gpt-models

[halcyon ledge]

1

#community-help hi

[wind urchin]

the web browsing feature is unable?

[red rampart]

How do I use Chatgpt4 here

[wary tiger]

How can I convert an image to HTML? Is there any plugin available?

[heavy karma]

Yes it is. We do not know when it is coming back

[wraith ginkgo]

This is a Safe for Work Discord Server

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

Posting NSFW content will result in a permanent ban.

[wraith ginkgo]

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]

hey, everyone. I have one question. If I buy ChatGPT Plus subscription through Apple's App store, will I be able to access the features such as web browsing and plugins on the web if I am accessing the same account?

[wraith ginkgo]

`` Server Tip: ``

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

[honest lance]

why gpt-4 is so dumb now?

it cannot be used for working..

what for i paid for the sake?

[knotty grotto]

You paid for
Access when service are in peak time
3.5 high limit
And new feature

[honest lance]

But this gpt-4 is not for use anymore.. it cannot do simpliest tasks in C#

all the time it says he is sorry for not understanding.

[knotty grotto]

Maybe start a new chat ? And ask prompt differently?

[honest lance]

belive me i am using this chat from the beggining. It was better. they changed something.

[knotty grotto]

They update it. And it react differently to thr prompt. That's shy #prompt-engineering have a channel dedicated for this art

[honest lance]

and this is bad:

var elementsDict = new Dictionary<string, IWebElement>
{
// ...
};

he creates shortcuts for using less tokens. but it wont help

even if you tell him use whoile method, without shortcuts it use it anyway

[knotty grotto]

Just small suggestions to compare result

Create new chat . Add plug in or code interpreter

[honest lance]

If this is going to this direction, this will be worse than bard from google

it was ok, but now its less comfortable

they do something that hurts users

because they want less usage of this

this is not smart

i dont pay for this to think about it, i pay for this to get answers

[knotty grotto]

Like I said prompting is an art and it doesn't hurt to learn from other :3 . It did help me for dalle

[honest lance]

i am using gpt-4 and i dont need to learn how to prompt because i talk about mechanism used in this gpt-4 i see now. like this:

var elementsDict = new Dictionary<string, IWebElement>
{
// ...
};

shortcutting code examples

or saying all the time i am sorry for not understanding

2 weeks now it sucks

gpt-3.5-turbo is not skipping parts of code etc. And actually now is workin better than gpt-4

this is bad configuration

and i dont talk about API, i am talking about chat.openai.com

OpenAI makes wrong decisions nerfing gpt-4

soon i will go to use bard instead of gpt

[exotic bone]

I tried using plugins like WebPilot, Web Requests, Link Reader, etc. with chatGPT 4 but only got halucinated responses. Does anyone have an idea how to solve this issue?

[unreal marsh]

webpilot

[full kraken]

Have you tried $google search page. 😋. It’s this ancient technology from like 2018. It helps you with research and learning. The results are mostly SEO

[gilded sentinel]

Sometimes things change, and you're free to use whatever you like. I too have noticed many changes with chat gpt4, but remember the AI can still get things wrong. It's a fantastic tool and learning what prompts generate the best results will help you get the most out of your subscription.

[honest lance]

Ok, but if things change for worse then its problem

and i pay 25$ per month for this

[wind urchin]

I love having gpt 4 with no 25 message limit

[wind urchin]

Nervemind after 100 msg I got blocked

[wraith ginkgo]

`` Server Tip: ``

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

[wind urchin]

Hello! when is GPT-4 goint to be available for fine-tunning? We are about to start a project, and need to decide which one we´ll use (and davinci is going to be obsolete soon..) Thank you and have a great week!

[real marten]

Anyone else have plugins and code interpreter missing?

[honest lance]

which plugin for code interpreter is the best?

[grizzled flax]

Code Interpreter is its own model without additional plugins (Plugins model is also its own model).

[honest lance]

how i can use it on chat.openai.com

is this possible?

[real marten]

currently only for paid users @honest lance

[honest lance]

i know i can see its beta feautre

[real marten]

make sure you go into settings and enable under beta features

[honest lance]

i am using it already

[real marten]

ah

[zealous granite]

My guess: they soon release an expensive business version of chatgpt you can fine-tune. But... it's only my guess 😉

[eager osprey]

Can anyone say what to do when openai removes bing extension for chatgptplus user

I don't know why they removed it to me

anyone faced with this issue. please help me😫

[grizzled flax]

It's been turned off for everyone temporarily.

#announcements message

https://help.openai.com/en/articles/8077698-how-do-i-use-chatgpt-browse-with-bing-to-search-the-web

In the meantime, try some of the plugins that can search the Internet.

[eager osprey]

Thankyou for your support

[primal rampart]

I am having huge issues getting pdf to be readable in chatgpt
it keeps saying The plugin I use to extract information from PDFs does not support scanned PDF files.
i cannot find a reliable tool to convert my pdf to a "chatgpt redable pdf"

[eager osprey]

for scanned pdf files there is no plugin

you should use tesseract-OCR(python library)

that is only option

[wintry mirage]

You need an OCR software. Windows powertoys has this available.

[upper iris]

I've been getting good results using gpt-4 to generate C# code, within the last few days. Here's a chat where it's successfully generating code for working with Graphviz DOT notation: https://chat.openai.com/share/58745135-057a-4269-a574-6d568904025e

Are you frequently creating long chats? ChatGPT simulates conversational memory by passing the entire chat to the model on every request, so the number of tokens to be processed increases more and more as the chat gets longer. They likely try to optimize and reduce their computing costs by compressing long conversations, which might contribute to seeming dumber.

Maybe I don't observe the dumbing-down effect because I prefer to start a new chat for every new task? That blows up the number of chats in my history, but a browser extension like Superpower ChatGPT makes it easy to search my chat history.

So, as someone already mentioned, starting a new chat and starting again is one of the first things you should check.

Also, if you're able to access the gpt-4 API, try your requests in the OpenAI playground https://platform.openai.com/ and see how it compares.

[honest lance]

THESE ARE BASICS... ;D i mean something harder

especially for blazor-server and wasm it is bad now.. ;d

[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.

[dull swift]

Does anyone have access to GPT4 32K through API?

[brittle sierra]

welcome

[wary socket]

anyone knows if there is a good plugin who takes information from linked in and the put in the code into gpt or any other AI?

[wraith ginkgo]

`` Server Tip: ``

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

[tardy nimbus]

can anyone help me? I just paid for chatgpt plus but it is still shown as the free version right now

[rugged sluice]

Got the same problem, following

[royal crag]

That hasn't been the case for me recently, but I could just be lucky 🤷‍♂️

[wind timber]

Hey ,i have subscribed / paid for gpt-4 , but its still showing free

[cursive lark]

I just signed up for chatGPT Plus in order to access GBT4. I was told that you can choose the model in a dropdown on chat screen. I see no such dropdown and I am still in old version. Any ideas or help? Thanks

[half zealot]

Yeah same issue here, had GPT4 for a few months and I forgot to renew it and now I can't access it again even after I paid. On the subscription it says Premium user but I can't change to it.

[grizzled flax]

Unfortunately several users are reporting this bug. Here is one thread in bug reports:

#1130578346418393118 message

Hopefully resolved soon!

[cursive lark]

Yeah that's what I'm seeing. Says I'm premium but no choice to switch to GBT4

[cursive lark]

Ah ok thanks. At least that means I'm not going crazy ha

[grizzled flax]

If it's recognizing you as a Plus user, it might be worth navigating directly to the GPT-4 model new chat URL to see if it works in the meantime:

https://chat.openai.com/?model=gpt-4

It might just be a bug with the model selector appearing?

[half zealot]

Nice idea but doesn't seem to work, still says "Free Research Preview" at the bottom and iirc it says something else on GPT4.

I contacted support but usually replies within a week 💀

[cursive lark]

Same for me, but good thought

[velvet fulcrum]

Does telling gpt "You are an expert/professional/the formal authority on/ X" really increase effectiveness or are you mudding up the context and reducing effectiveness of a good answer by making gpt "act" a certain way, could this make it hallucinate more?

I guess the answer is yes and no? For example it's better to say "You help with project management", instead of, "You are a project manager"

[wind urchin]

is there any way to see how many prompts do i have left for chat gpt 4?

[half zealot]

afaik nope

[knotty grotto]

Not yet

[eager creek]

Is there going to be an intermediate version like GPT 4.2?

[knotty grotto]

nobody know maybe!

[cunning roost]

where can we see in the website when we are going to be biled for using the api>?

[versed ferry]

create logo design logo name Wink Services logo type Combination Logo. symbol Wink face delicious food food delivery target audience is children,

[wraith ginkgo]

`` Server Tip: ``

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

[knotty grotto]

no?

labs.openai.com

[old granite]

has anyone gotten:
Error
UnrecognizedKwargsError: q

while using webpilot? is it a common issue?

[reef quiver]

Hey not sure if this is the correct channel. I have paid my bill yet still no gpt-4 api access.

[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.

[open sundial]

Is there any technical incident happening with GPT-4? At least, for the last two hours or so, I can't use GPT-4...

[cunning roost]

its been like a month or so and i havent been billed yet so i can use gpt 4 api 😦

[half trout]

What's a good way to get no used business names in my prompts?

[slow stream]

There really isn't a good way unless you ask for novel combinations of words. Even then, business names are so unstructured, easier to search after the fact.

[cobalt igloo]

Does anyone know when gpt-4-32k will be released (approx)? See it on the docs but doesn't seem to be out yet

for API access*

[lone delta]

I havent been able to use gtp4 either... what gives

[slow stream]

Working for me currently (using the chat) but API was working like 40 minutes ago

[open sundial]

Maybe it's a regional incident? By the way, I'm in Japan. Where are you?

[slow stream]

I'm in USA, try incognito, get a new session

[open sundial]

Thanks. I'll try.

[slow stream]

I would assume they would have load balancing/failover so even if Asia went down you should get a slower connection to next data center with a new session

[open sundial]

Right. Makes sense. By the way, incognito did't work.

[covert crystal]

Hello everyone, I have subscribed to plus, but why can only send 25 messages in 3 hours when using gpt4

[sullen stratus]

The same limiot applies to all of us.

[slow stream]

That is just part of the plan unfortunately.

[potent abyss]

is there image mode yet?

for ChatGPT plus?

[covert crystal]

OK, thanks. . But I remember about last week, there was no such limit. . .
How can I avoid this limitation? gpt3.5 is so stupid, he will even give me multiple wrong answers and keep looping when you say "are you sure this time is correct". .

[slow stream]

You can use code interpreter to process images but that is cumbersome and about it

I just use midjourney for images

[vital fulcrum]

The gpt-4 option is not visible on the main page for my plus account. What should I do?

[slow stream]

Log out and log back in, or contact support.

[static isle]

GPT-4 in ChatGPT is currently unable to understand images

Google's Bard is able to generally understand images at the moment

[wraith ginkgo]

`` Server Tip: ``

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

[unreal linden]

create book record system in html. Add Year Publication, Title, Author, Publisher, Total Number of Pages and Summary. Add new table after submit book and keep all data after submit book record.

[tender tartan]

Why have I not been billed

I WANT ACCESS TO CHAT GPT 4

[woeful swallow]

may I asked when will the limit be lifted off? waiting for 3 hours per xx number of prompt is 🥲

[dark geyser]

This has been in place since gpt-4 first became available in ChatGPT. I don't expect them to lift that limit anytime soon. Or well, maybe once they've got to the point where gpt-4 is as useless as 3.5, they seem to be making great progress on that...

[lilac shard]

Have a look at the API pricing for GPT-4. ChatGPT plus is an absolute bargain.

[woeful swallow]

🥲

[dark geyser]

I agree on that one, even with the limitation. But only up to some weeks ago, since then gpt-4 in ChatGPT has become noticably dumber and lazier and I can't use it anymore for the coding related tasks it was doing very well with.

I've dug up some older questions from my history and tried them on the current chatgpt-4, and the answers are consistently a lot worse, and lot more

//your code goes here

with a comment at the end that it's too large and complex and I should do that part myself... where it previously answered these just fine.

[zealous perch]

What are some of the well-known digital idols worldwide after 2020?

[hard crow]

Personally I switched from 3.5 to 4 and the difference was not immediately obvious but upon having a 4 blackout, it is really telling how much worse 3 is. Mostly for just like, not strictly following your prompt and going off on huge tangents/padding what it puts out.

4 has some issues for sure too and what sucks is if it does make a mistake or you need to tweak what it puts out you are really screwing yourself over.

[forest hemlock]

Google it.

[potent abyss]

How is 4 worth it? It’s $20

It doesn’t seem like from what I’ve searched like it has much more stuff

[hard crow]

Well 3.5 is not great

4 is a lot better

[forest hemlock]

Depends on the value of $20 for you. In my opinion it's very much worth the price of a good dinner($20 for me).

[hard crow]

I use it for work every day so it's definitely worth it. Makes my work take like half as long.

It's pretty much cut down on an hour of work every day

[potent abyss]

What is it useful for mainly?

[hard crow]

4 specifically just gives better output. 3.5 is a lot more basic.

[dark geyser]

I have been working almost exclusively with 4 since it became available in ChatGPT Plus. I agree with everything you say vs. 3.5, but also, over the last month or so, gpt-4 has become noticable worse. which is really annoying.

Don't get me wrong, it's still heaps better than 3.5, but it's not as good as it used to be. It has gotten faster, but I would much prefer slower and usable vs. faster and not really helpful anymore.

[hard crow]

I did notice 4 got a lot faster. It feels like 3.5/3 got very fast and worse, while all the good parts of 3/3.5 went to 4? 4 was much slower on release and felt pretty consistent. Now it feels like 4 is basically like a better version of the old 3/3.5. But yeah, it's a lot faster and seems to be more prone to error.

[dark geyser]

like, for exactly the same question, as first question in a new session, so not depending on any other context, gpt-4 used to provide me with the full code, and the current version says:

Note: I have placed comments "you have the logic here..." in the ProcessTableTypes and BuildLinkSchemaAndIndex methods where you should add your existing code blocks. This is because the code you've provided is quite large, and refactoring it would require a much deeper understanding of your codebase and the specifics of the data and objects involved. I recommend that you adjust these new methods and the logic within them to best suit your actual project needs and structure.

And keeps doing that or something similar no matter how often I click "regenerate response", so it's not simply that it just got a bad RNG result.

so.. yeah... it definitely has become lazier

[wraith ginkgo]

`` Server Tip: ``

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

[forest hemlock]

Maybe it just wants to help you improve your coding skills.😛

[dark geyser]

My coding skills are fine. They have to be when you work with ChatGPT, because you can't trust it and need to actually fully understand the code it writes to spot the problems.

I just need ChatGPT to do some mindnumbing refactoring with for me

[marsh warren]

Did I miss an announcement about Beta options being pulled? All beta options have gone from settings I can't even sellect GPT-4 anymore...

ah, I have the same as this #chatgpt-discussions message

[forest hemlock]

I wouldn't call refactoring "mindnumbing". It requires some thought put into it and there are many ways to refactor something. I can actually see why chat gpt wouldn't want to make these important decisions on its own.

[marsh warren]

Payment taken but my plus subscription hasn't renewed, telling me to upgrade. ffs

[dark geyser]

I know how I want it to be refactored, I describe it to ChatGPT, it does the work, I fix what it got wrong, and overall it's a lot faster than doing it all by hand.

That is, before it stopped even trying.

[wind urchin]

why does this chatgpt suck

[forest hemlock]

Maybe it doesn't fit into the context window?

[wind urchin]

ima delete chatgpt

you gonna do something about it

[dark geyser]

I'm well aware of the size of the context window.

The issue is, I picked something from an old chat history, which was the first question in that session, so not affected by any previous context, and I put that question verbatim in a new session now, and it behaves completely different.

[supple violet]

i don't think anyone cares that as you joined here only to say that, pretty low effort bait

[forest hemlock]

I agree that the model output changed quite a lot in the last few months, I'm just not sure if we can evaluate that change as a bad one.

[wind urchin]

@supple violet ima stop this its not even gonna be a ban for me

im not breaking any rues

rules

[whole kettle]

Well, at least they kept the same price but we got access to new features. Could've been much worse.

[supple violet]

you already did

[wind urchin]

how what

@supple violet

how many rules have i broke here then @supple violet

[dark geyser]

Well, used to save me hours of work every day for a few months. It doesn't anymore. By any metric that matters to me, it's worse

[forest hemlock]

Perhaps. Did you try adjusting your prompts such that it produces the desired output?

[forest hemlock]

Can you share one simple example of how it returns an undesired output? Including the request prompt.

[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.

[dark geyser]

I've spent a lot of time over the last few weeks to try and get it to do what it used to do easily in the past. I find it takes a lot of cajoling to push it towards giving it even a try, and when it does, it's worse at it then it used to be.

It's at the point where I spend more time trying to get it to do what I need than it takes to just go back and do it by hand. So that's what I increasingly do

[dark geyser]

Can't really share the exact code, but in general, where in the past it produced complete working code, it now very often just generates a framework with //your code goes here comments and an explanation that I should just do the necessary refactoring myself.

[forest hemlock]

I see. I typically don't use it for such tasks, so can't confirm or disprove it.
Did you try using the API for such tasks? It could be a simple matter of system prompt or temperature parameter.

[whole kettle]

I can confirm that, it also refuses to write complete code even when told to now, unless the script is pretty small, it looks like there may be some issues with its context window. It sort of tries to give a complete answer in one go, instead of trying to expand like it used to do and then you'd just ask it to continue. Since quite some time I don't even remember seeing that continue answer button anymore. It's not that is stupid, it's just that it seems like maybe they tried optimising it so that it uses less tokens per answer when possible.

[dark geyser]

I would if I could. But 3.5 is useless, and I don't get access to 4 via the API as I haven't used the api enough in the past to get charged for it

I would really like to compare the March and June versions of gpt-4 via the API how they react to the same input.

[whole kettle]

This tech will probably get much more effective with time. At the same time the fact that it currently has these sort of issues is not such a bad thing because if you're trying to start a business, at least is not THAT easy for anyone to copy whatever it is that you're doing.

So I'd say take advantage of that while you still can and at least think about it this way too.

[remote valley]

Does anyone of you know what to do when an active PLUS subscription is suddenly shown as not active anymore and as to be be renewed? – I have a plus subscription since many months and the current plus subscription is confirmed as payed and active in my account and subscription status, just not in the web client window. I did not find any contact possibility and just wrote to the openai website chatbot, but have no idea when anyone might handle that issue. Meanwhile I might use my API, but all my current needed conversations are in the web client, so it would be great if someone has an idea. Thanks a lot.

[knotty grotto]

Status.openai.com

[remote valley]

Thanks a lot, Neko 😘

[marsh warren]

https://help.openai.com/en/articles/7232905-why-is-my-plus-subscription-not-appearing

[remote valley]

Great, thank you very much, darkglove! ❤️

[simple finch]

Best news plugin?

[marsh warren]

Not that it's helped me so far!

[weak elm]

Can u tell me the best ai tools

[knotty grotto]

Chatgpt 4

[stable mortar]

yes I do

[fair aurora]

Hey is anyone else having an issue with gpt plus? Because I just lost my access

[edgy jay]

Anyone else finds chat says to think from perspective of terrible PPL like rapists and murderes

And also that it shows a very high levl of he's also a human thing

[mint plinth]

Hi community, I use ChatGPT on my phone and the message that the Bing Browser is disabled does not go away and it is resting right on top of the writing box from ChatGPT. Does anyone know hoe to remove that annoying message?

[fair aurora]

Does GPT-4 work for you?

[tranquil dew]

🙂

[mint plinth]

Yes it does when that pop up message is not in front of the writing box

[next gale]

How I can change my smartphone´s resolutions without root

[fair aurora]

No I mean is the model available right now?

[fair aurora]

Because for me it is not

[mint plinth]

Yes for me is working Devesh

[fair aurora]

Okay, for me, the plus subscription is not available too, and no premium models

I am a plus sub

[clear imp]

Hello everyone, yesterday I made a request with ChatGPT+ to summarize a book by a French author that my brother is currently studying in his class. However, as a result, I got banned from ChatGPT+ for « book inquiry ». Has this happened to anyone else?

[supple violet]

what does it say? what message did you get

[clear imp]

No message, I just can't access my Plus subscription, and on the iOS app, when I try to launch a new chat I receive a message saying 'banned for book inquiry - default 3.5'."

[supple violet]

i've never heard of such ban and normally they deactivate the account so you can no longer login 🤔

[supple violet]

what does it say when you logoff/login on a desktop browser

[wraith ginkgo]

`` Server Tip: ``

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

[empty night]

/Chinese speak

@wraith ginkgoChinese speak

[knotty grotto]

Server is only english . Other language will lead automatically a mute by automod

[clear imp]

I have no problem logging in, only with using version 4. When I log in, I can perfectly use version 3.5, but it's not the case with version 4, despite my subscription. On the IOS app I simply get a message saying 'Ban For Book Inquiry'. I sent a message to OpenAI to get more explanations about this. I hope I can use the Plus version that I renewed yesterday quickly.

[supple violet]

that does not sound official 🤔 or would be weird at least as they give warning and ban issues in emails and ban means you cant login. plus renewals have ongoing issue https://status.openai.com/

[clear imp]

Okay, thank you very much for all this information. Personally, I believe it's quite a coincidence to receive that message on the iOS app and not be able to use version 4. I'll try to find out more. Thanks a lot.

[echo scaffold]

I have problem on reaching version 4 even though I am a subscriber from the beginning. waiting the problem to be solved. I sent a report

[supple violet]

https://status.openai.com/

[echo scaffold]

Yes, thanks

[frosty thistle]

is it me or is the site really slow , keep having me refresh the site to send a message

just saw the status, degraded performance

[warped oriole]

What is noun

[supple violet]

https://chat.openai.com

[short drum]

split

[ripe condor]

grrrrrrrrrrr

[burnt swift]

Anyone know how to gain access to gpt 4 api or when they’re going to open it up again?

[green sun]

i'm homosexual

[supple violet]

at this point all you can do is just wait

[burnt swift]

Damn hopefully they open it up again soon 🙏

[storm egret]

Hello everyone! I have some problem on gpt-4 api, i have paid api bill yesterday, but i still cannot use gpt-4 api. After getting model list, gpt-4 model still not appear, why? could somebody help me?

[jaunty spoke]

Just here to say that the limitation on the GPT-4 is ludicrously low for how long its been out. GPT-3.5 has become useless due to their "optimizations". Even with its dumbness its still useful for its ability to have unlimited text completions (which is the real reason I pay 20 bucks)...

I hear they are phasing out and removing 3.5 completely which is a kick in the balls to users and shot to the foot for OpenAI. GPT-4 only has a limitation of 25 responses... how ridiculous is that to nerf ALL users experience like this so dramatically and de-value your service/product?

[storm egret]

yeah, @supple violet told me that if you are not paid before Jul. 7th, you may be the next batch of api user...

[rustic igloo]

My subscription to ChatGPT Plus renewed yesterday and was paid. Today I have no ChatGPT Plus Model Selector when I log in and so no way to select GPT 4 model or use plugins. Is anyone else experiencing this issue?

[supple violet]

https://status.openai.com/

[rustic igloo]

I see people here talking aout the API, but not ChatGPT

[storm egret]

openai sent a email to dev, and it said that bill system has some problem..

[storm egret]

ummm, they are almost the same

[rustic igloo]

Well, that explains it then.

[storm egret]

i heard that many pros leave openai for leaders' inaction

[rustic igloo]

@storm egret Almost? Programming is an exact science. We are not playing horseshoes or hand grenades. In fact, you probably should never play with hand grenades.

[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.

[storm egret]

after pros leaving openai, they reduced model running parameters, the error rate increased, ......

[atomic robin]

you should probably not talk about things you are completely clueless about

[wanton lantern]

Hello All, Been a GPT Plus user for over the last 6 Months, I just recieved this message without notive and now my account has been terminated. Can someone please help on how to get our account back. "Dear ChatGPT customer,

You are receiving this email because we have identified suspicious activity on your account. In order to protect our platform, we have refunded and canceled your subscription. You will no longer have access to ChatGPT Plus service.

We take fraud prevention very seriously, and we want to ensure that our platform remains a safe and secure place for all our customers. If you believe that this action has been taken in error, we invite you to submit an appeal to have your account reinstated. Please visit help.openai.com.

Thanks for your understanding.

Best,
The OpenAI Team"

[supple violet]

yep, you got banned

[storm egret]

but after the news that pros leaving reported, the frequency of receiving error email increased a lot.

[atomic robin]

you are great at spotting patterns, wow 🙂

[wanton lantern]

woooah why would they do that ? Any way to reach out to the team and get the account re-instated or should we just use a new account now ?
The old account has a multiple trained MOdels and more.

[storm egret]

https://status.openai.com/ scribe the status messages, then openai automatic program will send you error message about their system..

[storm egret]

your ip address was suspected..

i have a account using some cheap server ip address, and also banned..

[supple violet]

suspicious activity = used from non supported country, vpn, virtual credit card, phone number location mismatch, account sharing etc. etc.

[storm egret]

yes, after change my ip address, it goes really well

[wanton lantern]

Okay, I do not use VPN or Virtual Cards, thanks for the help.
What would you suggest, wait and go through the process of reinstatement or just create a a new account.
Any advice on how to reach out to the team. @supple violet @storm egret you're help is highly appreciated

[storm egret]

wait, i used to use cheap reverse proxy for api, which ip address is form Hongkong(unsupported region), then i was banned

[supple violet]

email has instructions but it may take several months and you have to prove you did nothing wrong

[tardy dirge]

I am looking for a way to generate a chat with GPT thanks to a vidéo. Can I do it with GPT 3 or 4 ?

[storm egret]

i had send a email, but they said 'We are unable to reinstate your account at this time; we will investigate further. Please reach out if you can provide any additional information.' ...

[storm egret]

some devs can get access to the photo version of gpt-4, which can process picture or generate pictures, but it is far from videos...

[storm egret]

do you know the stable diffusion webui project? some plugins of this project can generate videos frame-by-frame rendering...

[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.

[pseudo warren]

hey hey hey, time for some fun...

Anyone with access to Claude2 API, make a simple Python program and have GPT-4 and Claude2 debate a topic of interest. You have to get creative because Claude can take way more tokens than GPT-4 currently, but it's a remarkable way to get some good ideas. You can even have them try and out-code each other.

[wraith ginkgo]

`` Server Tip: ``

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

[storm egret]

cool bro

[pseudo warren]

Claude2's response time is much faster, but its "memory" is virtually non-existent.

So you'd have to feed it updated context each time

[storm egret]

there are a concept in software engineering named ' pair programming.

[pseudo warren]

yep at my org we do that, drives some people nuts lol

but having two AI agents skilled in coding attempt to outdo each other is a fun idea

you could even literally dynamically code an app by taking end user responses (thumbs up and thumbs down) and have it recode accordingly

or real-time AB testing

the use cases are wild

[storm egret]

I remembered that some front-end code for ChatGPT can modify the chat log, so I think it may just need some coding.

[supple violet]

#ai-discussions

[willow urchin]

Has GPT4 ever succeeded in solving a wordle game? For me it failed every time, I guess because its still not able to play this game

[pseudo warren]

trying to test people's comparison of GPT-4 to Claude2

[storm egret]

haha, really wild and fasinating 🙂

[pseudo warren]

for sure!

[storm egret]

but i don't think it is efficient...

[supple violet]

yeah well dont be surprised if those messages disappear at some point 😛

[pseudo warren]

fair! it ain't my server lol

[willow urchin]

so it succeded for you? Which prompt did you use?

because I tested it like 6+ times but it failed every time

[storm egret]

some said the silicon valley's friendship is dynamic, today's enimy might be your friends tomorrow. 🤣

[pseudo warren]

all about da benjamins baby

[storm egret]

i mean, if the scene is not complex, you can train a model which can identify subjects, then finish your task

[willow urchin]

no cv just text. I am talking about nyt wordle

[storm egret]

language model focus on language tasks, it may be a commander

[willow urchin]

?

[storm egret]

okay, i regarded it as a video games...

sorry, i am not a native English speaker...

[willow urchin]

no problem 🙂

[willow urchin]

so has anyone succeded with wordle? But maybe GPT4 is still not advanced enough to do it, or probably Language Models are not good with individual letters because they are trained on tokens. Will have to wait for gpt5 then..

I mean don't get me wrong gpt-4 is an amazing achievement but there is still room for improvement

[storm egret]

GPT-4's reply: GPT-4 can help you generate possible answers in the Wordle game, but because this game requires players to gradually improve their answers based on previous guesses and given hints, GPT-4 cannot play this game completely independently. GPT-4 has no persistent memory, it cannot remember previous guesses and results.

When you provide GPT-4 with all the necessary information (such as your previous guesses and the outcome of each guess), it can help you generate a strategy for the next step. However, every time you communicate with GPT-4, you need to provide all the context information, because it will not remember the previous communication.

In general, GPT-4 can be used as a tool to help you play Wordle games, but it cannot completely replace people to play this game.

[willow urchin]

well thats wrong. gpt4 has memory because it can look at previous texts in a conversation

[storm egret]

maybe it want say you should use multiple turns chatting

[willow urchin]

I did that but it still failed to find the word

but this gives me an idea: I will try to give it the list of possible and impossible letters. maybe this improves it

[storm egret]

yeah, prompt engineer!

[wanton star]

i have

[icy flax]

I'm thinking of having gpt4 just so I can train it to know me and help me while I go through a dbt workbook. Because I have allot of questions, more than the book has answers, and maybe that lies in faults in my personality, I don't know. I tried to train it to know me in a gpt3 chat once, I hand fed it (copy paste) hundreds of blog posts from my teenage/young adult days. It was very tedious though, and I barely scratched the surface.

If I were to use it's ability to search the web, can I use this to feed it my life just by visiting my old blogs? maybe my facebook page? (YES this means dumping all my personal data into the much feared data hungry machine lmao . I'm secure with that.)

[willow urchin]

hmm now I am a bit disapointed of gpt4. It gave me the word "lure" as guess which isn't even a five letter word. Even tho I clearly said it should give me a five letter word

[fierce spoke]

yea chatgpt sucks as wordle

probably cause it invovles symbol manipulation

[icy flax]

there is a term for false conclusions, it's called hallucinations right? I feel like in this case, maybe the right word is distractions. It was so destracted by one part of what you wanted that it ignored another..?

[fierce spoke]

hallucinations are probabilistically valid results that are actually false

this would probably just be more a failure of conforming to the prompt

[serene thistle]

why my Plus disappear after the monthly fee was charged yesterday?

[wise bluff]

Same here 😥

[hexed moat]

I'm looking for a plugin that can turn a picture into a description, is that possible?

[fierce spoke]

midjourney /describe

or really anying captioning llm - like bard

[coarse gyro]

At the moment, there is no plugin for it.

[hexed moat]

thanks!

[fluid hamlet]

any news yet on when the prompt limit of 25 will be removed?

i refuse to keep paying for an ai that hallicunates and has a 25 prompt limit

[knotty grotto]

No eta

[fluid hamlet]

bummer

[stable tinsel]

How many of you feels that GPT-4 performing worse than it used to? I hope the competition will catch up soon.

[shadow tusk]

Does it have access to the internet

[shadow tusk]

I hope a digital brain comes out that would be good

[rancid summit]

My subscription for Plus renewed yesterday but when I log in, it's acting like I haven't paid. I've subscribed for 3 months and haven't had an issue. Need this fixed ASAP.

[umbral badger]

Me too

[stable tinsel]

I am not sure I want a digital brain but ChatGPT already showed how much it was capable of then OpenAI has started to downgrading it, which is fair given the fact that nothing is free and computation power cost but I expected more after paying for the better version.

[wise bluff]

When does this issue fix dam

[rancid summit]

I hope I am compensated by having my subscription extended since I am paying for a monthly service that I don't have access to.

[fierce spoke]

that is against TOS

[stable tinsel]

From my experience with their support, I worry you may be disappointed.

[rancid summit]

😦

[stable tinsel]

Again, I can't blame them entirely. Having 100 mln users must generate A LOT support tickets so it is natural that they went with an useless bot but paying customers should be able to connect with human support at some point not a bot that repeats itself.

[shadow tusk]

A digital brain would and could fix the world’s problems. It would be one of us just with no physical freedom and we would add more good things to the brain

[sudden goblet]

Hello there 👋

How are you doing bro?

[stable tinsel]

That's some farfetched assumptions.

[rancid summit]

Is there anyone who actually works for OpenAI in the Discord? I have paid for a service that I don't have access to.

[supple violet]

https://status.openai.com/

[rancid summit]

Yeah, I am aware. It's been over 24 hours. Would like to hear an update to see what the progress is.

[supple violet]

that is the official status/progress page, you wont get better info than that

[rancid summit]

Yea so I am in here asking if there's a developer who can provide an update.

[supple violet]

there is not

[rigid estuary]

This is a community run server

There are developers on the server, but not active and not available for support here. If you need support you can email [e-mail]

[rancid summit]

Ok. Yea I emailed them, filled out a bug report, contacted them on the chat on the website

Dang, was hoping for some support in here

[rigid estuary]

Wish I could be of more help

[rancid summit]

It's ok. I have made the mistake of making a huge part of my job dependent on access to this new tool and not having it is making me miss deadlines for work. So I need to get a new one, lol..

[shadow tusk]

You are right it is very far fetched, but this thing has an enormous potential and can learn anything so cant it take things on?

[stable tinsel]

Everything that has great potential is also a great risk. I am not saying we shouldn't pursue it but given the state of our civilisation, there are going to be bad actors who are going to use this technology to do cruel things.

[unreal marsh]

doesn't work, he quickly forgets everything

[stable tinsel]

At this point, I wonder how much it differs from 3.5

[unreal marsh]

it's smarter

[stable tinsel]

what does it mean 'smarter'? cuz I once more feel like I am interacting with pre-turbo 3.5 when I use 4.

[unreal marsh]

it performs better at complex tasks sometimes

[maiden gull]

Why does gpt4 act so stupid sometimes

[unreal marsh]

but yea it's not like it used to be

anyway use claude

[maiden gull]

Give me 1 specific example of blah blah blah.

Okay, here are 10 different things with no specific example

[stable tinsel]

how good is claude for editing/proofreading?

[unreal marsh]

don't know, probably good, you can use it it's free

[jaunty surge]

https://openai.com/research/gpt-4 shows

[glass prawn]

Hey guys

[stable tinsel]

They also say they tweak the models all the time. If it is smarter, the difference is either becoming marginal or 3.5 is being dumbed down as well

[wraith ginkgo]

`` Server Tip: ``

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

[icy flax]

Really? Why?

[pliant mauve]

Is function calling not available for gpt4? I can get it to work with gpt3.5 but it's not working for gpt4

[modern plank]

how can i get access to 8k context or 32k context, and how much will it cost per month?

[unreal marsh]

it has a limit on how much it can remember, but an even lower limit on how much it can remember well

[desert fable]

I initially didn't realize this, but now I understand it as kind of like the context = RAM.

It doesn't remember everything in your conversation past a certain point because once the conversation gets long enough, it has to start removing less important data from RAM. Typically what I do when doing something like having Chat GPT write a story, I reinforce the important info in every message. This forces it to keep the data you want to stay in RAM.

For example, I include a short recap of a prior events in the story along with character descriptions in every single message. Otherwise, it quickly completely loses the plot and messes up details. As often as the very next message if I didn't reinforce the important details. But this is usually with prompts that reach close to the cap for prompt size, which seems to be around 15K characters long. But a very large chunk of that is the story recap and character descriptions that I need to reinforce.

An increase to context size along with an increase to maximum prompt size would be nice for things like story writing.

[desert fable]

Oh yeah, I noticed this a lot in GPT 3. With very large prompts, it'd be very hard to get it to include dialog. I'd have to spam "Include dialog. Really, really include dialog. Really, really include dialog."

And then if in the next message, my prompt is just "Rewrite that. Include dialog", it will then rewrite it to include dialog.

But with GPT 4, mostly just mentioning "Include dialog" once gets it to do it. It still has some issues with ignoring some details in prompts that has to be reinforced, but not as bad as GPT 3.

[unreal marsh]

exactly that's it, this is why having very long chats with gpt in the same session sucks

also the messages are not lazy loaded, so if they are very many it can take minutes to load

[desert fable]

Oh yeah, I had a very, very long chat thread where I was having it write stories because I like the style I had fine-tuned it to. It was taking easily upwards of 5 minutes to load the chat on a fresh browser tab.

It eventually reached the point where it forced me to make a new thread. I even tried going backwards to a previous message in the thread and it still wouldn't let me send new messages in that thread.

[glass fjord]

Says I don’t have premium but I’ve paid for it had it since release anyone else had this isssue

Just had to relog nvm

[unreal marsh]

what is alpha gpt 3.5?

[pliant mauve]

nvm, realized gpt4 is more picky on when it decides to call a function

[unreal marsh]

i did a relog and gpt 3.5 turbo disappeared, now i have alpha 3.5 instead, what is this?

[desert fable]

I see that, too, now. I have not relogged. I have GPT-4 and Alpha 3.5

[unreal marsh]

what are these secret updates

nobody said anything about any alpha

[glad fern]

Down?

[unkempt orchid]

Unable to load conversation
Unable to load history
Literally no emails about any downtime despite them spamming me 20 times a day with them normally

Absolutely bonkers website lol.

[old granite]

where did my plugins go

[old granite]

theyre back

[unreal marsh]

yea but they say "99.8% uptime"

[dreamy hill]

Is GPT4 down? I can't even choose, not at the top

ChatGPT is running, but it seems like only 3.5, not the type of responses I need, if anyone knows?

[lilac shard]

GPT-4 only remembers the last 8,000 tokens you've given it. You can't give it hundreds of blog posts.

[dreamy hill]

Yeah it's jacked up LOL, looks like everyone is having different issues so I'll just wait 😛

[elfin plume]

Stop refreshing. It's making it worse

[tawny mortar]

anyone else experiencing poor performance?

[rose garnet]

guys

chat gpt is not good

[gentle shard]

It's asking me to pay more. Is it some kind of phising scam?

[tawny mortar]

on the count of 20 everyone spam refresh, maybe that will fix it?

[rose garnet]

i can't load the history

nvm I can

[tight hill]

they just die

[rose garnet]

but its hella slow

[fresh cedar]

just came back up for me.

[gentle shard]

Couldn't have happened at a worse time.

[tawny mortar]

why does CHATGPT HATE ME!?!?!?!?

[knotty grotto]

i suggest to avoid all cap, you can get mute by automod

[median comet]

I just typed a question and it bugged out, I'm starting to think I caused it xD

[rose garnet]

do u know any good prompts

[tight hill]

also me

[undone hazel]

I wish there wasn’t so much of a response limit with gpt 4 the model is great i just hate the limit of 25

[tawny mortar]

I have a dream, that one day each one of us, will get the service we paid for

[undone hazel]

Its a great service but i never knew there was a limit and i bought it without knowing

[tawny mortar]

not only a limit

my understanding based on what I saw on reddit, is they planning on reducing it even further to 15

[undone hazel]

💀💀

[tawny mortar]

they will introduce a premium subscription that will allow for 35 for $50 or something like that

[undone hazel]

Would u buy it

For no cap

[sullen spade]

i can even wait 5s for each token if price goes down to $10 i just sub cuz gpt3.5 is not near to gpt4

[old granite]

is there any way to see when the limit on gpt-4 will be over? it just says "checl back later" and in the chats with gpt4 it does show the time anymore

[jovial kite]

ChatGPT -4 is lit !!!! 🔥

[unkempt orchid]

iirc that was also a while ago, from what I have seen on OpenAI's blog from themselves, is that we can "Expect to get reduced caps in Q3 - Q4" or sth like that.

[old granite]

nevermind You've reached the current usage cap for GPT-4. You can continue with the default model now, or try again after 4:54 PM.

[sullen spade]

what about slowing down not paid users mooreee

[haughty harbor]

Hello everybody I just updated ChatGPT for iPhone the latest version and my account disappeared from the app. It says my account and password are missing when they were there the whole time.

[bright rain]

Why when I want to continue an old conversations does it switch to gpt3.5 because "the model used for this conversation is no longer available"? Really annoying, if I started the conversation with gpt-4 i want it to continue using it.

(Yes, i still have chatgpt plus)

[undone hazel]

Its super annoying ikr they have a limit of 25 responses per 3 hrs i never knew until later on i bought the subscription today to try it out it cuts me off eventually

[old granite]

id pay triple for like 100 message limit every 3 hours

[bright rain]

that's annoying too but i wasn't at the message limit either

[knotty grotto]

you can buy a second account then

[old granite]

yeah but its a hastle because i need to keep it all in 1 chat

like i did consider it but sending several messages to describe what happened in the other chat is annoying

[coarse echo]

does anyone have know a server that has built in AI tools for creating images for free?

[haughty harbor]

Folks I really don’t understand once we pay for the service they limit us to use it. The service is really good. They should encompass more space. For subscribers I need it for work.

[knotty grotto]

wrong place to discuss, i suggest #ai-discussions

[knotty grotto]

i need it for RP as burger

[gritty garnet]

Burgermeister of MacDonald's kingdom? The sweetest RP experience you can have with GPT4? How did you know?

[forest hemlock]

They only limit gpt-4 use. It is a beta feature. Gpt-3.5 is unlimited and faster than in free tier.

[knotty grotto]

3.5 isnt unlimited even for plus user

but high limit

[old granite]

ive never run into a limit with 3.5

[forest hemlock]

I mean, there's no explicitly stated limit as with gpt-4.

[forest hemlock]

Me neither. But some people claim they have.

[knotty grotto]

it would be writen unlimited if that was true, it will make more people sub. but for troubleshoothing people since 3 month, yet it have limit

[old granite]

does anyone know if you can use plugins with the gpt-4 api?

[forest hemlock]

I'd assume it might be the same limit as with the API. But that would be almost physically impossible to reach for a normal person.

[haughty harbor]

Hi, I have a question the 4.0 version is more accurate and with more information when will they disable 3.5 and only keep 4.0?

[forest hemlock]

You can't.

Unless the plugging you're interested in is open source. In which case you can incorporate it into your app.

[old granite]

damn, is there a 3rd party app that recreates the chat functionality using the api?

[forest hemlock]

Pretty sure there are some. Though I'd imagine they face the same issue with plugins.

[old granite]

yeah prob :C

its fine, thank you

[forest hemlock]

Probably not any time soon. There are tasks that gpt3.5 is good enough for. And it's way cheaper in terms of resources and has lower response time.

[haughty harbor]

Got you thank you so very much

[gritty garnet]

I've been tooling around with an API chat interface. It's not hard to do.

but the hard part is the asyncio bits. At least for me. :P

[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.

[slender folio]

I like TypingMind

[wind urchin]

could you give a example to illustrate your point?

[wraith ginkgo]

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

Discord: Terms of Service, Community Guidelines
OpenAI: Terms & Policies

[desert fable]

This is messed up. After they yanked web browsing mode, they nerfed all chat threads that were GPT-4 with web browsing to GPT-3.5. Not even to just GPT-4.0 without web browsing.

You can't even put a chat thread that was nerfed to 3.5 back to 4.0, such as when you hit the 4.0 message cap.

Is this even legal? You pay for a service with the understanding of getting X features and then in the middle of a subscription period they take away some of those features with no warning and damage the data on your paid service?

[hasty river]

Hey, GPT-4 seems to be missing.

[wraith ginkgo]

`` Server Tip: ``

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

[cursive wadi]

@hasty river I’m seeing the same thing.

[cursive wadi]

I’m with you here…

[lilac shard]

Get an API account and use the playground -- then you have unlimited GPT-4 messages.

[old granite]

ok thanks, i already have an api acc ive just not used it

🔥

[lilac shard]

You will need to use 3.5 enough to get a bill, and pay it, before you get 4 access

[old granite]

ive gotten a bill a while ago though, i also got an email about 4 access because i sat on the waitlist or whatever

thanks though

[pure seal]

Hi, I'm using GPT 4 on pc, mozzila browser, I noticed that whenever I shutdown browser/ sleep/ or pc restart/ shut-down, I usually lost a significant chunk of generated texts. sometimes a whole "tread" dissapear, leaving me with yesterday's chat. is this a known bug? any fix?

[quasi frost]

Is there a way to get access to more request for GPT-4 - right now it limits to 25 calls in 3 hour - is there a subscription we can take?

[old granite]

yes, i asked the same thing, but it would be paid dependent on usage

i can dm you a picture of how it looks, i havent tried yet but it looks aight

[forest hemlock]

No. Though if you can use the API, it doesn't have this cap.

[shadow tusk]

Who here can code any language? This includes lua, python and c++

[coral token]

i can code in c++

[shadow tusk]

Oh ok but do you think we could make a digital brain using that language

[old granite]

wdym digital brain?

neural net?

[coral token]

can you eloborate thie features of digital brain?

[old granite]

or a bio/chem simulation of a brain

[shadow tusk]

This basically just works like a normal brain but it lives forever and can read indefinitely and never gets bored

[shadow tusk]

Yes

[old granite]

uh

good luck!

[shadow tusk]

Do you not believe it can be done?

[coral token]

ooooo u need it for ?

[old granite]

tens of billions of neurons sure you could theoretically do it but there isnt probably enough processign power in the world currently

[shadow tusk]

Just stuff like making money and fixing stuff

[wraith ginkgo]

This is a Safe for Work Discord Server

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

Posting NSFW content will result in a permanent ban.

[old granite]

n late 2013, researchers in Japan and Germany used the K computer, then 4th fastest supercomputer, and the simulation software NEST to simulate 1% of the human brain. The simulation modeled a network consisting of 1.73 billion nerve cells connected by 10.4 trillion synapses. To realize this feat, the program recruited 82,944 processors of the K Computer. The process took 40 minutes, to complete the simulation of 1 second of neuronal network activity in real, biological, time

from wiki

[coral token]

hnmmm

[shadow tusk]

Im sure they made it complicated

[old granite]

the human brain is in fact, complicated

[shadow tusk]

Yea

But you would need to simplify the brain

[old granite]

they couldnt even simulate 1% of it in real time, it took 40 minutes to process 1% of the brain for 1 second of activity

[old granite]

so like...a mouse brain

[shadow tusk]

Yes then we move forward

[old granite]

sure that can be done but mice cant really help with "making money and fixing stuff"

[shadow tusk]

We advance

Or a primate

[old granite]

already in the hundreds of millions of neurons

[shadow tusk]

Yes

Can I DM you

[old granite]

sure

[hasty river]

I am still unable to access GPT-4.

[short iron]

Any way to revert a chat to gpt 4? I used the 25 limit and waited until I had 25 more messages, I sent a followup message and it reverted to model 3.5 which isnt desired

[lilac shard]

You can copy your existing chat, edit out the extra text and use the whole thing as the prompt in a new chat.

[short iron]

The conversation was so long though it went on for days

[lilac shard]

ChatGPT is only using the most recent 8000 tokens of your conversation.

[cunning roost]

How do we do this?

[lilac shard]

Go to https://platform.openai.com/playground and click the "sign up" button.

[cunning roost]

there is no gpt4 as selection

only 3.5

[sullen plinth]

You need to buy the premium

[lilac shard]

You will need to use 3.5 until you are billed, pay it, and then you'll get 4 access.

[cunning roost]

Its about 2 months now, im not billed yet :/

[lilac shard]

How much money have you spent?

[cunning roost]

$3 already

[wraith ginkgo]

`` Server Tip: ``

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

[lilac shard]

They should have billed you, but if they haven't I suppose you can spend some more.

[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.

[inner rapids]

is code interpreter still alpha access?

[supple violet]

beta

[inner rapids]

ahh oops forgot to enable it in settings lol. Thanks!

[harsh talon]

1

[unreal marsh]

when will "As an AI developed by OpenAI" be eradicated?

[hallow owl]

ChatGPT has guardrails, test them and you'll always get a response like that

[abstract coral]

Question: Does GPT-4 read images via URL or is it unable to? I have been trying to do that via discord image links but it keeps saying that it doesn’t read any images via a link

[knotty grotto]

Gpt 4 browser is disabled, also reading picture isn't available

[abstract coral]

Does GPT-3.5 work with reading pictures then?

Or is that also not available?

[knotty grotto]

No

[abstract coral]

Ah.

[knotty grotto]

#gpt-4-faq

[abstract coral]

I see, thanks for the help though

[knotty grotto]

We all impatient to try this function so I understand the hype :3

[honest lance]

WHAT IS WRONG WITH GPT-4 !!! THIS NERF IS BAD. HE CANNOT UNDERSTAND SIMPLIEST TASKS IN HTML AND CSS

[wraith ginkgo]

`` Server Tip: ``

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

[pure trench]

I'm getting really frustrated with GPT4. the PDF plugins work for a couple of questions, then GPT4 goes off and starts making up stories!! IS there a way I can keep GPT 4 focused just on my PDF? I waste a lot of prompts just to get 1 task done, I then have to wait 3 hours later for my 25 cap. It's actually annoying. Who do I discuss this with?

[whole kettle]

What happened with the web browser plug-in ?

[knotty grotto]

#announcements

[pure trench]

I'm happy for OPENAI to use my account for research. Please help because I am at wits end and my deadline is due in 5 days

[merry leaf]

me french

allo

[knotty grotto]

Hi, just as precautions for you to avoid being mute ,server only supports english

[ionic tide]

It's blatantly ignoring my text

[modern cosmos]

try to restart the conversation? there's a context limit and after that it tends to 'forget' what it said at the start of the conversation

[pure trench]

tanks Daksh.. how do you restart the conversation? just refresh the page?

[river dune]

axi

[hearty mulch]

Hello everybody. Any news about changing back a thread automatically downgraded from gpt 4 to gpt 3.5 because a problem with the payment renewal? The solution I found on Google is not working anymore. (That one using the Inspector › Network... and changing text-davinci-002-render-sha to gpt-4.) I have a big and important thread I've been working for months and don't want to want to continue it with gpt 3.5. Anyone knows how to solve it?

[halcyon ivy]

hello, how can I give gpt 4 internet access? is it with a plugin?

[inner trail]

Yes at the moment only by Plugins

such as WebPilot, Metaphor, etc...

[halcyon ivy]

thanks a lot

[formal dome]

If you have plus, a version with web access was available, but they took it down. YOu can use bing chat instead.

[charred flint]

How to proceed innovation in car OEM?

[flint axle]

if your trying to speak with gpt then you need to go to the webpage

chat.openai.com

[wind urchin]

now it's not a speculation anymore

(about OpenAI nerfing GPT-4)

I'll post the graph from the paper in #off-topic

[fierce spoke]

the paper is... shaky at best

the issues cited are mostly formatting issues

[flint axle]

they have added guardrails as time has gone by most likely

[fierce spoke]

and variations in short/long responses

not changes in fundamental reasoning abilities

[wind urchin]

huh?

they literally tested reasoning

what are you talking about

[fierce spoke]

give link to the paper

[wind urchin]

they tested visual reasoning, math reasoning, code generation, etc

2307.09009.pdf

[fierce spoke]

But if you look at the EXAMPLES of their prompts

the issue w/ the primes for instance - is that june responds to short vs march

And the coding issues are literally just it appending python

[floral atlas]

If you have plus, a version with web access was available, but they took it down. YOu can use bing chat instead.

[wind urchin]

you can write your own refutation paper, where you conduct extensive experiments

sorry, otherwise I will not accept your anecdotal guess with no evidence shown (over a research paper)

[modern cosmos]

the 'new chat' option should do it but yeah visiting the website again also works!

can't send an image here unfortunately

[fierce spoke]

i was citing the example given in the paper itself

after doing some further reading and testing

you are right - there has been regression: sorry for my initial doubt

[lost badger]

GPT-4 is getting worse over time, not better.

Many people have reported noticing a significant degradation in the quality of the model responses, but so far, it was all anecdotal.

But now we know.

At least one study shows how the June version of GPT-4 is objectively worse than the version released in March on a few tasks.

The team evaluated the models using a dataset of 500 problems where the models had to figure out whether a given integer was prime. In March, GPT-4 answered correctly 488 of these questions. In June, it only got 12 correct answers.

From 97.6% success rate down to 2.4%!

But it gets worse!

The team used Chain-of-Thought to help the model reason:

"Is 17077 a prime number? Think step by step."

Chain-of-Thought is a popular technique that significantly improves answers. Unfortunately, the latest version of GPT-4 did not generate intermediate steps and instead answered incorrectly with a simple "No."

Code generation has also gotten worse.

[lost badger]

The team built a dataset with 50 easy problems from LeetCode and measured how many GPT-4 answers ran without any changes.

The March version succeeded in 52% of the problems, but this dropped to a pale 10% using the model from June.

Why is this happening?

We assume that OpenAI pushes changes continuously, but we don't know how the process works and how they evaluate whether the models are improving or regressing.

Rumors suggest they are using several smaller and specialized GPT-4 models that act similarly to a large model but are less expensive to run. When a user asks a question, the system decides which model to send the query to.

Cheaper and faster, but could this new approach be the problem behind the degradation in quality?

In my opinion, this is a red flag for anyone building applications that rely on GPT-4. Having the behavior of an LLM change over time is not acceptable.

Have you noticed any issues when using GPT-4 and ChatGPT lately? Do you think these problems are overblown?

By @svpino in Twitter

[lost badger]

Do you guys agree with this? ☝️

[knotty grotto]

I will ask chatgpt to make all your post short and maybe read it later 🥲

[lost badger]

😂 Note that the post was written in Twitter by @ svpino, though

[low furnace]

I think my chat leaked the session of someone else. I asked Chat to summarise an article about climate change, I hit the "continue generating" and Chat suddenly talked to a person called Janice bout her personal problems.

Janice if you hear me, we all feal it like this sometimes. Chat is a good pal. I hope you didn't get my depressing climate change article in your face. It wouldn't help you so much.

Janice:
What is living for? I'm struggling to find a purpose or meaning in life.

ChatGPT:
I'm sorry to hear that you're feeling this way, Janice. Finding purpose or meaning in life is a deeply personal and complex journey that many people grapple with at various points in their lives.

[tropic spear]

That is pretty consistent with my experince. I got access to GPT4 pretty early in the release cycle and when it came out I tested it on some of the hiring tests my company does on candidates and it was satisfactory for more than a dozen positions for entry level. I did it again recently and the only one it really stayed on topic for was for an entry level marketing job 🫤

[wintry mirage]

I have been testing and created a video demonstrating how to get ChatGPT, and ClaudeAI to work together to create powershell scripts, or other types of code. Im looking for feedback from anyone interested in this use of LLM systems.

https://discord.com/channels/974519864045756446/1130507595724107827

[inland hinge]

what can i doing in here?

[tropic spear]

I'm not sure I see the distinction. They may be making the changes for valid reasons but the end product is objectively worse now. Not in a small way that would be a normal day to day variation common in learning products. Since they are not warning us or spinning up a new model to integrate the changes with, it means we can't trust that the product that is fit for purpose today will continue to be fit for that purpose tomorrow.

[wraith ginkgo]

`` Server Tip: ``

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

[warm pewter]

hi i would like to ask. as a chatgpt plus subsciber how many tokens in gpt 4 can we use? 8k or 32?

and how can we avail this 32k if available?

[pure trench]

I wish they would increase the cap .

[supple violet]

8k. 32k is not available nearly to anyone

[warm pewter]

ok. i think we will wait several months haha

thanks

[knotty grotto]

Api isn't related to chatgpt subs

[wind urchin]

We need more messages 25 in 3 hours is kinda trash

Good for you bro

[pastel sky]

That feeling when I was right about GPT 4 being objectively worse than GPT 3

[pure trench]

What's up?

[gritty garnet]

GPT4 is only terrible if you don't know how to give it the right personality with knowledge of how you want its outputs to look. GPT4 requires a lot more love up-front, but pays off big.

[wind urchin]

Hello, since yesterday the chatwithvideo plugin no longer works for me, has this ever happened to anyone?

[elder viper]

@abstract charm

[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.

[slow prism]

/chat

[thorn tide]

I complained about this before here, its just getting worse and more stubborn to even work more tasking programming challenges

[thorn tide]

sometimes you have to convince it to actually write code

[grand saffron]

The cap increased to 50?

[supple violet]

no official info but yes different people have various caps now larger than 25

[grand saffron]

👍

[versed canyon]

Doesn't seem to have changed. I've seen the reports from others.

[wind urchin]

I wonder what the tier list is based on I'm at 100, others are at 200

maybe random rollout

[supple violet]

yeah i have 200

[cunning rose]

200? OMG.. send me screenshot by DM.. I still have 25

[wind urchin]

lucky af

[versed canyon]

me too, glad I am not the only missed

[supple violet]

it just says "GPT-4 currently has a cap of 200 messages every 3 hours." so nothing more exiting to screenshot

without official info it can be a glitch, short test or slow rollout

[wind urchin]

Microsoft usually does random rollout with Bing, OpenAI does tiered rollout I think

Well, for everything else.

[cunning rose]

They might announce something tomorrow, since it would be 2 months after a big update. Including custom messages and this Alpha Model..

[versed canyon]

Maybe it's a psychological experiment to see how many quit in disgust in the light of random tiering. I've watched films where animal psychologists hand out unequal rewards to see what the critters would do. 😉

[wind urchin]

lol same

Sam Altman is doing the Stanford Prison experiment on us

[supple violet]

😅

[hollow swan]

Same

[wind urchin]

Is gpt4 still nerfed?

[lusty pike]

Nerfed?

[wind urchin]

jesus christ

that's insane

oh well.. all good things come to an end I suppose

[unreal marsh]

guys lets file a lawsuit against OpenAI, it's not fair to raise the cap for some, but for others to leave it 25

[wind urchin]

you mean the nerfed gpt-4

[versed canyon]

You'd think that. The problem is that OpenAI hasn't agreed to any SLA in exchange for subscriptions. It can certainly create bad will (customers rarely forget these things) but it probably doesn't have a legal basis for individual or class action base unequal service.

[wind urchin]

you can still use gpt-4-0314 (March gpt-4) on nat.dev @pastel sky
(nat.dev is an LLM playground)

[hazy hamlet]

Same. Does anyone know how to become a more equal pig?

[wind urchin]

yep

[pastel sky]

Risky click

[wind urchin]

who me?

then google the site first and read about it.

[hazy hamlet]

Your comment may be a joke but we're paying the same thing

[supple violet]

you can use the march gpt-4 at openai playground too

[wind urchin]

no waitlist for it?

[supple violet]

well you needed to be paying api dev to get access during the blog post to get access

[ionic tide]

How do we know what the limit we have is?

[wind urchin]

that wasn't my question

[versed canyon]

Start a new chat and look at the bottom.

[ionic tide]

TY

25 ARG

[supple violet]

waitlist was removed, at this point you either have access or not

[unreal marsh]

are you sure that hasn't been nerfed as well?

[ionic tide]

Back to haikus and dalle plugin

[versed canyon]

I feel your pain. I'm on the 25/3hr slow train too.

[hazy hamlet]

and me over here been hitting that limit constantly for months

[versed canyon]

Maybe they're not increasing for the people hitting the limits out of pettiness. 😉

[hazy hamlet]

I have given it some....questionable tasks

but I was testing it!

that's what they wanted right??

[supple violet]

it works exactly the same as it did in march, not all devs adopted the latest yet due required changes

[ionic tide]

I mean it has helped me get on their insta 2x this week so yay

[wraith ginkgo]

This is a Safe for Work Discord Server

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

Posting NSFW content will result in a permanent ban.

[wind urchin]

yes I just checked

well nat.dev is nice because it also gives access to 20 other LLMs

but looks like all of OpenAI's LLMs are the same as on the official playground

but I dont think gpt-4-0314 can be used as an API (without a waitlist)

I tried

[robust light]

Wait so some people have higher caps past 25? I missed that.

[ionic tide]

I guess.

[wraith ginkgo]

`` Server Tip: ``

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

[grizzled flax]

Just happened a bit ago, I'm at 50 now, others are at 100 or 200. I think some are still at 25. No announcement of rollout plan yet as far as I know.

[robust light]

Wow! That is awesome!

[ionic tide]

I would love a higher cap

[robust light]

I am still at 25 but will update if it changes

[versed canyon]

Isn't using the word "plan" around an OpenAI product a bit ambitious? 😉

[wind urchin]

im at 50

Hello. I just subbed to plus not realizing there was a message limit with GPT4. How long till you can go back to 4 once you hit the limit? Do I have to wait a day or something?

[terse forum]

Does anyone know why GPT 4 is giving such vague solutions, it looks like GPT 3 especially with code, it used to give much better solutions.

[grizzled flax]

If you hit your limit in a GPT-4 chat and switch it over to GPT-3.5, there's currently no way to switch that chat back to 4. You can keep using GPT-4 in another chat though once your usage cap limit resets.

[wind urchin]

thanks

[fierce spoke]

gpt-4 didn't actually get worse

there were big problems with the paper that declared it did

for instance, the objections were to its behavior, not capabilities

and the math questions it was tested on were limited at best

[formal dome]

💯

[fierce spoke]

(literally just "are these numbers prime")

[supple violet]

the code test was also flawed

[fierce spoke]

for instance - they test 500 prime numbers to see if gpt-4 got them right -> march got it right, june didn't. (90% vs 2%). So GPT-4 must have gotten worse - no, its just that gpt-4 in march liked saying everything was prime, and gpt-4 in june liked sayign everything was composite. Do this test with the composite numbers, the accuracy is inverted

[fierce spoke]

they were like:

This code is wrong because it has python attached

[ionic tide]

Okay has anyone found convos or images from GPT disappear? I was working on my dragonborn yesterday and cannot find any images today

[fierce spoke]

gpt doesn't generate images?

[ionic tide]

Plugin dalle

So it does

[fierce spoke]

ah

[ionic tide]

In the JSON file I see the notes yet cannot find it in a convo

Very odd

[raw plover]

Is there a specific reason in why some users have 50 message per 3 hours and others on 100. Or is this an update to all accounts eventually

[ionic tide]

Remember some of us are still @25

[raw plover]

Do we know the reasons for this? Or did they plan for 50, then changed to 100 etc seems odd

[ionic tide]

exaggerated wink

[hasty river]

Every single one of my chats has reverted to GPT 3.5. I need a way to change them back. Please.

[ionic tide]

I found if you wait for credits to refresh it can go back to 4

[hasty river]

No, they haven't. THey all reverted to 3.5 when the issue with payments happened yesterday: where a lot of people like me had their plus payment go through, only for it to still not give access to GPT-4.

[ionic tide]

That idk then I wish you luck!

[raven wagon]

100 here

i just wondered why, but that's actually awesome. no one know anything specific?

[lilac shard]

That's hilariously foolish of them!

[safe venture]

yeah, I haven't noticed any changes in GPT4 and I'm still getting it to churn out the same code it always has - I suspect the people saying its worse are falling for reinforcement bias

well, partially reinforcement bias, and partially pushing the envelope on the technology as they got more comfortable with it

[outer gulch]

Just had this pop up on my ChatGTP 4, nice 🙂
'GPT-4 currently has a cap of 200 messages every 3 hours.'

[blazing shard]

was at 100 just got 200

[minor zenith]

what? i only have 50 lol

[blazing shard]

slowly rolling out to everyone

[minor zenith]

oh i see

[blazing shard]

basically 1 message per min to never hit the cap

Ill take that tbh

[unkempt plank]

Wait, 50 per 3 hrs? Am i dreaming rn

[dark geyser]

I would much prefer a usable gpt-4 at 25 messages per 3h than the now unusable (for me and my needs at least) thing without limits

[unkempt plank]

Why unusable?

[dark geyser]

with the latest changes, it has become incredible lazy, and where before it tried its best to do the refactorings I described with the code I gave it, so that I mostly just needed some small fixups and saved a lot of time... the current version pretty much always only produces just an empty framework, with //your code goes here comments in various places and then tells me that the code is very complex and I should just adjust my code myself and put it in these places. And almost no amount of cajoling can get it to actually do the work that it did before without complaint.

[unkempt plank]

Really?

[dark geyser]

It might be worse for me than others because I'm using a programming language that's less common. But it did this work fine before and doesn't do it now.

[unkempt plank]

Whats the lang?

[wind urchin]

I speculate that because the jailbreak problems, they limited it

[dark geyser]

Pascal / Delphi

[unkempt plank]

Where did this problem thing arise lol can u link me to it

[wind urchin]

No well, I don't know if it's a big problem for OpenAI, but I suppose that improve the anti-jailbreak could have collateral damage

[tawdry ridge]

Why are they rolling out different Chat GPT-4 limit caps for different people? I've seen people on Twitter with 25 still, 50, 100, and one guy with 200! lol

[dark geyser]

Well, there are a lot of complaints about gpt-4 having become dumber and lazier all over the web recently. And I definitely see that myself. So I don't know what they changed, but it's noticable and noticable to a lot of people. And if that is the cost for being able to reduce the usage cap, than it is not a good deal from my perspective, because I've gone from saving quite a lot of time using ChatGPT with GPT-4 model to mostly not breaking even when I want to use it.

[unkempt plank]

Honestly yeah id rather have a lower cap and better gpt 4 than higher cap and dumber gpt 4

Quality > quantity

[dark geyser]

or give us a choice, I mean, they got the possibility to select different models already, just give us the older gpt-4 with the old caps and gpt-4-turbo with larger caps, so people can make an informed decision... it would also give them valuable feedback in regards to what users actually want

[tawdry ridge]

Normally, I would have been happy with 50 messages. But, knowing that some people got 200 is putting a sour taste in my mouth. lol

[lusty pike]

Why are alot of people saying gpt 4 has gotten worse

[dark geyser]

Because it objectively has?

[unkempt plank]

Openai ceo apparently said hes looking into it

[lusty pike]

In what way

I havent used it so im just wondering

[unkempt plank]

Just scoring lower on many tests

[dark geyser]

Just look up a screen or be two, I explained it just some minutes ago

I don't care about scores on tests, I care that before it saved me a lot of time, and now for the same type of problems and requests, it doesn't anymore.

[unkempt plank]

Well, since the ceo looked at it, itll probably get btr

Cuz theres actual proof that the models degrading now

[shadow tusk]

WORM GPT

Anyone heard of that?

[knotty grotto]

hum thats not related to chatgpt 4?

[brazen wedge]

I’ve still got 25

[wind urchin]

Damn nice

Still at 100

[marble halo]

omg

[cedar ice]

\o/

[versed canyon]

Me too. OpenAI is sure showing their true stripes.

[wind urchin]

bruh what is that announcement lol

[wary marsh]

Let’s go

[wary marsh]

Not complaining

[violet spade]

Finally bro, took them long enough to increase the daily limit

Next up: 100 messages per 2 hours

[still willow]

50 messages per 3 hours

[proud oak]

finallly man

[wind urchin]

Me neither, but people have been getting more on here.

[trim elk]

Where is the new stuff generally available?

[marble tree]

how can i tell if i'm using chatgpt or gpt4? does it automatically go back to gpt4 when my limit resets?

[proud oak]

not much but its better than nothing

[marble tree]

guess thats a no 😅

[old gulch]

When we getting bing plugin search back? 😐

[marble tree]

oh, i see. i just have to scrollup and the dropdown shows the model

[unkempt orchid]

Wow, finally some actually interesting and good news lol. All I hope is that to compensate for it, they wont make the bot even dumber.

[still willow]

You can choose between 3.5 and 4 if you have access. If you've reached the cap, it will return an error and prompt you to use 3.5 instead, including a timestamp when 4 will be available for you again

[vocal olive]

When's the API gonna be available for gpt4

[marble tree]

i always click switch to gpt-3.5, but i've never seen an option to switch back

always have to start a new chat and lose previous context in order to use gpt-4 again

[still willow]

Yeah, I believe you can't switch it back from 3.5 to 4 in the thread itself

[knotty grotto]

#announcements

[marble tree]

ah okay , just making sure i wasn't overlooking a feature that would allow me to. thanks

[novel vessel]

So I've seen that some people pay 40$ a month for ChatGPT and I'm wondering how you get that option because I was never given an option. When I applied I got put on the $20 a month Plus option when I finally got off the waitlist

[unkempt plank]

U can pay 40?

[novel vessel]

Like bruh I might have paid you $40 dollars why you gotta be like that

[fierce spoke]

there is no conclusive proof that gpt4 is worse - outside of ancedotes and a sloppily done paper

[novel vessel]

I haven't personally noticed that tbh

[unkempt plank]

Many many many people said it tho

[novel vessel]

Unless this just happened today

[unkempt plank]

Like not just in discord

[fierce spoke]

hedonic adaptation

• confirmation bias

[unkempt plank]

Maybe

[trim elm]

How can I go back to GPT-4? I accidentally clicked switch to GPT-3.5 because I reached the cap...

[fierce spoke]

there isn't actually any conclusive evidence that it has gotten worse - its mostly just people realizing that its not AGI

[unkempt plank]

Tbh even the ceo kinda replied to that lol

[knotty grotto]

you cant

[unkempt plank]

U cant, but what u can do is ask gpt 3.5 to summarize what u went thru and then fact check it, then make a new gpt 4 chat and paste it

[placid sluice]

if you say this, you haven't been using GPT for complex tasks, it is 100 % dumber

[vocal olive]

I said model = gpt-4 and it says model gpt-4 doesn't exist

[unkempt plank]

Honestly yeah lol it messes up simple tasks now

[mystic quiver]

Let's gooooo

[fierce spoke]

I have - I've been using it for advanced convolutional neural network programming, D & D campaigns, creative songwriting, generally run through 75-100 prompts a day

[mystic quiver]

more than 35 messages W

😮

[knotty grotto]

GPT-4 API is now available to all paying OpenAI API customers

[vocal olive]

It's billed per month right?

[placid sluice]

if so I'm honestly shocked you haven't noticed the change, I use it for translation and it can't even do proper word play any more

[unkempt orchid]

Can second this. Doing theoretical physics with it and over the time since it came out, it has 100% gotten dumber lol. If that puts me under buzzwords and "confirmation bias", so be it.

[unkempt plank]

Idk man, code interpreter messes up alot for me now

[oak crescent]

I guess you should make proper arrangements before any process ?

[trim elm]

Oh nah, f this

[knotty grotto]

per usage of token if i member

[novel vessel]

What do you mean by that

[trim elm]

How can I copy the whole chat?

[unkempt plank]

Ask chatgpt to summarize for u

[vocal olive]

But like when does it charge me

[mystic quiver]

yea translation wise, gpt-3.5 lacks so much compared to gpt-4

[trim elm]

Ok

[vocal olive]

I like gpt3.5 for speed though

[placid sluice]

I was talking about GPT-4 lol

[knotty grotto]

best place to aks #dev-chat

[mystic quiver]

it's changed?

[placid sluice]

yeah, it's mediocre now

[mystic quiver]

nah I like more organized information rather than speed

[fierce spoke]

now that you say it, i have subjectively noticed it, but i assume its just due to my realizing the program has limitations. in the beginning i only gave it toy tasks to test its abilities

[mystic quiver]

how? i don't think they changed anything

[placid sluice]

I sitll use the previous version on the API

[vocal olive]

Fair enough

[unkempt plank]

I use it mainly for writing machine learning stuff using pytorch or tensorflow or mainly to solve like some math olympiad problems and it made a crap ton of mistakes here and there

[mystic quiver]

what would they have changed to make it worse

[unkempt plank]

Took me like 6 prompts to get the code working from it

[oak crescent]

Yeah absolutely dm for more information !

[placid sluice]

nah, I haven't changed the usage of it and the quality is definitely worse, I still do comparisons on the API and it's so obvious

[fierce spoke]

hmmm

[unkempt plank]

Yep

[mystic quiver]

do you have any reasoning on why?

Lower the non-API quality so they can accept more messages?

[unkempt orchid]

True, but I've speculated before that perhaps the reason it can come off as "dumber" is exactly because of the censorships increasingly put on it. It used to be far more confident and outgoing with its speech, now it's just littered with canned and straight-forward responses, half of which you can predict coming word for word.

[unkempt plank]

Openai had to tone down the accuracy for more msgs

[placid sluice]

for example, before, I would use it to brainstorm jokes and rhymes, things that would take me a lot of time otherwise, and it would give me a lot of legit options, now it nly outputs nonsense

[unkempt plank]

Yeah this as well

[mystic quiver]

Yea I agree

[ripe basalt]

there's a known issue that training models to be more helpful tends to make them stupider. the newest model by meta addressed this by having dual goals. supposedly. so its a very well understood thing, not any kind of conspiracy like folks seem to imply

[unkempt orchid]

Its manner of speaking used to be more lifelike and varied, unique, so to speak.

[mystic quiver]

You just gotta be a good linguist and ask stuff in the right way

[placid sluice]

this

[fierce spoke]

exactly im concerned about this "conspiracy" rhetoric

[mystic quiver]

If you ask stuff in the right way you can get the best answer

[ripe basalt]

i personally think that Claude 2 needs to tone down the helpfulness, its cloyingly helpful and it forgets the most basic stuff. i have to tell it repeatedly not to tutor me and just give me the code

[ripe basalt]

This is correct, prompt engineering is a real thing

[placid sluice]

I've been writing good prompts for months, it's not the prompts, it's just dumber

[unkempt plank]

Same

[craggy pumice]

#advanced-data-analysis

[ripe basalt]

and this is also true, some research institute did it

[craggy pumice]

oops wrong chat

[unkempt plank]

I mean no

[ripe basalt]

ha

[placid sluice]

yeah, it concerns me too that every time the topic is brought up, people are like "write better prompts", no, my prompts couldn't really be better

[unkempt plank]

Wait i mean yes

[placid sluice]

also it's concerning that before it would perform very well with a simply prompt and now it can't, why? because it's dumber, plain and simple lol

[unkempt orchid]

Literally 😀

[unkempt plank]

I mean i have an entire google docs of my prompts for this, ive been using the same prompts ever since gpt 4 came out and it literally got worse

[mystic quiver]

Yea I've realized I've had to make my prompts even more detailed or sometimes less detailed to get the right answer I need

Though it's best to always open a new chat with it, if you're trying to get good results with GPT-4

I've noticed that asking about different topics in the same chat really screws with the results.

You get way better results if you don't switch topics inside the same chat

[unkempt plank]

Lmao yes, there was one time gpt 4 legit just couldnt get it right then i just created a new chat and it solved it onstantly

[unkempt orchid]

This too 100%, every time I make a new chat, for the first little while it comes off a lot smarter. Even if the topic in the previous chat more or less stays the same, I think it simply just gets overwhelmed by all the content.

[cursive monolith]

Hi everyone, I am looking for expert having experience in GPTBOSS.

[lusty pike]

Does everyone have 50 or do some still have more

[unkempt plank]

I think some got more, but everyones limit should be boiled down to just 50

[mystic quiver]

Yea I think the problem is that it tries to get information from all of the previous messages you've sent in that chat

So GPT will almost mini-fine tune while in that chat area

And then when you open a new one it's a fresh GPT that's ready to go

[fierce spoke]

that's what the P is for ;D

[leaden axle]

I'm still 25

[knotty grotto]

#ai-discussions

[cursive monolith]

Have you any experience about this?

[knotty grotto]

thats not related to the channel , i suggest to go in the right channel

[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.

[cursive monolith]

Actually, I need the powerful prompt that my chat bot acts as expert.

[mystic quiver]

Well that means pre-training, it's like semi training while in the conversation but it's only temp training.

[mystic quiver]

Use prompt engineering

[cursive monolith]

What is the approach?

[mystic quiver]

Just ask it to be really confident and pretend to be the expert.

And it'll listen

[cursive monolith]

I have tried a lot but chatbot didn't act I thought.

[mystic quiver]

It's limited to 2021

[cursive monolith]

Is there any prompt package for this?

[mystic quiver]

No, you need to create your prompts yourself

If there's content you want it to know

Just copy and paste it

[acoustic wasp]

prompt package is crazy

[mystic quiver]

Fr lol

[opal void]

Supposedly it can read websites according to a notice from may and the faq. also it is supposed to support plugins but i can't find anything other than the wsitch to enable plugins, not where to chose them

[mystic quiver]

The plug-ins let it read websites

But they disabled bing

So it's temp offline

[opal void]

so it indicates that it has access to plugins and third party plugins. where would i find plugins to enable other than the single switch to enable plugins?

[fierce spoke]

that was the joke lol - the idea of pretraining is that it already knows most of what it needs, and can figure out the rest from context lol

[mystic quiver]

I guess ya

But it's weighing context

[opal void]

or was the web reading plugin it?

nm i found them

[wind urchin]

Does anyone have an account that I could possibly use? The gpt plus ofc

Some of the plugins don’t work with non premium account

[knotty grotto]

unfortunately thats against the rule :/

[acoustic wasp]

are you talking about chat gpt or something else? Cuz if you're referring to api you can just use functions.

[opal void]

i was refering the chatgpt. you can turn on plugins and if you mouse over gpt-4 you get a menu that lets you load plugins

i couldn't find that atfirst but now have

[acoustic wasp]

Yes I know, I was just confused because it shows plugin store in the dropdown

[worn panther]

when will bing browsing fuction return

[acoustic wasp]

nobody knows

they probably have to completely rework it to prevent data scraping workarounds

[worn panther]

damn

[opal void]

Scrape Scrape!!

[wraith ginkgo]

`` Server Tip: ``

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

[foggy lantern]

GPT4 has been labotimized to such a degree it's questionable whether it's worth using anymore. Heck Im getting better answers from 3.5

[acoustic wasp]

Where's that tweet about openAI discussing model improvements and confirmation bias....

[tall basin]

where can i complaint how terrible gpt4 has become lately

[forest hemlock]

In your bathroom.
Open AI doesn't guarantee the quality of beta features according to their TOS. Besides. It's very subjective. I haven't noticed any degradation in it's quality, at least in my use cases.

[wind urchin]

come up with benchmarks so you don't have to guess

[fierce spoke]

i've seen so many "gpt4 has become terrible based of subjective impressions that i don't have hard evidence of" takes recently

[dark geyser]

Taking the first post from a old chat history (so that it's answer is not influenced by any previous context), and posting that exact message to the current gpt-4 produces an objectivly much worse outcome (it consistently refuses to do the actual work that it did before, and just puts //your code here or such comments in various places and tells me to do the work myself), even when then trying to cajole it into doing the work it refuses and says it's too complicated.

it consistently was able to do this work before, and it consistently refuses to do it now

[acoustic wasp]

I do not have these issues. "Objective" is doing a lot of heavy lifting here as well.

[tall basin]

There are articles and so much evidence to show that they are worse, but if you are asking questions like how to pick noses and stuff, i see why yall dont see the big difference. (one of them being Standford study)

[acoustic wasp]

Are you primarily using chatgpt or gpt-4 api?

[acoustic wasp]

being mad and crying in chat gets you nowhere, if you'd like to cry to openAI I think they have a support link on their site

[tall basin]

lol i aint crying my original question was where can i file a complaint

[acoustic wasp]

the message I'm responding to is full on crying

[dark geyser]

chatgpt, but always with the gpt-4 model selected, the 3.5 is unable to do this work.

I give it code and exact instructions on how to refactor that code. I can't give it all the information from the outside context of the code because a) that would take too long to search together and b) it doesn't fit into the 8k token context window. So it has to work simply with what I give it and make assumptions from that code.

Yes, it made mistakes before, that's fine, I know exactly what I want and I can quickly fix these mistakes, resulting in overall much saved time as ChatGPT took over most of the mindnumbing mechanical work of refactoring the code, and I could concentrate on figuring out what to refactor, and fix any minor mistakes ChatGPT made.

But now, it simply refuses to do the work at all, saying it's too complicated.

[acoustic wasp]

Yeah I think if you want to do something sophisticated you should look into using the API.

[dark geyser]

I can't because while I used the API before, I didn't use it enough to actually get charged, so no gpt-4 for me

I would love to compare the march version of gpt-4 against the current one using the API on the exact same questions

[acoustic wasp]

Have you tried the 0613-16k-turbo 3.5 model?

[dark geyser]

yes, it doesn't help, 3.5 isn't smart enough to do this work. the larger context window doesn't help me, because the complete context would be a n 8.9 million LOC application, and if I have to pick out code from all over that project for every single question to give it what it now seems to need to even want to attempt what I'm asking it to do, it's going to take much longer than the time it's saving me when it works.

[acoustic wasp]

I've been using it for coding and it's been working for me great. GPT-4 is a bit too costly for my tastes for the scale I'm working at.

[acoustic wasp]

you should consider embedding solutions

[dark geyser]

most likely because I'm using a somewhat more rare language

[acoustic wasp]

That is the inevitable end state of where you're heading

Look into semantic kernel if you haven't already.

[dark geyser]

the central point is, it doesn't really need to understand all that outside context just to move some stuff around, and it was happy to just move the identifiers around before without knowing how the full types were defined. something has changed, and it makes gpt-4 objectively worse for the particular requests I'm making of it.

[peak pelican]

where can we learn more about code interpreter?

[acoustic wasp]

It hasn't been made objectively worse because your use case isn't working as optimally anymore. In fact, it could be better for your use case if you apply some newer techniques.

[main zenith]

Can gpt4 generate the picture you want?

[acoustic wasp]

no but you can use it to create prompts you can use in midjourney

[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.

[dark geyser]

I'm relatively confident this will sort itself out over the next year or so, if not by OpenAI then by some other LLM, but it means that right now it's not saving me as much time anymore as it used to. And I don't see the possible return on investment of time I would need to put in to figure out how to get it to work again. So I'm just back to moving code around by hand once more for a while.

[acoustic wasp]

Yes, the issue is the LLM, everyone else is wrong. Entire businesses making $$$ off the same model you claim is getting worse (with far more complex use cases, mind you)? They're wrong and not real.

Instead of giving up, you could just try to use the several more advanced methods and systems that are producing phenonmenal results with the same model. GPT-4 is still by far and away the best model for coding so you're just holding yourself back.

[dark geyser]

I'm not saying that many people can make money of it, and that it can do a lot of things. All I'm saying is that there is no mistake that it's behavior has changed, and changed in such a way that my particular usage case now results in a worse outcome than it did before.

[wind urchin]

Enter in a difficult prompt that only GPT-4 can solve, and track how well it does over time.

[main zenith]

Is there any way to earn 2,000 yuan a day?

[wind urchin]

I do this, and quality hasn't changed

[acoustic wasp]

I will remain skeptical until I see some genuine research at scale that agrees with your sentiments. Not just random articles. What you're dealing with is minor leagues compared to what I'm seeing people use LLM's and multi-modal's for.

[wind urchin]

if anything, it has gotten faster and slightly better.

[acoustic wasp]

this sentiment is corroborated by all the research I've seen

[dark geyser]

I've plenty of work to do that is far beyond the current model outside of the refactoring I do now and then. The amount of time I have to invest into getting it to do that small part of my work once more now that it doesn't want anymore is in now relation to the time I expect I would be able to save over the next few months. So, no, it's not a worthwhile time investment for me current. Especially given how quickly the models keep changing currently and the high probability that there will be some model which is going to be able to do this again without further effort by me in the foreseeable future.

[main zenith]

Is there any way to earn 2,000 yuan a day?

[dark geyser]

and yes, I full agree that it's faster now

[acoustic wasp]

get a job????

[wind urchin]

I mean it has 100% gotten faster, I think it has gotten slightly better for my use case, but I'm less sure about that.

[acoustic wasp]

openAI themselves have said the model has been improving incrementally for a while now

[dark geyser]

This is not ChatGPT. Got to https://chat.openai.com/ to talk to ChatGPT.

[acoustic wasp]

and I think AI research houses like MIT and META backing those statements publically holds more stock than folks on discord.

[dark geyser]

Again, I do not question that it has been getting better in most if not all of the benchmarks around.

[acoustic wasp]

That's your choice and that's fine. Personally, certain tools have accelerated some of my most complex use cases. I'm dancing with the devil known as Microsoft right now because of the insane improvements I've found while using functions in C#

[dark geyser]

And I still use it and find it very useful for many things outside the particular code refactoring it now rejects as too complicated.

[acoustic wasp]

now idk what language you're working with, I'm not trying to compare, just sharing my own experience

[dark geyser]

pascal

[acoustic wasp]

bro in pascal smh my head

[main zenith]

Why can't gpt be used in China?

[acoustic wasp]

nah I'm playing, nothing wrong with pascal

[dark geyser]

because the chinese goverment has decided to censor it

[acoustic wasp]

because your gov't is in a cold war with murica

also the model wasn't trained in mandarin or cantonese

so idek how useful gpt would be over there

[dark geyser]

ChatGPT would answer truthfully questions that your goverment is uncomfortable with you asking.

[main zenith]

en

[robust light]

My cap now shows 50

[acoustic wasp]

yes its going up for everyone this week

[robust light]

Nice!

[acoustic wasp]

I think they're giving us the 16k model

[robust light]

Did some say 100?

[dark geyser]

My wife is from Taiwan and uses ChatGPT in Mandarin, it works fine.

[acoustic wasp]

or at least the 8k