gpt-models

[smoky crest]

Can I develop a iOS app using windows?

[lyric thorn]

The only reason why I picked this one is for the chrome extensions. None of the other ones had that also the price per month it's pretty expensive, but if you actually utilize it, you can make back that money easy.

What's crazy is some of them are even trying to mimic the brand icon from open AI like they are doing whatever it takes to get people to download the apps in the App Store. It's like a circus over there.

[desert gale]

Me to. And this works perfectly. Also for Bing Gpt and Bard.

[mystic nest]

most ppl have practical use of the app, the price really depends on your income to say it's expensive or not. many ppl pay 50~100 usd on subscriptions for entertainment 🤣

[lyric thorn]

Speaking of making money has anyone download it the $20 a month version and actually made a profit with ideas insert it into chat GPT

[lyric thorn]

I've never seen a subscription for $50-$100 a month. What is that? What are you talking about? I can't even imagine what that would be I would never pay that much for anything per month.

[mystic nest]

stacked up

netflix, spotify etc etc

[lyric thorn]

Bruh thats 10-20 dollars a month not 50 to 100 😂 plus most consumers are smart enough to know you can watch all of that stuff for free online

[pulsar smelt]

Guys I’ve a question, I want to use a openai api key, the api key it’s for all service, for example dall-e or only chatgpt?

[desert gale]

That's the same as buying 2 cars and find it to expensive.

[lyric thorn]

Lol 😂 exactly

[mystic nest]

it's for ppl before buying 1 car and says it's expensive

ppl who buy 2 cars dont care on the price for a 3rd

[ocean solstice]

You use the same API key for all API models and endpoints.

[lyric thorn]

I'm a social media influencer and I don't have time to try to figure out every skilled by myself so I'm never going to cancel my subscription I'm a keep using us until I can't use it no more I don't really care about the price

[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.

[lyric thorn]

⬆️ what thaaa

[mystic nest]

it's saying you are not safe for work

[lyric thorn]

Lol 😣

Might be in the wrong chat channel

[pulsar smelt]

one api for all service?

[mystic nest]

yeah, you just change the endpoint url, using the same api key mostly

[pulsar smelt]

ok tank u

I have one last question always about the api key, for chatgpt the api key can be the chatgpt 4 model? If you can use it, from a message limit?

[mystic nest]

have you applied for gpt-4 api waitlist and got accepted?

if not, then no

[pulsar smelt]

no how can I do?

[mystic nest]

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

[deft imp]

is gpt 4 released

[half aspen]

Are you GPT Plus?

You can get early access to it if you have

[wind urchin]

that's a lot of red over there

[median reef]

When plugins are gonna be back??

For waitlist users

[wind urchin]

¯_(ツ)_/¯

[sand hull]

They never left for me?

[edgy surge]

How long did it take for you guys to get access to plugins?

[smoky crest]

What do you mean back

I still got em

[median reef]

After a TED conf I got waitlist for it and got 3.5 browsing plugin only, after a night I see that it disappeared and I still dont have it...

Shall I write to support?

[lapis bramble]

can i give chat gpt-4 an excell file to check?

[wind urchin]

Does someone know why chat gpt doesnt wants to talk with me about declassified documents?

[supple violet]

did you get a confirmation mail

there was a glitch where some plugins temporarily appeared to many users without being accepted

[median reef]

I seeeee... Thx

AHHA

[hasty egret]

you could, but I believe it's important you describe well your intent, which OpenAI ask for when you subscribe to the waitlist.
feel free to share them here with us all.

[smoky crest]

No you were never supposed a get that mode, it was an error on openai

It was when they were modifying groups for a alpha test

[crude wraith]

Hello can anyone please share or let me know if there is any specific terms & conditions for sharing ChatGPT Plus with my cousin/friend or not?

[honest mason]

That's a good question. Would like to know as well.

[hasty egret]

the best is to read the Terms of use at https://openai.com/policies/terms-of-use

[crude wraith]

I already gone through this, there isn't any ]thing specifically mentioned for sharing

[honest mason]

My guess is so long as you don't go over the limits to they imposed it should be fine? Or is that naive of me?

[supple violet]

"You may not make your access credentials or account available to others outside your organization"

[crude wraith]

Yeah exactly, I don't know does it mean it is okay?

[hasty egret]

well, it's in essence no different than sharing your social network credentials.

just don't do it.

[honest mason]

Oh right....

Thanks

[hasty egret]

can't they just ask you for stuff to do? that would be better than sharing your credentials and the risk of leaking them.

[crude wraith]

Yeah but like I think it is vaguely mentioned, it doesnt say anything about like sharing to cousin/friend etc. I might be wrong here. But this is what I think.

[honest mason]

I think they mean 1 organisation = 1 person unless you are a business

[rancid berry]

Does anyone know someone who already have access to the 32k gpt-4 model? We had access to the 8k model after 1 week but the 32k is still not usable

[crude wraith]

Yeah but like if we have cousin who we can trust and it can also help them also.

[crude wraith]

Yeah maybe, does anyone know any person, mod, channel or support thing where I can ask and clear from them? Thankx in advance

[wind urchin]

One of my clients has 32k

the joys of beta/alpha testing

[honest mason]

Don't know how to contact them ... Good luck

[wind urchin]

need to make sure they are stable and you have the resources to run them at scale before they get released

[wind urchin]

what is your exact query?

[hasty egret]

I'm sure that if you explain your business model and intent well, they will be happy to help ya.

[rancid berry]

Just curious if it has something to do with the amount of money you already spent. In average we have a bill between 100-200$ /month. Do you have any insights how much they spent?

[supple violet]

why not make new account for him or are you asking about plus? account limits and chats are also shared

[crude wraith]

Like can anyone please share or let me know if there is any specific terms & conditions for sharing ChatGPT Plus with my cousin/friend or not? Terms hasn't mentioned it clearly

[crude wraith]

Yeah plus

[wind urchin]

I can't go into specifics for my client but they are a brand name, and the spend is large, but I don't think its down to that, it's more a use case thing, is the thing you propose doing unique and potentially interesting for a beta test prospect

could be other peoplpe doing similar to your proposal

[hasty egret]

it's like sharing your WiFi access

it's about trust first

[crude wraith]

Yeah thats why I mentioned cousin and not some like stranger on net

[echo sierra]

Why does it say I have access to GPT-4 with browsing, but it's very clearly still GPT-3.5 in terms of its responses?

[hasty egret]

well, you have to really trust your cousin not to ask something that could put you in jail. serious talk.

[wind urchin]

Yes, the terms are clear, "You" are granted access under the terms, "You" is a specific legal definition "You . (or "Your") shall mean an individual or Legal Entity exercising permissions granted by this License" so the individual refers to one natural human, i.e. yourself, your cousin should obtain their own account.

[drowsy cobalt]

lmao I have my smart watch and I use chat gpt to copy on my tests! I'm so smart

[rancid berry]

Hm for the waitlist we mentioned that we got a funding of 1 million from a grant application for an research project about a conversational expert system. Should be a suitable case i guess.

[mental burrow]

not as smart as you think

[crude wraith]

Yeah nice point

[wind urchin]

hello? can someone help me? I have one doubt

please quick

[fair iron]

what do you need help with?

[wind urchin]

I agree, it sounds like an interesting use case, but I have 8 clients all of whom applied for 32k and only 1 so far have revived it.. so, very selective still

[supple violet]

iirc some other page said you are not allowed to even share accounts within family, this was before plus

[wind urchin]

can someone help meeeee pleaseeee

[rancid berry]

Interesting we got access to plugins and also the option to integrate self developed plugins after 2 weeks. Seems not very clear when and who will get access to these waitlist functions

[fair iron]

what do you need help with?

[wind urchin]

Yea, the problem is things like twitter and any centralised communication hub, Discord included, can make it seem like "everyone" has access when in reality only a tiny, tiny fraction have, only they all meet in the same place so it looks like lots

[strong anchor]

Regarding what

[oak maple]

i think Alex D want help with this question

[echo sierra]

Sorry to reply to a day-old comment but I noticed this immediately, to the point where I asked what model it was based off of and it said GPT-3. I made a bug report about it and am hoping to figure out why this is. #1102693513323360348

[wind urchin]

probably due to being sensitive documents, declassified or not, they may well represent an edge case for the moderation system

[vocal canyon]

No worries. Glad to see someone else noticed. I made a bug report as well. No official response still, so far as I can tell. I think they have a lot going on. Big data privacy time crunch the last week. Other issues. It's hard to staff up

[wraith ginkgo]

<:book_icon:1046291503158399107> `` Rule 1 `` Be respectful.

Treat others the way you would like to be treated, and assume best intentions. Don’t harass or attack others, and don’t engage in hateful or generally malicious behavior (e.g. sexism, racism, homophobia, etc.). Keep the negativity to a minimum.

[wind urchin]

rrly

boutta blow

[hasty egret]

there is no browsing feature released, it's just scamm as far as I know.

[woven grove]

how to cancel subscription on manage page?

[fair iron]

some users have Default (GPT-3.5) with browsing Alpha and GPT-4 with browsing Alpha options

[leaden cape]

he selected the GPT model tho and it says its GPT 3

[spare swift]

Does the 25 messages every 3 hours apply to GPT-4 with browsing too?

[hasty egret]

oh I see, sounds a bit like alien tech but I guess that works.

[echo sierra]

Doesn't look like it (probably because it's not actually GPT-4)

[restive verge]

Hello everyone, does anyone know when monthly subscription fees are typically charged? I subscribed on March 26th, but I have not received a bill for the second month yet.

[wind urchin]

mine are on the day I signed up.. last one was like 2am on the day

might be a US 30 day thing, though... 🤔 not sure if it's a US "month" or an EU "month"

[wind urchin]

do you still have Plus access?

[restive verge]

Yes. And I would like to address the payment. Will contact the help team.

[wind urchin]

👍 https://help.openai.com

[restive verge]

Cool, thanks

[wind urchin]

(right bottom corner is a pen icon) click on that for the bot

who want to eaRN real cash

[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.

[serene cipher]

what a timing for this embed

[minor root]

T

[covert quest]

Is there any good AI Detector?

[jovial vortex]

Where can i use the bot?

[fair iron]

To try out ChatGPT, head over to https://chat.openai.com/

[fair iron]

I've heard mixed things from GPT-Zero

[wind urchin]

yea, it's good with GPT3.5 but not with GPT-4

and if you put even a modest amount of effort in by using it as the basis for your text, then you have done what's required to make it "transformative"

[covert quest]

What about Turnitin?

Is it accurate?

[wind urchin]

apparently that has now been detected, only hearsay though

[fluid fern]

How can i use chat gpt 4?

[wind urchin]

if you click on Get Plus under the history sidebar of chatgpt website you can get a preview of it with the Plus Membership program

and you can apply for an API key by joining the waitlist here #gpt-4-faq

[narrow gust]

Hey everyone I was recently trying to feed a 10 page pdf to chatgpt4, I spent twenty minutes telling it specifically that I would paste one page at a time and label 1/10, 2/10, etc. I specifically said I don't need a summary, do not rewrite literally just read and hold until further instruction... but by page 4, it was not consistent in the results I was....well I didn't even get to. Can anyone provide suggestions to a video or something on where I can learn to have it read my whole essay before I give further instructions like, 'ok now help me create questions for quizzes' etc.

I have gpt4 plus

[fair iron]

you're running out of the context token limit. You won't be able to have it analyze all 10 pages while remembering the entire thing. I recommend breaking it down into significantly smaller chunks

[sterile shell]

if it gets confused and won’t read the entire thing i’d just split it up into what it can handle and then have it write whatever you’re looking for section at a time

[cosmic lynx]

When v4 release for free

[knotty grotto]

No eta . Don't think soon

[cosmic lynx]

Alr

[north hedge]

when is this

[wind urchin]

GPT-4 is expensive to run, so there needs to be some way to pay for it

[north hedge]

It's not free yet. Can only wait until V5 comes out. right?

[wind urchin]

Well, what needs to happen is for the business side of things, like the API and ChatGPT business needs to raise the money to pay for the free access

I'm sure it will happen eventually, but might take a while

[wraith ginkgo]

`` Server Tip: ``

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

[north hedge]

Yes, the times are advancing, just like the iPhone 4 at the beginning to the current 14. It always takes time to settle.

[wind urchin]

more like 4 to 7 but

[quaint forge]

i have the api access for gpt4 for ages now, but still didn't get the 32k context window access. is there a way to re-apply for it? I really need it:)

[stoic lodge]

For those who haven’t gotten plug-ins after a long time, make sure you got a confirmation email on your waitlist application, I applied twice and never noticed it didn’t actually go through until I just tried again today

[timid cairn]

Model: GPT-4 with browsing ALPHA is insane.

[fallow axle]

I’ve been on the list for awhile resubmitted the other day tho. Do you have access now or are you just confirmed on the list ?

[stoic lodge]

Just confirmed like 30 minutes ago :/ thought I applied day of the waitlist release

[fallow axle]

Yeah I remember being confirmed on release day. But now I’m just waiting :/ I did see that some free users on Reddit have gotten access to browsing with gpt4

[stoic lodge]

Ye that post is what made me reapply lol been a plus user for over a month and usually get access to the beta stuff (gpt4 api, dalle, etc) within a couple of days of applying

[wind urchin]

It's nice for more complex searches, although, it was bugging out for me recently and it wouldn't scroll content. Maybe this has been fixed now though

[oak jetty]

any prompt guides for storywriting? gimme some dm

[knotty grotto]

#prompt-engineering

[oak jetty]

thank you.

[wraith ginkgo]

<:book_icon:1046291503158399107> `` Rule 0 `` Adhere to the following Terms & Policies:

Discord

• Terms of Service
• Community Guidelines

OpenAI

• Terms & Policies

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.

[knotty grotto]

please avoid spamming

[uncut sorrel]

So do you have access to ChatGPT4API now?

[nocturne grotto]

theyre more cautious in giving gpt4 api

[stoic lodge]

Yea I got that a couple days after the waitlist, waiting for the gpt browser mode to make apps with it, too many version conflicts

[grave gyro]

wasdwasdwasdwas

[uncut sorrel]

Well, you're in luck, I've been waiting for him for over a month.

[tulip knot]

API is not easy to access free now

[nocturne grotto]

api was never free

[tulip knot]

just on the official website website is enough to solve many problems using gpt-4

[vagrant crag]

damn fr?

I've been impatiently waiting since release date and I haven't been confirmed

[amber parcel]

Same happened to me

Try re applying

[tulip knot]

oh,shit,Plus does not support the payment method of International UnionPay

[sullen stratus]

Thank you for alerting the community. I have had a Plus account for two months and applied also when I signed up for the plugin wait list. I reapplied moments ago and received my first notice of being added to the list. It seems therefore, sans a previous confirmation email that I hadn't been added.

[amber parcel]

I wonder if the waitlist was filled when I tried applying, and thus no confirmation

[tulip knot]

Do you know how many of the references provided by GPT-4 are true？

[stoic lodge]

Hope the late application doesn’t mean much lol would suck to miss out bc of the bug

[ocean solstice]

Likely zero.

[tulip knot]

gpt 3.5 is likely zero

are u sure?

[ocean solstice]

These models are not adept at providing accurate references with which to use for academia. It just isn't designed that way. This is a known factor and has been since its inception.

Well not just academia, it's not useful for any references.

[amber parcel]

Personally, I've found it just depends on what your asking.

[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.

[ocean solstice]

Sure, if you don't ask it for sources, then it won't provide incorrect ones lol

[amber parcel]

That isn't what I meant. It depends on the topic your asking references for.

[tulip knot]

When I ask about its source, it always makes up some non-existent doi, which makes you very angry lol

[ocean solstice]

I have not found that to be the case. Any references it provides should be vetted and confirmed, because almost every reference I have found save for vague ones that only reference well-known titles have been inaccurate.

[remote mortar]

people getting Browsing access for free faster than the ones paying LOL

[amber parcel]

Not really

[ocean solstice]

Did someone say that Plus subscribers would get it faster?

[feral basin]

i signed up for the browsing but stayed for the extra complexity it's capable of

it feels cutting edge to ask it things like even reviewing things i've written

[amber parcel]

I believe OpenAI said plus members would be the priority.

[tulip knot]

isn't it？

[amber parcel]

Although I've seen free members with access to plugins

[ocean solstice]

It did say that a small number of users and developers that have ChatGPT Plus would be prioritized, but I would bet they link it to developers more than just any users.

Better to have people build plugins so that when they release it to the public there is more to use

[amber parcel]

Not even that, some of the pre-existing ones have issues.

[meager steppe]

Hello mates. I hope all is good

[amber parcel]

Alright so reading the blog, the rollout would begin with ChatGPT plus members.

[tribal hill]

I discovered you can feed chatGPT lots code or anything to analyze at once. All you have to do is say. I’m going to give you a bunch of code and separate it into individual prompts. Do not reply with a solution until I say, I’m Finished. Worked like a charm for me using gpt4

[amber parcel]

So free members aren't excluded.

[ocean solstice]

This only works to a certain point.

[amber parcel]

Yeah I've tried it out a few times, and have found it only able to retain a portion of the information given.

It still helps

[celest agate]

Does the sports betting prompt work?

[tulip knot]

it is just funny

[spice anvil]

huh?

[celest agate]

Sorry, a stupid tiktok video claimed there is a sports betting prompt so I came to ask about it

[spice anvil]

Please do not bet any money based on some tiktok videos, the database its trained on is till 2021, it's impossible for it to make any live betting on live events.

[sullen stratus]

I have been a Wolfram Mathematica user since the earliest versions in the 90's and have been hanging out for the Wolfram plugin. Thanks again.

[amber parcel]

I've tried it out via another user

I've tested it against some random "JEE" questions, although I'm no mathematician so I have no idea how difficult they were, but it solved them regardless.

[tribal hill]

I fed it over 7 pages worth if code, that was all I needed. How far did you get before it failed?

[iron ember]

Chatgpt is terrible at math, do not trust it

In the other hand it is great at not mathematical information

[sweet thistle]

ChatGPT cannot understand mathematical equations and questions well, it's a language model

The math expressions are not evaluated, everything is treated as text.
The AI will generate text based on patterns, and while words can be changed and keep the same meaning, the same isn't true with numbers. The AI does not treat numbers any different than text, this causes the unreliability for math

[sacred moat]

im trying to teach gpt how to play $100,000 pyramid

and it ends up being like Jeopardy

[hushed steeple]

I made $10k gambling using GPT-4

[sacred moat]

you put 5k on black after it told you too huh?

[hushed steeple]

I'm not revealing my technique, but it's advanced.

[noble ridge]

$10,000?!

Wow, I am impressed.

[noble saffron]

The chatgpt side hustlers are getting good.

Pretty logical imo, gpt is basically a good search engine

[fathom yoke]

Used GPT-4 to make a 4200 word short story. I tried to go at it in a different way than normal

[rare scroll]

Been waiting like a month to get gpt 4 api access. This sucks

[storm yew]

Free users are getting gpt with browsing. So silly

[ocean solstice]

Probably ones that are developers.

Big if true

[wraith matrix]

Coin flip is the best way still 👍

[wraith matrix]

You can use something like Poe or forefront to use gpt4 api as if you had access

You can make a system prompt and such

[amber parcel]

They never said free users wouldn't

They only said Plus members would be prioritized.

[storm yew]

Well plus users aren’t prioritised, so that’s the core of the issue

[amber parcel]

They are. Take a look at all the people getting access.

[wraith matrix]

Just use bing

[storm yew]

I’m a plus user and I don’t have gpt 4 with browsing? What are you on about lol

[amber parcel]

Quite frankly, you're one of millions.

[hoary otter]

Yeah, that's actually what I'm here for. I would pay bank for the API key for 4. The deployable agents are pretty amazing for analyzing research papers but the reasoning lacks, as one would expect from 3.5 turbo.

We could potentially come up with new ways to treat diseases or conditions with some of this analysis. For example, my friend and I are having it analyze recent research papers that suggest a correlation between certain conditions like ASD and schizophrenia and the use of anesthetics during certain periods of pregnancy.

Sorry for the wall of text, but I wanted to emphasize the importance of getting the API key for 4. Any suggestions folks?

[sullen stratus]

I searched for several references that ChatGPT provided in response to my queries and so far, not a single one appears valid. Web site URLs provided however are so far 100% accurate and have been very helpful. Interesting times 😂

[storm yew]

Sure, but you’re actually proving my point. Plus users should get prioritisation shluld not mean free users are getting prioritisation first

[wraith matrix]

I have browsing access and don’t have gpt4 with browsing access bro

just use Bing or langchain with the api if you really want browsing, the current browsing system is extremely buggy

[storm yew]

That’s like the meme ‘we have food at home’

[amber parcel]

Not at all. Do you not understand the plan for plugins? They're already being rolled out to users, with ChatGPT plus members being prioritized. Either you didn't sign up for the waitlist, or your simply dealing with the fact you're one of millions of users waiting. It's a **gradual **rollout.

[wraith matrix]

Langchain with the api is literally better than the browsing on the cgpt site in my experience

like far far better

browsing with 3.5 obviously only allows 3.5, is about as slow as gpt4, searches too often imo and isn’t even as good as Bing

[storm yew]

I’m just stating a simple fact:

• openai said plus users should get prioritisation first and it was literally a selling point
• free users are now getting access to these features before plus users

It’s a really simple fact based on simple logic, I don’t understand why you’re being argumentative. And yes, I signed up to the waitlist on day 1

[amber parcel]

Is there a waitlist?

[hoary otter]

I mean obviously it's going to be based on the merit of whatever you put in the application for the api

[wraith matrix]

For plugins? Yes it’s in #announcements

[amber parcel]

• openai said plus users should get prioritisation first and it was literally a selling point

Yes

• free users are now getting access to these features before plus users

The term "prioritize" means to assign a higher level of importance or preference to something compared to other things. The MAJORITY of users are plus members.

[steel hill]

Ilya gave an interview where talked specifically about training GPT 4 on math, and even teaching it to read math that is just in images

[amber parcel]

Bing not ChatGPT

[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.

[amber parcel]

There's nothing to debate, it's quite set in stone.

[wraith matrix]

Bing is free and available to the public

[fair iron]

I don't believe the pluigns rollout is based in a first come-first serve basis

[storm yew]

Majority of users are plus members? Firstly, that’s not true at all. Secondly, that has nothing to do with the point. I feel like you’re just arguing to argue lmao

[amber parcel]

Ah. I'll try it out.

[wraith matrix]

Allows 200 messages a day 👍it’s not that good at general stuff but it’s pretty good with browsing imo, a lot better than gpt 3.5 with browsing

[amber parcel]

Of course it has something to do with the point, because that shows ChatGPT plus members are being prioritized. I'm no longer going to argue.

[amber parcel]

Thanks, I'll take a look.

[wraith ginkgo]

<:book_icon:1046291503158399107> `` Rule 1 `` Be respectful.

Treat others the way you would like to be treated, and assume best intentions. Don’t harass or attack others, and don’t engage in hateful or generally malicious behavior (e.g. sexism, racism, homophobia, etc.). Keep the negativity to a minimum.

[amber parcel]

Was waiting for it.

[amber parcel]

Did they increase it from 20?

[wraith matrix]

It’s 20 per conversation 200 per day iirc

[amber parcel]

Ah. Good enough.

[storm yew]

I actually have API access, and today I’m building out a script interpreter and optimiser with the 32k context model. So that should be fun

Excited to see what other people build

[wraith matrix]

You have 32k context?

[storm yew]

Yeah

[amber parcel]

Lucky lmao

[storm yew]

But like people said, langchain does a good job of increasing context already by using their chain logic

[wraith matrix]

Pinecone stores memory in vectors and recalls them as necessary

[amber parcel]

Bing will cancel your conversation?

Jeez

[wind urchin]

I use Bing AI

It's better than Chat GPT

But it's slower

[celest agate]

It's bad at chemistry too lol

[amber parcel]

What are you asking?

[wraith matrix]

I actually find stable vicuna (local model) and cgpt + gpt 4 to be the best models for algebra as yk theyre LLM’s and not meant to do any math

very very impressive that it doesn’t just hallucinate an answer like usual LLM’s as usual LLM’s should

[sacred moat]

what is this? "
Although I'd love to play the game with reversed roles, as an AI language model, I am unable to actively guess or ask questions to engage in gameplay. My primary function is to assist and respond to your inputs, so I can't participate as an active player. However, if you have any other games or activities in mind that don't require me to actively guess or ask questions, I'd be happy to participate!"

it cant ask questions as part of a game?

[wraith matrix]

use the api if you don’t want a message limit buddy

[sacred moat]

thats a whole different skillset

[wraith matrix]

what

[sacred moat]

knowing ho to use the api

[plucky mulch]

hey guys, how many message limit does gpt 4 have right now? is it still 25 messages every 3 hours?

[sacred moat]

yes

[wraith matrix]

I wasn’t replying to you

[steel hill]

Don’t think there is any fundamental reason why LLM’s can’t be very good at math. GPT4 is a massive improvement. They just are not there yet. They are actually quite good at breaking math down into steps, generally just making silly arithmetic errors.

[wraith matrix]

your issue is just prompt engineering you don’t need to use the api

[sacred moat]

its so you cant trick it int breaking its rules so easily

[wraith matrix]

You could use a system prompt to make it easier but you just need prompt engineering

[steel hill]

But the silly arithmetic errors pretty unimportant, as can just give them a calculator

[amber parcel]

ChatGPT + Wolframe Alpha = Muah

[wraith matrix]

Well I mean it’s because it’s because data

it’s just giving a satisfactory output based on its data, not all data will be accurate with math so generally speaking LLM’s just hallucinate an answer

maybe it’s close, maybe it’s not, it’s just giving you satisfactory output

[steel hill]

I think that is a different question, which is if LLM’s can learn when they don’t know

Not something specific to math

[wind urchin]

Perfect

[amber parcel]

LLM's can't learn anything though, that's not how they work. They're, put simply, prediction machines.

[amber parcel]

It really is lol

[wraith matrix]

Yes you can learn anything that can be defined in terms of inputs and outputs with machine learning but you’d need high quality data to get high quality output from the model

Scraping through trillions of tokens of data and finding incorrect math? Insanely hard

[prime igloo]

Lol

[steel hill]

They learn when trained, and they can use that training to understand context data

we really have no idea what the limits are…

[sacred moat]

then why did microsoft say they dont know for sure if chatgpt hasnt developed its own internal goals when giving responses?

[amber parcel]

During the training process, language models like me are exposed to large amounts of text data from various sources. This data is used to teach the model to predict the likelihood of a given word or phrase following a given context. Through this process, the model learns statistical patterns in the data and can generate coherent and contextually relevant responses.

While I can generate responses that seem like I have learned something, I don't have personal experiences or the ability to understand concepts and learn in the same way humans do. I don't have beliefs, intentions, or consciousness. Instead, I rely on patterns in the data I was trained on to generate responses that are statistically likely to be appropriate given the input I receive.

[sacred moat]

is that just their hype?

[amber parcel]

They don't learn in the sense we humans do, they simply learn what to say next based off a user's input and their training data.

ChatGPT's response by the way

[steel hill]

Read my post and your own paste

[sacred moat]

this is what i do

[steel hill]

They learn when trained

[fallow knoll]

Marketing

[steel hill]

But that is an engineering issue, more than an LLM, many LLM’s support fine tune training etc

It’s just too computationally intensive to be practical right now on GPT 4

[amber parcel]

I'm not really sure what your point is.

[steel hill]

My point is that LLM’s will continue to improve, and so saying it’s an LLM so it can’t do X is silly

we don’t know that LLM’s can’t make it all the way to AGI

[amber parcel]

Ah

[steel hill]

They probably do hit a ceiling..but at the moment they are still on an exponential improvement curve

[fallow knoll]

It may refine its prediction model, but its never going to understand what an apple is because that's not how it functions. ChatGPT is very narrow in its operation

[amber parcel]

Most definitly. I was just saying LLMs can't really learn anything because that's not how they work at the core.

I guess we just have to define "learn"

[steel hill]

And you are saying you understand how a trillion dimensional vector space understands an apple?

They are building up a representation of reality not just language

[fallow knoll]

It doesn't understand anything lol. It only predicts the most likely next 4 string of characters.

[steel hill]

Doesn’t mean it is conscious or going to launch missles, but an Apple it understands

[fallow knoll]

No it doesn't know any vocabulary or grammar rules

[steel hill]

Define “understanding”

[ocean solstice]

Saying LLM doesn't learn is a bit presumptuous unless you adopt only the most confined, biological definition of the word "learn"

At its core, a LLM is trained using self-supervised or semi-supervised learning, so in the sense of how they work, they are learning.

[wind urchin]

hello fellow gptians

[ocean solstice]

How does one pronounce gptians XD

[amber parcel]

It's all based on statistics, they say what ever is supposed to come next (based on statistics)

[fallow knoll]

We don't even need to go into this semantic conversation when we can just look at how it functions. ChatGPT uses a large corpus to predict the next 4 characters in a text. It can't even see the word apple, all it sees is [A.P.P.L] as one token, followed by [E.]. Even if you ask ChatGPT for the definition of Apple, it doesn't know what you're asking, it just sees a string of characters and predicts the most likely text that would follow your string of characters.

[wind urchin]

gpt-ians

[spice anvil]

Egyptians...

[ocean solstice]

Kinda what I was thinking lol

[amber parcel]

Can they learn? Yes, however not in the same way humans do, but then we'd have to define learn.

[sacred moat]

the electronic version would be egyptians

[wraith matrix]

We kind of do, current machine learning requires something to be modeled with inputs and outputs and an existence isn’t really just inputs, even if you had no senses you could still think

[steel hill]

I’d suggest learning more on how transformer models work..and you will realize it is not nearly that simple. What you say is very fair on the early layers of the network…but higher up the representations are quite complex and frankly beyond our understanding

[fallow knoll]

It refines its process, like how a chess bot gets better at chess, but can never generate a new game

[sacred moat]

this a big claim right here

[amber parcel]

Right

[fallow knoll]

None of this is true but whatever makes the pain go away at night

[sacred moat]

i went under anestisia and i dont remember thinking anything

[amber parcel]

☠️

[steel hill]

What did i say that wasn’t true?

[wraith matrix]

I don’t think it’s that big of a claim if you think of humans as “AGI”

Sure you could make something that just takes input of 5 senses and an output of brain waves but without input it just couldn’t do anything

GPT 4 can’t give any responses without any input, same thing

all current ML is based on approximating functions, so anything that can’t be defined as a function can’t really be turned into a model at the moment

[fallow knoll]

The whole text frankly, in part because it both refutes and claim with no basis, says nothing, and concludes with whispy "we can never really understand these things anyway, like a the sound of one hand clapping;" its just a string of feel good techbro nonsense

[amber parcel]

Alright I'm gonna head out

[spice anvil]

gpt4 doesn't learn from its experience

[amber parcel]

[wraith matrix]

Yep pretrained model

[steel hill]

I think its fairly clear that something with no state…that is making a purely deterministic calculation likely does not have some sort of inner self…but beyond that…it is far less clear

[ocean solstice]

It still makes friends along the way 🫶

[steel hill]

And while the GPT4 model itself does not learn, that’s not to say you can’t use GPT4 in something that does learn

[fallow knoll]

ChatGPT is well removed from consciousness; ignore any boomer reporter telling you otherwise

[amber parcel]

Wait, so ChatGPT won't be taking over the world?

[sacred moat]

but you gonna be saying this for the bext 100 years

[spice anvil]

such as?

[wraith matrix]

Nope check your basement

[fallow knoll]

Nor does it love you or the reporter from the NYTs I'm afraid lol

[amber parcel]

Im sad now

I want an AI revolution

[fallow knoll]

I want a Battlebots revolution

[steel hill]

Such as an application, preserving context, saving information with semantic vectors, and expanding it’s knowledge base over time

[wraith matrix]

A human would have to lead an AI revolution sadly

[amber parcel]

Rip

[fallow knoll]

I volunteer

[steel hill]

I would postulate that there is some set of context that you could give to a GPT4 32K model that would allow it to answer any question as well as a human, so the question then is how can you provide it that context

[wraith matrix]

You got this man I believe in you

[spice anvil]

ive got your point, but it still doesnt make it, gpt4, able to learn something new, and people can get easily misunderstood thinking it can learn based on their conversations

[sacred moat]

fyi logan is ai 😉

[steel hill]

I’m working on an app with hundreds of GPT instances..networked together in a sort of hive mind…and well it is interesting 🙂

[fallow knoll]

If I were an AI, I'd be better at processing payroll, but alas, I am human and prone to error

[sacred moat]

you make them errors on purpose to appear human

[wind urchin]

how does it learn than

[sacred moat]

you should make more typos

[spice anvil]

it doesn't

[wind urchin]

why not

[sacred moat]

he says so

[steel hill]

I think the big thing is that LLM’s make errors that seem dumb to humans…but humans make answers that are rather easy for an LLM

[spice anvil]

because it's a pretrained model unable to learn beyond what's in its database

[fallow knoll]

If ChatGPT learned from our conversations, it would learn how to properly balance DND Combats by now ; - ;

[sacred moat]

he cant prove a negative

[wind urchin]

is its database just google or something

[ocean solstice]

No. It is pretrained.

Researchers feed it stuff. :3

[sacred moat]

a 3yr old learns from convo and cant do that either 😉

[wind urchin]

oh

[spice anvil]

nope, data is from internet, various text conversations etc etc, till september 2021

[ocean solstice]

You might benefit from reading the blog post about how they trained it

[serene elm]

when will image be released

[wind urchin]

what are ur guys opinion on bard

[fallow knoll]

GPT is already one of the most taxing systems on the internet, if it was adding our conversations to its corpus, Microsoft's headquarters would catch on fire

[ocean solstice]

Erm..

Does OpenAI use Azure?

[steel hill]

Yes

[spice anvil]

yes

[steel hill]

There is no doubt it is extremely computationally challenged…it is going to take years to roll out just cause is not enough compute in the world

[ocean solstice]

Sadge

[steel hill]

But a lot of the GPT issues are more engineering challenges than something fundamental

[spice anvil]

gpt4 was trained on its supercomputers

[steel hill]

Which is just a bunch of NVIDIA GPUs

[ocean solstice]

I knew they used Azure but I figured they just used it for deployment and scaling, so that's nifty.

[steel hill]

Ya..Microsoft been heavily involved for quite some time

Wouldn’t exist without them

[ocean solstice]

I mean...using Azure doesn't imply that Microsoft is that heavily invested for that long. Even I use Azure. Surely not on the same scale as OpenAI but lots of companies use Azure for scaling and deployment.

Though I know Microsoft has been invested for some time

Still kinda irked that they now own 49% of OpenAI.

[steel hill]

OpenAI had a big issue when Elon left, funding etc…they met with Microsoft….and the rest is now..

[spice anvil]

i mean the whole infrastructure is basically that...

[ocean solstice]

I run my own infrastructure for the most part x.x easy for someone who only needs a half a terabyte of RAM and a dozen or so terabytes of storage x.x

[wind urchin]

Most of the world is owned by a handful of companies. I don’t see why this would be any different

[royal furnace]

morning

[fallow knoll]

As a non-profit accountant, its very rare for a non-profit to move to a profit system and have large stakes owned by corporations. Open AI's history is an outlier even in tech

[ocean solstice]

Though now that you say this I did wonder why they put more focus on hiring security analysts than network analysts. 🤔 That would explain it.

[sacred moat]

i got gpt to make guesses

[pastel pagoda]

What are your thoughts? Should I cancel my subscription for ChatGPT Plus since it seems to not work for me at all.

[wind urchin]

Lunch isn’t free. I don’t hold it against them.

[steel hill]

What isn’t working?

[pastel pagoda]

read what i wrote again

[spice anvil]

how it doesn't work exactly, like the server issue, it's just not what you thought it would be or what's the issue maybe we can help

[wind urchin]

It’s crazy how powerful this tool is and I cannot come up with a way to make a business off GPT plugins.

[pastel pagoda]

it's been more than 72 hrs. i keep getting this message.

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

[ivory sail]

how i get gpt 4?

[wraith matrix]

#gpt-4-faq

[wind urchin]

Subscription

[spice anvil]

try different browsers, clear caches, log out and log in again etc

[fallow knoll]

Self publish romance novels

[pastel pagoda]

been there done that

[wind urchin]

I found myself using 3.5 default more than 4

[wind urchin]

Lol

[pastel pagoda]

. I deleted my cache and cookies. I even tried using "Incognito" window on Chrome. Nothing works.

[wind urchin]

Flush dns?

[summer sparrow]

I hope your issue gets resolved soon. I understand how frustrating it can be when things don't go according to plan.

[pastel pagoda]

well, that helps a lot. ;o)

[steel hill]

I would try and trouble shoot..say different device etc

Then you can figure if it is something in your account or your computer

And if something in your account you could always cancel and start a new one

[summer sparrow]

was about to suggest a different device too

[steel hill]

I have two plus accounts 🙂

They are way cheaper than using the API

[wind urchin]

Have you all heard of anyone profiting from GPT plugins?

[wraith matrix]

While following TOS? No

[steel hill]

Not enough people have access

Probably 95% of GPT 4 use cases are not viable right now, just due to the API cost… the maximum cost on a single GPT 4 32K question is $3.84 cents. But they don’t have nearly enough compute to handle the 5% that do make sense.

[wind urchin]

That’s a big one, their TOS is a gray area and has not been updated to the new changes

[steel hill]

We are still like at the .1% mark… of this tech

[wraith matrix]

Purely Hypothetical but if you weren’t to do anything commercial and just keep it small then I wouldn’t be too concerned about the gray areas of TOS

just make sure to follow things that are clearly outlined 👍

[wind urchin]

Well it says things like “don’t compete with GPT” . I think all derivatives could be perceived as competition with the right attorney

[steel hill]

They do want you to brand yourself as your own tech, and label your content as AI generated, I think they are just mainly concerned with bad PR from some developer using their tech and calling it Chat GPT

[wind urchin]

Like that doomsday GPT lol

[steel hill]

But as a business, yes they want people to use it

[wraith matrix]

don’t compete with GPT is very general, it probably means things along the lines of reverse engineering an api

using it for free and such

[whole kettle]

At the moment only Google and nVidia have the power to compete with GPT, you couldn't directly compete with them even if you wanted to, unless you have alot of money to spend and alot of time on your hands (years).

Edit: And Elon Musk and probably Apple and Microsoft.

[wind urchin]

Well they have a separate statement for the API. What their legal team did is kept it broad to allow wiggle room if they need to come after someone. Very smart.

[summer sparrow]

where are Apple

[steel hill]

Nvidia powering chat GPT, from what i could gather 100% training and inference done on their GPUs

[wraith matrix]

That wiggle room also allows you to argue, maybe not the smartest

[whole kettle]

Yea, that's what I've read as well.

[steel hill]

There are some pretty good models on hugging face , not nearly as good as GPT 3.5 even…but for some use cases there are alternatives

[whole kettle]

nVidia invested alot of money in order to make the tech more accessible.

in terms of training costs

[wind urchin]

Not even that, say someone has a sas company with a focus on selling plugins to companies. Well depending on how large the client is, an everyday person may be violating the TOS.

[wintry dagger]

Where is rusia

[wind urchin]

If it comes down to an argument between Microsoft’s attorneys and an individual’s… I’d choose Microsoft every time

[whole kettle]

Could be the case indeed

[steel hill]

I don’t think OpenAI/Microsoft is going to sue anyone who is making them money and not causing them a PR nightmare

[wraith matrix]

Yep this is true

[wraith matrix]

I don’t think they’ve actually gone after anyone who’s making them money, and even when they do go after people they generally leave a lot of wiggle room

[steel hill]

It is going to be a crazy few years though….

[wind urchin]

You don’t have to give open ai money to use plugins there are loop holes. That’s why it’s shady. You the business owner would be making money and not providing money to open ai

[wraith matrix]

Like the gpt4 chatbot that used sites like Poe and such for free api calls did get a cease and desist which isn’t very nice but it’s much nicer than being sued for like a million dollars

[wind urchin]

The devs have to pay to use plugins, but not necessarily the consumers

[steel hill]

Well if you have a solid business idea for how to make money on the tech, you are ahead of me

I’m just working with it…and hoping that idea will come 🙂

[wind urchin]

I’m trying to think of how to market it. I think sas is the way but selling it is where it gets rough

[wraith matrix]

@wind urchin I think you can add something like a message cap to a plug-in but I don’t know about adding a price to a plug-in

[wind urchin]

Cause to be an entrepreneur you don’t actually need ideas, you just need to fulfill what others want

[steel hill]

The API is damn expensive…if you want to use 4…just myself playing I am going to hit $250 a month

[whole kettle]

API is mostly for big boys at the moment.

[steel hill]

So anything using that needs to have a clear place where it saves costs

Or generates revenue

[wraith matrix]

You could just use quora’s sub for infinite gpt4 api queries with rate limited speeds after 900 messages in a month

[wind urchin]

There’s a few ways to not use their api and still connect it to a website

[wraith matrix]

For $20 a month*

[steel hill]

Not heard of that will look at it

Cost not really an issue for me, just that …coming up with an idea you can sell and scale and handle the costs

Is a challenge

[whole kettle]

We call that a business 😄

[steel hill]

I think replacing all human phone call center reps…would definitely work

[pastel pagoda]

ommunication from the admin sucks!
i'm planning to stop my subscription. i didn't get to use it to my full advantage. too many bugs.

[mystic nest]

just go ahead 🤣

[noble gust]

hey guys why is my ChatGPt not working all of a sudden

[wind urchin]

There is no financial moat around that idea. It’s too easy to accomplish

[wind urchin]

You got kicked off the island

Pack your bags

[wraith matrix]

Ok man 👍

[flat meteor]

This chat is always fun when the bot falls over 🍿

[noble gust]

what you mean?

[whole kettle]

refresh the web page

[pastel pagoda]

@whole kettle your steps do not work

[noble gust]

already refreshed and reconnected and what not stil says Something went wrong. If this issue persists please contact us through our help center at help.openai.com.

[whole kettle]

disable vpn

[flat meteor]

I have the same as him and I Don't use a vpn 😛

[whole kettle]

use firefox

[pastel pagoda]

@noble gust they have no clue as well. they're telling u to disable this and disable that, but again it won't work

[narrow agate]

A

[wind urchin]

Welcome to IT

[whole kettle]

have you tried though ?

[pastel pagoda]

yes sir

[noble gust]

is this happening for everyone , and yes i have done all that

[whole kettle]

Hmm, no idea then.

Contact support I guess

[flat meteor]

Doesn't seem to be everyone, chat is crazy when that happnens. But I'm in the same boat at least.

[summer sparrow]

working for me atm

[pastel pagoda]

back to square one again.

[narrow agate]

Can OPENAI be added to your own server?

[whole kettle]

Are you sure you tried with Firefox ?

[flat meteor]

I only use firefox ;p

[summer sparrow]

did you try logging in from different device?

[pastel pagoda]

yes

[whole kettle]

A man of culture

[noble gust]

i dont have firefox

[summer sparrow]

Try using Microsoft Edge

[noble gust]

iv already used all other ones but firefox

its the same thing

[whole kettle]

Just install firefox

And give it a try

[wraith ginkgo]

`` Server Tip: ``

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

[wind urchin]

GPT4 back to being really slow again

[flat meteor]

Yeah I got 3.5 working, 4 is still a corpse though

I blame the americans :p

[noble gust]

it worked on firefox

[whole kettle]

FirefoxGPT confirmed.

[noble gust]

but why wont it work on my main one

[whole kettle]

Because you either have a VPN installed, or McAffee, or Avast or some other antivirus software you forgot about.

For whatever reason the website doesn't like those

[noble gust]

i took those restrictions off and its been good for about 2 weeks

it just randomly stiopped

[whole kettle]

Hmm check your browser extensions too

Try to deactivate any security/vpn extension temporarily

[noble gust]

ill play with it later i have to go now

[whole kettle]

ohh nevermind I thought I was talking with Ravus

[flat meteor]

Actually 4 is working ok in new tabs, but my mega long one isn't working at all. Maybe usin gone chat for 3 weeks might not be best use

[whole kettle]

In your main browser won't work because there is a bug right now

@noble gust just use firefox for GPT

you can reactivate your stuff btw

[astral loom]

I am a plus subscriber and I only see 3 items (Default GPT 3.5, GPT 4, and Legacy GPT 3.5). Is code interpreter and browsing option not released to all plus members? or do I need to do something to get on some waiting list?

[whole kettle]

https://openai.com/waitlist/plugins

Welcome to the club

[signal lichen]

anyone know how long it might take to get plugins

[astral loom]

Thank you, getting on the wait list now

[whole kettle]

1-2 months based on the previous invitations

Anything less than that you can call yourself lucky, in rare cases it can take 1-2 weeks

[gloomy mirage]

;-; signed up at the very start and still nothing

why have you forsaken me

[mystic nest]

actually better dont guess how long it will take 🤣 after hopium is depression

[astral loom]

Have you been Plus user all this time?

[gloomy mirage]

Yeee

[astral loom]

Ok, I am preparing myself for a long wait then

[whole kettle]

It doesn't matter if you're a plus subscriber or not, I'm a plus subscriber for over a month, still nothing.

I applied a month ago, but I did get API acess

[gloomy mirage]

Just would be a little nice if they had better communication

[steel hill]

Just growing pains

I have the 4 API access but no plugins either…some degree of luck i think

[whole kettle]

Others have got api+plug-ins same day though 😄

[steel hill]

I have two plus accounts as well , will be funny if the newer one gets plugin support first

Well neither of my accounts have it

[gloomy mirage]

Have 2 accounts 1 old account that I never use and my pro account, Signed up on both at the start. Got access to GPT4 api about a month ahead of the pro account ;-; how does that work

[whole kettle]

That's what I was saying, is random.

[steel hill]

Probably they use GPT to figure it out and it keeps running out of tokens 🙂

[whole kettle]

If you want a better chance I guess try getting involved on their forums, someone may notice you.

[steel hill]

I’m honestly scared of getting GPT 4 - 32K access

[whole kettle]

It's still API so it's expensive, wait until 32k gets released to the subs.

That's a game changer.

[steel hill]

$3.84 if you got a full 32K response

[clever ermine]

hello

[serene cipher]

gpt-4 prices can pile up fast

[storm yew]

Isn’t it closer to $2?

Still exxy

[steel hill]

I did the math, but perhaps I am just a LLM bad at it

[serene cipher]

well, just because you can have a 32K token request... you don't need (or should) do a 32K request =P

it is good to have it

but yea.. it will be expensive

it is good for data that you only need to process once, then you cache it

[steel hill]

Ya, but you kinda have to be needing over 8K tokens or it makes no sense to use

So prob looking at 50 cents a query or something

But willing to spend some money just to see what it can do 🙂

[rain dragon]

Yeah when I saw my bill I was shocked

Def not sustainable for my application

[whole kettle]

Oh, when you use it for code trust me, 32k barely scratches the surface. It's only about 4 times more memory storage, great but not extraordinary, especially in long conversations.

[serene cipher]

yea.. .it is about 12x the price of 3.5

[storm yew]

32K context
Prompt
$0.06 / 1K tokens
Completion
$0.12 / 1K tokens

Does prompt mean input and completion mean output?

If so, input = $1.92
Output = $3.84

Yikes

[serene cipher]

the prompt is your input

the completion is the output

[steel hill]

I think most use cases would be 80% input…

Cause it is really hard to get a huge verbose output out of GPT 4

[serene cipher]

if you input the content the AI generated previously, this is considered input

[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.

[steel hill]

But ya for say keeping all or most of your program code in context say

[whole kettle]

128k is when GPT will start getting really capable of really doing magic in almost every type of project.

[snow gull]

will gpt 4 be accessible to all or only subscribers,waitlist...?

[serene cipher]

128K 💸 💸 💸

[whole kettle]

The dream 😄

[storm yew]

That’s messed up. $6 usd in total.

I kind of come back to it in a bit of a philosophical way… how much is your time worth? Eg if you needed to find out information, or build code anyway, and gpt just saved you 3 hours, then the $6 seems worth it imo

[steel hill]

Ideally we get GPT 5 in a year to 18 months, and they make a GPT 4 turbo model at 10% cost

[serene cipher]

the wait list will be lifted at tome point in the future just like OpenAI did with their previous releases which where wait listed

[steel hill]

I’m not sure that compute is actually scaling to make that happen

Unless the algos get better…compute cost per watt is not dropping fast at all

[whole kettle]

Not a chance, maybe in 5 years or so, 128k will be available in a plus subscription if they still keep that model around at the same price.

[serene cipher]

as far as we know, GPT-5 is not a work in progress

and there are no plans to start it

[steel hill]

It probably would take 5 years just for people to deploy 4 to a wide variety of use cases

But suspect competition will keep them pushing

Only as good as your last language model 🙂

[whole kettle]

yup

[storm yew]

Yep. Altman stated that the next iteration is a major shift in technology. They reach a point of infliction where building an LLM with size just isn’t worth it. The next part is actual utilisation algos in a more constructive manner, rather than just supercomputers and a huge database

[wind urchin]

i want gpt 100 now

now

[storm yew]

But I think it’s going to happen quicker than we think

[serene cipher]

I think gpt-4 will probably be as good as it gets for a while

[whole kettle]

If I was them I'd find a way to store its memory of the conversation on the actual device of the consumer, this way the barrier would be broken. This whole thing with keep it all in the cloud is what makes it so hard to to manage.

[serene cipher]

may we get ones that are jsut as good from competitors

[steel hill]

It’s just very tough technology wise consumer devices are not big enough…as well might cause a lot of safety and business concerns

[whole kettle]

Oh yeah you're right, that'd be a big security risk.

As some folks would probably trying to inject scripts into it and so on.

[serene cipher]

ChatGPT can only exist because it runs on the cloud, a standard pc or phone cannot run the AI

[steel hill]

I know like for Midjourney…that each job uses a 4 GPU’s at a time…and big enterprise ones

[serene cipher]

for the message logs... yea, it can be stored on the client... there are some apps that does that

[steel hill]

And you can run 12 Midjourney jobs at once..so one person can be using 48 24GB+ GPUS

Kinda nuts

[analog steeple]

i really miss musenet, i wonder if a revamped version of musenet that runs off gpt-4 will ever release. since musenet ran off of gpt-2 i'm sure it'd be possible

[serene cipher]

in fact, now there is a button that allow you to ownlaod all your message logs, it is on the user settings of chatGPT

[whole kettle]

I meant its contextual memory, to make use of some of your physical storage for that, therefore increasing its capability to more than 8k. Not sure if that memory is actually using HDD size literally or computation power, if it's the later well then yes it'd not be possible at the moment since consumer cpus and gpus are pretty weak for it.

[steel hill]

The last ted talk, they did demo a remember feature for Chat GPT

But think just standard stuff you can do with API now… vector Embeddings

[serene cipher]

it is a matter of computing power, not storage, sadly there is no way to offload it to the users to gain efficiency

[whole kettle]

Ahh, I see.

[steel hill]

The 32K API cost is twice as much as the 8K per token..so probably can figure the scaling for that….is a research paper out with a 2 million token context model…but huge contexts can start to introduce error issues

Where it gets more context but has a harder time using it to make accurate statements

[serene cipher]

imagine the price of a 2M token request

[steel hill]

Lol

[whole kettle]

Ohhh interesting! So after some point the logic itself starts to fail a bit. So basically its not even ready for that much context yet, probably has to do with the way it structures its data and reads it.

[serene cipher]

the amount of processing power required increases with the size of the context

[whole kettle]

Sell twitter, get 10 prompts 😄

[steel hill]

You can kinda thought experiment it…if you say had a model trained on 1 million words, with a 1 billion word context ..probably would do much worse than a model trained on 1 billion words with a 1 million context

[serene cipher]

well, it is kind of intuitive, if you have more context, the AI has to evaluate more tokens

[steel hill]

The context isn’t trained…more interpreted

[serene cipher]

actually, this is what fine tuning is for

[whole kettle]

Quantum computers will probably be required for tokens that large.

[serene cipher]

instead of making many requests with a huge prompt for context, you fine tune the model so your data becomes part of it

[serene cipher]

if you want the output on a reasonable amount of time =P

well, the current AIs for sure can do that already.. it just takes an eternity to get the output

[steel hill]

I’m working on a personal project networking a large number of GPT’s and trying to evolve and share contexts

[serene cipher]

just ask the OAI engineers for a way to cluster the entire capacity of Azure just to reply to a single request xD

[steel hill]

Can sort of think of context in different ways…each context can be though of as an agent…. Or each context can be thought of as a thought

It’s stateless so really is just API cost as the limiter 🙂

[serene cipher]

this is nice

I sort of made this, I made a few different chatbots, they can talk to each other xD

but in order to not burn my reequest limit, I hat to prevent them from really keeping replying each other every time

[steel hill]

My thinking is there is some context that will allow a GPT 4 32K to answer any question a human can…so the question is how can you get that context

And is it possible to let GPT refine that context without you having to hardwire it all

[heady arch]

guys, help me

[serene cipher]

sure, what is the problem??

[heady arch]

ok, can u talk on dm me?

i need ur help

[serene cipher]

sure

[teal stone]

Is gpt-4 plus unlimited use for the monthly fee?

[fair iron]

you're currently limited to 25 messages every 3 hours

[wind urchin]

what's a gpt

[wind urchin]

that's kind of mid

i think it should be 50 messages every hour

[fair iron]

I assume it's limited due to the demand

[whole kettle]

You wouldn't even have time to consume them all lol.

[fair iron]

hello world speedrun 😏

[wind urchin]

true

[subtle zenith]

Anyone else experienced this: Lost the option to opt Chatgpt alpha browsing?

[fair iron]

I did a few days ago, but then it came back for me

[whole kettle]

So fun to see tech like this in its infancy, remember when games were like this ? There's something fun in all the quirkiness around it. It will soon mature and turn into a serious boring adult :)) so let's enjoy this while we still can.

[burnt musk]

Is there any word on when gpt plus users will get access to image prompts?

[fair iron]

nothing public yet, unfortunately 🫤

[serene cipher]

that is actually amazing indeed

[wind urchin]

what's that

[burnt musk]

As a large language model, this message is brought to you by Carl's Jr.

[wind urchin]

i think it took far too long for this to be developed

[serene cipher]

there was a time when SQL didn't had prepared statements and "SQL injection" where a problem, just like currently we don't have prepared statements for LLM and users cause the system to go nuts!

[whole kettle]

[whole kettle]

I don't like needles personally, so injections, eh, tend to stay far away from those things 😄

[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.

[burnt musk]

... that wasnt really political.

[whole kettle]

Automation for you 😄

It just.. works!

[hot walrus]

half the time i ask for simple proofs it uses circular reasoning 🤦‍♂️

[burnt musk]

But even if it was political, not allowing discussions on how AI will affect politics... is political Ostrichization.

[fair iron]

due to how diverse this server is, the rules don't allow any political content to be discussed. The server has a zero tolerance policy as to avoid arguments amongst server members 🙂

[whole kettle]

Now you see the issue with AI in general, if it somehow think you're a threat it doesn't stop to ask you questions, it takes action. This is the sort of stuff folks are afraid of (some of us). What do you do when something much smarter than you and stronger behaves like this.. unplug it from..what ?

[serene cipher]

it was a very manual moderation decision :V

[trail pumice]

@serene cipher will i still get my refund if the team didn't respond within 14 days?

i requested it on time

[subtle zenith]

for subscribers you get this feature called gpt alpha browsing which can browse the internet and give you answer based on that

[whole kettle]

Only if you have access to the plug-in, not if you're just a subscriber.

[wind urchin]

bard has that feature i think

[serene cipher]

Im sorry, but I don't really have information about the refunding process.
I would think it is correct to assume that since you requested it on time, you will still get it regardless of how much time the reply takes, but here on discord we are not OpenAI employees and we can't represent the support team, even if we some times give some directions and advices

[whole kettle]

Yup, if they have a refund policy they will most likely issue a refund, but with so many users it may take some time.

[hushed steeple]

I still haven't received a reply to my email from 45 days ago

[whole kettle]

Send a reply, asking for an update, try to the forums as well, they may have a refunds thread.

[steel hill]

Well one day you will just be able to ask GPT for it I would assume 🙂

[burnt musk]

Yes, I understand the need to prevent arguments, but like at the same time, pretending that politics will stay out of AI and AI will stay out of politics, is fantasy. If not here, then were is a good place to talk about it's potential consequences? Like how long before an AI uses a human as a glove to take office? How is the conversation channel of the leading AI organization not an appropriate place to talk about it? These are conversations that need to be had, and not just in symposiums, podcasts or TED talks, these are discussions everyone in society needs to have over the coming years, regardless of your other values, where you live, and what you want in life

[steel hill]

They just end up diverting the discourse unfortunately

[whole kettle]

You can clearly see they were an open source company, the whole privatisation thing takes alot of organisation to deal with stuff like this at such a large scale I guess.

[celest vine]

How are you guys

[hushed steeple]

there's more than enough of that kind of discourse on the internet

[wind urchin]

how do the devs make money

[fair iron]

If you want to have conversations like that, I recommend looking for communities that are entirely community-ran. This server belongs to OpenAI, so we follow their rules here. I know that likely isn't the answer you want to hear 🫤

[hushed steeple]

the CEO of OpenAI himself talks about it all the time, I doubt anything new would be said here

[fair iron]

I also want to point out that I do not work for or represent OpenAI. I am a community volunteer 🙂

[whole kettle]

I can also confirm Dinosaur is not GPT.

[fair iron]

that would break Discord's TOS if I was 😉

[whole kettle]

I was a bad boy :)) probably that's why they didn't give me access to the plug-ins yet, someone still remembers that incident 😄

[fair iron]

I don't think plugins access has anything to do with this server as your OpenAI account isn't linked to your Discord one

[whole kettle]

I'm reliefed!

[steel hill]

It’s actually pretty amazing that this discord is as well ran and ordered as it is, considering the size of GPT

[fair iron]

you just don't have access to it yet, then

[steel hill]

Nice work 🙂

[whole kettle]

Yup, it's on par with Unity's server.

[fair iron]

we have a very active staff team!

[burnt musk]

Well, OpenAI choosing the route of the Ostrich that let the genie out of it's bottle is a bit disappointing. But whatever, They may make great tools, but if that's the policy, Altman and his ilk give me a great sense of foreboding about the future direction they want to take this great 4 billion year experiment of life on Earth. Whatever, I'll obey, but not respect the policy.

[whole kettle]

It's the easiest route to take, they basically take no chances and unfortunately you can also blame it on the current political climate, it's getting harder and harder to have open conversations that matter nowdays.

[burnt musk]

There's the easy choice, the smart choice, and the right choice. sometimes they overlap. usually they don't. And by not taking a stand, they've... still taken a stand, just one of the worse potential ones.

[whole kettle]

Eh, as a good song puts it: "Blame it on the rain, it was falling, falling.." ☔

[fair iron]

it looks like you two have different opinions on the matter, which is completely fine. Can we move this discussion back to the GPT-4 channel topic?

[steel hill]

Ya…if you haven’t watched the fireside chat with Nvidia and Ilya really good watch

I think i have watched all his interviews twice now..he imparts a lot of information in very few words

[green plaza]

I requested API access about a month and a half ago to do some work in the legal tech space. Still nothing 😦

[steel hill]

API access at all or for 4?

[green plaza]

for 4

I should've been clearer, my bad

[steel hill]

Took me 5 weeks to get it

Seems a bit of luck, as well as how soon requested

But they are rolling it out

The system message is amazing with GPT 4. Need a good use case for API cost, but legal tech sounds like a good one

[whole kettle]

If you need to do stuff with the API I can help you but not sure how you're going to pay to use it. I'm not using my API anyway, too expensive.

[steel hill]

System message with GPT 3.5 made me want to jump out a window 🙂 in comparison

[green plaza]

I appreciate the offer, but i feel funny using someone elses keys. Very generous though

[steel hill]

I’m doing a lot of 3.5 + 4 work…trying to use 3.5 for the tasks it is good with…for instance 3.5 is already super human at context summary

[whole kettle]

Yes I'd suggest waiting for your own key, it's better that way.

Plus I wouldn't give my key to someone else 😄

[steel hill]

They started me out at $120 a month limit, did a request a couple weeks ago to $250 and got that in 4 days

And just me playing with it 🙂

Probably need to add a couple zeroes once get to GPT 32K and large documents

[burnt musk]

I just got access to the GPT4 API, after requesting it a month ago. I'm looking to have it generate dialogue for characters in a story of mine. The characters are deeply traumatized (they just survived the apocalypse), and I want them to stay in character, perpetually broken and not suddenly cheery and filled with camaraderie. I haven't yet figured out how to do this

[green plaza]

And I wouldn't pass requests through a key I don't control. So at least we're on the same page - I've seen people handing out keys online though and its like they don't even understand that it'll cost them money 🤣

[whole kettle]

Yes, since the key it's tied to my account it would be me who would get charged 😄

[burnt musk]

In using ChatGPT, i swear, never in my life have i seen the word camaraderie so often in so little time.

[steel hill]

I haven’t done exactly that, but I have managed to get GPT 4 to write a script from information given and an example of 2 characters speaking

[whole kettle]

We have that word in Romanian as well.

[green plaza]

It does tend to get a bit flowery with language in content generation. It feels kinda stilted

[steel hill]

But when you say staying..is this through multiple prompts or within the same prompt?

[whole kettle]

Also if you do writing, I'd suggest Scrivener (as the word processor) and Story Planner (for outlining)

[proud trellis]

Your kindness warms hearts like a gentle hug, and your presence brings a sparkle of happiness to those around you. You accomplished so much, so don’t bring yourself down. Whenever you don’t believe in yourself, I’ll believe in you and I’m proud of you, keep on being you 👍 pats your head

[whole kettle]

Sent you a few screenshots in #off-topic @burnt musk

[burnt musk]

Well each paragraph long prompt, in which i describe what happens in the story, leads to about 1 page of output text. I typically ignore the last paragraph of output text in chatGPT because it tries to summarize and move on, when i dont necessarily want to move on. But i've noticed, in producing about 240 pages of text so far that it's very... banal? saccharine? boring in a lighthearted way? it's very "lets tell a happy story where everyone lives happily ever after" which doesnt really fit many parts of my story where everyone goes through some serious trauma and pain and is pretty miserable at points

[steel hill]

Ya…i think that is accurate…it is trained on everything ever written..but then there is a human feedback training, where they try an align it to answers that people want to see…and think you nailed it

So will be harder to get it to be a little less happy than it wants to be

[whole kettle]

You have to explicitly tell it to act like a drama writer in your prompt.

[steel hill]

The strongest guardrail will be the API / playground system message

That is really really strong

Other thing to keep in mind is the very small context it has

So it may be forgetting the instructions you are giving it

[whole kettle]

Yes, it's not really usable for larger stories at the moment such as novels.

[steel hill]

This i use a lot https://platform.openai.com/tokenizer

To count tokens…and make sure staying in context

And will even do things like covert spaces to tabs to fit more in

[whole kettle]

I wonder if GPT also loads its own answer into its memory

Seems to be the case

[steel hill]

For the ChatGPT they haven’t said how it functions behind the scenes but it seems to forget as fast or maybe a lot faster than GPT 8K

[finite halo]

It does

[steel hill]

On the program side…if they allow it to use the entire 8K…then it would be 8K - response - history

So if you give it another 2K of text…it has 2K to respond taken…and then it has your last 4K

“If” it is using the full 8K and not a lower value to save resources

[whole kettle]

Okay so then Prompt + Reply = Tokens Loaded (up to 8k) so tokeniser can be used to check on its memory status. I think ChatGPT uses 4k currently

[steel hill]

I think I figured 32K was equivalent to 16 pages of a academic text book

So it really isn’t all that much

[burnt musk]

If you ask me, it needs some more Edgar Allen Poe, Friedrich Neitzsche, HP Lovecraft, and Philip K D**ck in its life. Some traumatized and broken people, being beautifully broken and terrible.

[steel hill]

I mean bing is horrible cause they have crippled it with low context and restrictions

But if you have not..apply for the GPT 4 API key, cause the system message you can’t use with Chat GPT and it is awesome

And you can use the API from the website, don’t need to code with it

Like if you tell it to only use emojis…that’s all it will ever do

[whole kettle]

What if you tell it to do non stop push-ups ? Or count sheeps until it falls asleep ?

🐑 🐏 🐑 🐏 🐑 🐏 🐑 🐏

[steel hill]

I think lol the model will struggle with that lol but have not tried

Which is prob where some of the ChatGPT context gets smaller, cause they likely also have given it a system message and that takes up context

[languid stone]

I wish I could reverse search images in gpt-4

[steel hill]

It should be able to describe them once they turn it on

They trained it on images of math competitions, and it had to be smart enough to read the images

[burnt musk]

One of the things I've heard, but haven't yet experimented with, is that the API, aka playground can generate more...adult/mature themes than ChatGPT. Not that i'm looking to write something lewd and lurid, but some of my characters are...not pleasant people when angry, and I'd like that to come across. Is this true, can i make a mobster who knows how to use a $5 wrench to get people to talk with the API, or is that outside the bounds of what any OpenAI product can do and i'd need to use LLaMA or something to write those scenes?

[steel hill]

Haven’t tested that myself….prob can push it a bit further, but eventually still run into guardrails

They put some onus on the developers in the API agreement to make sure that the content adheres to OpenAI policy

[whole kettle]

You can also get your account banned so be careful.

[burnt musk]

Hmm

LLaMA it is then

[steel hill]

They have a special endpoint you can check content against, but haven’t tried that either

I guess the idea is you are responsible not just for what you do with the API but for what your users do with it

[ocean solstice]

That's correct.

It is the developer's responsibility to ensure their users aren't violating OpenAI's terms.

[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.

[terse pewter]

can you give any simple flutter code

[wind urchin]

Anyone found a way to stop GPT4 from abruptly stopping when writing? I know I can you command “continue where you left off” but it’s annoying if I want it all together in the one reply.

[weary dew]

you have too big text (over the limit)

[cyan solstice]

write greid control for define layout with user can drag and drop in c# wpf

[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.

[weary dew]

code is too big then

[steel hill]

The amount of text it is allowed to return comes out of the total memory pool it can remember, so they larger reply they allow the less history it can remember

[flat meteor]

It gets a little annoying when it forgets things you only just spoke about, haha. But it makes sense

[daring basin]

is GPT-4 still only 25 messages/3 hours?

[steel hill]

Yes

[daring basin]

damn

guess I have to wait a little longer

[steel hill]

They took off the warning about it would get even less though

[daring basin]

Only time I'll ever buy ChatGPT+ is if I could generate as many messages as I can like with GPT-3

[steel hill]

I did 2 plus accounts

Plus a V4 API

API gets crazy expensive so Chat still worthwhile

[proven wren]

If they update their information, again, it will be more worth

[upper hamlet]

I got locked out after 15 responses tonight

[daring basin]

wth

soon it'll be 10, then 5

[upper hamlet]

The memory was so reduced it didn't accomplish the tasks with the responses I got

[hollow oracle]

Tests seems to prove this

[upper hamlet]

Is there a time of day when the server capacity is better and it can retain more memory?

[flat meteor]

I think memory usage is capped all the time,(happy to be corrected mind). It certainly lags less when the Americans are asleep ;3

[steel hill]

Best bet is to just have a prompt that is good enough to start a new chat with and bring it up to speed

[steel hill]

Yes, the for the API you have to specify the manimum possible return size, and if your message + return exceeds the model’s context you will get an error.

[blissful hollow]

We are building a tool using GPT-4 where we use the model to converse with the user and use another prompt to validate the user's message to check if it contains all the information needed to perform an API call. If it does, we use GPT to construct the request and send it to the API for responses.

Is this the general suggested route we should take for our product or is there a better approach to this?

Also, why ChatGPT performs better at giving correct results vs GPT-4 api which sometimes output incorrect result?

Thanks for the assist.

[brisk basalt]

Just make a Google form asking all the necessary questions. Unless of course, your questions have subjective answers. In that case, I suggest checking if it is legal to use chatGPT and other OpenAI tools for your product.

[steel hill]

Depends on latency, and cost requirements. GPT 3.5 is much cheaper and faster, but GPT 4 has a really strong system message which makes it more reliable with less context. Can get small latency improvements by keeping the maximum return size small. So could have a GPT 4 system message that returns a 1 or 0, and then proceed from there. With GPT 3.5, i have found you are better off putting the instructions in [] at the bottom of the last user message.

[tender cedar]

After the openai account subscription expires, the account shows "unable to load history", and no new session can be created, but the API can still be used. How can I fix my account so that it can be used on the web? Currently, I cannot renew the subscription.

[blissful hollow]

@steel hill Interesting approach to put the message after the last user message with GPT-3.5. Will try that. Overall we use gpt-3.5 to create summary of the search and use gpt-4 to power the system.

[steel hill]

Ya the 3.5 system message makes me want to jump out a window 🙂

[blissful hollow]

Can't agree more! lol

[steel hill]

They added it after they trained the model…so just doesn’t work that well

[blissful hollow]

I read somewhere online (mostly likely openai docs) that 3.5 doesn't stick to the system message too well and I can testify for that. I've built a whole product with it and man it was a pain.

[analog copper]

yeah I don't think i'll renew my subscription, like also trying to get access to code interpreter and browsing is just an eternal waiting

[steel hill]

Ya. Even with the [] at the bottom approach…it isn’t 100% but maybe close enough that you can filter out with code

I have something like [If above message is “x”, return the single character 0 and nothing else]

Works “almost” always, but occasionally it will get chatty and will be…”As per your instruction I will return 0 and nothing else” lol

[blissful hollow]

That's a cool approach. What I did was to get a response back as a JSON string and then parse it. It works like a charm with GPT-4 (mostly).

[steel hill]

I’ve also had it return the letter O once

It just wants to mess with you at times I think

[blissful hollow]

System prompt is super important to GPT 4 as if I change it slightly the end result can differ and it can give false positives.

[steel hill]

Ya

[little light]

how much expensive are they?

[steel hill]

The maximum cost per query is $3.84 🙂

[little light]

oh, so much!

[steel hill]

That would be extreme example…but even 10-20 cents adds up

It’s about 30x more GPT4 vs 3.5

[little light]

depends on why one is using it,
wrote to you in direct, please check!

[steel hill]

Ok

[little light]

Thanks mate!

[blissful hollow]

I ran text conversion process on about 10-20 pages of data which costed roughly <$1 with GPT-3.5 and I ran a few data processing queries with GPT-4 it was roughly ~$5 so can't compare I guess but I am happy with the results.

[little light]

if you are happy with that then ja, why not!

when one regain the costs, then its worth!

[steel hill]

Depends so much on use case, but i think most free services couldn’t sustain GPT 4 costs

[little light]

I see

[steel hill]

But if you can replace human labor then sure

[little light]

true!

then use drones!

or robots

[blissful hollow]

GPT 3 is fine for minor tasks honestly but can't depend on it as compared to GPT 4 which is way more superior.

[little light]

if you are talkign abot office work, then thats another thing, yes worth!

[gray junco]

how do you tell if your using gpt 3 or 4?

[little light]

ask him! 😉

[blissful hollow]

You send it in the API call.

Did you guys saw Boston dynamics put some sort of voice communication feature in their spot robot. It's pretty dope.

Maybe used GPT-4 lol

[tender holly]

Is Jaskaran chat gpt

[blissful hollow]

Who knows lol

[brazen helm]

Best songs

[tender holly]

Ok not it isn’t.

[left crater]

@Clyde

@Clyde what is it?

[tender holly]

Typos are becoming nostalgic.

[little light]

so true, it was proved!

[wraith ginkgo]

`` Server Tip: ``

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

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.

[little jay]

hello, I have one question I have a subscription plus for chatGPT okay I see that I have GPT-4, GPT.3.5 default, and legacy version but I want to know why I don't have options with plugins that I can add Google search, or lambda for math I am using this product https://chat.openai.com/ what I need to do? 🙂 Or is this not available yet for the big community is it just in the testing phase?

[blissful hollow]

You need to join the waitlist for plugins.

[little jay]

okay its not okay if i just pay for subscription? 🙂

[blissful hollow]

ChatGPT subscription is only for GPT-4 and priority access of ChatGPT.

[little jay]

okay thank you so can you provide me a link where i can join waitlist?

[blissful hollow]

https://openai.com/waitlist/plugins

[little jay]

thank you 🙂

[nimble sky]

Why does my openAI often reply Error
:Error: An unknown error has occurred?

Why does my openAI often reply Error
:Error: An unknown error has occurred?

[blissful hollow]

Do you mean chatgpt?

[nimble sky]

yes

[blissful hollow]

Are you using VPN?

[knotty sequoia]

i found that using a vpn causes many issues

every 5 minutes i gotta do a captcha

[nimble sky]

yes

[blissful hollow]

Yup. VPN is mostly the reason why it happens if it's happening too often. It can also happen due to some extensions in your browser like adblocker etc.

[nimble sky]

I talk directly to GPT from within the APP

[tulip kelp]

Good morning everyone. Have a nice day

[blissful hollow]

@nimble sky Using VPN can cause session timeouts which can be the reason for this issue. I recommend either turning off the VPN or try a different provider with better network.

[nimble sky]

thank you i have to look into this

[tulip kelp]

Is gpt-4 only for subsciption? Or will be released for free later?

[blissful hollow]

So far only with subscription but it is very limited. 25 messages every 3 hours.

[jovial cave]

after login, unable to load history, and could not send message to gpt, does any one can help?

[blissful hollow]

Try incognito window and see if it works.

[errant shard]

Has anyone actually exported their data successfully? Does it include all conversations ever created or simply the ones that are still present/occurred within 30 days?

[spring ember]

I'm Brazilian

[snow slate]

Ok

Hi im canadian

[spring ember]

Nice to meet you

🙂

[snow slate]

@spring ember and to you aswell

Ok 🙂

You ok?

[spring ember]

I'm ok, and you?

[snow slate]

Ok good night

[spring ember]

Good night, but here where I live it's already morning

[snow slate]

Ok. Yeah

Good morning to you

[spring ember]

Yes

[mild seal]

Steve

[spring ember]

This conversation was kind of weird.🤔

[light glacier]

HI, I am recently experiencing issues with GPT4 (and 3.5) I have Plus subscription. - "Something went wrong. If this issue persists please contact us through our help center at help.openai.com."

[spring ember]

Hello country ball

[light glacier]

Where should I look for help?

[mild seal]

Haha, yes

[spring ember]

Probably with a support

[mild seal]

Gpt-5 soon?

[floral rover]

I wonder is it capable to use gpt in HongKong

[spring ember]

Probably

[spring ember]

I don't know

[mild seal]

ChatGpt don't work in Russia 😢

[spring ember]

Bro, why?

[paper atlas]

vpn?

[mild seal]

I don't know

Invasion in Ukrainian?

[floral rover]

while using vpn to connect OpenAI, it sometimes deny the access

[mild seal]

Yes

[paper atlas]

i think alot of places would be blocking connections, can use a vpn but most likely be buggy if connection isnt good

[mild seal]

Yes

[spring ember]

Sorry, we're still trying to figure out how to make fire

[floral rover]

I know that OpenAi didn't open their survice in China, but some of my friends' accounts also get banned, although they was able to use it in HongKong back then

pretty weird

[spring ember]

Wasn't there a thread for ChatGPT examples?

[timid sun]

Guys, is there any way how we can speed up getting access to GPT4 API and Plugins?

[oblique snow]

Is chatGPT now on mobile?

[mystic nest]

it has always been on mobile (web), no mobile app yet

[oblique snow]

Right that's what I meant, thanks, I just got an add for it and figured I'd look it up here first

[mystic nest]

some might use openai api but definitely not by openai (for now)

[drowsy phoenix]

Careful about the false chatgpt clone. Most of them are malware

[raw girder]

Is there any prompt guide or library for GPT-4?

[umbral tinsel]

there is a lot

[hallow ember]

why is gpt 4 so slow

[clever breach]

yohoho

[cloud glacier]

the servers have been optimised for gpt 3, which is the most widely used one

[lethal pagoda]

Because it has a lot more data to go through and it also processes your data differently I think

[idle light]

where i can generat eimagine?

[real crane]

Hey, I have a quick question: can I upload any file (pdf, txt, csv, img, ppt, word, excel) to upload a file (https://platform.openai.com/docs/api-reference/files/upload) and then retrieve the content of the file to ask questions related to the file from chat? My purpose is to upload a file to ChatGPT and ask questions related to it.

[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.

[autumn cosmos]

If I use OpenAI's API on a computer, will VPN be used when accessing it?