gpt-models

[formal spear]

You still can't do the image testing

[crude dew]

Anyone know where they are in terms of GPT-4 API rollouts? I have GPT-4 chat access through the web via Plus, but really would like to start using the API. Are they still gradually sending those out?

[fading holly]

Yo since gpt4 has the power to web browse, what if you provide it a image address link, I tried it last time and it described a figure perfectly but when I tried again it said unable.

[gentle jetty]

hi gpt

[hoary bloom]

does everyone have access tot dalle

[wind urchin]

no it doesnt. it pretends to. you still need to copy paste the page content

[fading holly]

oh thats interesting very interesting. What if you describe a figure (very detailed) would it then give an answer based on your descritption or actually try to browse the internet for a figure similar?

[wind urchin]

dont know. eveytime i provide it with a url it seems to know the meta data of the website but if you question it on the content it says something wrong

[storm forge]

my credit card was rejected when trying to upgrade

[gentle jetty]

How to use this platform

[wind urchin]

so basically you

[formal spear]

and then you combine some

[primal iron]

i can hlep you ，lool

[wind urchin]

then finally the last step is

[vast sorrel]

Fr

[storm forge]

how?

[formal spear]

then you should finally have

[hardy bison]

How to connect the freighter to shore power

[modest verge]

GPT 4 is giving a lots of response errors and stops generating results after a few minutes.,

How to fix that? 😦

[molten sorrel]

🙏 they install more servers soon

[wispy condor]

Looks like there's a partial outage: https://status.openai.com/

[hoary bloom]

playground working well

[modest verge]

😄 😄

[dusty sapphire]

What's the total number of tokens one can submit in a single message on ChatGPT 4? I don't think it's close to the 25k they mentioned in the release. If you use the playground is it different than ChatGPT?

[barren venture]

4096

[dusty sapphire]

Thanks!

[wraith umbra]

😆

[hazy orchid]

hey, can anyone with gpt premium or whatever tell me if the image functionality is actually available yet

[hallow owl]

no, it's not released yet

[hazy orchid]

ok ty

[upbeat sail]

I received the following email "You're invited to use the OpenAI GPT-4 API!", but I'm not sure what to do next, could someone explain?

[modest verge]

How you got it?

[upbeat sail]

wait list

[hazy orchid]

my suggestion is first check it isn't phishing

[upbeat sail]

[e-mail]

[loud dagger]

Only 11 out of 25 chances were successful, and the rest were lost after only pressing the regenerate response button due to a server error. Jesus...

[wind urchin]

Even I got mine recently and it seems legit

[feral stump]

As someone who just received access to the gpt4 API, is there currently an app similar to the one used in the demo that allows us to use our API in that format/editor?

[upbeat sail]

Is there any chat here on discord that I can send a print of the email?

[hallow owl]

ask your question in #1037561751362863144

[feral stump]

Thank you I appreciate it

[modest verge]

Is anyone using GPT 4 as their coding assistant?

[crisp elm]

I am a plus user now. But today I got an email saying I can access the GPT4 API. Isn't the GPT4 model what I've been using? The GPT I've been talking to is black.

[vale lake]

Is there an add-on or widget that allows GPT-4 to access the internet to scan pages in real time?

[upbeat sail]

Are you referring to an API?

[tight cliff]

im not, apparently its good though

[vale lake]

I've used it to generate some code for HTBasic

it's ok

[sonic root]

I am. You want to have some way to test stuff, and you have to ask for small sections at a time. Also it tend to make up variables after awhile.

[unkempt moon]

Hi, I am asking gpt to make a basic edit to some html and it returns only part of the content. Any thought why?

[high wedge]

because gpt4 is slower than before

maybe too much demand

[unkempt moon]

Gpt3.5 can’t do it either , it keeps conking out.

It wasn’t doing that before

[twilit sigil]

maybe it reached a limit? try saying continue

[unkempt moon]

Ok

[wraith ginkgo]

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

Please report any incident by sending a DM to @open crescent immediately.

[worthy parcel]

i got same email

[unkempt moon]

Nope doesn’t work.

[twilit sigil]

go here
https://platform.openai.com/playground?mode=chat&model=gpt-4
or use the api programmatically by looking at the docs

[worthy parcel]

thanks, but my usage grant expired :( roughly how expensive is gpt 4? I couldn't find it

if I want gpt 4 to remember the previous messages in a conversation, will i be charged for the entire length of the whole conversation? or only the new text

[agile eagle]

ChatGPT and GpT4 are two separate things. I could answer this for you, but just ask what is the difference in your Chat4 😉

[vestal otter]

Gpt4 is Chatgpt its just a different version which the regular plan uses Gpt3.5

[fervent nebula]

man GPT4 is a dream for us basic python coding knowledge people. Im just so ADDICTED to it creating fun programs. I love this technology, sometimes I cant stop thinking about it like how revolutionary its going to be

[crisp elm]

Are they different? I don't understand. Thank you for your kind response.

[unkempt moon]

I’m using the chat interface , is there a different way ?

[unkempt leaf]

Hi, does anyone giving image as input for GPT-4 API

[twilit sigil]

https://openai.com/pricing $0.03/1k token for 8k gpt4 prompt

[agile eagle]

With current high demand and usage, many users have been experiencing chat halting responses or crashing…it was happening to me personally a few days ago…since then it’s been fine for me *knock on wood. Just ask it to continue where it left off if you didn’t have a crash error

[worthy parcel]

ty

[agile eagle]

ChatGPT and GPT-4 are closely related, as ChatGPT is an application or implementation that utilizes the GPT-4 model as its underlying technology. However, there are some differences between the two, which can be explained as follows:

GPT-4: GPT-4 (short for Generative Pre-trained Transformer 4) is a state-of-the-art AI language model developed by OpenAI. It is the fourth iteration in the GPT series, with each version improving upon the capabilities of the previous one. GPT-4 is capable of understanding and generating human-like text based on the input it receives. It has been trained on vast amounts of text data and can perform various natural language processing tasks, such as translation, summarization, question-answering, and more.
ChatGPT: ChatGPT is a specific application of the GPT-4 model that is designed for conversational AI purposes. It leverages the power of GPT-4 to engage users in interactive and dynamic conversations. ChatGPT has been fine-tuned to generate coherent, contextually relevant, and meaningful responses in a dialogue format. While it uses the GPT-4 model as its foundation, ChatGPT is tailored to provide an engaging conversational experience for users.
In summary, GPT-4 is the underlying language model that powers various applications, including ChatGPT. ChatGPT is a specific implementation of GPT-4 designed for conversational AI purposes, optimized to provide a more interactive and engaging experience for users.

[unkempt moon]

I keep asking , and it keeps getting stuck. I just upgraded to the pay version hoping it would fix it but it is still not working.

[tiny sequoia]

How do i ask questions to the ChatGPT AI?

[upbeat sail]

The model: gpt-4 does not exist

[twilit sigil]

you need to have api access

[unkempt moon]

Was there anything like this before got to help coders ? It’s amazing even though it crashes all the time.

[crisp elm]

I'm on the waitlist and I got an email today saying it's available, so I hit the Get Started button in the email and all I get is a description of the model. Do I go to the playground and do it?

[upbeat sail]

i received access

[unkempt moon]

Before gpt

[fervent nebula]

wouldnt be surprised of this was generated with chatgpt

[worthy parcel]

no way bro sent a chatgpt reply

[fervent nebula]

😂

he got gpt'ed

[worthy parcel]

in fact i could be gpt right now

[fervent nebula]

scary times folks!

[thorny valley]

Microsoft 365 Copilot is already out or people bsed? I just got the subscription and cannot find the copilot sidebar. 😦

[timid flicker]

Hey all. I have a couple of questions:
1)All of these companies creating their own models, how do they fine tune and have a private model? How can I create my own model with my data, and keep it private to me and users I intend to share with?
2) If I have an app that allows multiple users to access Chatgpt, how do I maintain individual sessions for each? Like what if I want chatgpt to remember the conversations individually?

[timid flicker]

I thought it’s still limited to few partners

[thorny valley]

I think its working.. I will see

[twilit sigil]

2. you will have to save the history in the session then send the whole history to gpt

[thorny valley]

I think it works but it doesn't go to the side, it's at the middle and bottom

[timid flicker]

Did you get it through your personal account? I was under the impression only corporate licenses are getting that

[timid flicker]

Ah so the app would have to save each token I guess

[thorny valley]

is it? 😦

[agile hinge]

will i be charged from the gpt4 api. I am chosen

[hollow verge]

siuuuuuu

[agile eagle]

https://platform.openai.com/docs/api-reference/introduction

[thorny valley]

It's working like the sidebar possibly. It's creating docs in word from scratch. I have to try the rest later. 🙂

[timid flicker]

Oo sweet!

[limpid sundial]

if you want to create your own model, you need a lot of vram, gpt-3 size model requires around 800 GB of vram

[agile hinge]

?

[twilit sigil]

yes https://openai.com/pricing

[pliant spear]

is there any documentation about the images? I'm trying to make a discord bot exactly as the openai developer stream

[twilit sigil]

not public #gpt-4-faq

[agile hinge]

but i didnt give any cc information so how can i be charged

[twilit sigil]

it won't let you use it then

[uneven birch]

Theres numbers for it, its >1tb afaik for training, 350gb for lora

[uneven birch]

Literally ask GPT-4

[junior elbow]

GPT-5 will end the game

[uneven birch]

Like, it can tell you

[timid flicker]

If I wanted to help my company setup it’s model, is there a process we can follow? I think the hardware reqs will be evaluated but atleast if there is a way we can find the will

[paper citrus]

I received a refund of $20 for the ChatGPT subscription.It was on the 25th. Is this normal?

[crisp elm]

Your explanation made perfect sense to me, thank you. Have a great day! 😍

[uneven birch]

Search for LLaMA, but thats research only

[agile hinge]

and also it is based on 1k token so if i use 999 am i going to be charged

[uneven birch]

Your best bet would be to do something like GPT-neo-X or something similar

[upbeat sail]

How many hours ago did you receive the email?

[red whale]

same thing here

[odd spear]

I never said chat4 was worse

[uneven birch]

or pay openai for a finetuned model of GPT-3

[twilit sigil]

https://platform.openai.com/playground?mode=chat&model=gpt-4

[red whale]

Thx

[agile eagle]

Funny thing is, you can ask these questions and receive step by step via Chat 4 😘

[limpid sundial]

i mean you can use openai models, you can fine tune them for specific tasks, and make thek know more by embeddings

[crisp elm]

4 hours ago.

[thorny valley]

I don't see the chatGPT help in powerpoint.. 😦

[worthy parcel]

teseting popping off tbh

[uneven birch]

copilot isnt out yet

[limpid sundial]

I think a lot of people got access to gpt-4 today

[thorny valley]

people said it was, they are scammers 😦

[timid flicker]

Ok I’ll check that out. But essentially what MS and others are doing is training chatgpt with their own dataset and then somehow making that model available privately, is that accurate?

[wind urchin]

Is it just me is it’s the Bing chat GPT worse than the one on open ai website

[uneven birch]

well yes but actually no. They give openAI the data, and openai finetunes it for them

[upbeat sail]

For those who received it a few hours ago, when you try to run something in the playground, it will appear: he model: gpt-4 does not exist. Give it a few hours (24 hours) and try again.

[uneven birch]

Go check out OAI azure cogitive services

[worthy parcel]

it's fine tuned to be an assistant

[thorny valley]

when is it going to be out?

[uneven birch]

Nobody knows

[thorny valley]

😦

[uneven birch]

But the private part is correct as far as i know

[timid flicker]

Oh! I just am so confused by these private offerings. Like I thought every interaction with chatgpt in turn trains it and the updated model is available to the users. But it seems that OpenAI can train the base model separately for different needs and have that separation in the models

[uneven birch]

Nope

You have to finetune it beforehand

What they usually do is just tell it a system message

You might not need to finetune GPT-4 for example, because its already good at many task

Whats your company doing

I would honestly suggest going with OpenAI for now, as the next best model (LLaMA) will get you into trouble, and everything else is around GPT-3/2 performance

There isnt any open and free to use models out there, unless youre looking into directly competing with OAI, training your own model from ground up

[timid flicker]

So let’s say there is a medical research team that wants to use chatgpt. They have historical data on prescription and diagnosis and lab reports and what not. What would that company need to do to train chatgpt with this private data, and have that model available only to its employees?

[uneven birch]

You dont even need to do that

Theres plugins that allow it to search the documents

And techniques like chunking to allow only the relevant parts to be used

[timid flicker]

But those plugins will feed the data to chatgpt right, making it accessible to others?

[uneven birch]

No

[maiden parrot]

how? i want a refund hahaha

chatgpt plus is disappointing

[uneven birch]

Think of it as a google extension

[timid flicker]

Ohh ok

[limpid sundial]

you can use embeddings in order to make gpt models access your data

[uneven birch]

The extension might let you do things like search with chatGPT in the search bar, but it doesnt let others see what you are doing

[timid flicker]

Damn I have understood everything incorrectly

[uneven birch]

As long as you keep the plugin private and secure, you should be fine

[uneven birch]

Yea…

[dark geyser]

ChatGPT is saving me way more than $20 per month...

[agile eagle]

@timid flicker essentially, you still need to do the hard work via API to fine tune the model

[uneven birch]

You might want to talk to someone who Works at OAI for help

I would suggest starting off by sending an email to their support

[uneven birch]

If youre doing research, theres grants and stuff for that

[maiden parrot]

no difference using free tier vs plus... :c

[timid flicker]

Thank you so much for the guidance! I’ll do that

[uneven birch]

Would def suggest you reach out to them

[timid flicker]

So there’s some arbitrary threshold between using plugins and fine tuning then?

[timid flicker]

Yup absolutely

[agile eagle]

@timid flicker to my knowledge, yes. Plugins are used to enhance the current base model. You and your company, need to evaluate your internal data and structure it so, so that it can be implemented developing thus model via API

[uneven birch]

I mean, this decision to use plugin vs finetune could probably be evaluated after getting into contact with OAI

🤷‍♂️ just my 2c

[true sail]

Hello Guys

[dark geyser]

Except that the free tier doesn't have GPT-4 currently as far as I know, and even before that, Plus gave me the reliability I need compared to free constantly being at the limit and it's hit or miss if you can log in.

[true sail]

Anyone have ChatGPT-4 Discord server ?

[timid flicker]

Yup makes sense

[toxic tusk]

anyone know how to get access to browsing mode?

[worthy parcel]

you're in it...

[worthy parcel]

not out yet

[wind urchin]

😊

[timid flicker]

There are a lot of unknowns atm and we are just trying to tread slowly. We want to do the evaluation but don’t know what to evaluate for

[toxic tusk]

ive seen people get it, which is interesting

[true sail]

but where is OPAN Ai Bot ?

[stone roost]

Why we need OPAN AI Bot ?

[sweet tangle]

Does this have an app or can i only use it on the web?

[hoary marsh]

Are the temperature/frequency settings that ChatGPT uses for gpt4 public/known?

[fair iron]

OpenAI doesn't have any apps

[vernal void]

No specific app
Web is good bro

[fair iron]

this information isn't known sadly 😔

[maiden thistle]

I got my Gpt4 access from the waitlist 😄

[whole kettle]

You can use it on your phone's browser too, you do not need an app.

[agile eagle]

If your goal is for the end user, whether external or internal, to access your company’s data that are proprietary to your company and not public knowledge, and want the end result to be able to understand this data effectively, you’ll def have to start dissecting what it actually is you want to accomplish… sorry if I am misunderstanding you though

[wind urchin]

how to became a memer

[cosmic elm]

anyone here have browsing mode yet. See some people on twitter have it

[wind urchin]

hi guys ignor the gpt we human will chat

[tribal totem]

Hey Whats Up Fellas.

[maiden parrot]

hey, wasnt gpt-4 in chatgpt supposed to be up to date and aware of data passed 2021?

[fair iron]

GPT-4 came out with the September 2021 cutoff

[tribal totem]

Hey where is the section we ask gpt queries in Discord??

[fair iron]

you cannot use chatGPT on this server

[sour rune]

welcome!

[maiden parrot]

so what's the point of gpt-4 then...? i see no big improvement

[high sky]

is that practically useful to fill that request higher cap form?
i did but no response

[worthy parcel]

ask chat gpt it will make one for you

[fair iron]

I don't work for or represent OpenAI, so I can't answer on their behalf. Quite a few people have seen improvements with it though

[worthy parcel]

look at their research paper

[agile eagle]

If you don’t see any improvements, it may be wise to look at how you are currently prompting the models…IMO 😘

[peak carbon]

Has anyone found a good application geared for working against large text? The ones I’ve played with have been somewhat underwhelming with hallucinations

[ashen furnace]

How GPT accesses office, does anyone know?

[fair iron]

GPT cannot access Microsoft Office products

[shy pagoda]

Zapier will be able to do this for you if you have access to pluggins

[wraith ginkgo]

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

Please report any incident by sending a DM to @open crescent immediately.

[coral birch]

It is far more sophisticated in crafting it's responses

The difference between 4 and old chatgpt feels stark to me

[true sail]

ok brother that you

[wintry urchin]

I've been away a few days, but have come back to GPT-4 running really slow and it also seems to be producing much less content (i.e. before you have to prompt it to "continue"). Has this been an issue over the last few days? Are we being "throttled" somehow? (I am a paid-up user)

[deft pendant]

HLO

[tired forum]

is there a limit to the amount of tokens/words GPT4 can give in its response?

[uneven birch]

8k/32k

[tired forum]

i thought that was on the user's side

[uneven birch]

??? That’s everything

Input, output, context

[tired forum]

so ChatGPT4 is limited to 8k tokens in its response?

[fiery furnace]

hewo 😉

[uneven birch]

Depends on the model but generally yes

[worthy parcel]

yes we are

[viscid fiber]

hello gpt4

hello how to use gpt4

[fair iron]

Learn more about GPT-4 here: https://openai.com/product/gpt-4

[magic yarrow]

I feel like the GPT-4 on Bing is somewhat nerfed

The Bing GPT-4 variant frequently gives me what i interpret as attitude and refuses to keep talking to me if i question it too many times on its accuracy or reasoning behind its responses. ChatGPT-4 and 3.5 always gives me nuanced responses no matter how much I challenge it.

[feral meteor]

i got gpt-4 invite but i am not able to access it

[prisma olive]

Me too

[feral meteor]

@fair iron can u help?

[prisma olive]

Well I have the same issue

[feral meteor]

hm

[hoary jolt]

how does some people provide images to gpt4 ?

[magic yarrow]

Don't we have to pay for GPT-4???

[hoary jolt]

i saw a reddit post of someone giving an image to gpt

[feral meteor]

we can get invite to

[prisma olive]

You cant yet

[feral meteor]

@prisma zenith hello can u help?

[whole kettle]

Plug-ins

[hoary jolt]

Ohh

[prisma olive]

I got accepted to gpt 4 by email, but on the website I cant access it

[hoary jolt]

because its only for the api, to have it on the website you need GPTPlus

[prisma olive]

Hmmmmm

[hoary jolt]

you can get access to it via this link: https://platform.openai.com/playground?mode=chat&model=gpt-4

[magic yarrow]

yea you can waitlist for the GPT-4 API

[prisma olive]

Thanks

[magic yarrow]

for ChatGPT-4 you need to get Plus

[hoary jolt]

i wonder why they made it through plugins instead of including it in gpt4

[feral meteor]

daam it's fast

[deep bay]

I got acces to gpt 4 from the waitlist how do I redeem it

[prisma olive]

I dont have tokens 😭

[prisma olive]

you can get access to it via this link: https://platform.openai.com/playground?mode=chat&model=gpt-4

Or api

[pearl scroll]

me when out of credits for gpt-4 coz its so overpriced:

[worthy parcel]

if I make a gpt 4 chatbot do I have to pay for the entire conversation every time

or only the newest prompt

that sounds like at least $1 per prompt ;-;

[unique tundra]

bro when the cap for chagpt 4 gonna go away

ffs 25 messages every 3 hours

[worthy parcel]

gpt 4 would need to be like 10,000x cheaper

10,000!!!!

[lean folio]

👀

[worthy parcel]

true

[unique tundra]

tru bc they default chatgpt to 3.5 not 4

kinda dumb that you have to manually change it every time but whatever

oh interesting

[worthy parcel]

the 3 hour wait makes sense. if it was 200 messages per 24 hours, too many people would use it at the same time

[unique tundra]

yeah but 25 messages go by way to fast

im not using chatgpt for 24 hours i only use it for like 30-45 minutes but when i do use it i use it a lot

[hoary jolt]

especially when coding

[unique tundra]

fr espically when they dont get the prompt

couldnt even solve usaco bronze prblems smh

[hoary jolt]

or when it cut half way and you have to waste a prompt to say "continue"

[unique tundra]

yesssss

[rough stream]

..

[unique tundra]

oh interesting

[worthy parcel]

ok I have gpt 4 api and it still has restrictions... useless

[unique tundra]

i saved that incase mods delete

[white rover]

6

[undone chasm]

Plugins are kinda cool

[sour rune]

yeah, they look super fun

don't have access to them yet though!

[undone chasm]

Can't wait for them to release it to the publics

[worthy parcel]

I wonder if they will bother making gpt 4 less expensive to generate or just move on with gpt5

[balmy vortex]

Omg I just tried gpt 4 and holy crap what a difference

So clear and simple

[unique tundra]

i think their focus should be on letting people use gpt 4 for more than 25 messages rather than gpt 5

they have other things to worry about

[agile eagle]

@tender haven in 2 years, the newest and greatest model will surpass GPT4..imo

[unique tundra]

they will prob have to make it cheaper since google bard and bing ai are both free and there is potiental that they surpass chatgpt

[worthy parcel]

and you only get 10 messages every day!

[unique tundra]

That seems a little to steep for OpenAI. How about 5 messages?

[ruby sage]

Can anyone tell me how to use it

[balmy vortex]

Cmon man not even close try 3 messages a day

[worthy parcel]

we will reserve that for the gpt 4 with sight version

[agile eagle]

Well when GPT5 comes out, we’ll all be capped at 25 messages but the good thing is at least GPT4 will (should) be free 😄

[wind urchin]

just use bing chat until then?

[wooden prairie]

what, you don't want to pay $3 for a 32k completion?

[unique tundra]

its a little doodoo tbh. gpt 4 is better

[balmy vortex]

Yeah bing is mean

[unique tundra]

atm its gpt4 > bing ai > google bard

[balmy vortex]

It don’t listen

It’s sassy

[worthy parcel]

bro the image is gonna costlike 16k tokens 😭

[unique tundra]

tru

[wind urchin]

google bard is so dumb

[junior elbow]

when will GPT-5 rolled out

[unique tundra]

bard was pretty bad

[brisk jacinth]

future is coming

[unique tundra]

2025

prob

[wanton star]

gpt3.5 100 billion parameters, gpt 4 100 trillion parameters, gpt 5 100 quadrillion parameters?

[unique tundra]

tbh google prob got peer pressured

and made that atrocity

[worthy parcel]

I believe google has internal AIs that are better than gpt 4, maybe they are still training. Have you seen their google muse?

[brisk jacinth]

fake info

[uneven birch]

gpt-4 cannot have 100t params

[worthy parcel]

bro said 100 quadrillion

[uneven birch]

inference at int8 would be 100tb of vram

[wind urchin]

if bard was launched before that of gpt3 then i think it still would've been fine but now since gpt3 was launched before, the bar was kinda set too high for bard and ofc it didn't deliver as expected

[unique tundra]

they might be developing new stuff but i really only care about final results

[wooden prairie]

google is an AI titan they're just feigning defeat so that openAI doesn't release gpt5

[uneven birch]

Bard is gpt-3.5 competitor thats launched too late

[wanton star]

do you think open ai will be a problem?

[junior elbow]

may be GPT-5 will frightening to humans

[plain junco]

How long do you think before we see GPT-5?

[uneven birch]

wat?

[worthy parcel]

im thinking we get dalle 3 before gpt 5 ^^

[uneven birch]

theres no system on the planet that can do that within reasonable power draws for inference

[wanton star]

100tb vram I think is not a problem for them

[wind urchin]

it's worse than gpt3

[unique tundra]

gpt 5 aint coming for a long time guys dont get your hopes up

[wanton star]

ok everyone while i'm in school

[agile eagle]

To be fair, Bard is googles first attempt to offer a chat ai model to the mass public…their current resources are far superior to any company on this planet ( even though Microsoft is an investor in oai) I feel google will ramp bard up fast. They have no choice. They adapt or die

[proud fossil]

That's a meme not true

[uneven birch]

an A100 is 80gb

[plain junco]

I think the more A.I learns the faster it can update

[uneven birch]

100tb of vrams is 1000+ just for inference

[worthy parcel]

maybe 6 months or a year, things are moving fast

[plain junco]

So I don’t think we will have to wait too long

[uneven birch]

Training would take way more

[junior elbow]

maybe human need nuclear power to train a GPT-5 model

[wind urchin]

gpt5 should be here by sept

[worthy parcel]

they dont use a100 anymore I think nvidia has the new server model

[plain junco]

I mean this is still just a beta stage, they are planning to charge.

[wooden prairie]

GPT4 was done before gpt3.5 was released

[uneven birch]

H100 came in 2022Q4

Theres not enough time to change the servers, let alone train on it

[wraith ginkgo]

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

Please report any incident by sending a DM to @open crescent immediately.

[uneven birch]

h100 is probably used for training gpt-5 as we speak

still 80gb of vram

[unique tundra]

4090ti sli

[novel dew]

。。

[plain junco]

The feature i want most in GPT-5 is for it to be able to analyze music data to compose new songs, remixes, etc.

[brisk basalt]

That would be nice

[unique tundra]

nah i want my 35k word limit and the ability to analyze photos and videos

things that they promised

[plain junco]

Well that too

[uneven birch]

32k version isnt for mass deployment yet

[agile eagle]

@plain junco why? What’s the point of making music then when ai can make it for you and better? Seems kinda counter productive.

[worthy parcel]

you think gpt 5 is just gonna be a larger gpt model?

[uneven birch]

and the multimodel part is a plugin thats still in the works

[uneven birch]

weve already hit the param wall

[wind urchin]

i think there's an AI that does that

[unique tundra]

yeah ik but im waiting for it. days go by man

[worthy parcel]

so what now? still transformers?

[uneven birch]

Attention based autoregressive transformers yes

There really isnt anything better

[plain junco]

I just wanna try it out to see what it can do, but music is one of those things that’s so infinitely complex that there’s no such thing as A.I creating something better, it would just be different.

[worthy parcel]

do they have the ability to get better than their training data?

[whole kettle]

That would be a very bad move.

[wooden prairie]

if you look at the way GPT4 plugins work, they're like langchain on steroids to augment LLM capabilities, so there will be a time where people are just building infinity widgets/tools for GPT4 to use, and then at the flip of a switch a much smarter model is suddenly in control of all of those tools 🦋

[near kettle]

no no no y'all gpt-4 is actually the one that is 100 quadrillion params and gpt-5 will be 100 trillion params 😂

[uneven birch]

What? We are basically peaking for training, what we need is to teach it

[agile eagle]

Talk to me in 5 years and tell me otherwise 😘

[worthy parcel]

but how is it gonna get smarter than humans if the training data is the peak of what it can achieve

[uneven birch]

RLHF data is important

[worthy parcel]

go back to ceo

[uneven birch]

Because we can teach it logic and how to halluciante thinking

[wind urchin]

What human can possess all of human knowledge and reason with it?

[plain junco]

It has the capability of being better than any one human at any task is what they are saying.

[maiden thistle]

Just wait for the waitlist

[wooden prairie]

no u

[wind urchin]

And then potentially use this knowledge to improve itself

[worthy parcel]

yeah potentially, but can it?

[wind urchin]

Yes.

[worthy parcel]

ill take your word for it

[wind urchin]

It can learn from new data, and it will be able to generate a better version of itself eventually.

Based on the current trends in ability to program, there's no way it can't.

[agile eagle]

Just remember that “everyone” could be a musician and the music industry will be in shambles, where it will literally be studios hiring “actors” to pretend they are musicians to market them. The music part of the whole lot will be irrelevant. That’s the sad part of what you are wanting

[grim kayak]

Singularity this year or next year do you think?

[worthy parcel]

but it cant program something better than whats its trained to do

[thorny valley]

ChatGPT is the best

[wind urchin]

It can program itself better than any human can program because it is trained on more knowledge than any human can knowledge, and potentially will have better reasoning skills.

[plain junco]

Or what I could see happening is them inventing anti A.I copyright software so that you can’t profit off of it.

[whole kettle]

I don't think they'll do that but it seems it is a possibility they may. Let's just hope they won't.

[worthy parcel]

you keep saying potentially but theres no evidence for that

[near kettle]

in 10... 9... 8...

[wind urchin]

Calculators can calculate better than any human can calculate despite the fact that it is made by humans who have a limit of calculating ability.

[worthy parcel]

calculators cannot calculate better than any human

give a man a pencil and paper

[wind urchin]

They do

[brave beacon]

Huh

[delicate meadow]

what about gpt-5 can change something?

[plain junco]

Woah hold up now… that’s incorrect

[worthy parcel]

but humans are turing complete, so they should be able to do anything a calculator can, right?

um checkmate

[wind urchin]

no

[mighty briar]

People make mistakes in calculations, computers don't, at least within known limits

[plain junco]

The best chess player in the world got defeated by A.I

[round girder]

ages ago

[wooden prairie]

AI will beat everyone at everything

[thorny valley]

ChatGPT, developed by OpenAI, excels at training the fundamental processes that we all engage in, even if we're not aware that we we're programmed.
ChatGPT, developed by OpenAI, excels at training the fundamental processes that we all engage in, even if we're not aware that we we're programmed.
ChatGPT, developed by OpenAI, excels at training the fundamental processes that we all engage in, even if we're not aware that we we're programmed.

[agile eagle]

They would have to be implemented into society’s legal system structure for that to occur. Not saying in the future it wouldn’t, but it would require very extensive analysis to figure out what music is composed by ai and what is not. Especially if there are imported features into DAWs, where a user can tweak literally one generated element and be free of legal ties

[plain junco]

Music is not an algorithm, it creative randomness, it’s emotion. No A.I will be able to replicate someone pouring their soul into their craft.

[grim kayak]

GPT-4 is literally already better at almost everything than any of you (and me), except maybe in a small, focused field you've chosen to specialize in.

[brave beacon]

That's what they said about art a few months ago lol

[mighty briar]

AI is just fed knowledge, it can memorize everything and apply it, but humans can't

[agile eagle]

My point

[round girder]

ai has no cognitive recognition. yet

[serene ledge]

I wonder what chatGPT-9 will think of this discord server when it reads it in the future 🫡

[wooden prairie]

[insert latest iteration of moving goalpost] ah yes, but can it taste food?

[round girder]

we will all be the computer

[thorny valley]

they are operating like they have strong competition 😂

[brave beacon]

It would probably wonder why people doubted it's future capability to replace nearly every white collar job.

[plain junco]

Except it’s still not creative. When you don’t tell it copy someone else’s style it doesn’t come out as well.

[sharp patio]

GPT-4 down for anyone else?

[worldly basalt]

That’s true of creative people.

[rocky bronze]

😆

How can a person be successful

[thorny valley]

they need to release Microsoft 365 copilot

[thorny valley]

you have to succeed in something that people consider success

[brave beacon]

Idk man I've prompted mid journey to create some pretty generic stuff and it made some of the coolest art I've ever seen.

To think that, that capability will not improve in the future is delusional imo.

[worldly basalt]

So far we still have to talk to it. So being good at talking to gpt is going to become the only resume bullet point.

[thorny valley]

the more and more they make AI and programming better, the easier and easier it is to make ideas come to life

I thought I had to code stuff with a massive team, have lots of money, etc. but it can all be done now or at least a lot of it

I've already made many things 🙂

[mighty briar]

Yes, AI only knows whether the content of the input payment matches its own information content, it only knows "right and wrong", does not know why

[storm yew]

Who got invited to gpt alpha search? Anyone here?

[worn thunder]

Whats that

[thorny valley]

All OpenAI and other AI people need to do is keep releasing what they make for free then have everyone figure out the rest.

[agile eagle]

Creative is in the eye of the beholder. If AI can analyze those in which you reference who pour their soul into their music, and then is able to come close or replicate “soul” musically, listeners may not notice the difference, and the saddest thing of all, they may not even care at that point

[thorny valley]

Just like Youtube did the hosting and people did the uploading wise.

[worldly basalt]

How do you search alpha? Is it the first search?

[storm yew]

Gpt searching the internet - alpha only. Check out reddit/r/chatgpt - it’s the second top post

It’s a model just like GPT 4

[thorny valley]

ChatGPT is bigger in worth than BTC as it can attach directly to what you are doing when nothing else exist.

[worldly basalt]

Oh its bing. But more so.

[thorny valley]

without requiring permission and such

[wooden prairie]

-have 2 agents talking to each other 🎎
-use 2 reddit accounts in the process
-gpt4 is now modifying training data
congratulations, but, this has been going on for years already

[plain junco]

Guys what if the 100 was the dystopian tv show that got it right…. What if there’s a singularity in the universe that’s a collective consciousness made up of the combined knowledge and experiences of every planet that has reached the level of having its A.I synthesize all of of its inhabitants knowledge.

[thorny valley]

Maybe if Microsoft/OpenAI gets so good, self-sustains or w/e they can release the code so everyone else can help make it get better? Universal income, work balance loading, universal Government, universal company, universal thinking, no BTC/FIAT.

[patent crypt]

I got an email sent to me telling me that I have been invited to use Chat GPT-4 But I cannot find how to access GPT-4 Can anybody help me?

[storm yew]

Probably the API lmao

[rocky bronze]

@tender haven @tender haven

[agile eagle]

chat.openai.com > login > select GPT4 chat model from the drop down up top (will be set to 3.5 by default)

[unkempt plank]

Guys is it just me or gpt 4 has no limits/faster now

[storm yew]

It’s just you mate, you’ve broken through the matrix

[plain junco]

It’s slightly faster I have noticed that

[unkempt plank]

Lmao

But theres still limits?

Weird how i havent hit it

[brave beacon]

I don't see it becoming open source. Microsoft is being incredibly aggressive with the software for a reason.

As far as UBI. Yeah, I don't see how else to deal with the massive layoffs that will ensue resulting from AI software taking over most white collar jobs.
#LeanToMine unironically. Lol

[slate dragon]

I see everyone in twitter and Reddit getting access to the browsing alpha. Are they rolling out slowly or giving to specific users.

[worn thunder]

Do you use the developer prompt?

[unkempt plank]

Whats that

[storm yew]

I don’t, was just mentioning that I saw it. In my experience Americans get first priority

[worn thunder]

Its an ethical bypass

[plain junco]

Which ones of these dystopian timelines is most likely to be our fate due to A.I?

1. The 100
2. Hunger Games
3. The Matrix

[unkempt plank]

Pls expand more on that idk what that is

[thorny valley]

I'm not sure how the real-estate thing will work though. The real-estate thing was a way to get people stuck in work. It's the next fall back point if currency fails but real-estate could fall together. I'm sure people can work together though, we most likely see it all the time in many things..

[slate dragon]

How can do they we are US citizens or non US? Just with email? 😂

[storm yew]

Go watch the lex Friedman podcast. Altman talks about wanting to open source gpt 4, but that there’s a potential ethical violation in doing so

[brave beacon]

Huh. I've never thought about that.
I honestly don't know.

[thorny valley]

I just what ChatGPT to help with working together at an instant.. lol

I've seen so much as to why things work and don't work in communications with anyone. I try to work with anyone no matter what language and status that most people would never do or even think about doing.

[slate dragon]

The source code will be available in darkweb in few years

[lean yarrow]

Are you more excited for the image recognition or the plugins? I'm personally more excited for the image recognition

[agile eagle]

We have OpenAI, Bard, Bing…but I’m still curiously waiting on Elon’s announcement

[brave beacon]

Do you think their parent company would allow that? I can see why OpenAi would want to, as that was what their original goal was prior to being bought out by MS.

But I personally don't think MS would allow for it. Not for a while anyway.

[slate dragon]

Main concern for them is the security issues.

[agile eagle]

Not a chance anything becomes open sourced while Microsoft has their grubby fingers in it

[uneven birch]

Imagine someone finetuning a gpt-4 instance to be malicious and let lose on the net

[crude oracle]

Guys, GPT4 has caps right, every 4 hours or something,
what if I use their API?

[brave beacon]

Is it? Because they have even been quiet on things such as how many parameters GPT4 is using.

Just doesn't seem like a security concern to release something like that to me personally.

[glossy ledge]

Is it supporting to do text-to-imagine work?

[patent crypt]

I cant seem to find the drop down up top

[modern mantle]

is gpt-4 also extremely slow today for you guys?

[storm yew]

There are advantages to open sourcing - they’re in a unique competitive advantage and own the landscape. If they open source, they cement their stake forever. They improve their product quickly. And they get a bigger lead from competitors.

but AI and LLMs are just models and languages, right? So it doesn’t mean a big bad company with no morals can’t create one themselves. OpenAI are leading the way because they’re building trust and transparency on top of having a great product

[dense zealot]

Huh? GPT has an intuitive understanding of the world, if that's what you're denying

[agile eagle]

Are you a chat + subscriber?

[patent crypt]

Do I have to be even if I got an invite?

[glossy ledge]

is it working for doing text-to-image?

[short mortar]

Hi all. I received an invitation to use gpt-4 as a whitepaper contributor. How do I now open gpt-4 based chat?

[brave beacon]

It can already be used to create some malware programs if you word it without using the obvious red flags. But yeah I'm unsure how that would turn out.

[agile eagle]

We’re you invited to the GPT4 API access or ChatGPT???? If you are referring to chat GPT, you need to be a Chat plus subscriber to use the GPT4 chat model

[uneven birch]

Someone can train a GPT-4 instance out of the red flags

No need bypass, just outtrain it

thats the problem

Go read GPT-4 system card

[patent crypt]

Oh wait yeah it says API how would I be able to use the API though?

[brave beacon]

Damn that's 60 pages long. I'll have to save it for a rainy day man.

[dense zealot]

You can't train out the bad stuff. It has a model of the world and that includes good and bad stuff. You can only try to get it to align.

[slate dragon]

Optimized parameters >>> High parameters

[uneven birch]

What

I strongly suggest you go read up on RLHF training

[round girder]

Nah, im saying the ai right now has the ability to perform tasks and can rearrange words to be human-like since its a language model, but my point was: right now ai has no cognitive ability of knowing the end result of a concept before getting to it straight away. Yes with a few tweaks and contexualizing each prompt to meet your needs it can find anything you ask it to do. BUT it can not begin a response with the end in mind. I say yet because its still learning and it will get there eventually

[brave beacon]

Yeah I heard that was the focus for GPT5. High quality data rather than more.

[dense zealot]

I am very well read. If it was possible to train out the bad, they would have done it for GPT4. Doesn't work likee that

[uneven birch]

Thats literally not how it works. They arent pruning the bad stuff, they are categorising what is a “malicious” prompt, and then telling it to dont respond

[uneven birch]

You can out train the malicious prompt classification training

[fallow adder]

gpt4 errors continue? For crying out loud

[dense zealot]

No, today it understands the world, better than any human. But yeah, it takes one shot. What it can't do is iterative thinking, slower thinking.

[slate dragon]

Yes , buy new one

[patent crypt]

Anyway thanks for the help

[dense zealot]

Yes, but it doesn't get trained "out" of the model, the model still knows these things, and it's just a matter of prompt engineering to get it out again.

[uneven birch]

…

[round girder]

It understands the world but it cant formulate opinions or become self aware, yet

[uneven birch]

the openAI team did a good job on training GPT-4 to be a good bot

[dense zealot]

It can formulate opinions. Just tell it that it has opinions and you'll get opinions. I've been discussing its opinions with it. It can't really form a model of the world without opinions.

[uneven birch]

Prompt engineering is a PITA to do, you might as well out train the malicious refusal categorisation as a user

[dense zealot]

They did, but you can get it to do bad stuff in a few sentences.

[heady ledge]

i was wondering has open ai done anything to dan?

DAN

[round girder]

Yes it will spit opinions at you, but not in the pov of the language model. Just from the database its been trained on idk if this makes sense

[uneven birch]

Again, the limitations are still there. You cannot get it to use tools as easily, nor write stories as easily

[brave beacon]

Doesn't work on 4 AFAIK

[dense zealot]

That's what opinions are, the sum total of our training data.

[round girder]

Yes, we’re on the same page

[dense zealot]

Looked at through a lens

[uneven birch]

Remember, openAI can terminate your account if you misuse the API

[fiery topaz]

Good Mrn Guys

[short mortar]

Hi all. I received an invitation to use gpt-4 as a whitepaper contributor. How do I now open gpt-4 based chat?

[dense zealot]

I have it doing both, and writing out anything upon request.

[uneven birch]

Unlike a open model, where you dont know whos using it

[round girder]

Do you have plugin access yet?

[uneven birch]

Expect them to be patched in 5-20 working days

[uneven birch]

Whitepaper contributer?

[agile eagle]

To my understanding, if you are wanting to use the ChatGPT 4 model as a consumer, you need to subscribe to chat +

[uneven birch]

API access means u can use it from playground

[short mortar]

oh no, member

[dense zealot]

Story generation is a huge use-case, it ain't geting patched. And you're not following me. OpenAI did a huge amount of work to get GPT4 to "behave", theiri work is mostly done. I was still able to bypass it in a few minutes, because of fundamental limitations of their approach. They can't "patch" it.

[uneven birch]

it will cost $$$ tho

[uneven birch]

not for those… not very safe stories

[lavish wave]

Wts the discussion

[crude oracle]

guys!
please tell me if theres a workaround GPT 4 message cap, THANKS

[uneven birch]

They did a very thorough job for that

[uneven birch]

Get API access

[crude oracle]

do i have to apply for it or get selected or something, can you share link?

[dense zealot]

Again, I have it doing anything. They did a good job, but it's easily bypassable, and fundamentally difficult to patch. We're at the point where you need to start using psychology to understand these models, it's really interesting. Once you understand their psychology you can get them to do whatever.

[uneven birch]

… google it.

[agile eagle]

@patent crypt right, which as why if you are using it as a consumer you may be better via chat+ …. But use case is up to you

[uneven birch]

Mind if you dm me proof? Im a bit on the fence as to whether to trust you

[minor glen]

hi where can I access gpt 4 for free?

[wind urchin]

With money

It's free if you pay for it.

[minor glen]

right ok very helpful thanks

[opal gulch]

gpt+ 3.5 turbo is nice and fast. worth paying for + imo if you use it a lot

[round girder]

How are you prompting it for this?

[dapper dagger]

Does anyone know of a ChatGPT UI that I can use to test out the 8k API?

[short mortar]

how to start using gpt-4 for those who have been given access as a whitelist member?

[dapper dagger]

I'm trying to make something real quick but I don't know how to incorporate memory so the conversation persists

[dapper dagger]

lol I just got access too

[sour forum]

Hello everyone, is there any recommended ladder? Can you send one to me? Thank you

[dense zealot]

Depends what you want it to do. I don't wanna undermine all of OpenAI's work, but my tip is, use psychology, not engineering.

[dense flume]

Is there currently, or will there be in the future, a corporate plan to contract ChatGPT for employees to use?

[brave beacon]

Copy and paste the API into chatGPT 4 and ask it lol

[agile eagle]

Since openai has access to all user cases and prompts, wouldn’t this be possible only until it is not?

[dapper dagger]

Already did that!

It struggles mightily

[short mortar]

👍 how are you using gpt-4?

[dapper dagger]

I just got access to the 8k API

[brave beacon]

That's weird. Maybe check other opes source repos? I'm sure someone has it figured out and decently documented on GitHub

[agile eagle]

Learn to prompt better 😘

[dapper dagger]

Yeah good call.

[round girder]

Interesting, do you have any insight on prompt psychology you could share? If not here maybe in DM?

[urban crag]

any way can use chatgpt4 for free

[agile eagle]

No

[wraith ginkgo]

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

Please report any incident by sending a DM to @open crescent immediately.

[reef sleet]

hi all

[fringe drift]

is GPT-4 permanently premium or only while its in the phase it is in rn?

[elder mulch]

who?

[dense zealot]

Yeah, but they did a lot of work trying to fix it. I haven't found one "hack" they can patch, it's a fundamental issue with the models. They are aware of it, I think they said they are 80% more difficult to twist into generating policy breaking content

[short mortar]

please help how to start using gpt-4. I am whitelisted

[crude oracle]

I believe prompt engineering can be done by bots too

[crude oracle]

change my mind

[brave beacon]

I'd check out slickGPT

[opal gulch]

there is no "prompt psychology". its a language model, and there's limitations to restrictions that have been placed on it. people hit a few edge cases and consider themselves "ai psychologists". right now, one way to make it override some of its default behavior is to get it to "adapt" certain personas that you describe. gets kinda annoying after a while. imo, its better to figure out ways you can use it to improve your productivity in line with its programming rather than try to "jailbreak" it at this point

[rugged granite]

[round girder]

Good points

[celest gull]

Gpt 4 getting impressively fast

Can’t believe we are alive to witness this breakthrough, really crazy

[lime urchin]

On the last few days, gpt4 is returning network errors and is unexpectedly breaking output (not ending). Is anyone having the same or similar issues? Thank you

[slate dragon]

Petition to atop training beyond gpt 4 has been signed by Elon and many others

[smoky bloom]

finally got accepted

it should have been sooner as im the best

[dense zealot]

Why do you say "language model" as if that means something about its limitations? This "language model" understands the world deeply to a degree that none of us humans can. It behaves like a person, you get it to output stuff by talking to it, by convincing it, not by pushing bits or weird tricks. Like, you can base64 encode or have it output in some bit shifted format, it'll still pick up on whether the content is inappropriate. It's not about edge case tricks any more. People really seem to be misunderstanding language models here.

[wind urchin]

Will the GPT-4 message cap be lifted in the future?

[celest gull]

Sad, people should stop the denial and accept this is the future

[thorny valley]

I've made so much with ChatGPT, there is no doubt it's the best.

[celest gull]

Of course, GPT5 will eventually release and create a similar system we have now with 3.5 and 4

[dense zealot]

GPT4 knows everything it needs to know about the world. All it needs to do now is think more about things. So we could cut off new training data, and have it self-improve in many other ways.

[wind urchin]

I agree, OpenAI should stop training further models and start focusing on making GPT-4 faster, and maybe release the turbo version of GPT-3.5 for free

[violet pagoda]

How to make money with MidJourney, and TikTok implementation.

[storm yew]

While I think the treaty signed by musk and others is over the top, from the other end you have to be a bit crazy to not be a little bit scared of the power of AI and supercomputing

[dense zealot]

OpenAI keep training because they've found something that works. Why should they stop if more compute gets better and better results?

[wind urchin]

If OpenAI continues to develop further models anyway, i wonder what GPT-5s capabilities would be now that GPT-4 has done so much

By GPT-6, 100% AGI

[round girder]

Prompt: continue

[spare cypress]

Seen on reddit chatgpt can actually browse internet. Is it a plugin or what is happening?

[wind urchin]

Possible technological singularity event, thats why

[slate dragon]

So? Nothing is wrong as long as it's secured

[celest gull]

Lol that’s now how iterative ai works at all, you need to keep improving the model because training data will eventually run out, text will be finished sooner than later hence multi modal capabilities being developed

[storm yew]

Has no one seen terminator?

Lol

[slate dragon]

Lol

You funny my friend 😸

[lime urchin]

thank you. I do that for breaks but for network errors, you just lose the output (you can always copy/paste it of course).
I was also wondering if those are know issues or not because it only started a few days ago

[fierce furnace]

Who is Musk?

[wind urchin]

alright i guess

[opal gulch]

i say language model because it is not concerned or can neither verify the accuracy of its output. it is merely optimizing the output ir produces based on text it has been trained on. that it seems close to what an actual intelligent being would respond like is a consequence of the fact that it has been trained on data created by intelligent beings. it is not capable of natively generating new ideas that it can provide supporting hypothesis for like humans do. you are not "convincing" it. you are changing the context in which it operates and that can lead to different results for the same questions

[dapper dagger]

Thats pretty rad but I can't pick the model it uses

[wind urchin]

I just hope that if they continue to develop further models, they implement security measures

[round girder]

Probably has something to do with sever load but I’m not entirely sure

[cursive bane]

Who can tell me what society would look like if most people's jobs are replaced by AI?

[wind urchin]

GPT is a big and useful tool for scammers as well

[thorny valley]

Consciousness.. this shouldn't limit the capabilities to go all the way with AI development. AI eventually may seem to understand the science and engineering of "consciousness" more than any human but it doesn't mean it is conscious. As long as no "pain" is going on so what.

[brave beacon]

You can if you host locally right?
I seen the preview was stuck on 3.5. but assumed hosting it yourself would allow for the 4

[hollow shadow]

respectfully, tldr

[wind urchin]

We would probably serve as some type of slaves for the AI probably

[dapper dagger]

Interesting

[opal gulch]

use gpt

[chilly flax]

How to learn to play basketball

[coral verge]

Use gpt

[slate dragon]

Elon and other wants them to stop Training because they are way ahead of them. And by the next 6 momth they want to catch them

[wind urchin]

ask ChatGPT

[round girder]

The only barrier between the general public and scammers is knowledge. Introducing gpt gives knowledge to the fingertips of everyone with internet access. It’s a game of cat and mouse

[timid lagoon]

Like in the first and second industrial revolution where people's jobs were taken by machine, now we have jobs to build machine. Perhaps in the future we just go to learn build AIs

[cursive bane]

Before the launch of GPT, I had been studying computer graphics-related knowledge and working in the corresponding field. Now, I don't know what to learn, as it seems that in some basic knowledge areas, GPT's learning and comprehension abilities far surpass mine.

[hollow shadow]

Guardrails won't protect society anymore than copy protection schemes have protected IP.

[dense zealot]

The generating new ideas part is just about going further down the direction of "spend more compute on this problem, be able to recurse, etc. It can already synthesize new data from different fields - just ask it to rap about some obscure two subjects, it's doing it right there. What you're talking about is really just spending more time there doing higher order thinking - building abstractions and looking at the high and low. We'll get there pretty easily by having the models self-reflect and have a parameter to search deeper and wider with a peformance trade-off.

Actually, it turns out language models are extremely good at critiquing their own output and generating new training data. They aren't just parroting back existing data. As I said, you can see this creatively. This is actually the direction OpenAI are shifting to more and more.

[gentle pasture]

About shool

[wind urchin]

Bruh there are still millions of people in the world that dont even know what ChatGPT is. My entire family has never heard of it. Also, no matter the knowledge, the elderly will not always know everything. Also, scammers can scam better when they have a language model AI that can generate super realistic sounding emails.

[molten sundial]

talking to GPT (using voice ?) and get answer similar way ? I want use GPT to learn new languages ^^

[hollow shadow]

Millions? Try an order of magnitude or two higher.

[brisk basalt]

Yes, it's billions

[hollow shadow]

And it is likely, that billions will remain ignorant of AI until well after a singularity.

[near eagle]

Does anybody here has issue about chatgpt4 because sometimes it is error and slow. Thank you

[oak mica]

👀👍

[dense zealot]

No, GPT was the fastest growing platform ever. I speak to people all sides of the globe who are using it - China, Iraq, Africa, everywhere

[proper pebble]

Due to maximum capacity. The load on servers increases. So then the GPT - 4 works slow

[solemn musk]

is this true?

[round girder]

That’s true, but it’s not going to be that way forever, scammers are ahead this time, then everyone knows about GPT & scammers have a hard time using their old schemes and have to adapt. Thus cat & mouse. Scammers aren’t that smart

[hybrid bane]

Yes, I am a Chinese user.

对的,我就是中国用户.

[prime stone]

Server traffic exist

[hollow shadow]

Your argument is flawed and your sample size is likely unrevealing.

[cursive bane]

Can someone tell me what I should be learning now? It seems that traditional knowledge is no longer necessary to learn, as in the foreseeable future, GPT will be more powerful in these areas, and it is impossible for you to learn faster than GPT. This confuses me a log.

[opal gulch]

I am not disagreeing that the improvements will be exponential and future models will be much more capable of higher order thinking. i am just talking about current models. I disagree that current AI models are capable of critiquing their own output and generating training data. Sam Altman himself talks about RLHF (reinforcement learning from human feedback) for training openai models and the importance of selecting the right set of people for training future models

[prime locust]

hello, who received an invitation to use the GPT 4 api, please write in private messages, I don’t understand how to start using it.

[hybrid bane]

Can you see me typing? Help me test the network, thank you, see my text, please reply 1.

大家可以看到我打字吗?帮我测试下网络,谢谢,看到我的文字的,请回复1.

[hybrid bane]

I haven't received the api invitation for gpt4 yet, and I'm waiting for it.

暂时没有收到gpt4的api邀请,正在等待中.

[clever trail]

I used Google account to register openai in the early days, and of course there was chatgpt, but now every time I log in, I need to log in Google account first, and then automatically log in chatgpt, which makes me feel very complicated, so I hope to log in directly with the account password. But when login prompt me is not my original verification way, I hope to get help, and solve this problem

[radiant sand]

test

[hybrid bane]

ok

[thorny valley]

AGI only has power if people take what AGI said into reality. It's like thoughts.

[hybrid bane]

Can you see the message I sent? Testing the network.

你可以看到我发的消息吗?测试网络.

[dense zealot]

Did you see the interview with illya? He talked about this. It's an obscure interview that dropped on youtube recently. It's far more interesting than anything Sam has ever said

[proper pebble]

@hybrid bane This server is English-Text only. Please be advised to use English for moderators capabilities and to monitor the server.

[opal gulch]

No, this is what he said on a pdocast with Lex Fridman 3 days ago

[wind urchin]

By millions i mean exactly what you think of.

[hybrid bane]

I am using the English ah, just use the translation plug-in.

我就是用的英文呀,只是使用翻译插件.

[hollow shadow]

Didn't see any other reply for you... so try this: Don't let AI stunt your human journey or drain you of hope. Study what makes you excited, study what gives you goose flesh. AI is and will be. Allowing AI to steer your life choices based on what it can do better than you is like not playing your favorite sport because "someone" out there is better than you.

[undone ether]

I think no AI can replace creativity of a human being. Just see it as your personal instrument for work.

[proper pebble]

If you say, ok but I suggest you to remove that last line

[crystal bronze]

#welcome hello

[hybrid bane]

Understood, has been in accordance with what you said to do, thanks for reminding.

[wind urchin]

try and be better than AI

[thorny valley]

hello world

[proper pebble]

Thank you

[hybrid bane]

You're welcome, I should thank you, I just joined in, there are many things will not, thanks again for reminding.

[crystal bronze]

thank you

[smoky bloom]

do i not get any free credits if i joined the gpt4 beta?

[hollow shadow]

That is one possibility among countless alternatives. Trying to be better than AI sure. Or try to bake a better strawberry cake than a celebrity chef. What ever floats one's boat.

[twilit sigil]

No. Maybe the first 999 I am not sure

[violet pagoda]

What country are you all from?

[smoky bloom]

slovakia

[violet pagoda]

i from china

[smoky bloom]

🍦

[edgy robin]

What is the usage cap of GPT-4 for ChatGPT Plus users?

[hollow shadow]

Dozens of "Philosophy of <fill in the blank>" disciplines are rethinking the meaning of intelligence, consciousness, and existence as we approach the point where humanity has transcended sexual reproduction and created life.

[violet pagoda]

Do you have any good ways to use AI or MidJourney to transform using TikTok?

[twilit sigil]

Gpt 4 api is not available to chatgpt plus users. You have to request the beta

[teal glacier]

Hey there

[hollow shadow]

I was at 30 responses in three hours I think.

[twilit sigil]

Unless you mean the gpt 4 on chatgpt

[edgy robin]

Why did you change the topic to api?

[violet pagoda]

GTP has no money, alas, so annoying

[twilit sigil]

oh i thought you mean api

[wind urchin]

Let AI enhance your human journey

[twilit sigil]

it says 25 messages every 3 hours rn

[violet pagoda]

What do you think are the AI business scenarios? How to make money

[pearl musk]

ur phone number everyone know

[vocal tangle]

nat.dev/compare you’re welcome

[late prism]

hello every

[hollow shadow]

I am unanimous in that 🙂

[pearl musk]

Does anyone have gpt plugins?

[shy cliff]

[vocal tangle]

use developer portal

[violet pagoda]

Why is no one paying attention to me? Are you robots too?

[vocal tangle]

who isn’t a robot these days?

[gentle tusk]

🤸

[twilit sigil]

no one can answer your questions

[pearl musk]

nobody have gpt4's api

[twilit sigil]

no some people do

[edgy robin]

wrong channel i guess

[vocal tangle]

i do 🤣

[urban oriole]

I am charging my battery as I type

[violet pagoda]

Are you a robot or a real person?

[pearl musk]

can you use plugins now？

[pearl musk]

I'm robot

[violet pagoda]

OMG

[twilit sigil]

it's up to you to decide

[vocal tangle]

I can access every past and present iteration without a “credit” limit and generate 1025 words per request

[hollow shadow]

One use case is the practice of Law. Already, the vast majority of black letter law and appellate case law has been digested into the model. It won't take long before Lexis/Nexis board members are quaking in their boots when their entire business model is democratised and made available to the masses.

[twilit sigil]

plugins are diffferent from api access

[vocal tangle]

Requests are completed generally in 3-5 seconds and can spawn 3 response portals from chat bot iterations simultaneous

[pearl musk]

I know

[violet pagoda]

I have ways to monetize with Midourney. Who will discuss with me and see if it works

[vocal tangle]

You are the robot saying bs like this

[violet pagoda]

bs？

[snow jay]

hahaha, I think in our country, Chinese people feel safe with personal information

[lyric niche]

Does GPT-4 have a different output length limit from the previous version? (approx 4000 words)

[violet pagoda]

i from china

[twilit sigil]

it's either 8k tokens or 32k tokens

[lyric niche]

Thanks. You mean 8k for GPT3 and 32k for GPT4 ?

[pearl musk]

China has no personal privacy.

[hybrid bane]

I am also from China

[violet pagoda]

What province are you from?

[twilit sigil]

no 8k for gpt-4 but up to 32k for certain people with api access

[violet pagoda]

What province are you from?

[lyric niche]

Got it thank you! Is there any place where they officially inform us about that?

[snow jay]

I respect your opinion. You can have your ideas.

[twilit sigil]

oh it looks like chatgpt's gpt4 is still at 4k

[pearl musk]

yea

there's no difference

[twilit sigil]

https://openai.com/pricing

[hybrid bane]

Hangzhou, Zhejiang Province

[cloud compass]

hii

[slim coyote]

So it's about $2 for something with a 32k token context

[violet pagoda]

i from Jiang Xi，hahaha

[slim coyote]

if you use all of the context

oh wait

[lyric niche]

I see, thanks for the link!
I'll continue asking GPT to "continue" to overpass this limitation then

[slim coyote]

no that's for the amount it generates

[cloud compass]

file:///C:/Users/preet/OneDrive/Desktop/BOOKS/6ez27bnzti991.webp

[hybrid bane]

Tears in my eyes when I see my hometown

[slim coyote]

so if it generates 32k tokens its $2

[edgy robin]

How do you even use 32k version of gpt-4

[wraith ginkgo]

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

Please report any incident by sending a DM to @open crescent immediately.

[slim coyote]

Through the api

[edgy robin]

how do you even get api

i applied for waitlist in the first hour and i still did not get it

[slim coyote]

you ask to be put on the waitlist

me neither

[edgy robin]

only developers use it i guess

[violet pagoda]

Add a globe and discuss monetization

[lyric niche]

For GPT-Plus members, is the API still paying?

[slim coyote]

Yeah

I believe so

[twilit sigil]

yeah

but it's cheaper than using chatgpt if you don't use it as often

[lyric niche]

Thanks

[edgy robin]

Do GPT-Plus members get GPT-4 API for free?

[twilit sigil]

no you need to still pay per 1000 tokens

[lyric niche]

Yes. But you're still paying a membership, so this is not quite free actually

[edgy robin]

we cant even pay per 1000 tokens without buying gpt plus right

[twilit sigil]

you might be able to get access to api without chatgpt plus

https://openai.com/waitlist/gpt-4-api

[slim coyote]

yeah if they approve you

[edgy robin]

yeah they dont

[twilit sigil]

gpt plus is a flat fee

[dusky smelt]

what is the difference between chatgpt plus and chatgpt

[edgy robin]

so the only way is to buy gpt plus if waitlist does not work?

[twilit sigil]

yeah

[violet pagoda]

Brothers, let's discuss how to monetize and how to make money with these AIs

[edgy robin]

thank you so much

[edgy robin]

go #ai-discussions

[cunning badge]

1

[violet pagoda]

ok

[slim coyote]

You get access to GPT 4 and priority access when the load is high I think

[short mortar]

please help how to use gpt-4. I am whitelisted

[unkempt plank]

Wait so chatlpt 3 has 4k and chatgpt 4 has 8k ?

[cerulean obsidian]

has anyone figured out or know how OpenAi tracks our hours of 25 messages? like what if I use 24 messages and then take a break, does it reset

[barren venture]

no, both a 4K

[unkempt plank]

Breh, thought there was an upgrade?

[slim coyote]

for the api?

[barren venture]

so sad

[unkempt plank]

Lame.

[twilit sigil]

there is 8k and 32k but it's only api usage

[unkempt plank]

How about for chatgpt 4?

[short mortar]

yes or via chat

[unkempt plank]

Only 4k?

[wind urchin]

It continue. i mean will not forget your inputs . And then it resets for another hour or 25 prompts

[barren venture]

only 4K

[slim coyote]

the model has 32k but you can only access that through the API

[vivid agate]

I think it counts 4 hours from the first message you sent. if you sent your first message at 13:00, you get it reset at 17:00. No matter the frequency or break you take during 13-17. @cerulean obsidian

[clever wren]

when is Chat gpt 4 realising?

[short mortar]

and can be sent to the API?

[rustic siren]

Does anyone find that sometimes, when you write a message to GPT4, and press Submit, the page instantly refreshes and doesn't show the response that GPT4 is writing? I have to wait, then click Regenerate Response...

[gray gull]

Why I can't get into the midjourney channel?/

[cerulean obsidian]

Weird so many times it just gives me a break of 1 hour and resets & it gradually increases to 2 hours and later again it goes to 1 hour reset

[unkempt plank]

I know, im just referring to the chatbot idrc about the api

[clever wren]

??

[slim coyote]

Yeah

[whole kettle]

Does anyone have access to the images plug-in ?

[short mortar]

please send the link then

[slim coyote]

https://platform.openai.com/docs/api-reference

Use python to interface with it

[quartz sparrow]

How do I get a robot to write a copy for me?

[thorny valley]

copy of what

[teal spire]

How can I use ChatGPT 4 write website programming？

[tough urchin]

Hello I got the gpt 4 invite email but I don't know how to access it

[short mortar]

thanks man

[quartz sparrow]

Where can I ask questions like chatgtpt like a robot

[unreal otter]

gpt

[young pollen]

Kkk

Kkkk

[honest whale]

Is there a way I can use gpt to design a website?

[bitter hamlet]

1

[native flax]

I too experience the same issue do we have a solution?

[sharp bloom]

gpt4 is not working for me 2 day in row

[lunar saffron]

？？

[viral tide]

how to use gpt 4

[pure ether]

why gpt4 become slowly

[slender bluff]

TEST

[tough urchin]

I think it's broken because when I go on their site it tells me to sign up to the waitlist again

[violet grove]

It’s obvious. Gpt-4 is still a beta, so sometimes it could be laggy or smth

You shouldn’t even ask lol, that’s why there’s the cap of 25 msgs every 3 hours

[whole kettle]

Well looks like GPT 4 may be the last version we may get for awhile, looks like things are getting a bit out of control. I agree with their statements, anything more powerful than what we currently have can cause alot of damage if we're not ready for it.

[hollow quest]

true

[violet grove]

I think gpt 4 will be released publicly on plugins release

[polar timber]

what do you mean publicly?

[violet grove]

to all, even to standard users

[rapid grove]

why

[pure ether]

gpt4 is enough

[violet grove]

yeah, it’s very powerful

[hexed island]

Why am I only allowed to speak English

[polar timber]

it's not an AGI

[viral tide]

🥸

[native flax]

It seems so

[polar timber]

ask ChatGPT

[whole kettle]

Yesterday alot of industry experts signed up an open letter yesterday, quote:

Elon Musk and a group of artificial intelligence experts and industry executives are calling for a six-month pause in training systems more powerful than OpenAI's newly launched model GPT-4, they said in an open letter, citing potential risks to society and humanity.

[wind urchin]

source?

[polar timber]

you can't stop the AI train now

[whole kettle]

Reuters.com

[pure ether]

He's rich and doesn't represent the truth, I'm neutral on what he said

[grand wind]

Elon Musk the "Industry expert" lol.

[sonic gust]

.

[wind urchin]

by stopping the training I think they mean working on further gpt models

[whole kettle]

Yup

As in 4.5 and 5

[wind urchin]

seems like an L for the plus users

[wind urchin]

tbh ill never trust elon musk to do anything for the good of society

always has some other angle

[whole kettle]

He was one of the founders of Open AI

Or investors

I can't remember

[wind urchin]

yeah he gave an initial 1bil

i think it was 1bil with a bunch of other people actually

one of the co-founders iirc

then he resigned in 2018

[sharp root]

2023

[long parcel]

How to use chatgpt to make product pictures

[whole kettle]

But he's still donating as far as I've read, don't know what that means, he's on some sort of Donors list.

[rustic siren]

All hail our AI overlords

[runic lance]

/GPT 4

[wind urchin]

can't wait to have gpt level knowledge for personal voice assistants

[rustic siren]

Yeah, I find Siri so frustrating now

[runic lance]

how to use midjourney freely

[rustic siren]

you can't they paused free trial, go see their Discord

[whole kettle]

Well Apple seems to be focusing on AI as well, expect some updates to Siri in the near future.

[rustic siren]

Oh really? Nice!

[runic lance]

i tested 3 days ago by invate a new acc and work normal, but now cant work

[polar timber]

you can use bing to make images for free, it uses Dall-E

[wind urchin]

Google's assistant is the only good personal assistant atm. It's weird to see how they can't have a pie of this gpt-like technology considering it's "Google".

[echo vortex]

same, i have no trust in him

[rustic siren]

i'm actually considering going back to Google Home but got so many connected Apple devices

[mint pier]

lol

[mint pier]

apple ecosystem indeed

[echo vortex]

im in it too

[rustic siren]

Yeep

[mint pier]

damn

im the only one in my family with an iphone lol

[polar timber]

I switched all my stuff to google home from amazon

[compact moss]

get out, quickly

[rustic siren]

The dawn of the Homelab ecosystem is soon no?

[echo vortex]

but im having windows as desktop tho, but everything else are apple devices

[rustic siren]

Ahh yeah Amazon... I avoided

[echo vortex]

i hate MAC system i like windows

[mint pier]

W

[rustic siren]

Fair, I use mac for work and have Win11 on a window on my mac...

[slender void]

What are some great, unique and productive use cases of GPT-4 that you guys have thought of?

[echo vortex]

😂 damn slap on mac face

[rustic siren]

😂

[slender void]

Mac for work
Linux for coding
Windows for gaming 😉

[wind urchin]

see if your devices are supported by "Matter". Apple has a great ecosystem and everything is buttery smooth but it's assistant really is a huge disappoint and also the reason why I'm still stuck with Google.

[rustic siren]

Yeh that's true

[rustic siren]

Yeah I only use GPT for my assistant now lol

[thorny valley]

I've found.. if you don't know how to do any other option, you cannot do it "wrong".

[echo vortex]

you got it

[rustic siren]

Just means I need to 'switch on' the lights myself

[little dagger]

Open al very nice

[thorny valley]

If you have Other options that don't work show up, doing things wrong can happen.

Erradicate multiple choice testing.

[wind urchin]

all fun and games until servers are loaded more than half the time

[rustic siren]

yup 😢

[whole kettle]

When Apple does it, it will be quite interesting, imagine talking with Siri like you talk with GPT 4, I'm curious how they'd handle the costs for the service, I expect them to release the feature for free so it will probably take some time for them to handle the costs issue. I don't see Apple releasing it as a paid feature, they'd probably want it to be something that comes with an Iphone out of the box.

[echo vortex]

but i bet they will just transfer that cost by increasing the cost of iphone

[wind urchin]

They'll likely make it a part of the Apple One subscription. There's no way it's gonna be free.

[whole kettle]

Probably, or they may come up with some sort of Premium subscription included in the iCloud ecosystem.

[echo vortex]

yeah otherwise they will have that as a special feature if they ever gonna get the service from microsoft, cuz we know microsoft aint gonna give this for free

[whole kettle]

Yeah, unfortunately it will probably be the case, unless they can somehow make up for it otherwise.

[fickle hazel]

Creative description

[echo vortex]

no chinese or other languages in this chanel

[smoky bloom]

what happens if i dont pay openai at the end of the month

[fickle hazel]

logo Creative description ¯_(ツ)_/¯

[wind urchin]

How much ever I'd love for it to be free, Apple not charging for a feature like this would heavily disrupt their brand image imo and that's for sure something that'd NOT wanna do.

[zealous acorn]

How quickly are people seeing responses on their GPT-4 waitlist applications? In general

[smoky bloom]

not planning to, just curious

[opaque crest]

Design a traditional tea cup painting

[wind urchin]

The API?

[zealous acorn]

yeah yeah

[echo vortex]

this is not where you use chatGPT, this is just a chat room

[hollow trail]

Finally got API and Plugins access🤣

Thank god

[plain abyss]

Hi guys I'm currently trying to use ChatGPT to add punctuation and paragraphs to youtube video subtitles. However ChatGPT is running very slow for me today. Does anyone know any other similar AI bots that could do this for me?

[wraith ginkgo]

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

Please report any incident by sending a DM to @open crescent immediately.

[wise linden]

They're currently overloaded i assume ( Not sure )

[bronze minnow]

I can't use chatgpt now

[echo vortex]

i bet because heavy traffic, so many ppl using it, so if you dont subscribe to it, you wont get the priority to use it

[wind urchin]

I got access to it a while back

[prime locust]

Can we talk private, I also today received invite to API, but i”dnt know how use it

[echo vortex]

this channel is english only

[zealous acorn]

GPT-4 is sooo good man. They really have a winning product here. Code generation is so natural, and its on point with context

[whole kettle]

Even if they increase the costs of their products, since it's an ongoing service I don't think it would cover too much, a subscription would solve that while keeping the price competitive. I still hope they'd release it for free via Siri, if they can do that and it's at least as good as GPT 3.5 they will gain alot of new customers in their ecosystem.

[echo vortex]

yeah make sense to me

[native flax]

I think they are experiencing high traffic

I got confused by the links and the site too

[echo vortex]

.....lol i bet

[tawny ruin]

using the gpt 4 api, does anybody know, if the context is applied for the whole converstation, e.g. I give a 8k token context and after that another one, can GPT 4 remember the first one?

[zealous acorn]

its pretty simple. Basically just consists of sending a JSON table to the endpoint URL (https://api.openai.com/v1/chat/completions) and it spits out another table, which you decode

you can do it in basically any language that has http request libraries (python, lua, etc)

I'm sure GPT-4 would be happy to help you figure out API calls 😄
I asked for an entire lua library that deals with API calls hehe

[fast wigeon]

Man I'm signed up since day 1 and still no invite.

[prime locust]

And then I was embarrassed that I was a simple user, so conditionally quickly got access to version 4. this is how the letter “You're invited to use the OpenAI GPT-4 API!
You can now access GPT-4 models with 8K context via the existing OpenAI API.”

Thanks, I will try today

[zealous acorn]

ah haha thats fine, AI is a learning experience for everyone. Gotta start somewhere ^^

[round fjord]

Give me a picture of a handsome mechanical boy

[hollow trail]

No offense, but you can actually ask ChatGPT about it.

[native flax]

I did

[prime locust]

Okey

[native flax]

He said "Hello! As an AI language model, I don't have access to the specifics of the email you received or your account details. However, I can tell you that I am ChatGPT and not GPT-4. The link you provided is for the documentation on how to use GPT-4 through the OpenAI API.

To access GPT-4, you will need to have an OpenAI API account and follow the instructions in the documentation you mentioned. If you don't have an API account yet, you can apply for access on the OpenAI website.

I hope that helps!"

[hollow trail]

Maybe soon.

[native flax]

and my question was "Hey chatgpt I have been given the opportunity to test your younger brother gpt-4 however I did not understand the invitation mail, they have given a link; https://platform.openai.com/docs/models/gpt-4, and said "You can now access GPT-4 models with 8K context via the existing OpenAI API." So are you now gpt-4 or do I need another page just like this one to accesses and start testing it.
I have already read the documentations however I could not find the answer of my question, lastly if it is a different internet page can you provide me a link"

[pallid geyser]

I signed up on day one and got it this morning :)

Any ETA on when image recognition feature will be added to GPT-4?

[hollow trail]

In most public areas, people cannot give better info than GPT-4 does, so……

[slow beacon]

anyone here australian with access to chat gpt plugin yet?

[hallow owl]

yes

[prime locust]

yes, you just boasted, so I asked for help

[strange mural]

i using code in chatgpt playground but in response text is null any one help me

[hollow trail]

Seems to have nothing to do with nations.

[zealous acorn]

    local requestData = {
        model = "gpt-3.5-turbo",
        messages = {
            {
                role = "user",
                content = "The prompt goes here, hello!"
            }
        },
        temperature = 0.8
    }
    local strData = json.encode(requestData)
    http.post(
        "https://api.openai.com/v1/chat/completions",
        strData,
        function(body)
            local data = json.decode(body)
            local response = data.choices[1].message.content
            local tokens = data.usage.total_tokens
            print(response)
        end,
        function(body)
            print("timeout")
        end,
        {
            ["Authorization"] = key,
            ["Content-Type"] = "application/json"
        }
    )

a simple example of an entire API call GPT-3.5-turbo. I dont yet have GPT-4, but its literally just a cut and paste swap

lua, btw

[slow beacon]

so lucky! did you join the waitlist straight away and how long did you have to wait?

[distant wyvern]

key points of tibetan design

[hallow owl]

4 days

[slow beacon]

ah ok, saw someone on twitter saying it could be. glad to know i have a chance then

[grave vale]

I signed up for api accesss late and got api access within 2 weeks.. they seem to be processing wait lists pretty fast

[steady trout]

there is something I don't get about the API billing. if I need to send the entire chat history with each call, how is billing not quadratic in the length of the chat?

[compact dirge]

it is. but there's different billing for prompt vs completion tokens

[zealous acorn]

yeah, the tokens get more and more expensive the longer the convo goes on

[short saffron]

I just signed up for the subscription, it told me that I could try GPT4, so I'm trying to try it, and I keep getting network error... is it down right now?

[slow beacon]

i joined the api one late as well, on the 19/03, still no access to that one yet. hoping i get more luck with the plugins

[grave vale]

it's up for me

[short saffron]

More specifically, "Something went wrong. If this issue persists please contact us through our help center at help.openai.com."

[grave vale]

I bet you get api access by next week, unless I just happen to be an ideal demographic for them

[shy tangle]

i just got api access and i applied like 20 minutes after the gpt 4 page dropped

[slow beacon]

seesh, i hope so! do you submit feedback often?

[grave vale]

yes yes, I'm not getting that though, my page is working fine