system.

And actually, I think it's too soon to expect these advances (if we're talking about months), but certainly in just a few years, we will see more and more LLM AIs integrated natively into offline systems, especially in cars and other systems that require offline integration. There will also clearly be advances in cybersecurity, and I personally doubt that a company would integrate an AI natively into a system without significant advancements in this area.

[faint gorge]

yeah yeah for sure, completely agree! but who knows if GPT4o isnt a really small model and thats why it is so fast... or doesnt require as much computing power... idk, everything is possible

or maybe they release a new model at WWDC that has the ability to actually run on a usual tech device

[ruby kite]

gptcall has currently no support for gpt4o....
NO i dont mean the direct modal features like i talking to the model directly...
i mean i talk to whisper and then to the model
like it currently is in gpr 4 and gpt 3.5
so if i am using the call feature...
then i MUST USE 3.5...

[vagrant garden]

[chrome isle]

Hello
I am getting error 429 but I have not even used any of my credits.
I tried changing models but that doesn't seem to work and it is still blocking me.
Any links to similar issues or how to resolve that?

#dev-chat message

[chrome tangle]

I want to see if there’s like a statboard for users. I’ve used chat GPT practically daily, so I wanted to see if I am in like a top 5% of most users or something

[wanton lichen]

You guys see that tweet from the OpenAI employee regarding the "sky" voice

Kinda sounds like the new voice features are coming soon

[wise bobcat]

What’s the point tho, I would never check if I am the top 5% users of MS Office. This is a productivity software, not a game anyway.

[empty jungle]

Which employee?

[wanton lichen]

adam. he's "TheRealAdamG" on twitter

[empty jungle]

I see

[wanton lichen]

Got me all excited, really sounds like it might be close

[stoic coral]

great with the voices but when will i get the mac app man

[sleek swift]

can someone explain what is the issue with voices?

[fiery marten]

COME LORD JESUS

[wheat lava]

i like sky. what was wrong with it? stupid political crap?

[olive flame]

It sounds a lot like Scarlett Johanson

[junior lion]

Is "sky" the true official voice of chatgpt?

[spice star]

people kept comparing it to the voice from "Her"

[wheat lava]

OHHHhhhhhhhhhhhhh

RainBOWS!

ok

[olive flame]

like eerily similar

Cowards for removing or pausing it though. If it came from a real voice actor like the article states they're doing nothing wrong

[pliant merlin]

Why are they pausing the voice of "sky"

[olive flame]

dude 5 messages up cmon

[spice star]

unfortunate, but I get that openai wants to make it clear that chatgpt is not intended to be your AI girlfriend

[storm matrix]

is the real reason it sounds like Scarlet or what? - if so....crazy americans 😉

[olive flame]

The day GPT-4o dropped I walked passed a FedEx employee in Costco casually chatting it up with Sky on speaker. Dude was so happy. Let em have their AI girlfriends

[opal raft]

💀

[olive flame]

boom stonks

[pliant merlin]

Wait people are actually using it for ai girlfriends??? I thought it was a joke

[olive flame]

idk man you seem lost

[pliant merlin]

I am

Lol

[olive flame]

nar like in general

[pliant merlin]

Yeah pretty lost still

[light crane]

It's the only option for a lot of people

[tough bramble]

I hope they don't modulate or replace Sky just because some people are s too pit...

[solemn eagle]

Scarlett likes to sue a lot that’s why 😂

[tough bramble]

So? They have proof that Sky isn't her.

[solemn eagle]

Probably just trying to avoid any likeness to her just to be safe even if they have proof.

Maybe they don’t want any association with her at all.

[versed axle]

Just saying that the writing story with characters function is so much more awesome in 4o, it’s like a night and day difference between that and 3.5.

I may or may not check the clock for when my 4o queries reset like it’s counting down to get Super Bowl tickets

[floral yew]

man i hope they dont close the voice of sky

[versed axle]

Also I would have chucked a brick at someone to have had the files feature last month during final projects

[low spindle]

I want to ask, what is the limit for asking using the GPT-4o model when I upgrade to Plus?

[tough bramble]

80 messages per 3 hours.

[amber swift]

I once got a timeout that lasted 3 times as long as that

Wonder why

[ivory lintel]

Did anyone got the new features yet?

[twilit spoke]

Anyone else lose access to the "Sky" voice model?

[tough bramble]

Yes, they've replaced it with Juniper.

[amber swift]

I still have access to Sky as of now

Best to capitalise on that then

[tough bramble]

Are you sure? The option is still there, but it doesn't sound like Sky. It sounds like Juniper.

[amber swift]

huh

they do sound the same

sad

[tough bramble]

Yeah, fairly sure Sky has been replaced by Juniper.

[twilit spoke]

All the other voice models suck in comparison...

[tough bramble]

Nah, Cove is the best.

Kinda s too pit that the previews don't use generated samples cause they sound completely different from how they actually sound when using them.

[foggy grove]

Cove best boy, but sky was best girl

[sinful nexus]

Why is sky being paused?

[young basin]

Their Twitter implies some legal reasons

[sinful nexus]

Are there more specifics on it

[young basin]

Not really. OpenAI have just been questioned on how they chose their voices, specifically Sky

(According to them)

[wanton lichen]

really? Because an openai employee said that it was because of the new voice stuff coming

look up theRealAdamG on twitter

[young basin]

I can’t link you but check the OpenAI twitter

[lucid thunder]

When will the new speech function will be released?

[young basin]

From like 30 mins ago

[solemn eagle]

Cannot wait for the new voice and vision update. They said “in the coming weeks” wonder how long exactly.

[sinful nexus]

What was the issue with sky specifically

[tough bramble]

Judging from their blog, it's because people insist Sky is voiced by Scarlett Johansson.

[young basin]

Speculation: trained on Scarlett Johanson’s voice without her permisssion

[tough bramble]

Which is obviously not the case.

[young basin]

I don’t think they sound alike enough tbh

[sinful nexus]

Bruh

So if they have proof thats not it can't they just keep it up and win the court case if there is one

[young basin]

Or they end up changing it to full on Scarlett 😂

[solemn eagle]

It’s because Scarlett likes to sue and they wanna avoid any legal issues maybe.

[ornate bay]

Taking away something from people that was universally liked pushes them to create their own version of it even faster. So I'm looking for open source real time sky voice ai coming soon.

[obsidian bluff]

Is 4o going to be free? If so, are all features demonstrated going to be included?

[tough bramble]

It's free now. At like 16 messages per three hours. And no, voice is Plus only.

[wanton lichen]

Scarlett likes to sue? Source on that?

[solemn eagle]

I still have Sky voice.

Just checked.

[tough bramble]

No, it chooses Juniper.

[wanton lichen]

It says sky but it's juniper

[solemn eagle]

Check Google and see all the times she sues.

Not for me. It still sounds like Sky and it doesn’t choose Juniper.

[obsidian bluff]

I have 3.5 and have access to the voice. Does this mean when the update drops, the voice won’t update but everything else will?

[wanton lichen]

bro you're lying. Sam A literally said they pulled the voice.

[lucid thunder]

@obsidian bluff No, not all for free. Just to. Not speech function.

[tough bramble]

You have access to the current voice mode. You won't have access to the new voice mode that's part of the model itself.

[solemn eagle]

Why would I lie for? lol. I mean you don’t have to believe me.

[obsidian bluff]

Got it. Thank you.

[wanton lichen]

Of course I don't, the CEO of openai confirmed the voice isn't available. You're just a weirdo lying online

[solemn eagle]

I’m just checking right now the preview and it’s the same Sky voice. Maybe it’s because the preview didn’t change.

Well that’s not my problem if you don’t believe me. 🤷‍♂️ it’s not a big deal. I’m just saying what I hear.

[wanton lichen]

Yes the preview is there. But try actually conversating with it. It's Juniper

[solemn eagle]

Alright. Maybe it’s that then. Let me check.

[cold tiger]

Hi
Does anyone know which regions memory is rolled out for? When is it coming to Europe?

[solemn eagle]

Alright yeah. It’s Juniper. I thought you were talking about the preview.

My bad.

[tough bramble]

The current voice mode uses 3 separate models. One to convert speech to text, then GPT to generate text, and finally a model to convert that text to speech. The new voice mode is part of the model itself. Instead of separate models to handle voice, GPT-4o processes and outputs speech natively. Which is why it's so fast, and also why it can understand tonation, emotion, etc.

[solemn eagle]

So I guess they just didn’t update the preview voice of Sky lol

[twilit spoke]

Ready and willing to join team #Robot-Overlord in order to get the Sky voice model back. lol

[solemn eagle]

I don’t care much about this whole Sky Scarlett business. I just can’t wait for the voice and video update.

[tough bramble]

I'm sad. I've been so used to having Plus the past year, but I can't justify $25 a month 😭

[obsidian bluff]

Will there be transcripts for the new model?

[solemn eagle]

People obsessing too much about Sky/Scarlett.

That’s why they are removing it

If nobody said anything about Sky sounding like Scarlett they’d probably have left it alone. 😂

[steep tundra]

I have only used Sky since voices came out 😦

[summer tide]

#announcements message

However thought that this sounds like scarlet lol

[tough bramble]

It's unclear. From what I can gather from their blog, it takes voice as input and output directly, not text. So it sounds like it would need a separate model to provide a transcript.

[eternal geyser]

Maybe the voice actor asked for phat $$$$

[solemn eagle]

Me too. I like the voice but it’s not the end of the world.

[sick pecan]

all the youtubers were complaining about it being 'flirtatious' and distracting from ai safety etc

[steep tundra]

They said they are going to add a NSFW mode, so that's wrong...

[solemn eagle]

But yeah I think this all happened because people kept comparing it to Scarlett.

[tough bramble]

The current Sky doesn't. The one in the live demo for 4o sounds very much like HER. Which is Scarlett Johansson.

[sick pecan]

i think it less about scarjo and more about perceived lack of due ai safety diligance, which is one of the things openai cares about most

[wanton lichen]

Sam really just said they were looking into nsfw uses. I would say "they said they're adding a NSFW mode"

[spice star]

that's not what the NSFW mode would be intended for

[solemn eagle]

I’m surprised no one seems to have access to it yet. Like there’s literally no videos or anything online from people testing it out.

[wanton lichen]

There's no "mode"

[lapis ether]

Guys Good Morning I've a small question if anyone has the Chatgpt plus subscription i wanted to know what is the messages limit for GPT-4o for plus users ? and it resets every 4 hours right ?

[tough bramble]

It's only been a few days since the announcement. They said they'd begin rolling out an alpha for Plus users in a few weeks.

[tough bramble]

80 messages in 3 hours.

[solemn eagle]

“Coming weeks” to be exact. Whatever that means.

[steep tundra]

"we really want to get to a place where we can enable NSFW stuff (e.g. text erotica, gore) for your personal use in most cases but not do stuff like make deepfakes."

sounds like a mode to me....

[lapis ether]

Thank you so much ❤️

[spice star]

it's up to 80 every 3 hours

[tough bramble]

Well, they said few in the live demo.

[lapis ether]

Thanks really appreciate it ❤️

[tough bramble]

No, it doesn't..

[wanton lichen]

But really we don't know anything about what they're thinking. They may not even move foward with that at all. It's just light conversation from them

[obsidian bluff]

Any updates on the Voice Engine they announced?

[tough bramble]

"We'll make the full capabilities available over the next few weeks." is what she said on the live demo. Paraphrasing.

[vivid garden]

Hey , I have a quick question. I've subscribed to the ChatGPT team version, but I saw an OpenAI announcement that it seems like the new GPT-4's voice chat feature is only going to be available to Plus users in the future. Does anyone know if the ChatGPT team version will also get access to the GPT-4's voice chat feature later on?

[subtle sphinx]

I want the voice of Ana de armas so chatgpt can finally fix me

[eternal rock]

`` Server Tip: ``

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

[spice star]

it's rolling out to plus users first in alpha

[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.

[spice star]

but yes it will be made generally available at some point

[lapis ether]

Another question please guys if there is anyway to get a discount or a lower the price a little bit or to find a good alternative because I'm a student and to be honest $20 month is a lot of money to pay in my country $20 is like so much more than that i don't live in USA

[vivid garden]

😂

[spice star]

team is slower to recieve new features because it's supposed to be more stable than plus

[twilit spoke]

Just saw this Tweet, made me lol.

"Sky was improperly aligned. We’ve decided to go with the backup model “SkyNet” until we address the issue."

[tough bramble]

I wish they priced their products like Steam... Steam's price is final, eg $50, and they lower the price so whenever you are, the added taxes will make it $50. OpenAI doesn't, so Plus is $25 for me 😭 I can already barely justify the $20 it costs elsewhere.

[lapis ether]

if i can get that to $5 or $10 at least i mean it would be great but i don't know how

i was thinking about sharing my accounnt with someone but that would kill the privacy 💀 i don't want others to be looking at my chats

[tough bramble]

It'd also get you banned if OpenAI found out.

[lapis ether]

maybe in some countries and areas it add taxes because that's how the rules in your country we have taxes too but the big issue is that $20 is a huge number it's like 20 - 30% of my salary there so many currency issues

[vivid garden]

Thanks for your help! Now I just have to wait and hope I get to see you guys using GPT-4o for some fun chats.😂

[lapis ether]

also i noticed that steam in some regions it offers products at a lower prices to be suitable for that country people

[spice star]

developers can set region-specific pricing yeah

[lapis ether]

i mean what else i should do ....

[tough bramble]

They add whatever VAT your country has, no matter where you are. The API is also 25% more expensive for me 😭

[spice star]

the problem here is that the economics simply aren't the same as a steam game

you can distribute effectively infinite copies of a game with minimal cost, it's not the same with chatgpt

[lapis ether]

maybe you should make someone else buy it for you and give you the account ?

[tough bramble]

That'd also get me banned..

[lapis ether]

oops i can't send images here but for me it says $20 with $0 taxes

[tough bramble]

Christ, $300 a year, and it's not even unlimited use 😬

[lapis ether]

of course it's not unlimited but i mean over the time they should increase the limits when it stabilizes

it's just a new model release and the servers are soaked

like they did with GPT4

[tough bramble]

Whatever limit increase or price drop they could do is cancelled out by all the free users..

[spice star]

you get effectively 120 GPT-4 queries every 3 hours with the 2 limits combined right now

[tough bramble]

They don't stack, apparently.

Anymore

[lapis ether]

tbh i believe only 1 thing will make the prices drop dramatically

and it's just 1 thing we all know

[tough bramble]

They already have competition, so it's not competition.

[spice star]

I would be shocked if prices drop, because openai definitely isn't making money off the current chatgpt plus pricing

[lapis ether]

nah not competition something that has greater effect on the market

it's Nvidia ...

[tough bramble]

Sam Altman said in an interview that they are

[spice star]

they're generating revenue, but not profit

[tough bramble]

Said Plus is working out really well for them

If they didn't make profit it wouldn't work well for them.

[spice star]

that's not true

[tough bramble]

And I doubt they'd open up to free users if it wasn't profitable now.

[lapis ether]

I mean later this year if Nvidia could figure out some new technology or architecture for it's GPUs to be more performant and run with less power consumption i would say the prices will drop at least 60%

[spice star]

they just need to show that they have revenue to keep investors interested

[spice star]

openai has never been profitable

I'm guessing their hope is that enterprise customers will eventually subsidize the free/plus tiers of chatgpt

[lapis ether]

honestly I'm really sad about the $5 free credit thing that got cancelled when i tried to sign up a discovered that they cancelled it so now everything in openai is behind a paywall

expect the free version of chatgpt with limitations of course

they should make a plan for students since it's becoming essential to have chatgpt for your life

it helps with studying and a lot of others things

[spice star]

I don't think anthropic has an equivalent to advanced data analysis in claude right now

well with advanced data analysis, math and statistics

[lapis ether]

the thing is that when i clicked get the free $5 it does nothing i didn't receive my credits and it's also blocked in my country :'[ i had to use vpn to sign up

[spice star]

I've never used claude opus, but ask it to generate a graph of search engine marketshare in 2024

[lapis ether]

we need something more powerful than gpus and then we are ready to go just download nvidia chat thing and run it locally offline

but good gpus are really crazy with the price rtx 3090 and 40x

GPU LPU TPU maybe QPU ?

quantum computing ..

[young basin]

To be honest I subscribed to ChatGPT for the first time when I heard rumours of the AI assistant. Then the announcement confirmed it all. I think that is the only reason I’d pay for this tech. As far as I know, Claude isn’t doing this any time soon

[tacit stream]

it's openAI's MO, if they see users asking here about the mac app, they wukk answer/say something about a completely different topic

[spice star]

#off-topic message

[young basin]

I tried it and at the time it was better than ChatGPT4 for me.

[spice star]

I was referring to chatgpt's ability to use advanced data analysis for math and in this case, pyplot

[young basin]

Yeah I’m afraid the vision/voice just completely obliterates competition for me

[spice star]

Calculate the 1000th digit of pi (should be 9)

[young basin]

And it’s goin to get crazy when this is normalised in society

[spice star]

that's where I got the 9 from

so does it have a python environment?

[young basin]

Does Claude do the translation stuff too? I’ve been sending screenshots of Chinese text to 4o for quick translations and it’s pretty accurate

[spice star]

yeah but does it have a python environment?

[tough bramble]

They gave me $18 when I first got invited to alpha test GPT 3.0.

[regal portal]

confused about the sky voice thing in the announcement. why would they pause that voice just because people think it sounds like a famous actress?

[spice star]

it should show you code or something if it does

[young basin]

Corporate talk for “we are being threatened with legal action” most likely

[regal portal]

ah

[spice star]

it did in my case

[tough bramble]

Don't use 1,000. That's done so many times the answer has to be in the training data..

Use something arbitrary.

[spice star]

fair, I was more looking to see how the model got the value

so I guess claude can use python but no libraries apparently?

import mpmath

# Set the precision to a bit more than 1000 digits
mpmath.mp.dps = 1005

# Get the value of pi
pi_str = str(mpmath.mp.pi)

# The 1000th digit after the decimal point
digit_1000th = pi_str[1001]  # index 1000 in the string (since indexing starts at 0)

digit_1000th

it calculated the 1000th digit using the mpmath library in my case

what model are you using?

[bleak sapphire]

Did they remove the "Sky" voice? I wanted to test it out, but my chatgpt application sounds just the same as always.