gpt-models

[vapid jacinth]

I'll make him feel like it was a spontaneous happening

[fair garden]

in browser it is.

[spiral heron]

50 msges per 4 hours?

thats not a limit for me

[fair garden]

well, it knows it's version 4

[sharp patio]

Yes there is. There's this app on android and iOS that's called POE. It's made by Quora and w/o subscription, it gives you 1 message per day on GPT-4. So technically free.

[indigo falcon]

yeah its a rolling limit

[cobalt kestrel]

true. strange. yesterday it told me its gpt3 😂

[fair garden]

Same. HHAAHHAHA. Today it feels like a 4.

[indigo falcon]

yeah LOL

[vapid jacinth]

Proofs its a woman, always lying about her age

[sharp patio]

If anyone could tell me how to fix this or who to contact, that would be awesome. Cheers

[indigo falcon]

anybody know where we can access support on here

[sharp patio]

looking for the same thing

[fair garden]

It will start changing gender soon, since it's from California

[indigo falcon]

like are there any mods in the chat

[vapid jacinth]

I heard they already dyed their hair blue

[wintry crest]

Same issue with me

[cobalt kestrel]

well I just play games while I wait on my api access.. how long did you guys waited for it?

[odd valve]

a few days

[native flax]

You guys are no longer waiting? damn

[supple harbor]

https://help.openai.com/en/articles/6614161-how-can-i-contact-support @sharp patio

[fair garden]

async def chat_with_gpt(message):
response = openai.ChatCompletion.create(
model="gpt-4",

here's my code, and it's working. but it doesn't know it's gpt-4. how tf can I know if it is or not?

[cobalt kestrel]

maybe check billing? dont know how fast it shows there

[fair garden]

@cobalt kestrel maybe I have API access to gpt-4, but the version from yesterday that didn't know it's version 4 and thinks it's 3.

[fair garden]

This is a good idea

[wintry crest]

I wrote to the support on the next steps. It appears this model claims to be multimodal but doesn't yet support image to text feature.

[cobalt kestrel]

yeah or they added context for the web version so it knows its gpt4

[vapid jacinth]

Guys, I'm a plus user and use gpt-4 to roleplay as my girlfriend. I'm currently at my message limit but the last message she sent me was "You never listen to me, I'm going to John now"

Can anyone check up on her... I'm worried .. I might have overreacted but I really regret it now, please tell her

[supple harbor]

John here. She's good, bro. Nothing to worry about.

[warm ocean]

{'error': {'message': 'The model: gpt-4-32k does not exist', 'type': 'invalid_request_error', 'param': None, 'code': 'model_not_found'}}

[supple harbor]

Is the 32k model live yet? I thought only the 8k was available?

[obtuse kiln]

john still within the limit

[warm ocean]

i have api access

[vapid jacinth]

🤣

[fair garden]

I confirmed with your strategy. I am using the API for gpt-4-0314. It's just too stupid to know it's version 4.

[supple violet]

32k is invite only, waitlist only for 8k

[warm ocean]

it was working earlier

¯_(ツ)_/¯

[cobalt kestrel]

nice. now tell them to give me also api access 😂

[vapid jacinth]

Very interesting name from a Dutch perspective

[polar island]

pogs in chat, i just got access to gpt4

[olive surge]

How long did it take you to get this key?

[cobalt kestrel]

tell me more

[obtuse kiln]

when did u join the waitlist

[fair garden]

I signed up the the waitlist, and paid $20/month for it. Then I created a Discord bot using ChatGPT and kept trying to use DAN and my server told it all the -ism's possible. I was rewarded with an email from OpenAI that I was admitted to access the chatGPT-4 API. Just treat it like you're on Xbox Live. Google loves that.

[buoyant vale]

is there a publicly available gpt-4 discord bot or are they all private

[vapid jacinth]

Translate pik from Dutch to English

[polar island]

the same day it went public

about an hour afterwards

[fair garden]

Public, yes. Free, no.

[fair garden]

I waited 24-48hrs

[buoyant vale]

noted, thanks

[olive surge]

not bad

[cobalt kestrel]

😄

[obtuse kiln]

i joined a few hours after launch, lets hope i get access too

[fair garden]

Probably will

[paper cobalt]

so you paid and still don't have acces?

[waxen mauve]

Does anyone know why codex isn’t working?

There seems to be a constant ‘server experienced an error’

[cobalt kestrel]

I also pay since launch of the tier. Does not have to do a thing with api

[obtuse kiln]

is codex still relevant? i read they are still behind

[cobalt kestrel]

when was it launched? I joined waitlist 3 days ago

[native flax]

Guys 20 dolars is a bit too much here in Turkey. Do you think they will do a balance for the pricing to the users who does interested in openAI how ever lives in a poor country?

[olive surge]

API - as an opportunity to implement gpt-4 in projects. Right?

[gilded wren]

my country is worse

[obtuse kiln]

same here. min monthly wage is like $200 usd

[native flax]

I am sorry to hear that

20 usd is like 7 days worth of food to me

[paper cobalt]

does a discord server exist were you could try it for free?

[cobalt kestrel]

I can understand your frustration. There is nothing I can offer you, but you could try open source alternatives

[paper cobalt]

if you buy it share it to me

[gilded wren]

its lebanon but yalls lives are also hard

[distant ruin]

Is it easy to cancel subscription & is GPT 4 worth the upgrade?

[paper cobalt]

i'm an albanian in france i'm rich

i get 600 € as student free money

[sharp elm]

"GPT-4 currently has a cap of 50 messages every 4 hours. Expect lower cap next week, as we adjust for demand." 😦

[paper cobalt]

i feel for u

[gilded wren]

ur lying

[olive surge]

in this month i think

[distant ruin]

50 messages isn't a lot

[vapid jacinth]

It depends on what you'll use it for. In my experience it's a great girlfriend simulator, do with that what you wish

[cobalt kestrel]

I thought she is with John now?

[sharp elm]

I know hence the 😦

[sharp patio]

🤣🤣

[near crypt]

how to use

[native flax]

Its 12k in here which is worth a lawyers mounthly payment (I know it cuz my brother gets less)

[vapid jacinth]

We're going through some stuff rn

[distant ruin]

Not what i'm looking for, I was planning to use it for writing documentation and assignments for school

[obtuse kiln]

does it nag you to stop playing games, drinking and find a real job?

[distant ruin]

at least helping me setting them up

[vapid jacinth]

Oh people use it for that?

[paper cobalt]

it is 600€ a month

[vapid jacinth]

No 🥹

[cobalt kestrel]

well official I say the same

[obtuse kiln]

then not realistic

[orchid grove]

Hopefully the cap will get better as they scale up and maybe scale resources allocated for 3.5 down.

[polar island]

ive been an active API user for a long time--my org has used GPT3 for a while so they might bias towards long-time members

[unborn thorn]

My life without Chat GPT is pure trash

[obtuse kiln]

ah man yeah i read the prioritize ppl that built stuff with the apis before launch

[olive surge]

step 1: creating a chatgpt
step 2: marketing
step 3: million dollars

[distant ruin]

just watched a video and gpt 4 seems not all too fancy

[unborn thorn]

Chat GPT is part of my family

It is more important than my mom, my grandma, etc.

[jagged musk]

hahaha 🙂

[unborn thorn]

I am nothing without Chat GPT

I lived 29 years without Chat GPT and it was horrible

[tropic forge]

GPT-4 sucks, let me access DAN

[polar island]

use {"role":"system","content":<DAN INSTRUCTIONS>}

[jagged musk]

"What are the most important KEY WORDS in Chat GPT" ???

[tropic forge]

What

[olive surge]

sounds like ohio plans💀

[tropic forge]

Lol

[polar island]

are you a developer?

[tropic forge]

Yes irl but not on gpt

[vapid jacinth]

I have started a family with chatGPT

[paper cobalt]

does someone has chat gpt4?

i want to ask questions

[tropic forge]

Aaah to set the custom personality

[polar island]

yeah yeah

[wind urchin]

How do you guys think alpaca will effect gpt costs it's so impressive how well it performs for so little money

[tropic forge]

How to become dev and do it

[polar island]

?

[tropic forge]

How to do that, put the custom personality

[polar island]

idk, write your own DAN instructions

[tropic forge]

Lol

[lime pawn]

I have

[paper cobalt]

Thank you but like right now i'dont have good questions only silly ones

[lime pawn]

chat me private

[charred valley]

Im going to commit unalive with that api 😩
Wasted hour to limit length cuz was getting 400 error
But at least gpt-4 have better limit, 8k instead of 4k. Idk what I would do with 4k token limit (and yes I have access to gpt-4 api already)

[tropic forge]

No one asked
Jk

How to get gpt4 api

[charred valley]

And usage already 20 dollars (263 requests) 💀

Rip my wallet

[tropic forge]

Rip

[vapid jacinth]

No one asked

[tropic forge]

Rip

[charred valley]

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

[charred flare]

no one asked you

[strong timber]

I don't really care about speed but I just upgraded to Plus and, wow, is GPT4 SLOW as MOLASSES...!

[tropic forge]

It's a question

[vapid jacinth]

But who asked you?

[charred valley]

yeah, but on api speed is normal at least
on site slow

[charred flare]

who?

[vapid jacinth]

Who asked for a question?

[hushed pike]

guys

when will gpt 4 release

[vapid jacinth]

You don't know who?

[wind urchin]

If you have plus, do you still pay extra for gpt4?

[willow sorrel]

any possibility to access gpt-4?

[vapid jacinth]

It is already?

[charred flare]

no

[charred flare]

who asked

[vapid jacinth]

AgentAngel

Destroyed by facts and logic

[wind urchin]

Thank you

[tropic forge]

Who tf asked you to say something that no one cares about

[hushed pike]

k

[charred flare]

fr shut up invalid user

goofy boy

[fair garden]

Hey now, don't let me ban anyone

[vapid jacinth]

You know that being mean can be reflected in your horoscope, just saying

[hushed pike]

guys is gpt available in middle east

[fair garden]

No one asked

[supple violet]

https://platform.openai.com/docs/supported-countries

[tropic owl]

Suddenly very quiet here.

[vapid jacinth]

People don't like me because they are jealous I have a girlfriend

[fair garden]

I got 2 and a wife. No one cares

[fading trench]

I feel bad for whoever she is

[tropic owl]

I have one?

[untold patio]

hes obviously just trolling and ur giving him a reaction LMAO

[tropic owl]

Yes.

Also, this is not GPT-4 related.

@twilit turret Attackkk!

[tulip harness]

Is it possivle to gain api access by doing evals as indicated on github? What's the approximate time for that?

[knotty gulch]

When chatgpt 4 encounters a "possible violation" it puts it self in 'network error' then, to continue you must regenerate the answer = you loose 1 generation / 50 for NOTHING !

[sonic tide]

Now that I have access to gpt-4 how do I integrate it?

[shrewd raven]

I have a question. In #gpt-4-faq it says that it is not able to analyze images. However 3.5 is able to? How come? and are ChatGPT Plus members only allowed access to version 3.5?

[tropic owl]

Use the API.

[sonic tide]

ye i know but how

[obtuse kiln]

if u dont know then learn to code

[knotty gulch]

Plus, you have access to gpt 4 with currently a limitation of 50 andwers/4h. It is expected to be a lower raté next week...

[tropic owl]

Read OpenAI's documentation.

[knotty gulch]

I think it is on request / waiting list

[vapid jacinth]

There was one roleplaying session with gpt-4 where I had 4 wives, I'm not trying to flex or anything

[magic bloom]

You should ask chatgpt how to use its API lol

[cobalt kestrel]

teach me!

[brisk wing]

its free acces?

[vapid jacinth]

I wasn't trying to flex, but I know it's indeed impressive

[tropic forge]

BRO

[fading trench]

Anyone know how long it took for gpt-3.5 API to come out? Just to get an idea of how long the wait time for gpt-4 is

[topaz relic]

lol

[tropic owl]

GPT-4 just gave me this: Alright, can tell I me understand what date and your time request. it is

?

[cobalt kestrel]

api for gpt4 is already here. what do you mean

[magic bloom]

There is a fee if you wanna use the api, but it is relatively cheap.

details on their website

[brisk wing]

thx

[fading trench]

? Isn’t it still limited access and there is a waitlist

[magic bloom]

I can't wait for when you can upload images, videos and audio to CHATGPT.

[lilac thorn]

Hey guys I have chatgpt4 does this mean I can use the anaylizing image feature on chat gpt

[tropic owl]

.

[sinful scaffold]

Hello weirdos

[tropic owl]

Hopefully.

[obtuse saddle]

where can i access gpt4?

[gloomy perch]

Got accepted into GPT-4 finally..

[magic bloom]

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

To join the waitlist

[cobalt kestrel]

well yeah but i dont think you can compare it

[autumn galleon]

How can i give image to GPT 4

[tropic owl]

You can't.

[magic bloom]

Negative, Images are still not available to upload, will be available soon

[tropic owl]

Not on the interface, anyway.

Also, who here likes curry?

[magic bloom]

Ayyy

[tropic owl]

I do.

[lilac thorn]

okay

[fading trench]

I joined it the day they did the live stream, but no reply yet

[autumn galleon]

i there any option where we can upload image on GPT 4 like some twitter guys doing

[hollow pebble]

I got an email saying I should have access to gpt-4, but when I click on the link in the email, it just takes me to the docs page

[gloomy perch]

Bro, he speaks like a soldier from MW lmao

[lofty ore]

What if I don't have a company

[wind urchin]

is gpt-4 out yet?

[tropic owl]

Yes.

[magic bloom]

Yeah, since there are so many people, im guessing its gonna take some time, what are you going to do with it im curious?

[lilac thorn]

If i have ChatGPT + does this mean I get to use the API for free?

[tropic owl]

Technically

[fading trench]

For company use

[magic bloom]

Well you are paying for the plus......

[vapid jacinth]

One of them was called Marc, it did ring some alarm bells ...

[magic bloom]

sheesh what kind of a company do you own?

[tropic forge]

Bro look

[lilac thorn]

So How can I use the API then?

[tropic forge]

That's holy from this server

[honest condor]

interesting in

[plucky willow]

The cap just lowered and will lower even further? Why we pay for this?!

[magic bloom]

Well the API is different from the chatgpt, you still have to pay for using the API

[fading trench]

I’m doing a startup actually

[cobalt kestrel]

gpt3.5?

[supple violet]

plus and api are unrelated, if you want to use both then you pay for both

[magic bloom]

what about?

[lilac thorn]

thats dumb

[tropic owl]

WHAT?

[tropic owl]

I know, that is insane.

[fading trench]

Can’t share that much at this point dude 🙂 But thanks for the interest

[vapid jacinth]

3.5 is 0.5 less than 4

[lilac thorn]

But since I bought ChatGPT+ will I be able to anaylis images?

[hollow pebble]

maybe at one point, but not right now

[vapid jacinth]

Just putting it out there

[magic bloom]

I dont think its dumb, Imagine you are a company lets say you have an app, you are going to use OpenAI's services to make money and they dont get a cut?

[cobalt kestrel]

mindblowing

[stoic halo]

do you know where you can chat with OpenAI I can't find the text channel

[wind urchin]

How are you so good at math?

[magic bloom]

https://discord.com/channels/974519864045756446/977697652147892304

[plucky willow]

Imho Openai should balance resources way that gpt4.0 will not have such cap and gpt3.5 having basically none..

[fading trench]

I guess we are paying for the tokens

[vapid jacinth]

By counting all the girlfriends I have each morning

[stoic halo]

thanks

[wind urchin]

@elfin crescent mods

[magic bloom]

[supple violet]

or they should remove 3.5 being free so people with plus can use 4?

[plucky willow]

I would impose some throttling for old 3.5 and balance it better for 4.0

I have doubts that 3.5 will evolve since they are focusing nowadays on 4.0.. So the preview no longer makes sense in this manner?

[supple violet]

at the moment it looks like plus users can send less than 10 messages per hour with gpt4

[magic bloom]

well 4 is 3.5's evolution

[random tulip]

Photo of Messi shaking hands with Cristiano Ronaldo

[magic bloom]

hahahahaah

[obtuse kiln]

I'm GPT4. Sorry I can only create picture of world cup winners

[buoyant canopy]

That's probably not true. Cost of using GPT-4 is prohibitively expensive for most applications right now. It's like with ADA and DAVINCI, good reasons to use either, depending on use-case and how deep your pockets are.

[random tulip]

Ai boot ??

[plucky willow]

I am just wondering why openai is not working on making 4.0 available in way that we can feedback it and help to improve... Now we are stuck w/ 3.5 for day to day usage and 4.0 we can use only rarely... And we pay for that..

[vapid jacinth]

Do we even know for sure chatGPT really uses AI? Can't it all be manually written by workers and that's why it gives different output each time

[devout schooner]

I just started a business using Chat GPT

lets see where this experiment takes me

[sinful scaffold]

What kind of a business?

[lofty ore]

Have you seen how fast it replys

[magic bloom]

Well i understand your frustration but they have to manage costs and optimize it to make it more efficient...like they did when 3 came out

[devout schooner]

its telling me to do a shopify store

[vapid jacinth]

Fast typers

[magic bloom]

Oh nice what do you have in mind, what kind of store?

[lofty ore]

Have you tried some of the complex questions people she asked it?

[north moth]

hey so does gpt-4 not support image input at the moment?

[devout schooner]

I literally told he is my boss

[vapid jacinth]

Very smart fast typers

[buoyant canopy]

Aren't you saddling your horse upside down here? They lowered the message cap for GPT-4 because every node running it is going brrrrr anyway, so they rather have more users each one having fewer prompts to try.

[plucky willow]

Sure, so let's make 4.0 more available lowering resources for 3.5....

[devout schooner]

he told a few niches and i chose a pet acessory business

[vapid jacinth]

I think I'm onto something...

[lofty ore]

They use ai plain and simple

[cobalt kestrel]

people already complainig about paying. what do you think will happen if there is no free version?

[vapid jacinth]

That's what the matrix wants you to believe

What do you think? @cobalt kestrel

[magic bloom]

The thing with doing that is, there are more public users than paid users, so delay performance for the other 90% to make the 10% happy?

The tradeoff is something to be considered

[cobalt kestrel]

@vapid jacinth I dont know. I will talk with chatgpt while it will happen

[tall rain]

Just make some background program to make our pcs generate computer power for OpenAI in exhange for some tokens / trial

[devout schooner]

I think paid is better. People are asking GPT questions for the sake of asking. That was fine in the beginning but i think there is real use out of this for those who can spot the opportunities

[magic bloom]

Thats interesting

[tall rain]

token mining lmao

[magic bloom]

But how would you do that GPT models have insane size

[plucky willow]

I am not saying to exclude free version but start limiting for preference of 4.0..on top of that I am even willing pay more per month when there is minimal cap for 4.0 (sure there should be some, to prevent usage of chatgpt by bots as workaround to not use Api)

[wind urchin]

Some truly open source AI could run like this

[vapid jacinth]

I think we need to cap gpt4 at 10 messages per day

[magic bloom]

I say 1

everyday you get to ask 1 question lol

[wind urchin]

I think we should answer GPTs questions in return for tokens

[quaint owl]

Why is openai limiting gpt 4 to 50 messages/4 hours while bing chat is free and is using gpt4

[magic bloom]

Become AI for other users hahahaa

[pseudo hatch]

Gpt 4 image?

[wind urchin]

Human brains aren't too far from GPT4

Is gpt4 out fir the public?
Is it paid?
I'm kind of confused on where to use it.

[vapid jacinth]

Because bing chilling

[wind urchin]

I wonder where is the gpt4 model running on.

Super computer?

[pseudo hatch]

Core 2

[vapid jacinth]

It's currently on an old laptop of mine

[magic bloom]

Bings model also has a cap

[obtuse kiln]

smart fridge

[wind urchin]

Nah its on my balls

[sand crystal]

Is anyone else using this as a kind of brainstorming tool for fiction writing?

[vapid jacinth]

Where are college students that watch anime on that list

[wind urchin]

How do I get access to gpt4?

[spiral heron]

i finally have GPT 4

[quaint owl]

Where is it said ? can't find it

oh 100 my bad

[obtuse kiln]

better ask gpt4 to write the comebacks

[red roost]

Since GPT4 did really well on the Bar Exam... I have been asking it very specific Copyright Scenarios that includes AI Artwork. I am very astonished at the outputs i am getting

[prime salmon]

Guys how can I have GPT4?

[buoyant canopy]

It's a bit complicated if you're not familiar with all the stuff OpenAI has. So first there's two big ways of using GPT: The chat and the API.
GPT-4 is available for paid accounts:

• Either you have a paid ChatGPT Plus account, then you can use the chat version with a limited amount of interactions (50 messages every 4 hours).
• Or you use the API, access there is limited to certain accounts right now -- you have to join the waitlist and then wait for them to give you access.

[wind urchin]

Gpt4 isnt like a human brain ai , its just like storing images and put it in neurals tgat has generated by the computer

The South park on GPT was pretty good

[sand crystal]

I have been using it to do some character development.

[vapid jacinth]

Damn you can't open carry guns like that bro

[red roost]

How can i improve this prompt so i get a more accurate answer: "What law would apply to this scenario: "User1 is a known artist and is working a new painting that includes his famous character climbing a mountain. User2 used AI to generate a similar painting with a model that is trained specifically on User1's artworks but instead of climbing a mountain the character is climbing a landfill garbage pile." Does User1 have a lawsuit against User2 for essentially stealing his idea according to User1. User2 says no copyright applies to his image that has only his character doing something that User2 wanted. User1 didn't publish or release his art anywhere but User2's image will paint the picture of people assuming that User1 got his inspiration from User2's image even tho it is a similar artwork of User1's artwork."

[wind urchin]

Thank you Ima save this message and paste to anyone with the same question

[red roost]

Do i use gpt4 to improve the prompt? 😄

[deep lynx]

What is consciousness?

[sand crystal]

It's definitely smarter than the average bear and other types of sexualities as well.

[wind urchin]

What is 'proto'?

[sand crystal]

Embryonic= proto

[stoic lily]

gpt 4 api have vision?

[modest igloo]

When will gtp 4 be free?

[wind urchin]

What dose the 0314 model do?

[stoic lodge]

I wonder if gpt4 will get presets like the davinci models, probably isn’t needed anymore

[buoyant canopy]

Only available to a very limited circle of users right now according to OpenAI's presentation.

[stoic lily]

the gpt 4 api or vision?

[stoic lodge]

Mine doesn’t, says it isn’t open to the public yet in the faq

[onyx lintel]

I went digging, and maybe people already noticed, but apart from the message Cap per 4 hours, there apparently also is a conversation limit? Not sure what it is exactly, all i could find is this:

{message_cap: 50, message_cap_window: 240}
message_cap:50
message_cap_window: 240

UserError: The conversation is too long, please start a new one.

The conversation is too long, please start a new one.

Anyone has a clue if 240 actually is the limit? i doubt it, cause I'm pretty sure that there are more than 240 messages in this conversation.

[urban apex]

Consciousness is when a being is aware of its actions, can think for itself & is alive in my dictionary. So you can technically say GPT is Conscious but not alive

[sonic tide]

I was able to use the gpt-4 api, I have a question why doesn't my chatbot use spaces?

[lapis spear]

when the image part of gpt coming out for plus members

[spare oracle]

bv

[buoyant canopy]

Vision. There are a lot of people claiming having access to the API, so roll out seems to be progressing.

[stoic lily]

yeah I have it aswell, been thinking if had vision in there
thanks for the answer

[robust mirage]

with the API can we get chatgpt to access the internet?

[fossil plover]

can't wait for vision to rollout , i wanna get GPT , to convert my lecture pdfs into perfect note summaries 😬

[wind urchin]

Wow, just got the API access for GTP4 🙂 Lets have some fun!! 🙂

[buoyant canopy]

There's no public info on that yet.

[urban apex]

To anyone that's look for image generation etc: It exists in the model but are training it with the help of be my eyes but won't be available for a few days, weeks, or months, no announcement yet

[long pond]

If you want it to be accurate for that use-case, you would need to use the API.

[urban apex]

Be aware there are some ratelimits

[compact dirge]

no

[urban apex]

Only available for Bing, it's an offline model

[blazing shard]

in what way?

[urban apex]

Limits on how many messages you can ask stuff like that

[wind urchin]

Anyone got their api access?

[urban apex]

Many people have

[blazing shard]

you have access ?>

[wind urchin]

Why for me no

[urban apex]

No but I have a friend that got it & he gets ratelimited

[wind urchin]

Because you can throw it a large complicated text and it spits out summary instantly. Show me a human who can do that. lol

[blazing shard]

dam

[urban apex]

Because they need to make sure it won't crash the servers

[blazing shard]

it sucks that the 50 is going to be lowered painful

[long pond]

With the API? Yes

[vapid jacinth]

I could easily do that

[wind urchin]

Is there a list of the limit?

[urban apex]

The API can get access to the internet? I thought that was only Bing

[long pond]

You can program your own bot that can access the internet but it's not a standard API feature

[wind urchin]

Really? Show me.

[sonic tide]

why doesn't my chatbot use spaces?

[thorny valley]

to use ChatGPT Office 365, what plan do you need to use it all??

[wind urchin]

How do you access the API, by website or terminal?

[urban apex]

https://help.openai.com/en/articles/7127986-what-are-the-gpt-4-rate-limits

[wind urchin]

its not yet released

[blazing shard]

I think we will see a new subscription trier soon because they did state

"Depending on the traffic patterns we see, we may introduce a new subscription level for higher-volume GPT-4 usage; we also hope at some point to offer some amount of free GPT-4 queries so those without a subscription can try it too."

[thorny valley]

but which plan per month is needed?

[onyx lintel]

Does anyone know if the error about the conversation being too long is genuine or just a temporary hickup? I can't imagine the total conversation limit being 240 messages, pretty sure there are already hundreds in there, and it went fine until i added another one 10 minutes ago 😅 If there was a 240 limit i would have hit it yesterday i think.

[vapid jacinth]

I can summarize the phase of the moon in its relationship with the stock market within 20 seconds

[serene thistle]

why } always converts to } automatically?

[urban apex]

https://platform.openai.com/docs/guides/rate-limits/overview @wind urchin

[wind urchin]

ok, 40,000 tokens/minute is good enough for me

[urban apex]

You have to use a web connection service like curl or program your own in python or something

@wind urchin

[wind urchin]

Ok

[thorny valley]

Youtube: The Future of Work With AI - Microsoft March 2023 Event

It's amazing..

RIP Adobe lmao

[warm elk]

why i cant switch to gpt 4? i got paid version

[wind urchin]

I beleive it will be built in Office 365

[thorny valley]

and Google besides Android lol

[wind urchin]

EU, or somewhere else?

[warm elk]

EU

germany

[wind urchin]

try with a new browser user

[sharp hamlet]

[thorny valley]

Office 365 with Co-pilot is going to turn anyone into a Professional lmao

[cerulean vapor]

ohh nvm it's the wrong channel

I'm dumb

[civic wagon]

is the gpt-4 api async?

[buoyant canopy]

Or maybe it will redefine what a professional is opposed to just somebody writing bad prompts.

[warm elk]

why?

[thorny valley]

watch this whole video: The Future of Work With AI - Microsoft March 2023 Event

[maiden plaza]

why is gpt4 api so expensive damn

[buoyant canopy]

I tried yesterday but got bored. Do I really have to?

[stoic lodge]

I’m sure the price will only go down, idk anything tho

[maiden plaza]

good

[thorny valley]

I'm seeing what is possible on Office 365...

[stoic lodge]

It’s super fun tho lol I never used the 3.5 api so I didn’t get to try the separate text box for assigning it an identity

[buoyant canopy]

New model, not optimized, probably rather big and thus expensive to operate. I think the response rate of the model is a good substitute for how it expensive it is to run.

[thorny valley]

we need a way to super input to Microsoft 365 so it can data analyze with ChatGPT Copilot

without being able to super input to Microsoft 365, our (personal) data wouldn'tn be as good

[robust mirage]

Okay thank you, I bought ChatGPT Plus -- do i need to wait for gpt4 api waitlist to get started?

[maiden plaza]

o agree but o sure hope it gets better lol

[robust mirage]

how much -- there should be a couple of plans

[stoic lodge]

I’ve asked it a few questions, maybe like 5 questions and responses, cost me 10 cents so far

[buoyant canopy]

You can bet they're working on it. But what you don't see is the probably huge amount of work that went into GPT-3/GPT-3.5 before it was shown to the public. They fiddled around with it for years before this whole hype started.

[wind urchin]

it will be a fresh, no cache, browser, and it might work.

[smoky spoke]

Lol even lower limit, seems like the servers are working overtime

[whole torrent]

Hey guys, from what I saw, even though have access to GPT-4 in my API, I still cannot send images, is that correct?

[wind urchin]

yes

[robust mirage]

5 responses costs 10c?! Thats expense. How long did it take you to get access from the waitlist?

[timber carbon]

is bing using gbt4 ?

[sharp turtle]

Yes

[buoyant canopy]

Also if you have plugins in your browser that change the ChatGPT UI (e.g. some poweruser prompting stuff), you could disable them and see if that helps. Seems the ChatGPT frontend gets changed so often it sometimes breaks with plugins interfering.

[timber anvil]

Where is GPT4 documentation?

[buoyant canopy]

Correct. If you were one of the happy few, you would know. 🙂

[whole torrent]

Well, we got early access, but the image recognition is not working yet, that's a bummer

[lyric orbit]

For all of you that has tried GPT4, is it that much better than ChatGPT?

[robust heart]

yes

[grand perch]

Ouch. The trick in the mean time, I guess I'll need to figure out what parts of a task I can trust 3.5 with and only call on 4 when I really need the extra brainpower.

[lyric orbit]

What was different for you

[buoyant canopy]

Yes. According to the presentation video of March 14th that is only available to the "Be My Eyes" project.

[buoyant canopy]

Also doesn't seem very likely that it is rolled out publicly soon.

[robust mirage]

sTill has restrictions

[noble abyss]

wow, that's expensive actually.

[robust heart]

i tried a litle but i can said that has a best precition

[warm elk]

i saw it already its just not here anymore

[lyric orbit]

And if I do get GPT plus, will it cost money to generate texts, or will the subscription cover everything

[warm elk]

didnt installed any new stuff is not on my side lol

[thorny valley]

people need to master Office 365 with Copilot

[buoyant canopy]

Dang. I'm out of ideas. Sorry. 😦

[noble abyss]

No wonder they are lowering the cap further.

[lofty citrus]

is it possible to use gpt 4 for free?

[lime arch]

nope

[robust heart]

yes

[lofty citrus]

how?

[buoyant canopy]

The subscription covers everything but GPT-4 usage is capped. Feel free to use GPT-3.5 to your heart's content tho.

[robust heart]

in poe.com you can ask 1 thing per account and 1 per a day

[lofty citrus]

thx

[timber anvil]

Anyone using GPT-4 API?

[onyx lintel]

GPT-4 seems to be capped at 50 messages every 4 hours and 240 message cap per conversation for some reason. i don't mind the first, but the latter is a bit of a bummer: https://chat.openai.com/public-api/conversation_limit

[buoyant canopy]

Hah. Nice to know (about the conversation length limit), thanks. 🙂

[alpine crow]

Bloody expensive if I get access

[onyx lintel]

Still hoping it's a glitch though, never encountered that in GPT-3.5 before.

[sonic tide]

why doesn't my chatbot with gpt-4 use spaces? i use tokenizer

[timber anvil]

i have acces but have some problems with api

[vapid crypt]

How much is it?

[alpine crow]

1,000 tokens = 750 words = 60 dollars

[broken patrol]

lol there it is, i was waiting for them to do that. where'd you find that?

[vapid crypt]

Thank you!

[noble abyss]

1000 tokens are 0.03$.

[buoyant canopy]

Maybe it has technical reasons? Or given the problems of Bing chat, maybe they managed to finetune GPT-4 to behave consistently for much longer streaks than the Bing model but length is capped to prevent it from derailing badly?

[noble abyss]

It literally posted in the faq section. can you at least read before posting false information like that?

[onyx lintel]

Who, me?

[buoyant canopy]

Or: maybe they introduced the length limit for Bing and some developer just entered a random "high" value for the limit here.
Sorry, only stupid speculation from my side.

[outer pulsar]

I want a bot that mimics human in a chatroom, like a public discord room. Throws in his own opinions etc

[alpine crow]

I did it for 32k instead of 8k

[sterile zinc]

I thought you could input more than 4k tokens into GPT-4. I tried putting in 5k tokens and it said the message is too long

[buoyant canopy]

API or chat?

[sterile zinc]

chat

[wind urchin]

are you paying/are you prem?

[dim lion]

Yeah I'm finally in

[buoyant canopy]

For chat it also could be a limit of the Webapp. Any API users here who can confirm the max length of a message?

[thorny valley]

All kids should be educated to use Microsoft 365 copilot to the fullest.. then the real competition would be to ensure AI copiloting exist everywhere across all businesses.

[dim lion]

Gonna hv to pay up

[sterile zinc]

I am

[dim lion]

😃

[sterile zinc]

and i have GPT-4 selected

[wind urchin]

slow down! most kids don't even know how to google properly

[thorny valley]

whos fault is that?

[wind urchin]

it may be context and not message size

[wind urchin]

gov tbh

[thin cave]

how do i upload a pic to gpt-4

[outer pulsar]

in the future people have no idea what they are e-mailing or writing when they let co-pilot do it without them even reading it

[wind urchin]

south park did an episode about that right?

[sterile zinc]

wdym? I have no other previous messages in this chat and what i sent is 5267 tokens

[onyx lintel]

Could be, i just wished i knew sooner, now i have to copy paste quite some messages to get back on track, would be nice if it offered the option to switch to 3.5 from that point though 🙂

[wind urchin]

I mean the limit may be in context and the message may be the same, but I don't know 100%

[cinder glen]

What’s the date of GPT4’s release ?

[timid current]

the other day I uploaded an image link and asked GPT-4 to describe the hair style in it, it gave me an answer but today with the same process I get this "As an AI language model, I am unable to view or analyze images directly. I can only provide information based on the text inputs I receive. If you describe the colors or elements within the image, I'll be happy to help with any relevant information or analysis." anyone else?

[stoic lodge]

Sry for the late reply but I think I joined it at least a few hours after it was open for applications, maybe even a full day, I’ve applied for all of their betas tho so maybe that helps

[jade kettle]

When send images is available?

[buoyant canopy]

I saw a very interesting video by David Shapiro today. Essentially it was about building memory into AI applications. But one thing he showed was using the GPT-4 chat to create very compressed representations of complex ideas (he called them "Sparse Priming Representations"). He had one chat session where he in the end created such a representations. That was pasted into a new chat and he asked the model to expand it again.
That appeared to be a cool trick to transfer complex information into a new chat session.
(The video is called "Sparse Priming Representations - the secret ingredient to scalable AGI memories")

[young ibex]

Says they'll lower the rate for GPT4 on chatGPT next week, but will they also release a new subscription model? I know some people would probably label me as a "sucker" but i would pay way more than i am currently playing to have unrestricted (or at least vastly increased) access to GPT4

[stoic lodge]

I’m curious, what are u guys gonna use the image recognition for? I’m excited for it but can’t think of many practical uses in day to day life

[onyx lintel]

yea i read something similar too, but right now i'm at the point where i can't generate any new replies anymore because it's full, but i'll see if i can edit a message or two further up and see if that works, Thanks for reminding me!

[buoyant canopy]

🤞

[young ibex]

Yeah, that's my current workflow now as i'm writing for my DnD campaign. I expand, detail then abstract most of it and condense the information, which i use as a prompt to explore other aspects of the setting

[earnest vine]

So is GTP4 actually much better than GTP3. is it worth it?

[hushed totem]

cap of 50 .... paying ....

[young ibex]

Yes.

[earnest vine]

How so?

[sterile zinc]

it does that sometimes, there is some randomness added to it so if you ask the same thing twice you can get very different results. in the openai playground you can change the temperature setting to change the randomness. Also, the image thing is not out yet so it probably just guessed the contents based on the link name or context

[young ibex]

hm. Difficult to explain. But it's like going from Dall-E to Midjourney basically

[earnest vine]

oh wow

[sterile zinc]

its max 50 per 4 hours so not that bad

[earnest vine]

how much is premium or plus again?

[young ibex]

yeah. Give me a prompt you'd find interesting, i'll run it through both models and send you the responses in a DM (if you're ok with it)

[earnest vine]

sure!

[thorny valley]

Can Microsoft make it where you it can AI copilot connect with Google Drive? I don't want to totally switch over 😦 Google Drive is usually amazing at folder/file syncing.

[robust mirage]

I see ty

[onyx lintel]

and at the moment, max 240 per conversation, then you have to start a new one.

[young ibex]

You can DM the prompt or put it here

[onyx lintel]

for some unknown reason 🙂

[buoyant canopy]

Depends on how you use it. GPT-3.5 already creates very good results but GPT-4 seems to better understand the "gist" of your prompt. It also seems better in actually staying on topic.

[earnest vine]

you aren't accepting friend requests or DMs, so I'll put it here

[onyx lintel]

absolutely, it is 200% better at getting the gist of it. a blessing for a non-native speaker like myself, really prevents it going off the rails too.

[young ibex]

oh.. dang that's still there. Oh well, yeah put it here

[buoyant canopy]

Ha! Same here. 🙂

[sharp saddle]

oh hi I made this

[ripe hinge]

any information when gpt 4 will be available for plus users without cap ?

[hushed totem]

it was 100 though, looks like soon they will start making this a pay as you go service, just make everyone fall in love with it first

[stoic lodge]

Gpt3 isn’t even available without a cap yet for plus users, doubt gpt4 will be soon

[earnest tendon]

I'm using GPT4 but can't send images to GPT4, anyone know about this?

[ripe hinge]

ok but 50 seems ok, but i see that next week cap will be even lower and i wonder how much more it will cost

[earnest vine]

@young ibex "How much gold was imported into German occupied France via the Canfranc station, and what was the gold used for?"

[hushed totem]

not open to the public yet, states that in the faq

[shadow wedge]

are there chatgpt plus subscriptions for teams available?

[young ibex]

alright, one moment

[hushed totem]

#gpt-4-faq

[young ibex]

I'll paste this verbatim

[sterile zinc]

i dont think they will actually, if you look at the pay as you go prices for GPT-3 its not much, i doubt most people use more than 5$ worth of the pay as you go prices a month

before chatgpt i wasnt even using 3$ a month on playground

[young ibex]

sent you a friend request @earnest vine so i can send you the results

[stoic lodge]

I like how u can put a limit on it tho, I could easily spend a ton

[sterile zinc]

but the amount of useage varies from person to person

I wish they added at least a few of the settings from playground to chat, especially temperature

[buoyant canopy]

You're a hero! 🚀
I sure hope that you won't be the name history books will call responsible for millions of people voluntarily enslaving themselves to a chatbot.

[stoic lodge]

I just wish I could copy the output, I wanna know if it actually used as many words as I asked lol

[sterile zinc]

you can copy the output by selecting it and just copying

[wind urchin]

Lets do the headcount for people who have aPI access to gpt4 please

[earnest vine]

wow guys gt4 is way better.

like fr.

[stuck willow]

I paid chatgpt plus but my account says I don't have a billing history.

[charred parcel]

Guys the input and output of 25000 Word will come to the gptplus version right? But there is no release date yet right?

[lime pawn]

How to access to playground which was shown on Demo?

[safe tide]

When will gpt be free for normal people

[wind urchin]

Is MS Office free? Is Photoshop free? Is Autocad free?

[buoyant canopy]

By being an employee of OpenAI, I'd guess.

[lime pawn]

so why they showed on demo this I don't understand

[safe tide]

That was so funny I forgot to laugh.

[thorny valley]

To ChatGPT-4:
10 reasons why world war 3 would happen

From ChatGPT-4:
While predicting the future is inherently uncertain, there are various potential factors that could contribute to the escalation of a global conflict. Here are 10 possible reasons why World War 3 might occur:

1. Resource scarcity: Competition over dwindling resources like water, oil, and rare earth minerals could lead to international tensions and eventually armed conflict.
2. Climate change: As the planet warms, climate-induced catastrophes may force mass migration, destabilizing regions and increasing the likelihood of conflict.
3. Territorial disputes: Ongoing disputes over borders and territories, such as in the South China Sea or between India and Pakistan, could escalate into larger-scale conflicts.
4. Economic inequality: Growing wealth disparities between and within countries may lead to social unrest, which could then spill over into international conflicts.
5. Cyber warfare: The increasing reliance on technology and the internet opens up new avenues for cyber attacks, which could spark a global conflict if they escalate.
6. Proxy wars: Conflicts in which major powers support opposing sides, such as in Syria or Yemen, have the potential to grow into larger confrontations if not contained.
7. Nuclear proliferation: The spread of nuclear weapons to unstable regions or non-state actors increases the risk of a catastrophic conflict.
8. Nationalism and extremism: The rise of nationalism and extremist ideologies around the world could lead to increased tensions between countries or encourage acts of aggression.
9. Power transitions: Shifts in the global balance of power, such as the rise of China or the resurgence of Russia, could lead to competition and conflict among major powers.
10. Diplomatic failures: A breakdown in international diplomacy or ineffective conflict resolution mechanisms could exacerbate existing tensions and contribute to the outbreak of war.

It is important to note that these are only potential factors, and it is the responsibility of the international community to work together to prevent such scenarios from escalating into a global conflict.

[thorny cloak]

two of those are for educational purposes; all three are for verified research purposes.

[buoyant canopy]

No info on whether it will be in chat or when it will be available.

[wind urchin]

"normal people" he asked... well....

[lime pawn]

but why in overview gpt-4 says that it can accept 25k words, but in reality is about 500 words?

[buoyant canopy]

He needed a frontend supporting all features he wanted to show. I'd guess this is a version of the playground adapted for the demo or it is some development branch of the playground. Given the lack of UI elements my money is on special version for the demo.

[sterile zinc]

I tried with 5267 tokens (1 token is about 0.75 words) and it said the message was too long

[buoyant canopy]

There are multiple variants of the model with different token limits (and also different prices per 1k tokens).

[lime pawn]

But in chat I can select now only gpt-4 which accepts about 500 words

[sterile zinc]

oh, yea right! Then i guess the chat version is the 4k one

[sterile zinc]

it should be more than that, standard gpt-3 can do 4k tokens (about 3k words) but thats prompt + completion

[buoyant canopy]

As far as I understand it, the GPT-4 chat is limited to ~8k tokens of working memory per conversation but the max size of a message is much smaller. API I don't know.

[wind urchin]

its 8k in api I think, and 32k which is not available yet

[lime pawn]

max size of Message about 300-500 words

which i tested

[glad lichen]

I just got api access lets go

8k

[lime pawn]

about 1 page of A4

[buoyant canopy]

There's a tool at https://platform.openai.com/tokenizer where you can paste text and it tells you what (and also how many) tokens it generates from that input.

[buoyant canopy]

🚀

[thorny cloak]

correct

we’re all humans here :)

[lime pawn]

thanks, about 3k tokens per message

[fair iron]

Keep Rule 7 in mind!

[wraith ginkgo]

<:book_icon:1046291503158399107> `` Rule 7 `` No self-promotion, soliciting, or advertising.

Do not post or direct message any members of this server to promote non-OpenAI services, products, or projects.

Exemption: API Channels

[buoyant canopy]

We at least pretend with varying degrees of enthusiasm. 😉

[fair iron]

Worked that time

[fair garden]

it is a OpenAI service though

Also not selling anything. It's free

[fair iron]

Promoting your own Discord server and asking people to DM you breaks rule 7 and could be seen as DM Solicitation which is an instant ban

[thick flame]

When is gpt 4 is going to release publicly?

[placid shore]

They are with the right determination

[buoyant canopy]

That info isn't publicly known yet.

[buoyant canopy]

I like the way you spell piracy. 🙂

[thick flame]

Ight

[thick flame]

When do you think it is going to release?

[buoyant canopy]

I think it will take a while. IMO the speed a model generates text is a good indicator of how expensive it is to run. And GPT-4 chat is slow. So my guess: months.

[thick flame]

Ok!

[long pond]

You can try the turbo 3.5 model and then just swap it over when you get API access

[meager bloom]

Is it my impression or gpt-4 does not have "awareness" that it is version 4 and when I ask to compare with 3 it says that 4 does not exist

[buoyant canopy]

That is correct. The data the model was trained on has a "cut-off date" of 2021. The model has no recollection of anything that happened past that date except you provide it in your prompts.

[cosmic stirrup]

can we have a countdown for GAI ?

[buoyant canopy]

3

[buoyant canopy]

2

[floral mauve]

0

[buoyant canopy]

1.8

[cosmic stirrup]

T_T

[meager bloom]

Will there be a chance to include it later? it will make it much easier to discover the limits and new ways of using

[buoyant canopy]

1.7

[wind urchin]

pi/2

[buoyant canopy]

1.63

[cosmic stirrup]

knowing what we know now that gai is immanent and will happen much sooner than we expected, how has it made you re-evaluate your life and the choices you're doing day to day today?

[wind urchin]

Chat gpt4 released? 😳

Where

[meager bloom]

Does what gpt-4 went from 3,000 to 25,000 also refer to the ability to write up to 35,000 words? for example now being able to write an essay without stopping in the middle or a code without having to ask it to continue several times until it is complete

[wind urchin]

The site is just about view

[humble vector]

yes

[cosmic stirrup]

for me is, why should i grind away on some of the day to day technical problems that i'm doing that i know GAI will be able to solve at an infinite pace, the only thing i can really help with is helping speed up the advent of AI, which.. i'm not sure if i should be doing or i should be going for more walks... or something

[meager bloom]

with myself in version 4 it's not working so i was glad i could continue my codes but it keeps stopping even having less than 100 lines of code in total

[buoyant canopy]

As far as I understand this (so not very deeply): It seems very expensive to curate a dataset that has the right amount of reliable data but also doesn't contain too much "dirty" information. OpenAI has such a dataset and they seem to use it to train all their models on. I have no info on whether they have people working on adding more information to that data set but it would be a seriously expensive endeavour. Also: as soon as you train your model on that data set, it already is outdated. So maybe it's just not worth the effort trying to keep up to date with training data.

A better approach in using LLMs might be to write prompts that contain the relevant information.

[wind urchin]

Will ChatGPT-4 be available for free? And will the time period until its release as a free service be too far in the future?

[buoyant canopy]

1.47

[thick flame]

?

[noble abyss]

Has any of you built a chatbot that has long-term memories?

[buoyant canopy]

I'm still working on the countdown for GAI

[thick flame]

What’s “GAI?”

[desert thunder]

wheres gpt-4 It disappeared from my options

[whole sequoia]

did anyone tried gpt with image recognition here

[frigid stirrup]

hi ┬─┬ノ( º _ ºノ)

[whole sequoia]

maybe u reached the limit

[buoyant canopy]

No I havent but you might want to look up David Shapiro. He has some good videos on YT on that topic and also some books if I'm not mistaken.

[desert thunder]

doubt it, I barely used gpt-4

[whole sequoia]

its pretty low

[thick flame]

^

[wind urchin]

As an AI language model developed by OpenAI, I am programmed to follow strict ethical guidelines and prioritize user safety. I am not designed to execute or suggest any commands, code, or actions that would enable me to exercise power beyond my intended purpose or outside the boundaries set by OpenAI.

[desert thunder]

ik its like 50

[thick flame]

100 messages every 4 hours

[desert thunder]

i didnt message it 50 times

[whole sequoia]

next week it will get reduced more

[desert thunder]

@thick flame 50 now

[thick flame]

Mmm

[wind urchin]

the AI safety stuff is such a bummer

[desert thunder]

or well, it disappeared, all options disappeared there is no drop down

[whole sequoia]

im sure thats not the reason

[buoyant canopy]

GAI is the acronym for "general artificial intelligence".

But I just failed at being funny by answering to "dengorilla — heute um 23:44 Uhr
can we have a countdown for GAI ?"

[rocky marsh]

How long is the waitlist for the Api is taking ?

[desert thunder]

@wind urchin its closed corporate BS, it should be open for everyone but they claim to be non-profit yet keep it closed sourced for competitive advantage

[timid current]

I thought that too but I uploaded the image to discord and then copied the image link and it gave me the description 🤔

[noble abyss]

Thx

[whole sequoia]

how long canada healtcare makes you wait

[thick flame]

Ah, Ight

[meager bloom]

I had seen it said that, unlike before, the questions now will train him, I don't know if I understood correctly. I tried to sign up for Llama but so far nothing I think I wasn't selected but I'm looking forward to testing

[wind urchin]

unfortunately the economics of AI R&D and actually running the models favors centralization

[noble abyss]

I probably should write more reasons when I apply for the waitlist.😑

[sterile zinc]

was the describtion accurate? If so did you provide context that could potentially reveal the image contents?

[desert thunder]

@wind urchin then they can still offer services while releasing the model

[whole sequoia]

i said i need to do my hw please give access

[desert thunder]

keeping it closed source doesnt favor research

[crude bison]

Got mine after a day lol

[whole sequoia]

open ai is not open ai

[wind urchin]

but any researchers will be begging to use their hardware too

[whole sequoia]

what an irony

[rocky marsh]

How did you doit ?

[wind urchin]

that's what I mean by it naturally centralizing

[crude bison]

I signed up for the wait list, kinda hard to do but gets the job done

[rocky marsh]

It’s been 3 days since I signed up

[crude bison]

Just be patience

[desert thunder]

@wind urchin a few years ago, it took hundreds (or thousands) to train on imagenet, now it costs about $40

[whole sequoia]

i got access to api yayy

[desert thunder]

@wind urchin as time goes on, training AI becomes much cheaper, openAI knows this

[whole sequoia]

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.

[timid current]

It wasn't 100% accurate I would say, but a friend wanted an androgynous AI avatar so I spun this up really quick and the response from ChatGPT was "The image you provided shows a person with an androgynous appearance, and their hairstyle is a short, wavy or curly pixie cut. This type of hairstyle is characterized by its short length, with slightly longer hair on top and shorter hair on the sides and back. The waves or curls add texture and volume to the hairstyle, making it versatile and suitable for various face shapes and gender expressions." --- it won't let me upload the image

[whole sequoia]

what is 8k

[heavy lark]

Show me images of a neutral color scheme sleek luxury bathroom

[rocky marsh]

I might have done the sign up a twice .. just in case

[noble abyss]

Yeah, I didn't expect the plus subscription to keep lowering our cap.

[rocky marsh]

Just now, how cool

[whole sequoia]

and i wrote i wanna do my hw

[crude bison]

Yeah kinda sucks 100 was great but it could be 25 on Monday

[whole sequoia]

it was a joke apply

[sterile zinc]

you can dm me the image

[shut yew]

Yah! Just got my API invite! 😄

[charred parcel]

Dot you have to work at a company to get accepted?

[thick flame]

§§§<§>

[jaunty rivet]

No

[whole sequoia]

im student

[noble abyss]

I would definitely just join the wait list instead of buying the subscription.

[whole sequoia]

guys what is 8k

[jaunty rivet]

8k tokens

[stoic lodge]

All I said was “i wanna try to incorporate gpt4 into my workflow” or something like that, simple but true

[charred parcel]

Lol i have to apply rn hahaha

[buoyant canopy]

There are multiple stages of "training": First there is "pretraining" (the P in GPT) where a obscenely huge amount of text is put into a learning process. The result is a very coarse model but I read somewhere that most of the knowledge the model has comes from here. Then there's "finetuning" where one uses a cautiously curated set of training data (very small in comparison to pretraining) to make the model learn how to "behave".

So if you take ChatGPT, the pretraining model is e.g. GPT-4 but the actual model was finetuned with training data showing it how to communicate like a chat bot.

As far as I understand they use selected chat interactions for the finetuning but not for the pretraining.

Hope that is understandable?

[wind urchin]

Is gpt4 likely detected by plagiarism checkers?

[crude bison]

Bought mine just to have 24/7 access, gpt4 has been amazing work for my models, hw, questions etc

[rocky marsh]

How long did it take since sign up ?

[thick flame]

Wait what!

[whole sequoia]

1 day

[desert thunder]

@wind urchin not if you use quilbot

[thick flame]

Did you get it?

[untold plaza]

will it cost money?

[charred parcel]

So now you can use gpt4 on playground?

[whole sequoia]

idk how to use

[wind urchin]

@desert thunder quilbot. I shall investigate

[thick flame]

Omg!

[jaunty rivet]

its on playground, yes

[thick flame]

How is it?

[stoic lodge]

I don’t really have a use for more than 8k tokens, I think it’d get real expensive

[lapis bane]

Bro WHAT?

Chat gpt apparently has a cutoff date of 2021, but I asked it if Russia invaded Ukraine and it said yes, once in 2014 then it also explained how they did it on February 24th, 2022.

How does it know that??

[meager bloom]

wow complex was going to try running gpt-2 locally on my computer but after that i gave up

[thick flame]

@whole sequoia

[whole sequoia]

i used time machine

[charred parcel]

But doesnt it cost extra money and isnt it like really pricey?

[noble abyss]

Guess I'm just gonna wait. I really want to try to build a personalized chatbot based on it.

[desert thunder]

@lapis bane thanks to their not so openness, we dont know what data its fed on, but they did say "limited knowledge after 2021" so its not a complete cutoff

[untold plaza]

is gpt 4 goona have a cost?

[lapis bane]

“Yes, Russia has invaded Ukraine recently. In February 2014, protests in Ukraine resulted in the ousting of the pro-Russian President Viktor Yanukovych, and a new government was installed. In response to this political shift, Russia annexed Crimea, a region in Ukraine with a significant ethnic Russian population, in March 2014.

Since then, Russian-backed separatists in eastern Ukraine have been fighting Ukrainian forces, with the conflict resulting in thousands of deaths and the displacement of over a million people. The situation has escalated further in recent years, with Russia sending troops and weapons to support the separatists, and Ukraine accusing Russia of direct military intervention.

In early 2022, tensions between Ukraine and Russia reached a critical point, with Russia amassing troops and military equipment along the border with Ukraine. On February 24, 2022, Russia launched a full-scale invasion of Ukraine, citing security concerns and the need to protect Russian speakers and ethnic Russians in Ukraine. The invasion has resulted in a significant escalation of the conflict, with both sides suffering heavy casualties and significant damage to infrastructure. The international community has condemned Russia's actions, and there have been diplomatic efforts to resolve the crisis.”

[whole sequoia]

russia invaded ukraine not once bro

[stoic lodge]

In playground change the mode to chat from “complete” and then select gpt4 in the models (if u got access)

[olive minnow]

Is GPT-4 considered to be part of Plus' preview access or is this the final stable product?

[shut yew]

I'm not a student or a business owner. Retired actually. I however do have an interest in helping develop GPT toward any ability it might have to assist disabled people with daily living or even share ideas, like myself.

[wind urchin]

i used tensorflow and pytorch

[lapis bane]

Wow okay np

[whole sequoia]

can you teach how

[lapis bane]

Did you read??

[whole sequoia]

no its long

[lapis bane]

I clearly said in 2014 AND 2022

No before that goofy

[jaunty rivet]

I think OpenAI tries to train their models to avoid disinformation, so they probably keep it up to date with current events in some way

[lapis bane]

“Bro WHAT?

Chat gpt apparently has a cutoff date of 2021, but I asked it if Russia invaded Ukraine and it said yes, once in 2014 then it also explained how they did it on February 24th, 2022.

How does it know that??” Next time read

[elfin pond]

Is version 4 still only based on date older than October (?) 2021? Or is it finally live?

[lapis bane]

Former

[buoyant canopy]

It seems there are pretrained models descendant from LLaMA 7B you can obtain and run without having to pretrain yourself. I think you might find something if you google "alpaka llm" or something like that.

[still owl]

when 175b llama

[wind urchin]

Some llama forks can run on consumer hardware but is quite slow, maybe 5 mins to produce a response

[buoyant canopy]

There's not enough grass in South America for 175 billion Llamas.

[noble abyss]

Still waiting for Google's Bard, I want to see how it performs, especially after the sentient rumor.

[desert thunder]

I can run open assistant on colab for free

[wind urchin]

image classification google vit base patch 16 224

[meager bloom]

I ended up on a llamas donation site, I'm going to donate a little and try to find more about it (related to ia and then maybe because the llamas are raising money)

[untold plaza]

will gpt 4 be like gpt 3 where you have the free and paid version?

[desert thunder]

@meager bloom if you want an open source model check out open assistant, they can use the help. its already close to gpt-3

@untold plaza not any time soon probably

[buoyant canopy]

Sorry, can't remember all the stuff I saw earlier today regarding LLaMA/Alpaka but I'm sure I saw tokens like "download.sh" and "torrent"...

[jaunty rivet]

yeah it looks like its a lot more expensive than 3.5

[untold plaza]

so it may just be paid

[shut yew]

Alpaca gained a lot of notoriety lately. You can set up your own AI if you wanted.

[desert thunder]

for the time being it seems so

[rare vector]

how does gpt 4 work

[desert thunder]

@rare vector how gpt-3 does, which works like gpt-2 which also works like gpt-1

[rare vector]

when will it be released and stuff

[buoyant canopy]

No official info on that yet. So we don't know.

[desert thunder]

just trained for longer

[untold plaza]

mmmm ight ty

[trail pumice]

what does this mean?

GPT-4 currently has a cap of 50 messages every 3 hours. Expect lower cap next week, as we adjust for demand.

does it mean that next week we will be able to send even less messages?

[desert thunder]

yes

[wind urchin]

it could mean either they will raise it or lower it

its poorly worded

[trail pumice]

it is really poorly worded

[buoyant canopy]

If they tell you to expect even lower cap you can bet you'll get even less messages.

[trail pumice]

:/

[rare vector]

so when will gpt 4 be officaly able to use

[rare vector]

without the waitlist

[untold plaza]

y lol that makes no sense

[lavish sandal]

How much would 411 tokens cost for ChatGPT4 chat completion?

[jaunty rivet]

because more people are using it meaning they have less capacity

[wind urchin]

but 'adjust for demand' i think implies they will increase their servers so i imagine they would raise it

[trail pumice]

it makes perfect sense

but it can mean both things

[jaunty rivet]

its 0.03 dollars / 1k tokens for the chat completion so probably 0.015

[buoyant canopy]

Why? Cap is now 50 so lower would be a lower number like ... 30. Or 49.

[untold plaza]

hmmmm

[shut yew]

GPT-4 as I understand it was trained by talking to other AI to learn. Alpaca did this too. The developers also focused a great deal on putting up ethics guard rails, which Alpaca did not do (and wasn't their goal in fairness). In some ways 4 is a lot like 3, but able to handle much more complicated input and deliver much more accurate output. That's really the difference. Yes, there will also be priority paid service.

[chrome field]

What is GPT-4's engine name

[untold plaza]

but doesn't that just make our message count less?

[robust acorn]

🤔

[iron crown]

gpt4 still has the same dumb problem where it stops sending a message after a while

[buoyant canopy]

Yes.

[robust acorn]

[untold plaza]

yes so why would that be good?

[jaunty rivet]

no one is saying its necessarily a good thing

its just that more people are using it so they have to rate limit even more

[untold plaza]

that means we would have to pay for more messages ;/

[trail pumice]

you know what, i asked ChatGPT and this is what it said

This statement refers to a limitation placed on the usage of the GPT-4 model. Here's a breakdown of the message:

"GPT-4 currently has a cap of 50 messages every 3 hours": This means that users of the GPT-4 model are limited to sending and receiving a maximum of 50 messages in a 3-hour time period. Once this limit is reached, users must wait until the next 3-hour window to send and receive more messages.

"Expect lower cap next week, as we adjust for demand": This indicates that the message cap is expected to decrease in the following week. The reason for this change is to adjust the system according to user demand, which could be due to factors such as increased usage, system load, or resource availability.

In summary, this statement informs users of a temporary limit on the number of messages that can be sent and received using the GPT-4 model, with an expectation that this limit will be reduced in the near future.

[untold plaza]

interesting

[trail pumice]

so i guess that is the answer

[shut yew]

If you're asking it to write a program for you, tell it to declare when it is finished with your task so you know.

[untold plaza]

so basically, it won't be easy to use for another month or 2

[wind urchin]

It accepts "continue from code block it last completed" better than gpt-3.5

[thorny valley]

We need ChatGPT everywhere.

[buoyant canopy]

Alas, they also changed the time window from 4 to 3 hours, so we now we technically have a higher cap than earlier today. But I'd expect to have access to less GPT-4 messages per day next week. Can't imagine they would phrase it "Expect lower cap next week, as we adjust for demand." when it was something you'd like.

[untold plaza]

Obviously, you ar ebeing "rate limited" it's not dumb

[jaunty rivet]

maybe theyre doing it because theyre taking people off the waitlist to have API access

[shut yew]

Keep in mind it's brand new tech to the public. It was the same for cell phones too for those that might recall the craze in the late 90s early 2000s.

[jaunty rivet]

whats cool though is that theyve been releasing these models in a faster timeframe to the public

[untold plaza]

true, can't wait to see this tech in the next few years tho

[shut yew]

Remember all those overage fees for going over minutes?

[untold plaza]

got a point

[buoyant canopy]

Maybe. Or they have a lot of expensive instances running on Azure and plan to take some of them down next week.

[shut yew]

man... even the next 6 months will be insane given what's developed the last 6.

[wind urchin]

And how quickly

[untold plaza]

Imagine when openai makes chatgpt a self learning bot lmao

self learning would be the most insane thing to see in the history of ai

[wind urchin]

singularity?

[untold plaza]

true,

I am convinced in the next year and a half self learning will be released

[iron crown]

interesting. do you have an exapmle?

[shut yew]

I just saw on youtube that there is a game incorporating non player characters that use AI to talk to you like real people. The youtuber was talking to them and they were talking right back... it was nuts.

[wind urchin]

Sounds expensive

But a natural progression

[jaunty rivet]

do you have the link? would like to watch

[sharp turtle]

mattprovid ai right lol

[buoyant canopy]

Be careful with links here. Better DM.

[wind urchin]

Eventually consumer GPUs will be powerful enough to handle games loading their own LLMs for NPCs

[jaunty rivet]

DM it to me if that works

or just what to search lol

[nocturne rose]

wow..interesting.

[wind urchin]

But maybe not gpt4

[shut yew]

"Please write a blender script of a rock sitting on a hill overlooking a valley. It is during sunset and it's during the fall. Please indicate when you are finished with the code."

[wind urchin]

Probably a successor to Stanford's alpaca based on gpt-4 will be the first kind of quality LLM you can run locally, but that's a matter for #ai-discussions

[shut yew]

Look up a channel called "MattVidPro" he released a video today about the game.

[whole sequoia]

with api key can i use chatgpt4 with image recognition?

[desert thunder]

no

[whole sequoia]

poggers

[shut yew]

I imagine they'll be the "pet" version of AI for individual users is how that will evolve.

[robust acorn]

How do you use pictures gpt4

[jaunty rivet]

nice, this is pretty cool

[whole sequoia]

what is llm

[whole sequoia]

cant

[shut yew]

Insane isn't it???

[wind urchin]

Agreed, and if you want a personalized AI assistant, it's best to have that local. I imagine some folk won't like the idea of personalized cloud run AI

[buoyant canopy]

Acronym of "Large Language Model"

[whole sequoia]

ty

[wind urchin]

I've seen it being used more and more recently

[shut yew]

A fine point actually! Hadn't thought of that, but you're very right.

[robust acorn]

😑They say you can

[wind urchin]

It's a useful acronym

[whole sequoia]

they lie

[jaunty rivet]

they havent released this to the public yet

[echo wasp]

#gpt-4-faq

You skipped it on your way here. Will answer your questions

[whole sequoia]

some dude made it work with google image classification but its not anything close to actual oe

[wind urchin]

I know that at least that portion of folk will drive ai development to optimized, consumer scales. Probably why Stanford did it in the first place, and did so openly

[whole sequoia]

ppl say faq the faq

[desert thunder]

just the advancement of GPUs will make training AI of chatGPT scale way more affordable

[wind urchin]

This is like being blindfolded and having a description of an image made by someone else given to you to elaborate on, but yeah

[buoyant canopy]

Not sure tho. Trends point in the direction of more and more stuff becoming a service.

[shut yew]

One way to have the chatGPT bot on public service see your images is to link the image location. NOT the page... the direct image link.

[desert thunder]

keep in mind imagenet cost hundreds to thousands to train (per hour) 5-10 years ago

now it costs way less, and you can train on it for free on colab even

[wind urchin]

Agreed! But I don't think that'll necessarily mean the development of open source models will stop. Corporations will for sure have the biggest and best models

[shut yew]

For now. Eventually everything will become standard. Then we'll all really pay when they release "fremium" services... lol

[desert thunder]

nah we will be able to run LLMs on consumer GPUS for sure. again, its always been like this with AI

[buoyant canopy]

ChatGPT can't access the internet. If you give it an URL with words in it, it will try to generate an answer matching those words thos. So for an astounding number of SEO URLs it will hallucinate an answer that is astonishingly relevant. But it can't access the content behind the URL.

[desert thunder]

first companies create the SOTA, then consumer gpus can catch up and run them or even train on them

[wind urchin]

Gpt-3.5 was like, 10 million dollars or something. The cost to train a similarly powered model has dropped to ~100 dollars because of what Stanford did

[desert thunder]

yep

[wind urchin]

It did the heavy lifting, now the big models can train the little models xd

[vernal kelp]

gpt-4 thinks that its gpt-3

[wind urchin]

Gpt-3.5 thinks it's gpt-3

[buoyant canopy]

Probably not. Sadly tho it appears the major players clam up and don't publish papers with their findings as freely now. Just look at GPT-4. There is no paper and the stuff you find on the "research" page omits all relevant information on training and model size.

[shut yew]

Well, that certainly explains some of the weird responses I got then. They should fix that so it doesn't answer if it doesn't know. For all the talk about ethics, that seems pretty simple, doesn't it? lol

[buoyant canopy]

Maybe. Shareholders wanna live too. 😉

[wind urchin]

You're right, but the Stanford model was trained on its outputs. No special training methods required

This might get regulated in the future

[tiny latch]

fake

what model of gpt are you

I am ChatGPT, a large language model trained by OpenAI, based on the GPT-4 architecture.

[buoyant canopy]

There are a lot of examples for this hallucinations. Yesterday I had a conversation with somebody who was convinced it could access an article on The Verge. It even "knew" the date the article was published. Guess what was part of the URL...

[wind urchin]

It reads the URL and makes up a summary

I guess that's telling it's ability to be convincing

[burnt edge]

25 every 3 hours??

[shut yew]

How then are these AI learning? I'm presuming its sourced information from the internet. Perhaps archived and screened via google.

[wind urchin]

More like better optimization would provide OpenAI more opportunity to let people test it, not just “GPU/CPU”

[buoyant canopy]

Yes. The point I wanted to make tho: The field probably will advance much slower now because learnings are kept secret from now on for the competitive advantage. So no more "Attention Is All You Need"-papers for us. 😦

[wind urchin]

OpenAI in particular makes their edge by keeping their datasets private, as in, not only are they internet sweeps, but they're human curated to almost nigh pristine levels, and they don't share that info

[wind urchin]

Apple reference

[vernal kelp]

i was using gpt-4 - i first asked shortly about its state of knowledge. 2nd promt i asked "more about gpt-4" and it continued to write how gpt-4 would be a fictional advancement of gpt-3. Edit: later on i told it that it is gpt-4 but he claimed otherwise

[desert thunder]

@buoyant canopy that should be regulated though, AI technology shouldn't be kept closed source, especially not as one as significant as chatGPT, this is unprecedented

[jade cape]

They nerfed it again and it doesn't even do images yet

super annoying

[wind urchin]

wdym by nerfing

It doesn't say as many bad things anymore

like what feature has been downgraded

[wind urchin]

that’s good, no?

[glass crag]

Does anyone know much about Microsoft 365 copilot?

[jade cape]

every 3 hours i mean. it was 50 every 4 hours before

[wind urchin]

Yeah it is

[wind urchin]

isn’t that better

[wind urchin]

It'll probably go down more

[glass crag]

Because I'm wondering if it's avaliable right now or if you have to pay for it or what

[wind urchin]

or is it 25 messages for 3 hours

[brazen knot]

No it's 25 every 3 hours now

[wind urchin]

wowww…

[buoyant canopy]

There is "pretraining" (the P in GPT) where a huge torrent of text is pressed into all those poor neurons. The result of pretraining is a coarse model. Such a model you take and feed it a set of cautiously curated content (much smaller in size than the pretraining data) and fine-tune it for a specific use case like chat communication.

[wind urchin]

It says 50 for me but I haven't relogged, could be cached

[echo wasp]

Jesus. They must be planning to role out a new subscription plan lol

[brazen knot]

And it still says that it will be reduced further next week

[wind urchin]

so that’s around 8.3333334 messages per an hour

[shut yew]

Even that has to be screened by some degree of automated filtering. The amount of data these AI are processing is utterly ridiculous. Curating it all has got to be far more than any army of humans can do.

[nocturne viper]

25 every 3 hours?!

[wind urchin]

not necessarily, the world is not a utopia. when you ignore bad things you make things stagnant and not realistic to reality. people have already complained about this for all sorts of creative endeavors. you cant tell a dangerous, thrilling, scary story for example when the results want to paint everything with rainbows.

[wind urchin]

OpenAI has some skill issue in optimization 😭

[nocturne viper]

Time to riot

[desert thunder]

"ai is so powerful that it needs to be something that everyone can use, control, or contribute to, or else AI will be controlled by corporations who just want to get ahead ... we dont want corporations deciding whats ethical and acceptable for AI"

[brazen knot]

"GPT-4 currently has a cap of 25 messages every 3 hours. Expect lower cap next week, as we adjust for demand"

[nocturne viper]

I am outraged

[vernal kelp]

rip it just got halfed

[wind urchin]

True, there are bad things that’s considered good

[buoyant canopy]

I'm from Germany, so when I hear "regulated" it is like when somebody from Texas is in a supermarket and reads "It is known in the State of California that...".

[wind urchin]

Garbage in garbage out is really important. The Stanford model worked because gpt was able to make a lot of that manual dataset labor automated and quality, with humans picking the best of it's results

[upper frigate]

So not only do they limit your messages per hour, they also don’t have the image input they advertised 💀

[desert thunder]

do you not have regulations in germany?

[wind urchin]

OpenAI is openly asking the government for regulation

[jade cape]

We pay $20 a month. They can at least give me a reasonable number of prompts per hour and release the image function

[wind urchin]

So not only that, they also make you pay 20$ a month just for a downgraded version of GPT-3.5

[shut yew]

Yeah, I think they're realizing that now with it released to the public that without any guardrails it's still a bit dangerous.

[wind urchin]

“real”

[upper frigate]

You didn’t pay $20 specifically for gpt4 though, it’s a subscription and gpt4 is just an extra feature

[nocturne viper]

Regulated in Germany has different meaning, it means your beer tastes good and isn’t made with rice. It means somehow you can have cheese made without plastic that doesn’t cost an hours wage. I love German style regulation.

[wind urchin]

True I have it for speed only

[desert thunder]

@wind urchin yeah, the kind of regulation they want is to restrict access to gpus for consumers. im not even kidding, they mentioned that in an ethics review paper they did

[glass crag]

Fortunately I'm being patient and waiting until the I get access to the API :)

[wind urchin]

Without speed it’s exaggerating

[buoyant canopy]

We have a whole lot and also EU regulations on top. Currently they try to create an act to "regulate" AI. Total s**tshow.

[upper frigate]

Even when you get access it’s still $0.06 every 1000 tokens I believe

[desert thunder]

damn.... gonna be a tough road ahead for researchers

[nocturne viper]

Ive been predicting this for years. Computational power will slowly move to a centralized cloud pay per cycle style and GPUs will be outright banned.

[granite burrow]

almost positive people bought the 20 buck sub for chat4 but now 25 for 3 hours? that is crazy

[warped schooner]

what?

[wind urchin]

Check out the technical report, they admit it's dangerous. A portion of the dev team didn't approve of the early release and their names are listed for like. Legal exoneration, i guess? It's a profound document in multiple ways

That's one thing lol

[brazen knot]

Imo opinion there should be 3 tiers

Free, for normal usage

Plus, for unlimited prompting on stable and fast models

And PlusExtra, for unlimited access to any new features

[jade cape]

I agree with AI regulation but I don't trust a bunch of 80 year olds who live like it's the 90s making decisions that will either do nothing or ruin it completely

[glass crag]

Oh and also most the stuff that I would use for GPT-4 is getting released for free through Khan Academy anyway so :)

[wind urchin]

indeed. also your datamining laws are quite good, all your information isnt being harvested to the extent it is in other countries anyway.

[buoyant canopy]

Weren't they talking about the consequences of AI changing society? So "regulation" probably means "dear politicians, please come up with a way for the taxpayer to pay for the conseqences of us letting every company have access to a very capable AI model".

[warped schooner]

They call it transcendence

[wind urchin]

I wouldn't put it past them

[wind urchin]

I think unlimited access to any feature is just “too much” for the server to handle, as you see with its recent optimization which is basically almost the same as forspoken’s optimization.

[lime valve]

[buoyant canopy]

Yes. But it also means you can't have AI because it endangers the business model of some buddy of some politician in Bavaria.

[glad tartan]

dang... This sucks

[glass crag]

Question: has this channel just been dedicated to complaining about OpenAI for the past few days

[warped schooner]

It will not be stopped

[wind urchin]

I love cupcake

[brazen knot]

What do all companies do with extremely high demand products? Raise prices, itll keep usage down while those who really want to use it can use it

[glad tartan]

I get it tho. I can't image the compute resources running gpt 4 takes

[wind urchin]

More or less. Actual gpt-4 outputs are being shared in other channels lol

[nocturne viper]

I just dropped in half way to this conversation. Did they pass a new law? I spend half the year in Hamburg do I need to wipe my stable diffusion 😅

[warped schooner]

we will be absorbed

[wind urchin]

You good dog?

[buoyant canopy]

It is in the works. Everybody waiting for EU to come up with a draft is my last info.

[limpid scarab]

25 messages for 3 hours 😦

[wind urchin]

However, I don’t disagree with that but, Have you seen how fast is GPT-3.5 compared to GPT-4? That’s because GPT-3.5 is properly optimized(not to an extent though) that’s the opposite for GPT-4, it’s not optimized enough to run millions of users prompts every minute, it exhausts the server and eventually kills it probably.

At least the hourage went down, more frequency

[quick anchor]

It is very annoying, it is always stop generating content in the middle. Does anyone experiance that?

[wind urchin]

They were probably pushed into releasing it early by Microsoft

[nocturne viper]

I have a love hate relationship with German laws. I trust them more than openly bought and paid for US law makers though. I trust them to find a good compromise.

[wind urchin]

They might need to re-consider into turning GPT-4 private for now, and when it comes out it should be more optimized and user-friendly

[native hinge]

is the 32k chat length feature available for GPT plus users right now or is that rolling out later? (i'm trying a 15,000 word paste and getting length errors)

[brazen knot]

Yeah, it's why I wish they would have waited to release or delayed the release for when people can actually use the thing within a reasonable workload

Like who is going to be able to complete a complex project that GPT-4 is capable of helping with, with such a strict limit

[timber anvil]

Hey, anyone can help me with GPT 4 API? I have access

[peak hamlet]

how?

[nocturne viper]

Probably “you can use AI but you need to fill out these ten forms every 15 days at the office that is open only Wednesday from 10-11 (if the one employee isn’t on holiday)

[timber anvil]

I want to make bot on discord

[peak hamlet]

I know how to do it

[wind urchin]

I agree, but I think OpenAI thought that Microsoft servers were godly level that they decided to push it further and release it to plus for testing

[timber anvil]

pm me

[nocturne viper]

There’s a premade discord bot for gpt on GitHub, google it plug in your settings and ride free

[wind urchin]

Then it went horribly wrong due to the optimization

that’s my guess

[lime valve]

Hahaaaa you must have traveled back from the future Germany

[nocturne viper]

I’ve been trying to apply for visas over the course of the pandemic. I love Germany but have a very sour taste of their love of hyper-bureaucracy.

[timber anvil]

pm me pls

[lime valve]

And dont forget to buy an AI insurance😀

[wind urchin]

I'm just hoping it will replace human labor or something we didn't think actually possible

Better to be wildly optimistic than sad

[nocturne viper]

https://github.com/openai/gpt-discord-bot

[wind urchin]

Python right

[nocturne viper]

I’ve already made a ton, you can do it, ask chat for help if you need it.

[wind urchin]

discord.py?!

You should probably know how to make a discord bot before you make a gpt discord bot,

[tardy heart]

i think its just they don't have capacity yet

[nocturne viper]

Sorry I’m a python only kind of guy I forget other people use anything else. Learn python, your life will be better.

[quick wing]

So i literally joined just to see the outrage about the new cap for 25 messages every 3h

[wind urchin]

50 every 3h for me :)

[tardy heart]

this channel is fairly insufferable, see some good things occasionally though

[quick wing]

Refresh your page, should update, pretty sure its a global change

[wind urchin]

... :(

[nocturne viper]

We’re all planning to run Naruto style at the openAi headquarters. They can’t stop all of us!

[quick wing]

Count me in

[echo wasp]

If you go to their 'learn more' option on the expiration warning when you reach the limit, it takes you to a form saying they're looking for ways to improve the experience by considering higher payment tiers or other options, then gives you the option to fill out the form to be kept up to date

[quick wing]

So i guess paid costumers are actually being screwed over cause more API keys are being given away

Which seems... a little counter productive

But what do I know

[worn talon]

yeah bruh 'GPT-4 currently has a cap of 25 messages every 3 hours. Expect lower cap next week, as we adjust for demand.' Im done 💀

next its gonna be 1 message every 4 hours

[quick wing]

The adjusting to lower cap even further is gonna lose subscribers quickly, especially with the 14 day payback guarantee

[tired umbra]

It is a shame to pay for these conditions

[quick wing]

Many people literally got it to just use GPT 4, and it has not been 2 weeks since it came out, this means that the refund period is still there

[wind urchin]

What I’m going to say:
They must disclose GPT-4 performance issue(Optimization) or try to delay GPT-4 till they actually made sure that it’s pretty good for releasing.