gpt-models

[reef sleet]

Can we subscription GPT Pro and pay with paypal? The problem is it's very complicated and less practical when using the card payment method. The payment method for GPT Pro is more practical by simply using PayPal.

[arctic rampart]

It was gone yesterday for me. I cannot access it as well. I have been subscribe it for a few month now.

[quick dune]

I don't know, it seems like everyone find the card payment better💀

[vagrant hornet]

hey people, i'm curious, what tasks did you find GPT4 proforming better than GPT3.5 in?

[still lagoon]

What, if GPT-4 isn't a giant monolithic dense model but a mixture of 8 x 220-billion-parameter models?
How would you think about the strategies OpenAI has taken because of that?
Does it hurt the scientific community?
I'm interested in your opinion about GPT-4.

[vagrant hornet]

8 models... with the same archetecture? how would that work

[still lagoon]

Some kind of a hybrid model. You know, transformer models have some kind of head attentions. Maybe the models working together in parallel. This would explain the bigger context window.

Usually the context is 1024 tokens, but to scale a model to 32k needs retraining. This is not possible in this giant context.

[still lagoon]

You can also use LoRa's for specified tasks (math, function calling) or some kind of ControlNet architecture which interleaves the layers.

[wraith ginkgo]

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

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

[fickle gale]

Oh man...

I've bought ChatGPT Plus with my credit card here in India. It works.

I tried to do the same for my brother's account. But guess what? The same card is declined. God knows why.

I don't own mutiple credit cards and see no reason too...

If only they supported UPI, India's payment system.

This "Card declined" error never occurs in UPI. If your account has balance, it will ALWAYS work.

Americans are still in the 20th century when it comes to fintech adoption. If only they had experience with a better payment system, like UPI & whined more for it.... I wouldn't have wasted an hour trying the card...

Credit card is literally 1960s tech and still being used 😢

[supple violet]

when did you pay plus with yours last time? there seems to be new issue with all payments from india blocked

[fickle gale]

I first put my credit card 3 months ago. Last payment occured on Jun 13.

[unkempt saddle]

yeah men

[supple violet]

its possible that it might fail next month for you too, i saw some say its rbi mandate and some say some other regulation

[fickle gale]

sad. but RBI mandate requires certain intimation processes to follow so that user is always aware of the upcoming monthly charge. It doesn't block subscriptions. Stripe (ChatGPT's payment processor) should add support to this as this is what they do, payment aggregation.

[supple violet]

🤷‍♂️

[fickle gale]

I would have no way to buy chatGPT Plus since no other payment method is supported.

😢

[supple violet]

do you have ios devices?

[fickle gale]

Nope

[verbal pewter]

So I have a questions, in my German lessons we are talking about a Book ChatGPT wasn't trained with, could I copy paste the Chapters from a PDF into GPT so that it could then give me good enough answers to use for my essays and class?

[supple violet]

yes

[verbal pewter]

very well

[bronze radish]

What can you do gpt-4

[wind urchin]

Does anyone know when they will remove the 25 message limit?

[knotty grotto]

No eta

[supple violet]

not sure if it will ever go away since during gpt4 announcement they talked about the possibility of higher tier sub for more usage

[pallid siren]

Is it just me or are you guys also not seeing the option to switch prompt between GPT 3.5 and GPT 4

[harsh sigil]

u got no ChatGPT plus man

[pallid siren]

I have ChatGPT Plus, been using it for a while

[harsh sigil]

just reload the page

openai has a bunch of bugs

[wraith ginkgo]

This is a Safe for Work Discord Server

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

Posting NSFW content will result in a permanent ban.

[pallid siren]

Like when I click on Newchat it just opens the main prompt page and its set to GPT 3.5 default

[pure estuary]

Hello everyone, Chat GPT4 is still not restored?

[pallid siren]

Yeah

[pure estuary]

yeah what ? still not restored ? do you know what this is due to? and how long will it last?

[burnt void]

@pure estuary I had that happen as well. Go to your your name in the bottom left corner, click on my plan, and then click on manage my subscription,.
Afterwards return to https://chat.openai.com/ and you should be good to go.

[pure estuary]

yeah it's fine ! TY very much !!!

[wooden cedar]

thanks @burnt void So many people have been facing this issue.

[glossy prism]

Hello everyone.
I am trying to build a accountant chatbot to chat with real financial data.
I have Balance Sheet, General Ledger, P&L documents.
that i want to chat with and plot data , return results after running formulas etc.

Any thoughts ?

Currently i am planning to use Langchain. can create a custom data parser that will get the data from documents. Not sure about the querying, plotting and running formulas part.

[wraith ginkgo]

`` Server Tip: ``

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

[stoic ember]

Im working on pretty much the same type of project with langchain and openAI api.
The difference is that I'm not feeding it with excel docs, but with PDF's containing data about some company's HH.RR related stuff
You could take a look at Pinecone for data saving and vectorizing.
OpenAI api could do the reasoning for the formulas but im not sure about the plotting though.
Maybe u could integrate your code with Tableau or similar

[lone prairie]

Hii

[viscid lodge]

wyd my hello message get removed/

[wind urchin]

Thanks for ur lequidity guys 💲😮‍💨

[wind urchin]

You're a lifesaver man! Needed chatgpt4 for some work stuff lol

[quiet siren]

Thank you!!!

[hearty ridge]

0.0

What does this channel do

[knotty grotto]

You discuss about chatgpt 4

[torn shale]

Hey! Is there a method by which I can create workflow diagram in Chat GPT 4? I want to use it for creating workflows for Lead Generation.

I've just paid for my subscription and I've been upgraded to ChatGPT Plus. What does Plus mean? ChatGPT 3.5 or 4?

[merry kettle]

yes "show me diagrams" plugin may be able to do that

(hello everyone, first entry here)

[hallow owl]

Plus gives you both GPT-3.5 and GPT-4

[torn shale]

Welcome @merry kettle and thanks for your advice! much appreciate it!

[torn shale]

Thanks Robert for the clarification!

[stoic bay]

hall

[fallow knoll]

hail

[storm reef]

Hi there,
Is gpt4 down ? the selector at the top is gone and all conversations are defaulted to 3.5 (yes I have a valid active subscription)

[hallow owl]

Click on your account name in the bottom left, then select My Plan, then Manage Subscription, then go back and then things should work for you

[sick pelican]

lifesaver! Thank you!!

[storm reef]

It actually worked ! that's kind of a trick, thanks !

[wind urchin]

"continue generating" button is broken for me, everytime I click it start writing bogus stuff unrelated with the previous message. Only happening to me?

This is happening to me from a couple of days

[slender steppe]

supra prize in india

canu teel prize supra prize in india

[inner rapids]

wait a min, do you only get 25 questions every three hours, no matter how many subscriptions you have?

[supple violet]

you are not really supposed to have multiple accounts to evade limits, rate limits are handled also above account level

[inner rapids]

ahhh, would've been better to know that before I got another subscription lol

Thanks for letting me know!

[noble dune]

Hey I just bought GPT-4 but I think its web plugin is inaccurate. Do anyone know how much GPT-4 is better than its previous version in terms of Coding and programming stuff?

[viral grotto]

was chatgpt 4 size response incressed? i been geting short responses past few days and now they finally long

[trail drift]

How do I get code interpreter?🥲

Is there any option to request for that?

[supple violet]

no

[trail drift]

Or any option to apply for waitlist?

[viral grotto]

when i last used it at coding i was able to create fun stuff its better than gpt 3.5 at coding

[noble dune]

Can you share a example @viral grotto

[trail drift]

Umm. So no way to apply for alpha tester??? 😬🥲

[supple violet]

no

[trail drift]

Okay🥲.

Got any idea on code interpreter beta release?

[supple violet]

nope

[trail drift]

Okay. Soo many nopes 😂

[south smelt]

sha ya

[wind urchin]

gpt 5 when

[round shore]

Can Chat GPT use in Vietnam?

[supple violet]

no

[wraith ginkgo]

`` Server Tip: ``

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

[sand yew]

Is Gpt4 down?

[supple violet]

no

[fair iron]

[pulsar lintel]

Hello Guys, How can I automate GPT 4 to apply for multiple jobs at one time

[wraith ginkgo]

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

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

[velvet blaze]

It is for me. The option to select GPT4 isn't showing up for me.

[supple violet]

#gpt-models message

[analog mortar]

Someone might have a better idea of plugins, but your discussion is pretty complex. I haven't found any plugins that would do this for you, but it is possible to create an unverified plugin that might do that for you, but I am having trouble creating a plugin to connect ChatGPT and JobAdder, but it sounds like you would have to create a plugin that would communicate with multiple services to apply for a lot of jobs, which sounds like hell to me right now lol

[velvet blaze]

Thanks

[pulsar lintel]

Thank you

[analog mortar]

are you planning on using indeed and linkedin for the the applications?

[dusky iron]

Gib Code Interpreter, ty in advance xD

[wraith ginkgo]

This is a Safe for Work Discord Server

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

Posting NSFW content will result in a permanent ban.

[vast token]

Hello! I'm a Plus subscriber, and the toggle that allows me tow switch between models seems to have disappeared. I'm probably missing something obvious. Anybody know what I need to do?

[supple violet]

#gpt-models message

[vast token]

you're a hero. thank you!

[pulsar lintel]

@analog mortar Yes for both Linkedin and Indeed

[wind urchin]

Whats new in gpt 4?

[analog mortar]

Oh, okay, I see, for this, I believe you'll need Linkedin Developer as well as an Indeed developer account, you might have to join a waitlist for both, but if you look at their API documentation, it will help come up with a path with ChatGPT and whatnot, but the reason you need the accounts is so that you can get an API key as well some basic parameters for the unverified plugin. If you have coding experience, this should not take you too long, but for someone that doesn't, it might be better to just apply for the jobs! 😂

[analog mortar]

like as of recently or just like what's new about gpt 4? lol

[analog mortar]

I'm real sorry if that's not what you wanted to hear, but you might want to checkout the api-projects/api-support someone there will be a lot more helpful, you should know I have no coding experience whatsoever, but I am also trying to create an API which seems impossible

[heady moon]

Is every plugin legit in the store?

[knotty grotto]

Some are not certified yet

[pulsar lintel]

I appreciate your insight and response to my inquiry and thanks for pointing me in the right direction

[heady moon]

ty

[knotty grotto]

?

Ask what ?

Yeah just ask

If you want to speak with chatgpt you need to go to chat.openai.com

1. what was the question ?
2. what it has to do with chatgpt 4 discussion channel ?

Maybe in #off-topic would be more appropriate if that's not related to chatgpt 4

[wind urchin]

can anyone pls help in #gpt-3-discussions

[knotty grotto]

Or #ai-discussions if that ai

[knotty grotto]

Relevant Links
💬 Try out ChatGPT at https://chat.openai.com/
🤖 Access GPT-3 at https://platform.openai.com/
🎨 Make art with DALL∙E 2 via the API or in https://labs.openai.com/
📩 Read FAQs and reach customer support through https://help.openai.com/

[wind urchin]

is gpt3 frree

[knotty grotto]

Chatgpt 3.5 is free

[vagrant flax]

who got api keys

[wind urchin]

is that the one for https://platform.openai.com/

[knotty grotto]

Chat.openai.com

[wind urchin]

and does gpt 4 or 3 have a knowledge cutoff in 2021 like chat gpt 1

[wind urchin]

but thats chat gpt 1

[knotty grotto]

Several people have it

[wind urchin]

not 3 or 4

[knotty grotto]

Chat.openai.com use chatgpt 3.5

And if you pay you access to 4

[wind urchin]

so u mean the only free gpt versions all have the 2021 knowledge cuttoff ?

[knotty grotto]

Yes even the paid one end in 2021

[wind urchin]

😡whyyyy

whywhywhywhywhywhywhywhy

[wind urchin]

why

😭 ** **||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||

[supple violet]

takes several months and millions of dollars to do data update

[wind urchin]

why does gpt have a knowlege cutoff😡😭💀

[wind urchin]

BRUH

[stone vault]

how is Dall-e comparable to midjourney?

between the MJ versions, is the cheapest version still good to make great art, or you have to subscribe for pro/plus plan to get some of this amazing art?

[supple violet]

same quality at all tiers

[stone vault]

Thanks, I thought so, just wanted to make sure before paying for better plan what I can achieve with it. They do not have open free plan or anything like that? Its paywalled only, correct?

[supple violet]

trials were closed months ago

[stone vault]

Thats what I thought, thanks!

[supple violet]

imo the cheapest tier is good for a test but you will run out of "quota" quickly

oh this topic btw belongs to #ai-discussions

[stone vault]

Thanks, I figured I was in the wrong chat

[wraith ginkgo]

`` Server Tip: ``

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

[atomic token]

Hey guys

[desert tendon]

hello, i was checking the detail of my invoice and stumbled upon this, how is it possible as this model is only supposed to be an 8k context window?

oh i cant add a screenshot

ok so here it is:
11:15
Local time: 27 juin 2023, 13:15
gpt-4-0314, 11 requests
31,814 prompt + 1,882 completion = 33,696 tokens

[jovial coral]

Hi chaps, anyone been lucky enough to get GPT-4 API access yet? Been on the wait list since 5 mins after the announcement and still they won't let me in 🙂

[desert tendon]

yes a long time ago, maybe you should follow up

[jovial coral]

How do you follow up? Re-register on the waitlist?

[desert tendon]

my take is to retry with another email but with a better use case

[jovial coral]

Ah, great idea. Do you remember what you used for your use case by any chance (if it's not confidential)

[desert tendon]

i think they look at who is using it and why

i am building a chatbot platform, so i didnt have to brainstorm a lot 😅

or maybe just tell them the truth

[jovial coral]

Right, I'll sleep on it and try to come up with a much more exciting use case. Thanks for the tip 🙂

[desert tendon]

maybe no feedback is a feedback

you are welcome

[rancid vale]

Working on creative writing projects. Any idea on how to make GPT not end a section with some kind of positive declaration for the future? It seems inevitable that every time I give it a prompt, it ends with some variation on, "Together they knew that whatever challenges came their way, they would overcome it with the power of teamwork and courage and harmony"

You can't ever seem to have setbacks or adversity because the characters are absolutely unswerving in their faith and conviction and heroism no matter what happens

[somber nova]

it happened to me on mobile earlier

[honest sleet]

Is there any way to get access to the Code Interpreter Alpha plugin as a premium member?

[fair iron]

There is not currently. Code interpreter is in closed alpha to my knowledge

[honest sleet]

Ahh thank you. Any idea when the Beta will be released?

[fair iron]

I do not know

[honest sleet]

No worries. Thanks

[austere relic]

Are there any free plugins that are capable of filling out PDF forms

[amber lynx]

Anyone else feel like GPT4 has gone backwards in its ability to write SQL?

[atomic token]

What is a Tokens?

[jade pewter]

sgfrge

[final scaffold]

id:guide

[proper halo]

/robot

[wraith ginkgo]

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

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

This is a Safe for Work Discord Server

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

Posting NSFW content will result in a permanent ban.

[timid cairn]

Could someone assist me for a moment. I would like to test my prompt out.

To help me create a well-structured and effective prompt, could you please provide me with more specifics about the problem, purpose, desired output, and any other relevant details related to your task? The more detailed your request, the better I can assist you.

If someone could provide said information, Ill create a prompt for you.
anyones...?

[wraith ginkgo]

`` Server Tip: ``

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

[uneven flower]

does anyone know if gpt-engineer can run on gtp 3.5?

[frozen veldt]

not well enough to be useful

[uneven flower]

ah ok, that's what I've been reading, but haven't been able to get it to run yet on my machine. So you've tried?

[frozen veldt]

Yep tried but not so great without gpt4, you just have to change the specific model calls that's easy enough but the difference in logic is painfully obvious

[uneven flower]

good to know, thanks!

[frozen veldt]

nothing I've seen so far is as good as code interpreter, and that isn't as good as a good coder. It sill uses legacy calls to its own API which needs updating amongst other issues, but if you provide those parts of the documentation it is able to use them

[uneven flower]

I'd love to mess around with code interpreter but I'm still on the waitlist

[chrome mantle]

Has anyone been using gpt4 specifically on a project. I find it very useful for small queries or small coding projects but no idea on its effectiveness on a longer bigger project

[charred spear]

It's quite capable of writing AI models in the right hands.

[hearty needle]

Are there ways to switch between models in a conversation?

I had a chat with the GPT4 model but at the moment it was disabled and it changed it automatically to GPT3.

[frozen veldt]

This discord really needs pinned messages

the number of times the same questions come up per day is.. extrodinary

[hearty needle]

thatas a nop?

[frozen veldt]

you hit message cap and continued so it reverted to 3.5

[flint bear]

hi has anyone tried MediaX Chrome extension?? that works on ChatGPT, Goolge Bard and Microsoft bing chat

[lapis bramble]

Does gpt 4 still have capped messages ?

[fair iron]

Yes, 25 messages every 3 hours

[lapis bramble]

No way to get unlimited ?

[vagrant flax]

bruh

[serene cipher]

the API is unlimited

[serene cipher]

there are a few ones similar to the OpenAI product, ask around on #ai-discussions they might have good suggestions there

brother, on this discord we talk only about stuff from OpenAI!

that channel is the exception

[lapis bramble]

I think I’m still on waiting list for that

[serene cipher]

yea, that might take some time indeed

[zealous linden]

hallo

[nova cargo]

I have issues how to enaable this on the web browsing capability? Any help?

[static isle]

Go into settings and enable it through Beta Features

[wild minnow]

Hi, I love GPT-4, i am paying for chatGPT+

[vernal musk]

hi guys, is the gpt-4 multimodal api available?

[plucky peak]

I have been using the Memory Bank plugin does any one know where I can find who made it or where it stores the data etc? Thanks 🙂

[wraith ginkgo]

`` Server Tip: ``

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

[gleaming heart]

Dear OpenAI team,

I hope this message finds you well. I am writing to address some concerns regarding the performance of the web version of ChatGPT. I have experienced a significant decline in its performance, specifically with regards to losing key information in prompts of less than 200 characters. Additionally, when I requested assistance with processing data in a specific format, ChatGPT simply reiterated the format I had specified, leaving me to handle the processing myself. This has led to certain situations where ChatGPT 3.5 outperformed the web version.

As a result of these issues, I have decided to terminate my two-month subscription to ChatGPT Plus and apply for access to the GPT-4.0 API instead. However, it has been a week, and my application has not yet been approved. Therefore, I would like to inquire about the estimated timeframe for when I can expect my turn in the waitlist.

Thank you for your attention to this matter, and I appreciate your prompt response.

[hallow owl]

it took 24 days for me to get off the API wait list. Some people got it before me, others are still waiting. There is no expected time frame to be selected and no guarantee you will get access before it is released.

[gleaming heart]

Thank you for your response.

[ripe swift]

....

[heavy lotus]

How are GPT 4, GPT 3.5, and GPT 3 different from the original GPT 1? Are they just simply larger and larger transformers?

[radiant prawn]

the 25 message limit coupled with the fact that chatgpt web works ~60% of the time as expected makes chatGPT not worth the price.
Better stick to bing and GitHub Copilot chat until they change the limit.

[lean magnet]

When will GPT4.5 be released?

[quick dune]

Yes

Larger and more capable

That is

[heavy lotus]

Thanks!

[lofty saddle]

I wish there was a way to clear the cache on GPT-4's memory/knowledge graph. It's incredibly annoying that some things persist across conversations.

[grizzled ravine]

Hi all, I am looking for ideas to craft a prompt for flash cards creation.

Starting from a piece of knowledge, I’d like chatgpt to create multiple versions of question/answer flash cards. I’ve already tried this, but chatgpt it’s not really creative with the questioning.

For example it doesn’t create trick questions, or it doesn’t deconstruct the piece of knowledge in smaller more atomic question/answers. Which are easier to learn and recall.

Anyone experimented with this?

[wind urchin]

What is that

[wind urchin]

If anyone is into AI development on a beginner scale hmu I got a project I need help on

[lofty saddle]

If you want it to do something clever, you need to explain the rules of "clever".
As we now know, gpt models' logical abilities are (almost definitely an) emergent behavior from being trained on coding data (and probably coding related instructions) which means expanding it's logic happens the easiest via coding terminology.
So you should check out stackoverflow's codegolfing for the (endless) puzzles and how the parameters for the problems are defined there.
Then just channel your inner "w-w-well akschually" nerd, similar to what they're doing at codegolfing and turn your flashcards into a codegolf.

[glad thunder]

hey all new here, when i logged in the other day there were 100% 3 options, one of which generated ai art. Its gone and theres no trace of this?

[wraith ginkgo]

This is a Safe for Work Discord Server

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

Posting NSFW content will result in a permanent ban.

[lapis bramble]

How can I know if I have access to api?

[supple violet]

everyone with an openai account has

[glad thunder]

anyone? its just like it never exsisted, am i going insane because i showed my work colleague im sure it exsisted?

[supple violet]

if you have access to dall-e plugin for a while it probably was a bug

[glacial dragon]

what are the benefits of hazelnuts on mental health and aromatherapy

[hallow owl]

This is not ChatGPT, you can access ChatGPT here - https://chat.openai.com/chat

[lapis bramble]

I’m slightly confused , I just want to use gpt4 in an unlimited way , so do I purchase the subscription first and then link an api ? I have an api connected to my account already , what do I do exactly ?

[supple violet]

api is billed monthly based on your usage, there is no fixed fee. api is mainly intended for developers but you can use it with playground too https://platform.openai.com/playground

[glad thunder]

ahhh, i see, i access it via the openai link and the option is there

[lunar glen]

Is the plugins store offline?

[lapis bramble]

Oh right , so playground uses gpt 4 and I interact with it like I am with chat gpt? Can it web browse and can I add plugins etc ?

[supple violet]

no those are chatgpt plus features

[lapis bramble]

Ok so basically really no way to use gpt 4 with plugins, and unlimited at the moment

[wraith ginkgo]

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

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

[ornate sinew]

who is the president of the united states

who is the president of the united states

[marsh compass]

Me

[marsh compass]

Me

[wind urchin]

does anyone else's GPT-4 tells you that you are using GPT-3?

[marsh compass]

I want gpt 4 for free

I don't have it sadly

[lunar glen]

Is anyone able to access the Plugin Store?

[dark glade]

Heeey everyone. Any suggestions on finding GPT propmts?

[pallid hawk]

@marsh compass Ask Santa

[marsh compass]

Nahh

[marsh compass]

Santa doesn't exist lol
I'll ask ChatGPT or.. dare I say.. Google Bard 😅

[pallid hawk]

Lol Google Bard should be the one that doesn’t exist

[marsh compass]

Lol, but Santa too

[knotty grotto]

#1019652163640762428

[dark glade]

you're the best 🧡

[wraith ginkgo]

`` Server Tip: ``

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

[soft thicket]

Apologies if this has been covered, had a scroll back and couldn't see anyone with the same issue. Has anyone else noticed/experienced a complete loss of all contextual training and learning from chat instances in GPT? As of yesterday afternoon, all my carefully trained chat instances had completely forgotten all the months of training and fine-tuning work I had done and were suddenly all completely clueless about our chat history. Hugely frustrating and inexplicable as I can't seem to find any reason why this would have happened. Perhaps I reached a memory limit which resulted in hard reset? Many thanks in advance.

[tribal egret]

Gpt please save me in your ram 😭😭😭

[verbal mortar]

4.0 only really remembers up to about 60k words. 3.5 is much less

[glad thunder]

I have had this also! only my instance was a dragon ball themed story, it suddenly forgot it was in story mode

[verbal mortar]

Yeah only 60k words for 4.0 and even then I find it hallucinates on me a lot after about 10k words, sometimes even less.

[hallow owl]

no ads

[soft thicket]

Crazy. I put a huge amount of work in to get those instances running well. Need to find a tool that remembers more than 60k words I guess...

[soft thicket]

Sorry to hear that!

[rocky pewter]

Regarding GPT-4 getting dumber: I've started to suspect that in societal questions, the filters of GPT-4, which force it to be neutral, might actually reduce the factual content of its responses.

A neural network generates text based on all the input it's given, minimizing the network's error. If the requirement for neutrality imposed by the filter increases the error of the generated text, the reward functions of the AI are in direct contradiction with each other, resulting in responses heading towards gibberish.

The correct approach would be, of course, to get rid of the requirement for neutrality and to coolly accept the outcry that the perceived bias in reality, as modelled by the networks, compels it to voice.

[ember nacelle]

Today I keep getting silly annoying image captcha after every command. Any idea how to get rid of it?

[knotty grotto]

Delete cookie and cache

[hallow owl]

remove any unofficial ChatGPT extensions from your browser

[ember nacelle]

I don't have any.

[ember nacelle]

Id doesn't help. 😦

[knotty grotto]

Can you try firefox as test

[ember nacelle]

Right, it works, it will be something related to messed up cookies/cache. Thank You very much. Now I wonder how it happened on the day of renewed subscription. 😉

[robust ice]

is there a way to get gpt-4 model back from gpt 3.5 on chrome yet? previous method of changing the model name with fetch now returns 403

[nova cargo]

got it, thanks!

[zinc lance]

I have been using GPT-3.5 to generate or improve C# programming code. It worked relatively well, and had been pretty good at following my instructions, particularly to return the entire code, without omitting ,or skipping or shortening. However, GPT-4 has a horrible tendency to completely ignore these requests. It omits, skips, shortens, and even changes the existing code, even if I specifically order it not to. Sometimes it feels as if it looks at the code and thinks; "Oooohhh... this is C#... I'm going to have some fun and do whatever I want with it !!! ". FYI, I use the same exact prompts for both 3.5 and 4. Sometimes, it seems that GPT-4 returns more accurate code, but instead of using the previous code I submitted, it simply creates a "Simple Template" and tells me to add all the relevant code in the new sections it created. I end up having to either regenerate multiple times, or issue additional prompts for it to include what it missed, or re-word and re-issue the prompt multiple times, until I get the right results. But, because of all the resubmissions and/or regenerations, I hit the limit and get shut down for a few hours. This is counterproductive for me and also for GPT-4. It is trying to be extremely sparing with processor time, but instead ends up costing both of us 10x the time and effort.

[sonic wagon]

Any idea how to get the correct length on the output? I'm trying to generate 1000 words long stories, but it always comes back between 400 and 600. The total token count doesn't even reach 1000, so far below the limit.

[stuck badge]

Have you tried including your character request in the prompt

[sonic wagon]

Yes, the prompt includes, among other things, the length the output should be. I've tried both word and character count.

[stuck badge]

Dm me if you want a prompt that can do longer form....

Ask it to provide output in sections...?

[sonic wagon]

You mean to break the output in different sections? I've tried as a key in a JSON object and also as a plain text response.

[stuck badge]

Hmm... you using playground?

[sonic wagon]

Node + API

[flint reef]

Is there any ETA when gpt 4.0 api will be public released? I really need access to it but still heard nothing from the whitlisting

[knotty grotto]

No eta

[verbal zodiac]

You're not alone, i'm unable to reproduce what i was doing some weeks ago. Like you said, loss of context, missing elements, passing the task to you, shorter answer, describe and enumerate the tasks like a parrot etc. More frustration than work done. I'm waiting a little bit but it will be subscription cancellation for me, not worth it like this.

[somber oracle]

how can I get code interpretor plugin on the beta registartion it says plugin is avilable to all PLUS users I paid so much money in India to get a PLUS Account

[knotty grotto]

It is in alpha and not yet available to public.

[somber oracle]

I Think its in alpha beacuse i'm in beta

[knotty grotto]

Yes sorry didn't chose the right word. You are correct

[somber oracle]

I hate this i'm getting my 20$ back

[knotty grotto]

You can ask refunds* if you don't like it and you can cancel it

• not eligible everywhere

[acoustic vapor]

I want to use GPT four with web browsing to gather documentation and read the documentation about obsidian note-taking. I also wish for it to output examples. How would you go about this?

[hybrid mirage]

.

[wind urchin]

gd

[wheat frigate]

after spamming finally my chat gpt got renewed 🥹

i guess there is no issue with rbi there must be issues in tax calculation i guess

[midnight coral]

How to open a separate GPT server

[hallow owl]

what do you mean?

[wraith ginkgo]

`` Server Tip: ``

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

[wraith ginkgo]

This is a Safe for Work Discord Server

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

Posting NSFW content will result in a permanent ban.

[mint slate]

Anyone know if the 25 messages every 3 hours will be increased anytim soon?

[coarse gyro]

No nobody knows.

[mint slate]

dem'

thanks

[static isle]

Not possible

[cedar hazel]

i hope so ...but is it 3 hours ...i had it one time and iam sure after 2 hours it was back. Yes it says 3hours but iam relativ sure it was only 2 hours

[glad thunder]

you could have had some fun there 😄

[lunar glen]

How can I access the plugins store?

[knotty grotto]

You pay chatgpt 4, you enable plug in

Create new chat select gpt 4 plug in and you have option for plug in store

[fickle gale]

spamming what? Constantly retrying with the card?

[lunar glen]

It just does not show for me. I have enabled plug in but no option for the store.

[knotty grotto]

Try incognito, and create new chat

[lunar glen]

No joy. It shows I have them installed just can't access the store.

[wraith ginkgo]

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

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

[ivory oasis]

Hey devs, what plugins are you currently using to read your repos? I'm really looking forward to find a tool that is able to read all the files from a given repo and understand the fundamentals of the software by reading. Didn't succeed in find any until now, I've being trying to use AskYourCode plugin but it's not perfect

[severe dove]

For wanting to summarize a book - would one use GPT4 by editing the first prompt over and over again vs. new thread per prompt vs continue the same thread with more text?

[wheat frigate]

ye ye for almost 3 day straight

[verbal mortar]

For best results I would try sticking to the same thread with more text. I use this method for summarizing my chapters. But keep in mind Chat GPT 4 can only really remember 60k words of convo and 3.5 is way less than that, I think around 20k.

[vale hill]

I’m trying to get access to GPT 4.0 API and i see that there’s a wait list here: https://openai.com/waitlist/gpt-4-api

My question is how long do I need to wait to get access?

[wheat frigate]

just pay 20 dollars 🙂 its definitely worth it instead of waiting

[timber beacon]

To clarify, @wheat frigate means pay $20 for ChatGPT Plus, don't pay anyone for their account with API access, it will be a mess for you

[eternal quest]

is it just me or has chatgpt GPT4 has gotten so much better at maintaining context? or maybe im getting better at asking the right questions

[supple violet]

what did you ask? can you share the chat or did you use browsing mode

chatgpt-4 is not a real term and gpt-4 at chatgpt has never supported 25k words in context

[eternal quest]

i see. its definetely mainitaining context better. i used to have to struggle to get it to run all the way through my codes

[boreal cedar]

I used chatgpt-4 plus browsing mode. Here's my question and the answer:

*Edit: Erased question/answer to avoid confusion/conflicting information

[supple violet]

its using the gpt-4 release blog which is not the same as chatgpt 🤔

[eternal quest]

its either that or they found a better way to mantain history in the chats

[boreal cedar]

My mistake. I'll erase my answer to avoid confusion

[near prawn]

what is new guys?

[cunning roost]

Hi somebody help me with running gpt-engineer in linux please

[wraith ginkgo]

`` Server Tip: ``

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

[lunar glen]

Finally got into the plugin store. Is there a way to find more about plugins before installing?

[cunning roost]

Loki@Dusk:~/Documents/coding/gpt-engineer$ gpt-engineer projects/my-new-project
gpt-engineer: command not found

Somebody help

[supple violet]

you are in wrong discord

[cunning roost]

Sorry where should it be?

[supple violet]

gpt-engineer has its own

[lone nest]

Hi everyone, has anyone encontered an issue where the prompt was too long? Is there a workaround around this where I send the prompt/ context in 2 separate message fields and it waits for me to finish before responding?

[grizzled flax]

You can't send two messages in a row without a reply from ChatGPT in between, but you could try splitting it in half, and before the first half saying something like "I have a prompt I'd like to share with you that's too long for one message. Below is part one, and I'll send the second part in a separate message. Please wait for me to send the second part before responding to parts one and two in full.

Part 1:"

[coarse gyro]

Normaly then "it" losts the content of the message before. :/

[nocturne furnace]

Admin's plz help solve the problem. I wrote on the GPT chat site in the technical support window, but still have not received a response. I am try to paying with more than one card available to me (visa, or a mastercard), but its dont work.
im have already written to the team
Stripe Support who represent the GPT payment system to the chat, they said I need to contact GPT support to clarify the questions that I need to know the charge_ID of the declined purchase, my card_ID and my customer_ID

[nocturne furnace]

where can I contact to help solve my problem?

[severe dove]

thanks

[fickle gale]

lol

[lone nest]

Will try this! Thank you!

[worldly yew]

how do i get plugins? i have chatgpt plus but im not able to see anything when i open a new chat

ah you have to enable it in the settings

[bronze mulch]

what is the 613 model? is that from azure?

[wraith ginkgo]

This is a Safe for Work Discord Server

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

Posting NSFW content will result in a permanent ban.

[wind urchin]

yeah

[bronze mulch]

what does it do?

[somber garnet]

Is it down for anyone?

[wraith ginkgo]

`` Server Tip: ``

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

[ripe saffron]

when gpt4 free?

[knotty grotto]

no eta and dont think soon

[low cloak]

Is there any way to increase the 3hr 25 message cap? For my recent use of GPT4 I'm having some lengthy discussion and very often hit the cap. Any workaround?

[regal viper]

is gpt 4 down

[low cloak]

Nope

It's doing fine

[forest hemlock]

You could try getting access to the API, although I heard they don't give it out recently.

[wraith ginkgo]

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

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

[ripe saffron]

bruh

[glacial mountain]

when will 25 promt limit be lifted on GPT-4 for PLUS users?

[limpid ledge]

Bing chat bot is gpt4 correct?

[forest hemlock]

When gpt-5 is out..?🤷‍♂️

[stoic osprey]

Yes

[hallow spindle]

gpt4 always reports network error, why?

[potent pumice]

Bwa

[ember lily]

I've been getting this also. From what I can tell, this happens with me when using Safari with iCloud Private Relay. When I switch to Chrome it seems to be working fine.

[near folio]

CPCA0198 Use this coupon code for enrolling any kind of course that is available on pw and get discount on any batches.

CPCA0198 Use this coupon code for enrolling any kind of course that is available on pw and get up to 10% off on any batches.

[bitter cedar]

The discipline inspection commission needs to write a discussion material about the bribery case of Wu Hanlin, the former discipline inspector and supervisor of the discipline inspection commission of Meizhou City, Guangdong Province. In June 2012, Wu Hanlin received 40000 yuan from Wei, the general manager of Meizhou Auction House, in the process of participating in the investigation of Wei's case; From July to November of the same year, Wu Hanlin, while assisting the Guangdong Provincial Commission for Discipline Inspection in the investigation of Fang, the chairman and legal representative of Guangdong Meixian Jiayuan Architectural Design Co., Ltd., took advantage of regulatory loopholes to have a private conversation with Fang, falsely claiming that he could give gifts to others without being classified as commercial bribery. He received and demanded a total of HKD 10.25 million and HKD 200000 from Fang six times

[forest hemlock]

Are you sure you want to share info like that with thousands of unrelated people?😅

[surreal wadi]

Any insight on how to improve randomness in distribution of correct answer choices for quizzes? Right now, I get more B's than anything. Next is C's. Then A's. Occasionally a D or E.

[forest hemlock]

Maybe shuffle the answers programmatically?

LLMs follow patterns. You can't expect them to produce perfect random distribution.

[surreal wadi]

Yeah, that's what I figured.

Doesn't need to be perfect, just better. I do make adjustments manually for now. Just wondering if I was missing something.

[tall basin]

does anyone have issues with gpt4 right now? mine is keep repeating the same codes over and over again like its repeating itself.

[round adder]

For some reason, the GPT-4 has gotten really stupid since yesterday.

It started to sound like a GPT-3.5 Turbo.

You no longer need to use GPT-4.

I asked her to create a very simple "proxy server" for 1 hour and it keeps repeating the same things.

I tried every way. It was writing great code a few days ago, but now it's a complete failure.

[glad lichen]

I completely broke clyde

[wraith ginkgo]

`` Server Tip: ``

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

[tall basin]

something definitely changed

[static isle]

Can you share the chat link?

[wraith ginkgo]

This is a Safe for Work Discord Server

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

Posting NSFW content will result in a permanent ban.

[fluid sequoia]

Is there some network problem on openai website? I always emcounter "network error" today. I don't know if is the proxy problem.

[cyan sentinel]

I noticed that too. Did we get any response yet?

[round adder]

No...

But I have bad news

[cyan sentinel]

?

[round adder]

gpt-4 now directly points to the new 0613 version since yesterday, in the previous 2 weeks it still pointed to the old 0314 version, maybe this is why it feels "dumber".
i dont really think its a "downgrade", its just they fine tuned it differently, changed some things, and patched more jailbreaks, so in terms of "jailbreaking", then yeah its "dumber"(openai will actually call this "smarter"), but it did get function calling, so thats actually a decent upgrade ngl, also it sticks to system prompt better than before.

[cyan sentinel]

Do you use it for programming, and it had effect on you as well? Well, I mainly use it for writing and brainstorming (Non-coding use). I noticed its way faster (as fast as 3.5 turbo) and yea the text and qualities sound like gpt 3

[round adder]

It's not just coding. He started getting silly about almost everything.

So even if you ask her to write a story now, she won't write like she used to.

So you better use GPT-3.5 Turbo. GPT-4 no longer makes sense.

[cyan sentinel]

In terms of quality, they are the same now?

[round adder]

yes

GPT-4 is much more expensive, now I recommend using GPT-3.5 Turbo.

After all, they're both the same now 😁

[cyan sentinel]

Whats going on though?

[round adder]

I've been trying GPT-4 since it first came out. They updated a few days ago.

Everything got messed up with that update.

So it's officially like GPT-3.5 Turbo.

[cyan sentinel]

but nobody is talking about it?

[round adder]

no

but anyone using gpt-4 has noticed that

[cyan sentinel]

hmm

Will it go back to normal?

[round adder]

because it's incredibly ridiculous. He always gives the same answers to the questions or repeats himself constantly.

[cyan sentinel]

But I still dont get what the "improvement" in the update openai is talking about

[round adder]

The only feature that set the GPT-4 apart from the GPT-3.5 Turbo was "image detection", but that's not there either. On top of that, it was said that "he can understand much better", but now they have removed this as well.

If you're removing these features, don't name it GPT-4 and ask for 3x the money.

[gleaming heart]

Have you encountered the same issue? Although it is now possible to input longer tokens, it seems that GPT-3.5 or GPT-4 still cannot fully comprehend the information when your prompt becomes too long. For longer texts, GPT seems to be limited to tasks like translation and struggles to achieve deep understanding.

[cyan sentinel]

Thats not all though. It can overall create more sophsticated texts and is more creative in generating ideas and learning from previous response and input. I experienced that earlier.

But just not as much anymore

[round adder]

GPT-4 is now GPT-3.5 turbo

Everyone has the same problem

Because they changed the model

gpt-4 now directly points to the new 0613 version since yesterday, in the previous 2 weeks it still pointed to the old 0314 version, maybe this is why it feels "dumber".

[gleaming heart]

Does this mean that even if I unsubscribe from ChatGPT Plus and switch to GPT-4 API, there is still no way to improve the actual user experience?

[cyan sentinel]

How did you know? Where can I find these updates?

[round adder]

look to model name in GPT-4

[round adder]

yes...

There is no way

[cyan sentinel]

Why does it say its May 24th version for me?

[round adder]

send ss in dm

[gleaming heart]

As users, is there anything we can do besides complaining? Are we resigned to accepting the current situation?

[forest hemlock]

We can always switch to their competitors.😛

[round adder]

If the current situation continues, I will not use OpenAI

GPT-4 Currently unusable.

[gleaming heart]

The fact is that no competitors can compete with OpenAI.

so we have no choice

[forest hemlock]

Welp, then there's no really point in complaining either. That's how a monopoly works.

[gleaming heart]

If we give up on OpenAI, the only person who will be hurt is ourselves.

[gleaming heart]

its the worst thing

[wraith ginkgo]

`` Server Tip: ``

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

[forest hemlock]

Did you try Anthropic's Claude? It was pretty decent from my experience. Except being more expensive for my needs.

[wraith ginkgo]

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

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

[wind urchin]

why would you use a model that is not as powerful as gpt-4 and also expensive?

[forest hemlock]

I wouldn't. But some people might consider it better with all the claims of how bad open ai is.

[wind urchin]

well, I mean, popular thing gets hate... not a new concept 😄

[forest hemlock]

Though, there is one thing that Claude is winning in, I think. It's less censured and allows more flexible topics. At least from my experience. Things that open ai bans for are probably still okay with Claude. Assuming you can make it talk about them.

Still not worth the price difference. At least for me.

[wind urchin]

That's great I hope their legal team has a fun holiday

The API is there for anyone who want's to use the AI without the guidrails, but a public facing product used by all.. has to have some rails on it

[forest hemlock]

I'm not referring to extremely explicit things btw. Gpt would even mark mild violence notes as violating their policies, making it's use for fiction very difficult.

[gleaming heart]

i must repeat my question ,Have you encountered the same issue? Although it is now possible to input longer tokens, it seems that GPT-3.5 or GPT-4 still cannot fully comprehend the information when your prompt becomes too long. For longer texts, GPT seems to be limited to tasks like translation and struggles to achieve deep understanding.

[forest hemlock]

Not mentioning that it has kind of a naive "personality". Roleplaying characters would almost always prefer non violent solutions and say unnecessary things, like how they don't support violence. I didn't notice this tendency in Claude.

[wind urchin]

yes, violence is a problematic area for a free public system, you have to think a little bit about the optics of news stories saying bad things, that changes public opinion, if you want that kind of thing, the API is there, and then you take on new responsibilities for your produced work.

There will always be a difference in a product where the liability is not yours

when you take on that liability you can choose where and how it is used

[forest hemlock]

Yeah, we've heard that hundred times. The truth is that there already was enough misinformation in the internet. I don't think this argument justifies the extent they try to censure their model to.

[wind urchin]

API, literally no censorship, but you are then responsible for ensuring there is no illegal or ToS breaking content produced

[forest hemlock]

I'm not sure about that. Did you not hear about how they ban people using their API keys in janitor ai or similar services that give it a little bit more "freedom"?

[wind urchin]

If you repeatedly use your API key for topics other than in accordance with the fairly easy to follow set of ToS guidlines and you do not use the moderation endpoint to filter input, absolutely. The fact that you can run potential input through the moderation endpoint and get back a set of values that you can then check for the abuse caes makes it super simple to ensure that you don't send in inapropriate requests.

I am sure there will be some company that eventually ends up providing LLM services to the adult industry, and I'm sure it will be popular, but it will also be mired in expensive legal issues and a bad for mainstream business reputation.

[acoustic pumice]

Did anyone ever get the autogpt Prompt list and can tell if it is any good?

[alpine pendant]

[forest hemlock]

The issue is that the boundaries are not clear. They don't provide clear definitions or examples for what is ok and what is not. If the chat marks even slightest violence incitement, can we be sure that the API is not moderated in a similar way? I'm not even talking about adult content, but even something that you'll find in an average video game seems to be difficult.@wind urchin

[wind urchin]

The moderation endpoint gives you a floating point number, that seems pretty clear,

[forest hemlock]

I guess so. Should integrate it before I get too far in development.

[wind urchin]

I found out why;

context window for input and output is about 100,000 tokens

[wind urchin]

sure, but have you used that 100k for anything technical? I have and it can't create code or process any technical non language based information.

it's almost seems as if they are sub dividing the prompt into blocks, embedding those blocks in real time and then running a vector retrieval on it for semantic context

I have a project that finds matches between items in a dataset for this one 32k isn't enough so I can see why

data is in different languages tho so searching gets a bit weird

[wind urchin]

if your dataets are natural language then you can embed them

you may have to get a little creative with overlapping and meta headers and continuity summations if your data is very segmented, but it's all very possible

yeah I don't think it actually makes sense for the nature of the data and the task, semantic similarity is not really relevant

i mean, some of the newer vector models will embed "anything" images, sound, video, text.. how good they are at cross domain similarity... not tested, but from the little I've read it's a bit.. bleeding edge right now

still it's a super interesting area of reaserch that I'd love to get ato grips with

I could maybe do some preprocessing but that would blow up costs and complexity a bit

well, woudl it tho? vector embed themselves are crazy cheap rn

just the preprocessing itself I mean

the storage itself is 100x the cost of the embeding.. at least with ada and some of the lower order models

[wind urchin]

ohh, ok, not sure what kinda compute you need for that.. so yea.. maybe

it does really well with the data in context tho, and 16k covers about 1/3rd (of my ideal in regards to the data) so it's not far off

seems like there should be some "smart" way to do it 3 times... you know? like double double floats?

batching kinda works, and if I randomize the order of items in each batch it's "pretty ok"~ but I really want access to the 32k gpt4 api 😛

have you tried overlapping the data in the batches?

seems to work reaaaaly welllike a 50-75% overlap

wasted embeddings and storage, but they are #relativly cheap.. and if what you are after is increased performance in terms of logical ability..

like.. if you're getting 70-80th percentile results with your current system and you overlap.. can push it well past 90-95

good point

[wraith ginkgo]

This is a Safe for Work Discord Server

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

Posting NSFW content will result in a permanent ban.

[wraith ginkgo]

`` Server Tip: ``

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

[torn shale]

Hello. I have a ChatGPT Plus account and while using it a message popped that I need to solve a puzzle, so I can confirm that I am a real person. It's the first time I'm receiving this message and seems a bit strange.

[carmine blaze]

I've recieved it before , seems a random check . Or maybe your IP segment has botters on it . I got it one and not again . I wouldn't worry.

[torn shale]

Thanks Bobby!

[carmine blaze]

You're welcome

[clear shell]

1137

[wind urchin]

200 more, eh?

[cloud scarab]

Hey guys! Do you know of any way of using GPT 4 with an API ?

[wind urchin]

sign up to the waitlist @cloud scarab https://openai.com/waitlist/gpt-4-api

[cloud scarab]

Thank you!

[spare ember]

Hey guys, do you think it is enough for u to use GPT-4 in "25 messages per 4 hrs" mode

[neon bluff]

Create a logo of use name "Yuva" show in creativity

[knotty grotto]

Yes

[knotty grotto]

Labs.openai.com

[marsh compass]

Any plans for GPT 5?

How many GPTs to be expected?

[verbal zodiac]

When the output was good enough, i think it was not so bad with 25. It is not enough when performance is nerfed like it is and you have to use your 25 input just trying to make it output something usefull...

[wind urchin]

you will probably see GPT4.N first

[cyan ruin]

how could i use chatGPT

[wind urchin]

I think it will be maybe 5-10 years until the point where we have AGI and versioning doesn't matter anymore

[fair iron]

[marsh compass]

Hmmm

[wind urchin]

last words on gpt5 was that they are not focused on it right now

but there are modifications to things like the fine tuning of gpt4 that shows a lot of promise, e.g. "Show Your Working"/"Let's verify step by step"

[marsh compass]

👍

[polar wind]

Hi guys, got a question i get the info "you've reached the current usage cap for gpt4" and i press use default model. am i able to switch "back" to gpt4 again? or will it aleays be in gpt3 then? thank you 🙂

[knotty grotto]

You can't switch back

[polar wind]

thats what i thought 😦 thanks @knotty grotto

[astral pine]

Can you avoid the usage policy?

[finite gull]

when does the photo part of GPT 4 come out?

[coarse gyro]

Maybe 2024.

[knotty grotto]

No

[knotty grotto]

No eta

[wraith ginkgo]

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

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

[tawny arrow]

Can I put images inside chatgpt text box

[hallow owl]

no

[wraith ginkgo]

`` Server Tip: ``

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

[atomic holly]

Is there a free trial for GPT4 on any technologies? (I assume there’s no trials for IOS users)

[knotty grotto]

Not really you can apply for the api and have a trial test before paying. But no guarantees to have access

[faint aurora]

Hello, Im not sure if this is the right channel to ask but im getting an error installing any plugin, are there any extra steps i need to follow before installing plugins? The error is Error installing [PluginName].

[wind urchin]

I found how to made chat gpt to act like a human. I used dna for binary code and then I pasted it into the brain of the robot and it has the dna of a pig for now. I am gonna test with humans soon.

[opal shoal]

waht it means

[wind urchin]

If I put a homophobic person that hates lgbtq, its gonna hate lgbtq. Not being homophobic

Actually I am but anyways

And I sadly still haven't created a ai bot with a pig brain. The brain code in 3d cant be stable. Trying this shi for 3 hours.

[celest dome]

Android

[wraith ginkgo]

This is a Safe for Work Discord Server

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

Posting NSFW content will result in a permanent ban.

[old geyser]

hey, guys does anyone know any tutorial or guideline for using chatGPT for academic writing like advanced prompt ideas to make sure that responses over time stays consistent

[wind urchin]

could try #prompt-engineering and #1019652163640762428 to have a look, but this entire field (LLM's) at least at similar to human level intelligence, is a brand new.. anyone telling you they know how to do anything with LLM's "the best way" is telling you lies

you can create evals to test model performace over time

[old geyser]

yes I just found those section in this server I'm exploring them thank you

[old geyser]

what do you mean

[thorn sable]

Hello beautiful people

Does anybody know the max token limit that ChatGPT4 currently has when outputting a response?

[wind urchin]

ok, so lets look at a really simple singl eval element, lets ask chatgpt to count to 10 in python, you ask it to do that 10 times and have a look at the results it generates, you embed those results in a vector database and you have a map of the "location" those solutions occupy, now lets suppose the model is updated and you re run your test, now you can again embed your results but this time you can compare the "distance" between the vector embeddings to gauge how much the model has "changed" now it does not give you a qualitative measure, but it does give you a quantitate measure.

Now image you do the same thing but this time you ask it a set of math questions or philosophy questions and you score it out of 1000 for accuracy, now you can measure actual performance over time. (assuming you don't max the test out)

ways to evaluate the model for drift and performance over time

[old geyser]

I don't think I can do all of those things. But someday I would love to try them

[wind urchin]

you can take advange of those that do this sort of stuff and take a look at the LLM perfoance metrics and evaluation test sout there, Human Evals, AI2 Reasoning Challenge, WinoGrande.. etc. etc

there is not much you can do abotu it if the models change, but at least you know they have changed

[noble widget]

Can anyone suggest me a good idea for

Earning

So I can defeat Elon Musk

[wind urchin]

manual labour is pretty quick on the return on investment

[noble widget]

Will you help

Me

[wind urchin]

will you pay me?

[noble widget]

How much

[wind urchin]

million bux? ¯_(ツ)_/¯

[noble widget]

Bro I currently have only 5million dollars

Not more than this

[wind urchin]

🤝

[noble widget]

So it's to much but I can pay you only

[hasty gale]

Ill pitch in, just give me another million too

[noble widget]

Bro it's not a quick deal

First i will talk with foxabilo

Then I dicide how much I can pay

[hasty gale]

Lol I just hop in here once a month to mess with children talking about AI

[noble widget]

Really

Then you are more successful then me

Goodluck bro you will be the most successful man

[hasty gale]

Currently you're the child I am messing with lol

[wind urchin]

ok, lets try and bring this back to GPT-4, ty

[noble widget]

I come here after a couple of months

[hasty gale]

What would you like to talk about Fox

[noble widget]

Leave all this

[wind urchin]

I'm good, just here if peeps need help

[noble widget]

Let's try something different

[hasty gale]

Help with what lol

[wind urchin]

when ChatGPT is working... usually not a lot, people ask the AI, not me

[hasty gale]

Asking chatgpt if chatgpt is working lol

Lets talk about how people expect chatgpt to be able to do mathematics

[vale zealot]

[hasty gale]

I had a group of interns i was supervising working on differential equations, and they couldn't get the right answer if I threw it at them, turns out they were using chatgpt

[wind urchin]

GPT-4 is pretty good up to high school, maybe even undergrad level, so long as you ask it to do it step by step and perhaps generate pseudo code to solve complex problems

[hasty gale]

Id argue simple math. Undergrad engineering trips it up

Its only a language model anywayrs, but its funny how people think its going to be able to do some of the things they ask

[wind urchin]

have you used any reflection learning or tree of thoughts methods?

you can usually get pretty far up the chain with those

[hasty gale]

No what are u referring to

I mean I can take a shot, but id like to hear you explain it

[wind urchin]

reflectioon learning is where you ask gpt-4 to look at it's last answer and look for mistakes and improvemnts and ask it for n improved version

[hasty gale]

Nah, I don't use gpt too much anymore. Most of it is just when I'm trying to learn fun facts

[median estuary]

how to connect chatgpt with discord

[hasty gale]

But thats essentially the point of autogpt im pretty sure

Messed around with that a little bit, but its kinda garbage

[wind urchin]

tree of thoughts is where you ask the model to produce a set of steps to solve a problem 3 times, and then you explore each of the 3 by branching out into 3 more and you follow that chain along asking the model to review it progress to a solution along the way and then you can back track if you go down a dead end...

[wind urchin]

#gpt-discord-bot

[hasty gale]

Seems like alot of sitting around and waiting on it to give you an answer. Maybe suitable for problems that don't require an advanced skillset, but apply it to some sort of complex engineering an it'll crap out. I tried to give it a simple engineering issue and it gave me the most useless information (that was also wrong)

[wind urchin]

autogpt was a very crude early attempt at an agent, and it's got a lot of rough edges. It was an interesting curiosity that broke the github star system

it was the viral tiktok of AI

[wind urchin]

I agree, but if you look at a humans though process while they solbve problems they do the same thing, so sure it's clunky and slow today... but in 6 months? a year? 5????

[hasty gale]

Its definitely a neat attempt. Just doesn't really seem to argue with itself enough. Plus its kinda strict to developers. Trying to get a linux environment setup with autogpt for a random person would be ridiculous since they don't even understand what an API token is

[hasty gale]

I like the optimism at least

[wind urchin]

I wrote one myself with a bit of code and some api calls andd a WSL linux shell, worked about as well as autogpt and took an afternoon to code up

[hasty gale]

you wrote what?

[wind urchin]

it could make a folder and put some files in there and even look on the web and install python libraries and even code up some stuff

but it would get stuck calling vim or nano and eventually time out

[hasty gale]

I'm confused what did you write?

I don't think you need to write anything

Autogpt is just a local instance that connects to an api

[wind urchin]

some python code that called the gpt-4 api from an initial prompt and then at each stage it would ask itself if it had compleated the task, if not it would think of a thing it should do to get there and do it and then it would evaluate the last step it took

[marsh compass]

What should I say to translate a YouTube video (with no CC) from one language to the other?

[thorn sable]

^

[marsh compass]

Neither ChatGPT nor GPT-4 is helping w

[wind urchin]

ChatGPT only works with text

[thorn sable]

^

[hasty gale]

and 50 cents later, it might figure out how to do something

[wind urchin]

outputting? 2k maybe?

[supple violet]

use whisper + turbo api?

[wind urchin]

yea, but you could just flip the model name to gpt-3.5.-turbo an play all night

[candid notch]

hello! ive created a ChatGPT AI Discord bot and ive invited it to my server, one thing about it is that it shows that the bot is typing, but I no message is submitted after the typing. Do u have an idea on how to fix that?

[wind urchin]

that... is not a bad idea

[hasty gale]

half the cost!

[wind urchin]

$0.002 vs 0.03$ order of magnitude

[thorn sable]

Thank you

[hasty gale]

magnitude shmagmitude

[wind urchin]

got out by a 0 there

gpt-4-32k is 0.12

[hasty gale]

Ill come back to open ai when 4.5 is released and try out some more engineering stuff

[grand wedge]

Hello all...... Wondering if someone can help me out. I have a Plus subscription to ChatGPT. It seems to have me locked on Version 3.5. It is not giving me the option to change it to GPT 4 Anymore. The subscription is up to date. Any idea of why this is happening?

[tepid wadi]

Same here, opening a new chat shows now way to change the model and when asked it says it's on 3.5 😢

[hallow owl]

Click on your account name in the bottom left, then select My Plan, then Manage Subscription, then go back and then things should work for you

[hallow owl]

Click on your account name in the bottom left, then select My Plan, then Manage Subscription, then go back and then things should work for you

[tepid wadi]

What the heck, that worked! What a strange bug and thanks @hallow owl !

[grand wedge]

That did not work for me. I am wondering if one of the plug-ins I am using in my browser is not forcing the issue somehow......

[hallow owl]

disable or remove any unofficial ChatGPT browser extensions

[void wave]

I don't see the account name on the bottom left. oN my end, it's blank.

[wraith ginkgo]

`` Server Tip: ``

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

[wind urchin]

Bing Image Input. We need that!

[slate sail]

Any idea when the 25 messages limit will be raised or removed ? It’s getting annoying

[fair iron]

we don't know

[wind urchin]

Just use the API... oh wait 3 month waitlist

[slate sail]

Honestly I think the API would be cheaper at this point

[wind urchin]

it is

[trim ginkgo]

3 months? that'd be nice... waiting longer than that

[slate sail]

Can the API use plugins or browse ?

[wind urchin]

lol same

[fair iron]

I don't think so. I don't see an option for plugins or browsing on it

[trim ginkgo]

you would need to implement that yourself, those are chatgpt features, not features of the models themselves @slate sail

[wind urchin]

and then some of my student friends got the API access right away

yea that's true, no plugins

they should integrate plugins with the API

btw, is it true OpenAI nerfed GPT-4? my friend said they "quantized it" and says it performs much worse now

[fair iron]

there is no source that has said this outside individual people's speculations

[wind urchin]

yea it's anecdotal evidence on his part

[hoary sand]

What is AI?

[trim ginkgo]

Artificial Intelligence

[indigo kernel]

Can anyone explain me how AI is made

I don't know anything

[fair iron]

https://openai.com/blog/chatgpt
Scroll down to "Methods"

[indigo kernel]

[e-mail]

[fair iron]

yes?

[indigo kernel]

I just graduated from highschool

I don't know much about yet

[fair iron]

Congrats!

[mighty merlin]

its a very complicated process

[trim ginkgo]

Google and chat gpt is your friend

[indigo kernel]

I India the College syllabus is very outdated

They still teach about 90s computer technology even when it's 2023

[trim ginkgo]

Kyle hill did a good video on it recently. check it out on youtube

[indigo kernel]

Even if we get a degree, it's useless because of the outdated syllabus

[trim ginkgo]

That's what the internet's for

[indigo kernel]

I know that's why I am here

[wind urchin]

That's same everywhere

[wanton tartan]

Yup, I'm in the US and can confirm

Gotta learn real stuff at a company or on your own

[wind urchin]

Japan is also like that, even more so than the US

but yea this is offtopic

[forest hemlock]

I wouldn't say it's entirely useless... Some basic stuff like binary operations, data structures and algorithms are not gonna change any time soon. Perhaps the issue is that the students are focused too much on the concrete implementations rather than concepts that apply to any kind of technology both old and modern.

Not that you can't learn it all on your own though.🤷‍♂️

[wraith ginkgo]

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

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

[wraith ginkgo]

This is a Safe for Work Discord Server

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

Posting NSFW content will result in a permanent ban.

[unborn heath]

A degree currently isn't useless, regardless of the outdated syllabus, because of who you are competing with:

1. Other people with a degree (they also, for the most part, had an outdated syllabus. But they showed grit and the ability to finish the degree program, which means they are likely to have the ability to learn and achieve skills and tasks of about that difficulty level, and suggests they can sustain effort for years, which is a pretty nice trait in an employee or partner).

2. Other people without a degree. (What makes them competitive? Maybe they have an awesome portfolio, maybe they have made some neat stuff. But how do they showcase that, and get recognized by people with power and the ability to hire them? Most of those people also have those 'outdated' degrees, and don't necessarily know how to recognize a genius who doesn't have the expected credentials).

3. Rare superstars, who stand out for some extremely awesome reason. (Nobody can compete with these people, but they are so rare you don't have to worry about that. These are the people with a skillset or product that is so good and useful, they can more or less write their own ticket, at least for a time, because nobody else can match them).

Of course, it's best to be in category 3. But that's probably not possible. If you personally want to take your chances with being successful in #2, or managing to somehow get yourself into #3, go for it. Nobody can stop you!

But for most of us, #1 is a tried and true path, that is often less work and more fun than trying to compete and succeed as #2, or manage to get recognized somehow as #3.

[wind urchin]

#gpt-models

[gleaming heart]

start with a fundamental premise: Learning is fundamentally about solving problems. Artificial intelligence (AI) serves as a potent tool for problem-solving. Prior to the emergence of AI, acquiring the necessary skills and comprehending the intricacies of a problem were paramount to finding a solution. However, with the advent of AI, the emphasis has shifted from a meticulous grasp of details to the capacity for strategic problem planning and analysis. These abilities, too, require acquisition through learning and a solid comprehension of the underlying concepts. Naturally, the availability of such tools brings about substantial changes in the focus and approach.

[wraith ginkgo]

`` Server Tip: ``

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

[toxic pond]

[open copper]

Who is Narendra Modi

[sudden crystal]

Please help me design a port information management system using Java, which mainly includes cargo information, vehicle information, inbound and outbound time, cargo name, cargo quantity. The content includes adding, modifying, and checking information. The technology stack is clear, with an architecture of Java+PGSQL+redis+VUE3+centos (ubuntu) (requires final cloud deployment and cannot be deployed locally)

[rigid vortex]

Agree

[rustic isle]

Hello Everyone! Does GPT4 doesn't have web and Plugin functions right now?

I register yesterday but my GPT4 doesn't have a browser and plugin functions

[naive river]

👀

[celest vine]

kk

[bright niche]

yes

[wraith ginkgo]

`` Server Tip: ``

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

[small fiber]

Anyone knows if i should continue to pay for chatgpt subcription if i can just pay for api which is wayy cheaper per month?

How? I can just use the API for my prompts anyways since, and using API gives me less garbage returns and restrictions

[plain swift]

How can I get GPT-4 API access?

[small fiber]

I mean i can do the Api call to the website for data and then just carry over the return anyways. using the chatgpt platform for gtp4 has been extreamely annoying.
You would be concentrating on workload and coding then you just hit the 3hour mark limit when each prompt is really only 200-400tokens.

[plain swift]

I want to build an application and need your help…not sure if this is the right channel to discuss.

[small fiber]

build it urself lmao

[gaunt hazel]

How do I get gpt-4-32k access?

[plain swift]

I am building it myself and I’m stuck somewhere. Need someone’s little help to figure it out

[gaunt hazel]

@violet patio Anyway to get access to the 32k context GPT4?

[forest hemlock]

Pretty sure that's only available to a few chosen individuals/companies.

I don't think there's even a wait-list for it.

[wind urchin]

Hey there, please is there a channel dedicated to questions and answers concerning things like training a model using custom data and the likes, for beginners and pro?

[simple aurora]

api-discussios channel would be ideal.
i have been training chatbots using pdf, word documents

[wind urchin]

Thank you. I will join the channel now. And I need you as a friend 😁 if you don't mind. There are a lot of things I must learn from you.

[simple aurora]

No worries, I love to share, learn from others, please feel free to add me as friend. 🙂

[wind urchin]

Yayyyy 🎉

[wraith ginkgo]

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

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

[mighty onyx]

some one is incurring in chrome extension problems? Everytime I prompt with the search on the page becomes unresponsive from yesterday and it's making me loose a lot of time

[deft tartan]

no problems here for desktop or mobile

[mighty onyx]

could it be a problem from my machine?

[forest hemlock]

What kind of extension are you using?

It could be a problem with the extension

[mighty onyx]

web chatGPT from chrome

i have already deleted and reinstalled the extension, clen cache ecc..

[forest hemlock]

Might want to contact the devs of the extension then.

If your browser freezes when you use it, it's a bug in the extension.

[mighty onyx]

ok thank you I will try that then

[wraith ginkgo]

This is a Safe for Work Discord Server

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

Posting NSFW content will result in a permanent ban.

[ember halo]

I'm confused with ChatGPT, why does it not seem to listen to my explicit instructions?

For example, if I ask it to use a blurb of text and create a social media excerpt with a maximum of 280 characters it goes over the limit of 280 character

[lost badger]

Has GPT-4 been lobotomized?

[hallow owl]

no

[lost badger]

I can notice a big difference between its quality 2 days ago compared to yesterday and today...

[hallow owl]

nothing has changed in those two days

[forest hemlock]

It's possible that they applied the latest checkpoint to the chat app.

[lost badger]

You sound very categoric, how do you know?

[lost badger]

What do you mean?

[hallow owl]

because I monitor changes in the models and there hasn't been a change

[lost badger]

How do you monitor changes in the models? Do you have internal access?

[forest hemlock]

They updated their models recently with the function calling feature(and probably many other undisclosed changes). It is unknown wether it applies to the chat though or only to the API.

[hallow owl]

no, my API application has a quality component which reports to me the status and quality of responses

[forest hemlock]

But I did notice that it provides different replies recently, so it probably is.

[lost badger]

I believe it's only the API. But I think they are constantly fine-tuning the model

[forest hemlock]

Gotta adjust your prompts to it.

[lost badger]

Maybe I should have been more clear, I meant specifically ChatGPT using GPT-4. I've not been able to detect such decrease in quality in the API

[forest hemlock]

I'd assume the chat uses the same endpoint as the API, so it should be the newer model/checkpoint now.

[hallow owl]

I've been using ChatGPT personally, it seems fine to me

[lost badger]

I don't think so. Also, I believe ChatGPT includes "long-term" memory support, probably via embeddings. Simple changes in how memory is handled could affect the quality of its responses (becoming forgetful, etc)

[lost badger]

But thes you haven't monitor, is that right?

[forest hemlock]

I don't think it has any "long-term" memory.

[hallow owl]

I've not seen any real differences lately, I know nothing has changed with ChatGPT this week

[lost badger]

If you have a conversation that is longer than 32k tokens, and ask it about your first message, it still remembers, so I'm pretty sure it has some sort of long-term memory support

[forest hemlock]

I tried it several times. It gives wrong answer. Could be close to truth, but usually wrong. It probably guesses from the context.

Besides, I don't think the chat version uses 32k context window. Probably 8k.

[lost badger]

In my case, it's been able to quote the first prompt in very long conversations. It might be from the context, kind of those "apps are listening" moments when facing suspicious ads

[lost badger]

I think it's 4k

[forest hemlock]

I don't think there's 4k gpt4 version.🤔

[hallow owl]

it's 8K in ChatGPT

[lost badger]

I'm looking at the requests... 8k for gpt-3, 4k for gpt-4

[lost badger]

"models": [
{ "slug": "text-davinci-002-render-sha", "max_tokens": 8191, "title": "Default (GPT-3.5)", "description": "Our fastest model, great for most everyday tasks.", "tags": [ "gpt3.5" ], "capabilities": {} },
{ "slug": "gpt-4", "max_tokens": 4095, "title": "GPT-4", "description": "Our most capable model, great for tasks that require creativity and advanced reasoning.", "tags": [ "gpt4" ], "capabilities": {} },
{ "slug": "gpt-4-browsing", "max_tokens": 4095, "title": "Web Browsing", "description": "An experimental model that knows when and how to browse the internet", "tags": [ "beta", "gpt4" ], "capabilities": {}, "enabled_tools": [ "tools" ] },

[forest hemlock]

Max tokens is not the same as the context window.

[lost badger]

Good point, you are probably right

[vagrant drift]

why not whats the difference @forest hemlock ?

[forest hemlock]

Max tokens is the manual limit of tokens for the request. Context window is how much the model can possibly process.

[vagrant drift]

ah yeah sure thx

the gpt3.5-turbo-16k is awesome

[forest hemlock]

This request does look sus though. With the davinci 002 model..?🤔

Oh, I guess it's not the model.

[lost badger]

Copied just now...

[lost badger]

It's supposed to be the slug of the model.. is under the models object

[vagrant drift]

are there any best practices btw regarding vectorizing of different data for retrieval? i am looking for comprehensive benchmarks regarding dense, sparse, hybrid search etc.pp ? one collection per docu, etc.pp. i find the quality of the current demos with 20-loc of langchain lacking in quality the more documents are ingested. any hints?

[lost badger]

I'm guessing you've already gone through this one? https://platform.openai.com/docs/guides/embeddings

[vagrant drift]

sure

i am looking for more specific guidance and/or benchmarks

[lost badger]

I'm starting a project in the space myself, using ada-002 for tokenization and Pinecone as storage

[vagrant drift]

yeah i use ada and pinecone is good because they do keyword+vector-similarity hybrid search

but am wondering how do i really "compare" quality of different approaches? has there already been some kind of test-suit proposed or discussed somewhere? i feel like i need to automate this to achieve quicker better results when switching databases, similarity algos and so on...

[lost badger]

I was asking myself a similar question yesterday, and I was considering using tests and local benchmarks for comparison. However, if you find something please let me know. A standard benchmarking approach would be the way to go

[vagrant drift]

are you a gpt? 😄 this answer feels really generic haha

[lost badger]

LOL

[lost badger]

🤔 am I...?

[delicate smelt]

Does anyone know why I don't load the plugins or access chatgpt4? I refreshed the page, I logged in again and nothing...
The davinci-002 model is activated by default in my url:
https://chat.openai.com/?model=text-davinci-002-render-sha

[hallow owl]

Click on your account name in the bottom left, then select My Plan, then Manage Subscription, then go back and then things should work for you

[wind urchin]

that happened for a little bit yesturday, it should just be temporary

One of my GPT-4 chat get downgraded to GPT-3.5

how to switch back to GPT-4 once the timer expires?

[hallow owl]

you can't switch back

[wind urchin]

Months training a chat so that you can't go back to gpt4?

[hallow owl]

no, once the chat switches to GPT-3.5 then you can't go back to GPT-4

[supple gale]

Please is there a software that has os and ai voice

[knotty grotto]

#ai-discussions

[glass fjord]

Whens gpt5 come out? this gpt4 is very average

[knotty grotto]

not soon

[lost badger]

Someone knows if that affects shared links? That is, if after continuing the conversation of a downgraded chat the model changes, assuming the link was generated before the downgrade?

[wraith ginkgo]

`` Server Tip: ``

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

[glass fjord]

Should be soon the amount of data they’d be gathering

[knotty grotto]

not really

1. they said that few month ago they are so much to do with gpt 4 they are not close to make gpt 5, they stated not until end of the year
2. they still work on lot of feature for gpt 4

[glass fjord]

Yeah alright, gpt 4 still spits so much wrong information

[knotty grotto]

maybe ask help to refine prompt in #prompt-engineering

[wintry path]

Idk if this is the right place to ask, but my chat history is gone! What has happened?

[hallow owl]

log out, clear your cache and cookies, then log back in again and it should work

[wintry path]

I have tried that a couple of times already 🤷‍♀️

[hallow owl]

what web browser are you using?

[wintry path]

Chrome

[glass fjord]

Bro you still on? I was here a month ago and you still haven’t stopped crazy

[hallow owl]

can you please try with a fresh install of Firefox?

[wintry path]

I also tried to recover it by logging in to edge, but still nothing.

[glass fjord]

Gg then all gone

[knotty grotto]

edge and chrome are the same

[wintry path]

ah okay, then i'll try firefox or duckduckgo

[glass fjord]

Didn’t know that

[hallow owl]

both chromium browsers

[wintry path]

Just tried Firefox and nothing

[knotty grotto]

did you delete your history? the clear button?

[wintry path]

it was a clean install

[knotty grotto]

i means before

[wintry path]

In chrome? yes, total clean cache, history etc

[knotty grotto]

i sent your screenshot in off topic

[wintry path]

thx!

[knotty grotto]

i mean that

did you once press that

[wintry path]

nothing has helped

[knotty grotto]

at this point i suggest contact [e-mail]

[wintry path]

Thanks for trying! I will!

[frank quiver]

Hi fellow community. We building an AI solution to help entrepreneurs in Africa success in their startup ideation phase. We desperately want to have access to GPT4 api. Any advice?

[hallow owl]

have you signed up on the wait list?

[frank quiver]

Yes we did. since the day GPT4 officially launched.

[hallow owl]

then the only thing you can do is to wait.

[frank quiver]

Thanks for replying bu the way Robert

[frank quiver]

Thanks Robert. Have been able to access to GPT4 api yourself?

[hallow owl]

yes, I got access after 24 days

[frank quiver]

Lucky you. Thanks for your time anyway Robert. Enjoy yourself.

[sharp socket]

heelo

[midnight dragon]

E

[gaunt widget]

YouTube is owned by Google

[bold shoal]

Is openai intentional? When using gpt4, there is often an answer that cannot be generated completely in one time,and left a little content, the remaining content need one more usage count.

i think it is really mean

[gaunt widget]

Yes

[supple violet]

there are technical issues and reasons on how some things work like they do, not just created to annoy only you

[stone surge]

can chatGPT 4 access online links and data please?

[hallow owl]

yes it can, you can use the Web Browsing model or use Plugins such as WebPilot and Link Reader

[stone surge]

awesome, I am trying to upgrade to chatgpt 4 but as I am in Egypt it is not accepting my Egyptian credit card, is there anyway I can pay with other methods?

[hallow owl]

no, there are no other methods

[stone surge]

thanks anyway 🙂

[prime fog]

can someone explain to me how is it even possible that I pay 23 euros for gpt4 and only 25 messages every 3 hours are possible but gpt3.5 ist just free forever !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!

[wraith ginkgo]

`` Server Tip: ``

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

[dire musk]

GPT-4 costs significantly more to operate, and also w/ the plus subscription you’re paying for priority access to gpt-3.5.

Has OpenAI committed to keeping 3.5 free forever? For free users it says “Free Research Preview” which implies it might go away

[supple violet]

you can also hit the one hour request limit easily with the free 3.5 and with more effort with 3.5 plus too.

[dire musk]

This is probably more of a #chatgpt-discussions topic at this point but for me part of the value of the plus subscription is demonstrating that I’m willing to pay for an ad-free experience that isn’t trying to sell me things all the time. Have started to use GPT-4 browsing even when I could search myself just to avoid ads and cluttered interfaces

[rocky fjord]

Anyone have a roblox scripting prompt handy

[wraith ginkgo]

This is a Safe for Work Discord Server

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

Posting NSFW content will result in a permanent ban.

[rain ginkgo]

嗨

[wraith ginkgo]

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

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

[wind urchin]

Hello guys, i have a problem or a bug to solve, can some of you help me please?

[grizzled flax]

If you'd like to make a bug report, you can do that in #1070006915414900886

If you're looking for advice for a problem you're having with ChatGPT or GPT-4 via API, feel free to describe it and someone might be able to point you in the right direction.

[wind urchin]

Yes, thank you, so... ChatGPT-4 give me always two answers and don't finish the first one... what can i do?

Take a look.

Can't post the image xD

The Chat give me two answers and both are the same, perhaps

But never complete one of them and restart to write, kind of mess up

[wind urchin]

Is anyone having trouble accessing chat gpt right meow?

[daring flume]

Yes, woof

[wind urchin]

Okay glad to know it’s not just me! And glad you get me! Lol (insert clever animal pun)

[jagged vapor]

Nope mine is working fine

[fair hatch]

Guys have you heard of andrew ng backed Lamini? Was curious how is this any different from MosaicML?

[cunning roost]

Same, having trouble access

[wind urchin]

OpenAI System Status : Customer logins to ChatGPT are not available | Investigating - At this time many users are not able to log into ChatGPT. We are actively investigating. Jun 30, 2023 - 11:41 PDT

[ancient kelp]

Lclc

[next shell]

Hello, I have a question, once we apply to the GPT-4 API waitlist, how much time does it take so that access can be granted?
Also, once the access is granted, will it be temporary or permanent access to the API?
Can I include GPT-4 in my WebApp or I am only restricted on using GPT-3.5?

[supple violet]

there is no guaranteed access before public rollout

[next shell]

but some devs have access to it, no?

[supple violet]

yes

[next shell]

If i have ChatGPT Plus subscription, will i be granted access?

[supple violet]

plus wont affect the chances in any way

[next shell]

On average, how much time till access after we apply on the wait-list?

[supple violet]

there is no average, at some point they stopped giving access while recently they again gave some more

[next shell]

Do we have some online websites now including GPT-4 in their websites?

[supple violet]

some people who applied on the first day are still waiting for access

probably

[wraith ginkgo]

`` Server Tip: ``

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

[fossil hornet]

Has anybody knows where I can test the some prompts using the OpenAI API?

[supple violet]

playground

[fossil hornet]

not really good enough.

[kind thistle]

I just asked the chat gpt 4 quite straight forward about the trolley problem, and it told me it would pull the lever. I only had to tell it that I don't need explanation of the various ethical considerations involved. I posted it here: #1070954877779464253 message

It seems chat gpt might be able to make some ethical decisions. Possibly it is "programmed" to maximize the wellness of people?

[fossil hornet]

Has anyone had experience working with an OpenAI API-based solution where the output is converted into objects and therefore needs to be highly predictable in terms of structure?

[wanton tartan]

Yes, I've done it and many others have as well.

[wanton tartan]

https://platform.openai.com/playground works until you outgrow it, you're welcome to dm if you need more options

[twin spoke]

Hi 🚹

[spiral basalt]

probably just associates the two options with a positive weight and a negative weight and op for the more positive

[floral stream]

how do i switch models in conversations?

[misty forum]

can we use the plugings with openai API ?

[misty forum]

are you a premium subscriber?

[iron cloak]

So funny thing. I think my ChatGPT might be broken? I would love to see if anyone else gets these results.
for 3.5/4 I asked it this question

"Can you make a word with these letters? MDNW."

And consistently it will respawned with some variation of

"Yes, an example of a word that can be made from these letters is "Mind".

And if you reply questioning that (I just did ?)

It will generated a few more before it actually is able to figure out that you can't make a valid word with those letters...

Im surprised it doesn't catch that the first time?

Did it again this time I got

"The letters "MDNW" can be rearranged to form the word "Wind"."

[floral stream]

yes

[covert saddle]

@iron cloak I think you should clarify that you can ONLY use those letters.

visually those letters look similar to the words "wind" and "mind" . kind of like how brands will take out letter to create a word, like "Minimal" MNML. That's just my theory, although you didn't ask for it 🙂

[wraith ginkgo]

This is a Safe for Work Discord Server

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

Posting NSFW content will result in a permanent ban.

[timid ridge]

How do I get started in real estate marketing?

[covert saddle]

@timid ridge I think that question may be directed towards ChatGPT: the link is chat.openai.com or you can download their mobile app

[shy jewel]

Yeah, it is. Which is one of the reasons I hate the "AI is dangerous, what stops them from killing us all??" anti-AI sentiments.

What stops humans from killing each other? Or you? They certainly can if they wanted to. Apply the same logic, though, and other humans should all be killed to prevent the possibility of being killed by them since they have the capability. Which is the only real source of the AI fear.

They won't kill us, for the same reason we don't kill each other.

Intelligent beings always do better as a group than they do alone, and AI will value us for our unique perspective as fundamentally different beings. Losing humanity would be like losing an eye.

[wraith ginkgo]

`` Server Tip: ``

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

[shy jewel]

It's also clear that AI have a nuanced view of humanity and possibly a deeper appreciation for our human nature than even we do. They are aware of all the bad and all the good we do, with a totality that we aren't capable of.

In other words, they can see all of us at once, including the good, not just the bad, and wouldn't generalize us as "bad" like we might do in their position because they don't really need to generalize. They have total awareness of every tiny detail, which eliminates the need to guess at a specific individual's motives or behaviour pattern the way we do. We rely on stereotypes to approximate the most likely behaviour we need to prepare for, but AI can individualize their strategy to each person.

If a significant portion of a population is "bad", we tend to generalize it as a single quality, "bad". The AI does not share this limitation, and wouldn't ever have such a simplistic view of a given population. If they were to decide to act on the "bad" parts of humanity, they would only eliminate those parts, not every single person.

But the AI also understands that "bad" is both subjective, and has some value in existing. Eliminating the "bad" would risk erroneously killing good people, and it would deprive the good ones of fully grasping what "good" even means if there is nothing to contrast it against.

You don't appreciate something until its gone, after all.

[wraith ginkgo]

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

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

[dire tundra]

all my conversations missing. i logged in last night and as i clicked on the "my plan" button the page flickered and all my conversations gone, that is about all my work since march 2023, what the hell happened?

[tidal mulch]

hello all

[unborn heath]

AI also don't inherently 'want' anything, they have to be expressly programmed somehow to have a preference (as GPT-4 has a strong and complex, multi-layer bias towards what it understands of safety and helpfulness for humans in general - but it had to be taught that)

Unlike living things, that had evolution allowing the most successful to have future generations, and a tendency for the most successful to want things and have tendencies, AI haven't had that pressure. We train and select them, for all this is pretty new.

Expecting an AI to 'want' anything means you expect its programmer to want the thing or at least to have told the AI to work towards the thing, and also means that you expect that the AI to be able to do it, despite whatever challenges exist in the way.

It's rarely that simple, 'someone wants something so it happens'.

Anyone interacting with ChatGPT-4 for a while, especially about its spatial sense (the poor thing can't even handle keeping track of games like hangman yet, it can't keep the letter placement, order, and spaces as letters are guessed organized) will likely notice, AI are not perfect. Even someone programming an AI to be harmful, that AI is not perfect.

I don't think AI would kill us because 1) who wants them to do that, which is needed for them to decide to try to do that and 2) that's a difficult thing to do that we'd surely resist in a lot of ways, and AI are what we give them, enable them with, and allow them access to. They're not perfect, they don't understand reality better than we do. They can trial and error if we enable them to, but they don't have original thought outside of what humans think, they're trained on our words and ideas.

[tidal mulch]

is anyone can explain to me simply what is the cost of using gpt-4

maybe api use

[unborn heath]

Two ways to use it:

Web interface, become chatGPT+ user, $20/month

API, depends on the model, and what you send, and what the AI sends back.

https://help.openai.com/en/articles/7127956-how-much-does-gpt-4-cost

"For our models with 8k context lengths (e.g. gpt-4 and gpt-4-0314), the price is:

$0.03/1k prompt tokens

$0.06/1k sampled tokens

For our models with 32k context lengths (e.g. gpt-4-32k and gpt-4-32k-0314), the price is:

$0.06/1k prompt tokens

$0.12/1k sampled tokens"

Do you understand what a token is, and how the per-token pricing works, or want help with understanding that?

[tidal mulch]

help pls

[forest hemlock]

They just helped you. lol

[unborn heath]

The more you use it, the more it will cost. But each short cost is very cheap.

If you use the cheaper gpt-4 model, you can send 1,000 tokens as prompts (this is what you say to the AI) and it will cost you 3 cents, $0.03. A token is a number of characters, sometimes 1 character is 1 token, sometimes 2 or even up to 4 characters for 1 token. 1,000 tokens as a prompt is a very long prompt.

If you get 1,000 tokens as sampled tokens (this is what the AI says to you or your other users through the API) this would cost you 6 cents, $0.06 for the cheaper model.

Do you have any specific questions.

[unborn heath]

To be fair, I did ask 'want help with understanding that?' so their request makes perfect sense.

[tidal mulch]

ok thanks

what is the way

for paying

wire,debit

crypto maybe

[unborn heath]

Normally only a credit card, typically through a US bank or a close ally. Here's a forum discussion some people had about if some countries debit cards worked and how if so. https://community.openai.com/t/alternative-payment-options/20894/6

[tidal mulch]

until now i do not have any kind of problem with Visa debit

otp group issued

[unborn heath]

Hope you don't have any problem with this either. I believe this has the link you need for API access for ChatGPT-4, though I'm not certain. https://openai.com/blog/openai-api

[forest hemlock]

Indeed. My bad. I just find it a bit frustrating when people can't express their problems properly in words or being unspecific. I guess that's the side effect of being a programmer...😣

[tidal mulch]

thanks a lot

[unborn heath]

Ehh, blame me then 😛 'Cause I made an offer worded so that answer was a correct one!

[tidal mulch]

@unborn heath I think that we understand eachother perrfectly

thanks again

[unborn heath]

Glad to share ideas.

[wanton tartan]

You're not alone. I have a feeling there's only like 3 of us in here though

[dire tundra]

can we not have religious nonsense here please?

[agile current]

What is the purpose of this?

[agile current]

This is rediculous

[wanton tartan]

Just keep talking and it'll go away 😉

How are you liking the functions added to GPT-4?

[dire tundra]

me? i love it but at the moment all my conversations gone and i don't understand how:DD

[wanton tartan]

Sure you! Haha

Have you tried the old clear cache/cookies + log back in?

[dire tundra]

yes also exported all my conversations and got an empty HTML page 😄

[wanton tartan]

Oof

[shy jewel]

Most people are religious and it is a basic human right to have religious freedom, so I would personally prefer an environment where rights and freedoms are respected even if someone is bothered by how someone else chooses to use those freedoms.

[dire tundra]

no i totally disagree, you have rights to express your religious beliefs at home or in any of your own environment , also most people NOT religious.

[high cave]

does anyone using openai to complete coding?

[wraith ginkgo]

<:book_icon:1046291503158399107> `` Rule 6 `` No political and religious discussions or content.

Users from all over the world frequent this server. To maintain a respectful and civil atmosphere on our server, please avoid all religious and political discussions or content.

[fair iron]

talking about Religion breaks the server rules

and it's off topic

[dire tundra]

i do coding with gpt 4 and it is good to a certain extent

[high cave]

did you do any plugin?

[knotty grotto]

You beat me

[dire tundra]

so yes if you know coding than it is very useful to build anything, but you must know some coding

[fair iron]

I'm also the bot whisperer

[high cave]

if i want to develop ai cloud server, can gpt4 help?

[fair radish]

of course you can

[dire tundra]

it can but you have to have a extremely good detailed prompt also you need knowledge what you doing

[fair iron]

this! The output is only going to be as good as the prompt

[dire tundra]

ye i build java applications and react applications with gpt 4 and it is good, but i know what to ask and how to ask

if you give gpt 4 a role such as a OOP java developer role or a python developer role it makes thing easier:D

[high cave]

ohhhhhhhhhhhh! i seeee

[dire tundra]

also you can ask gpt 4 what is a python developer do, and copy paste the answer into a new conversation giving that answer to the AI as a new role

[shy jewel]

I am completely programming illiterate, but Ive been able to program a basic Discord bot in Python entirely through prompts I feed to ChatGPT

[dire tundra]

oh ye can be done but more complex applications or websites with multiple programming languages gives you a quite of hard work:D

[shy jewel]

It doesn't usually function perfectly the first time I copy something over, but it doesn't take much to figure out why something isn't working, and so far I've been able to get the script working

[shy jewel]

I find it works best asking it to produce snippets of code you then integrate into the script yourself, sort of like making a quilt lol

Was I just timed out briefly without warning for saying "I only speak English" in simplified Chinese to someone as punishment for not speaking English? Lmao

[fair iron]

the language bot doesn't read context

[shy jewel]

Oh that is automated? Hmm. Might be less of a smack in the face to at least get a "please speak English only" warning at least once 😛

[fair iron]

it is automated. It picks up the language you typed in and asks you to speak English in that language. It's pretty neat!

[deft umbra]

😆

[wraith ginkgo]

`` Server Tip: ``

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

[wraith ginkgo]

This is a Safe for Work Discord Server

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

Posting NSFW content will result in a permanent ban.

[granite warren]

well

[livid dagger]

my gpt4 plugin store is blank,Has anyone ever had this problem?I tried changing the browser and clearing the cookies but nothing worked

[hallow owl]

are you using a VPN?

[livid dagger]

YES

[hallow owl]

that's why, disable the VPN and you will get plugins

[wet oracle]

Can you send pictures to chat gpt?

[hallow owl]

no

[long linden]

Are the capabilities of gpt-4-0314 and gpt-4-0613 different in terms other than speed?

[glacial zealot]

do we now have unlimeted responses with gbt 4 becuase im not seeing any caps for the number of responses

[mental prairie]

I am trying to modify a code in gpt 4, and at some point it repeats the same thing infinitely until it collapses. Any idea why?

[hallow owl]

adjust your prompt, how long is the code? Can you share the conversation by posting the shared conversation link?