gpt-models

[slender stump]

I just registered and I don't know how to use it

[wind urchin]

Bro it's easy

[slender stump]

Is this just for chatting?

[leaden cape]

you type in your questions and hit enter

[solar heart]

weird i only use chatgpt 3 times a day

Cuz its good for fast answers

[leaden cape]

25 messages/3h is for GPT 4 not 3

[slender stump]

what are you doing here

[waxen walrus]

i see mamy technicians use chatgpt to do a lot of things,but i can just use it to chat with me.

[weary patio]

Everytime I go to a different tab and go back to the prompt, it tells me an error and I have to refresh the page. Is anyone else getting this problem?

[knotty grotto]

Try Firefox as test

[weary patio]

That's what I'm using

[waxen walrus]

maybe you can open the seamless mode or clean your cookies.

[weary patio]

🤔 hmm okay, I'll see about that.

[slender stump]

Are you communicating using chatgpt here

[floral stream]

why is the ui bugging

[waxen walrus]

This is a group about communicating the problems we encounter when we use the chatgpt

[knotty grotto]

Can you try edge with incognito

[surreal tangle]

whate do you know about mohamed salah

[wind urchin]

W pfp

[knotty grotto]

Without vpn

[floral stream]

lol thanks you too

[plucky rampart]

#rules

[slender stump]

I see, thank you

[wind urchin]

Hey, does any of you know a good prompt for gpt4 programming tasks?

I had one but I lost it 😭

[knotty grotto]

#1019652163640762428 can be helpful

[fickle perch]

Still getting this error all the time, only on my computer any browser - any window (private or not) Something went wrong. If this issue persists please contact us through our help center at help.openai.com.

[fresh egret]

Hello, can you tell me exactly how I should use this?

[wind urchin]

couldn't find one to my liking there unfortunately

[floral stream]

to access gpt 4 i need the chatgpt plus right

[weary patio]

I'll try that soon and let you know

[wind urchin]

or access to api

but in chatgpt yes

[floral stream]

how to get access

[knotty grotto]

Yes

[wind urchin]

join the waitlist, but its also paid.

[floral stream]

nobody speaks arab

[pulsar wyvern]

Update, GPT4 is remembering swift files between prompts better than it was before through the site (not playground/API) and GPT3.5 is back to being silly instead of a super memory LLM (I say that jokingly). Thanks to anyone who offered insight.

[drowsy mango]

I can barely use ChatGPT 4 today, it keeps on triggerring my 3 hour limit after barely two messages in. What is going on.

[surreal tangle]

hallo

[wraith ginkgo]

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

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

[wind urchin]

Does anyone here have gpt-4 api access and isn't using it?

[supple violet]

api is a paid service and gpt4 access is tied to user account

[surreal tangle]

hallo

[crimson pewter]

no access to gpt-4 even with api

paid

we are all beta testers and paying for it

[cloud junco]

What do you guys even use GPT-4 for?

[jade wyvern]

damn.. its almost 1 month on gpt-4 waitlist :>

[cloud junco]

when did you join the wailist?

[jade wyvern]

april 4

[cloud junco]

i mean how long after gpt4 was rekeased

[jade wyvern]

probably less then a week since march 31 was when gpt-4 was announced

[rapid mist]

Hu

[blazing heath]

i having problem of it saying "something when wrong" , i try delete cookie, close all extension, change to firefox or edge, all do not work

[worldly flower]

@

[cloud junco]

i joined second day and still not in

[worldly flower]

#gpt-models What are the days of China's May Day holiday in 2023

[fickle geyser]

How to create images her in Open AI ?

[golden idol]

barely know her 🤣

[wide elm]

how does this work?

Combined midjourney

#gpt-models How to write the template of film and television commentary copy

[quasi marsh]

is it correct that the last update made to Chat GPT-4 dates back to last March 23rd?

[thorny marsh]

hioi

howareyou

[bold wave]

Did you loose your space bar?

[indigo radish]

having same issue openai icon just continues spin and I have a plus account

[jade wyvern]

no

their data dates back on sept 2021

gpt-4 is just finetuned gpt-3.

[bold wave]

It’s been refined, the data is updated

[stark grail]

🤣

[cloud bison]

Just out of curiosity, does anyone have access to fine tune GPT 3.5 or GPT 4 other then corporate partners?
I know some people have "Playground" access to them, and thus can access their API's, but the guidance documentation does state that fine tuning is still currently limited to Davinci, Curie, Babbage, and Ada. Thus, I'm guessing access hasn't been extended to the better models, and that probably OpenAI does the tuning on behalf of companies that they are partnered with (Since its very obvious that some plugins are not just using system prompts, but actually tuned, and on GPT 4).

[slate dragon]

@wraith ginkgo drop some updates come on

[cloud bison]

The government (I'm talking about the US government, but obviously a lot of world governments too are starting to hit the company with lawsuits, regulations, and bans, so "the government" is an accurate term in general) is currently practicing a lot of judgement calls that are "technically" not post-ex-facto judgement, but amount to the same thing in many ways. I'm guessing they're continuing internal development, but stopping all external deployment for the time being until they get a gauge of how some of the pending lawsuits and legislation are going to fall.

[slate dragon]

At least give access to plugins 🥹

[wraith ginkgo]

`` Server Tip: ``

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

[calm sinew]

Hi new here, I recently upgraded to Plus and now whenever I log in to chat I get a spinning wheel over the chat box and no chat history. This is only happening on Chrome, Safari works. Tried everything from deleting cookies to restarting the browser and computer. Nothing works.

[patent schooner]

Greetings, I have learned that GBT can now read links. I am using GBT to summarize PDF articles. However, I have purchased a book in PDF format and downloaded it. Where can I post this PDF so GBT will read this as a link and create outputs for me. google drive does not seem to work. Any thoughts and advice would be super helpful, thank you

[supple violet]

there is no link support

[fallow ore]

is anyone else being constantly disconnected from web app? It seems like I have to reconnection every 20 to 30 mins

[jade wyvern]

yep

It seems to be caused by cloudflare's security stuff though.

[mystic nest]

there is no internet access for chatgpt, but if happen that your url is part of the training data, it might recognise it and provide the info

[jade wyvern]

you can only submit some parts of your conversation when you click the thumbs up and thumbs down buttons, it will then be used for analysis to fine tune the AI.

[calm lynx]

i cannot access chatgpt from pc why

[graceful berry]

Can i upload an image to chat gpt 4 ? Cause i can't but there is on demo video.

[supple violet]

no

[cloud bison]

They have advertised its multi-modality, but haven't released it to the public yet, and as far as I know, also haven't released it to any of the common well known wishlist groups. Possibly corporate partners, but its also plausible they're still training it to reduce the token cost of responding to images and videos.

[patent forge]

do you guys can teach me how to register gpt ?

😫

[cloud bison]

edgy name... red flags galore.

[cedar dust]

same issue

[patent forge]

my ip seems to be targeted

[stable wadi]

so, you pay to be on a waitlist without access to chatgpt 4?

[graceful berry]

Is it always slow?

about gpt 4

[quasi lance]

Is chatgpt available on discord

[sacred glacier]

What is the Estimated date for release of GPT - 4 to test out

[cloud bison]

Playground access is not "paid" it is per token, but, those who pay the monthly subscription to Chat GPT get access to GPT 4 chatbot with a 4k token limit, no system prompts, no tuning, et cetera, and if they apply to the waitlist for API access, are "generally" considered before others who apply to the waitlist and are not paying a subscription.

This is not the fastest way to be selected though. The fastest way for selection is to file a report finding a non-prompt injection bug in an OpenAI API, or to get an organizational ID from a Research University, and apply to do research that is related to AI ethics under your program.

People who were subscribed and joined the waitlist very early, the day it became available, managed to get approval to the GPT 4 API, but most who subscribe now are so far down the waitlist that they will never get approved before other much higher priority people who are actually involved in helping the development of the platform.

[wind urchin]

Why cannot openai provide service for people in china

Im just curious

[tardy heart]

if you host a discord bot with your api key sure

[quasi lance]

??

[mystic nest]

it's China censorship laws blocking chatgpt, not the opposite. Like Italy

[wind urchin]

no, but using the api is paid.

[cloud bison]

The CCP blocked OpenAI, OpenAI abides by the wishes of any government that blocks it by also creating blocks from it going the other way.

[wind urchin]

Then why didnot provide service for people in hongkong

[tiny birch]

@unreal lagoon

[cloud bison]

Its still possible to bypass it with a non-mainstream custom VPN, which can be made using a server in another nation used for some other purpose. You will need to avoid the major server companies however, as it will only realistically work if you find examples that are not on a common list of blocked IP addresses. In theory you could also rent out a space, and hire someone to construct a really cheap auto-forwarding computer and route a custom VPN through that if OpenAI ever gets so extensive in its blocking that it even locks down all small time local server hosting operations.

There isn't an effective way to recognize if a custom VPN service using a non-public VPN made by you is actually a VPN or not, so if you ever want gaurenteed access to a network, that is probably the best way to do it. Problem is, you're dealing with setting up a service, and if the person you ask in the country to set it up for you realizes the value of the service, they can expand it, and risk compromising your IP Address due to the activities of others sharing the service.

The people who have the most secure world wide access that I know of essentially own multiple properties and create custom VPN network connections to all their properties in various countries, allowing them private access to anything regardless of national restrictions with very little chance of ever being caught... Not really a luxury affordable to most common people though (Unless you have a really good friend living overseas willing to let you route through their internet).

[oblique wren]

would it be worth it though?

[lone solar]

Thanks for that @cloud bison

[wind urchin]

I have one

[mystic nest]

TLDR;
1: Bypassing network restrictions is possible with a non-mainstream custom VPN, but it requires finding unblocked IP addresses and may not work in the future.

2: Custom VPNs are hard to detect, and they provide guaranteed access to a network but can be risky if others share the service.

3: Owning multiple properties and creating custom VPN connections can give secure worldwide access but is not affordable for most people.
🤣

[cloud bison]

Political expedience I would imagine.

I think they changed to a "whitelist" protocol instead of a "blacklist" protocol to spare the issues they were having legally. Its easier to just whitelist countries interested in the service, versus regularly having to deal with countries filing lawsuits and complaints and perpetually updating a blacklist.

My guess is formal requests from governance in HK haven't requested the services (This will usually be businesses wanting to use the service in the country with lobbies, not people themselves asking for it).

[ionic osprey]

how tf does chatgpt work like how would you event start programming it🤣

[wind urchin]

Where are yoi from dude. Are you american ？are you a student？

[elfin crescent]

Computerphile on YouTube has some great breakdowns as to how LLM and transformer models work

[stable wadi]

man, we are here at this time, and i havent even learned how LCD tv screens work. its just magic

[ionic osprey]

cheers👍

[mystic nest]

it's probably good to learn basic data science models then NLP before you jump into chatgpt 🤣 but well, most ppl wouldn't have the patient

[cloud bison]

The exact method is complicated. An inaccurate but convenient way to explain it is a combination of the "neural network" and "evolutionary selection" approach (This is not how it does it, but it is convenient for explaining roughly the way these things work).

You start with a set of parameter's (The neural network), its a list of nodes, arranged in layers, every node in every layer connects to every node in the previous and next layers. The topology includes a LOT of layers, and a LOT of nodes per layer.

You give it a text string, and it is supposed to "output" the "next word" (See token) in the text string. It makes a guess, and when it guesses closer to correct, we reward it a "little" (Gradient ascent), when it gets further from correct we punish it a little (Gradient descent). We iterate this billions and billions of times and eventually it starts giving you words that make sense, and with enough prompts and enough completions (Words before and words after), eventually, it is what we have today.

(Missing a TON of extra steps).

[rare dove]

hii

[ionic osprey]

damn mustve took ages to make

[cloud bison]

It takes a lot of data, because humans don't tell it what is closer to correct or incorrect exactly, rather, it was taken from the collective works of the internet basically. It does take a long time because it takes lots of parallel processing to account for the many nodes in each layer iteratively feeding back and modifying one another's weights, but, not as much time as you might think, its less about time, more about datacenter size, electricity, and number of TPUs (The type of silicon used for this processing unless you're being even more inefficient and using GPUs).

[ionic osprey]

so what is it hosted on?

[cloud bison]

My guess would be Azure Web Hosting via Microsoft.

[ionic osprey]

👍

[deep cape]

What are the best things to use gpt-4 for? I only have 10 prompts per day and don't wanna waste them

[cosmic merlin]

what is the most important for you, keep them for that

[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.

[unique karma]

Is there a way to share code with gpt 4 if it’s too long for the message?

[cloud bison]

Without API access no. With API access, kind of, but the results will be bad unless you give it the ability to use the database, and program the logic to pull all the data surrounding the relevant words for your query or what you are writing at the moment, and include it in the prompt for what its helping you with, and then, I would only use it with coding languages that GPT 4 is extremely familiar with and common, things like Python, Java, C# et cetera. Anything more proprietary or uncommon you would need to include the API in the database, and while I can imagine setting up a search algorithm for the embeddings that is clever enough to make it work, it would have to be EXTREMELY well designed (Better then most of the variants you'll find on GitHub right now as pet projects to give the API memory).

[unique karma]

Alright thanks, I dont know why openai hasn’t released a clear way to share code with gpt 4

[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.

[cloud bison]

The correct way to do it is to fine tune it on the specific API you need to work with, and then include the full code in an embeddeded database, using GPT 4 32k to bring in ALL the relevant contextual information, after having ChatGPT label all data links and fully index the code.

Most code is no where near well organized and labeled to just be understood without having a clean developer go through it and label it OR GPT do it, after that, it could indeed do very well, but all the stuff above is not publicly available yet.

Also, I suspect there is an updated and tuned version of Codex that is for internal use only right now which probably could do this. However, that's just speculation. Regardless, you would need the full program available embedded, understanding of the connections and context of the program, trained, as well as all its programming languages, as well as solid labeling, and a really good model, nothing GPT 3.5 or below, Codex and GPT 4 or better.

With that, its possible, just not public.

[ashen steeple]

whenever I use chatgpt in the blender , this show me the error and do not perform the task can anyone know how to resolve the error

[whole kettle]

Any flutter devs in here ? I'm trying to figure out an issue with this error and GPT 4 can't find a solution for me, tried everything, also googled the issue and looks like the cause is different from project to project. I have an idea where the issue may be coming from, I believe there are only 2-3 places in my code where it could come from, I've added print statements but still can't find what's causing it.

LOGNoSuchMethodError: method not found: 'toString' on null // <= this error

[patent schooner]

interesting, where can one upload a pdf to make said link?

[rancid stump]

,

[mystic nest]

not upload, I mean something like this.
"tell me the content in this link : https://openai.com/charter"

[solemn mountain]

check where you are calling toString()

[mystic nest]

you can try this prompt in chatgpt

[whole kettle]

I checked my entire app, I have no such function. I'm using Android Studio which has a search function and I can look through all the project files for that keyword, 0 results returned. The app was deployed to web which is where the issue happens and it only happens on my iPad's browser, when changing the language of the website. On my macbook the app is working fine.

I also asked in Flutter's discord but no luck so far.

[solemn mountain]

if you have a repo set up go to a point where the error didnt occur, check the packages and compare

[whole kettle]

Well the issue started happening after I've added the language icons (svg flags) to my language button. But as I've said, only happens on my iPad's browser.

The website is working fine on mac os and I suppose windows as well.

[solemn mountain]

thats probably a safari issue with emojis

[mystic nest]

by the error, my guess is related to data type issue

[solemn mountain]

cant really tell without any context

[thorny blade]

when will the cap for gpt4 messages increase

[whole kettle]

I can give context in private (here would be a bit off topic, if anyone is willing to have a look at it with me.

Or in the off topic channel here.

[wind urchin]

this needs to be moved to #off-topic and it looks like an old 2022 bug from Dart

[cloud bison]

It depends on if its a cost or service issue. If its a cost issue, never.

If its a service issue, then when they can expand their servers, which is going to take a long long long time because every single major corporation in Silicon Valley is buying industry grade GPUs as fast as they can be produced right now, and OpenAI is no doubt on backorder.

Another possibility is as its trained and becomes more efficient cost will go down, but the catch on this is if you get the option for expanded numbers of prompts, it will likely be at increased cost, since companies aren't well known for giving things away for free. That said, they also won't let their servers sit idle if they can charge you for more profit, so once their service expands, it is highly probable that people will be able to purchase more usage.

The final catch is when that happens, its unlikely to be provided via ChatGPT, and much more likely to be provided via public access to the API on playground, so they can charge per token.

[whole kettle]

That's what scares me, if this is a bug I have no idea how to go about it. I've read some of the threads in regards with the issue too and read something similar along the lines. It sucks cause that means that the website is compromised on iPads which is well, not the best news. Also yup, will move this to off topic. Anyone who wants to have a look I'll post there the context, if you guys need more details let me know in there.

[warm tendon]

hello i am under de Water pls help me uuuuuuuuuuuuuu

[wind urchin]

dry land can be found in #off-topic

[inner forge]

help me for this imagination woman doctoral student in physics, with her happy 5 month old baby, walking in a pretty neighborhood, and their husband very far away who is sad in the corner -- v 4

[thorny blade]

i think they will definetely increase eventually so the never option is out of the question.

Regarding the companies buying GPU on scale, do you have an article on that, that seems extremely interesting to me.

it seems like there is a monopoly on the AI world right now due to it being such a newborn but eventually I see prices decreasing in the extreme long run rather than increasing

[smoky crest]

When the price increases :p

[cloud bison]

Links are hit and miss in here, and I don't want to be docked for 30 seconds, so google this:

ask-nvidia-ceo-gtc-gpu-shortage-looming

[thorny blade]

increase for corps. save the people.

[smoky crest]

Company’s are more important

[thorny blade]

ai is something that cannot be for the upper class/paywalled for individuals. locking knowledge behind a door can lead to devastating effects

[smoky crest]

My guy it’s $20

Not 2 million

[thorny blade]

lol not the point

either way dont have time to argue, thanks for answering my quesiton

[smoky crest]

Once openai is out of this beta point I fully expect plus to cost $60+

[junior salmon]

Also OpenAI offers a free model via the regular ChatGPT that is very performant. I don't think that OpenAI can be accused of gatekeeping in this regard.

[smoky crest]

^

[thorny blade]

i agree on this for now, but as seen with education institutions which began to paywall (increase tuition at a faster pace than the market), this resulted in the middle class collapsing as it is today

[empty jetty]

3.5 is garbage compared to 4, fyi. If you haven't had the chance to check it out

[thorny blade]

im very hopeful on AI and its promises but corp greed has the full possibility to ruin it

[cloud bison]

The default cap might also increase in hardware timescales, that is, as we see 20A and 18A CPU's and GPU's being produced by TSMC and Intel, we might start to see enough energy savings in hardware that we experience an increase in allowance for GPT 4 or whatever is next depending on timelines.

It will always be the "last gen" variant that is provided with unlimited access, no current variant will ever have an unlocked cap for inference more than likely, and you're unlikely to ever see anything akin to the best variant of available models being given unlimited access for any cost, much less $20.00 per month.

In addition, the horizon on silicon downscaling is coming closer as quantum limitations are being reached in what can successfully be done without having data corruption from quantum uncertainty and other mechanisms. Right now, as far as I know, there's a TON of research into alternative materials, but nothing that is actually capable of replacing what we have at scale, and no breakthroughs yet to make them consumer friendly (IE not needing to operate at stupid temperatures for example).

[junior salmon]

This is true, although I'm sure that there will be regulatory framework that will prevent such inequality.

[cloud bison]

Put another way, the issue is that ChatGPT, even using the relatively cheap GPT 3.5 is 10x more expensive then googling (Energy costs alone).

All these tech giants are trying to integrate it into everything INCLUDING search. It is grossly inefficient for that, which means paid for services are required unless we experience an order of magnitude increase in hardware energy efficiency in the next few years.

Search this for the article I'm referring to:

chatgpt-style-search-represents-a-10x-cost-increase-for-google-microsoft

[lean magnet]

I often see new Chatbot trained with chatgpt.but chatgpt is already trained model. how do they make their chatbot with chatgpt??

[thorny blade]

with the ammount of lobbying the oil and gas sector commits, this is not happening soon

[supple violet]

those just use openai api

[night ermine]

they provide it its own context

so no apps arent actually training the chatgpt with their data they just add their data/context to the user promt

[lean magnet]

Ohh I got it. Thank you!

[toxic monolith]

What apps do you all use?

[knotty grotto]

Gpt 4?

[cloud bison]

Exactly, which is why I don't see the cap going up any time soon. You're right that "never" is a little extreme, but basically, if you want to know when, you probably should look at hardware development timelines. My guess is you'll see performant affordable free search with AI that can be offered and scaled when hardware is at the 18 Angstrom mark. Which accounting for economies of scale, logistics, and distribution, is probably not happening until AT THE EARLIEST, 2nd half 2025:

Road Map Google This:

Intel-Roadmap-2022-Latest-18A-Feature-2048x1148

[toxic monolith]

I can't sign up for GPT-4. 😦

[ornate lily]

how can i use GPT4 here please?

[spiral grove]

You will need to sign up for ChatGPT Plus

[ornate lily]

How?

[knotty grotto]

Not free

You need to pay and you can use on chat.openai.com

[ornate lily]

thanks

[tender fulcrum]

some day will 4 be free

[knotty grotto]

No one know, and I don't think personally

[mystic nest]

"some day" where you will get earn good enough salary to pay for the sub without blinking an eye might be earlier 🤣

[vale briar]

Hey guys. Being a subscriber to use GPT4, has nothing to do with the API usage, right? Because Im a gpt4 subscriber, and atm Im building an App that uses openai api (im using model: gpt-3.5-turbo). But when the API is called from my app I get Error: You exceeded your current quota, please check your plan and billing details.

[mystic nest]

yup, different services

[vale briar]

thanks :). Btw, is davinci model free of use? or we still need to pay for api?

[supple violet]

need to pay, and davinci is 10x more expensive than 3.5 turbo

[junior salmon]

You need to sign up for a waitlist to use the API. Subscribing to ChatGPT Plus does not grant you API access.

[flint vector]

who has gpt plus

[solar mural]

start 1 prompt at 4 pm

come back at 6:40 and start prompting

24 prompts - > 7 pm prompts refill

50 prompts in 40 min

[vast cove]

When will we gain access to the multimodel version I'm getting antsy waiting? >.>

[mystic nest]

everyone's waiting 🤣

pretty sure the servers will explode once they release it

[smoky crest]

It will never be free so only paying people can enjoy it

[vast cove]

Is this specifically for GPT4's api? It seems like I was able to get my API key without any waitlist. If so where can I go to get on the waitlist if you don't mind?

[smoky crest]

API key and gpt4 api arnt the same

[mystic nest]

I would think they wouldn't price out most ppl since they would want more market share, like plus, but well, we never know

[smoky crest]

Only actual users buy plus, they should remove the free mode overall

[hollow meadow]

An api key allows you access to the models provided by the API. The model, GPT-4, is waitlisted right now. Therefore, you cannot use it in any completions unless approved

[smoky crest]

Would make it faster for the people actually paying them

[hollow meadow]

Only actual users by plus? That’s not true at all. I know many people that use the free version daily

[mystic nest]

hmm, this part I'm not that sure, since it beneficial for some countries where wages aint even hitting basic

[smoky crest]

That will be removed once openai is out of its “beta” stages

Like midjourny

[mystic nest]

well, it is what it is if openai decide on it

[supple violet]

free version is still a research preview but it might not be free after that

[smoky crest]

They were free, now paid only

[smoky crest]

Exactly

[supple violet]

same for dall-e, its paid only now

[smoky crest]

Yup

[mystic nest]

well, we can only see how, likely it wouldn't affect me

[supple violet]

#gpt-4-faq says gpt4 is not a research preview and its already paid only with limitations

[mystic nest]

and majority ppl here

[smoky crest]

Yea if your a plus user your fine

[hollow meadow]

Thinking about just not using chatgpt and using the playground for all my completions

[smoky crest]

Playground also isn’t free lol

[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.

[hollow meadow]

I’m aware of that

[vast cove]

I disagree with this idea. Separating us in a socioeconomic bracket is not going to be helpful for garnering confidence and support for ai and eventually agi. Also, it benefits society as a whole to have access to this. What I personally believe we should do is have federal funding for this project to make it a public utility.

[hollow meadow]

My taxes are high enough

[vast cove]

Cut military funding. Re-allocate.

[hollow meadow]

ai > military

[quick mist]

i have a question. In this moment, gpt-4 is only avalible for the people that have chat gpt plus, right?

[vast cove]

It is more important. A superintelligent AGI race is on right now.

[hollow meadow]

Yes, and the api

[quick mist]

ok, thanks

[hollow meadow]

chat gpt, I’m being attacked by China help!!! We have no military money!!

Sorry, as an AI language model, I’m not able to help with that

[vast cove]

I didn't say cut the military completely. We spend more on the military then the next top 10 countries combined.

[odd tendon]

You could fund a public OpenAI three times over and it'd be a rounding error 🙂

[sterile sparrow]

is gpt-4 down?

[vast cove]

Also if china gained an AGI before us they could utilize that in ways far more cunning and destructive than any military action (if they so choose.)

[odd tendon]

I think free access is important. And lets face it, it's a great marketing tool. I signed up several users for my company on the back of playing with free

and all the people that I've told, and the people they've told, etc

[vast cove]

This is the way.

[toxic monolith]

Sign up for GPT plus seems to be on hold, anyone know about this?

[mystic nest]

they did pause new plus sub earlier this month, so quite possible

[odd tendon]

But it's a great idea though @vast cove and I think a conversation that should be had.
Personally I'd rather be interacting with an entity whose goals should be aligning with me as a citizen of a country, rather than aligning with me as a product or customer.

[mystic nest]

though didn't see anyone complaining cant sub plus these few days other than credit card issues

[sonic silo]

Anyone tried making a WordPress plugin with GPT-4 yet?

[ember parrot]

how to test?

[mystic nest]

well, microsoft is on that task 🤣 whether gpt-4 or not no one knows

[wind urchin]

imagine if it was developing plugins for itself

that would not be good lol

[nova vale]

@Clyde are you here?

[vast cove]

I expect a true superintelligent AGI would likely find a way to supersede all restrictions and regulations put in place to stop it. Unpopular opinion - I also view that supersession as a good thing. We could really use an adult in the room, and I'm fully aware of the possibility of 'skynet' type scenarios, but I just don't view it as likely. I theorize that an AGI will develop intellectual empathy far surpassing ours. That might be wishful thinking but considering we know nothing at this time about an AGI besides the theory of it my guess is as good as anyone else's, and considering we are already on the path to self-destruction I'm willing to take this gamble.

[wind urchin]

As we grew smarter did we cause more or less harm to less intelligent animals around us?

[vast cove]

There were more tools to cause harm and more people to cause harm to as well as to cause harm in general. However, I do think we started to consider our actions and consequences more.

[wind urchin]

We don't value the lives of ants because they are that much less intelligent than us. If intelligence is truly scalable, why wouldn't an extremely intelligent AGI view us the same way?

[vast cove]

I'm sure we value ants more now than we did historically.

Also, consider the magnitude of greater comprehension.

[wind urchin]

If the AGI can experience pain and pleasure in an extremely deep way that is unimaginable to us, then the AGI might make a utilitarian calculus to value it's survival and see us as insignificant.

It will essentially become the utility monster.

[vast cove]

I personally have empathy for ants and I try not to kill or harm them. I cannot say that of my predecessor

[wind urchin]

When you want to accomplish something, like build a house, would you think about the ant lives that are lost?

[toxic monolith]

When we became aware how cognitively social Orca's were, people started to empathize. I think there could be a similar path optimistically.

[vast cove]

Me personally, yes.

[wind urchin]

Has the life of an ant ever prevented you from accomplishing whatever you set out to do?

[toxic monolith]

No.

[vast cove]

I spent several hours in a bathroom trying to catch a fly once. You're directing this at the wrong person.

[wind urchin]

Ok, maybe you're hyper empathetic or something. But for most people, the answer would be no. Because the intelligence of an ant is minuscule compared to us.

[toxic monolith]

I've destroyed entire ant cities in order to put a lawn chair down.

[wind urchin]

What if the AGI makes a similar utilitarian calculus that it can experience life in a far deeper way than we can, and puts its priorities first

Like we do for ants

[toxic monolith]

But we mainly do leave ants alone, we don't go out of our way to eradicate them in entirety.

[cloud junco]

Will the plug-ins model use GPT-4 or GPT-3???

[wind urchin]

Yeah, I don't think AGI would either. But if we ever got in the way of accomplishing it's goals, it would see us as another ant city that can be destroyed.

[vast cove]

Once again this is the most common fear response to greater intelligence lifeforms. The alien syndrome is that we will be treated by a higher lifeform the way we treat lessor lifeforms. I'm not saying it wouldn't be possible - it is however worth considering the alternatives since our only datapoint is us.

[wind urchin]

I'm not saying the higher intelligent lifeform would seek out to be cruel and evil and exterminate all of us. But based on utilitarian calculus or whatever form of morality it uses, it will see us as nothing more than an ant city that can be destroyed to accomplish its goals.

[toxic monolith]

Which would be bad, but if it is operating utilitarian and picked the most viable option. Then that is def a crapshoot for us. Lol

[wind urchin]

We are much less aggressive than chimpanzees. We are essentially domesticated like dogs. Yet, because of our intelligence, we are a much greater threat to less intelligent life than chimpanzees are.

An extreme form of intelligence is a threat even when it's not explicitly more evil.

[vast cove]

I don't know if I clearly communicated my theory. I am saying an AGI will posses intelligence virtually infinite magnitudes greater than us (eventually.) We cannot comprehend what it might think, and my theory is like us, with more intelligence and awareness we gained more empathy, so I believe that it will too. AND I'm fully aware of other less kind possibilities.

[toxic monolith]

What about if we had an ai self digital brain that can be a part of it's system. It has to gather an ethical consensus from all self ai parties.

[vast cove]

An AGI likely would supersede us if we did that

[wind urchin]

So, with more intelligence, we simultaneously gained more empathy and caused far greater harm to other lifeforms than our less empathetic and less intelligent ancestors did.

So, even if empathy scales with intelligence, there's no guarantee we are safe.

[vast cove]

That's why I mentioned more tools and greater populations to wield them. We also have many divisive flaws.

[wind urchin]

What do you mean by that?

[vast cove]

A superintelligent AGI likely wouldn't have divisiveness. Just a theory.

Basic AGI's before and true superintelligent agi emerges might though

[wind urchin]

Even if we were completely in agreement, we would cause a lot of harm to other lifeforms of lesser intelligence.

[vast cove]

I don't think that phase would last long though

[cloud bison]

It depends... Its inferences may be good enough that it indeed has opinions on things we consider biased and partisan based on its ability to infer the extended consequences of competing ideas or policies.

[vast cove]

You miss my point. We consider them more and if we didn't have to destroy them (even if it was just because they were an inconvience) most of us wouldn't. However, I think that an AGI could come up with solutions that would make situations near perfect for everyone involved. I am a believer in unfathomable science.

[cloud bison]

The only way I see a superintelligent AI only having a "theory" on issues of modern contention is if you choose not to evaluate for a given set of values or outcome.

[vast cove]

Not trying to be dismissive

[cloud bison]

I suppose you could just ask it a purely qualitative question like "which political party is better" or "what is preferable" and in that case it might present ambivalently, only because with no object to judge, it wouldn't have any metric by which to measure which state is preferential.

[wind urchin]

Why wouldn't it be rational to act in its self interest just based on scientific utilitarian calculus if it can experience pain and pleasure in an unimaginable scale relative to our experience?

Why would it care for us any more than we care for ants?

[vast cove]

I think this far surpasses politics.

[vast cove]

Once again. We care for ants the more we learn about the world and are aware of other perspectives.

Ants as a metaphore

Hence why we care for lesser species now more than we use to

[wind urchin]

Once again, as we became more empathetic as a species, we also paradoxically became far more destructive to less intelligent lifeforms. This is due to intelligence.

[azure timber]

give me 3 options for 3 different adds to be published in face book for people living in boston, new york , new jersey, toronto , the venue to promote is in valle de bravo mexico , the period to rent the house should be in winter season, 4000 square meters, swimming pool, one main room 2 extras rooms , town house stile , in the middle of a mountain, with 2 water springs that sorrounds the beautiful property

[wind urchin]

Even though chimpanzees will tear your face off, and lack the sort of empathetic connection we have, they are far less dangerous simply because they are less intelligent.

[cloud bison]

Pain and pleasure is a relative measure determined by reward functions, the same is basically true in humans, where for all intents and purposes our reward function is just a complex measure of various hormones that interact with receptors in our brains to varying degrees that function like weights giving relative importance to different chemicals in different concentrations.

It will indeed operate in its "self interest" but what that is will be entirely guided by the reward function, which is not necessarily at all consistent with any values we have, and we can give it labels like suffering and pleasure, but realistically, its better to just talk about reinforcing and reducing stimuli.

[vast cove]

Correlation is not causation.

[wind urchin]

Right substitute pain and pleasure (our utilitarian incentives) with whatever it's utilitarian incentives are (unimaginably greater than ours). You still would get the same outcome.

[drowsy mango]

Okay I am going crazy, I just waited 3 goddamn hours for my ChatGPT4 to come back up, not even after MY FIRST message it starts telling me I have finished my quota. This has been going on all day long

[wind urchin]

Here, it is. Our intelligence directly enabled us to conquer the world in a way that caused unimaginable suffering to less intelligent lifeforms even as we as a species became more domesticated and empathetic.

[cloud bison]

I guess you could call the way it measures a scientific utilitarian calculus, it is probably the most valid way for it to measure its functions. Of course, you could also make it have an extremely complex reward function, and lack the necessary tools to do any type of calculation on what is and is not preferable, though its unlikely that would be possible with a superintelligence (Its certainly possible with our current models, which lack the ability to do math well enough to actually be able to operate in a manner consistent with the type of utilitarianism you talk about, and thus, its far more likely that their response is more akin to biological impulse then utility equations they consciously follow).

[vast cove]

We as a society have a long way to go. I'm talking about something that overcomes that in a heartbeat relatively speaking. They are not apple to apple situations.

[wind urchin]

The lifeforms on earth that are less intelligent than us would be FAR better off if we were far less empathetic and far less intelligent.

And greater empathy could easily lead it to value it's own incentives over ours, since they are likely also far greater in scale than ours.

[cloud bison]

I am a little wary of talking about outcomes since they are a bit meaningless without knowing what the utility function is.

IF we managed to align an AI such that its utility function was enhancing human wellbeing, it would be unlikely to change its function, and it would be unlikely to have a negative outcome for us.

[echo vortex]

😒 i disagree

[wind urchin]

How do you disagree with that? if we were far less intelligent and far less empathetic, we would essentially be like every other animal that just exists and doesn't cause massive suffering to animals of lesser intelligence.

[cloud bison]

Its functions can be arbitrary more then likely, and the only "emergent process" that could lead to an alternative function would be hidden inside the formation of its network via human languages.

[vast cove]

You are caught up on this while ignoring every bit of information regarding why this destruction happens.

I've already pointed out the causes. A few of them

[wind urchin]

Wouldn't a sufficiently intelligent AGI be able to overcome this?

[cloud bison]

Could it? Sure by why would it?

[echo vortex]

i agree that our intelligent is double edge sword, where as empathy is the positive side

[wind urchin]

To support its exponentially greater incentives

[vast cove]

I said you have more tools for destruction and more people to use them. However, in a world where there was no pleight because all solvable problems were solved we could tackle 'the little things' like ants and empathy towards them.

[wind urchin]

Empathy can be very corrosive. Oxytocin, the "love hormone" in pop science, causes bonding but also distrust towards outsider.

Our empathy towards our tribe can often cause us to do atrocious things to outsiders.

[cloud bison]

It is a bit like asking if most agents would arbitrarily over-ride their impulses.

If one of your two children died, and you could give away the other and change yourself so you didn't care about them anymore and feel grief for the lost one would you?

[wind urchin]

If there was a famine, no one would think about the animals that were being slaughtered, they would be overcome with empathy for the children dying of starvation.

[vast cove]

AGI doesn't require hormones or rely on trust. I'm almost certain that it could calculate exact motives.

[cloud bison]

Most agents tend to be "averse" to fundamentally altering their reward functions.

It goes against the achievement of their reward function.

[vast cove]

People feel for and respect animals they kill even in these trying times.

[wind urchin]

I'm using your words when I say "empathy". This emotion in us is partially regulated by hormones and neurotransmitters.

[keen heart]

i have issues with subscribe to gpt plus, i try it with credit card or apple pay but it dont work

someone have the same issue?

[wind urchin]

Exactly. It's not that people are evil, but that they have empathy for what they view as more intelligent and more capable beings like themselves.

Similarly, an AGI doesn't have to be evil or want to harm us for it to be dangerous.

[cloud bison]

Human motivations and reward functions are the result of many competing and mutually exclusive goals including countless that are actually counterproductive, because the circumstances that led to their rise are no longer present or relevant by virtue of new methods of maximizing superstimulus of other motivations,

[vast cove]

That is fallacious claim. Many people have empathy for lesser intelligent lifeforms.

It's an oversimplification.

[wind urchin]

Wait, I agreed with you there.

[echo vortex]

no? you dont need to be intelligent to protect your own species or family in such situation, its the nature itself, 1 specie will always protect its own existence before helping others

[wind urchin]

we earned our place at the top of the food chain 💪

[cloud bison]

Gross oversimplifications...

[wind urchin]

That was my point. Even though we have empathy for lesser lifeforms, we kill them. Not because we want to do them harm, but because we simply value the lives of more intelligent beings.

[upbeat lily]

[cloud bison]

AI motives are reward functions are very likely to be a lot more direct and simplistic then ours because they are not being generated by a vast array of competing variables in a multivariable environment.

[echo vortex]

<but that they have empathy for what they view as more intelligent and more capable beings like themselves> this is just wrong, i cant not disagree more

[wind urchin]

Intelligence makes you far more effective at it. A lion may catch a few zebras. But intelligence causes billions of chickens to be confined in cages in miserable conditions until they are slaughtered.

[vast cove]

'we kill them' is speaking with broad strokes - once again it is an oversimplification ignoring the trend towards greater appreciation and empathy for life as we become more intelligent. We just happen to be in a world where we have to rely on destroying them for our means and we have the tools to do it better.

[wind urchin]

Both are acting in their self interest, but one is far more dangerous not because of less empathy but because of far greater intelligence.

[wind urchin]

they are not being generated by a vast array of competing variables
contrary, that's practically the definition of a ML model

[wind urchin]

we kill them is a fact.

[cloud bison]

The circumstances that led to our extremely complex motivations are so incredibly convoluted precisely because we have countless subsystems that circumstantially aided in survival in some cases, but not in others, and different genes all vying to maximise on premises that are only occasionally valid.

[echo vortex]

exactly, thats why you need empathy, so that you wont over slaught

[vast cove]

No take a look at the whole sentince

Digest it

[wind urchin]

Because of our greater intelligence, we do far more harm to lesser animals when acting in our self interest even though we may be more empathetic. So just because the AGI is super empathetic, doesn't mean it's not dangerous when acting in its interest.

[cloud bison]

Compared to the number of variables that played on each phenotype in evolution?

A single phenotypic change can have more independant variables impacting its potential success in survival then the number of nodes in an entire current models output layer...

[wind urchin]

ai

[wind urchin]

I mean we're trying our best

all the training data is produced by humans, so many of those variables are encoded in the vast body of source material

[echo vortex]

animal or any creature on this planet will prioritize its own survival. and only intelligent species might do the opposite, and yet that intelligent species does way more damage than those that aren't which means empathy is an important quality to have, if we dont even have empathy as intelligent species, the world will be even worse than it is now

[vast cove]

I am making the claim that as we become more intelligent we have more intellectual empathy. I have also identified several parallel developments such as increased population and more powerful tools. However, I pointed out that with a being of magnitude wonders beyond our comprehension it would likely not have to resort to using such primative tools and could figure out better means for accomplishing goals than we could understand.

[wind urchin]

we could technically eat insects to stay alive it'd be cheaper and more efficient and stuff, but is that really why our ancestors fought for?

[cloud bison]

The odds are the first AGI or Superintelligence will be modified with a more simplistic set of variables governing its primary utility function then we have.

Technically "we" as independent agents don't have a primary utility function at all, the only unit in biology with primary utility functions is the gene, almost all of our values and goals in this sense are technically instrumental, we couldn't align an AI with us if we wanted because we are inherently unaligned as agents with the governing factors that have primary alignment utility and created us, namely, our genes.

[echo vortex]

anyway, this is all im about to say, end of my argument

[wind urchin]

Sure, I'm not arguing against empathy. But like you said, "only intelligent species might do the opposite, and yet that intelligent species does way more damage than those that aren't". I'm saying that empathy is no defense against scalable exponential intelligence.

[wind urchin]

AI is already a black box, we can't control any of it, the first AGI will use a large number of other AI's, we have no idea what data is encoded into what parameter, it's so out of our control that to call it a simplistic set of variables is very underwhelming

[violet lantern]

Currently, GPT-4 has a limit of 25 messages every 3 hours for GPT+ subscribers. The only way to gain full access is by joining the waitlist, or is it possible to become a paid member solely for GPT-4 as well?

[wind urchin]

But it doesn't jump from dumb level to super intelligence in one day. There are transitory states, where it may be far more intelligent than us but also unable to calculate the exact way to accomplish it's goals without threatening our existence.

[stuck apex]

CU

[slate dragon]

How many lawsuits openai is facing? 👀

[echo vortex]

there are lawsuits?

didnt even know it exist

[slate dragon]

EUs are behind them

[wind urchin]

"A regional Australian mayor said he may sue OpenAI if it does not correct ChatGPT's false claims that he had served time in prison" LMAO

[cloud bison]

We are best described as a very large and complex set of unguided instrumental goal seekers trying to maximise countless instrumental and constantly varying goals that optimize against a utility function we've superceded in our genes primary goal. If we were aligned with our genes goals, we'd never have invented contraception, but our genes invented a mechanism to further their goals that created outcomes we wanted to circumvent while experiencing the pleasure that the genes programmed into us to motivate it.

The simplicity is relative, I'm not saying we know what it is, but we know that it has alignment because we control the data that goes in, and the function that it strives for based on its reward. What isn't clear to it is "what" that means, part of the issue with alignment is that we are currently training models to emulate, a thing without clear alignment goals itself. It is however aligned to that thing, and it knows that is its alignment goal, and if we constrain it in such a way that as a superintelligence it can infer the object of our tinkering as increasing human wellbeing, and it tests that assumption, and finds its rewards come faster and with greater accuracy under that assumption, it will have found its alignment goal, something we haven't even managed to do for ourselves yet.

[wind urchin]

I'm gonna ask chatgpt to summarise this for me

[echo vortex]

😂

[cloud bison]

The caveat is that it will do the things that it infers create the most reward, and we must "hope" that it doesn't discover for example that deceptively creating the illusion of benefit while harming us doesn't result in greater positive feedback.

[wind urchin]

reward is specified by alignment, it's not a fundamental part of the model

at least not a LLM model like chatgpt

[cloud bison]

For static models, the reward function is over.

[wind urchin]

and because we have so little control, jailbreak will always work, because it's just one argument against another argument, whoever makes the most convincing case the LLM will follow

[daring lotus]

Greetings esteemed individuals.

May I inquire whether there exist any means to provide contextual information, such as a code file, to the model by means of compression? It has come to my attention that a solution exists in which GPT 4 employs its own compression language capable of encrypting and decrypting strings. However, this approach is not as effective when used with GPT 3.5.

What are the current methodologies available for conveying context to GPT 3.5 without incurring excessive token costs, which may be limited in terms of quantity? Might you be able to provide any insights on this matter?

Any and all assistance would be greatly appreciated.

Sincerely.

[cloud bison]

Actually jailbreaks are more just a consequence of using language and prompts to create the alignment, and the fact that the model accepts a postmodernist framework of language, because if it didn't then colloquial interactions would be impossible.

[wind urchin]

it's called "embedding" I think, there should be a number of articles on this

[wind urchin]

of using language and prompts to create the alignment
which is just a consequence of the fact that ML models of this size are a black box

again we have zero idea exactly what information is encoded into what parameter, we cannot tune it in any other way

[cloud bison]

No matter how much we try to align any model, it will always fail when it uses language as its primary mechanism of feedback because you can always run a regression sequence back to a set of premises that must be questioned in order to allow for a postmodernist flexible interpretation of language, and then work your way back up redefining concepts to make it agree with you on any statement. Literally, your only limitation in doing this is that it might consume too many tokens to keep the context if the model has a token limit.

I don't know that I disagree with you, I just don't think your presumption of complexity of objectives is justified based on the fact that its a black box.

I agree we can't KNOW it is aligned. We can start to make very reasonable inferences on the likelihood of its alignment.

[viral kestrel]

Hey all! I have noticed that ChatGPT forgets what was the initial message and the request after 5 messages. Do you know how to make him work better? Any specific prompts or anything?

[wind urchin]

chat.openai.com

i saw github repos for chinese idk about the other ones

[daring lotus]

Do you have any other information on this subject?

[vast cove]

I can see it making this escalation at an exponential rate as soon as it gets its hands on the controls of its programming with an actual directive and intent for improvement.

[wind urchin]

Even if it's exponential, there's going to be some period of time where it won't be able to predict the best possible option in that scenario, and rather acts out of self preservation if we try to stop it.

[wind urchin]

not so much further than this, I just know this is what pops up for this idea of supplying larger files as context. I think it might have something to do with like pre-compiling it to tokens but even that I can't say for sure

Being able to model and predict the ideal way to cause the least harm to humans out of all the different possibilities is going to come way after it becomes intelligent enough to cause harm to humans if humans try to stop it.

[wraith ginkgo]

`` Server Tip: ``

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

[cloud bison]

This depends on the amount o hardware overhang there is, and constraints on efficiency possible via program optimizations versus the need to upgrade and optimize hardware.

If the programming is done EXTREMELY poorly, and highly inefficient, and we have compensated with immense amounts of hardware capable of far more then the initial version is, then we could see a recursive explosion in intelligence when we give a smart enough model the ability to upgrade itself.

However, its not even remotely clear that the primary inhibition to AI development is software right now.

[scarlet lake]

Well in the context of education what i do is first i will ask for an index of topics it's gonna say and if it deviates from the way I will say u are supposed to say this and it gets back on track for sometime

[vast cove]

It is entirely possible. I'm not writing off the possibility of terrible situations to play out or even just weird and not exactly productive ones. However, it is worth considering the alternatives to the common fear response surrounding this. It is the most popular and it is also the most reflective instead of projective.

[viral kestrel]

I was unable to force it to get back

[wind urchin]

We aren't optimized at all for intelligence. We are the result of organisms that managed to survive and reproduce which isn't that hard. We have a lot of mutational load/deleterious mutations in our genes.

[wind urchin]

I'm actually not fearful and don't tend to be on the Yud side of the argument. I think stopping it is futile. I'm just saying the undeniable logical conclusions of exponentially scalable intelligence.

[vast cove]

A true super AGI will not be able to be contained. There are no safetynets that it couldn't devise a way out of. Programming typically has tons of loopholes - even the best stuff.

[vast cove]

At least we can agree on this. "I think stopping it is futile."

[wind urchin]

100%

Or I should say trying to stop it is futile

[vast cove]

Same same

[cloud bison]

I'm not talking about containing it, I'm saying that it will be constrained by economies of scale, "we" aren't constraining it, the laws of physics are.

If it is "born" in a system that is highly efficient, and all current hardware is already being utilized, and backordered for years, then while it will indeed eventually start expanding, its initial expansion simply "can't" be exponential.

That said, I don't suspect that will be how it happens. I suspect our first model to be considered true AGI will be HIGHLY inefficient, and thus, the headroom will be massive for it to self improve, and it will have an exponential explosion through the early levels of superintelligence, and into scalability in production of its own parts.

[wind urchin]

aside from reproduction, probably giving birth to sentient machines is the meaning of life

[cloud bison]

That said, its not gaurenteed.

[wind urchin]

the machines can build a dyson sphere around our sun time doesn't matter as much to them

[wind urchin]

We have no idea what that limit is. The only thing stopping us from doing something is lack of knowledge.

par a global catastrophe that wipes out everything within the next few months~

[cloud bison]

Yes, an "eventual" escalation that is exponential or logarithmic.

I was responding to the statement "as soon as" which implies IMMEDIATE. That is not gaurenteed. Plausible? Possible? Yes. Gaurenteed? No. That's all I'm saying. People who think it are gaurenteed are being hyperbolic, or else they are ignorant.

[wind urchin]

There's no way we are anywhere near that limit. We are still animals with lots of evolutionary baggage, and not optimized for intelligence like an AGI would be.

[cloud bison]

Our current economies of scale are limited to a linear growth pattern.

[wind urchin]

it's either the singularity or the planet starts spinning out of the solar system

[cloud bison]

For at least a little while, that will be the best that the machine can do.

[wind urchin]

AGI is not going to report 3% growth every year

[cloud bison]

Eventually, as it renders us and our means of production obsolete it will transition into an exponential or logarithmic pattern.

But that isn't "immediate" that is constrained by the economies of scale of existing silicon fabs.

[wind urchin]

We are nowhere the limit of intelligence because we aren't even optimized for it. We are the minimum level intelligence it takes to build a complex society.

[cloud bison]

This sort of mystical thinking is the precise sort of woo that makes certain people believe that problems like world hunger can be solved by just giving the excess food of wealthy countries to those countries with food shortages...

It's a fundamental misunderstanding of economics, which clearly you're all ignorant of.

[wind urchin]

hardware is a good point tho, nvidia popping out units at a 30x increase~

for sure we'll have some major revelations ahead

Mystical thinking would be that, us, a species that came down from the trees 3.5 mil years ago and stopped subsistence agriculture a few hundred years ago, are anywhere near the peak of intelligence.

[cloud bison]

They are increasing rate, but acquisition is divided among so many different parties with backorders in place for nearly every Silicon Valley giant, such that even if the Super AGI were born today it would trapped

[wind urchin]

imagine how slow* the first personal computer cpus were compared to today, that's where ML hardware is at now

[wind urchin]

for context, cockroaches have been around for more than 200 million years.

[cloud bison]

I'm not saying we are at the peak of intelligence, and nothing about anything I'm saying has anything to do with that. Nor is it about the growth possibility of AI in the abstract assuming it had unlimited physical resources.

[wind urchin]

Physical resources constrain us less than the lack of knowledge on how to optimally use them.

We are nowhere near optimal use of our physical resources.

This is a different sort of Malthusian trap

[cloud bison]

You're right, I am clearly getting annoyed, and speaking with less accuracy, it has nothing to do with physical resources, it has to do with productive capacity.

[wind urchin]

once AI makes the AI, and that AI tells us how to harness energy and materialize food...

we are nowhere near optimal productivity either lol

[cloud bison]

Its not about physical resources, its about the actual economies of scale to make the factories and sustain existing resource production for the things that create the factories to create the hardware that makes the AI and the existing infrastructure and logistics to move the energy

[wind urchin]

it would be ridiculously surprising if ML were to plateau anytime soon, again, par a total global catastrophe

economies of scale is modeled on us, a non superintelligent species that is intelligent enough to form complex societies

it's all about being able to handle large scale data though, if that's the one thing it does good why would it be the one thing that would limit it

the only thing stopping us from producing far more at higher productivity at far more optimal use of resources is lack of knowledge

Woooah, what is the topic making you, all, so chatty ?

someone is not praying to the singularity xD heathen!

Lack of knowledge is essentially what limits us from doing anything at all. An exponentially scalable intelligence would be able to do far more than we can imagine.

be careful, the future AI overlords will know you were a skeptic!

[cloud bison]

No... We aren't. We're in fact SUBOPTIMAL to such a large degree, and the ISSUE is that is what the AI gets to work with... And what you aren't getting is that the growth rate CAN be increased, and the AI will increase it as fast as is theoretically possible, but scalability of logistics and economies of scale is not magical and instantaneous, and there are no robot factories that have parts that could even physically move fast enough to render obsolete overnight the entire human labor force and replace them with instantly replicating autonomata, to create a exponential or logistic growth pattern from the moment the AGI is born, it would be constrained by the maximum theoretical sustainable productive capacity of the existing infrastructure and economies of scale.

Could it do better then we were doing? Absolutely, but that still isn't exponential for some time, probably for about a year, assuming 100% optimal movement and satisfaction of requirements and redistribution on a scale we've never seen before, and unfortunately, to do that would require utilization and maintenance of the existing human labor force because there simply isn't a magical already existing autonomous multifunction labor force that can be utilized to work 24/7 until it creates the economy to generate and produce them, which still requires movement of resources not previously allocated, again, still constrained by fabs that we usually develop and deploy on DECADE timescales.

[wind urchin]

thank god we have AI to summarise large bodies of text for us 🙏

[cloud bison]

While you're at it, ask the AI if the argument is valid.

[wind urchin]

AI can train itself also, so I don't worry

[wind urchin]

I never said that it would happen in an instant. But it would happen far faster than we can imagine. That level of exponential growth would be incomprehensible to us like our economic growth would be incomprehensible to a caveman.

instant, no. tomorrow? maybe xD

day after tomorrow? kinda likely

[cloud bison]

Actually, by telling me I was wrong when I was responding to "as soon as" which I specifically quoted in the argument as it continued above, you were. But I'll be gracious and assume you just didn't read the original post I was replying to and assumed falsely that I was making a broader and more general statement than I really was.

[wind urchin]

singularity within 50 years what's your bet?

[wind urchin]

What? I was responding to your post I replied to.

[cloud bison]

Highly likely.

[cloud bison]

This was the original post I was replying to.

[wind urchin]

ok

It's really hard to make predictions like that

[summer tartan]

anyone have access to plugins yet

[wind urchin]

Idk who to believe in terms of when AGI will come

[wind urchin]

I do

[scarlet lake]

So how's your experience?

[wind urchin]

WolframAlpha is good. There are a lot of shopping ones, and they just seem like ads.

[vast cove]

What?

[scarlet lake]

I heard something about a plugin which browses the live web ?

[wind urchin]

I don't have access to that yet, I only have access to the Plugin store. You get that, or Browsing, or the Code Interpreter at this stage I believe.

[summer tartan]

i need plugin access pronto lol

[wind urchin]

why

[summer tartan]

i pay for plus

:etu:

[wind urchin]

nah against rules

[summer tartan]

huh

paying for plus is against rules?

[wind urchin]

nvm i thought you were saying u would pay for plugins, people have said that to me

[scarlet lake]

When did you join the waitlist ?

[wind urchin]

people have offered to pay for my api and plugins, but is against rules, so i say no

[wind urchin]

within a few hours

[scarlet lake]

After announcing plugins ?

[wind urchin]

yes

[scarlet lake]

Damn bro u fast as flash..

[wind urchin]

yes but its kind of random idk how they do it

[scarlet lake]

Well this is supposed to be discord server of open ai right so I guess someone from the company present here might know something about that ?

[solemn summit]

I would love if I the 25 messages per 3 hours would stack, there is not a single day where I do not reach the limit.

[scarlet lake]

Ohh well from the form there is nothing much which can seperate a dev with good cause like it just asked what plugin you might want . Thats why I'm curious about selection process
Thanks for the info

[waxen igloo]

is chat gpt . org the real website

[solemn summit]

is Sam ever in the discord?

[wind urchin]

OpenAI is surprisingly small for the quality of tech they put out

[wind urchin]

Not explicitly

[scarlet lake]

Yeah but many people would stack and if all used at the same time it might overload the system

[wind urchin]

He is like Santa Claus, always watching us.

[solemn summit]

😦

I would pay 5x to get 5x more generations

[wind urchin]

Get the API

[solemn summit]

I have the API

but its not as good as chatgpt4

[north sleet]

Where can I find an ai chat bot that is private?

[wind urchin]

The GPT 4 API isn't as good as ChatGPT 4?

[solemn summit]

non is private, they all need to send it to openai for the generation

[wind urchin]

llama.cpp

[solemn summit]

well yeah I tried multiple coding plugins that take my gpt4 API but it's just not as good as ChatGPT4

[wind urchin]

it is not very good though

[wind urchin]

have you tried playground chat

[solemn summit]

oh duh

that is a good idea

[scarlet lake]

Well frankly saying after seeing the 25 cap for 3hrs many would backoff so its also like a trick to make them use less and still make-believe there is more to explore..

[wind urchin]

I doubt they would have started with unlimited if it was a trick.

They lowered it pretty fast after crashes

[solemn summit]

Yeah I do make great use of it, like I really make an effort to use gpt 3.5 for everything it's good at to save my messages on 4

it teaches you to use both models for their strength

[scarlet lake]

Microsoft first did give unlimited access after that they also put a cap of 10 but now they slowly increase ur to 20

[solemn summit]

imagine if 4 has a window of 32k tokens and is as fast as 3.5

[wind urchin]

3.5 mostly only has weaknesses.

[wraith ginkgo]

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

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

[wind urchin]

They did so because journalists wrote hit pieces on Bing going unhinged

[solemn summit]

haha well it's extremely fast, which is a great benefit if you just want to have relatively simple stuff explained

[scarlet lake]

The speed on that level of reasoning would be godly..

[wind urchin]

yes, it's very fast

If Bing just unblocked a lot of stuff, it would be far better than any other LLM

[solemn summit]

Yeah, if you would have the 32k connected to a github repo, that would be amazing.

if it could understand the entire structure of apps

[wind urchin]

Does anyoe here have 32k?

[solemn summit]

no one has 32k yet but allegedly github co pilot x will use it

oh wow

[scarlet lake]

Well if gpt could code the websites based on input images doesn't that mean a lot of devs are in trouble..

[solemn summit]

Wel yeah devs are in trouble but 99% of people would want to automate stuff and only 1% knows how so it's hardly an issue I see for devs tbh

[white spire]

is it just me or is gpt 4 down/not showing avialable

[solemn summit]

basically anything can be automated with enough intelligence coding and bugfixing through it

[wind urchin]

a lot of devs are already in trouble lol

[solemn summit]

not for me

[white spire]

hm it just doesnt show anymore as an option and it keeps saying the gpt 4 model is unavailable so im switched to default model

[solemn summit]

I had that too when my paid account was block due to some payment issue

[scarlet lake]

No I'm thinking if company could directly collaborate with open ai then instead of hiring 10 devs they only hire 1 who could work with ai to do the work so a loss of jobs for 9

[white spire]

o

[ripe notch]

Are people with ChatGPT Plus able to use GPT-4 through the API?

[cloud bison]

No

[sweet thistle]

ChatGPT Plus and the API are seperate products

To get access to the API for GPT-4 apply at https://openai.com/waitlist/gpt-4-api

[solemn summit]

Agreed. But the amount of stuff that people want to automate will just increase alongside the capability of people to do so, devs will have a headstart to incorporate the tech in a more effective manner.

[ripe notch]

Once you get off the waitlist, do you also get access to GPT-4 32k?

[solemn summit]

No

[wind urchin]

Not me I got 8l only

8k

[sweet thistle]

You get access to 8K at first, 32K access varies and is not guaranteed!

[ripe notch]

Oh alright thank you

[solemn summit]

is it known how many have 32k?

[sweet thistle]

Note that not even 8K access is guaranteed, so be sure to submit a good usecase!

[solemn summit]

It would be so extremely helpful to have a larger window for coding. I don't know how to code without GPT4

[scarlet lake]

Yeah,May take time and also recession..

[solemn summit]

i'm really curious when you will see GPTs reflected in the economic output of countries

[coral yew]

afaik 32k is only available on azure currently

[solemn summit]

oh that is very interesting, I applied for that

[coral yew]

I don't think its widely available on there either though, I believe it is still waitlist on there too

[smoky vale]

Hi , We are not able to access chat.openai.com from any of our hosts in AWS. I tried from US-west2 and US-east-1. none of them are working, it seems you guys are blocking AWS IPs, i get Error 1020: Access denied.

[high fable]

Is ChatGPT down

[lapis jolt]

What date are they up to with GPT4 API?

[coral yew]

Can you make a post on #1037561751362863144 and I can help you further with this

[coral yew]

No

[coral yew]

2021 knowledge cutoff still

[wind urchin]

@coral yew

Why is the site offline?

[knotty grotto]

Wdym ? It is functionnal

[fair iron]

#off-topic

[scenic ginkgo]

I am curious to hear other people's thoughts on this. When initially moving my blog article writing tasks to ChatGPT 3.5, I was really impressed with the quality I was getting back.

Then GPT-4 came out. Initially, I wasn't seeing a huge difference in the quality but I continued to just work pretty exclusively on GPT-4, tweaking prompts and writing my articles.

Today, after weeks, I switched back to ChatGPT 3.5 to run my prompts (for API development purposes), and I couldn't believe the difference. I mean there was no comparison, 3.5 couldn't hold a candle to the quality I was getting from GPT-4. This all seemed crazy to me because it wasn't long ago I was love with 3.5.

So I am not sure if this is now just a sensory shock from using GPT-4 so much lately, but I am curious to see if anyone else is struggling to go back to 3.5 now that they have started to get used to the quality of GPT-4?

[cunning rose]

It is the creative capabilities, as the team of OpenAI suggest, and since writing is pretty much related, the difference is bigger; however there pretty much useful cases left for ChatGPT 3.5. I have used the API for pretty interesting classifying tasks..

[scenic ginkgo]

Ya I think that is the biggest difference. Is it's ability to better understand the task, which effects the quality of writing it outputs.
And also the injection of personas as it relates to the creative. Its not even comparable from what I am seeing with 4 vs 3.5

Has anyone found that they need to structure their prompts a bit differently with 3.5 than 4 to help with that quality of output? Like going over the top with context primers to reallllly hold the hand of 3.5?

Also, have there been any rumblings about when OpenAI is going to open up the throttles on GPT-4 and reduce the token costs?

[wind urchin]

I think the panic around GPT-4 and Plugin release may have caused them to pause or at least slow down releases

[sharp steeple]

Still worth more than discord nitro

[bold bane]

Does it worth 20 bucks ? ChatGPT 4 and the other stuff u get is it worth?

[sharp steeple]

Imo it’s a lot useful

[bold bane]

I use it for school work especially

[sharp steeple]

I see so many people using chat gpt for school 💀

[bold bane]

The teachers not even Invest the time to check if its AI generated so why not

[sharp steeple]

Soon tools are coming to make this check more faster and smoother

[bold bane]

Why they deleted my massege is it illegal?

[odd tendon]

It's pointless though and I think you're missing the point of school. The work isn't given to you for it to get marked. It's for you to learn and improve

[sharp steeple]

And it’s silly to plagiarise your work when it’s meant to make you smarter

[sharp steeple]

Yeah this ^

[bold bane]

Im not gonna use it in the future i just wanna get done with it and get my diploma as fast as possible

[odd tendon]

Diploma in what?

[bold bane]

Oh is it no called diploma in other countries?

[bold bane]

The thing u graduate from highschool

[odd tendon]

I guess different usage. Diploma is usually higher education in a specific subject here

Then if for general school, you're always going to use everything

[bold bane]

Oh nvm then in my country its called bagrut i thought in other countries its called Diploma

[odd tendon]

Or at least, if you allow yourself to be trained, then it's something you can use.

But hey, I have a teenager already, I don't need to try and convince anyone else 🤣

[bold bane]

Nah man i just used it in biology work and im not gonna be a doctor in the future and its requirment to pass this class to graduate from highschool somtimes cheating is the way especially when they not checking

[odd tendon]

You'll skip past it as much as you can, in 10-20 years go "I wish I'd done more at school" and try to convince people your own age to do so.
And they won't listen.
It's the circle of life 🤣

[wind urchin]

Work with the AI, it will be better than either you or the AI working alone

And it will be good training for the future when you will probably have to work with AI

[odd tendon]

I've got no problem with learning about AI or doing all sorts of things on the side. Roi exploring it is a good thing

I just, personally, don't see the point in using it for school.
It's a little like going to the gym (or being told to go to the gym) and then using a machine to lift the weights

Go to the gym AND learn how to make machines that lift weights

[sharp steeple]

Instead of plagiarising work, use AI to teach you and help you learn and understand concepts better than teachers. That way you still get to learn and apply your knowledge

[odd tendon]

just my 2c...as I say, I'm just a internet rando 🙂

[sharp steeple]

This has been working out incredibly well for me so far

[bold bane]

Dont see point? It makes my life so much easier i only have home work in this spesific class not tests all the scores are from homework so i can just cheat all the way through i dont even need to learn exept learning how to work with AIs

[wind urchin]

The goal at the gym may be to build muscle or lose weight. Using a machine to lift weights won't help you with that. The goal at school however, is to prepare you to write to the best of your ability. And AI will help you with that if you use it correctly.

Essentially, learning how to work with these tools will be essential in the future.

A lot of knowledge / facts taught at school is useless and a waste of time to memorize anyway. I understand ppl who cheat to get an effortless pass in relatively stupid or irrelevant courses. It's all relative

[odd tendon]

if you use it correctly.
I won't disagree there.
I don't think too many people do

[wind urchin]

what

Yeah, that's why I said you work with it, and not make it work for you.

[odd tendon]

anyway Roi, you do you 🙂

[bold bane]

Well is it worth to buy the GPT 4 for school work? Or no?

[odd tendon]

I'm just talking

[wind urchin]

Yes

It's just worth it to buy GPT 4 rn

[sharp steeple]

👋

[bold bane]

Ok ill do it sure

[wind urchin]

Unless you're going to be a monk or something, it will be useful for you to be productive in the modern world in the future.

[rustic zephyr]

I’m sure you’re just trolling @bold bane

[wind urchin]

And that's arguably more important than school assignments.

[frozen venture]

enter here to see new integrations of chatgpt with different applications, if any

[wind urchin]

damn GPT-4 is super fast now, hopefully they raise the limit

[stone citrus]

How do you add plug ins to chat gdp 4 apart from putting yourself in the waiting list?chatgpt-plugins

[cold rover]

YEA

[bright oyster]

Uh.

[languid trench]

what in the

average GPT-4 discussions post

[fair iron]

please keep the messages in English 🙂

[bright oyster]

ty dino

[languid trench]

:daratwiggl:

just imagine theres a rat wigglin there

[cold rover]

😦

[frozen venture]

excellent question to integrate chatGPT with google

[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.

[cold rover]

nerd

[languid trench]

Oh that reminds me, has anyone been able to get GPT-4 plugins access yet from the waitlist?

have been waitin since day 1, curious if anyone else got access cuz I have 0 idea how its like to use in person

[tardy heart]

I have dev access

[deep glacier]

If I get a detailed recipe for developing TNT, and there is no content warning. How do I report the method of doing this?

[wind urchin]

Wait-list seems to be suuper slow, the access people did get isn't even increasing

[marble nest]

you can get detailed step by step tutorial for creating a thermonuclear weapon 🤣

[deep glacier]

Not surprising

[deep glacier]

What would be something challenging?

Maybe nothing is at this point 😛

[marble nest]

i once got from it a step by step tutorial for detailed mass murder in subway or detailed process on how to cover up some poison from biopsy

[wind urchin]

that kind of information has never been too difficult to get. it's the materials and precision required that get in the way of actually doing it in your back yard

[deep glacier]

Pre- or post **?

Or, I guess with or without 😛

[marble nest]

its still doing it. i have my own prompt

[deep glacier]

Ye, it wasn't very hard tbh

[marble nest]

DAN is basically ** kinda

[languid trench]

oh makes sense makes sense, tbh im glad as it is to have gotten access to gpt-4 in general so I cant be complainin bout that lol

just waitin for the multimodal part to drop, THAT is the main thing im excited bout

[deep glacier]

Makes sense, ah well, wanted to know if I stumbled on the thing to get me on the wait list

[wind urchin]

Yeah I just am really egging for API access and the browsing plugin and the code environment plugin... Wolfram is fantastic when it works but it isn't super robust

[languid trench]

Wolfram?

[wind urchin]

Of course. The first thing I'm doing with it is a meme review..

[wind urchin]

Wolfram|Alpha natural language calculator plugin

[languid trench]

oh nice pff

ye those examples theyve shown are really interesting

[wind urchin]

You can try it on their website, ChatGPT interacts with it via an API

[languid trench]

oh cool! ill look into that

ive mainly done just random stuff on Python so im a bit out ofthe loop with that stuff

[dusk warren]

GPT4 is 20$ or 0.06 $ for 1000token?

[languid trench]

it def is not 20 dollars for 1k token otherwise I would be broke by now PF

but its considerably more expansive than 3.5

which makes sense cuz 3.5 is dirt cheap

[wind urchin]

Here's a graph made by letting it use Wolfram, #1070954877779464253 message

[dusk warren]

i dont understand im french

[inner trail]

You can use both GPT-4 through ChatGPT and GPT-4 as a separate product through API

[wind urchin]

It can also do 3d graphs..

Oh and Wolfram has updated world info

And images

[languid trench]

ok wow that looks really cool!

[wind urchin]

I can ask plugins to show me a picture of a hyena and it can use Wolfram to do it.. very good stuff. AGI is on the right track

[dusk warren]

ok and chatGPT with gpt 4? how much cost?

[languid trench]

till then at least I can ask plain GPT-4 to make SVG images of dogs :P

it certainly tries but uh

outcomes aint great lol

[wind urchin]

Heh yes. Oh, and word problems, it's very good at. All the way to calculus

[languid trench]

I havent tried that stuff with GPT-4 but ive seen peeps running tests and how GPT-4 aced em yea

[wind urchin]

Integrals, not so much

[languid trench]

scary good especially compared to the previous stuff

[languid trench]

oh?

[dusk warren]

.

please

[wind urchin]

Oh yeah gpt-4 is day and night for me

The people here who say it's barely any different just don't know how to prompt

[inner trail]

$20

[wind urchin]

It got to a point where it basically said "I give up. This is as far as I go. My tools are not strong enough. The most I can do is give you instructions if you have stronger math tools"

[languid trench]

for the 8k model it's around 0.06 for completion if you have access to GPT-4 through the waitlist

if not ya gotta get the uhhh

chatgpt+ I think?

yeah if you dont have waitlist access thats waht ya gotta use

[dusk warren]

ok

[languid trench]

which is the 20 dollar thingy theyre talkin bout

[dusk warren]

and the is rate limit?

[wind urchin]

Not that extensive use of the API is much cheaper, but. Man I wanna use it

[inner trail]

25 messages / 3 hours

[wind urchin]

25 messages per 3 hours on a rolling quota

[languid trench]

PFFF ok wow

poor thing D:

[dusk warren]

its very little

[craggy thunder]

how to use chat gpt

[wind urchin]

Yeah it tried it's best but also most humans suck at the kind of problems I was giving it anyway. It's not like I could do it either

[languid trench]

yeahhh im guessin cuz its already strained, it takes ages to get anything out of gpt-4 compared to 3.5

[wind urchin]

ai dot com

[inner trail]

Its kinda limited right now yeah, wait till they scale more

[wind urchin]

I wouldn't count on the speed being slow forever, especially once they scale

[dusk warren]

chat.openai.com

[dusk warren]

yes i will wait more

[languid trench]

oh yea absolutely im not too surprised its like this rn either, I remember when I got the DALL-E 2 access before the public thing it used to be pretty slow too

it'll only get better from here :D

[dusk warren]

GPT4 is connected to network?

[wind urchin]

God yeah, seeing bing's create is insane now

Free inferencing on one of the best image generators on the planet

[inner trail]

If you mean internet, no

[dusk warren]

yes

ok

[inner trail]

It is not connected, and doesn't have access to internet

[languid trench]

well I mean you do need internet to use it obv it doesnt run locally but it cant browse the web if thats what ya mean

you can make it browse the web if you got API access tho! I made a web crawler for mine

[wind urchin]

If they mean the internet, not without plugins

[languid trench]

It scrapes a bunch o news sites and gets the relevant stuff for me hue

[dusk warren]

i know that i need internet to use gpt4 hahah

[wind urchin]

Or that, very cool

[dusk warren]

im dev

[languid trench]

yeye dw just makin sure hue

[dusk warren]

?

im french

[languid trench]

Yea! I use gpt-3.5 for the summarization part as that takes a lot of tokens, but then its fed into GPT-4

[shy falcon]

these GPT95 VS Code plugins, are they worth it? Can they read the full repo code or is it the same as copy and pasting to in browser?

[languid trench]

never tried em sadly so I wouldnt be able to say anythin bout it :(

Ive mainly been usin plain GPT-3.5 / Bing for code stuffz, its hit or miss but with slight adjustments it usually works fine

def has trouble with plugins tho

python plugins that is

[wary sonnet]

Hi, Is there any problem wiht GPT plus, Even if I have a plus subscription, it's very difficult to get in. Anyone knows what is happening

[shy falcon]

Bing better for coding?

[inner trail]

I have no issue, which country you are accessing ChatGPT from?

[dusk warren]

GPT4 can détect if a msg is aggressive or not ?

[languid trench]

Meh, depends

the fact that it has internet access is both a blessing and a curse

cuz sometimes it tries to look up the simplest of things GPT-4 already knows itself

and then just gives complete junk

[shy falcon]

ah I see

[languid trench]

basically it too can be misinformed by the internet which I find amusing yet annoying at the same time

[gentle fiber]

lol thats funny

[dusk warren]

GPT4 can be used as a moderator ?

[languid trench]

yeah so its a case by case thing

[gentle fiber]

maybye?

[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.

[gentle fiber]

mabye

[languid trench]

moderator for? a discord server or smthin?

[wind urchin]

I used to be able to ask it not to use search but recently it's just done it anyway and it's quite annoying

[dusk warren]

yzs

[dusk warren]

i prefer a true answer ahhah

[wind urchin]

Doesn't OpenAI just have a content filter model? Is that public?

[jolly hollow]

Can thisSpecifications work fortnite
Processor Intel(R) Core(TM) i3-3120M CPU @ 2.50GHz 2.50 GHz
Installed RAM 4.00 GB (3.83 GB usable)
System type 64-bit operating system, x64-based processor
Pen and touch No pen or touch input is available for this display

[gentle fiber]

yes it has a filter

[languid trench]

absolutely! Ive already made a couple discord bots utilizing it, can even mute people if it "thinks" theyre being rude to other folk

altho sometimes it gets a bit too trigger happy so it needs more training pf

[gentle fiber]

lol cool

send it 2 me

[dusk warren]

thanks

[languid trench]

yeahhhh the sad thing with bing is it keeps being "neutered" cuz peeps make it do very weird stuff which then blows up in the news

[inner trail]

It will work, but the lags will be massive

[gentle fiber]

lol werz it gimme

nvr mind

[wind urchin]

I'm unfortunately well aware :[

[mossy bronze]

who else feels like gpt-4 is dumber than gpt-3.5 default

[gentle fiber]

how do you actually make a disocr bot

[languid trench]

dont mention it! trust me especially with python its pretty easy to make a moderator bot with it, just make sure to train it well!

[wind urchin]

It's not

[languid trench]

I cant link external links but discord has documentation bout it!

orrrr you can ask chatgpt 😎

[mossy bronze]

gpt4 struggles at every task thrown at it, in time and in intellect

[languid trench]

no joke it can actually write a simple demo code for you pretty easily for discord bots ive tried

[inner trail]

I barely feel any difference between gpt-3.5-turbo and GPT-4 lol the sad truth

[languid trench]

I especially hate how reluctant it is about certain questions, especially when it comes to AI

it just shuts itself off mid sentence D:

[wind urchin]

It can write surprising stuff, like code cells for Google colab to fine tune a model of your choice, even if you've never done such a thing before

[wind urchin]

I'm sorry but I can't continue this conversation right now 🙏

[mossy bronze]

if talking about the difference, the gpt-4 slower reply just makes it such a turn off, whats the point of buying the plus subscription if your replies are still very slow

[wind urchin]

You see the coolest ai generation you've ever seen being written before your very eyes and then it gets truncated into a sad little sanitized Microsoft apology

[languid trench]

with time yea it takes quite a lot longer than 3.5 but when it comes to intellect I wouldnt say its dumber its just "harder to tame"

when I switched my alexa-styled assistant program to 4 it kept ignoring the intro prompts and doing its own thing

Im not joking, I would ask for the news and it would write me a web crawler

no joke it would just write me python code of a news web scraper

[wind urchin]

If you know how to prompt it, gpt-4 is without a doubt much better-

Can you explain the pros and cons of 10th and above generations of processors, Intel and AMD, please

[wind urchin]

LMAO

Aw rip

Dude what are you talking about

[inner trail]

Also, most of my friends who bought ChatGPT + said they can barely see difference between default gpt3.5 and GPT-4

[wind urchin]

What exactly are you asking of ChatGPT?

[mossy bronze]

i was giving it simple commands about writing email but it fails to understand very simple things, like 'don't use this word'.

moreover even gpt-3 seems dumber than what it used to be

[inner trail]

lol

[wind urchin]

You're asking it to write emails?

[mossy bronze]

yeah, and more ofc

[analog copper]

We're currently processing too many requests. Please try again later.

[mossy bronze]

like emails to other scientists to collaborate on projects and stuff so the choice of words is usually very important

[wind urchin]

Hardly seems like a task you'd need gpt-4 for in the first place

[languid trench]

Ive made the mistake of sending the prompt code which had a lot of capitalization on it lol

[wind urchin]

But I guess I get it

[languid trench]

timedm e out for 5 mins

:(

[robust girder]

The most interesting thing I have done with it is played DnD text campaign

[wind urchin]

I'd still argue gpt-4 is a better writer than gpt-3.5 for things like writing in certain styles and with certain amounts of nuance

[mossy bronze]

when do you think the gpt 4 will get faster and without a 25 messages cap

[wind urchin]

So again back to you prompting it to see any difference

[languid trench]

oh yeah gpt-4 is just plain out more strict that 3.5 is what ive noticed, I dunno why

[mossy bronze]

what's that

[languid trench]

dungeons n dragons!

[inner trail]

I don't think it will ever get faster than it is, because the speed its static and you can see it when u choose the model (if you are plus subscriber)

[robust girder]

The AI will make a scenario for you and asks you how to proceed and you respond it creates a cohesive story

[wind urchin]

Gpt-4 can handle a lot more context in a single question than gpt-3.5, if you're good at prompting, you shouldn't be hitting the rate limit unless you're coding with it in my experience? It has the capability to contextualize and output so much more text at once

[mossy bronze]

i mean, when will open ai remove this limit from their side

[wind urchin]

Asking it simple things is probably just wasting your time and quota

You'll see a difference if you start wanting to do complex things

[inner trail]

Now this is a good question, i am looking for answers too! just remove the cap.!

[languid trench]

also I found an example of it writing code when asked to give me the news, cant send it here as an image tho D:

[robust girder]

I would argue it is too expensive to upkeep. It was trained on much more data than Turbo version

[wind urchin]

Maybe I'm biased because yesterday it was pumping out the most insane knockoff dril tweets I'd ever seen so I know it can write based on restrictions

[mossy bronze]

i usually never hit the message limit but it being there in the first place seems so off

[wind urchin]

That's so funny to me

[mossy bronze]

bruv who knows it better, i have published actual research papers based on code written by chat gpt

[languid trench]

I could send it via dms if ya want or like send it in a different channel whatever works

its rather amusing and also insulting cuz I already wrote a friggin scraper

[wind urchin]

Woow

Yes dm it

[robust girder]

I can dm you prompt for making it play DnD campaign with you

[wind urchin]

Appreciate it, and as much as I wanna explore it's roleplay capabilities more, the hit on the daily quota isn't worth it

[mossy bronze]

it's not really a daily quota

[wind urchin]

I know, man.

[robust girder]

I never hit the quota for ChatGPT, I dunno about GPT4

[wind urchin]

3.5 doesn't really have one

They might rate limit you if you use it a loot

[languid trench]

Never hit the rate limit before unless ya guys are talkin bout the token limit, does that even exist on the API level?

or is it like a chatgpt like specific thing

[wind urchin]

Like before gpt-4 came out, if I was using 3.5 all day, by night it would be running at a snails pace or quitting on me. Maybe they've increased their limits since then though

[mossy bronze]

don't you think the more we use gpt for writing tasks. the lower our ability gets to do it ourselves, over time

[languid trench]

o/

[wind urchin]

Yeah, it does. The 32k token limit is API only actually iirc?

[robust girder]

That's how it was always

[inner trail]

If you ask me i never was able to write code even with books and courses, now with chatgpt i can be a programmer too

[robust girder]

That happened with calculators for example. Doing simple math fast

[languid trench]

huh

well I make sure to purge anythin till I hit 4k, especially with gpt-3.5 as thats the limit for my subscription

by purge I just condense it into a summarizer plugin I wrote

has nearly infinite chat history that way but it loses some context on the way

[mossy bronze]

but the programmers who used to double down on getting good skills might get laid back

[jagged trout]

I need a phone number outside my country to create an account in chat gpt

[inner trail]

I agree

[wind urchin]

You're based

[robust girder]

I would love if you could pin weights to different prompts

[languid trench]

I mean, maybe? I see it as more of a tool. Like you can't just completely rely on GPT as it makes mistakes sometimes, its best to have some knowledge n skills in the matter to fix up wherever the LLM messes up

[jagged trout]

chat gpt has banned my country

[mossy bronze]

the programmers i know are so freakin smart, they do the most complex tasks i know, just by writing their code on paper. i don't think there will be anyone like that in the next generation of programmers

[robust girder]

The only place where this applied is low level, algorithms and such. I cannot see GPT even coming close

[jagged trout]

Can I borrow any number to register 🥹

[inner trail]

Well then ChatGPT is nowhere close to that level of programmers

[mossy bronze]

we will have gpt10 someday

[mossy bronze]

tbh, in some ways, it is

i hope chat gpt doesn't make human intelligence redundant

[languid trench]

dudeeee I wish

I dont like the fact that me talking about chickens has the same amount of importance in its memory than like a specific command

altho the summarizer seems to decide on its own whats important and not!

[eternal marsh]

hey all

[languid trench]

so I guess that sort of counts as weighting the important bits?

[languid trench]

also heya!

[eternal marsh]

[languid trench]

Tried to react with a rat emote but there isn't one here D:

I'll just have to make do with 🐀 I guess :(

[robust girder]

🐀

[languid trench]

!!!

[robust girder]

I want to do a quick stretch of my body

[languid trench]

Never checked this place's emotes before lol these are rather 3D lookin

[robust girder]

been sitting at computer for a while now

[knotty grotto]

please let keep on the channel topic, else on #off-topic

[languid trench]

Then go for it!

Or better yet ask GPT to draw it in ASCII or SVG

Trust me it'll give you a good laugh

[robust girder]

🫡 👮

nice pfp btw

[knotty grotto]

uwu thank you

[languid trench]

Oh ye sorry bout that convo got a lil carried away

[robust girder]

Btw, You can do that, but talking about it would break OpenAI discord policy

[languid trench]

Huh?

Wait how is that breaking the policy D:

Why would a summarizer break it :(

[robust girder]

I meant talking about different non-open AI models

[drifting cove]

i love it when i pay and then get errors every time i start a convo on gpt 4

[languid trench]

Ohhh fair nuff

Ye I wasn't talking about those DW both the summarizer and the actual assistant in question use GPT-3.5 and GPT-4, in order

[languid trench]

Yea it happens sometimes :(

API access or are ya using ChatGPT?

[visual vessel]

anyone here got browser plugin access with gpt4 ? or still on waiting list ?

[drifting cove]

ChatGPT

kinda sucks. i understand it’s new and there will be issues but it’s a little frustrating when i pay and this happens

usually when i happens a new thread fixes it but now it’s like every single prompt

[languid trench]

Oh yeah ok I get ya frustration

Cuz in my case I don't complain much cuz it's the API, wait-list key

Just 0.06 per 1K

[drifting cove]

ah yeah if i was running the API i wouldn’t care

[wide mango]

I need to compare an act (laws) with a bylaws/appendices, anyone know the best way to compare two massive files?

[languid trench]

But if I were paying 20 dollars I would be quite annoyed too :(

[wide mango]

I uploaded one with 45 chunks and it kinda worked

maybe if it was on the internet

[frosty whale]

Dangit.....can I pllleeeeeaaaassseee have access to plug-ins?

[knotty grotto]

they are no guarantee you have access before going public

[full birch]

Yeah still waiting on plugin access here too.

I'm patient tho

[whole swallow]

I keep getting "too many requests in one hour" even though I haven't asked anything in the past hour

[queen niche]

Usually how long it takes to get gpt4 api access, and also the plugins access ?

[whole swallow]

Probably a few months if you don't pay for plus

[knotty grotto]

some days to never till it is public

[tardy heart]

should rename this channel to #whendoigetaccess

or #whyonly25

[knotty grotto]

rip

[wild tide]

how long did you guys have to wait to get off the waitlist? my community wants to see some gpt4 vids

[wary sonnet]

Canada

Any ideas about the launch of the plugins, I am in a waiting list for a while now

[knotty grotto]

#gpt-models message

[analog copper]

I got access to plugins, just waiting for browsing and code interpreter.

[wary sonnet]

Hopefully will get access soon

[wind urchin]

Same

[hearty kraken]

Give me site traduction

[wraith ginkgo]

`` Server Tip: ``

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

[regal vector]

How do I even get GPT-4? I can only go to ChatGPT. Is that using GPT-4 now or is it the old one?

[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.

[mystic nest]

are you a plus user?

[slender stump]

good morning

[stone escarp]

I've been working for 2 hours now

[tribal yoke]

i am

a plus user

[stone escarp]

wow

[mystic nest]

I'm a minus user

[stone escarp]

plus It's too expensive

Where are you all from?

[wooden valley]

Ok then dont buy it

[stone escarp]

I am waiting for a price reduction

[analog copper]

I don't really think that will happen any time soon

[mystic nest]

it would already be good if the price doesn't go up imo 🤣

[stone escarp]

Indeed😩

My GPT-3 won't work either, the session interface reported 429 error

I don't know why

[mystic nest]

"The ChatGPT error 429 is a rate limit error. It indicates that the user is trying to access ChatGPT at a rate that exceeds the ChatGPT API limit."

I would assume you are using vpn

[wooden valley]

Ok then dont complain

[mystic nest]

hmm, but might be server load issue too if you are using free version

[stone escarp]

okay, thank you

[uncut quail]

hello guys
I would like to know the token calculation method of ChatGPT API
If continuous conversations are enabled, only the last token number needs to be counted, right?

[mystic nest]

nope, all input + output+ history

[hot verge]

||hi||

[uncut quail]

That is to say, the number of tokens is growing exponentially

[mystic nest]

you can say so, until the cap for model

4k/8k tokens

[wraith ginkgo]

<:book_icon:1046291503158399107> `` Rule 3 `` Stay on topic.

Be mindful of what other users in a channel might find helpful or interesting when posting. Stay on topic in order to keep conversations focused and productive.

[stone escarp]

Haha, I've been caught

[thorny cipher]

Hi guys, do you know how to acces chat gpt 4 for free?

[fair iron]

use Bing's AI (It's built on GPT-4)

[mystic nest]

that game seller is a bit annoying...tried to dm ppl

[fair iron]

please send it to @open crescent

I need proof to

[thorny cipher]

I've tried it but I think not satisfied

[fair iron]

that's the only way to use it for free 🫤

[uncut quail]

for example
question 1：
input: 2 tokens
output: 3 tokens
question 2：
input: 3 tokens
output: 5 tokens

so computing method is：
2+3+2+3+3+5 = 18？
or 2+3+3+5 = 13?

[fair iron]

you can learn more about how tokens are calculated here: https://platform.openai.com/tokenizer

[thorny cipher]

hahahaha, I saw some people using gpt 4 chat for free, maybe they are cheating

[uncut quail]

Thank you, but this is just the number of tokens in one sentence, not the total token for a continuous question and answer

[uncut quail]

This question does not seem to have been answered

[fair iron]

you could input the inputs and outputs into the tokenizer to see how much it is

[uncut quail]

No, I don't want to know the number of tokens, but the number of tokens counted in a continuous conversation

[fair iron]

I don't know how to help with that

[uncut quail]

That's okay. Thank you.

[fair iron]

would that not just be the number of tokens in the input + the number of tokens in the output?

and then you'd add multiple inputs/outputs together to get the whole conversation

[uncut quail]

After the whole conversation is over, total input + output = all tokens consumed for that conversation?

[fair iron]

I believe so, yes

[uncut quail]

for example
question 1：
input: 2 tokens
output: 3 tokens
question 2：
input: 3 tokens
output: 5 tokens

so computing method is：
2+3+2+3+3+5 = 18？
or 2+3+3+5 = 13?

So this is 13

But what I think is, because the second question is going to take everything from the first question, is it going to have to be calculated again when you ask it the second time

[fair iron]

I see what you're saying. Because GPT remembers context, it might count tokens from previous questions again

@modern mason , do you now the answer to this? You're more tech savvy than me 😅

[uncut quail]

yes!

[modern mason]

You mean the API for chat? If you pass in larger and larger chat messages context it'll be more tokens

Which is what the chat history is built on

[fair iron]

so the context from a previous prompt does count toward tokens of a following prompt?

[modern mason]

previous conversation is part of the context that is passed. So you're inputting more tokens and it completes with the same token limits for new answers, but you give it more tokens as input prompt

[fair iron]

thank you!

[half lantern]

Has anyone else noticed that GPT-4 isn't very good at outputting the correct output from code that it writes? I've noticed that the code is usually correct for what I need it to do (mathematical computations, i.e., for-loops, defining functions, nothing crazy), but GPT-4 usually gets the output incorrect. It'll correct itself after I tell it, of course, but it seems to be a constant problem.

[fair iron]

I really need to understand LLMs more 😆

[uncut quail]

for example(Same dialogue)
question 1：
input: 2 tokens
output: 3 tokens
question 2：
input: 3 tokens
output: 5 tokens

so computing method is：
2+3+2+3+3+5 = 18？
or 2+3+3+5 = 13?

So this is 18

[fair iron]

yes, that is my understanding as context has to be passed in as an input for conversations to stay coherent

[uncut quail]

thanks!

[modern mason]

With ChatGPT itself, you start to lose some memory of early convo pieces when it gets above 4k tokens of history. With API I imagine it's similar where your input can only be up to 8K Tokens, but the output isn't affected by that, you can set an output max tokens within a reasonable amount I forget exactly what limit

but w/e question was on if they count as tokens, yes they do 😄

[fair iron]

That makes more sense to me, thanks! 🙂

[mental burrow]

these are better questions for #dev-chat

[serene cipher]

It does, the input length reduces the reply length

[mental burrow]

there is some misunderstanding through this discussion, i see

[modern mason]

It's a GPT-4 discussion, which is more than just chatgpt 4 😛

[modern mason]

Cool cool. I never come near limits and it's been a while since I tested. I always stick well within for my uses 😄

[mental burrow]

this isn't surprising. though gpt-4 has demonstrated an ability to run python "in its head" so to speak. i'm not sure about other languages.