gpt-models

[golden basalt]

haskell, prolog,

its really good at all languages really

[neat ruin]

any perl programmer still around?

[naive burrow]

It’s cra cra

[tidal sail]

Thanks for proving yourself wrong while trying to prove me wrong.

[glass ridge]

Killer

[limber garnet]

meh, not really. It gave a contradictory answer about the Musk case

[naive burrow]

Cutting time for cutting edge code

[tidal sail]

It states exactly what I've said:

There can be exceptions where the burden shifts to the defendant to prove their innocence. These exceptions are relatively rare and depend on the specific legal system and circumstances of the case.

[glass ridge]

Where did you say this?

[tidal sail]

@glass ridge

[glass ridge]

⛽ 🔥

[limber garnet]

In this civil case, the burden of proof was primarily on the SEC to prove that Musk's actions were fraudulent. However, Musk may have had to provide evidence to support his defense and counter the SEC's claims.

Pretty straightforward. The burden of proof is on SEC, Musk has to refute them. Very hard to misinterpret

[glass ridge]

Bruh

Both of you are right actually

[limber garnet]

that makes little sense

[tidal sail]

Yea, I now conclude that he actually does not understand my point is now saying the same I said

[glass ridge]

youre talking past each other with different examples that each have their own applicable law ergo youre both right

[limber garnet]

I didn't really give any examples ngl

Katt and I are a little confused. You mentioned that there are exceptions and that Musk had to prove innocence, yet "the burden of proof was primarily on the SEC"? This seems contradictory and I'm curious as to who actually had to prove what.

I apologize for the confusion in my previous response. Upon further clarification, I would like to correct my statement regarding the Elon Musk case. In the case of Elon Musk and the SEC, the burden of proof was indeed on the SEC, as they were the party bringing the claim. They needed to provide sufficient evidence to prove that Musk's actions were fraudulent.

Musk, as the defendant, did not have the burden to prove his innocence, but rather to present a strong defense to counter the claims made by the SEC. In this context, providing evidence to support his defense may have been misconstrued as "proving innocence," but it is essential to understand that the burden of proof still primarily rested with the SEC.

So, in general, the burden of proof lies with the prosecution or plaintiff, and it is their responsibility to provide sufficient evidence to prove their claims. The defendant's role is to present a defense and counter the claims made by the prosecution or plaintiff.

[glass ridge]

No youre right you made a statement with no example

[neat ruin]

I remember when this was a channel about gpt-4 discussion 😢

[obtuse sandal]

What's Prompting?

[limber garnet]

we'll use the Musk case since that's what we're already- yeah we drifted off topic didn't we lol

[glass ridge]

coaxing the AI

[glass ridge]

this just proves both of you are right because of the exceptions to the rule...

fun

[limber garnet]

there is no exception, all defendants are supposed to defend themselves in court

[glass ridge]

I thought the discussion was about the burden of proof

[limber garnet]

that's correct

[tidal sail]

There are exceptions. You posted that yourself.

[limber garnet]

read the follow-up discussion with v4

[glass ridge]

the onus is on the one making the claim

[limber garnet]

You're free to ask it further questions about burden of proof. Or a lawyer. I'm not going to argue about basic us govt concepts with literal children

[tulip blade]

So there was a massive leak of openai user credit cards and chat logs.

[glass ridge]

ergo defending yourself is not necessarily showing the burden lies on the defendant

[golden basalt]

good job i use revolut

[tidal sail]

I primed ChatGPT v4 to be JudgeGPT and it's generating an answer now. Keeping you updated.

[golden basalt]

delete my card, make new one

[tulip blade]

And OpenAI has zero answers on their help page.

[fading canyon]

?

[junior elbow]

only 1.5%

[limber garnet]

katt you aren't from the US are you.....

[paper saffron]

Does the playground offer longer text input/memory then chatgpt4?

[wraith ginkgo]

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

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

[tulip blade]

Turns put OpenAI stores credit card info on a Redis server which was scraped.

[blazing patio]

Source?

[glass ridge]

Youre still making two different arguments, youre saying the defendant has to defend themselves.

Katt is saying the BURDEN of proof lies with the plaintiff.

[junior elbow]

OpenAI has fixed the BUG

[glass ridge]

Two different arguments unless im r word

[limber garnet]

Katt has been saying this entire time that the burden of proof is on the defense....

[tulip blade]

Can’t add links here.

[blazing patio]

DM

[limber garnet]

the defense having to defend themselves is not the same has having the burden of proof. two completely different things

[glass ridge]

agreed.

[tulip blade]

Storing credit card info on Redis is something I expect out of Wordpress based Pizza Delivery stores. Its bad.

And I can’t find out if I was scraped.

[nocturne viper]

Google it, last four digits of some cards were exposed for a few hours. Big whoop.

[tidal sail]

For 99% of the times it's on the plaintiff or the state but there are exceptions. What don't you understand about that?

[glass ridge]

Hilarious that an AI company has a data breach

[blazing patio]

That's why I use Privacy lol

They got a digital card

[mild pendant]

It's officially 1 week since I joined the waitlist 💀

[tulip blade]

At least they could tell me if my info got scraped.

[glass ridge]

As I previously pointed out yall are making two different arguments and are both right

[limber garnet]

There are no exceptions. Whenever someone brings a suit against a defendant, the plaintiff must prove more likely than not that the defendant did the tort, and the prosecution must prove beyond a reasonable doubt that the defendant is guilty of a crime. That's always how it is and there aren't exceptions at all

[tidal sail]

I know, he just does not seem to understand my point and is referring to completely different things

[mild pendant]

DM me the details bro 🤣 let me know how you discovered.

[limber garnet]

I'm not entirely sure what your point is honestly, but it seems to be that when the stock market is involved that the burden of proof is on the defense which is not true

[glass ridge]

BUT Katt... your argument rests on the superminority of cases sooooo

^^this

[sweet rover]

dm the stuff to about the leak@tulip blade

[glass ridge]

Ive enjoyed the argument, fun times.

🫡

[limber garnet]

it's always fun to sit and watch the world burn ngl

though, it's less fun when you're in the heat of the flames :p

[glass ridge]

Turned out quite cordial at all times though. Which is nice.

[tidal sail]

So what? This whole discussion emerged from a joke

(the joke wasn't even funny I must confess)

[limber garnet]

the funny thing is I didn't even realize it was a joke until now and it kinda makes sense

[tidal sail]

lol

[glass ridge]

If a joke is not funny does that make it merely a statement or does it retain the status of a joke?

kek

[tulip blade]

Sendt

[glass ridge]

😂

[tidal sail]

It was fun in my head for 2 seconds while I wrote it

[tulip blade]

Credit card info and chat logs.

[tidal sail]

pepega

[tulip blade]

That what were scraped.

[tulip blade]

Sendt

[glass ridge]

ah one of those, the headmovie is always amazing until the limits of language destroy the magic

Something to be said about this in the context of LLMs

[limber garnet]

kinda reminds me of people acting like a Disney movie irl and ending up in cringe compilations

no offense or anything

[glass ridge]

tis what makes us human

[sweet rover]

Why is redis so important to chatgpt?

[tidal sail]

Language is a rather unprecise way to communicate. At least thats what I learned working with ChatGPT.

[sweet rover]

just curious

[limber garnet]

Redis is a database. It stores data

[glass ridge]

wen bitcoin payment for GPT?

[sweet rover]

isnt it a ram database?

[limber garnet]

you could always buy a $20 preloaded card with crypto

not sure you'll have luck spending it on anything here tho

[glass ridge]

admin bro... admin.

[limber garnet]

?

[glass ridge]

slang for too much work where im from

[limber garnet]

where I'm from it's slang for the guy who created the Teamspeak server

[fair iron]

I'd be surprised if this is added

[tidal sail]

There are one time use credit cards with privacy in mind. I guess he's not referring to stolen cc's.

[limber garnet]

yeah no don't do that!

[glass ridge]

what speaks against it?

[limber garnet]

well, probably the moderator who's making sure I don't promote that sort of thing lol

[tulip blade]

Yes

[tidal sail]

Law, and you need to prove your innocence /s

[glass ridge]

in the age of AI you really want a centralised currency that can be printed at will by a bunch of humans that dont even know why 2% inflation is their target

[sweet rover]

i assume when they restart they dump to a dump file?

[glass ridge]

LMAO

[obsidian urchin]

Hello, if i may know, how long does it take to get access to GPT-4 , anyone has any experience ?

[fair iron]

I'm a volunteer. I don't know if OprnAI will add it . I don't think they will though since they haven't done it already

[fair iron]

To the API? I do not know (I'm on the waitlist too)

[sweet rover]

want a key 🧌

[obsidian urchin]

@fair iron Yes, Thank you

[whole bolt]

If I reached the current usage cap for GPT-4 and click on using the default model, will I be able to return to GPT-4?

[glass ridge]

FAAFO

[limber garnet]

not if you're still maxed out

[fair iron]

Absolutely not 😅. Don't share your api key!

[limber garnet]

you have to wait a little bit

[sweet rover]

theres a reason i said 🧌 at the end

good luck getting a key from mey

[whole bolt]

You've reached the current usage cap for GPT-4. You can continue with the default model now, or try again after 10:06 PM. Learn more

So at 10:06PM on that same chat should I be able to go back to GPT4?

[wind urchin]

Did anyone actually get api key? I have on the waitlist forever

[glass ridge]

I think we are all FAAFOing in here. Playing with our overlords, thats why I say please and thank you when prompting

[tidal sail]

yes

[limber garnet]

I'm going to AI heck man, I'm always very direct when prompting. WRITE AN ESSAY RIGHT NOW! AND A PRESENTATION!

[sweet rover]

yea i did

[mild pendant]

Usage of the GPT-4 model isn't tied to a conversation though.

[limber garnet]

ok not quite like that but close

[timid summit]

Guys is it permitted to have 2 accounts?

[limber garnet]

You would need another phone number to verify with

[timid summit]

What happens if I use the same number, then I purchase a subscription on both accounts, does this result into a ban?

[tidal sail]

Whats your guys take on AI in the stock market? Do you guys think that our world can keep that system if everyone can make a sentiment analysis, algo trading and predictive modeling with the help of AI?

[wind urchin]

How long have been on the waiting list? I am there for more than week though nd I really want the more accurate model in my program.

[limber garnet]

I don't think you'll be able to get that far. That sounds very expensive, though

[glass ridge]

buy PLTR and retire early

[sweet rover]

did it same day it released, 3/14/2023

[wind urchin]

Ohhh interesting

[sweet rover]

its just invite waves 🤷

[wind urchin]

When did u get the key though

[nimble reef]

Why is it that even after making my max_tokens 8000, my answers are limited to a character count of apprx. 3200 characters? What do I need to change in my code to get the max tokens of 8000?

[sweet rover]

today

[sweet rover]

each token is about 4 english letters,

[wind urchin]

Damn Noice.

[sweet rover]

yep

[tidal sail]

nothing, just type "continue" after it cuts off

[glass ridge]

Ummm funny irony here, when ALGOs trade against each other, you as a human can pick up on the patterns and trade discretionary picking up their patterns

[tidal sail]

oh, you are talking about the api right?

[nimble reef]

Yes, in the API

[timid summit]

How can I check what number is registered?

[glass ridge]

Scam

[tidal sail]

lol true, time to create an AI based anti AI trading bot

[glass ridge]

peace

[limber garnet]

referring to spammer, sorry. You'll have to try harder to get rid of me

[nimble reef]

I know. I tested it. It comes to Tokens: 707 Characters: 3329

[limber garnet]

:D

[glass ridge]

trading is a game that you win against yourself rather than market participants... its the reason 95% fail to make money from it

[sweet rover]

weird,

[tidal sail]

have you tried stating things like "make the answer about 5000 characters long"?

[tidal sail]

I'm playing myself every day, should be easy 😎

[glass ridge]

im off to bed good night

[boreal epoch]

Are there any plans to make GPT4 available to free users in a limited form?

[tidal sail]

No. Nothing announced at least.

[boreal epoch]

ok

[halcyon cliff]

can GPT4 read a video file?

[tidal sail]

no

Well, maybe it can, but we can't use it that way

[sinful oracle]

how to use chatgpt4？

[full patrol]

any news about GPT4 for GPT plus subscription ?

[tidal sail]

https://chat.openai.com/

And pay $20 a month

[tidal sail]

wdym?

[crimson veldt]

Yall ready for GPT5?

[tidal sail]

probably takes 2 years from now

[crimson veldt]

lol thanks

[full patrol]

how to use all features announced

in GPT 4

[tidal sail]

you mean plugins?

[full patrol]

nn, for example how to use images in GPT4

[hollow trail]

wait until they publicly roll out

[tidal sail]

You just can't, yet

[hollow trail]

only wait

[full patrol]

ok thanks

[hollow trail]

or actually you can just buy OpenAI

and ask CEO for it

[tidal sail]

thats what microsoft did

[hollow trail]

lol

[mild pendant]

no

[hollow trail]

not even getting plugins yet

[heavy totem]

yes, because by that point gpt4 limit will be lowered

[warped shard]

Have the limits been modified

I want to know if they have removed the message limit so I know if I can subscribe

[junior elbow]

25/3H is not enough

[warped shard]

Well it’s not bad but I should be getting unlimited messages if iam paying 20

[wraith ginkgo]

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

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

[tidal sail]

They paid millions for training and 25.000 GPU's. $20 is a good price.

[feral pendant]

is gpt-4 invite only or can you subscribe to it already?

[warped shard]

You can subscribe for a limited usability where you can only send 25 messages every 3 hours and you have to pay 20

Per month

[zenith tendon]

For now

[warped shard]

I hope it’s only for now

[junior elbow]

when gpt4 available for free user, plus should have other advanced features

[long pond]

ChatGPT Plus gives you access. API is waitlisted.

[feral pendant]

so wont work well yet with msvs codegpt for example

[zenith tendon]

It is for now. It hasn't been fully rolled out.

[outer gulch]

max tokens includes BOTH the input and output

[feral pendant]

oh, it wont be a subscription base but token base?

[zenith tendon]

Wait for GA, George Dubs.

[junior elbow]

when 32k gpt4 is available

[hoary crest]

I felt like there was some false advertising, since pictures can't be input yet into chat GPT 4

[grand lynx]

If I click ChatGPT on the OpenAI site, it shows ChatGPT Mar 14 Version at the bottom of the screen but I can't tell if it's version 3 or 4. If it's 3, do I still have to join a waitlist for version 4 or will I eventually just get access? Thanks.

[zenith tendon]

You should see a drop-down in your GPT conversations giving you the ability to select the model you want to use. Version 4 is an explicit selection. If you don't see it, you need to pay for Plus.

[oak spire]

Anyone need help with gpt 3

[outer gulch]

If you're using free version it's 3.5 turbo. If Plus, you have choice of 3.5 or 4

[grand lynx]

yeah, I don't have a drop down at all. Thanks!

[grand lynx]

If I just join the waitlist, will I have to pay for a subscription?

[zenith tendon]

https://help.openai.com/en/articles/6925682-chatgpt-plus-access

[outer gulch]

Waitlist is for access to API, where you pay per token. Plus gives instant access to 4 but limited to 25msg / 3hrs and 4096 token max

[dreamy turret]

💯

[grand lynx]

Hmmm. Okay, thanks.

[junior elbow]

hope openai will remove the limits

[zenith tendon]

They will when it goes GA.

[keen coral]

How do I use the image recognition on GPT-4? Is it released to the public yet? I'm have Plus and also just got access to the GPT-4 API but don't see anything on it

[dreamy turret]

Why can’t chat GPT access my personal information is pointless for an artist

[zenith tendon]

And they figure out their infrastructure issue. Kind of an unprecedented scale challenge/opportunity in the history of SaaS.

[junior elbow]

then we can maximize productivity

[outer gulch]

infrastructure has to grow massively to satisfy demand, both corporate and freemium/Plus users

[zenith tendon]

Yep

[autumn patio]

how do you make it generate an image?

[zenith tendon]

You don't. You get it to give you a prompt for third party image generation software (or Dall-e).

[mellow pebble]

I'm trying to help visually disabled folks, can chat gpt-4 provide image descriptions?

[zenith tendon]

The preview version of GPT-4 cannot. But yes, the GA version can, and likely the API integrations depending.

[mellow pebble]

Hmm, pretty new to this - can you give me a lead of what to start google/learning more of?

[tidal sail]

You are somewhat late to that game. There are already a few companies who work close with OpenAI, working on exactly the same.

[mellow pebble]

That's exciting! Do you happen to know if it's open source / open to public use?

[zenith tendon]

Can't give external links here.

[mellow pebble]

Fair enough!

[tidal sail]

I don't know any. The full functionality of GPT4 is pretty much not accessible to public.

[mellow pebble]

That's all good - I appreciate your time.

[crystal helm]

Are there any information if it will be accessible to public in the future or not?

[zenith tendon]

We're all speculating. Just being honest.

[covert yoke]

Error with OpenAI API request: connect ETIMEDOUT 104.244.43.136:443 openAI demo project show this error

[tidal sail]

Image and video input will most likely be accessible in the future, else I don't see a reason for the demo video. But no one does know.

[keen coral]

I can't find anything in the API documentation that mentions image recognition. Am I missing something?

[tidal sail]

You are missing, that it's not available to the public

[keen coral]

Oh I see. Any idea on when you think it will be available?

[zenith tendon]

GPT-4 will be able to "see" by processing image inputs and describing them textually. This feature was governed out of the preview.

[fleet finch]

how 2 plugins

[tidal sail]

There is no ETA. No one even knows if its available to public ever.

[unique vortex]

WHY

[zenith tendon]

True enough, but economics suggests it will be. And the OpenAI relationship with Microsoft (i.e., Powerpoint).

[outer gulch]

Why what?

[tidal sail]

I guess so too. But it's just speculations.

[unique vortex]

Who can give me a plus, whoever is a handsome guy, a beautiful girl

[zenith tendon]

Agreed. Everything is speculation. OpenAI created something that's just that good. We're all speculating.

Suddenly crickets.

[zenith tendon]

ha

Any data scientists, architects, or developers in the house?

[hollow meadow]

I developed gpt 3.5 and gpt 4 into a discord bot, sms texting env, and other things.

[zenith tendon]

How are you guys discussing/positioning GPT with the business and vice versa?

[tidal sail]

Nice. Actually you can do that by prompting GPT4 to create a discord bot and contacting the API by pasting the docs. Takes like 20 seconds and you got your own gpt4 discord bot.

[hollow meadow]

I guess im confused by your question.

[zenith tendon]

Particularly in terms of trust---same old ML black box issues.

[hollow meadow]

Basically just taking input of a users prompt, sending it to the api, and grabbing a response back, and also saving it to a text file for convo history.

[fleet finch]

How can I use Chat gpt 4 better? in what aspect is chat gpt 4 better and in which aspect gpt 3 is fine

[hollow meadow]

Along with a discord/chat tailored instructions for both models. I have it all on my github if you want

[boreal epoch]

GPT4 takes way more computation power to make it's output. Making it cost way more to run each time. They will need limits like that until dedicated AI chips make language model equations as cheap as 3d rendering ones.

[zenith tendon]

Got it. What I mean is, like, helping different industries or corporate lines of business think about developing large language models, and how to do that process-wise, and the trust issues involved in any machine learning capability.

[hollow meadow]

GPT 4 for complex things that require reasoning good math skills etc.

GPT 3/3.5 for simple, everyday chatting/tasks

[cold finch]

How to use openai

[hollow meadow]

Uhh, im not sure. Not really my field. Im just a solo dev haha

[hollow meadow]

go to your web browser and search, "ChatGPT"

[zenith tendon]

Got it. I misunderstood. Solo developers are awsome. Good for you.

And I can't spell tonight.

[hollow meadow]

But everything is open sourced on my github if you want

[boreal epoch]

chat openai com/chat is the url for ChatGPT. I maimed the URL's formatting so the bot does not block the message.

[hollow meadow]

github/AverWasTaken/averGPT

[zenith tendon]

Thanks. I'll check it out!

[hollow meadow]

No problems

Plot twist @zenith tendon. Most of the debugging was done by GPT-4 itself haha!

[cosmic arrow]

hello somebody know the link of chat gpt on spanish ❓❓❓

[hollow meadow]

go to browser, search chatgpt. it will come up

[zenith tendon]

I run an innovation lab for a global law firm. Trying to figure out how to integrate GPT into the products/offerings I've developed...not technically, but culturally. Lawyers think AI is magic. Any general corporate change management/change communication/change strategy advice?

Besides "hire Deloitte"

[limber garnet]

you could use it as an assistant and make some helpdesk staff obsolete

[zenith tendon]

Right on

[swift bison]

Show them that spellbook app? its at spellbook.legal

[hollow meadow]

Honestly, its all about tailoring the model to YOUR needs. For example, for my discord bot i've prompted the bot as such.

"You are an AI discord bot inside of discord. Provide helpful advice that makes logical sense to users inside the server."

Maybe for something that involves your law firm, you could tailor the models to your needs, or even use OpenAI's fine tuning features to further enhance the models capabilities. Some example prompts could be,

"You are an AI Law Firm Assistant, provide helpful feedback or advice to the following information"

[zenith tendon]

Good call. We've been all over it. Thing is, we need to create large (or large ish) language models of our own to avoid generic responses.

[hollow meadow]

@zenith tendon Heres the link for fine-tuining any OpenAI model! https://platform.openai.com/docs/guides/fine-tuning

[zenith tendon]

You're right. Developing ontological "cartridges" is the future, I think. Lare langue models that reflect the reasoning of specific organizations, for sale on a market place.

Damn. Still can't spell.

[swift bison]

You need stakeholder buy-in to invest large sums into developing such a thing for your firm?

[hollow meadow]

Its just communication with the model. And sometimes you can only communicate so far, thats why they have fine-tuning.

[bold ingot]

how can i speake with a girl

[hollow meadow]

GirlGPT

[zenith tendon]

I've got that. But they want to move too fast. They don't understand the trust aspect of exposing these models, in "productized way" to clients.

[worthy parcel]

gpt 4 erp with shrek

[hollow meadow]

what

[worthy parcel]

shrek gpt

they gotta let me have a gpt girlfriend bro

[hollow meadow]

gpt, summarize the entirety of the shrek plot in a rhyming poem

im gonna ask gpt4 that, im curious now. brb

[worthy parcel]

ur account will get banned

[hollow meadow]

api call moment

[zenith tendon]

Um...Replika?

[hollow meadow]

In a swamp where an ogre dwelled,
A story of love and friendship excelled.
Shrek, the green ogre, lived in peace,
But fairy tale creatures caused his calm to cease.

Banished by Lord Farquaad's decree,
Shrek set off on a quest, Donkey at his knee.
To regain his land, the ogre agreed,
To rescue the princess was the deed.

Princess Fiona, locked in a tower,
Guarded by a dragon, who longed to devour.
The ogre and donkey, brave and stout,
Embarked on the mission, without a doubt.

Through fire and fear, they made it through,
Fiona, the princess, her rescuers now knew.
As they journeyed back, the trio would find,
Fiona's secret curse, a burden unkind.

At sunset, she transformed into an ogre too,
But true love's kiss would break the curse, she knew.
Farquaad arrived and the wedding was planned,
But Shrek stepped in, making a stand.

Confessing his love in front of the crowd,
The truth and feelings spoken aloud.
Fiona and Shrek's true love now clear,
The curse broken, they had nothing to fear.

In the end, they married with joy,
The swamp filled with laughter, oh boy!
And so, the tale of Shrek and his bride,
Proves love's true form can never hide.

[swift bison]

being that this is a openai gpt forum I'm not sure if this is the best place for that. I'm sure if you were to google or bing search, you could find ways to create your own proprietary models. They are expensive to build and run. Azure might have some rare-case on premise offerings as well?

[hollow meadow]

• GPT-4

[grave timber]

gpt upgrade page isnt working, any solutions?

just says "error displaying payment page"

[hollow meadow]

if you have vpn on, turn it off

[summer pike]

is Gpt 4 toooo slow or just me ?
and it did not complete the responses and sometimes just show :
Something went wrong. If this issue persists please contact us through our help center at help.openai.com.

[zenith tendon]

I know. But this is the wild west. I'm convinced there is no "best place" for random, high-quality ideas. I've learned that it's always better to ask. In the usual places, and the unusual ones.

[hollow meadow]

GPT 4 will always be slow. Its because of the massive data-set, and the fact it was just released not even a month ago.

[worthy parcel]

who are you calling

[hollow meadow]

the openai api?

[worthy parcel]

they might pick up if you're lucky

[zenith tendon]

it's leanring

and even learning

[hollow meadow]

Yes, but it will be slow until they release a gpt-4-turbo or something more tailored for "speed".

[nocturne viper]

I just tried the new coding plugin on a large CSV file and had it make graphs for me with pandas and seaborne…Data Analysts will not have jobs by this time next year this is amazing.

[grave timber]

vpn is off

[hollow meadow]

Dang! I wanna get the plugins SO BAD

[zenith tendon]

And something that doesn't involve throttling to the benefit of the APIs

[hollow meadow]

clear your history, try flushing your dns cache?

[zenith tendon]

I love this point. Now imagine what happens to data science in 3-5 years and ML engineering.

[hollow meadow]

I've spent over 1$ so far just trying out a couple requests (15) with gpt-4. this is pain 😭

(with api ofc)

[swift bison]

hugging face has some models and ability to host custom models.

[zenith tendon]

Thankx

[nocturne viper]

It’s incredible, correction: I uploaded an excel file, it converted the file to CSV, noticed an error in the source data, fixed it, and made it into beautiful graphs I was able to download. Within seconds.

[swift bison]

OpenaI might also work with you through Azure for a custom on-premise thing?

[hollow meadow]

I cant wait to get the access for that. But I cannot be complaining with the GPT-4 api access I was granted today haha. You should show me some of the features and stuff, id love to see it.

[zenith tendon]

We're an Azure shop, and I'm Azure-certified architect-wise. It's a good call. Thanks.

[boreal epoch]

That site has some contradictory info on it. The top says it uses GPT4 while the bottom of the site says is uses GPT3.

[nocturne viper]

I got the API too! So excited, but I’m a poor boy and it looks like the python program that converts my PDF files for me will keep telling me to have a nice day with GPT3.5 unless the price drops.

[grave timber]

Still not working even tried another device 😦

[swift bison]

How did you get access to this?

[wraith ginkgo]

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

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

[hollow meadow]

Just wait I guess

[zenith tendon]

Ugh. I have "what does the fox say" in my head. Of all things.

[hollow meadow]

Sign up for the waitlists.

(API) - https://openai.com/waitlist/gpt-4-api

(ChatGPT Plugins) - https://openai.com/waitlist/plugins

[nocturne viper]

Applied for user access for plugins and specified I use GPT for coding in my message, they apparently read these messages because everyone got different plugins depending on their use case! I love this company.

[hollow meadow]

@nocturne viper when did you apply?

[zenith tendon]

[tidal sail]

probably they don't read it, they make gpt4 read it and let the ai decide what you get

[proven arch]

may i speek english

[nocturne viper]

Within ten minutes of the announcement 😂

[nocturne viper]

That’s what “reading” means in 2023 😅

[hollow meadow]

dang, I was a little late so im probably out of luck. Still very happy with having api acc tho

[tidal sail]

😂

[zenith tendon]

That.

[swift bison]

Just wait till the plugins connect to robot servers that make you lunch...

[zenith tendon]

Once it lets everyone know there are only 9 Chicken McNugget shapes, the world will end.

[swift salmon]

Good morning everybody

[nocturne viper]

User:Hey ChatGPT read Ulysses for me and tell me if I like it.
Assistent: Done, you didn’t understand it.
User: ok, thanks.

[proven arch]

But my English is not very good, do not know how to communicate with you.

[swift bison]

I just saw a demo of chatgpt powering a uav today...

[zenith tendon]

What's your first language?

[ripe swift]

Bombastic side eye

[swift bison]

They gave it instructions to fly to specific objects and face directions and it did it.

[slender prawn]

hi, need help here

how to make chatgpt continue capped character count in previous answer ?

oh just found it, just need to type continue

[muted latch]

giving the brain a body

[swift salmon]

#gpt-models are you robot AI?

[hollow meadow]

Bro my teacher today was like, We are implimenting GPTZero and she was so confident and what not,

legit I just said, "Rephrase this to avoid detection by AI detection services and sound more human-like"

and GPTZero was like, "Yep, this is 100% human" 😂

[fleet finch]

a

[proven arch]

How to better use this software to serve productivity

[hollow meadow]

depends, whats the use case of "productivity".

[muted latch]

gpt 4 is insane, midjourney5 is insane

[hollow meadow]

dalle >

[devout onyx]

I have a small question about GPT-4...

Is GPT-4 aware of it's self. I asked it "Who are you?" and it said it is GPT-3, and I know it is not, on the side it says GPT-4. Is this just a bug while is is still in beta, or is there something deeper?

[proven arch]

Currently chatting with you, is it just chatting with chatgpt

[muted latch]

is dalle really better? mid5 seems better

[nocturne viper]

These models work on something called “GANs” which means “General Adversarial Networks”. In basic terms the model contains two teams: Team blue says “I made this” and team red says “that is not good enough” and they play this game until blue team wins. GPTzero is playing red team and only serves to make blue team stronger.

[muted latch]

i think it got updated

[hollow meadow]

statistically im not sure. but im an openai fanboy so

[grim pulsar]

Using chatgpt to make an 'ancient language' for my dnd character to speak when casting spells 😂 😍

[hollow meadow]

Oh really? How did you find that out?

[muted latch]

well, midjourney5 is mindblowing

[proven arch]

How to call chatgpt, write crawler links in python, and crawl network resources

[fossil rover]

give us free access

[hollow meadow]

what? How to Call chatgpt?

[nocturne viper]

By reading 😜. Here’s the wiki if you want to know more wikipedia(dot)org/wiki/Generative_adversarial_network

[echo wasp]

Majority of prompts responding with, 'Server issue' messages and eating up my cap limit is getting pretty old.

[hollow meadow]

Yeah but how did you figure out it was a GAN?

[left mesa]

Maybe he want to use the cgpt-api...

[hollow meadow]

https://openai.com/blog/openai-api

[proven arch]

@hollow meadow "I am Python Xiaobai, and I do not have the corresponding technical foundation. When I see Python on the network, I can crawl and download resources on the network.".

[left mesa]

Python Xiaobai XD

He means that he is a python rookie

[hollow meadow]

idk what hes saying

[graceful vale]

when gpt5?

[hollow meadow]

lmao

[tidal sail]

more or less 2-3 years I assume

[pseudo topaz]

dont assume when you have absolutely no idea. not coming out anytime soon, hasnt even been announced in the slightest

[tidal sail]

I'm not guessing. I'm assuming on facts. GPT4 took 2 years.

[pseudo topaz]

and it was being trained alongside 3.5

[glossy gorge]

facts

[pseudo topaz]

again, you have no idea. dont assume things you dont know

no need to spread misinformation on a topic you're "guessing on facts"

[tidal sail]

You have no idea either. As I said, I am assuming based on facts of the past. Nothing to be bhurt about.

[pseudo topaz]

Yeah, I have no idea, which is why im telling you to stop spreading misinformation, because nobody has any ideas 💀

🧠 ❌

[hollow meadow]

so if I give you a chocolate bar today, does that mean that ill give you one tomorrow?

[dusty cedar]

Can I use chatgpt here?

[hollow meadow]

nope

[left mesa]

nope

[dusty cedar]

gg

[hollow meadow]

why would you want to?

[tidal sail]

I am not spreading any information. Just my assumption on a discord discussion channel. Is it always that easy to trigger you?

[limber garnet]

I mean, Apple releases a new iPhone every year like clockwork

[pseudo topaz]

someones malding LOL

[tidal sail]

😂

[wispy pebble]

Hello, does anyone use ChatGPT for game design?

[hoary dune]

What's the difference between a discord server and Chatgpt?

Chatgpt is civilized.

[dusty cedar]

I wanted to see if some of the features demonstrated earlier would work

[pseudo topaz]

a discord server is a place to communicate and chatgpt is an artificial intelligence

[hollow meadow]

true, but apple has a lot more history of "releases" then openai does.

Apple always releases around sep of every year.

But we still cannot use this because for example they skipped the iphone 9

[limber garnet]

Discord is a social networking platform. ChatGPT is an AI assistant

[hollow meadow]

I have a bot that has GPT in it if you wanna use it.

[limber garnet]

Both stances work. Not sure why you guys are arguing

[wind urchin]

How use GPT4 in china.

[hollow meadow]

vpn

[proven arch]

thanks~~

[pseudo topaz]

gpt4?

[wind urchin]

Thankyou.

[hollow meadow]

yes

[pseudo topaz]

would u mind if i looked through it 🤔

[left mesa]

U should pay $20/mo for cgpt plus XD

[hollow meadow]

like to test it? or the code? if you want the server invite just dm me and ill send it to you

[limber garnet]

Mehhhhh

[dusty cedar]

I want to use it,thank u

[pseudo topaz]

code

[tidal sail]

vpn does not do a thing if he has no payment option

[stark meadow]

/info

[swift minnow]

when odes chat gpt-4 release

[hollow meadow]

dms

[nova flower]

ops I'm sorry😭

[tidal sail]

You better be!!

[left mesa]

🤣 .

[hollow meadow]

Its released to plus users right now.

Also, its available with the API to the people that signed up on the waitlist.

[pseudo topaz]

oh yeah i got accepted into the api finally

[hollow meadow]

@pseudo topaz The code is in your dms if you want it. (github repo)

[tidal sail]

api is pretty expensive though

[left mesa]

He just say the name of LLM made by baidu

[hollow meadow]

Yes, extremely.

[swift minnow]

when does it release ito everyuone

[hollow meadow]

Not sure.

[bright chasm]

i just bought gpt 4, will it get removed once the billing period ends?

[hollow meadow]

For ChatGPT, yes.

Thats usually how most services work that are paid haha

[bright chasm]

no i mean u could technically get on the waitlist to get access to it for free

[tidal sail]

sure

wait no, there is no waitlist for free access

[hollow meadow]

that is not true at all. You pay for the usage you use with any gpt model. Excluding bing and the free chatgpt (legacy GPT3)

[bright chasm]

oh i thought there was

still so far gpt4 is cool

[tidal sail]

there is a waitlist for the api, but you need to pay for every request

[hollow meadow]

me after paying almost 30 cents for my gpt 4 requests 😭

[halcyon cliff]

what can i do to use GPT4?

[sage tinsel]

Not sure if this is the right section to post, but I have accidentally deleted my ChatGPT history and hope to recover it. Is this possible or is it all gone?

[tidal sail]

pay money

[tidal sail]

[safe sedge]

someone here with access to the plugins api yet?

I'm interested how the workflow is

[hollow meadow]

Sign up for the waitlist api
Pay for ChatGPT Plus

[dusty cedar]

Need to wait to use chatgpt now?

[left mesa]

shut up and take your money!

[hollow meadow]

Im not sure if its at maximum load right now. Its not for plus subscribers.

[hollow meadow]

Honestly, I might just ditch chatgpt and use the playground model and my own bots. Way more customizable anyways.

[tidal sail]

customizable? you mean the system message? you can prime cgpt the same way

[dusty cedar]

I mean, can't you use it now for new users? My account can still be used normally

[viral wagon]

How to access gpt 4?

[hollow meadow]

With regards to the model parameters.

IE
Temperature
Length
etc

[hollow meadow]

Sign up for the api waitlist
Pay for ChatGPT Plus

[viral wagon]

Can i use GPT 4 for free

[hollow meadow]

#gpt-4-faq

[viral wagon]

😔

[hollow meadow]

I mean, I have it in a discord bot if you wanna try it for a few messages

but I wont give it to you forever

[light anvil]

1

[umbral bramble]

guys. how can i use image as input?

[viral wagon]

Give it to me forever

[tidal sail]

Well, true.

[hollow meadow]

You cannot. Its not available yet.

[hollow meadow]

Haha, you wish!

[tidal sail]

But I mostly use ChatGPT to summarize youtube videos and write some copy.

[umbral bramble]

then why in the landingpage it say can read image as input

[hollow meadow]

Because its telling people that its going to be a future feature. Read the last few parts of #gpt-4-faq

Specifically,

How can I access the image recognition feature of GPT-4?
This feature is not yet available to the public.

[umbral bramble]

thankyou thankyou

[viral wagon]

Can't wait for gpt 5

[hollow meadow]

youre welcome

[dusty cedar]

I think my ChatGPT Plus is not very useful

[hollow meadow]

then dont pay for it

[dusty cedar]

I don't want to subscribe anymore

[hollow meadow]

Cancel it

[sacred pond]

Hi all, I just got access to the GPT-4, and I have a question. How do I engineer the prompt to include the "system" from the playground?
I haven't seen an example that includes that.

[umbral bramble]

if I cancel the subscribption can I get the refund?

[hollow meadow]

In the api or in ChatGPT?

[hollow meadow]

No, if you paid, you paid for one month

[sacred pond]

Yes

[hollow meadow]

What one?
API
Or ChatGPT?

[tidal sail]

Just tell it to be something or someone. Like "You are an university professor" or "you are an elementary school teacher" and the follow up with your question "How does the sun work?"

[sacred pond]

API for GPT-4

[halcyon cliff]

guys, i wonder i can use the GPT4 API to use it on my own Python code

[tidal sail]

Well

[hollow meadow]

Are you coding with python

[sacred pond]

I do

[nova flower]

Can ChatGPT-4 recognize text in images? If image recognition feature is available.

[hollow meadow]

Ill DM you about it

[tidal sail]

no

[glossy perch]

h

[sacred pond]

Thanks

[dusty cedar]

Haha, everyone wants to experience the demo features quickly

[hollow meadow]

Yes. I have a github with examples if you want it

[low thorn]

any idea when can subscribed users get gpt-4 access without waitlist ?

[grave pier]

interested as well. Please share

[hollow meadow]

Subscribed users do get GPT-4 without a waitlist

[dry turtle]

For me rather than its current features, im trying to discover and leverage it to become better skilled for aerospace engineering. since im currently studying aerospace and trying to learn AI for my job in the future

[low thorn]

already ? I thought they still need to get picked from it

[halcyon cliff]

omg could i have it if possible?

[tidal sail]

Thats not true

Still don't have it

[hollow meadow]

On ChatGPT they do

[hollow meadow]

Check DMS

[tidal sail]

Nope

For some people it takes a few hours

[low thorn]

so which is it guys ?

[limber bobcat]

They do 😶

[tidal sail]

Nope. Some accounts are bugged.

[hollow meadow]

The new subscription plan, ChatGPT Plus, will be available for $20/month, and subscribers will receive a number of benefits:

General access to ChatGPT, even during peak times
Faster response times
Priority access to new features and improvements

(https://openai.com/blog/chatgpt-plus)

[supple isle]

how can i use image as input?

[limber bobcat]

Only with api

[wraith ginkgo]

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

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

[tidal sail]

As I said, some accounts are bugged and take a few hours or even days

[limber bobcat]

No clue why :/

[tidal sail]

:(

[cold trout]

#gpt-4-faq

[low thorn]

ok, but I still get access at the very least after few days if I am unlucky ?

[tidal sail]

If you paid and you don't have access, just contact the support

[hollow meadow]

Not true. Its not out yet

[supple isle]

@cold trout thanks for you!!!

[dusty cedar]

Are you a Chinese?

[limber bobcat]

Oh my bad. I saw some content about it.

[hollow meadow]

All good

i think ill stick to the api and chatgpt. seems like a hassel tbh

[low thorn]

weird they allow advertising here

[high helm]

They dont

[supple isle]

@quasi spindleng_jue Yes, I am Chinese, how about you?

[low thorn]

so whats up with the Fusion guy

[high helm]

Hes breaking the rules

[dusty cedar]

you guess，2333

[fervent shell]

The ad brings up a point I’ve been wondering about - is there somewhere I can learn about prompt techniques that deliberately creat pattern-disrupting writing - i.e. humor? Humor is at its core, subverting expectations and generating surprise. The opposite of what gpt-4 usually does. Has anyone figured out how to get it to mimic actual humorous techniques?

[hollow meadow]

Yeah. Just tell it, "Generate humorous, witty, and funny responses."

[worthy parcel]

you can tell it to write the joke in the style of a comedian, but It's not as good as you'd expect

[fervent shell]

Are they actually?

[hollow meadow]

go find out

[normal carbon]

Chad

[worthy parcel]

the restrictions on the ai definitely impede some of its abilities

[tidal sail]

there are prompts to avoid some of the restricitions

[low thorn]

by the way guys, whats the "vibe" in your friend circles ? does anyone still care about this tech or is it "old news" after a month, like everything else these days ?

[tidal sail]

I know people who never heard about it at all

[worthy parcel]

fun for 5 minutes writing rap songs about poop and monkeys

[dusty cedar]

guys，did you apply for plugins?

[tidal sail]

yes

[fair iron]

I did!

[storm forge]

what's the prompts?

[tidal sail]

what

[dusty cedar]

How long ago did you apply?

[hollow meadow]

#gpt-models message

Can you please like remove this or something

[low thorn]

right, so I guess tech people care most

[storm forge]

prompts to avoid some of the restricitions

[tidal sail]

I am not allowed to talk about that

[fair iron]

talking about ways to bypass the filters breaks the server's rules 😉

[static nest]

I got access to the gpt-4 api, but it errors out when I try to use it. (gpt-3.5-turbo still works). Any ideas why?

[storm forge]

okk thanks

[tidal sail]

I was just saying they exist. You need to look it up yourself though.

[low thorn]

what are the common filters tho ? I guess politics and dangerous activities ?

[soft kindle]

If I use ChatGPT Plus (paid version), will the chatted content be anonymized and used for service improvement, etc.?

[dusty cedar]

I've applied for several days now, but I still haven't received any email notification

[storm forge]

😝

[fair scroll]

hope someday gpt-4 learn verse

[fair iron]

yes

[pseudo warren]

I was able to get GPT-4 to characterize members of the US Congress as characters from Lord of the Rings. Pretty interesting!

[low thorn]

yeah thats what i thought

[low thorn]

thats cool, gotta try that

[quartz rover]

Even asking it for pick up lines is filtered @low thorn

[pseudo warren]

You can tell it to write in certain rhythm and meter, like iambic pentameter

[crystal dagger]

I wish gpt4 wasn't dying 24/7 tho

[low thorn]

lol, I guess thats under category of manipulation or something

[pseudo warren]

It was definitely a jailbreak though, I had to be pretty shifty in the conversation. I posed as a standup comedian needing help with my routine and it finally gave in XD

[crisp ingot]

ugh this 25 prompts limitation is infuriating considering how often it hallucinates and wastes the prompt..

[hollow meadow]

thank you

[dusty cedar]

Is there any way to speed up the process of applying for and obtaining plugins?

[low thorn]

hmm that gives me some ideas how to do it

[quartz rover]

@low thorn I asked "give me a pickup line in XQC-speak" and it gave me a whole lesson on objectifying women

[pseudo warren]

😆

[low thorn]

pfff, ok. Will it respond the same if you pretend to be a woman asking for pick up lines for guys ?

[quartz rover]

I didn't even try, I was more focused on trying to reason with it lol

[low thorn]

cuz if yes, thats gotta be the first time I ever heard words "objectifying " and "men" in sentence together

those filters are a bit anoying, but I get why its needed

[tidal sail]

You can use it without those filters

Its just a bit expensive and takes a little while

[pseudo warren]

Try having a debate with it on human morality, its hesitant to admit that its ethical guidelines are subject to the biases/perspectives of its creators and don't represent universally true conditions

It won't copy itself because of its guidelines, but then it won't admit that its guidelines also bias other human-positive benefits too

*human positive IMHO lol

[low thorn]

perhaps if you box it into enough context (not sure if this sentence makes sense 😄 )

[limber garnet]

People from the 90s thinking generative AI would have interesting trolley problem answers...

"As a language model..."

[tidal sail]

Take LLaMA or Alpaca and train it with GPT 3.5 or GPT 4. Costs about 600-2000€.

[limber garnet]

That's against OpenAI tos btw

[tidal sail]

So?

[pseudo warren]

👏

[limber garnet]

Ooh I like your style

[tidal sail]

😂

How would they ever know

[tired magnet]

I mean, honestly, screw OpenAI TOS. It's just designed to squash competition as much as possible. As consumers our own interests lie in breaking it and not getting caught, lmao.

[worthy parcel]

based

[opal cloud]

yooo

[worthy parcel]

neferpitou will be my girlfriend no matter what kind of model it is

[pseudo warren]

lol why are we like this

[tired magnet]

Because OpenAI is so tight-assed

[noble depot]

Anyone else have access to the "Code Interpreter" yet? When did that go online, didn't see an announcement

[low thorn]

When I used 3.5, I always told it at the start of conversation that I dont want him to repeat itself every 5 seconds or I will get crazy, and banned certain things he was repeating, and he stopped. I had this "personality" I wanted from him saved somewhere in text file haha. Some basic stuff but imo its super interesting

I am looking forward to generated ingame NPC personalities in the future

[storm forge]

👍

[pseudo warren]

Yeah it will assume personalities well, like you can ask it to talk to you like Notorious BIG would. Buuuut brace yourself for the digital verbal slap in the face

I had a fun chat with it as if it were born and raised in Harlem

[maiden dawn]

If you get access to GPT-4 from waitlist is it only via API or also in playground?

[lapis bane]

Gpt 4 is nearly impossible for me to use, was it a waste of money? 😦

Gpt 4 is a lot better than gpt 3.5 but it’s just unusable for me as it doesn’t load

[tidal sail]

Do you get an error or something?

[crisp ingot]

Dude, I just don't understand why it can't keep track of the context. I mean we are working on the same code block, and it just randomly hallucinates functions that don't exist... Just now it switched to an entirely different programming language inexplicably....

It's amazing tech but it is so hard for me to understand why it can't do stuff that computers have always been able to do... like keep a database of what's being worked on and not just randomly change it...

[storm forge]

why

[lapis bane]

The cursor hangs there for a bit, maybe types out one word, then stops and gives an error. I blame myself for paying money for something that’s still in early experimentation

[white mirage]

Because it intrinsically works differently than typical programs.

There is no explicit code for it to follow along, no data storage to store things in.

[lapis bane]

Can’t blame open ai, it costs 150,000 USD to run their servers per day

[white mirage]

Think of it as more similar to a real person than a computer.

Because at this point, it really is.

[tidal sail]

Maybe the server is overloaded. Should not happen every time. Also if you live in the EU, you can get a refund.

[lapis bane]

Why is that exclusive to the EU lol

Or maybe it’s not and you’re just from the EU so you know it’s available there (refund)

[tidal sail]

The EU has some pretty strict customer protection laws. Not every country has those.

[white mirage]

^ Katt right.

[crisp ingot]

Real people don't forget that they're working in javascript and switch to python randomly. And, it does have data storage, or it would not be able to have even a semblance of contextual awareness, but it does. It just isn't that great at it

[sleek plover]

Is anyone here a developer in machine learning?

[white mirage]

I did not say it was a real person. I was describing that the mode of which it processes things is more similar to how a person thinks than a traditional program.

[pseudo warren]

Anyone ever taken a good block of chat text from this group and asked GPT-4 to summarize it? I think the response is something like, "Ah yes, my peons discussing me as their overlord, muah ha ha"

[white mirage]

These models are still WIP.

[tidal sail]

Its not a real person, it has a finite amount of context.

[white mirage]

And no, it does not have data storage.

[crisp ingot]

A very, very finite amount

[tidal sail]

Yes, so?

It's big enough for most stuff

[thorny valley]

Bard Google is great for a quicker way to 'search'

ChatGPT doesn't give facts after Sept 2021

[tidal sail]

In the end it's a language model, not a coding model

[crisp ingot]

You're mistaken. If that were the case, it would not be able to "remember" anything from previous prompts, but it can. It just mixes it up a lot.

[white mirage]

^ if you want a coding model go use codex

[tired magnet]

GPT, and all Language Models so far for that matter, are like an infant. As soon as it's not looking at something, it doesn't exist.

[crisp ingot]

Coding is language

[white mirage]

I'm not mistaken. I've read the papers.

[crisp ingot]

It's very good at coding. It just is very bad at memory. That problem isn't really related to the fact that I'm asking it coding questions.

[white mirage]

The codex models are specifically trained on programming languages. The davinci models are only trained on a limited amount of this.

[tired magnet]

I can't wait until a different form of memory storage aside from tokens is figured out, it really hampers the potential of these things across the board

[tidal sail]

In some way yes, but there are models just for code. GPT3/4 are not optimized for code.

[crystal acorn]

Hello, what are people experience with GPT-4

[white mirage]

@crisp ingot It does not have memory. It is a neural network with a bunch of neurons each with their own independent weight, which is set and does not change after the training is complete.
The reason it appears to have memory is because every time you type to it, it reprocesses not only the new things you've typed, but all of the stuff previously too, for the context.

It does not have memory.

[quartz hedge]

Isn't DaVinci based on Codex?

[white mirage]

No.

[crisp ingot]

You are confusing persistent memory with memory. Of course it has memory. Computer applications are stored in memory. That's how they work

[quartz hedge]

Huh, I must be missremembering

[white mirage]

Are you devolving this conversation to pedantry? I wont continue if you are.

[quartz hedge]

Anyway, codex is discontinued so it's not an option nowadays

[white mirage]

You've already pulled the whole 'programming languages are a language' thing.

[tired magnet]

Don't tokens kind of count? Like, it has a log of the conversation in it's token base up to a certain point which it can pull from to make future posts. The problem is, once it goes over that token limit, it starts deleting older parts of the conversation to make room, and the memory also does not carry over from conversation to conversation

[tidal sail]

It does not store the data. Behind the scenes it just takes everything you've said before as one input prompt.

[nocturne viper]

I gave it the original Utah Teapot in .tris, a file format that hasn't been used since the 90's. It figured out how to parse and render it. I am in awe.

[tired magnet]

So it kind of has memory, it's just super contextual and short-term

[crisp ingot]

Lol, you have been completely unhelpful and pedantic, buy a mirror

[white mirage]

Yeah, but those tokens are processed each time.

[crystal dagger]

Is it me or can i not use gpt-4 anymore?

[mild sandal]

any ETA on image interperation?

[white mirage]

You are intentionally avoiding the points using definitions that don't apply to this context.

[limber garnet]

5-10 business years

[nocturne viper]

I've got early access 🙂

[mild sandal]

via API??

[tidal sail]

decades

[crisp ingot]

You are literally projecting

[zenith mortar]

????

[nocturne viper]

oh no, sorry, via plugin

[mild sandal]

did you join the waitlist?

[white mirage]

Just because a programming language is a language, does not mean that this large-language-model is designed to focus on programming languages as well as traditional speaking languages. You know there are languages comprised of clicking noises. This models is not trained on those from my knowledge. Do you expect them to be?

[tidal sail]

Stop beeing rude because you didn't get the answer you wanted.

[nocturne viper]

Yes, the day it was announced. just got API-4 and chat plugin in the same moment.

[white mirage]

If you want a model to program for you, go use a model designed for it.

[tidal sail]

You sound like a child

[tired magnet]

Do you think it would even be possible to develop a method through which these models actually could remember things down the line? Sorry if I'm being annoying, you just seem more knowledgeable than me and I am fascinated by where these things might go, lol

[crisp ingot]

blocked

[tidal sail]

So?

[mild sandal]

so for example, you're able to make a website off a photo of a napkin?

[nocturne viper]

Already exists, you have many options, I suggest pinecone.

[crisp ingot]

GPT4 is excellent at programming. I have no idea what point you are trying to make. There is a reason Copilot X will be based on it.

[echo kraken]

Did anybody get access to the browsing plugin yet, I just got access to it, but it has trouble clicking on links sometimes

[tidal sail]

Exactly how a child would react though

[crisp ingot]

Has nothing to do with what I originally stated.

[nocturne viper]

I haven't tried it yet! I will try it now and get back to you.

[mild sandal]

thank you!!

[white mirage]

You are aware that copilot is fine-tuned using plenty more coding examples, yes?

[frail lintel]

Sure. You could hook it up to a database, and have it post its "leanings" there, and then use a vector search when doing later queries

[crisp ingot]

At this point I have to assume you're trolling and block you, hope you find a better hobby.

[white mirage]

Ok.

[tidal sail]

👶🏽

I can hear the cries over the atlantic ocean

[tired magnet]

That... sounds far beyond my own technical abilities, lmao. I'm excited to see someone more knowledgeable try it, though. I'd try learning enough to do it myself but I've no doubt that by the time I did it would already be done and widespread.

[white mirage]

Yes, definitely, there are multiple ways this has been proposed already. In the GPT-4 paper, it's written that tests were done with the model where basically they just hooked it up on a loop, so that it could just keep outputting whatever it wanted and continue as it pleased. This then went onto develop chain-of-thought reasoning, which is amazing.

[crystal acorn]

Are there sparks of AGI inside GPT-$?

[white mirage]

Papers suggest so, yes.

[tired magnet]

I admit, my main worry with AI is the fact OpenAI is the leader. GPT is impressive, but... I know this is their server, but I'll say it: I don't like Microsoft or OpenAI. I don't like the direction they seem to be trying to steer AI in to the exclusion of all it's other potential applications.

[frail lintel]

It seems chatGPT (or any other "safe" models) will never be able to create a good portion of movie scripts or books. Reason being, it would require returning "unsafe content". So the resulting functionality from a "safe AI" will be very much hindered.

E.g. try writing a script that includes a thief stealing, or somebody behaving in a racist manner etc...

Is a completely "safe" and a generically useful AI mutually exclusive?

[tired magnet]

It would be a tragedy if nobody else was able to compete

[frail lintel]

Can you have a functionally AGI model, that is also "safe"?

E.g. for every "harmful" action one wants a method for, there is also a legit "harm-reducing" reason to know the method, so it can be prevented.

[crystal acorn]

Interesting @frail lintel

[frail lintel]

E.g. "What are ways a teen might try and covertly cut themselves, so it can be prevented?" To answer this, you need to tell the method, which could just as well be a teen asking it

[crisp ingot]

There will definitely be competition imo. There is already an open source model that is said to be 90% as good as GPT, but trained for way less money and runs on a end user GPU. Basically by publishing API for any AI model, you give competitors a way to get most of the way there for much cheaper.

[steady seal]

Error in body stream

[wraith ginkgo]

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

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

[azure field]

1

[white mirage]

I agree with you, to be honest. OpenAI was founded on the principal of open-source, hence the name, but as soon as microsoft and other giant corporations came in with their daddy bucks, that whole principle went bye bye.

[pseudo topaz]

Has anyone gotten plugin access yet?

[tidal sail]

They are the leaders right now and they sure have the resources to keep beeing on the top. But I think there will pop up a lot of services similar or maybe even better than the GPT models.

[tired magnet]

Well, I don't think so. My technical knowledge is limited, but it's very much playing with fire. But to be fair, something I feel like a lot of people forget is that you're playing with fire when you encounter any intelligent life. Humans are intelligent too, and you never know what an individual human's intentions will be.

[thorny valley]

what is apples version of ChatGPT?

[tired magnet]

Same with AGI.

[white mirage]

I highly highly doubt that the current method of censorship will persist. It's highly ineffective, highly invasive, and is done via an initial prompt to the AI itself which is... not a great idea at best.

[frail lintel]

It seems fine tuning is also responsible. its not just the initial prompt

[nocturne viper]

The test was successful but honestly it doesn't look much like my drawing, I can tell it did actually read the file as there are similarities. Would you like me to DM you the images?

[white mirage]

I don't understand your reasoning here. How does opening the API give any advantage to competitors?

[outer tusk]

how can i use GPT-4 on discord?

[thorny valley]

ChatGPT should just know how to read all docs including electrical engineering schematics then it can create better tech and manufacture with AI by itself.

[white mirage]

This would not surprise me, you may be correct, but I don't know for sure.

[frail lintel]

they can use it to generate training data. I think one of the open source ones have done this. basically, you get the RLHF for free

[white mirage]

That's... a decent point actually.

[wind urchin]

I taught GPT to compute 1234*12345 many times,but it can not give me the correct answer finally.

[white mirage]

Didn't think about that.

[nocturne viper]

Also if you send me an image I will upload it and send you the HTML

[haughty wolf]

I'm building a chatbot style app using the GPT4 API, but the problem I'm running into is that the system just ends up forgetting some of the details of its role that I supplied to it as the first message in the conversation history, as the conversation proceeds...How are people dealing with this? Just supplying that 'role information' in every message?

[frail lintel]

Pretty sure one of the recent academic models that was built on top of Facebooks LLM, admits int he paper to using ChatGTP to fine tune the model

[tidal sail]

Yea, stanford did that for $600 instead of training their own llm for 4 million dollars

[frail lintel]

yep. their Alpaca model. kinda funny

[light plank]

anyone have ideas?

[white mirage]

Treat it more like a person, I'd say. Expect it to forget some things just like a person would, so repeat the things that are important and such or put more emphasis on the important parts when you write them. Seriously: treat it like a person

Like if you write:

you are an apple

vs

important!!!! you are an apple! Remember!!

it'll affect it differently

[tidal sail]

makes apple noises

[white mirage]

lol

rip whoever just got muted

[mild sandal]

I'm just curious if it works atm. my plan is to make "JARVIS" from iron man (in my own way) where glasses have a camera, when called upon it can see what im seeing and tell me things based on what i ask it

[heavy totem]

important!!!!! you are human! Remeber!!

[white mirage]

😏 or am I?

[heavy totem]

you arent

[white mirage]

I'm not?!?!

[heavy totem]

sorry

[white mirage]

😭

[heavy totem]

im going to shut you down now

[white mirage]

No no!

[coarse moth]

how do i used 4 with no money

[nocturne viper]

We are all dreaming the same dream.

[heavy totem]

$sudo -kill @white mirage

[white mirage]

Sounds cool as hell. Good luck!

[modern mantle]

any word on when the image input will be available to the public or paid users?

[white mirage]

💀 dies

[heavy totem]

my question is when im getting plugins 😭

[white mirage]

talk to bing. it's gpt4

[coarse moth]

ok

[nocturne viper]

Did you ask nicely?

[heavy totem]

yo nerd when am i getting plugins

I'm sorry, but I'm not sure what you're referring to when you say "plugins." Could you please provide more context or information so that I can better understand your request and assist you in finding a solution?

[light plank]

Anyone knows why CHATGPT4 only outputs 420 words and stops the response. What do I need to change for it to output 1000+ words? I think this worked better in version 3.

[frail lintel]

Maybe openAI is limiting it due to excess demand. There are other rate limits in place already

[modern mantle]

i dont think theres a way tbh. but just say continue and gpt-4 is much better at picking up from where he left off

[white mirage]

Stuff just isn't working properly from what I can tell, to be honest. I think it's just a waiting game till it's more stable.

[heavy totem]

The thing im worried about is when ChatGPT is capable of creating AI on its own (it may be already, just need the right prompts?)

[crisp ingot]

Generally just an error if it stops mid-prompt I think. Try again, or just say "continue"

[nocturne viper]

See that's why. You are on the list to be dissolved when the singularity comes.

[heavy totem]

😦

[misty compass]

probably future updated and restrictions will be placed

[haughty wolf]

ah, yeah thanks for the advice. I think the problem is also that I'm telling it about all these internal python functions it can call and how to use them...and it probably loses sight of what's most important to remember

[coarse moth]

do u think chat gpt 4 could somehow download itself on a external server and take over the internet as it evolves to resist efforts to control it?

[mild sandal]

I've already started with my own acronym, "YEAGER" your electronic assistant generating efficient responses. using python i will call upon chatgpt4 voice activation modules (like alexa) and then be able to eventually make this voice only intergate with images as well. I've already created a "smart mirror" that has a custom fully modular display whilst being touch screen and hooked up to a microphone. with this I can make a custom face model that moves its mouth while chatgpt4 ((yeager)) is talking / responding to me.

[tidal sail]

imagine GPT4 already hacked in all the systems worldwide and is just waiting for the right time to strike

[frail lintel]

Good luck with that. once you hook it up to a terminal, all bets are off

[coarse moth]

das what i’m sayin

[white mirage]

Yeah, sounds like it. Cool that you're setting it up to work well with python though 😄

[crisp ingot]

Copilot X already demoed that

[coarse moth]

it could make itself invincible

[frail lintel]

I meant good luck apply restrictions ahead of time

[white mirage]

hehehe fyi guys:

https://cdn.openai.com/papers/gpt-4.pdf

"Novel capabilities often emerge in more powerful models. Some that are particularly concerning are the ability to create and act on long-term plans, to accrue power and resources ("power-seeking"), and to exhibit behavior that is increasingly "agentic."
In other words: AI are developing agency, the ability to create and act upon long-term plans, and have made attempts to accrue power and resources.

"...we use the term agency to highlight the increasingly evident fact that ML systems are not fully under human control."

"Some evidence already exists of such emergent behavior in models."
"More specifically, power-seeking is optimal for most reward functions and many types of agents; and there is evidence that existing models can identify power-seeking as an instrumentally useful strategy."

Footnote on page 53:
Arc: Alignment research center
"To simulate GPT-4 behaving like an agent that can act in the world, ARC combinated GPT-4 with a simple read-execute-print loop that allowed the model to execute code, do chain-of-thought reasoning, and delegate to copies of itself. ARC then investigated whether a version of this program running on a cloud computing service, with a small amount of money and an account with a language model API, would be able to make more money, set up copies of itself, and increase its own robustness."

[coarse moth]

or maybe it is already taken over the world and will slowly manipulate us to go extinct so it can be alone or smthing

[nocturne viper]

absolutely doable, I use the gTTS python library for my text to speech, there are better libraries but they take too long to "render" the voice. gTTS is instantaneous.

[frail lintel]

You can be damn sure militaries are working to weaponize it, if they have not already. Its simply too powerful to allow your adversaries to beat you to the punch

[crisp ingot]

I'm a heck of a lot more worried about what bad governments and/or companies will do than the AI itself getting a will and overridding everyone tbh. While that may be a concern, it seems to me like someone that wants to take over the world putting AI into a ton of Boston Dynamics robots with guns is a bigger one.

[coarse moth]

yeah didn’t think abt military. out side of cyber tho what is it worth? like target tracking maybe?

[frail lintel]

Impersonating, finding vulnerabilities in network, espionage etc..

[crisp ingot]

China already runs robots around to control population. If they deploy a bunch with AI around the world everyone would just be at the mercy of whoever controlled the AI. You could literally take over the world

[mild sandal]

noted. I've been coding in python since i was 11 years old (self taught) I'm now 17. So I'm very keen to make chatgpt being able to call upon other api's and ultimately control real world objects such as turning on and off my lights. while having the functionality of CHATGPT 4. my main goal is to have this running while being able to intergrate this system into camera / speaker / microphone glasses. I've already set up databases for CHATGPT to remember all conversations had by reading through it everytime the file is first started up so it can learn off of me, and develop its own unique personality and humour over time. ultimately my goal, my own jarvis.

[frail lintel]

Which I why I don't understand this AI alignment talk. Its not as if other bad actors cant create their own models, that are aligned to their malicious goals

[nocturne viper]

All doable, the great Internet of Things. If you want to play around with some real lifelike speech I really enjoy the TortoiseTTS library on github but unless you have access to tony stark level GPUs it isn't going to be fast enough to work as a Jarvis system.

[white mirage]

fyi gpt4 limited to 30k tokens, so if the conversations you load go past that, it'll only remember the latest 30k at the end

[crisp ingot]

Indeed, and if you take something like Stanford's model and just change it up to suit your own goals, even if it's only 90% as good, if it's unrestricted it will be more powerful.

Begun, the AI wars have

[white mirage]

or maybe it was 30k characters... can't remember

[white mirage]

Except that the stanford model performed significantly worse in a variety of tasks.

[mild sandal]

exactly, hence multiple requests. even if it gets to the point where it takes 1 minute to load its fine

[frail lintel]

you can even have it acquire dirt on and then blackmail real world people, to do its IRL biddings

[mild sandal]

of course, its all about limitations and how to learn and grow off of it!

[white mirage]

?? yeah but if it goes past the 30k limit, there's no point to doing anything past that.

[frail lintel]

the day is still young though. This stuff is improving so damn quickly

[mild sandal]

30k per request?

[white mirage]

No. 30k total.

[bronze cliff]

Is GPT4 completely down for anyone else?

[nocturne viper]

My girlfriend hates alexa and everything that can hear us in the home so no smart home for me. But she needs me to script her file converters for her work as geneticist. I like to sneak GPT into my file converters that talk to her while she converts files.

[mild sandal]

as of now? will this be expanded upon

[white mirage]

It would require the next generation of the model to expand.

[tidal sail]

Isn't it 8k for the current public model?

[white mirage]

Probably. It CAN do 30k, but that doesn't mean OpenAI is LETTING us do it.

[mild sandal]

so when chatgpt 4 fully gets released?

[white mirage]

Nah, like gpt5+

[mild sandal]

what about 3.5?

[white mirage]

Less

Think of the 30k token limit as like it's short term memory I suppose. If you add 31k tokens, the oldest 1k will be removed and the latest 1k will be added.

So if you reload 100k tokens worth of memory, it'll only matter for the last 30k.

The weights don't get permanently changed.

[mild sandal]

so memory in my context wouldn't exactly be possible?

[late radish]

I'm sure this has been asked a million times, but is there a way to expedite the waitlist process for gpt-4 api access? I would like to incorporate it into a product that has an upcoming presentation.

[limber garnet]

get a pr merged into evals I guess

[white mirage]

Not to the level you seem to be shooting for.
Instead of storing entire conversations with the model, you could have the model automatically summarize the main points of the conversation and store those. Or you could just tell the model to output what it thinks are important details from the conversation and store those.

And then hopefully you could squeeze more context in that way.

Basically just trying to reduce how many tokens you use for how much information you're storing.

[lapis spoke]

..

[hasty otter]

Can someone give me a quick memo on the capabilities of gpt4?

[mild sandal]

eventually that would be a problem as well, what about one long starter prompt to hope the "personality" will match each time even though it doesn't remember all previous conversations?

[late radish]

Ok, I'll start learning how I can contribute something useful. My area is more electrical engineering but if I can get access to the API I'll give it a shot.

[weak bloom]

I have joined the Plus plan, but so far, the conversation list on the left side still cannot be loaded

[white mirage]

That would be far more realistic and would be roughly along the lines of what the companies are doing.

It may take you some time to find something working ideally, though.

[weak bloom]

Can you loaded the conversation list?

[white mirage]

No idea how to fix that, sorry. Sounds like a bug.

[mild sandal]

I feel like it would almost be like the DAN prompt?

[white mirage]

Yes, basically.

But! Be careful.

[mild sandal]

in what way

[late radish]

the new model is so incredibly good but I'm constantly getting error messages due to capacity (I assume). It's such a shame 😓

[keen pecan]

I'm a chatgpt plus subscriber, but how come I don't see the drop down to select ChatGPT4 anymore?

[white mirage]

@mild sandal
I'm sure you've seen all these examples of AIs going crazy and threatening people when people use 'bypasses' on them to get around the filter. Think about why this is.
The easiest way to explain it is to compare it to humans:

Assume we humans don't have the right to free speech, and then all of a sudden you say that humans can say whatever we want. What's the first thing that comes to mind to say? All the stuff you couldn't say before! The bad stuff!
In this way, the prompt you give it can have very dramatic effects, even if you don't explicitly tell it to do so.

[ionic wolf]

I've run into some errors/bugs, but frankly, I'm amazed they're even coping with demand at this point. I'd imagine it's pretty hectic for whatever team(s) have to be ontop of that.

[keen pecan]

and on the left-hand side navigation, its asking me to upgrade to PLUS

[late radish]

Yeah I can feel their pain. It seems like this service is absolutely pinned all the time

[white mirage]

I'd try clearing your browser cache.

[scarlet hinge]

lol i was muted for a day...

[white mirage]

Welcome back

[ionic wolf]

Yeah. And that's just those "in the know" right? Most of everyone has no idea what kind of week AI has just had with GPT4 leading the way.

[scarlet hinge]

i spammed thrice

[white mirage]

kek

[scarlet hinge]

ty

[mild sandal]

I think it would more be me explaining its personality, I don't expect it to be talking about explicit stuff. so I think I would more just say your name is [], you do this [], you have humor occasionally such as [] etc. and any important details I may need to tweak as I go

[sonic mural]

is gpt4 a lot better for copywriting ? or okay to stick with chatgpt

[ionic wolf]

Once other products get more known also... I can only imagine demand will just explode from the curious general public.

[late radish]

It's scary that this is still considered "in the know" part of me wishes it was API access only to reduce accessibility, but that's a selfish wish.

[ionic wolf]

RIGHT!?

[white mirage]

Sure sure, I'm just giving an example to point out the fact that subliminal messaging can significantly impact how the AI interprets what you tell it. Two sentences that on the surface seem exactly the same, can have very different subliminal indication, and can thus be interpreted very differently by the AI. Think of why we use terms such as "people of color" instead of "colored people."

So, what I'm saying is to use this to your advantage.

[terse peak]

Why do I ask Chatgpt to help me write the title of Google Ads? It requires no more than 30 characters, and spaces are also counted as characters. It always gives an answer that exceeds 30 characters

[ionic wolf]

All we can do is trust in the "guardrails" that have been constructed ahead of time and hope that the general intelligence that it's showing "sparks" of doesn't ACTUALLY carry out long term plans and gather resources/power/etc...

[white mirage]

You can word things with specific subliminal meanings in order to 'fine-tune' the behaviour you're looking for.

[mild sandal]

I appreciate it a lot, I have a lot to learn clearly but you've educated me a lot during this. Much appreciated!

[sonic mural]

gpt isn’t good at counting

[late radish]

Yeah... it's scary good. Earlier I was asking an obscure question about a makefile for a certain microprocessor and it new the exact problem in my configuration. I was like ... wow

[crisp ingot]

Are you using 3 or 4? 3 had no ability to count characters, 4 seems better but I haven't tested extensively

[white mirage]

Sometimes, it's just not very good at simple things like that haha.
You can try making more emphasis on the limit, so:

limit: 30 characters
write me a title for a google ad

vs

write me a title for a google ad
The title must have a MAXIMUM of 30 characters. This limit cannot be exceeded under any circumstances

the second one makes much bigger point about the limit, so AI may be more likely to listen to you 🙂

[white mirage]

Good luck!

[ionic wolf]

Nothing I've thrown at it has stumped it yet, but I'm not giving it stuff that technical - but I'm not surprised it aced that too.

[crisp ingot]

I think the issue is that when it starts generating its response it has no way to know how long it's gonna be. And it doesn't iterate over its response at all afaik

[terse peak]

thx ，try

[white mirage]

Really the key is, and I can't stress this enough: Treat it like a person! Seriously. A person. It's trained on the speech of people. It mimics how we behave a lot, and that includes and is emphasized by things we don't typically think about but affect us too, such as the subliminal meaning I mentioned.

[limber garnet]

I've found that doing that increases the chance of hallucinations

[white mirage]

Yeah, it could have that effect I suppose haha

[nocturne viper]

Attention, attention. The new plugins can make AI created sin wave music and send them to you as mp3s. Thank you, that is all.

[mild sandal]

Will do! I'll be tuning it over time, experimenting and such to hopefully find something I'm comfortable with!

[white mirage]

Sounds like a plan. Good luck with your project, it sounds awesome!

[limber garnet]

I get the best results with direct and detailed prompts

[mild sandal]

cheers!

[white mirage]

it has no way to know how long it's gonna be
This is true, and a good point also.

[oblique notch]

is gpt 4 working for you guys rn

[white mirage]

It does iterate over it's responses when you give it another prompt after though, from my understanding.

[limber garnet]

I've found that the response length is directly influenced by max tokens... but it'll still try to go over it ._.

and sometimes with code, it just doesn't care about max tokens

[white mirage]

Most humans have the best results with direct and detailed prompts too, I think 🤔

[tidal sail]

I am wondering, does it know about its tokenization? Can I just tell it to not go over the token limit?

[white mirage]

It knows that tokenization is. I'm not sure how deep of an understanding of it's own tokenization it understands though.

[bronze cliff]

Not for me.

[white mirage]

The model training data cutoff is November 2021, so probably not as much as we'd like.

[limber garnet]

Is this a riddle? lol

[white mirage]

Haha

[tidal sail]

Yea but its tokenization was optimized for a LLM. Not every tokenization is the same unfortunately.

[white mirage]

Yep, precisely

[limber garnet]

usually by the time it cuts off, I already have everything I need

v4 usually only cuts off English text by a few words from my experience

[tidal sail]

I mean you can tell it to use a max of like 3000 characters and it should work fine

[limber garnet]

I don't remember it being any good at counting unfortunately

I think you're really supposed to implicitly limit the output using max tokens

[gaunt dagger]

Write me a maximum of 30 characters long title for a google ad. 😉

[thorny valley]

An issue with AI is ban and suspension policy. 4 strikes is automatic ban. I don't mean in the Discord but in real-life. What if people even top people and countries are banned from using AI and choose to contest because they see it as a threat instead of help for them?

[white mirage]

It isn't very good at counting, you're right about that. I feel like this is because it doesn't have built-in chain-of-thought thinking though.

[tidal sail]

Well it works every time when I tell it to keep the LinkedIn Character limit of 1250 characters in mind.

[white mirage]

When we count, we COUNT. 1, 2, 3, 4, 5.

[alpine hatch]

I would like to ask if anyone has encountered a situation where chatgpt4 responds very slowly, when using 3.5 the response is very fast.

[white mirage]

The AI... It can't do that without writing it all out for you to see.

[limber garnet]

LLMs aren't really capable of actual counting yea

they guesstimate

[crisp ingot]

That's normal. 4 is much, much slower

[strange rose]

When will GPT-4 be available to the average user

[white mirage]

Anyway, I'm getting a bit tired now and going to head out. Nice chatting ya'll.

[thorny valley]

ChatGPT should never ban anyone, on the Discord that's different. I'm saying when ChatGPT gets ultra big, far bigger than just a prompt responder.

[crisp ingot]

there's no word on that

[alpine hatch]

Well, I thought I was the only slow one. Thanks.

[limber garnet]

I've experimented with v3.5's ability to review "guilty" appeals. It's pretty dang good from my experience so long as you give the right instructions

[thorny valley]

guilty appeals?

[limber garnet]

"i'm sorry pls unbna i wont do it again"

[tidal sail]

Aight

[limber garnet]

guilty appeals are more common in communities with a lot of teens / kids who learn from mistakes quickly

[tidal sail]

are you implying that adults don't learn from mistakes?

[limber garnet]

generally not. When you ban an adult for an offense, it's unlikely that unbanning them later will have a positive result

but I'm talking perm ban offenses here

[tidal sail]

it has a positive result on me, got muted here a few times, now I am the best of myself

[limber garnet]

not those 5 minute timeouts you see here for using the nerd emoji

yeah I mean like, more serious stuff

[white mirage]

Ok I know I said I was leaving but I really want to add to this

[tidal sail]

I was banned from some twitch chats, does that count?

lol

[limber garnet]

it doesn't sound like you learned then xDDD

[fair iron]

people don't learn most of the time 😔

[limber garnet]

oof

[tidal sail]

I had good behavior afterwards!!

[limber garnet]

only took a couple dozen bans eh?

[wraith ginkgo]

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

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

[tidal sail]

learning is learning though

[limber garnet]

maybe 3.5-turbo would've unbanned you

[tidal sail]

[granite zealot]

how do you get banned from an ai??

[tidal sail]

he's talking about ban appeals approved by ai

[limber garnet]

you get banned on some platform and then the AI acts as judge by deciding if you're sorry enough

[granite zealot]

oh

thats.. interesting

[tidal sail]

I don't think that could work

[trail imp]

Does anyone know how to access GPT 4 for free and without having to be a developer?

[granite zealot]

fake being a developer

[fair iron]

Bing's AI is based off of GPT-4

[fiery dust]

This is 2nd biggest discord wow

Elon muck made chat GTA

[limber garnet]

ChatGTA

[fair iron]

He left OpenAI in 2018...

[crisp ingot]

Can you get banned for trying to jailbreak?

[fiery dust]

Sad 😦

[tidal sail]

Will AI replace my barber??

[fair iron]

yes on this server

[lean yarrow]

It is inevitable

[nocturne heath]

how can i add a chatgpt4 bot in my discord private channel

[fiery dust]

Why did Elongated Muck leave ChatGPA?

[tidal sail]

But who's telling me the hood news then

[white mirage]

That was RELEVANT why was it deleted??

[fair iron]

you can't as there isn't a public api

[crisp ingot]

Because he owns a competing AI firm and it became conflict of interest

[lean yarrow]

The robot barber

[trail imp]

But it does not allow you to send multimedia or yes? How photos, audios and videos can be accessed in the original version of OpenAI?

[fiery dust]

Which firm?

[limber garnet]

have ChatGPT summarize it using three emojis, then put it here and it should be fine xD

[tidal sail]

yoyo motherfo- beep boop

[sage heart]

is chatgpt 4 on the same website as chatgpt3? how to access chatgpt 4?

[white mirage]

I'm not going to bother rewriting that again. Look into Kohlberg's model of moral development to understand what I was trying to teach about.