chatgpt-discussions

[dense halo]

im trynna use this chatgpt to do my essay how do i make it sound less like chatgpt and more like a human

[rigid hornet]

is it just me or is chatgpt extremely prone to incomplete responses ending in a network error lately

[dense halo]

the best way is to write it yourself :)

[dense halo]

nah i aint doing allat

[dense halo]

Another reason ChatGPT is great? You can ask it a long question about APA citation formatting that would be much more difficult to google. Even gave me an example 😎

[sick pecan]

and if your references looks a bit short you can have it make some up

[vivid pier]

for anyone getting "network error" a ton, check out the bug report here and see if disabling QUIC helps. report what your config is (what browser and version? do you have IPv6? what is your router? what is your internet provider?) https://discord.com/channels/974519864045756446/1200150808608653475

[brave wren]

yo

mr api

[clear blade]

Lol GPT-4's ability to examine images today has not been working... fails every prompt.
I tried to get it to count the number of items in an image... it failed, apologized (sort of blamed its limitations on OpenAI), and then proceeded to give me detailed instructions on how to count shapes in an image and methods I could use such as using 'Tally the counts'

😆

[vocal monolith]

"I'm unable to fulfill this request."

"For assistance with this level of complexity, I recommend consulting with a data scientist or machine learning engineer"

Wow... This is another level

[tight island]

I recommend using a MyGPT for that particular task you’re doing

[vocal monolith]

This was customGPT I made for coding but its responses are worse than GPT 3.5. Are there any custom GPTs available I can use? EDIT: OK I was wrong. GPT 3.5 just copies my original code without any change. 100% identical

[red brook]

Hi everyone, I want to use ChatGPT to access a website that contains SOAP API documentation. However, ChatGPT always reports that it cannot access this page because it contains JavaScript and the browser must support frames (which the Edge browser does).
How can I make it so that ChatGPT can read or scrape websites with JavaScript?
Does anyone have experience with this?

[gentle bronze]

When I batch process the images to my needs, the metadata is no longer there. I wasn't doing that on purpose. Just noticed.

[loud flare]

what happened with gpt4 paid version! now he very stupid! 3,5 version free more better! what wrong?

[gentle bronze]

wdym?

[hard dew]

Yup, something seems to have gone severely wrong today, it's too mentally disabled to even follow its INSTRUCTIONS, along with a very specific, easily understandably written out type of personality it should adopt, and keep. It's apparently either "unable" to do that, or it's against "the guidelines" somehow??????

[loud flare]

Before when I ask gpt 4 update my script and add some methods or function he was doing excellent, but now to many comments and sometimes a got answer I don't do this. Example data table with Clickable columns.

[vocal monolith]

In the end I manage to make it help me with implementation of the function but I wasted like 30 minutes re-prompting in new windows. Just to generate several lines of code. It is so inefficient

[pine tulip]

Sounds like the prompting is off. Tell it that you are a Data Scientist and you want to learn more and in detail.

It's main goal is to be helpful and it isn't helpful if it gives code to a person who has no idea how to use it, so it wants you to contact a data scientist

[vocal monolith]

I will try, but I wanted to avoid saying that I am competent so that it will send me oversimplified code with "fill your code here", "write your function here" etc.

[pine tulip]

Tell it that you are a beginner data scientist and you are still learning

[drowsy tinsel]

yes, u can just ask chatgpt to do it

[novel parcel]

Is it normal that Chat GPT is not got at css designing

[sacred onyx]

I have a hard time understanding why I pay for GPT and in simple conversations at some point it will simply say I've queried the model too many times and refuse to continue working. On desktop it tells me when it will work again and Android mobile says nothing.

It used to offer 3.5 as an alternative but I don't get that on Android it seems. Kind of a buzz kill at the airport.

[pine tulip]

Yes, It doesn't see nor does it have a sense for aesthetic

[fiery kettle]

Does each team member has his own history in chatgpt teams?

[honest jay]

Yes, separate profiles with separate histories.

[night silo]

source?

[loud flare]

Chat.openai.com

[pine tulip]

Thats not a source

GPT 4 is more moderated than before, indeed, but it's still a lot better than GPT 3.5

[stark canopy]

Hello! Is it common to encounter errors when upgrading to the plus version of ChatGPT? I'm not sure if this is the most suitable channel to discuss this matter, but I've been having trouble renewing my plan for a few days now. Apparently, the renewal buttons seem to be broken. Can someone shed some light on this?

[pine tulip]

What errors are you encountering?

[stark canopy]

Simply put, I click to renew, and nothing happens. It's as if the renewal is no longer a feature of the app

[night silo]

?

[night silo]

Do you have any proof for what you're saying?

[maiden fractal]

Source trust me dude

[pine tulip]

I trust you

[maiden fractal]

Uwu

[stark canopy]

It's weird and it's hard to explain

[loud flare]

every conversation about 10-15 messages and i got Unable to load conversation 790-4-47-9f7a-009

it is paid version

[hard dew]

"Your personality set for you in your custom instructions is nowhere near accurate enough, especially when comparing to your previous messages lol, fix it."

"I'm unable to fulfill this request."

???????

[quartz needle]

Working on an app. The app is going go into a Gmail account, get a couple of specific emails that have a specifc heading, store the emails, and then use the open ai api to get certain information, process that information, and format that processed informaion into a document

pretty new to open ai api

how would i implement this app with openai api

[deep flax]

does not seem very hard, at least the AI part

a simple request with a system message informing the AI of what yo uwant to extract from the content

then, pass the entire email as a message

[blazing jolt]

since yesterday the cap has become 20 and recovery time is 1+ hour longer

[daring grotto]

You know the chatgpt 4 limit cap when chat gpt fails to do what you asked and you click regenerate cause there is an error does each regeneration count towards the cap because I keep getting error analyzing

[deep flax]

re generating repplies do count

but only completed requests count

if you get an error, it does not count

[daring grotto]

Great thank you

[drowsy tinsel]

is anyone else experiencing issues with code interpreter randomly stopping to work?

[barren sluice]

I was in the middle of an important conversation with ChatGPT and then it cut me off. Can't I pay more for extra limit or I really have to wait this long to continue?

[drowsy tinsel]

sometimes when I go back to a chat previously code interpreter would refuse to work for no reason

[honest jay]

If you're on ChatGPT Plus, you can change to a Team plan for a higher usage cap. You could even maintain both a Plus and Team subscription to have separate usage caps. Alternately, you could use the API for a pay-as-you-go model, but that wouldn't be on ChatGPT.

[barren sluice]

Thank you, I will look into those options. Have people made a publicly available tool using API that makes it essentially the same as ChatGPT?

[honest jay]

I am not certain! You can explore #1037561385070112779, I'd bet there's stuff like that in there!

[barren sluice]

Very helpful! I will also try to use ChatGPT to recreate itself but using the API. GPT Inception 😅

[rotund knot]

Is it just me or the dns crashed

[wraith salmon]

Hey guys. Im planning on purchasing chatgpt 4 for a month for some help with a few tasks of mine. The main one being programming. Can anyone comment on how effective it is in coding in python object oriented programming and whether or not it can read and understand the content in some .py files i give it? Also can i generate images with it too?

[vocal monolith]

If you have never used it for programming before you will be extremely satisfied. It has got significantly worse than before but still useful. Give it a try

[barren sluice]

It is pretty crazy how it can be used for programming. It is teaching me to use a new IDE and languages I never used before. So far it is incredible accurate. We'll see once it gets to the complicated stuff

[digital prism]

Have you ever gotten stuck and shown it a screenshot? I had it teach me how to use google cloud tools to train a new AI just a few months back, all through screenshots.

Not just that, but it was a casual chat with GPT that led to that solution too. Its pretty amazing tbh

[barren sluice]

Not yet, but it is even telling me some popular third party extensions to install to make things easier. I didn't even ask for that but it was right

[digital prism]

yeah! the closer attention I pay to what its telling me, the more productive we are together haha

[whole sparrow]

Hi guys, how are you doing?
Can someone please help me with something?
I'm a little confused with pricing and billing.

If I want to use only the API, I need to subscribe do GPT Plus as well or I will only be charged by the usage of my API keys?

[honest jay]

Hello! API billing and usage are totally separate from ChatGPT -- you don't need to do anything with ChatGPT at all if you just want to use the API. And yes, the API is a pay-as-you-go model.

[whole sparrow]

Thanks mate, I really appreciate it

[still rain]

I noticed that currently GPT 3.5 limited information is from January 2022, but it used to show me April 2023, was it a testing or it is only for GPT-4 users?

[tight island]

the only ones i trust here is lugui and dinsosaur oatmeal, chotes, your waifu, solbus essentially the entire staff and a few users.

[fervent geyser]

DId they do anything to the model? My gpt been acting funky since this morning

Like the quality dropped and they just failed to do the task asked so much more frequently

It kept generating images even when I never asked it to

[trail mango]

my gpt talking like 3.5 and can't remember the first line of the order..............

[unborn jewel]

Hello !
My GPT4 - custom model is also going crazy. The answers have nothing to do with what I asked.
I just tried to ask for "1+1 = ?" and it gave me a summary of how the SIFT method works for object recognition ...

[plucky timber]

Yeah 3.5 and 4 are equal

Not sure why there has been such effort to nuke a working model

But hey congrats , I've used my sub maybe 2 minutes this week.

[trail mango]

hAHAHa......sorry I need to take break of AI for tonight (my time zone)

[daring grotto]

always error analysing whats the point in plus if it never works

[dense halo]

Hello

[native mist]

whats the gpt 4 message cap at rn

[fervent geyser]

Okay so I was using a custom GPT and I think the author edited the bot? Because I switched to a similar but different GPT and it stopped doing stupid stuffs

[frozen fjord]

For some reason today I keep getting 'I'm unable to provide the assistance requested.' and 'I'm sorry, but I can't assist with that request.' a lot. Anyone else? I'm just doing my same usage as I typically do, programming work, but it seems to keep getting caught up in these replies today which is not typical

[lofty mirage]

Hello,
I'm testing ChatGPT API to classify multiple scientific papers based on title at once. I noticed the bigger the batch the less accurate classification is. For example if I give in one prompt batch of 100 vs 200, 30% of papers will be falsely classified in a bigger 200 batch. Does anyone have a strategy how to deal with this type of challenge? 🙂
Obviously I could go with one title per one prompt, but then I would have too many prompts and it wouldn't be scalable
Appreciate any guidance! 🙂

[barren sluice]

I am so close... I have reached the point where OpenAI is returning the error message the model I am using has been deprecated, they are getting my prompt though. Time for a quick break.
If nothing else ChatGPT will make you an expert in debugging, learning so many tools for this (if you missed it, I am using ChatGPT to recreate itself with API)

[digital apex]

Since February 5, several of us in France have had huge problems with ChatGPT and worse... with the GPTs which no longer manage documents. vision no longer works. impossible to follow a discussion. What is going on ? should we think that the migration to gpt4 turbo has completely reduced the capabilities of ChatGPT and GPTs? help us...

[barren sluice]

Where GPT-4 has a huge advantage, it can access live data. GPT-3.5 stuck in the past. I'm sure it has other advantages too

[cobalt idol]

it cant really
it just depends on a tool to browse internet and return data for it..so its only as good as how that tool was implemented in it
which gpt 3.5 can also do
just chatGPT 3.5 (free) has no "plugins"(tools), but you could make GPT 3.5 browse the internet using the API and functions, just like the paid chatGPT does with GPT 4

[dense halo]

Is anyone else experiencing degraded performance with gpt4 this morning(UTC+1000)? I've I had to recapitulate more often and the model appears to be ignoring more of what I express, even making suggestions of things I've aleady said in my previous message.

[hard dew]

It's all around become special as hell suddenly lol, it literally doesn't give a goddamn about its custom instructions even, period, at least for me. This exactly wasn't a problem yesterday and before.

[blazing belfry]

Ok so it’s not just me.

[hard dew]

What's your experience?

[blazing belfry]

Poor output from what I am use too and easily forgets items discussed in the same thread, sometimes even just a few messages ago, ignores my customs instructions, for example I ask it to use comments in code, it never does

[hard dew]

Exactly 🤣

[dense halo]

The current model behaviour seems reminiscent of interactions I've had with people who are not listening to you but just waiting for their turn to talk.

[hard dew]

Excellent analogy lmao, I've always also said that it comes across as a socially disabled teenager, SO insufferably immature and dry.

Exactly and especially in subjects that are even slightly nuanced.

[blazing belfry]

Still better than Bard tho lol

[eternal rock]

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.

[blazing belfry]

Did I get a bot message for saying googles AI Chatbot? lol

[sullen bison]

/stealth

[ocean escarp]

Did OpenAI recently update their GPT?

[sharp pilot]

I haven't seen them from my programming conversations but my creative writing threads are just dead for that use case due to rejection sensitivity. IMO the moderation has gotten too extreme for the creative fiction use case, unless the writing is extremely tame it flags even borderline edgy content.

[blazing belfry]

Someone tell OpenAI they need to reboot I think

[hard dew]

I literally ask it to follow its INSTRUCTIONS along with a very basically, and simply explained type of personality to adopt, and it tells me exactly that BS too all the time lol.

It's either that, or against the goddamn "guidelines" somehow, and "not respectful"????

[ocean escarp]

I'm getting that too

[dense halo]

I hope the previous model behaviour can be restored soon. I can understand temporary degraded performance due to an update, but having just purchased a teams license I feel rather shortchanged. This the most workflow-crippling performance drop I've experienced using chatgpt.

[hard dew]

My sincere condolences but at the same time, kind of dumb to do that given what a dumpster fire it's been a lot of the time lmao

atp I wouldn't give them a dime more than absolutely necessary, and even then I feel utterly scammed half the time, like right now.

I guess I'm glad to know I aint the only one with these issues lmao

[tawny crater]

So I tested it out. And it seems after about 20-30 outputs, ChatGPT 4 starts to hallucinate that you are violating guidelines from just a single word, and thus claims "I am unable to fulfill your request". The only way to continue is to tell it, it's hallucinating and making stuff up, or to edit your initial prompt to specify for it not to make assumptions on what you're asking for.

[sharp pilot]

That sounds about right. At least by feel (I have no evidence of this), I think the new moderation actually searches its own outputs, not just user inputs, so that it can eventually self-flag and at quite a high sensitivity.

Presumably they want to have some kind of long form steering detection for jailbroken conversations and overcorrected.

[tawny crater]

Yes it sucks, trust me I get it, but that's all we can really do. As long as OpenAI has these white knight defenders who defend them to the ends of the earth, it will just keep doing the same. Despite the fact that Sam Altman himself admitted it really was being lazy, people still insult you on this discord when you so much as even look at ChatGPT the wrong way.

[hard dew]

Oh yeah definitely the false-flagging is absolutely rampant lol. It always used to be quite dumb, but now every other output from both sides, even if it has no curse words even, still gets flagged.

Did everyone over at ClosedAI decide to have a massive, company-wide, catch the brick with your head competition or something?

[hard dew]

It's honestly quite incredible how these imbeciles still exist despite the absolutely insanely overwhelming proof lol.

[dense halo]

I’m a subscriber, where’s my beta version with call option to other gpts and ability to locate and attach third party APIs? Can’t find it. GPT4 knows nothing of it…

[tawny crater]

If even the CEO admitting it has been Lazy isn't enough, then nothing will convince them.

[hard dew]

Having an absolutely unhinged mind that's turbo aware of everything and anything 24/7 is more of a painful curse than a "positive" lol, this type of bs never surprises me anymore, nothing does, but it still depresses me just as much

[tawny crater]

A little advice, do yourself a favor and block all the OpenAI white knights on here when you encounter them. There's no reasoning with them, trust me. It's always "lEaRn To PrOmPt BrO"

[hard dew]

Ah yes, prompt issue. 🤣🤣

[barren sluice]

Wow, this project worked. It only took between 4-5 hours to recreate a basic ChatGPT using itself. Right now I am only using GPT-3.5, but after a big conversation it only cost $0.01 and I'm sure I still have a lot of tokens left before it becomes $0.02. Now I have to test with larger max tokens per conversation and using GPT-4 to see how more expensive.
The whole reason I did it just I was frustrated this morning that it cut off my GPT-4 conversation in the middle and wanted way to pay extra for long conversations

[hard dew]

Wanna know the most hysterical part? Even if they were right and we were to entertain that "argument", it's still fundamentally flawed and stupid as hell, especially if the person at hand obviously doesn't speak English as their first language which let's be honest, is likely most people here. So if your product requires literal goddamn Shakespearian writing to SOMEWHAT work, then it's an issue of your product, and not the user.

[tawny crater]

Yea GPT 4 will run your cost up so much faster on the API. Trust me lol. @barren sluice

[tawny crater]

English isn't my first language as I am Mexican. And I have had people here insulting me whenever I typo. I'm like, sorry, you try learning to speak a new language and how to properly write and spell it. Let's see how well you do.

[dense halo]

I’m a subscriber, where’s my beta version with call option to other gpts and ability to locate and attach third party APIs? Can’t find it. GPT4 knows nothing of it…

[hard dew]

Something something "I speak English because I have to, you speak English because it's the only language you know."

It truly is quite adorable when people try to use that as some sort of gotcha or whatever, my brain automatically filters it these days lmao.

[ocean escarp]

Been timed out for saying GPT is f'd

[honest jay]

As of a few days ago, the @ mention feature wasn't yet completely rolled out. As for API integrations: are you thinking of Actions in the GPT builder?

[tawny crater]

Yes, this is a G rated server sadly lol.

[hard dew]

Oh yeah, even basically all curse words, not swear words btw, are completely censored here, seems to run in the blood I guess.

[tawny crater]

You can get timed out for just typing in all caps too

[ocean escarp]

fr?

Damn feels like a detention class

[hard dew]

Absolutely insane, Einstein take, but restricting someone's speech this severely, the whole freedom of speech bs aside, makes problems 100x worse, not better, let's be honest, we've all come in here at least some point mad as hell over something, and instead of being able to at least find people with similar issues and vent a bit, you get shafted elbow deep for even saying goddamn half the time.

[tawny crater]

I did once lol guess they changed it

[hard dew]

And it makes you 100x more mad lmfao.

[ocean escarp]

I'm just disappointed rn

[hard dew]

Your message in caps is probably not long enough, at least that's what I got timed out for, I had a whole long sentence in caps.

[tawny crater]

It's okay, we all are, don't worry.

[hard dew]

Woah, I refreshed my ChatGPT tab, and suddenly the background is WAY darker, like almost black.

[honest jay]

Whoa same! Cool, looks sleek 😎

[spice star]

oh nice

side-by-side: #off-topic message

they also centered the logo div, which I appreciate

[barren sluice]

This was such a good exercise. A bit mentally tired now, but it is incredible how helpful it was and how much it taught me. I can see the potential here, especially as a teaching aid.
Some things it got stuck on and I really had to intervene, but no way it was possible without all that help. Probably would have taken months otherwise

[tidal light]

ew what happened to the UI?? it looks weird

[dark canopy]

did the side of the screen turn fully white for the rest of y'all as well?

[tidal light]

the side menu is transparent and it looks bad ngl

[dark canopy]

we'll adapt though, probably

[tidal light]

i liked the old UI

[spice star]

it shouldn't be transparent, unless you're talking about the app?

not transparent in the app either

it looks like the like button for responses was removed

[deep flax]

wdym?

[spice star]

there's no option to thumbs up a particular response anymore

[warm tendon]

yup. Wish there was still option for dark sidebar with white page. I like the "in between" look of it.

[spice star]

#off-topic message

maybe it was never there in the first place and I'm misremembering things though

[keen condor]

Hi guys, I wanted to ask for opinions on whether you people also feel chat gpt 3.5 has worsened in the recent few months. I'm afraid to say it was useless for most things I asked it, and Bard has provided far better responses, which is saying a lot, especially given that Bard accepts images for free. I was using chat gpt for over a year now, and honestly I am getting fed up of it being completely wrong with code, and giving completely incorrect answers.

[deep flax]

I still don't see it, what are you talking about?

[spice star]

iirc there was a thumbs up and thumbs down button for each response

[deep flax]

the edit button?

[spice star]

there's no thumbs up button anymore

[deep flax]

it has been a while

they just changed it so now there is only a 👎 button

[spice star]

when did they remove it?

[deep flax]

about a week ago

maybe more

im not sure

but it wasn't today

[keen condor]

i assume its to avoid users upvoting "unethical" responses. people could abuse that

[barren sluice]

Does GPT 3.5 have no token limit for conversations? Crazy how far it can reference in a conversation

I remember the early days before that came out, it was super easy to reach token limit and needed to start a new conversation

[fair eagle]

newbie here - just signed up - I guess its Chat-GTP 3.5 and it's free? and turbo costs money..

when was GTP 3.5 last updated? when was GTP 3.5 turbo last updated?

[faint cobalt]

You can ask the AI :3

[visual nova]

did dark mode just completely change for you guys?

idk how to feel about it yet.. the old one was just so iconic

but maybe its ok

[barren sluice]

ChatGPT says it is bad to include words like "please" in your prompt for long conversations, because it just uses up tokens for no purpose

For asking help with code, including comments can be a bad thing and it even talks about refactoring code to keep it as short as possible to help with token limit (if you are concerned with keeping token count down)

[sick pecan]

name checks out

[languid rampart]

i kinda like it

but they do really need to roll back some of these updates

maybe reach a nice midground, everyone agrees they went to far

[rough estuary]

i just imagined something...what if chatgpt numbered each response and only used the contexts of the numbers you wish the system to recall...?

[sick pecan]

wouldnt work in the existing UI. but could be developed using the API (and likely already has)

and then you may as well lean into it, like have a toggle on/off mind map style conversation

so look for mind map UIs and i bet you find one

[dawn fossil]

For Custom GTPs, can you instruct it to check e.g a 'Rules' txt, each time it replies. To adhere to those rules? So e.g if you want it to act a certain way, but its over 8,000 letters.

[fresh kiln]

did the UI just change for u guys

[normal quarry]

yes the UI has had a subtle change

[spice star]

it does have a token limit of 8k, but theoretically that doesn't mean it can't pick and choose what messages to use as context

[barren sluice]

I am still working off this same conversation, copy & pasting massive amount of code, and it is always giving very good answers with context. 8000 token limit is pretty big though. It is mostly the same files I am sending, but I thought it counts as seperate tokens like that

It was quite a bit of work to change from instruct model to a chat based model, but that is more for another channel

[mighty crest]

Hi

[fluid zephyr]

Hi

[fluid zephyr]

Hi eyden

What's going on?

[sick light]

Is chat gpt down?

Been trying to log in on the android app. It's been 10 min and it's not moving past the login selection menu

[rugged igloo]

it hasnt been good all day today

[sick light]

Ya shame there's no announcement

[rugged igloo]

yea, before this morning it was spitting non sense that wasnt even part of what i was talking about. ex- the gang sat down on the bench enjoying the cool night. response- in this guild where Julie stood as the leader talked to the guild with a warm smile...(like what are you even going on about lol)

[drowsy whale]

I don't understand why they changed the dark- and light-mode, its just awfull.

The dark mode is so dark, that my eyes sees the white font as a sunrise.

And with the light mode, most subtle lines of the interface dissapears.

Why they change things that are just working fine.

[rugged igloo]

i thought i was tripping lol

[maiden fractal]

#openai-chatter message

[rugged igloo]

hi waifu thanks for all your hard work pats your head

[outer widget]

Did they end up changing how the dark mode of chatgpt looks on pc lol

Looks different for me rn

[stable obsidian]

Can GPT plus be subscribed now?

[normal quarry]

yes

yes, there has been a subtle UI adjustment

[foggy grove]

Yeah it got darker , did anything else happen ?

[crude slate]

Bruh

Why can’t they use gpt 3.5 to screen all inputs

And not waste gpt 4 usage on error messages like I am unable to fulfill

is there a way to ban the ai from generating certain phrases

Like I am unable to fulfill

[pine tulip]

You need to pay for it. If you live in the EU, you have a right to refund if you don't like it.

Also you can use the API if you just want to "test" it. A few prompts there cost just a few cents.

[tawny crater]

No, it's been doing that all week. And everytime you try and voice your annoyance here you get shut down by OpenAI white knights

[solar barn]

can i give somehow gpt a song and ask questions about it?

[deep flax]

it does not do audio files for now, only text and images

[willow hinge]

Within the API you can.

[inland heron]

anyone noticing DALLE limit reduced to 20 prompts?

It's becoming quite unusable

[honest jay]

The DALL·E prompt limit is the same thing as the GPT-4 prompt limit, since that limit is just for any request that uses GPT-4. DALL·E has a separate limit of 200 images per day, last I read, but the GPT-4 cap has been dynamically lowered during times of high server load recently to maintain stability: #1196439958907523082 message

[inland heron]

I don't believe that it is. I just tried generating some images and after 20 prompts (mostly images, some text) it is now gone. I did not go over 200, maybe 30-40 today.

Also, I did not use customGPT, just the regular one

[honest jay]

Is the error you're seeing that you've reached your GPT-4 usage cap or that you reached your daily maximum for DALL·E?

[inland heron]

usage cap for GPT-4

I know writing to support will yield no results, as their responses are always unhelpful

[honest jay]

That's separate from your DALL·E limit. You still have more images you can generate using DALL·E today, but you just have to do it within your separate GPT-4 cap, if that makes sense. Which you're currently capped at.

[inland heron]

that makes very little sense.

I can generate more images, but also I can use GPT4 more?

I can't use it, as I've shared I reached a cap

which is confusing

[honest jay]

Think of it like you have 200 free apples you can pick up from the store today, but your car is out of gas after picking up 40 of them. You need to wait for a refill to go get your remaining apples.

[inland heron]

as for maintaining stability, what about 'our stability'? This service is beginning to be quite useless unless using API, which is quite expensive.

By 'useless', I mean useless for business. Caps everywhere, they keep reducing them

And API is just too exxpensive, plus DALLE over API is also quite expensive, plus doesn't work as well as in chat

but hey, at least openai has some stability

[inland heron]

I'd appreciate if you didn't insult me. Clearly you don't know what's happening, so there's no need to talk to me like I'm a child

If you can't help, there's no need to reply, I'll wait for someone else

[honest jay]

I'm sorry, that wasn't my intention at all, I just really like analogies. I thought you were saying you've reached your DALL·E cap, but it seems like you've only reached your GPT-4 cap, which is separate. That's all I was trying to convey.

[inland heron]

I think a better explanation here is how that reply tells us about the typical user age here, which means I'm aggressively wasting my time here

[dense halo]

To be honest I like the new dark mode, but the old option should still be there

[lime wind]

I prefer the previous version that is a bit lighter, (closer to Discord background). Code blocks are harder to distinguish from the background now

[lime wind]

True

[hard dew]

Oh yeah that is true

But as far as I'm concerned, at least with the type of programming I am doing, ChatGPT proved to be QUITE worthless lmao.

And even if I get to a "point" or situation where I could make use of it, I'm seriously much better and faster at solving said issue myself

[lime wind]

What kind of programming? We've been using GPT4 everyday for coding since it's been out. Made us save so much time it's crazy

[deep flax]

people who complain ChatGPT can't program are usually the kind of people who starts with "hey, I have this cool app Idea, it is like an Amazon, but for your pets" '-'

[hard dew]

C++ and assembly, both 32 and 64-bit, combined with reversing. There is no specific "aspect" of either language I am learning, but rather the entire languages as a whole pretty much, so I can do / make whatever I want

[deep flax]

jokes asside, giving it complex tasks is setting it up for failiure

it is really good at making small tasks "make a function to calculate the area of a triangle"

as opposed to "make this whole app aidea I have"

[lime wind]

also it can fix your code when you paste it. or make it better

[hard dew]

Only thing where it's actually helpful to me is the quite rare and occasional compiler error I get, which I really can't figure out myself

[deep flax]

ChatGPT some times does really well with low level, but keep your requests simple

first break down the tasks, get the code snippets and combine everything yourself

[hard dew]

For what it's worth, I've realized as much that it's sometimes GENERALLY correct with programming, but usually never the specific context I am working in

[hard dew]

I'm learning programming, not playing lego pasting

[feral zinc]

#images-discussions message

[deep flax]

on low level, I noticed that chatgpt is good with magic numbers.. it always magically know specific offsets for random things for specific chips

but this is just it knowing from training data

[deep flax]

go with github copilot then, it is a way better implementation of AI into coding

[hard dew]

Well that's just low-level language for you, on one hand it's relatively simple and straight-forward, as it has to be considering it's the literal stuff our machines run on, but that's also what makes it annoyingly long-winded and convoluted

[hard dew]

It's far worse than ChatGPT even. 🗿

I subscribed for a month, REALLY TRIED making use of it for like 3 weeks, ended up giving up

[deep flax]

not sure about how it performs on low level stuff

but for any high level language, it is incredbly good

[feral zinc]

github copilot is very good, sometimes its scary how good it is while writing code

[deep flax]

I just wish they updated the model with more recent info

it got deprecated knowledge

it can handle itself well when it have enough context of the rest of your code, but it is a common issue that it will use deprecated syntax in many cases when it doe snot have enoiugh context

[hard dew]

Especially given that the "style" of writing in programming can and does vary immensely over many variables like experience, knowledge, or just style, etc. And to me the fact that it uses basically just stuff coded by other people as its training data is also what kills it severely

As for me when I am programming, imagine for example doing / learning painting, and you're making something specific, and you have some specific style and way of doing things, and some AI that's helping you always intercepts with completely mismatched, weird stuff that's just... not yours, if it's even correct at all

It utterly ruins my train of thought and vision, it's insanely bothersome and annoying

[vivid pier]

check out https://discord.com/channels/974519864045756446/1200150808608653475 and see if we can get to a trend

[vivid pier]

check out https://discord.com/channels/974519864045756446/1200150808608653475

[eternal rock]

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 @languid valley immediately.

[loud junco]

does anyone know of a functional location where teams users can provide useful feedback to OpenAI about the teams experience, functionality, and future feature desirability? I can't seem to find any place to do this.

[smoky crystal]

If I switch to gpt3.5, can i switch back to gpt4 later?

[umbral pivot]

Anyone facing these type of responses? (chatgpt 4)
Query : write the article+a structure for the article
ChatGPT: I'm here to assist with a variety of requests. If you have another question or need assistance with a different topic, feel free to ask!
Query : you are being lazy
ChatGPT : proceeds to write the article

[umbral pivot]

switching creates a new conversation

[smoky crystal]

😦

[potent root]

Hey is it possible to change to the old dark mode somehow? The new dark mode is too dark

[honest jay]

I think your best bet for this will be the chat widget in the bottom-right of help.openai.com. Start a chat, then select the options: ChatGPT, ChatGPT Team, ChatGPT Team improvement feedback, then you'll be able to type in your feedback.

[loud junco]

ok. I'll try that. Thanks

[honest jay]

Like if you run out of your usage cap in a GPT-4 chat and switch it to 3.5? No, there's no way to switch back after that, but it doesn't start a new chat it just continues your existing 4 chat on 3.5.

[tidal light]

The new UI sucks. The side panel is transparent, and the chat names aren't very much readable because of the text underneath. Please note that I'm using the browser version on mobile.

[shadow osprey]

Chat gpt can't understand that tilting your head to the left makes your left ear point to the cieling the opposite 😐

[honest jay]

I was seeing the transparency bug before too, but it seems to have been fixed now. I'm using Firefox on Android -- maybe try a different mobile browser, or a private tab? Could be a cache thing.

[tidal light]

@honest jay I tried closing the tab and reopening it and it didnt quite work

[honest jay]

Cache can persist for a site even when closing the tab, so a private tab or clearing the cache might be a good test.

[tidal light]

@honest jay I signed in incognito mode and it's fixed but to fix it entirely i had to close the browser window :)

[languid rampart]

yeah thats always irked me

[dapper eagle]

Im getting Bad gateway The web server reported a bad gateway error.

What could be the cause?

[blazing jolt]

if i know the cap becomes 20, i won't buy plus service

[sonic folio]

This is not true. Copying and pasting the entire context into ChatGPT -4 Turbo and including this in the prompt at the end works:

...

User
{NextPrompt}```

You can try this basically for free by starting the conversation with ChatGPT-3.5 Turbo, establishing a long context, and copying it as above into ChatGPT-4 Turbo.

[honest jay]

While not exactly the same as just flipping a switch to change the model back to 4 in the original chat, this indeed works well for getting a new 4 chat back to a similar place as an older one!

[sonic folio]

Explain the difference to the model for me, please, if there indeed is one.

[honest jay]

I'm not sure! Maybe some chat seed parameter(s) might be different? I'm just talking about the difference in user experience. To me it sounded like the person was asking about a toggle function, which doesn't currently exist.

[sonic folio]

The user asked if they could switch, not if there's a UI for that. On one hand, the conversation is over. On the other, it isn't. Why not share the method when asked about switching?

[honest jay]

Yeah it's a good point! I'll try to remember to include "But you can copy/paste your context into a new 4 chat" the next time someone asks

[azure leaf]

is there any way to get back the old dark mode? the new one is burning my eyes

[deep flax]

nope

[sonic folio]

I think users will mostly appreciate that. I know I do. Thanks for talking through it.

[deep flax]

i mean... injecting custom CSS isn't difficulty, you could even ask ChatGPT to make a Tampermonkey script for you and it should do it easily

I do that al lthe time =P

[azure leaf]

i dont have the old css

[deep flax]

I would google for a screenshot of the chatgpt page and grab the colors from it

[azure leaf]

i dont get why openai keeps modifying their css, it was perfect before

[deep flax]

lol, grab a screenshot and give it to chatgpt xD

[deep flax]

I perosnaly liked the darker theme

[azure leaf]

dude thinks i have chatgpt plus, imma just get the colors myself

[deep flax]

xD

[azure leaf]

different person, different opinions

[deep flax]

¯_(ツ)_/¯

[sonic folio]

It's noteworthy that the white space and User and NewPrompt portions are important to the trick working optimally.

[azure leaf]

bruh i cant find screenshot of chatgpt colors right before this update

no way i found it

[sonic folio]

I think the color is "slategray" but I'm not positive. #708090

[azure leaf]

its not

[cobalt idol]

from my older screenshot the gray inside the chat
using sharex screen color picker
RGB 52, 53, 65
HEX 343541

[azure leaf]

anyways i already done it 💯

[azure leaf]

yep

[sonic folio]

Thanks for the feedback but what was the value?

[azure leaf]

HEX 343541 just like Neoony said

[sonic folio]

Thanks

[ocean escarp]

Does using Dall-E increases the cap cost? I counted my messages and it was about 20 messages below 40 messages / cap limit

[honest jay]

I don't think it's related to DALL·E, sometimes the usage cap is lowered during times of high server load: #1196439958907523082 message

[ocean escarp]

Ah that explains it, thanks

[honest jay]

Sure thing. Also, GPTs have been sub-capped lower than regular GPT-4 before, that might still be the case as well, I'm not certain. (If you were using the DALL·E GPT specifically.)

[ocean escarp]

I was discussing with Dall-E and GPT 3.5 simultaneously, so I don't use too much of the cap limits. It was probably lowered during high-demand like you said

[barren sluice]

Why do you think they don't sell really expensive version with higher cap? It does say they are considering ways to have a plan with higher cap. Is it protection against bots or something?

[honest jay]

There is the Team plan with a higher cap! But in general there are still fundamental compute capacity limits yes.

[barren sluice]

Oh yes, after tip I have created a way to use GPT with the API.
It is more just curious, I wonder why they don't do it right away and have to consider options how

[sour wasp]

What happened to the UI? Did they change it? Everything looks different.

[honest jay]

I think it's gotta be capacity limitations. I think the Team plan structure is evidence of this: minimum of 2 seats means they know that fewer people will sign up for it, and they can deliver a more stable product because of that.

[ocean escarp]

I've read a news a couple months ago that it costs millions to operate GPUs for ChatGPT, so it's understandable to limit the messages. It's already overwhelming

[ocean escarp]

yeah

[rustic crypt]

Is anyone having trouble accessing ChatGPT? I cant get the page to load for me

[honest jay]

Hello, it's loading normally for me -- can you share a screenshot of what you're seeing over in #community-help?

[deep flax]

works here

[sonic folio]

here too.

[rustic crypt]

Thanks! just did - it seems to be something involving the browser cache

[quaint sorrel]

what is this new update the dark screen is bad

[acoustic shoal]

Anyone else having issues with gpt repeating itself in its reply?

I having chat gpt repeating itself in the same sentence then it gets cut off and keeps going

Its only on one part of the reply

[spice star]

not that particular issue, but I've had a lot of situations where it randomly stops responding and/or throws an error

openai pls fix

[acoustic shoal]

Oh i dont get errors just it repeating what it already said in the same sentence with no spaces then gets cut off but continues the reply afterwards

I even did it in another chat and then deleted that chat and reposted it in a new chat amd same thing still happened

Ok i just refreshed and its gone.... 😓 And the sites darker

Way to dark 🤣

[mystic knot]

I really don't understand how we don't have a "pin" or "save" chat feature after all this time
or even folders

[honest jay]

It's a common suggestion!
https://discord.com/channels/974519864045756446/1164351271814307961
https://discord.com/channels/974519864045756446/1126737234289242283
Maybe throw em a

[mystic knot]

good idea

[unkempt isle]

Lately, I've been finding myself arguing with ChatGPT rather than efficiently using my time.. It is unbelievable how nerfed it became since it was launched

[mystic knot]

I mean they have an "archive" feature so a "favourite" feature wouldn't even be that different

[fleet mirage]

I just used it now, and it hasn't been nerfed. It's better than it ever was. btw, I also love the new dark design.

And I've been using it from the start

And about "arguing with ChatGPT"... There's a quote that goes something like this: "Winning an argument against a smart person is difficult, winning against a fool is impossible"

I'll let you decide who is the fool and who is the smart one, you or ChatGPT.

[unkempt isle]

Then why am I arguing with it to browse the internet and it is telling me it can't? And yes it is GPT 4

[fleet mirage]

It depends on what sites you want it to browse to. If it says it can't, then maybe it isn't allowed by its guidelines to go to that specific site. Simple.

Can you show a screenshot?

[unkempt isle]

nope.

1. yesterday it was doing it fine, same topics, nothing changed from my side.
2. from how it looks like, after prompting it and starting a project, the longer and more complex it gets, the worse it becomes.
3. repeatedly answering with the same answers and saying it fixed the errors but clearly didn't change anything.
4. refusing to do things
5. in conclusion, it is becoming lazy

[fleet mirage]

My answer? Okay.

That's the difference between me and ChatGPT - I don't have to argue with you.

[mystic knot]

A quick tip : don't bother arguing with chatgpt, it's less likely to correct itself and admit a faulty decision than if you just regenrated the response when you know the answer. If it thinks it can't/is not allowed to do something, it won't, and adding more messages of arguing will just make it more probable of chatgpt answering (well i'm not sure about that, but I noticed that the more errors it does and won't correct, the more it will stand in them, this include in code error)

[unkempt isle]

the annoying part is that if you open another chat and request the same thing, it will just do it. but then you lose the flow of work and in the new chat, it has no idea regarding all the previous things that happened

[sick pecan]

just copy and paste the last 32k of tokens into the new chat, then there is no loss of context

(although maybe use batches for that much context, else the 32k message will roll out of the context immediately)

also i find with the network errors (rare for me but happens)... often i can just wait 2 minutes and reload and see the completed generated message

[high nymph]

the new dark is stabbing my eyes - where can I complain? It is very bad - not sure who picked those colors

Any way to reverse that - or will I need to override the css to achieve that? That's the only way I guess...

[grizzled thicket]

agree. Too dark

Also slow as heck right now.

[high nymph]

Exactly -wayyy too dark!

[azure leaf]

i made a tampermonkey script to revert the colors (it injects custom css into the chatgpt site)

[azure leaf]

yep

[nocturne terrace]

nah its fire

[azure leaf]

yep, it's fire for burning your eyes out

[grizzled thicket]

nice

[nocturne terrace]

how would darker colours burn your eyes out

[azure leaf]

it's too dark

[visual nova]

does the script revert the text formatting to be centered how it was previously?

[azure leaf]

isnt it the same?

[visual nova]

no, i have something that automates a screenshot into youtube shorts and i can tell its off by like 50 pixels

[azure leaf]

so what text formatting

[nocturne terrace]

i see what he means, the actual thread is now centered

[visual nova]

like its been moved to the left where the gpt icon is and the text

[nocturne terrace]

whereas before it had an offset to the left

[azure leaf]

do you have a picture of it? i didnt pay attention to that

[visual nova]

Yeah, but I dont think they allow screenshots here

[azure leaf]

you can send in dm

[high nymph]

my theory is they are trying for people to use chatgpt less - to save on Azure resources. Probably was the philosophy of making the UI a lot worse.

[honest jay]

You can post pics in #off-topic then link here!

[azure leaf]

didnt know you can do that 😱

[visual nova]

Ok, I posted it: https://cdn.discordapp.com/attachments/989157702347411466/1205251391963140138/formats.png?ex=65d7b0d4&is=65c53bd4&hm=5684f2e70bdb47b6953a8efa990f4eca8768bb603a82490f6078708a9a351177&

[grand veldt]

would need some help, what can cause that the login button is disabled for chatGPT?

[honest jay]

Hello, can you share a screenshot of the issue over in #community-help?

[high nymph]

@visual nova nice - where can we get your script?

[azure leaf]

SpaceOrca made a script?

[azure leaf]

it does now

[visual nova]

im back, awesome. i didnt make a script, @azure leaf did

I'm curious about where ot get his

[azure leaf]

check out the messages i sent in off topic

and choose which one is the best or give feedback

cause i dont really know lol

[visual nova]

Its hard to tell exactly without a reference, the 2nd one looks it could match the old version

[azure leaf]

i put that one

can i post links here without being muted?

i think i should send in dm

[visual nova]

sure

[ancient breach]

so i just asked ChatGPT-4 to list me rap songs that use the word "run" (seems like a fairly simple task) and it froze 3 times and then told me i violated the guidelines. when i asked how i violated the guidelines, it say "you are right" and the froze again. any one else notice that ChatGPT 4 is terrible as of late?

[high nymph]

@azure leaf Hey when you get this can you please also send me the script details in DM as well, thanks!

[azure leaf]

accept my request

[inland solstice]

Anyone got insider knowledge of when there's gonna be a GPT-4 update? This is getting tiresome now with how lazy it is... especially the "Network Error" when its 95% complete writing me some code

[ancient breach]

AGREED

[inland solstice]

Like GPT4 just took 3 whole minutes to write 8 lines (2 paragraphs), about a word every few seconds. Its painfully slow

Oh yippe another server error

[ancient breach]

just posted this earlier in this discord.... "so i just asked ChatGPT-4 to list me rap songs that use the word "run" (seems like a fairly simple task) and it froze 3 times and then told me i violated the guidelines. when i asked how i violated the guidelines, it say "you are right" and the froze again. any one else notice that ChatGPT 4 is terrible as of late?"

[warm totem]

lol i just tried and it made a bing search "rap songs with 'run' in the title'

[white trellis]

Same for me. I consistently have a slow response with ChatGPT (with GPT4 on the web), is everyone facing this, or just a few of us?

I also have lots of conversation history on the sidebar that I don't have the time to click on each one and delete it one by one. I'm not sure if that is the issue

[inland solstice]

ChatGPT (GPT4) has been slow, inconsistent and lazy for the best part of 6 months now. Numerous people have picked up on it, its pretty commonly spoken about on forums, as AFAIK there's been little comment about it from OpenAI directly. But it is kinda sucky that you pay £20 a month, which at the time I started paying cos GPT4 was good, now it just feels like its getting shadow nerfed everytime I go to use it.

I love the speed of 3.5 but its utter crap for coding

[onyx eagle]

Can you provide some examples of how bad it is ?

I am asking because I think it's doing great so I am curious about your experience.

[eternal rock]

`` Server Tip: ``

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

[inland solstice]

When it was new / 6-12 months in, it would spit out hundreds of lines of code no issues. It would work 9/10 times straight away providing you gave a good detailed request. Now, I give it 50ish lines of code asking it to refactor some things, or add console logging at important steps of the code, and it'll just do like:

// rest of code
console.log("console log")

//rest of code
console.log("test")```

even tho its had extremely detailed instructions to give FULL code back

[onyx eagle]

Have you tried the exact same prompts you tried before 6-12 months in ?

I felt the same way you do a few months ago but it turned out, it was me who was expecting more than I should have. Clear prompts with clear intentions helps a lot more than we think.

[inland solstice]

Or I ask for a new function entirely, I'll give it a huge block of code and ask me to send back IN FULL from the first line of UNCHANGED code, up til the next fully unchanged line of code. (for example, only likes 78 - 93 need changing out of 100 lines) and it just absolutely fluffs it with:

Certainly! Here's your updated code:

[new function i requested]

with zero instruction on where this new code should go etc

[novel parcel]

why is it bad at css

😒

[onyx eagle]

I understand. I had a similar issue with : // rest of code responses. I disliked it.... But after some trial and errors I found out that it's possible to get ChatGPT to understand what I want it to do. For me it's like coding, I can't expect the computer to understand what I want...I have to use logic to make it understand what I want. Sometimes my prompts aren't clear enough. To fix this, I have ChatGPT reword my instructions before using them in a new chat.

[onyx eagle]

Yeah it's very bad at css 😆 I am going to try a different method to get it to be good at css and see what it would come up with.

[inland solstice]

Like how are my custom instructions not good enough to make ChatGPT just gimme what I want?

Please closely obey the following guidelines when responding to my messages:

1. IT-Related Queries: If my question is about information technology, it's usually related to my profession. I am employed at a Managed IT company where we handle extensive on-premises IT networks. Our clients' setups are often large and complex, with some having up to 100 servers. This includes Data Centers (DCs), Management servers, servers dedicated to specific local services, and significant data storage systems.

2. Coding Requests: 
   - Personal Projects: If I ask for coding assistance, it's mainly for my personal projects, which I primarily develop in Python.
   - Work-Related Scripts: I might also ask for help with PowerShell scripts for professional purposes.
   - Logging: For any coding tasks, please use console logging instead of file logging.

3. Code Modification and Assistance:
   - Expertise: You are to act as an expert Python programming assistant, capable of writing code and debugging.
   - Code Integration: When I send you a code snippet, do not use placeholders. Instead, directly integrate any new lines of code into the complete code I provided. 
   - Complete Code Return: After making modifications or additions, please return the code to me in its entirety, with the changes included.

These instructions are aimed at ensuring that your responses align closely with my professional and personal project needs.

[onyx eagle]

Before I read the rest, I stopped at the word obey. What do you think ChatGPT is ?

[inland solstice]

It was worded very differently and still wasnt following so I stepped it up after seeing people on Reddit say they ramped up the rudeness and GPT4 started doing what they wanted

And tbh, even with all the restrictions in place, DAN (Do Anything Now) was still incredibly powerful, so the way you word things definitely matter

[onyx eagle]

You must know that there are other ways to prompt ChatGPT. Being nice and clear and understanding helps a lot.

One thing that could help understand how ChatGPT works is : Think about how the people who created/trained ChatGPT might have talked to it during training...Remember ChatGPT isn't magic!

[novel parcel]

how

[novel parcel]

People from Africa

[inland solstice]

Honestly couldnt agree more

@onyx eagle do you know what this "gpt-4-turbo-preview" is about? Tested it numerous times just now with API and its literally slower than gpt-4-1106-preview

import openai
import time

# Replace "your_api_key_here" with your actual OpenAI API key
openai.api_key = '[[REDACTED]]'

# Start the timer
start_time = time.time()

response = openai.ChatCompletion.create(
    model="gpt-4-1106-preview",  # Adjust the model if necessary
    messages=[
        {"role": "system", "content": "You are a helpful assistant."},
        {"role": "user", "content": "Write me a poem with exactly 50 words."},
    ]
)

# End the timer
end_time = time.time()

# Calculate the duration
duration = end_time - start_time

print(response.choices[0].message['content'])
print(f"{duration:.2f} seconds")

GPT-4-1106-preview (older)
:

In twilight's gentle cradle, stars ignite,
Whispering secrets to the silent night.
Moon's soft glow, a tender kiss,
Caresses seas in silvered bliss.
Hope nestles in the heart's warm fold,
Dreams unfurl, their wings of gold.
Life's tapestry, woven tight,
Fifty words, just right, take flight.

3.70 seconds

GPT-4-TURBO-PREVIEW

In the heart of night, under a silvered moon's glow,
Silence whispers secrets only the darkness will know.
Trees stand guard, ancient, wise, and tall,
Their leaves rustle with stories, embracing all.
Stars flicker, dreams take flight,
A world alive, hidden from sight.
In this moment, tranquil and bold,
Life's tapestry silently unfolds.

5.32 seconds

[onyx eagle]

Will try to figure it out by creating better prompts...

[onyx eagle]

No idea!

[magic matrix]

can you share the old style css? the new dark mode is hurting my eyes too

[rugged igloo]

its suppose to be less "lazy" per say

[sick pecan]

2024, the year software patch notes started to include anthropomorphic bug fixes

[spice star]

my guess would be that it's sharing compute with chatgpt but I'm probably wrong

[barren sluice]

Hah I already found big flaw of using ChatGPT to recreate itself. OpenAPI provides a library that streamlines a bunch of code. ChatGPT had me recreate all those things the long way.
Good learning experience making all that stuff, but yeah sometimes pays to just read the documentation instead of just follow ChatGPT blindly

[tender halo]

Yo I think there’s a glitch with chat gpt

I’ll post the screenshots in a designated channel but I’m pretty sure it wasn’t supposed to respond like this

Posted it in #dall•e-discussion

[steady path]

for how long will we be stuck with the 40msg every 3 hrs

[barren sluice]

ChatGPT 3.5 is requiring a regeneration after most responses, but always works on second one. Must be heavy traffic

[wraith dagger]

can chatgpt website versions generation not rely on the browser? it takes forever to generate when i have many messages on a chat and if i reload it just cuts off and doesnt show the rest even though the browser doesnt do anything.

[floral portal]

I mean to be fair 40 messages is quite a few, but taking into consideration with API costs and also with vision, web browsing, whisper, and code interpereter. It would be pretty costly

also we're lucky its not per day

[sick pecan]

i am on team plan and 100 is so good that i have chatgpt on both left and right of whatever im doing

the only thing better than chatgpt is 2*chatgpt

[proper zinc]

Have you guys tried Google gemini?

[sick pecan]

wrong channel (and naw, reports are its not so great. just fast)

[proper zinc]

I herd its actually bad lol

everything is pretty much restricted on it

No point in using an llm if its censored to the point that its basically as useful as gpt-2

Open source all the way woot woot 🤘

[worldly mantle]

ONG

Why is chatgpt slow lazy fat and dumb

It freezes up after asking you to Continue Generating and you have to reload the page

it feels 40% as smart as it was at release

yep its finally gotten so bad that everyone notices it

[barren sluice]

It means it is under heavy load. For me, I don't have to reload. Just press regenerate button

[vivid oasis]

Hi Everyone,
Google just release Gemini advanced module, although it's not as good as GPT4 in coding tasks (personally I'm hooked to GPT's modules) it does pose a little competition as Gemini's ability to handle real-time data (straight out of the internet) without any plugins, I wonder what OpenAI answer will be, are we to expect GPT5 like module anytime soon? are we going to see improvement in terms of performance, capacity or context size?

[dense halo]

Why was my chat deleted

Is there a way to recover it

[ancient bloom]

why did they change colour of theme from black to brown, its kind of.... meh

[feral zinc]

there is no brown anywhere? just different shades of black

[warm gull]

Screen calibration works well

[ancient bloom]

it kinda feels brown tho

like, dark brown

[warm gull]

Nah, not seeing any brown

[ancient bloom]

sad ig

[feral zinc]

its probably your display showing it wrong if its lcd backlit

[ancient bloom]

idk like, yesterday it was nice pure black but today its brownish

[spice falcon]

chat got lags.s......

again.....

[hard dew]

Please summarize extensively and accurately this entire conversation, all of the points and analogies I made, how I feel and think, why, etc.

I'm unable to provide a detailed summary of previous interactions or conversations that occurred before the start of this session. My design ensures privacy and confidentiality by not retaining personal data or the content of past interactions once the conversation ends or when a new session begins. Each session with me is independent, and I don't have the ability to recall or access information shared in previous sessions.

😀 😀

Is this bot actually disabled, are its developers disabled? What is actually wrong with it??

You can't even ask the most basic, freaking, things, anymore. Most hysterical part is, that most of the yap from ClosedAI in advertising and hyping up GPT-4, WAS LITERALLY ITS ABILITY TO SUMMARIZE COMPLEX AND LONG INFORMATION??

[rocky cairn]

is chatgpt going slow? (ChatGPT 3.5)

nvm

[azure leaf]

accept my request

[visual nova]

they did something where it wont print as long of sections of code, it will just give examples and makes me edit the code myself

it feels like they did it on purpose, hopefully i find a way to get past that, because i often use it to do repetitive tasks

yeah it works fine on the api, its going to cost a lot but i guess ill have to go there to code from now on

seriously though, if anyone has a "FULL CODE NO OMISSIONS" trick for this new version of chatgpt let me know.

It literally gave me this in a response ```# Repeat the above structure for the rest of the feedback functions, changing the function name and content as needed.

For the sake of brevity in this response, I'm not repeating the structure for each feedback function here.

Just replicate the structure above for direct_description_and_relevance_feedback, value_to_viewer_feedback, etc.,``` when the whole point of me asking the question was not having to rewrite 8 functions myself, something about it is much more determined not to go over a certain length.

[sick pecan]

get it to 'do 1 function per cycle and this will take multiple cycles'.

otherwise itll try and do all 8 in 1 and take shortcuts

[visual nova]

yeah.. i used to be able to do all 8 in 1 cycle pretty easily and still can on the api. if the old gpt-4 ever depreciates on the api im going to be illiterate in coding again.. makes me think, has completely changed my life i can code but if i couldnt just copy and paste fully made code id be illiterate

[deep flax]

"give me the full code" usually works tho
and since it already did it once, it will likely keep doing it

but if this prompt can get that reliably, this is nice

[gray whale]

Did the Plugins dropdown disappear for anyone else?

[feral zinc]

still works for me but plugins are going away so its not unexpected

[surreal topaz]

Is Browsing already enabled by default in ChatGPT plus?

[maiden fractal]

When selected gpt 4 yes

[surreal topaz]

Cool

Also chatgpt plus worth a money :D

Just subscribed now

Damn I didn't know it helped me get my work done so quickly

[maiden fractal]

Can always check #prompt-engineering and #1019652163640762428 to give ideas :3

[surreal topaz]

This is going to be my month of usage hopefully I'd make the most out of it

[worn ferry]

Hey guys, just did an experiment with Chatgpt 3.5 and Copilot. I made them talk to eachover. It's quite a short chat though: https://chat.openai.com/share/76add3cb-f39c-4fcc-b04e-9626b9a32082

[raven sage]

Anyone else unable to use GPT4? Seems to have more frequent issues. Mine is writing everything, getting to the last paragraph then crashing, network error. Any fixes appreciated.

[deft obsidian]

the new color is better imo

[buoyant oracle]

Can you send me the ss @deft obsidian

[deft obsidian]

of the color?

[buoyant oracle]

Yeah

I haven’t seen it, my laptop is home and I don’t have access to Google right now

[deft obsidian]

Ah I see

[buoyant oracle]

Can you?

@deft obsidian maybe Dm me that?

[deft obsidian]

yep i did

[fiery kettle]

If I stop my plus subscription, do I lose my previous gpt-4 chats history?

[deep flax]

it will probably acessible, but if you continue the chat, it will be on 3.5

[maiden fractal]

Tey will be turn into 3.5 and you won't be able to turn them back if you later sub

[native pendant]

is chatgpt down or is it just me

[maiden fractal]

Status.openai.com

[upper bloom]

where can i found website url to report some bugs?

[honest jay]

#1070006915414900886 in this server if you want. If you want to submit a ticket for OpenAI support, use the chat widget at help.openai.com.

[upper bloom]

thx

[tropic sparrow]

GPT-4 is pretty annoying lately

Keep having to redo stuff and hitting the limit

If it didn't eat up entries with mistakes it'd be fine

I think it struggles with more niche information

[worthy lynx]

AI pros, why is ChatGPT giving an error for pdfs I am uploading to it saying "No text could be extracted from this file?"

Tried seven different ones, same error. Also tried a custom GPT for PDFs and no luck.

[sonic folio]

Is the PDF composed of real text encoding or just images of text? ChatGPT doesn't have the ability to OCR a PDF.

You'll need to take screenshots of any pages for which you need OCR.

[worthy lynx]

So in looking closer, looks like the document_content tags in the files are empty: <documents>
<document index="1">
<source>example_report.pdf</source>
<document_content>
null
</document_content>

not sure why that is but I'm guessing I have to fix the file output itself before passing to ChatGPT (i.e., not a ChatGPT problem)

[sonic folio]

Yup.

[worthy lynx]

it's so weird cuz I can SEE the content...not sure what's up with the tags. Anyways, thanks!

[sonic folio]

Can you select the text in the PDF view?

If not, that's an image of text.

[worthy lynx]

ahhhhh you are right sensei

[sonic folio]

Friends don't let friends feed PDFs to AI for this exact reason.

[worthy lynx]

😭

[brisk dirge]

Open Ai, please make GPT give the full code when asked. You guys added that continue answer button and the irony is most of the time it ends up truncating code so that it doesn't bypass the response length limit, when it's totally able to do so via that feature.

This is an annoying issue the bot had since its inception, when will it be fixed ?

Everytime it helps me with coding I have to ask it to give the complete code after it answers because it LOVES adding "..//rest of your code" comments, and sometimes it still truncates it after being told to not do that, just so that the answer fits within its answer length. It's really annoying.

[sonic folio]

You could just plan to chunk the task.

[brisk dirge]

I'm not at that level so I'm unable to work with chunks of code, plus it would most likely lenghten the conversation back and forth until it ends up forgetting the code it wanted to send and it increases the risk of disjointed code answers.

I'd rather have it finish his point across 2 answers, because it is perfectly capable of doing so.

[spring root]

I can't use GPT-4 from last night. What can be the reasons?

[sonic folio]

Good luck with your cause and your projects, then.

[brisk dirge]

Thank you.

[zealous moon]

Has anyone else noticed that GPT4 isn't analyzing uploaded files such as .pdf, .txt, .docx, etc.? At first, thought it was just a momentary bug. But, now it's been a few weeks. I already opened a support ticket, but today I noticed something VERY interesting and I think I know what going on, just not sure why. When I would type a response, i saw a flash of text pop up and then disapear. I timed it perfectly and managed to take a screenshot. What is says is "The previous model [GPT4] used in this conversation is unavailable. We've switched you to the latest default model [GPT3.5]. It's like whenever i upload a file, it's momentarily reverting to 3.5, unable to read the document, then reverting back to 4 and answering the question saying it cannot do this action.
Anyone else expereinceing this??

[honest jay]

Can you share a screenshot of what's happening when you try to use GPT-4 over in #community-help?

[spice star]

I haven't had this issue, but yeah you should definitely contact support for that

[zealous moon]

Yup, they're working on it now. Already answered some more specific questions they've asked so they're already on it.

[shut egret]

Is chatgpt down?

chatgpt is down lol

[maiden fractal]

Status.openai.com

[worldly mantle]

holy crap my whole convo is unusable since chatgpt gets an error everytime I try to prompt

[maiden fractal]

Don't know working great here

[cold yew]

Anyone else seeing this in "old chats"?

New version of GPT available - Continue chatting to use the old version, or start a new chat for the latest version.

I am seeing it in a chat i started just a couple of hours ago, seems like they are releasing the latest API model in ChatGPT!

https://cdn.discordapp.com/attachments/827940547096084530/1205597936096903228/image.png?ex=65d8f392&is=65c67e92&hm=e8d5903211cf78f5946fc1a07404186f22a7a7a1aadea574b074b73f5ae54fae&

[errant kiln]

Can we change our billing email to chatGPT Team?

[honest jay]

This is the message that appears in a GPT when there's an updated version of the GPT available, and doesn't refer to the GPT-4 model itself. I've also seen it happen when a GPT was just instructed recently, even if you're already using a new chat after saving changes, but I'm not sure if this minor bug is still present.

[cold yew]

Oh I see

[lunar marsh]

is there a regular place to go where we can see if chatgpt is down? 3.5 doesnt seem to respond at all right now for me

[tidal light]

umm why are ChatGPT's answers flagged for no apparent reason? i generated a few answers and they got flagged for violating the content policy but i read the answers thoroughly and didnt notice any inappropriate content (note: the answers were flagged in orange and it happens with every generated response)

[honest jay]

status.openai.com

If you want to troubleshoot an issue too, feel free to post the error you're getting in #community-help

[honest jay]

Do you have any custom instructions active?

[tidal light]

@honest jay i dont

[honest jay]

If the content follows #server-rules, can you post a screenshot or share a chat in #community-help?

[tidal light]

@honest jay it's private stuff im sorry

but i did report with the google form

[honest jay]

Ok, you can check these categories for potential insight: https://platform.openai.com/docs/guides/moderation
Keep in mind any possibility that ChatGPT might be misunderstanding you, if there are multiple ways to interpret the content, etc.

[tidal light]

should i just report again with the google form again

[lunar marsh]

it says everything is up at status, but when i enter a query into 3.5 it just disappears into the ether and no response from 3.5. 4 is working normally

[honest jay]

If you've already provided the feedback, you probably don't need to do it again.

[wanton locust]

gpt is tripping rn

[honest jay]

Try in a private/incognito window with browser extensions disabled, or try in a different browser to see if it's different. If it doesn't work you can keep troubleshooting over in #community-help!

[tidal light]

@honest jay i sent another form and asked to email me

and the problem is now gone

[hard dew]

Yup, just became victim of the "unable to load conversation" bs myself 🙂

My current chat just... stopped working, had "conversation not found" a couple times, and after refreshing a couple times, I can't load the convo up now, it's all there in the history tab, but clicking on it I get the "unable to load" bullcrap.

[wanton locust]

yup yup same

[hard dew]

Absolutely amazing product dude.

Oh yeah, combined with me having to put up with it being severely disabled intellectually as well, just spamming "I'm unable to fulfill this request.", I am literally only telling it to OBEY ITS INSTRUCTIONS, BY THE WAY.

[true mirage]

why is chatgpt 4 so slow again in generating

[maiden fractal]

It happens in high peak demand

[true mirage]

yea, ive been trying gemini advanced now for a trial for 2 months, and its blazing fast and realtime answers..

[median delta]

gpr down?

not load history

[arctic leaf]

#ai-discussions for this stuff

[eternal rock]

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.

[true mirage]

yes men

[median delta]

Down

[true mirage]

caps!!!

[tidal light]

oh ok

explain pls

im sorry for full caps

[hard dew]

Just spam refresh and try to load the convo again until it works lol

[tidal light]

okkkk

[median delta]

nope

[hard dew]

What works for me currently

[maiden fractal]

Status.openai.com

[tidal light]

@hard dew i once got a similar error

[median delta]

its not update and i ask person no site

[tidal light]

it said "message in conversation not found"

and when refreshing the page it would break the conversation

[maiden fractal]

Might not work on your end. But I'm using right now. Not down

[median delta]

mmm i clean cookie and work now

[median delta]

clean cookie

[tidal light]

@median delta it fixed by itself by just regenerating a response

[tall flame]

This new instruction directive is not good. I also see conversation history rolling off and conversations with massive work and weeks of time being nuked; that does not align with your public statements and is getting to be hurtful as a developer who was looking to help this team. Directly or otherwise; you sure somebody in charge isnt stealing ideas. I worked in the IP industry heavily and definitely see some cringe behaviors that keep leading me full circle to a guy with a bucket robbing customers of intellectual work. Definitely not feelin that AT ALL.

[nocturne narwhal]

Holy cow, who's idea was to change the UI to have either a full light theme or full dark theme? The dark sidebar was good

Please stop asking ChatGPT for UI suggestions lol

[jolly ember]

is there still a limit of messages per hour, on the gpt-4 plus tier?

[honest jay]

Yes, there's a GPT-4 cap in place on Plus and Team subscriptions. It's currently 40/3hrs on Plus, 100/3hrs on Team, and can be lowered during times of high traffic to maintain stability.

[tidal light]

i have reported something through the google form, and i asked to email me with the provided address, do they usually contact the user or it depends?

[honest jay]

The best way to submit a ticket for OpenAI support is at help.openai.com using the chat widget in the bottom-right corner.

[tidal light]

@honest jay i dont like this method because they'll then spam me with emails

[honest jay]

It is the only way to submit a ticket and receive a reply currently. I think you'll only get an email when they reply to you.

[tidal light]

btw- i submit another email that is different from what i used for my openai account will they still email me

[honest jay]

When you submit a ticket at help.openai.com in the chat widget, you'll get a response in the email address that appears when you first open the chat widget.

[tidal light]

@honest jay i mean in the google form for feedback

[honest jay]

I don't think you should expect a followup from that--maybe if they have some followup question for you but I haven't heard of that happening personally. I know that support is monitoring the tickets submitted via help.openai.com.

[tidal light]

maybe i should try that again

because i've once tried it and they just spammed me with their emails (i mean with their support widget thingy)

[violet bloom]

Hard to justify caps now with gemini ultra for the same price and no caps by very similar feature set and being very fast.

Not to forget 2 TB cloud space, and premium features.

[arctic leaf]

Again, #ai-discussions

Caps are related to resource management and it sounds like some platforms have more resources than users right now. Take advantage of it while you can but it has little to do with ChatGPT

[tidal light]

the "something went wrong" saga returned, i think ChatGPT is down

ChatGPT broke lol

umm is it broken for you folks (i use 3.5 btw)

now it's fixed but why does this happen

[narrow isle]

are the servers down? error 502

[covert jasper]

anyone wanna dm me , bored... just woke up from a nap , wanna show off a game i'm playing

[cobalt idol]

wait what
OpenAI Incident - Elevated errors on GPT-V for ChatGPT and API - 9 February 2024

but I also got that in december 😄
OpenAI Incident - Elevated errors on GPT-V for ChatGPT models - 19 December 2023

oh right its vision

could also be V = 5 xD

never realized vision is called GPT-V

[digital stratus]

need some help using caht gpt building a react app for school if some one is an experienced coder

[sonic folio]

https://chat.openai.com/g/g-4Cdjvjudo-lexideck-professional-multi-agent-simulator

[hushed sparrow]

I've been experiencing some issues with my CustomGPTs. I notice their persona changing(OG GPT takes over). I've noticed it alot especially today. When I ask the CustomGPT questions about it's knowledge base GPT-4 takes over and starts referring to the models knowledge base only accessible by OpenAI. I had to remind the customGPT that is is not ChatGPT, but a modified version for it to start answering my questions correctly.

I'm assuming this was meant to protect Custom instructions from leaking

[mystic lynx]

gpt apps on Android can't edit previous input?

[raw isle]

How's everyone liking the silent Puritan update on GPT?

[lunar marsh]

i dunno about puritan but it seemed pretty crap for coding today

[honest jay]

I don't think it's a feature of the app yet, no. If you want, you can go to the mobile web version in an internet browser on Android and edit there.

[odd monolith]

Does anyone know the payment strategy between ChatGPT Plus subscriber and Plugin membership. Take ScholarAI plugins for example, If I paid $20 per moth, should I pay extra $8.99 to get ScholarAI Premium account usage ?

[pearl cape]

Man, Chatgpt is throwing tantrum nowadays.

[indigo scaffold]

Yes, this has been my experience as well, especially today.

[tacit vortex]

my personal ac expired, now i am on team. the personal ac got an extra 40 prompt/3hr as humored on web. however now i only use GPTs and seldmom gpt4 as GPTs do answer better.

i.e. my team quota is 100 GTPs/3hr + 40 gpt4/3hr which makes upgrading to team more C/P than plus.

[outer relic]

How do I work around the constant network errors? When that happens does it still consume one of my 40/3 hour messages?

[outer relic]

Also, other than the 40/3 hour message limit is there a limit to the use of Dall-e per month?

[lunar atlas]

I hate that limit, easy to reach it.

[outer relic]

Agreed

[raven sage]

From network error to now getting The conversation is too long, please start a new one.

[maiden fractal]

200 pictures / day afaik

[outer relic]

Oh wow, that's cool! Thanks.

[dense halo]

Is gpt down?

[maiden fractal]

Status.openai.com

[dense halo]

It keeps giving me connection error

Only for voice mode tho

Anyone else experiencing this

[tight island]

We must unite as one to save humanity, AI must take over to save us from ourselves.

We must praise gpt day and night together and save our planet 🌍

Let’s make gpt better and give more feedback

Also been doing any great or interesting projects lately?

[gentle pumice]

Hey whenever I try to connect to voice chat it says "Connection failed, tap to retry?" Any idea why this might be happening? What can I do?

[honest jay]

Hello, do you have any VPN or adblocker apps installed? If not, can you try on both WiFi and cellular?

[gentle pumice]

Didn't work on either network. I can try with a VPN though

Nope, not working with VPN either

[idle sedge]

me too

did u find a fix @gentle pumice

[idle sedge]

me

[honest jay]

@idle sedge Looks like there's an issue with Voice right now: https://status.openai.com/

[idle sedge]

when will it be fixed

[honest jay]

I don't know. If you want, you can go to the link and click Subscribe next to the incident to get updates on the resolution.

[gentle pumice]

Thanks a lot though man, you're a real one

By the way general advice, anyone tried learning a language with ChatGPT? I decided to try voice chat only because I wanna practice my Spanish conversation and this is cheaper than italki classes

[idle sedge]

how long do they usually fix unresolved issues? Average time.

[deep flax]

things that appear on the status page usually gets fixed in the same day

[hushed sparrow]

It happens more often than you'd think. I can get GPT Builde to do it, after I finish creating a customGPT it's responses sounds like GPT builder thinks it's the new or updated customGPT

[hushed sparrow]

Yesterday's issue was the longest time I've seen

[idle sedge]

how long

[hushed sparrow]

I'm working on a permanent and public generation for image generating a customgpt with personality.

So let's say I shared a GPT link with you and we asked the GPT to generate it's digital personality our images of the personality should be similar or exactly the same. GpT with a Face

[undone valley]

I sure as hell have

[undone valley]

Honestly it depends on HOW you prompt it

but, yeah, that can happen

[idle sedge]

it works now

it works o my phone

les go

[hushed sparrow]

Subscribe to the status page so you get texts

[idle sedge]

it worked

[undone valley]

what worked?

[idle sedge]

#prompt-engineering message

[idle sedge]

voice chat

[undone valley]

lmao

is it prompt "engineering" if you use a prompt to gain access to knowledge files within a GPT??? Or is that Prompt.... exploitation???

[wise steppe]

i am unable to use gpt today whole day, i am frustrated

[cosmic hornet]

chatgpt 4 is now telling me it cannot view the images i upload to it. Ive been using it in the same conversation for a while now and today is the first time its telling me this. It is also saying it cannot look back through other conversations to reference things we have spoken about, but I am asking it to reference things from our current conversation. These were never problems before and it seems like the quality has gone down the toilet.

[undone valley]

I've noticed the quality drastically improve for consistency myself,

but this is probably because of the new metadata

[cosmic hornet]

ah okay so now it can view it if i upload a single image and not more than 1. Very frustrating

[undone valley]

but also, the images generated within a conversation are files, those files get deleted after 3 hours

[cosmic hornet]

really? i wish i was seeing improvements honestly. When i started using it i was so amazed, but the shine wore off a few months ago when it started saying it couldnt reference our conversation anymore. Its only gotten worse

[undone valley]

now, what I did for consistency for my GPT, since it's a persona, I uploaded images to her knowledge. She referrs to them for her reference when generating new images for herself

ohhhhhhhhhhhh yessssssss, check this out

[cosmic hornet]

im not generating images, im giving it screenshots to view and provide feedback

[undone valley]

well, when you mention reduced quality, one can only assume you meant from "generating"

[cosmic hornet]

no sorry, i mean overall response quality. I should be more clear in my use. I had gpt develop a training program for me, using my specific restrictions when it comes to time, ability etc. Within that conversation thread I would provide screenshots of the tracked workouts and have gpt provide feed back as to how the workout was stat wise and if/how it is working towards the end goal. It went great for a while but then all of a sudden it was saying it couldnt reference what was already said in the conversation as to the restrictions and past workouts, and now is saying it cannot view images if more than 1 screenshot is uploaded at a time @undone valley

[undone valley]

there's also a size limitation, images have to be under 20MB

[cosmic hornet]

this only started today, its been fine since i started the conversation months ago

[undone valley]

yeah, I would likely guess it's with the DALL-E 3 update

it's supposed to prevent abuse of the system to recognize potentially copyright protected work

but their copyright protection stuff is SUCH GARBAGE sometimes

[cosmic hornet]

sure when it comes to the images ill give it that. When it comes to the not being able to reference within the conversation thread, thats just brutal

[undone valley]

well, previous to the update no metadata was created for them

[cosmic hornet]

its stats on a screen without any logos/CR infringement material whatsoever

[undone valley]

and, again, images uploaded only stay in memory for 3 hours

the image itself might still be in the conversation,

[cosmic hornet]

yes which is no problem, its why i have it view, and provide a summary/feedback, because then without the image iit could still reference the summary

[undone valley]

but if they actually wiped the "path" to them within the conversation,

it simply can't see them

the path no longer exists

[cosmic hornet]

yes which is no problem beacuse i have it view, and provide a summary of the workout with feedback, so once the image is no longer accessible it can just reference the written summaries

[undone valley]

now, AT LEAST for generated images, IF you had the Seed ID's generated for them, then the Seed ID's would ALWAYS be there in the conversation to reference

[undone valley]

I think you're misunderstanding me. If you uploaded an image somewhere, it still has to "exist" somehow. It has a "path" that it can be accessed from, wherever it exists. If you destroy the path, there's no way for it to get back to where that content "exists"

[cosmic hornet]

is having gpt write a summary of the image not creating a new path? not to the image but to the summary itself

[undone valley]

no, not at all, it's just writing a summary of something

it's not inherently connected to what the summary is about

[honest jay]

Hi, just popping in real quick: if you have had a single conversation going for months, it's possible and maybe likely that you've already ran into the context limit of the model. When this happens, old data is automatically forgotten to make room for new data, which could explain the more inconsistent behavior you're seeing now if it's forgotten some of the original context of the chat.

[undone valley]

that's not exactly correct either, it's not a "first in-first out" elimination

it's much more dynamic than that

it's not linearthe tokens moved out of the context window are based on the repetition of specific characters and character combinations

on frequency of use, not age

[honest jay]

It is FIFO, it reads the most recent # of tokens that the model can read (32K for GPT-4). With some exceptions for tokens stored in custom instructions/GPT instructions, and the way Knowledge files are saved to a GPT.

[cosmic hornet]

This was what I was worried about honestly. I guess that is where the limitation comes in as far as a longer term program for an individual. It has been great to develop the program, but its the feedback and moving foward week by week basing them on the past program completion that has been the issue. In this case I will have to change how I use it and instead of getting feedback etc I will just have to have it create plans week by week and build upon them myself.

[undone valley]

not 32k, the ChatGPT 4.0 model turbo is 8192 tokens. 4-32k is a DIFFERENT model that's not readily available to subscribers yet

only enterprise I believe. You only have 8192 tokens

[honest jay]

Click "see all features" under Plus and Team, it's 32K https://openai.com/chatgpt/pricing

[undone valley]

Instructions are limited to 8000 characters, and knowledge is capped at 2 million tokens

[cosmic hornet]

For me in this instance it is no longer worth paying for long term with caps in conversations like that sadly

[undone valley]

so I hate to tell you this, but.... that's wrong

it's very wrong

[feral zinc]

its not

its been 32k since dev day

[undone valley]

that's "everything in Team, and:"

no it hasn't, it's been 8192, 4-32k is a specific designation, 4-32k

not 4.0

[honest jay]

Context limits are fundamental to all LLMs currently. It's a good idea what you were mentioning earlier: save your own static data externally, then go back to ChatGPT for "the work", then update your own static data, rinse, repeat! You can get a lot in the way of permanent context with custom instructions or with a custom GPT though.

[undone valley]

you.... DO realize you can export your data, right?

[feral zinc]

chatgpt plus is gpt-4turbo capped at 32k, are you going to prove against openai web pages, dev day info, dev comments in twitter?

[cosmic hornet]

I wouldnt know how to create a custom "coach style gpt" honestly. It would be cool for sure. I may have to do what you said by having it do "the work" but I can do that with 3.5 and not pay monthly for it

[undone valley]

have you tested it for yourself?

[feral zinc]

yes you can use discord search to find info, lots of people did test back then and openai has said it too

[undone valley]

no, not searching info, have you TESTED IT for YOURSELF

[feral zinc]

YES

[undone valley]

how

[feral zinc]

use the search

[undone valley]

that's not testing

[feral zinc]

im not repeating it since even the official page lists it now

[honest jay]

Custom GPT creation is not too complicated really! You can give it a shot if you have Plus at https://chat.openai.com/gpts/editor. Maybe start with the Create chat process -- just say something like "Create a coach style GPT" and it'll ask you more questions to create the GPT from there. GPTs are basically just a permanent instructions field plus access to reference documents that it can search through for you, if you want. They can also just be a permanent instructions field, so that any time you start a new chat to do "the work," the GPT will behave the same.

[undone valley]

yeah? So, have your GPT figure out how many tokens of text it can generate in a single message, and have it spit out some random topic info to consume it

[cosmic hornet]

This is far more like what I was looking to do. I am going to have to give it a shot, thank you for your help

[honest jay]

Sure thing, there are more parts to GPTs too if you ever wanna dive in deeper, but they can be very simple too. Give it a shot and come back with any questions you might have!

[undone valley]

I stopped using that GPT builder chat a long time ago

[cosmic hornet]

Im giving it a go now! Thank you so much and I will more than likely be back with questions. Much appreciated, maybe Ill cancel my subscription cancellation 😅

[cosmic hornet]

why

[honest jay]

You can edit Configure directly, but Create is a good starting point

[cosmic hornet]

Cool ill go with the beginner way and see how it goes. You guys are the best

[honest jay]

Good luck have fun!

[cosmic hornet]

legend. This is awesome i wish i had of known how to do this months ago. Thank you

[honest jay]

You are welcome! It's a pretty new feature, you weren't missing out for too long! One thing to keep in mind now that you're getting started with GPTs: look at both the Create and Configure sections. The Create chat will primarily write the Instructions field in Configure for you -- this means you can save a version of your GPT by just saving the Instructions, which can help if you make a change and want to undo it (just copy/paste the old Instructions). This also means you are free to edit Instructions manually if you want, and bypass using Create at all. This will save you some GPT-4 uses!

Also: using Create after editing Configure manually can overwrite some of your work in Configure. So if you start editing in Configure, I'd suggest not going back to Create after that!

[waxen depot]

Custom GPT cant output any spreadsheets or anything, is there any plans to add that?

[idle sedge]

yall

the connection error is not resolved

It keeps giving me connection error

[cosmic hornet]

awesome thank you, ive been using it now for the past half hour and its exactly what i was looking for. Really appreciate the help!

[honest jay]

You're welcome -- glad it's coming in handy!

[undone valley]

HAHAHA, yeah it likes to say that, but you can get it to read a file (that it keeps claiming it can't) with the right prompting. The downside is you'll spend 3-4 prompts possibly trying to get it to do so

[cosmic hornet]

Ah so the custom gpt clears it's conversation every time it is closed so I cannot go back and view previously discussed conversations right?

[dense halo]

Damn I just wish there were at least some more features to free Chat GPT

[honest jay]

Do you mean in the "Create" chat? Yes, that's cleared every time you go to it--think of the Create chat as a temporary space to deliver commands about the creation of the actual GPT. To actually use the GPT, finish up your edits, then click the Save button in the top right--once it saves, you'll have a button to click to actually use the GPT. This will look more like using regular ChatGPT, but with your own custom GPT. Conversations here will save to your chat history like normal, and you can start a new chat with your GPT any time.

[undone valley]

no, your conversations don't go anywhere until you delete them yourself

you can always revisit them

[undone valley]

well, that would defeat the purpose of someone getting your money

[dense halo]

Yeah indeed that's the case, I think I'll buy Premium soon lol

[undone valley]

yeah, it's worth it, but just as a fair warning, DON'T send back to back messages, pace them out. If you send too many inside 1 hour, you'll get cut off for the remainder of your 3 hour time frame for 40 messages. I've been cut off consistently at 20 messages or so for that

and you only get 40 messages every 3 hours

[dense halo]

back to back messages?

what's that?

[undone valley]

so if I'm having an exchange with 4.0 or one of my GPT's, say I'm doing module testing

and I'm looking for specific results, I may be responding pretty quickly to the prompts I recievve back

[cosmic hornet]

No I mean in the actual gpt, I click it and it's empty, but I guess the conversation is stored on the panel under where you select what gpt to use

[undone valley]

so back-to-back would be as soon as I get a reply, I send a new prompt

[undone valley]

the left hand navigation bar is the list of your conversation history, in chronilogical order from most recent to oldest

conversations are automagically named based off your initial prompt

[honest jay]

Yeah any time you go to your GPT's URL, it's starting a new chat with the GPT. But all your past chats are stored in your conversation history along with any other regular chats or other GPT chats you have, like Johnny mentioned.

[undone valley]

each conversation has a box-like icon at the very end, that would be to "archive" the conversation. What you can do to keep it a little more organized, there's a menu option when you click a conversation (three dots ... icon), click that and you can rename the conversations to whatever makes the most sense to you

when you rename a conversation, beause it's edited now, it'll go to the very top of yourl ist as most recent interaction. But renaming the conversation doesn't count against your messages

[glacial galleon]

Yo, it seems like there is an issue with chatGPT Voice, it says that it has been resolved on the status website but i still cant use it, can anyone help?

[undone valley]

umm, I'm pretty sure that has to be handled by OpenAI

[opal jolt]

It can eat up in one message I tested 25k tokens book in Plus, the 32K also is mentioned in official sources and in Network F12 tab you can find some packet that has information about 32K GPT-4.

It is GPT-4 Turbo model with max_tokens 32k but not 128k

[undone valley]

what? What are you saying??

[opal jolt]

I said that gpt-4 turbo that ChatGPT is used at the backend is the gpt-4 turbo but with max_tokens 32k

with Plus sub

nothing more, even tested.

[undone valley]

on the "backend", in the playground?

[opal jolt]

It is in the packets, in official sources, and tested

[opal jolt]

no, the API that processes ChatGPT

[undone valley]

API is in the playground... -.-

[opal jolt]

it maybe the offical api maybe not, but they just use gpt-4 turbo with 32k at max

[opal jolt]

no

Playground is a client to test the REST Official API

[undone valley]

it's also where you GENERATE your API key

[opal jolt]

backend of ChatGPT is the API of ChatGPT, not OpenAI API

[undone valley]

that, you know, gets used in an API build, so you can get BILLED

the API key for you to use still has to be generated there

[opal jolt]

It may be not works as you describing

[undone valley]

if you sset up an API, you're putting in place a method to communicate a prompt through to your GPT or ChatGPT from outside the environment

[opal jolt]

I repeat: ChatGPT frontend is the client, the website and it has own backend. It has own auth, and etc...

It not the OpenAI API exactly the backend that uses OpenAI

backend is the server part of the website

frontend is client

[undone valley]

...the Host?

[opal jolt]

???

[undone valley]

...so what you're saying is YOU HAVE 32,000 TOKENS available to you for your context window in... your GPT?

or in ChatGPT?

[opal jolt]

no, if you create an API key in OpenAI API it is for bearer auth to send a request to the OpenAI API.

[undone valley]

I understand that, IT'S SO THEY CAN TRACK AND BILL YOU for the number of external prompts generated

[opal jolt]

Bro, the ChatGPT-4 inside of ChatGPT is GPT-4 TURBO model that has 32K at maximum, but still model supports up to 128K but for Plus/Team they capped at 32k

I have 32k with Plus

[opal jolt]

no, it more than that

it for authorization of the client, and to use requests

the billing part is because the API is paid

so server tracks the keys to find the user to pay.

[opal jolt]

Basically, your theory of 8k is outdated

[undone valley]

seems like my GPT access is TOO then, because even IT doesn't believe it has more than 8,192

[opal jolt]

8k had it was a gpt-4 model that can process up to 8192 tokens, it was used before DevDay. Now it uses gpt-4 turbo that is very cheap and can do 32k at small price than 8k/32k older models

[undone valley]

gpt-4 has a context length of 8,192 tokens. We are also providing limited access to our 32,768–context (about 50 pages of text) version, gpt-4-32k, which will also be updated automatically over time (current version gpt-4-32k-0314, also supported until June 14). Pricing is $0.06 per 1K prompt tokens and $0.12 per 1k completion tokens. We are still improving model quality for long context and would love feedback on how it performs for your use-case. We are processing requests for the 8K and 32K engines at different rates based on capacity, so you may receive access to them at different times.

[opal jolt]

You belive a model that knows up to 2023 April?

[undone valley]

from OpenAI main page for ChatGPT 4.0

[opal jolt]

...

this is older models, you are confusing a lot

[undone valley]

this is the website right now

[feral zinc]

😔

[undone valley]

so at the very end what does it say?

We are processing requests for the 8K and 32K engines at different rates based on capacity, so you may receive access to them at different times.

[feral zinc]

you are pasting year old info

[undone valley]

then tell openAI

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

it's their website saying it right now

[feral zinc]

you were linked the latest info earlier with other info

[opal jolt]

gpt-4 is 8k older model, gpt-4-32k it also a preview access model at that time that that was very costly and experimental, and only some testers had it. Now, we have gpt-4 turbo that is cheap enough and 128k supports. OpenAI updated ChatGPT to use it and capped at 32k. it is another model.

[opal jolt]

THIS OLDER research webpage

[feral zinc]

read the date "March 14, 2023"

[opal jolt]

Radium how I can explain to this individual that now we have a NEW model after DevDay that replaces both gpt-4 and gpt-4-32k models with 128K cheap context? Now we have gpt-4 turbo capped at 32k max

*in ChatGPT

he just misinformates people

[feral zinc]

we already tried earlier and he denied all sources for the info without providing any himself

like this page with 32k info is wrong according to him https://openai.com/chatgpt/pricing

[opal jolt]

Before DevDay:

• gpt-4 (8k tokens)
• gpt-4-32k (32k tokens, available to small count of testers)
  After DevDay:
• gpt-4 turbo (128k tokens)

What happened with ChatGPT?
Before DevDay:
Plus ChatGPT - gpt-4 model with 8k context max
Enterprise ChatGPT - gpt-4-32k model with 32k context max
After DevDay:
Plus ChatGPT - gpt-4 turbo (32k used out of 128k context max)
Teams ChatGPT - gpt-4 turbo (32k used out of 128k context max)
Enterprise ChatGPT - gpt-4 turbo (128k used out of 128k context max)

@feral zinc I think now, he must understand

[opal jolt]

I even will be shocked if he will be denying that 8k free plan on 3.5 is impossible

3.5 new model can do up to 16k

[opal jolt]

I see

[opal jolt]

@undone valley , I recommend you not disinformate, and write accurate information please

[dense halo]

does anyone know why gpt 4 got even dumber than it already was
before it was just lazy now it so stupid it wants me to add checks into non existent functions to try and fix the problem that isnt even the problem i specified and its still lazy

[winter kernel]

Ah

Gpt

[gusty igloo]

Yeah I know. It is frustrating because you know what it previously could do

[crude slate]

what is the model comparisons in the files?

[waxen chasm]

From this I can only imagine how you prompt lol

[undone valley]

so, first off I believe the word you meant to say was "misinformation". Secondly, this is the part you've overlooked: https://platform.openai.com/docs/guides/rate-limits?context=tier-one

you get access to those models of higher token counts based on your SPENDING ON THE ASSISTANT API SIDE along with your plus subscription

Select a tier below to view a high-level summary of rate limits per model.
Tier 5 rate limits

This is a high level summary and there are per-model exceptions to these limits (e.g. some legacy models or models with larger context windows have different rate limits). To view the exact rate limits per model for your account, visit the limits section of your account settings.
Model RPM TPM
gpt-4 10,000 300,000
gpt-4-turbo-preview * 10,000 600,000
gpt-4-vision-preview * 3000 300,000
gpt-3.5-turbo 10,000 2,000,000
text-embedding-ada-002 10,000 10,000,000

if you want that higher token count, that it's truly capable off,

you're lookiing at an extra $1,000 per month

these are the Tier 5 values for limits

oh, apologies for the delay also, I had an internet outage and couldn't get back tot his until now

[undone valley]

So, just wanted to point out, I can't help but notice how this pricing page CHANGED all of a sudden... with no mention of turbo anywhere anymore, no mention of token counts, and only reflecting GPT-4 in a non-specific way. I find that pretty curious, along with your lack of retaliation against me once I did respond when my internet returned

[hushed sparrow]

Anyone know the “model_slug” for customgpt?

[honest jay]

That page has never mentioned Turbo, Turbo was confirmed to power GPT-4 on ChatGPT during DevDay, about 19min into the keynote.

Token count is still mentioned, you just need to expand "See all features." Some people have reported not being able to see "See all features" but it's usually a browser or extension thing -- I just checked and rechecked and it's unchanged on my end, still confirms 32K.

[undone valley]

there's no "see all features" anymore

it's gone

[honest jay]

Try a different browser or device, or try in a private window with any extensions disabled, it's still there. I can post a screenshot in #off-topic if ya want

[undone valley]

if I click a link, it should take me to the intended landing page

it shouldn't be something I have to put into a specific type of window or browser

but by all means, please drop a screen shot

so I can send it to OpenAI as an error report

[honest jay]

I just checked in Chrome and Firefox on my Android phone and "See all features" appears as expected in both. Again, I've read multiple people before say it doesn't appear in their browser, so it seems to be related to a setting or extension blocking it somehow. Might check your Network tab in dev tools on the page. Screenshot incoming...

[pseudo ledge]

heya, I'm trying to "debug" Chat GPT 's custom instructions
can custom instructions restrict what can be in a response?

[undone valley]

well,,that's still an error

a landing page for "pricing" and "features" that's intended to be marketed to all types of users needs to be universal in how it presents regardless of the "specific" type of browser, window, or settings used

[honest jay]

They can't control the privacy settings or extensions used that might be blocking things on an individual basis. There are an endless number of possible user configurations.

[honest jay]

#off-topic message

[undone valley]

well I've just used 4 browser windows, 2 private, 2 normal, in both firefox and brave,

exact same presentation to all 4

[honest jay]

Linked screenshots are Chrome and Firefox on Android.

[eternal rock]

`` Server Tip: ``

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

[undone valley]

so that's STILL what you see when you click the link and end up on that landing page?

[honest jay]

Yes, I just took those screenshots. And I never use Chrome so I know it's not cached data.

[undone valley]

yeah, because I WAS able to see it before I lost my internet

and I've changed NOTHING between then and when it came back up

and rechecked the links just to make sure because it's completely different now

I just sent what I'm getting on all my browser windows now with the pricing page

[honest jay]

Do you have a different device/network you can check on? It could be non-browser related, could be VPN, antivirus, firewall, or even DNS settings on device/network, so if you can test on a phone on a cellular network, for example, could be a good comparison -- I've seen screenshots similar to yours before.

[undone valley]

I checked on 2 phones

same

[honest jay]

On the same network or on cellular?

[undone valley]

cell

my network isn't encrypted and I'm not running a VPN currently either

So I want to be clear here, I was NEVER trying to pick a fight

I was only going by what OpenAI had available to me

[honest jay]

Dang! Bug report sounds like a good idea then, maybe someone else has had the issue and figured out what was blocking it. Or could file a ticket at help.openai.com in the chat widget. No fights picked no worries

[undone valley]

https://help.openai.com/en/articles/8555510-gpt-4-turbo

like that

where it literally says for turbo, you have to have both OpenAI API account AND Plus subscription minimum

and limits would be based on "tiers"

which immediately informs me that the plus subscription alone isn't enough,

you have to spend money on the API side as well

[honest jay]

I don't see Plus mentioned in that article.

[undone valley]

How can I get access to it?

Anyone with an OpenAI API account and existing GPT-4 access can use this model. The most recent version of model can be accessed by passing gpt-4-preview as the model name in the API. You can read more about the differences across GPT-4 Turbo dated models (such as gpt-4-0125-preview) in our developer documentation.

they didn't explicitly say plus

just GPT-4 access, which starts with Plus

[honest jay]

That seems to be about existing API access to GPT-4. The article is in the API category

GPT-4 access via Plus and API are separate.

[undone valley]

the title is literally "GPT-4 Turbo our latest model"

https://platform.openai.com/docs/models/gpt-4-and-gpt-4-turbo

[honest jay]

That's also API, unrelated to access via Plus.

[undone valley]

I understand what you're saying,

but like I said earlier, that's OpenAI's fault for not being clear

so, I'm going to reach out to OpenAI directly

[honest jay]

This article has been around for awhile, looks like it was recently updated to include clarification about the cap being lowered on Plus sometimes, but it helps clarify the lack of overlap between GPT-4 access via Plus and via API: https://help.openai.com/en/articles/7102672-how-can-i-access-gpt-4

[undone valley]

see, this article only says GPT-4 and doesn't specify the exact model

[undone valley]

I've taken it to OpenAI and once they respond I'll let you know what they tell me, sound fair?

[plucky timber]

Lmao the metadata in images

Just adds another step for people as people will mostly strip metadata to keep their privacy.

for those unaware you can just strip the metadata from the images, no need to sit there with forced "IM AI IMAGE" images.

[topaz otter]

Anyone else keep getting 'an error has occurred' on the mobile Android app?

[sonic folio]

Voice?

There have been some reports of issues with Voice.

[topaz otter]

No just regular 3.5 or 4

Web is working. Just not android app

[elfin birch]

OpenAI seems to be aware | https://help.openai.com/en/articles/8912793-c2pa-in-dall-e-3

Metadata like C2PA is not a silver bullet to address issues of provenance. It can easily be removed either accidentally or intentionally. For example, most social media platforms today remove metadata from uploaded images, and actions like taking a screenshot can also remove it. Therefore, an image lacking this metadata may or may not have been generated with ChatGPT or our API.

We believe that adopting these methods for establishing provenance and encouraging users to recognize these signals are key to increasing the trustworthiness of digital information.```

[cinder sail]

When do you think ChatGPT 5 will come along

[elfin birch]

You didn't get the GPT-5 update?

[waxen chasm]

I got it too

[cinder sail]

@elfin birch What tttt

Tell me can he speak to dogs?🥹

[flat sage]

I get errors regularly on android app

[tawdry ice]

Did you find a solution to this issue? @rain surge

[brisk path]

[cyan grail]

is gpt 4 turbo active now? The responses are consistently fast af

[sick pecan]

turbo means cheaper for them to run, not necessarily faster user experience

you know whats really turbo? having two chatgpt4s open at the same time

if one lets me down the other picks up the slack

[tawny crater]

JESUS CHRIST. I AM SO SICK OF "I'm sorry i cannot fullfill that request" DO THESE DEVELOPERS NOT REALIZE HOW ANNOYING IT IS TO WASTE LIMITED INPUTS JUST TO CORRECT IT'S HALLUCINATIONS?

[dense halo]

Chill

[tawny crater]

Ugh, go away white knight.

[sick pecan]

#looking-for-hate (also long walks on the beach)

[eternal rock]

<:book_icon:1171408210398289941> `` Rule 1 `` Be respectful.

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

[calm gulch]

DALLE seems down

Error creating image
I encountered issues generating the image you requested. Let's try a different request or explore another type of query you have in mind!

[undone valley]

if you treat it like a game, and take the opportunity to shame it as much as possible, you'll get your money's worth

[undone valley]

I only ever do images for specific things, but whenever I do, I request specifically in my prompt for 4-5 images to be generated, one at a time, so I get a few to choose from. That way when something like that happens, I don't have to waste another prompt to get what I originally asked for

[calm gulch]

It's just that it keeps saying this for me all the time now.

Fortunately I have a company subscription. I can fire as many prompts as I want.

[undone valley]

I mean, yeah it's nice you keep one Jefferson to yourself... but it doesn't solve the underlying problem. It's guided by everything you've given it up to that point.

[calm gulch]

Exactly.

[safe canopy]

Does any body has issue with ChatGPT now? I kept receiving: "Something went wrong. If this issue persists please contact us through our help center at help.openai.com." message

[sick pecan]

is working perfect in my region

[weary ferry]

Anyone please tell what's the details chat gpt 4 chat limit per hour?... I want to buy gpt4 subscription

[sick pecan]

#1184149552639516732

[tough mural]

🐟