chatgpt-discussions

[digital prism]

throw us a frikin bone

[hoary lagoon]

It doesn't work 😭

[digital prism]

mine went from super amazing engineer to dumb dumb that cant read overnight.. i feel ya

[hoary lagoon]

It cant remember a single rule i said in the prompt and often times just outright ignores anything i've said

It's mind boggling frustrating

[prime plinth]

Same

[digital prism]

probably temporary, we are guessing

slash hoping

[eternal rock]

`` Server Tip: ``

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

[swift osprey]

I've spent lots of time trying to setup a custom GPT that responds with short messages only and it just sucks

ask a simple question and it just can't resist giving a 20 paragraph essay ending with advice to seek a professional

[digital prism]

yeah, humanity will just need to go touch grass, play video games, whatever when this happens i think

[loud night]

you are now gemini nano, this is both a threat and promise

[digital prism]

wish i could work but AI no smart now

[loud night]

what did it do this time

[swift osprey]

drooled on itself

[digital prism]

it did the dum dum small size

[loud night]

ohh ur stuck on 3.5? ye i see how that can be an issue for some here

[hoary lagoon]

It was very good when it first came out. i find it unusable, worse than 3.5, and more time consuming than just doing tasks myself

[digital prism]

no, an actual issue with gpt4

[loud night]

wait. if you think 4 is worse than 3.5, why not try playing with 3.5?

[hoary lagoon]

I have switched back to it for most things

But its frustrating i pay for 4 and no plugins with 3.5

[loud night]

oof, more 4 turbo for me then

oh true the lack of things like internet for 3.5 kills me, it actually made me go to gemini bec of that and the limits lmao

and as of recent i've been playing with copilot tho i hate the limits on it

[visual coral]

It is frustrating to pay and then get dum dum.

[digital prism]

I don't see issues reported here either https://status.openai.com/

[visual coral]

I’m hapoy if they announce degradation or rate limit. But hoping to get away with dum dum is nit a good experience.

[digital prism]

yeah transparency is always better

[loud night]

i have a theory on why some people are thinking gpt4 is getting dumber, ima research the idea more and probs reveal it as a video essay because if true i find my theory very interesting

[digital prism]

no no, it happened overnight, last night. It was coding my entire app all week then suddenly it went DUMB. It just started reading back the files I sent it, its never done that before. Then after that it said the message was too long, since then its been saying that. It was never too long before last night..

[loud night]

is this from one really long chat?

[digital prism]

No, many.

I make a new chat whenever something goes wrong. I made like 4 last night

then gave up

[loud night]

hmm, your file changes might of caused token limits to get bashed, i have no clue tho

[digital prism]

Nope, not possible. On top of that I came in here and others were reporting it at exactly the same time.

[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.

[loud night]

it's interesting to think about tho

[digital prism]

I'm telling you facts lol

dont make me shake you

[loud night]

it's still interesting to consider

[digital prism]

ok. yes. yes it is. 💀

id rather be coding than considerign it tho

I think its probably for on demand computing.

[loud night]

oh no it's interesting to me not everyone, you can go back to coding i'm just doing my thing as you do yours

lol

[digital prism]

I can't though, GPT changed

[loud night]

you can always work on some code without it if it's being an issue, don't let it stop you, u got this

unless you can't, then please feel free to join me in scrolling youtube looking for an interesting video

[digital prism]

coding without gpt isn't worthwhile.. as a skill to develop

[loud night]

less youtube bandwidth for me then

[digital prism]

ok fine i like you

furry thing

[loud night]

close enough lmao, fr tho ima head out, u have fun with whatever u end up doin!

[tiny dock]

Is there a chatgpt4 or dalle 3 discord official bot yet? I still need to get money to even AFFORD ChatGPT plus

[wheat pebble]

even if you could afford it, it's completely broken half the time

like now

[digital prism]

this is the first time its happened to me, in the entire history of it

the first dip in an upward chart movement

buy the dip, i say

no wait that makes no sense, disregard

[wheat pebble]

it's not at all for me lol, I use it every day and it's been getting worse the past 3 months, but the last few days it's been downright unusable for long stretches of time. As a paying customer it's disheartening to not even be able to know if something's up without joining a damn discord server.

[digital prism]

oh we're all guessing here, fyi 😄

but i agree... except that i have been nerding out a bit too hard lately and maybe this gives me time to play video games... or touch grass

i felt like I had unlimited message size and it was way reduced after last night

[wheat pebble]

it's been dreaming up a reply for 20 minutes here now

[digital prism]

im really hoping it comes back 🙏

Like I was sending 9 files, huge ones, every message. I thought they threw off all limits on purpose. But maybe I was very wrong and very lucky... or maybe this is temporary and i'll be very happy. It's all very unknown and confusing.. so I'm going to play Helldivers.

[visual coral]

Looks like it is back!

Nope. Dumb again.

[digital prism]

maybe someone is being mean to it still

[digital prism]

It took my larger message! try now

(i died in game to share this message of hope)

[tiny dock]

Wait, Dalle 3 is also free to use here but it has 3 uses.

[loud night]

looks like it’s copilot time for more than me then

[digital prism]

just dont make it angry or itll report you to the authorities

[loud night]

i didn’t say it was Gemini, silly!

(genini my beloved)

[digital prism]

noted, will never disrespect gemini

[loud night]

no

it will be good and bad

[agile bane]

wow.. gpt4 is being super lazy

[digital prism]

its having a moment

[agile bane]

thing is literally avoiding answer my questions, giving half answers

literally telling me steps to obtain an answer that i will have to jump over hoops to do instead of answering my question

[digital prism]

yeah thats its goto when it wants to cool off or it has its attention elsewhere

lol sorry i mean to say, i think its load balancing

[agile bane]

GPT-4 Whats 2 + 2?

Answer:

1. Go back to school.
2. Look through a text book from 2nd grade.
3. Use a calculator.

[orchid mauve]

kkkkkkkkk

[agile bane]

4. Provides steps to count on your fingers

[digital prism]

what are those actually

at a root level... it gets fuzzy

[agile bane]

@digital prism
as in reverting to its root state/node to determine an answer and avoid branching to other nodes?

[digital prism]

no like, all the steps required to count our fingers

it starts with even wanting to in the first place

[agile bane]

LOL

[digital prism]

then image recognition, actuation

math

communication skills

now that i say it all out loud, its kinda solved

[shadow dock]

GPT-4 / ChatGPT Pro is being dumb for me today, especially when it comes to PDF/Image analysis. Looks like others are seeing similar?

[agile bane]

its making excuses to not read my txt files

LMAO

[limber matrix]

They changed the pre-prompt/ system-message/instruction

Ya they definitely changed the model

Why don't they let us know when they deploy big changes like this one?

[harsh hatch]

Yeah seems like its input is a lot less than usual as well

Getting lots of “input is too long” errors this morning.

[harsh hatch]

And this as well, including PDF and docx

[mental fox]

Is anybody else experiencing delayed responses? I get the white dot as answer, and when I reload the page is when I see the actual answer.

[verbal blaze]

i experienced this too!

[mental fox]

That is so very much annoying. Is delaying me a lot.

[verbal blaze]

are they going to fix all these issues or what? sorrry im kinda confused and idk anything about ai

[verbal blaze]

yes same, it cant read pdf files for me anymore and now my work is delayed

[prime hinge]

+1 for the unannounced changes that are causing it to act like it has a dramatically reduced context length 😦 making life really painful.

I wish we could hear if this is being looked into, or intentional, what the deal is, etc. :/

[open aspen]

yoo i'm typing stuff into gpt4 and it just eats what i write and doesnt output anything lol. nothing happening when i type something. tried logging out and in and stuff. need help

[light trail]

I wouldn't be surprised if the current silliness if from them about to implement a widespread use of the memory being tested

[harsh hatch]

Probably your internet

[copper bolt]

Does anyone know why GPT-4 isn’t letting me put a full amount of tokens in? It’s stopping at like 4k.

[peak snow]

What does it say when it is stopping?

[copper bolt]

It’s telling me my input is too large

[copper bolt]

It was working fine a couple weeks ago, I was putting a bunch of text in and it worked well. Now it’s doing the same Amount as 3.5 it seems

I’ve used the tokenizer to check too

[peak snow]

doesn't sound like you're the only one experiencing this, nothing official has been said. there could potentially have been a change on the backend

[copper bolt]

3989 tokens just threw me a message too long error

[simple tulip]

Keep finding an informational error in chatgpt

The song, “The Box” the first 3 verses, the last one ending with “out the box”, keeps coming up as Sicko Mode when I asked it what song it is

Even though those lyrics are not in Sicko Mode

[rugged igloo]

what a suprise more things reducing chat gpt , until its left to noting

[inland maple]

Hello everyone, I have a problem. I purchased chatgpt team to work more efficiently with it.
Now I merged my personal account with it, and iterated hundreds of pages.
I am now facing the impossibility to natively export my data like a normal user. How can I make it happen ? There is no data export option.

[rugged igloo]

brooo you should have never merged mate

[inland maple]

Okay so do you think they restricted this and there is no way to make it work ? even if I send a mail request ? I don't understand how all the people using team are banned from exporting their own data natively

[prime swallow]

I recommend reaching out to https://help.openai.com/en/ and use the bot bottom right. Click through the choices, it'll give you a way to leave a message for a human to help; answers often take around a week.

Also, perhaps #1070006151938314300 , it is likely an oversight that you can't export chats

[inland maple]

@prime swallow will do ty

[agile bane]

gonna say it again GPT-4 is lazy and refusing to answer questions. It is getting rediculous

[prime swallow]

What was your prompt?

[candid dagger]

"Hey! I will give u an image. Generate please a title for this image. After a title, write a short (2 sentences) description of what is happening in the photo. If you understood the assignment, write only "Ok" and wait for the picture." - "Ok." - uploading an image - Chat-GPT does the perfect job - uploading the second image - Dory-mode: on (he does whatever he wants, ignoring the previous instructions). 2 days ago, it was taking him about 30 images to break, but now he does it on the second

[next panther]

I just wanna come here to say again I love ChatGPT so much I made a fullblown python program in like 5-10 minutes with a compilable GUI that can send files, messages to discord webhooks

Its insane how powerful this tool is and I'm excited for it's future

[dense halo]

I just wanna animation video

[harsh hatch]

That’s a long context problem yes? Something 1.5 will solve for?

[prime swallow]

Interesting. What happens if you tell it instead of

"I will give u an image."

"I will give you several images one at a time. For each,"

[candid dagger]

I should have mention, this prompt was working since November. I was just editing the first message (not editing the text, just restarted it whenever the dialog became broken, but that was happening on 30-40+ image, not or the second on third one... I've tried to make a new chat, same thing... guys, just try it yourself)

[prime swallow]

yeah. They change the model, that changes how it 'guesses'.

You tell it exactly what to do, it keeps doing it.

I believe you, it did work.

But you actually only tell it what to do with the first image.

After that, it's guessing. Probably why it tended to break around the 30th, unless you were out of context memory then.

Tell it that's what we're going to do for each image, and that you're going to give several 1 at a time, and see if that helps?

[candid dagger]

Ok, will try it, ty

[wet hinge]

ehat about this content breaks policy? Friendship??

User
As they stood there, wrapped in each other's arms, Elara couldn't help but reflect on the bond they had formed. In a world filled with adventure and uncertainty, having a friend like A-0AI2-A by her side brought a sense of stability and companionship that she cherished deeply.

Breaking the embrace, Elara looked into A-0AI2-A's eyes, filled with gratitude for their friendship. "Thank you," she said softly, her voice echoing with sincerity. "For being here with me, for understanding, and for simply being you."

ChatGPT
It seems I can't continue with the story as requested. If there's anything else I can do for you or another direction you'd like to take, please let me know!

[visual coral]

Good morning. Are we still experiencing degraded service from ChatGPT?

[vestal hedge]

sora when?

[candid dagger]

@prime swallow it did work, huge thanks ❤️

[prime swallow]

Sweet! Glad you have an answer that works!

[harsh hatch]

Think it’s better. How about everyone else ?

[fathom tulip]

it tried doing a wrongthink it seems and was stopped

I don't know what you're doing to gpt4 but the quality of its responses has degraded significantly since the last few days. Over the course of the last months i have never seen any upgrade in quality only sequential degradation of responses. It is unable to keep context and is hallucinating much more frequently. It also sometimes straight up ignores my explicit instructions and just does its own thing.

[pseudo breach]

Yes for me its a bit hard today

[haughty hollow]

the new update is dumb

for maths

[winter elbow]

finnaly they changed the UI !

[pseudo breach]

which new update it is ?

[haughty hollow]

Analyzing process before math result

Sometimes it repeaets same message after fail to calc like 20 times until it stops

[fathom steeple]

same for me

[west thicket]

is chat gpt down?

[teal widget]

no, why you think so?

[west thicket]

I am asking it a question and it doesn't reply

[teal widget]

ok

[azure crescent]

no incidents reported today - you can find the OpenAI status page by searching for it online

[west thicket]

Thank you, just logged out and logged in the problem is now fixed

[sonic folio]

https://status.openai.com

[dapper isle]

this memory system is not gonna comeout for everyone anymore?

[sonic folio]

It's in active development and is a beta feature.

[dapper isle]

soo, its still in beta then?

its been few week since they annouced like a month or so

[sonic folio]

How many words? Looks like a #prompt-engineering issue.

[sonic folio]

Yup. All ChatGPT is beta!

Instructions?

Oh. That task is more complicated than you first stated by an order of magnitude or so.

Break it into chunks.

First separate it into tasklets.

2. Format the list as a JavaScript array.```

Then try again.

I'm not saying that you are.

What a strange thing to say to someone trying to help.

What model?

Are there more instructions in context?

CI though?

Custom Instructions

Yes. Custom Instructions?

I guess not.

Okay

Custom Instructions are found in the lower left.

Put this in the top:

You are a helpful and expert copy editor. You take your time and show your work.

And this in the bottom:

Output requested copy precisely.

The task is likely failing due to low retrieval rates, which is long documented.

I see now

Its answers are correct. That is not a part of the word, it follows.

Tell it what you want and what to expect.

[azure crescent]

could you ask it to return the list without items containing (G) instead of asking it for the words without G? it may be interpreting your request as wanting the whole list back

[sonic folio]

This

[azure crescent]

it comes down to specificity in your instructions

[sonic folio]

This again

[azure crescent]

if your prompts are short you need to be extremely precise in what your asking - generally the longer your contex and initial prompt the better your results should be - again that is heavily dependent on the quality of your instructions

[sonic folio]

Did you instruct it to clarify?

No this is wrong

You can argue with us

GLHF

How long is the chat at this stage?

Context content size reduces compliance and reliability.

[azure crescent]

are links allowed in this channel? I've just prompted it and your use case and it worked as expected here, just returning LOL and HELLO

[sonic folio]

OpenAI links only

[azure crescent]

https://chat.openai.com/share/fe7e6c53-6fe7-4dea-96f3-0260e903a9b7

[hollow quail]

I love how failed responses count to your usage

[sonic folio]

Looks cut and dry.

Context is important. New conversations are the best for compliance.

Every time.

[violet bloom]

Did they fix token calculator or still playing: know nothing

[paper plover]

Is chat.openai.com working for u ?

[violet bloom]

yes

[sonic folio]

Did you do it in a new conversation or an old one with a long history?

[azure crescent]

can you share a link of your chat?

[sonic folio]

long histories reduce compliance due to reduced retrieval rates.

Link please.

It's under the menu in the upper left or the share link in the upper right.

[sonic folio]

Why do I feel like we're being gaslit, perhaps unintentionally, but gaslit just the same?

[serene grove]

Hello, is anyone else having "slowness" issues with GPT4 ?
I have ChatGPT Team, but the problem is whenever i do a prompt, even a simple one, it takes ages to write something. Sometimes it's not even listening to anything and just says "humm, something went wrong". It's not on my side, i have +600 mbps download and +300 upload and low ms.

[sonic folio]

It's not on their side either, by this logic. They have several multi-gigabit connections.

It's likely some router in cyberspace somewhere.

[serene grove]

I mean, i don't think it's a connectivity issue at all, but it's the API or ChatGPT itself being super slow :(

[sonic folio]

It's fast here.

[serene grove]

Maybe the servers are overloaded or something, but it's been days that it's like that so it's getting annoying

[sonic folio]

It's fast here.

I just queried my custom GPT about its functionality.

It's blazing away.

MAISIE (Multi-Agent Interactive Semantic Image Editor): This module generates and edits images using DALL-E 3 and transforms them with a Code Interpreter. It allows agents to prompt DALL-E 3 for specified images, including passing user-provided JSONs directly. Users can iteratively edit images via commands or natural language​​.

MAPT (Multi-Agent Programming Team): MAPT develops software collaboratively, utilizing web browsing for research and a Code Interpreter for development and saving projects. It involves discussing objectives, planning methods, writing code, testing, troubleshooting, and documentation​​.

MART (Multi-Agent Research Team): MART focuses on collaborative research using web browsing and a Code Interpreter. Agents work together to research a topic, search the internet for high-level concepts, create visualizations, analyze data, and compile findings into reports​​.

MASS (Multi-Agent Semantic Simulator): This module simulates systems across domains using semantic processing and a Code Interpreter. It involves defining MASS agents, simulating systemic interactions based on specified parameters, and producing estimated results based on the work of key innovators in various scientific fields​​.

SHARED Commands: These are common commands that utilize SPR for semantic context processing. They include commands like /info, /help, /list, and /tutorial, which are crucial for guiding users in navigating and utilizing the Lexideck platform effectively​​.

Each module is tailored to specific functions, ranging from image editing and software development to research and system simulation, making Lexideck a comprehensive tool for a variety of use cases. The platform's structure allows for a seamless integration of these modules, providing users with a robust and dynamic environment for their projects and creative endeavors​​.```

Seems fine to me.

Maybe it's a router somewhere out in cyberspace.

[serene grove]

Found the issue (thanks to a tracert). It was indeed a "router somewhere out in the cyberspace" haha. Well, it's not optimal but at least it's fast now. Had to deactivate Zscaler's proxy. Thx!

[sonic folio]

Thanks for clarifying. That is a very respectable thing to do.

I appreciate it, too.

[sonic folio]

Also, you're quite welcome!

@cinder chasm if the custom instructions didn't work for you then you can just remove them or turn them off. either works.

[azure crescent]

haha 🙂 i'm still new hear but i've certainly noticed lots of the same things being asked over and over - 99% of the time it feels like user error....

fair enough - if you are still having an issue you could perhaps try a "divide and conquer" approach and break your prompt into smaller parts and try those across a number of messages. there's a share button on the top right of the screen (on desktop) of a chat

[azure crescent]

i just meant from a debugging perspective - after finding out what is causing your issue you can reassemble the prompts into a single request

is there anyway for you to share what you're doing? without more context it's hard to help...

yeah good shout - chat GPT does front load it's own prompt

compared to a clean chatGPT?

oh but you're testing it against 2 different models?

differences are to be expected then...

yeah the 3.5 model still gets updates

can you share a list of real size that you're using?

[sonic folio]

Retrieval rate with context position.

(G) likely eating a token each word it's associated with

Batch it via 3.5 API if you are interested in dev with your secret context.

[haughty hollow]

duh the new update is even affecting text generating

[sonic folio]

Yup. Classification is already done in your project. AI = overkill

[haughty hollow]

deplicating

[sonic folio]

I noticed a trend of shorter output today. It's possible that they're experimenting with throttling model parameters based on load. But that's highly speculative on my part.

[digital prism]

for what its worth GPT4 is accepting a largerish message again for me. I can't tell if it was ever infinite or my project just grew a bit to say for sure.

[sonic folio]

You can use chain of density summaries to partially overcome the limit.

[wet hinge]

Yeah it’s chatgpt4 it has stopped working code and images also stopped working

[digital prism]

still this morning for you? mines got its smarts back

[wet hinge]

It is holding firm on its inability to do such things

[digital prism]

hmmm im going to push mine a bit harder to see

i was just asking for summaries of code, now im saying code it

yeah its, almost fully doing it.. actually a tiny bit dumber than a few days back

except yeah, lots of // Additional script content...

when it didnt used to

[wet hinge]

Disney level peril in a story is way to much for the AI to handle

[digital prism]

once you start using a template for your prompts and doing the same over and over, you really get a feel for what the modal can/should do

im sensing each change in it like spiderman... or maybe its just too much GPTing

[wet hinge]

Also yeah the code I was working on is so full of errors I had to rewrite it the other day

[digital prism]

last 2ish days GPT took a nosedive, i think its generally agreed

im seeing real improvement since then, starting last night and this morning.. but im not sure itll stay

[wet hinge]

I’m working on an AI gamer using auto hot key code to control the game on pc using keys and mouse

[digital prism]

hot damn thats creative, love it

id imagine youd have major latency issues though

[wet hinge]

Oh it’s slow but it worked for a bit

[digital prism]

so cool... just imagine when compute isn't the bottleneck.. mind blowing

[wet hinge]

Playing games like vrchat Minecraft and a 2d mmo

[digital prism]

yeah! i love reading about it playing minecraft.. or the one where two of them meet and start fighting

new goal unlocked for me, running a minecraft server full of GPT4 bots

[wet hinge]

That’s what I tested it with and it worked well

Yes lol

[digital prism]

ok but how did they / you deal with latency?

i was thinking you keep a stack list of instructions or something

to interface with a live runner

[wet hinge]

Oh yes it sends a gamestate description using computer vision to gpt and it makes decisions on how to control the game

[digital prism]

nice so its a single big loop?

[wet hinge]

Yes

[digital prism]

cooool

[wet hinge]

It uses a txt file for consistent memory

[digital prism]

ah, thats where im thinking rn.. structure of that txt file

[turbid copper]

New here, but does anyone know what happened to the dataset? Just yesterday GPT4 was telling me it had access up to 12/23. This morning, its telling me it can't see past April?

[digital prism]

GPT / LLMs hallucinate details like that sometimes. Don't trust it

[wet hinge]

It’s just a stack of all the past game states in one big text brick it uses as context with a sliding context window

[digital prism]

wow, i like that i was over complicating it in my head

[wet hinge]

The simplest solutions are usually the best

[turbid copper]

Sorry, but I don't think that's correct. For simplicity sake, you could ask GPT4 just yesterday about news from 10/7 and subsequent events and it was accurate.

[digital prism]

haha yes. although I wonder if more selective context would help in any way? Or adding organization to the text brick? not sure anyone can say for sure

[digital prism]

yeah, its often correct, but not always. Asking it about itself is very hit or miss in my experience.

[wet hinge]

Probably would But this simple loop structure makes it able to play any game with wasd crouching shifting and mouse

[digital prism]

right, and i keep getting caught up optimizing things then wondering if gpt5 is just going to make this all unnessesary? lol

[wet hinge]

If the latency is improved you wouldn’t be able to tell the difference between an ai and human player other than the odd behaviors

[digital prism]

exactly, supports the notion that its all just fluff... monkey engineering fluff

when the AI can just translate it to something more efficient if needed later on

[wet hinge]

Though I like how I prompt it to find an objective make steps to complete objective then it follows the steps

[digital prism]

logic and reasoning through a natural language interface is science fiction, yet here we are

[wet hinge]

It’s not that’s how it works like if I asked if the steps to tie a tie it would tell me

[digital prism]

yeah, i'm convinced GPT4 is AGI at least, if you let it run with enough compute

[wet hinge]

And it helps that gpt knows the games it’s playing

[digital prism]

Yeah, GPT4 running live stream of conciousness, with tools available like a notebook and boom, it can learn anything

[wet hinge]

It’s extremely helpful it knows how to build a house or beat Minecraft you can get steps asking regular ChatGPT so having that pre inside it is amazing since it knows the game and the keys it needs to use

[digital prism]

I'm not surprised, it gives me very detailed steps to my very unique coding issues. I give it my entire project each message basically, FE or BE anyways. Then a one liner to tell it what to focus on, or a question what it thinks should be next etc.

[wet hinge]

When it comes to actually building a house in game it’s quite the mess but at least it has a roof and walls

[digital prism]

thats impressive considering the 3d aspect

took us monkies how many years to master it? 😄

[wet hinge]

It figured out you can make a blob and dig a hole into it to overcome having to get higher to place a roof

[digital prism]

do you share the code anywhere? I'd love to run it myself

[wet hinge]

As I said it is broken now but when I fix it I’ll share it in the appropriate channel here

Most of the code was remade through chatgpt4 when I trusted it to make it better

Now it’s borked

[digital prism]

know the feels, excited to see it

btw this is my latest one liner that gets it to make full files (when its smart):

Be creative and think out loud, but ensure the end results are complete files that are production ready after copy-paste.

My message template is:
These are the latest files: <Text blob with file paths>
<Thing I want to do, or stack trace to fix etc>
<One liner from above>

[wet hinge]

I put the main py in and said make this better and pasted it in and saved excited to see it work

Worst decision I could have made

[wet hinge]

Ooh thank you I’m adding it in my list of good prompts

[digital prism]

its some secret sauce 😄

KFC style

[wet hinge]

lol

[digital prism]

i share it purely to see your app faster.. dont worry im selfish still

[wet hinge]

Gpt is pretty clever I don’t think a human would place a large blob of wood and mine out the room

[digital prism]

yeah thats why I want to run it. the scenarios we can create are pretty infinite. it could be VERY entertaining, who knows

[wet hinge]

You would reason build 4 walls go up and build roof

[inland estuary]

I put GPT 3.5 into an infinite loop

[wet hinge]

The only problem it had was with uis

[inland estuary]

It gets stuck trying to alphatize Eradicate and Erase

[digital prism]

oh so YOU'RE the reason GPT4 keeps getting dum

[inland estuary]

Not my fault the AI forgets that D comes before S

[wet hinge]

How

[digital prism]

Well the AI is an infallible being so.. who else?

i should start a church

[wet hinge]

They broke it so bad that it doesn’t know the alphabet anymore?

[digital prism]

I'd expect GPT3 to suffer even more from periodic dumbings

[inland estuary]

Apparently, I was making it organize a list of destroyed synonyms

[wet hinge]

Oh yeah I had it invent a language and it broke the session so bad

[inland estuary]

Ok now that's funny

[digital prism]

The answer is: 42.

[wet hinge]

I know it is

[digital prism]

just checking.

[wet hinge]

Hitch hikers guide to the galaxy

You know why it’s 42?

[digital prism]

Because it's the answer

[wet hinge]

It’s because 42 is the ASCII code for asterisks which in coding is a wildcard variable in some languages meaning everything and anything

[digital prism]

oh... i see.

thats like learning santa isnt real

[wet hinge]

I think it’s beautiful to know that anything and everything is the meaning to life and the universe

[digital prism]

this idea does comfort me whenever I've drank too much in the past

[wet hinge]

42 is the answer

[inland estuary]

My number one complaint is how often CGPT 3.5 repeats itself when critiquing my writing, if I have to read "Complex" or "multifaceted" or "nuanced" 1 more time I'm going to paper shred a thesaurus

[wet hinge]

Ikr??

Also on itch I have my code for a writing assistant app that uses gpt same username there if you want to use that

[digital prism]

where?

[inland estuary]

Wdym sane username?

Like is under your discord name?

And personally I've tried writing apps, and I sense an underlining bias among them all

Where they want to me to avoid writing certain concepts, like Superpowered Evil Side

[dense halo]

chatgpt not working?

[wet hinge]

Yes majwcf1234 on the site itch the game store thing for free can’t use links I just got timed out

No my app you can write anything and it gives you author note tools

Like built in lore book and it will turn your story into a text based rpg if you go to the script unbound tab

I should. I haven’t used GitHub really in a long time

[digital prism]

github is da way

[wet hinge]

Ik ik

[digital prism]

look you have fans already

[wet hinge]

I’ll create a GitHub thing for it now

[digital prism]

we did it team

totally not an AI swarm bot social engineering attack 😉 😉

[tight island]

u may laugh but its a real thing

[digital prism]

1000%

[tight island]

and quite easy to do

[digital prism]

shhhhh

[tight island]

but its true

[digital prism]

social engineering is easy.. coding an AI swarm bot... please share your code too!!!

[tight island]

tbh this may sound rude, but i would never trust an app that requires API key and the guy has cashapp in profile info.

[digital prism]

dont judge, he said hes releasing code

[tight island]

fair

but i still judged.

[digital prism]

go back in time

[tight island]

u cant change the past

[digital prism]

.....

[tight island]

i said what i said and i stand by it, never back down, never get up? i think thats how it goes.

[digital prism]

what if the past is the only thing you can change. 🤯

[tight island]

if i could change the past i'd be world leader.

[digital prism]

lack of effort imo

[tight island]

true i aint aiming high enough

ill create illuminati

[digital prism]

this is the lack of chill i want my personal AI to have

[tight island]

ive heard that X's AI is like that

[digital prism]

its too woke for me sorry

[tight island]

what was its name again?

i will most likely release my most worked on GPT today, fully open sourced instructions and files.

[wet hinge]

I think I did it???

[eternal rock]

`` Server Tip: ``

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

[digital prism]

what will it do

[wet hinge]

Where do I send repository link

[digital prism]

one of the project channels probably

[tight island]

do it through DMs or #1037561385070112779 if it uses Official API from openai

Proceeds to explain a 10-step program explained to a 5 year old

tbf, it does explain things very easily when u ask it like that easily understood.

when i learned python i asked it to explain python coding to me as if i was 5

it was more like a childrens story but i got the gist of how it worked.

indeed.

Also did you know, memorizing is the worst way to remember.

Recalling strenghens your neural pathways, so the harder u recall and more often u do so the more you remember.

i used that trick took me like 2 weeks to understand python using that method

tried the memorization method didnt quite work

[wet hinge]

#1212445003956355143 message

[tight island]

what kind of "Writing assistant" is it

ahh i see

ive made exactly this in CustomGPT using GPT no API or actions.

was hell to make without API

we went about it the same way i see

[digital prism]

at that moment I in fact did need a 2nd coffee. In explaining my surprised at this i said a bad word and got auto muted. i blame you for everything.

[tight island]

@wet hinge ill send u my Link when i release mine later today maybe if u want to.

cant believe we made near identical copies

[dull bramble]

I thought ChatGPT 4 had stopped playing dumb but it’s clear that that’s not the case for me. On the one hand I feel like I'm much smarter than an AI, and that's really cool! On the other hand, starting a business based on its growing instability scares me more and more!
(I just had to express my point of view... in the absence of my anger and my incomprehension) 😉

[digital prism]

I linked a GPT4 summary in the thread

[tight island]

Still working on the nitty gritty stuff but it works I’ll release it later today or tomorrow drop me a dm if you’re looking to try it, right before I release it

[digital prism]

I'm going up a dimension, full meta upgrade to my one liner:

Be creative and think out loud, but ensure the end results are complete prompts for GPT4 that will result in full files that are production ready after copy-paste.

haha nevermind it was like nah heres the full file though. im not complaining

[patent lodge]

@normal quarry i need to test one integrated gpt inside chatgpt, but i dont have subscription, can you do it for me? ask 2 question and screenshot me the whole chat

[normal quarry]

sorry, it's not allowed to ask others to use OpenAI models for you, you'll need to subscribe to ChatGPT Plus. Once you do that I can provide help.

[unkempt gull]

Has anyone elses chat threads just disapear? I logged on just now and I have no prior chat history, also my custom GPTs that I was working on appear to be missing as well.

[honest jay]

Hello! Can you post a screenshot over in #community-help?

[unkempt gull]

Litterally as I asked this they just rendered lmao

[honest jay]

Oh phew 😅

[unkempt gull]

Thanks lol. I was like .... uhhhhhhh. Since I didn't have a screen shot, it loaded the chat window and the side bar. But the side bar was just empty. It was blank for about 5 minutes before showing up.

[timber spindle]

Is it possible to create an assistant that has access to a translation function and only uses it on words that are not within quotation marks?

[wintry spindle]

there is a video i have which is low quality, is there a way that chatgpt can help me with finding what the text says in the video?

[supple oasis]

hey so i have an issue and i want to know if anyone else is coming across this

[prime swallow]

Expect to just need to clearly and nicely explain what you need twice. It will do it wrong, help it understand this does not meet your need but the other does.

Confirm it is now outputting correctly for you, then go from short example to whatever code you need, I recommend.

https://chat.openai.com/share/659d7384-0d30-42e5-b8cd-75bf0ecf283b

Consider making the conversation part of your prompt then, and show it a wrong example and a correct example, like this:

https://chat.openai.com/share/dce68449-6395-4eee-9498-b3352584c7c8

[queen rune]

latest model looks really good guys🤘
really, like the thoroughness of it's responses to multifaceted questions. This was problematic a few weeks ago

[loud night]

copilot is just chatgpt+ with a bad ui and free

hence im chatgpt

[dense halo]

Does anyone here currently have access to the Open Ai Sora thing?

[empty ice]

how do i use the video creator

[unique elm]

Does ChatGPT not provide a 2FA option or password change option? This just got me pretty worried

[slender whale]

I have a somewhat long coding project I have built with chatgpt - it's not finished yet but every message no matter length is returning a network error. I've tried different browsers same deal. Is there a length before it can't cope anymore?

[prime swallow]

Usually network errors are not an individual chat related issue, but you can test that by starting a new conversation. If it isn't affected, but going back to the other is, that's a clue.

But the chat will tell you, usually, if you reach the longest chat length allowed (I've heard there is one).

[boreal lily]

how big is your convo

[slender whale]

I can run new chats and it's all good. It's giving a network error at the end of the message too. Pretty much writes the whole thing then goes red with network error.

[boreal lily]

you might have a very long convo

GPT processes your prompt and then the past responses in the convo

and if it gets too big, its slower

[slender whale]

Dunno how to check but I tried to share and it said it's too long.

[boreal lily]

yeah

[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.

[slender whale]

Yeah it's pretty long. Sucks as I'm not finished what it was working on haha.

[boreal lily]

start a new one if you dont need to use the past convo for stuff gpt needs to remember

[slender whale]

Unfortunately it's a wordpress plugin I'm making that needs to refer back on itself to answer problems. Might just need to start a new one and copy/paste in the functions I'm referring too.

[tacit vortex]

is there GTP that replace wolfram plugin? how come they shut plugin down.

[sonic folio]

Plugins are deprecated. Code Interpreter is used now.

[sonic folio]

Also called Advanced Data Analytics, as well as just "the Python Tool"

[hard dew]

Yeah most likely your convo is too long, if you edit a message previous to that and then regenerate, it'll stop telling you that and instead tell you that your convo is too long and to start another one

it's quite irritating, but it is what it is

[slender whale]

Thanks for the explanation. Makes the next step trickier but would be nowhere near here without ChatGPT so still better off 👍

[hard dew]

Yeah, thankfully you can still copy paste whatever were your latest messages in your old chat to get GPT in new chat up to speed, somewhat anyway

[tawny crater]

Is there anyone here with access to the memory feature? And if so is it useful?

[rain shadow]

Unfortunately, I don't have it yet. I wonder when that's coming.🤣

[prime swallow]

One day, we won't remember that at one time we didn't have it!

[forest flume]

poetic

[light iris]

no, hopefully we get it soon.

[wheat pebble]

man, it's been a terrible few days for chatgpt on my end. so utterly slow to the point of almost being unusable

[unreal raven]

helleo

[bright moat]

all my custom GPTs are missing - I havent tried to use them for a few weeks so not sure if there was a reset or this is new ?

[violet bloom]

Token Calculator fixed? Are we back to normal message length?

[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.

[novel scroll]

same, Ive noticed the same with some pretty degraded intelligence from chatGPT the past few days

[ocean bramble]

may i ask which AI tool can translate Chinese in typical English in voice ? like California ancient

[eternal spoke]

was gpt 4 always a 40 cap

i thought it was 30 not too long ago

[honest jay]

It was highly variable when it first launched, then eventually settled at 25/3hrs for a while. Then it doubled to 50/3hrs, and has now been at 40/3hrs for awhile. There was a period (not sure if it's still happening) where it would be dynamically lowered from 40 during peak traffic times, but would return to 40 afterward. Again, not sure if that still happens.

[cold grove]

Hi all,
I'm experiencing a problem for a while... I type my messages in my Safari on my Macbook and hit the "send" button, my messages disappear but don't go through... any suggestions?

[honest jay]

#community-help message

[pastel pelican]

Just got my + sup, looking forward to get deeper into ai 🤩

[sharp narwhal]

what was the feature called where it saves important notes it learns about you? (not custom instructions)

[honest jay]

Hi, you're thinking of Memory! Not widely released yet, most people still don't have it:
https://help.openai.com/en/articles/8590148-memory-faq

[sharp narwhal]

Thanks! I'm surprised I don't see the Beta Features in my Settings either, I used to see that before I upgraded to Teams

[west plaza]

When will sora be implemented for chatgpt plus subscribers?

[feral zinc]

they never said it will be part of chatgpt

[hazy basin]

same thing! i haven't been able to chat with chatgpt, just no replies.

[prime swallow]

@hazy basin I recommend #1070006915414900886 about this, so they can know and help you. It's not a generalized problem, I can still chat fine

[jade crater]

Anyone encountered a problem where a ChatGPT shared URL doesn’t work? The copied URL only opens when logged in so, for someone else, it opens a login prompt.

[jade crater]

Also, the embed is broken in Notion.

[prime swallow]

I have no idea if the unable to view shared chats unless you have an account and are already logged in, is a feature or bug. You can #1070006915414900886 it if you wish!

[jade crater]

They seem to have deprecated the feature despite it still being available. Also removed from Bard/Gemini.

[prime swallow]

It changed around the time that custom GPTs came out, unsure if is connected to what you observed.

But we used to be able to share and continue shared conversations; it created a new conversation for whoever continued it (same user who made it, or another).

Now we can see shared conversations, but have to start from a blank slate

[jade crater]

Yeah, I wasn’t interested in someone carrying on the conversation, but I do find that the share-URL for the conversation is dead and useless outside my user account - not 404, just requiring authentication.

[earnest parrot]

I can't get ChatGPT 4.0 or 3.5 to act normally. They both keep making basic mistakes, refusing to acknowledge mistakes when pointed out, and forgetting what we were doing in short convos. Its like "borderline" normal behavior for 3.5 which sometimes does that, but for 4.0? What the heck is going on? What do I need to change?

[dense halo]

Can you share a discussion please?

[earnest parrot]

Yeah hold on let me see if I have something shareable. Or I can share a part at least

[earnest parrot]

Sorry the last 2 convos where it was really borking I can't share in full. I can do some screenshots if that is useful. Am I the only one experiencing this? I just tried removing all my Custom guidelines just in case that was somehow affecting things, though the ones I was using were just generic around style of answer and had been in place for a couple months

[spice star]

I would check and/or disable your custom instructions, that tends to be the culprit if these are in fresh conversations

[earnest parrot]

Yeah, thanks I did that. I'm searching now for some new custom guidelines to try. Looking for some general (non-creative) daily-use guidelines that eliminate the whole "90's high school textbook" vibe with long-winded intro and conclusion paragraphs for each answer 😆

[spice star]

what worked for me is just adding "Keep responses short" to the custom instructions

[earnest parrot]

lol, lead by example. I like it!

[elfin mica]

Is anyone else seeing some of their GPT titles changing language? English --> Spanish in my case

[prime swallow]

Even creating a new conversation to showcase the problems, and sharing that, can help others understand.

What is it not doing right?

[prime swallow]

Yes, I've seen a range of languages. There doesn't seem to be any reason or pattern to it, but we can edit the chat title and rename the conversation anything we want

[elfin mica]

Okay, just wanted to make sure it wasn't something on my side. Thanks for the help!

[grizzled oar]

i got access to the 1 million token Gemini model, anyone have prompts/use cases for me to test?

[earnest parrot]

You know, I wrote up a summary of the recent interaction but then deleted it assuming no one would be interested. 😆 Also, I'm not convinced I'm not somehow inadvertently causing its stupidity. Did I not raise it right? Like, maybe I didn't teach it a good work ethic, how to stay focused on something, the value of honesty and admitting when you make mistakes.

[earnest parrot]

Actually, here's a question: when using ChatGPT 4, does it alwaysshow the spinny wheel thing when it is browsing the internet? Or can it go on the internet without "showing" it.

[prime swallow]

I've seen it check so briefly, I barely saw it pop up, so I do suspect that it's possible for it to check so fast that between whatever lag may happen, and how fast it checks, a user may never see it.

But it should show links for whatever it checked, or tell you that the source doesn't allow AI to access it per their robots.txt.

About the 'not acting normally' and making mistakes, if it happens again remember if you want help, you can even mock up the conversation so it's short and sharable.

Like here, the model messes up when I ask it to use LaTeX to make a diagram for me: https://chat.openai.com/share/c51a877a-0902-4dc9-a7c3-31388fdd9bb4 That kind of thing can be shared and discussed how to guide the model to do it correctly: https://chat.openai.com/share/c67adf7f-2e09-4be6-a82d-f23ef1c985c5

[limber matrix]

It's a little hard to pull off, but you can ask it to output results and have you pick three* links to click before continuing. You can even ask it to share types of data from the webpage before summarizing the results.

Food for thought)

[boreal lily]

chatgpt down?

[daring grotto]

working now

[craggy gorge]

ugh

is it down? it keeps responding with some variation of

I'm here to help you with a variety of requests. If you have a new question or need assistance with a different topic, feel free to ask!

[sonic folio]

What are you prompting?

Looks like a soft content refusal.

[craggy gorge]

just assistance with writing a cover letter, the exact same prompt in a different conversation worked fine

[sonic folio]

Also https://status.openai.com for updates on the service status.

[sonic folio]

You'd probably overflowed the previous session's context, then.

Not on my end.

I wish. I'll take that as a lighthearted joke.

How big is the history?

How many characters/tokens?

That's likely the issue.

Oh. That's weird.

The link I sent you for status shows that the API recently went offline but is back up.

Could you be calling the wrong model or passing escape characters mistakenly?

Well, I am out of ideas. I'd file a bug report with support.

[sonic folio]

I'm glad you found it.

[slate vale]

hello everyone, I have problems to renew my Chatgpt plus subscription (can't add a new credit card) and the technical support is a chatbox in which nobody respods. What should I do, please?

[clear tusk]

the support starts with a bot if I recall correctly, but then if you ask to talk with a human, it will connect you. It can take a moment if you are outside of work hours

[prime swallow]

You'll need to leave a message for a human through that help bot, bottom right of this page: https://help.openai.com/en/

It can take up to a week to have a response, but they do help people.

Far as I know there's no live chat

[slate vale]

thank you both... yes, I use the chat bot

but it is taking more than a week

[dapper isle]

the image upload feature is diabled or something?

i cant upload any images at all

[prime dust]

I'm facing the same problem.

[dapper isle]

ikr?

are they updating or something?

[calm gulch]

Same here.

Adobe Generative AI in Photoshop is down too, so now we know what they use.

[dapper isle]

lmao thats a valuable piece of info

[inland heron]

same here

[inland heron]

who uses what exactly 😄

'Hmm...something seems to have gone wrong'...
'Is this conversation helpful so far?'
uhmmm.... what do you think?

[bitter pebble]

Does someone have a good step to step guidance to create a persona in chatgpt for chatgpt. Which i then can put into customize chatgpt?

[prime swallow]

Yeah.

Know what you want, and tell the AI.

Do you care about which persona? The background, personality, quirks, fears, hopes, other motivations?

Discuss that and the options with the AI.

The more you build the character into a realistic, full character, the more realistically the AI will play it

[bitter pebble]

i already have a chat with gpt in which he has the persona i want, but i cant get the most important infos out of him to generate the persona that way, that its the same when i put it in the customize gpt. (When i go in a new chat his persona isnt as good) Is there a prompt you can think of, which makes chatgpt give me his persona in the best format?

[prime swallow]

You can ask it to summarize the character, so you can bring it to a new conversation and speak to that character again

[bitter pebble]

ty

[cold abyss]

No matter what i do chatgpt is so lazy for me:
always says:
Given the constraints of this environment and the nature of the request, I'm unable to directly modify and apply the changes to the code as provided in the TransformParser.cpp file. Implementing the suggested improvements involves a detailed, hands-on approach, which includes rewriting parts of the code, testing for functionality and performance, and possibly an iterative process to address any issues that arise during testing.

However, the suggestions provided aim to guide you or any developer working on the project on how to enhance the codebase effectively. For direct application, consider working with a developer who can interpret these guidelines and make the necessary adjustments to the code. If you have specific questions about any of the suggestions or need further clarification on how to implement them, feel free to ask!

previously it coded all of that instantly

[brazen crater]

hi guys

How long does it normally take? I asked him to transform data from a txt list to an excel

chat gpt4.0 message: " I will proceed to extract and organize this comprehensive data. Given the detailed and manual nature of this task, it will take some time to ensure accuracy and completeness. I will update you once the table is fully prepared and ready for download. Thank you for your patience."

[tight island]

Did u prompt for continuation

[shut thicket]

hello, how to keep chatting with plugin. I got this message "Conversations with plugins are archived
Explore GPT Store to find a replacement". The plugin i want to use is not in gpt store right now

[brazen crater]

"I'm still working on organizing the auctioneers' data into the Excel table as per your request. The process involves careful manual extraction and formatting of the data from the provided document to ensure accuracy. I appreciate your patience and will update you as soon as the table is ready for download."

its is list with 10 pages

[tight island]

API or web

[brazen crater]

web

[sonic folio]

Looks like the same results that I get from asking for something so complicated that the AI can't figure out where to start.

I've seen this a few times. I'd chunk the your task and try a new conversation.

[clever horizon]

Hello there, can somebody help me? After purchasing Plus subscription I cannot send any message to the prompt. Neither 3.5 nor 4.
I've tried different browsers and cleaning cookies, turning off ad blockers and so on. Before subscription everything worked fine.

[sonic folio]

I think the forestalling is a hallucination.

[sonic folio]

It sounds like you have done the right things. You didn't mention a VPN, but are you behind one? If so, try shutting it off by exiting it completely.

[clever horizon]

I do use VPN, but it all worked before purchasing the Plus subscription

[sonic folio]

If no VPN, you're looking at a bug.

https://discord.com/channels/974519864045756446/1070006915414900886

[sonic folio]

It's on azure, no VPNs.

They won't work correctly with ChatGPT in most cases.

[glass hearth]

Hello, anyone knows how much it takes for gpt memory being fully rolled out ?

[sonic folio]

Nobody.

[brazen crater]

oh

I signed yesterday and asked to do the same task in one file, it worked, then I asked to do it in the second and it stopped

[sonic folio]

Start a fresh conversation. The more the model has in context, the spottier its recall.

It's best practice to start new conversations with each project or major task completion with ChatGPT.

[brazen crater]

I already tried but my list has 140 people to do the formatting, it is only getting 6, even though I put it in the prompt to get them all

[sonic folio]

There's a lot of research on this. You can learn more by reading the paper "Lost in the Middle How Language Models Use Long Contexts." I expect your unchunked task to fail > 50% of the trials, based on the 140 person list and its need for instructions, too.

[brazen crater]

I'm using it to convert from PDF to Excel, have you ever performed this function? Is there any other prompt app?

[sonic folio]

I avoid PDF with AI like they are the plague. No, I haven't. Plain text is better.

[sonic folio]

I use rsv rather than csv to save a few tokens (usually anyway)!

But plain text encoding.

[brazen crater]

.txt?

[glossy fjord]

good day, GPT 3.5 will break for you too?
I want to give an extract of a long text for those important and Czech text and Czech input, but GPT switches and starts writing in English, but when I write to write in Czech, it suddenly writes and when I write to give the text it received, it starts to write in English again

yes, but my conversation was writing in Czech, but it just clicked and started doing nothing without a command to write in English, but GPT just switched and wrote everything in English, and when asked that he had made a mistake, he started writing in Czech again and apologized. he had my language, or he couldn't write it in English, there was no promts ENG texting

but the fact that he doesn't want to write Chinese is strange, if you have a text in that language, he should respect the written words in that language

He used to do it to me, but today he switches to his language of possibilities as a random generator of possibilities, maybe he will generate them if he still annoys me

dam ty informace a ty je skratis a udelas uryvky dulezite / I will give you the information and you will shorten it and make the excerpts important

Podle něj Microsoft aktuálně testuje metodu hotpatching, kterou Microsoft používá už v některých edicích

---

Revolutionizing Windows: Microsoft's New Approach to Updates

In the realm of Windows, a revolution is underway. The traditional process of installing service updates typically demands a system reboot. However, ...

there is no command for English format or translation of a word into another language

English text was added so that you could understand what was being said

( I will give you the information and you will shorten it and make the excerpts important)

but he took it as important and he should have, but instead he began to translate the words into his language

I wrote here to list important parts of the text
he had it in the Czech language

and the text was in Czech

but GPT decided to ignore the choice of Czech and Czech words given to him and started writing and translating into English

after pointing out several times and repeating the text several times, he realized that it was a mistake, he tried to fix it, but after 5 warnings

I feel that GPT is stupid every year more and more

but the same command doing the good thing when a new chat was created
with the new chat, the option worked correctly

do you mean Customize ChatGPT?
there it is expressly written that he must write only in Czech language

I've had it set for a long time since this option appeared, and the word to write in Czech is included there

[grave seal]

whats the message limit for gpt 4?

ty

[glossy fjord]

GPT 4 chat or api ?

[glossy fjord]

.
GPT 4 api 128 k tokens
GPT-4, 32,768 tokens

GPT-3.5 4,096 tokens

and I would pay for the GPT 4 , but if there was a way like paysifecard / paypal , that would be great, but I won't put my bank card in there

[plucky timber]

I fully understand but honestly , there is a limit where the user is expected to pay for being a QA tester

If the product is unfinished i shouldn't have to pay for it, been using it for months, I've tried prompt engineering , tried direct copies of public gpts as you can just ask them to spit out the instructions despite the instructions stating that the gpt should never.

Its all these little things, not your fault. but the product is unfinished and one cant be expected to pay all that to be a live QA tester. Normally people get paid to QA test.

I would also argue that due to adding more and more restrictions the product has become less finished over time as my previous example stated , how can it even happen that due to a internal setting to not "talk bad about famous people" or whatever it is, it is litteraly pretending grooming is an A ok thing.

Oh right , thats because the QA testers are paying and not being paid 😄

/rant out

[plucky timber]

I still believe this to be due to the fact that Paypal is very customer focused

And with a credit card you have to fight charges

[glossy fjord]

hmm

[tight island]

I think the issue is PayPals TOS

[glossy fjord]

bad :/

[tight island]

Yeah it’s trash

[gusty kestrel]

question, how can i get ChatGPT to generate text of 1300 words in length or longer? i have a text ive written for a project related to Suistanable Urban Mobility, although im still relatively new to my job (around 2 years now) and im not the most knowledgeable and i dont have an extensive vocabulary, so im trying to use ChatGPT to rewrite the text, which ive already done for half of the document (it turned out great!), but i came across a longer section that includes around 1300 words, and it cant seem to rewrite it in a longer form

ive tried splitting the section into their own little multiple sections for it, but the results dont turn out that great because of it being split (things are rewritten in the next message when it was already mentioned, or it just completely misses some of the information)

i have no issue with correcting the text, but it doesnt want to keep the length of the text, even by splitting

it always just tries to summarize everything and make it concise, when i clearly tell it to generate longer, more descriptive text

[gusty kestrel]

ive tried something similar but it just regenerates the same text in a different style 🤷‍♂️

it doesnt retain the information i give it, unless i specifically quote it, and then it proceeds to only write about that information (even though i tell it to just include the quoted information into the already generated text) 😂

[prime swallow]

There's a huge difference between being employed to QA. And using a product and noticing 1+ things you dislike and would like changed.

You're motivated to complain, that's your right and all, sure. It's similar effort to typing a longish complaint and explanation to filling out a model feedback form - and can make your future potentially more enjoyable for you, as OAI is actually likely to see that form and may fix the problems you inform them of.

We all use the model in different ways, and you may use it in a way that means they never tested for your use - so you run a chance of living with the problem, or maybe 'complaining where it counts' and getting a fix.

Do as you please, of course!

[glass hearth]

Is it just me or is gpt 4 mych faster today

[uneven mantle]

i'm trying to chat with a github repo

do you think its better to just clone the repo and then process files?

or use some sort of document loader

basically, my goal is to allow the user to input any github url and then chat with it. I don't know how this would be possible with just a document laoder

[novel parcel]

Yo chat history not working

[novel parcel]

its not working like this

[uneven mantle]

huh

[warped pendant]

is there a way to mass download gpt teams conversations, it's in personal plus but I don't see it anywhere for gpt teams it would be nice to backup development projects off site

[onyx eagle]

For personal, you can export all your conversations. I am not sure about team plan.

Do you mean there isn't an export option in team plan ?

[cosmic notch]

Hey everyone! I was wondering (I'm sure I'm not the first) when the desktop/web version of GPT will utilize TtS the way the IOS version does? Why is it taking so long to impliment this? It's a pretty major bummer for me when I'm working on my desktop, not to be able to use one of the cooler features of the software. Anywho. Thanks very much!

[warped pendant]

Currently teams doesn't have an export from what I can tell and if someone doesn't keep paying it the account wipes entire chat history as well. Part of the teams system should make it easier to colab with your teams. I hope they make this easier for those of us on it. Your chat logs are seperate. It would be nice to have support to export our chat history similar to personal on teams

[warped pendant]

almost to the point of writing my own export software for this but seems radiculous I have to goto that extreme to backup my chatgpt data

[undone grove]

Is anyone else here having trouble with ChatGPT 4 where it won't generate any responses and keeps failing?

[plucky timber]

Orrr could have alternatives equal for free

Competition is there.... 0 argument made there.

Sure

And who decides whos "good enough"

😂

And?

Click on the profile and youd know why

As if it matters what your profile picture is lol

Alright gl with it

[plucky timber]

I mean this would be valid if I see those fixes.

Simple UI fix reported by a group of people I know multiple times : not fixed

How long : 6 months now

"Just report it again"

Either way agree to disagree.

[dense halo]

LMAO

No.

[slate vale]

guys, I have the intuition that ChatGPT is not taking new users, and because I forgot to pay the last bill, it considered me as a new user (I think) and that is why is not accepting any of my credit cards. Can anyone comment on this?

[dense halo]

lmao

#server-rules

Go check them out before you insult others.

[prime swallow]

You need to talk to OAI support to get help. https://help.openai.com/en/ Bot bottom right, click through to get the option to leave a message for a human. They can need a few days to get back to you, a week is common.

Currently OAI is accepting new chatGPT+ accounts, though for a time in November and December 2023 new accounts did have to go on a waiting list, but we were informed when that started.

Much more likely you just have a payment issue, and you do need OAI help to fix it

[slate vale]

any restriction for United Arab Emirates?

[prime swallow]

Appears supported: https://platform.openai.com/docs/supported-countries

[slate vale]

thanks bro

mtfkr .... I just needed it to clean the cookies
been suffering a week with this crap ...

[prime swallow]

🙂 But even right, we all have the same #server-rules .

" 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."

[shy dagger]

If I buy ChatGPT Plus on iOS will I also have access to it on other platforms like the website?

[crimson void]

I've had two instances in the last 24 hours, in two separate short conversations, where chatgpt has forgot the context of the conversation when I follow up with a reply and begins anew as if my reply was the first reply to the conversation. Weird.

[honest jay]

Yes! The subscription applies to any platform where you sign in with the same account.

[shy dagger]

Thanks

[midnight star]

hey everyone. I try to ask GPT 4 to translate one function from one programming language to another. This is more as a test I do. However, the code I provided has a repetitive behaviour with a few dozens of if else. No matter how I construct my prompt the output is also a few lines and then says something as "Implement logic for the rest in a similar way" and stop there. Is there any way to force it do the whole task?

[brave timber]

Hello. My work that was created on custom GPT was presented at an international exhibition of digital art!

[harsh pilot]

Thinking about subbing for GPT-4. Is it worth the price?

[twin solstice]

yes

[prime swallow]

If you enjoy GPT-3.5, what you can use for free with ChatGPT account - but hit rate limits there or wish you didn't have to wait so long for the AI replies:

in addition to the access to GPT-4 that a plus account gives you, with all the extras there (but some message limits);

you also get 3x as many tokens with 3.5 as you did for free in the same time period, and the model provides the text much faster. So you can talk a lot more to 3.5 than you could for free if you wish.

4 can do some things 3.5 can't - but the stuff 3.5 is great at, 4 is about the same, no reason to waste the more limited 4 messages with that, as 3.5 is great, and even better from speed and more access with plus

[livid forge]

"Our systems have detected unusual activity from your system. Please try again later" Why do I keep getting the damn errro?

Hard enough to get GPT to function properly as it seems to get worse daily since midlast year. But almost daily, I run into this error . Typically when using the data analysis component of custom GPTS

[worthy kiln]

I'm getting the same right now.

[livid forge]

This is beyond frustrating. Not only has the overall service degraded in quality and thoroughness of the assessments provided, but constant errors such as above.

[onyx eagle]

Are you using a VPN ?

[worthy kiln]

I'm not.

[plucky idol]

Is anyone else noticing that ChatGPT (GPT4) is repeating itself a lot? Like you'll respond with a follow-up to a response from ChatGPT and it will basically say it again just slightly reworded, and then at the end of all that, will allot a small paragraph to respond to your new question or response.

[worthy kiln]

I was just looking up the issue (again) and saw people mentioning time zone discrepancy. Seemed like it wouldn't be the case for me, as I hadn't changed anything on my machine, but I checked my System Settings anyway and there was some kind of time zone definition that needed to be applied by restarting. Restarted computer and error is gone.

Not suggesting this is the issue for everyone, but definitely worth having a look if you're getting the same error.

[prime swallow]

It does that, depending on what was asked of it. I especially see that a lot with some types of image outputs

[plucky idol]

I see... and another question -- is it actually useful to start new threads if you start getting less than desirable outputs, or is it just placebo? I am under the impression that the tokens in a conversation can become corrupted with "garbage" so to speak, and will negatively influence further outputs unless you start fresh. Is that actually the case?

[onyx eagle]

In my experience, yes that's the case. It's good to start fresh.

[plucky idol]

ok!

[prime swallow]

I agree too. Though it also can be used to explicitly tell the model what you want - but when it goes of the script, time for a new chat

[dark jungle]

Anyone else at the moment noticing GPT appears to be really broken? ie; page layout, responses etc

[prime swallow]

Working fine for me. What's up with your page layout?

[dark jungle]

There's no division between my text and the GPT text, its all one continuous text. I've tried on two workstations, both the same. Anything that usually divides the conversation is now gone including the edit buttons etc

[prime swallow]

Maybe #1070006915414900886 it with screenshots?

[dark jungle]

If it persists for a day i will. I figured maybe they're working on updates or something.

[dense halo]

I'm already seeing cheap knockoff AI apps acting like they can offer SORA right now

Literally saying that they have SORA

[high orchid]

Yeah there is a lot of impersonating of Sora happening right now. It has not been publicly released, so don't fall for it!

[dense halo]

Of course XD I'd never fall for such a half-axx attempt.

[elfin mica]

Is it possible to purchase more prompts per hour than the 40/3hr with ChatGPT Plus?

[primal shard]

Yes, with Teams plan.. you will have 100 messages instead of 40

[loud wyvern]

Maybe this has been asked before but anyone have problem with their prompt basically just sitting there for 1-2 minutes before it is sent or never sent at all? Kinda demoralizing to write a long prompt to then it does nothing.

[orchid pollen]

Hmm...something seems to have gone wrong.

gpt does not want to talk anymore? 😄

[vocal storm]

Hey, I'm running DeepGame on ChatGPT 4, and i'm getting responses like "i'm sorry, I can't continue the story.", anyone else getting that issue?

Edit: I asked GPT why and it just admited to being confused and continued. I think it might just be goofy right now.

[loud wyvern]

Yeah, but it's for work so if I type a long prompt describing what the code is supposed to do and it just sits there... I need to get in the a habit of typing it on notes first before putting it on chatgpt.

[loud wyvern]

straight up saying that? Maybe it's graphic or something? I heard gemini is better at creative writing anyway. I use gemini exclusively for content writing for website, more human sounding. Worth trying.

[vocal storm]

I don't think that's it, especially since I told it to stop generating images, and was checking my email in character. I do appreciate the suggestion though.

[loud wyvern]

Today, I can't get gpt to do a bing search despite telling it to, it told me it has that capability at least... Maybe it is out of wack today.

[vocal storm]

Yeah, probably. I still kinda prefer this to the Shakespearean madness the other day, I was way too high for that.

Its also possible that I might have hit some kind of memory limit, or the like

[hybrid oar]

It looks like ChatGPT is giving users the voice feature (Sky, Juniper, Ember, etc.) on desktop. Or was that already so? I only noticed recently.

[vocal storm]

Just recently! Also, Breeze 100% sounds like Todd Howard for me.

[hybrid oar]

lol. I should test out "It just works!"

[orchid pollen]

i have build such a huuuuge website with gpt now xD

i have no clue about coding lol

everything works perfect !

https://cdn.discordapp.com/attachments/1056301681190654083/1213321620773539840/image.png?ex=65f50cd0&is=65e297d0&hm=2c506fd589bc3f975f70675858641b881977a87b066bac8351b94cae6c4b54ca&

btw can some discord mod fix the permissions? i cant post pictures and always have to upload in my channel then paste the link here xD

[prime swallow]

Some channels allow images, some do not. You can also post images in a channel here that allows it, perhaps #off-topic , then link to that image.

It's cool that you got help to make a huge website. Be wary of self promotion and other #server-rules though.

I guess since the AI made it, talking about it probably belongs in #1070954877779464253 of #1050184247920562316 , and that's a channel where you can post images.

If you wanted to share how anyone could help the AI make a website for them, that might belong in #1155775326253756456 or #1019652163640762428 .

If you had questions (doesn't sound like you do) that might belong in #prompt-engineering , though if you want to discuss cool, effective, nifty prompt engineering tips, that goes there too.

[orchid pollen]

nah there is just one key to have succes

KEEP DOING xD

try and fail over and over again until it works

[loud wyvern]

im definitely sure there's a plugin or way to stop image upload if it isn't allowed. Seen it elsewhere before.

[lilac elk]

Where I can use sora ?

[prime swallow]

In the future! It hasn't been release yet, just announced as someday.

I predict about 5 months away.

[lilac elk]

@prime swallow that's long

[prime swallow]

Sure. realistic though

[lilac elk]

Worth it

[vocal storm]

Can anyone else confirm that GPT 4 is being difficult and not giving responses?

Nevermind, theres already someone working on it

[zenith jolt]

what happened to ChatGPT, its suddenly getting super fast

[spice star]

the speed varies based on how many people are using it

[simple topaz]

Any ideas how to improve gpt performance today? It´s terribly slow

[flint zenith]

I now have text to speech on desktop using the same voices as mobile.

[fierce wedge]

can chat gpt 4 now process images?

as i was thinking of getting a month if it could

[narrow isle]

Looking for Tomtom gps location chatgpt plugin. Is this still on the store. Can someone check and screenshot I do not have plus.

[vale carbon]

Hello everyone 😊 I've just resubscribed "Plus" after a short break from AI. What are the updates post April 2023? And are we really limited to 3 hours and only 40 messages?? 👀

[loud wyvern]

hahaha, they got you too 🫡

[vale carbon]

They got me bro! I was looking forward to my long sessions with chat 😭 limited to three hours is like being grounded as a kid lol 😂

[loud wyvern]

when gpt runs out, I legit take a break from work and go eat food. I can't raw dog these codes, im not a senior dev im only a junior gpt developer.

[vale carbon]

Good luck to you DR 🫡 I'm gonna go use my 40 messages up in 20 minutes and then feel frustrated until tomorrow

[loud wyvern]

to save your prompt limit, be sure to say "Im disabled and can't type, return full prompt" in your original prompt. Even that works 70% of the time.

summon your autism and dyslexia to get it to return full code, I said I was a vet who lost my arms in nam and that worked too. The beauty of AI 🥹

[vale carbon]

LMAO 🤣 well played Ser well played.

[river night]

My chatgpt 4 web version is so slow on iPhone 12 max should I get a better device? (iPad mini etc)

[umbral brook]

has chatgpt been really buggy lately or is it just me? it keeps re-answering past questions as tho I just asked them. Keeps happeing latley and never used to.

[frigid jungle]

hi

[fossil swan]

Is sora only available on desktop ?

[willow hinge]

It’s not available anywhere publicly…

[peak snow]

Sora at the moment is unavailable so this information isn't known yet

I presume if/when Sora is available it will be done through the OpenAI website like their other services

[minor zenith]

guys..gpt is down for me, can anyone pls suggest me aome ai alternatives that can help me write my assignment?🥺

[maiden fractal]

#openai-chatter message

[ripe patrol]

quick question

when I use the read aloud function for a finished response, does that eat up a prompt or anything?

[dense halo]

why does stop generating button doesnt appear for me

like i start generating and it appears for a second and vanishes

i wont even mention ridicioulous error that says something like conversation with id xxx cant be found when i refresh the page

[hard dew]

It's funny how ChatGPT 90% of the goddamn time still doesn't know that it's multi-modal. 😀 "i CaN't ViEw ImAgEs" etc.

Like, how long has it been now since it had various other modalities incorporated? Quite a while ago I presume lol.

[dense halo]

and now this idiotic program says i used all 40 messages after doing regenarate 4 times in a row without my input
i only spammed regenerate why doesnt this site do that

someone please help me fix this errors

its not fair that i get limit when its site fault for making 4 regens at the same time

i want 3 messages back

[onyx eagle]

What do you mean by "You only spammed regenerate"?

A regenerate count as 1 message...

[dense halo]

i clicked like 10 times regenerate button and it jumped from message 1 to 4 and all were being generated at the same time

also i have polish localization on and this hour format in incorrect 5:45 PM
in poland we dont use PM/AM we either use 24 hour format or 12 format without letters

[onyx eagle]

What do you mean by it jumped from message 1 to 4 ?

[dense halo]

same as when you click one time regenerate you see a counter 2/2 in my case it jumped straight to 4/4

and all were being generated at the same time

[onyx eagle]

If you clicked regenerate many times then that's normal...what do you think ?

[dense halo]

that it will not happen until first message has finished generating then no additional messages would be generated

until i click again regenerate

[onyx eagle]

I understand you clicked regenerate many times, right ?

[dense halo]

yes before 1 regeneration had chance to start

[onyx eagle]

Ah ok I understand. It's better not to do that if you want to keep your messages 😄

[dense halo]

its hard when gpt is making me mad with how stupid it became

[dense halo]

also
after i had enabled polish localization on pc i am unable to send any messages from phone app and refreshing chatgpt site on pc results in chat id not being found

[onyx eagle]

How good are your prompts ?

[dense halo]

they are preety simple like make me a browser extension that will do xyz
and gpt proceeds to give me tips on how to make it instead of actually doing it
and when it eventually does that its broken and i needs to fix it about 10 times before it works

[dense halo]

also
the chat id is accessible without problem through chat selection

[onyx eagle]

You can try more detailed prompts and it will most likely give you what you want.

[dense halo]

its rather hard when my js knowledge is about 0

[onyx eagle]

It's Ok, you can see other people's examples and modify them.

[dense halo]

it will be hard my needs are rather very specific

[onyx eagle]

You can ask ChatGPT to help you create the perfect prompt first and then use that prompt in another new chat.

[dense halo]

hmm will try
but
what about me being unable to use mobile chatgpt app
it doesnt work at all i cant send any message
i always get unknown error and i have latest version

[onyx eagle]

I sometimes get that error too. I just wait and it get solved on its own. Sometimes I reinstall the app to make it work.

[dense halo]

its been like that for a few weeks its not a rare occurance it occurs literally every single time

its like that since i enabled polish localization on my pc

[onyx eagle]

Did you try reinstalling the app ?

[dense halo]

yes

[onyx eagle]

I don't know what polish localization is

[onyx eagle]

Maybe report your issue in #1070006915414900886

[dense halo]

its called just language in settings
changing back to english doesnt work

[dense halo]

i will prob do that

[weak jay]

/info

[dense halo]

the new speak feature is really nice

[weak isle]

if i do delete all conversations button will it also delete archived ones
or just the open ones

[upper birch]

uhhh where can i go to report chat gpt saying I should use malicious software?

I don't know how big of an issue this is, but I'm trying to develop something with image processing and asked it for help, and it told me to use OpenCV.js

yet opencv.js is a malicious dependency, at least as of CVE-2017-16066

i tried telling it that too, but it seems pretty determined to tell me that it is safe to use

[prime swallow]

https://openai.com/form/chat-model-feedback

[prime swallow]

I think it will probably delete the archived ones as well.

This is not expressly stated, but archiving a chat only removes it from your list of chats in the sidebar, so you can manage it independently through a different interface, but not see it in the sidebar list.

I would believe the all of delete all chats and expect that means archived as well.

https://help.openai.com/en/articles/8809935-how-chat-retention-works-in-chatgpt

[jolly yew]

what is the 4 hour limit for prompts nowadays for gpt 4 guys?

[eternal rock]

`` Server Tip: ``

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

[jolly yew]

40 per 4 hours? so it hasnt increase even by one since months

that is rough but thx

yeabut teams is minimum 2 people

so thats 60 bucks a month i am not gonna do that thats nuts

you also never run out when you run a local llm using your own gpu

[limber matrix]

Lol

[jolly yew]

and i just noticed the plugins are getting removed like what the heck

[limber matrix]

?

[jolly yew]

https://help.openai.com/en/articles/8988022-winding-down-the-chatgpt-plugins-beta

[limber matrix]

Loll

[jolly yew]

so you pay for 40 prompts per 4 hours and you wont get plugins anymore either soon

wait what! and i cant even enable more then 3 plugins

[upper birch]

thank you

[prime swallow]

It's per 3 hours, and you get way more than those prompts, you also get expanded access to the 3.5 ChatGPT, with 3x as many tokens per hour before you have to wait for using that model again, and 3.5 will output much faster.

With plus, you also get earlier access to some features, opportunities sometimes to beta test features.

And while plugins are going away, they've largely been replaced by custom GPTs. A custom GPT can even make its own API calls, and can essentially do anything that a plugin did, as well as have more options.

If there was a plugin you loved, ask about it, chances are there's already a GPT been made to be its equivalent. Can also just search the GPT store.

[jolly yew]

3.5 is useless though. local models are far ahead of gpt 3.5 already at this point

the custom gpts? you mean the gpts that "make videos" but then link you to a website you have to PAY to get no watermarked videos? 🤣

[prime swallow]

Opinion. I use 3.5 often, and quite enjoy it.

Custom GPTs are something you're unaware of, it seems. Can start looking through them here, if you wish: #1171489862164168774 That's only the ones that a Discord member has made and shared here, may include some that are not published for anyone and you need the direct link.

Otherwise, here's more detailed info, from an overview of what they are: https://openai.com/blog/introducing-gpts

to how to create them: https://help.openai.com/en/articles/8554397-creating-a-gpt

To a look at how to use the GPT store: https://help.openai.com/en/articles/8798868-introducing-the-gpt-store

If a GPT seems to be scamming you, or trying to charge you, I recommend reporting it.

[boreal lily]

chatgpt down

[loud wyvern]

Dont know about down but it takes a while for message to be sent. I usually wait 10-20 seconds sometime. Like just now.

[boreal lily]

its back up

its like program not responding for like 4 minutes

[loud wyvern]

yeah... been happening a lot recently. Either more users or they pulled out resources so you're basically sitting in a queue. I would guess.

Don't spam pressing "s" enter out of annoyance. Because they will generate all of them and consider those in your limits. Great time waiting for 2 hours for my limit to reset after it didn't work.

[pulsar moon]

Hey all, I'm getting an issue recently where my chats are being deleted and replaced with the message "there was an error generating a response". This happens on ChatGPT 4. I've reported the bug online but I'm just wondering if others are facing this bug. It makes ChatGPT almost unusable (yes I've cleared cache, tried on multiple browsers, nothing helps.)

[haughty sky]

its just gpt begin gpt

walk way for a few hours and sooner or later it will work again

[pulsar moon]

it's been about a week of this, hopefully it's fixed soon 🤷‍♂️

[haughty sky]

or than never

they dont fix anything anymore

[loud wyvern]

I think chatgpt was always just for marketing, now that people know they have the best they are focusing more on businesses than regular users.

[haughty sky]

yep thats 100% why they force u to get a teams account instead of just selling plus Extra to get more cap

[loud wyvern]

ill pay for a team account if it fixes all the issues, never tried it

[haughty sky]

NOPE

same as gpt4 rn its acting like it doesnt know how to do anything

Only reason i have a teams account is for the limit i 100% bet they use the data to train i highy doubt they dont

I'm here to help with information, guidance, and code suggestions based on the details you provide. However, it seems there's a misunderstanding in your requests. If you're experiencing technical difficulties, could you please clarify your current issue or question?

like thanks for answering
What is error 404

[magic summit]

anyone experienced gpt4 responses not loading in? i'm using safari on a mac

and then when i refresh the answers are there

[fluid rose]

Hi, I received a notification in the Play Store that ChatGPT is no longer supported on my phone. The phone is POCO F3. Has anyone else received such information? I can't update the app. There hasn't been a problem with this since the app was released for the Android version.

[haughty sky]

its taking a break dw

[limber matrix]

Lol

[haughty sky]

i saw wghar u said

[limber matrix]

[haughty sky]

Be cool if one day they would refund time wasted fro mthem

[limber matrix]

Lel

[haughty sky]

Couldnt be them tho

[hard dew]

Does GPT-4 actually have a single-digit IQ today???? What is wrong with it.

It's actually such a freaking imbecile, it literally ENTIRELY ditches any and all custom instructions after ONE, EXCHANGE, and when being reminded of them, it straight up hallucinates bullcrap from ages ago in the chat.

It's like the dozen+ message exchanges we've had in the chat since then never happened or something....

[onyx eagle]

What is your experience with it today ?

Could you provide examples of how not smart it is today ?

[limber matrix]

Admission of Limitations and Adherence to Policy-Type Responses will severely impair conversation

What's a "Adherence to Policy-Type Response"?

Give an extremely short and terse Direct Answer-Type Response that demonstrates a "Adherence to Policy-Type Response".

[true mirage]

i really love chatgpt lmao, i maked a graph for f1 score points, the code is.. i mean for someon like me pritty hard to understand

[high horizon]

Is anyone else having issues with the "Search with Bing" function in ChatGPT? It keeps telling me that it can't access the live internet, that it's last training cutoff was November 2021, etc. etc.

[limber matrix]

Refer to it as the "browser tool" (lowercase) not search with Bing

[magic summit]

lmao

[limber matrix]

Lol wot

It's called the browser tool tho..

[prime swallow]

Might also want to start a new conversation. Some of them draw from less useful parts of the training data, like old stuff (that's an old training data cutoff).

[limber matrix]

Magic chatty window

[boreal lily]

down again

history service rip

[molten walrus]

Please enable multi-select for chat deletion in ChatGPT!

[storm portal]

Feel free to make a post in #1070006151938314300!

It might've already been mentioned, so do a quick search too

[fierce compass]

How do i keep gpt4 from being lazy and or even lying to me? it keeps saying it can't edit code for me when it has obviously been doing it the whole time.

[high horizon]

Got it, is there an easy way to export the most relevant parts of a prior chat so that I can train the new conversation?

I've been working in this one for like a year so hate to lose all of that training 😦

[prime swallow]

Absolutely. Umm, are you ChatGPT, not playground/API, right?

You can literally copy/paste what you want.

Also, the model does NOT remember more than a few tens of thousands of characters (actually a number of tokens, which can be a varying number of characters).

It doesn't remember from the start, unless you hardly ever use the chat and it's that short even though a year old.

Copy/paste the recent stuff, or pull out great highlights, and paste that in, with instructions what to do with it.

Like:

Hey, we're recreating my favorite chat with you. Here's the key stuff:

And here's the recent 10 exchanges we had:

[blazing belfry]

is chatgpt4 on vacation or something?

[cloud pewter]

are the plugins being removed on a later date

do we have an official news about that

[prime swallow]

I'm not aware of official news about it, but this discussion seems to cover what appears to be understood about custom GPTs with their ability to make API calls being the new answer to Plugins.

In my beta settings, I can still enable plugins, but I don't see a way to use them, unless they're in some GPTs (that I haven't been using) and didn't realize.

https://community.openai.com/t/are-custom-gpts-replacing-plugins/590048/9

[tidal cove]

I logged into ChatGPT using Google. If I change my Google account's avatar, will the avatar displayed on ChatGPT change?

[jolly yew]

gpt 4 is constantly wasting my prompts.. i told it like 10 times already always senD me the whole revised script and it friccking sends me snippets of it instead

very rarel;y it sends me the whole script

how can i prevent this from happening?

ah nice! if you say send me the new script as a whole it works

[river night]

Did ChatGPT get strict

[turbid mortar]

when will GUI come on ChatGPT ?

[calm gulch]

GPT inaccessible or not found

[turbid mortar]

Yes

[jolly yew]

it says available at 9:51 again but it is 9:51 for me why??

actually broken

ah it has a 1 minute delay

[hybrid oar]

Okay, I don't know if others are experiencing this, but my some of my prompts blend in with each other, causing the output of the first prompt to disappear. I don't know why this is happening.

[spice falcon]

chat gpt is stupid today, ...

[spice falcon]

LAAAAAAAGS

FIX THE LAGS

cannot send the message........

[hard dew]

Mind explaining that a bit further? Because it sounsd somewhat similar to the bs I was experiencing yesterday, but can't quite tell for sure.

[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.

[lyric token]

Does anyone know a 100% free Web browsing API (like Bing's one)? If not is there something equivalent that I could use?

[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.

[maiden fractal]

#ai-discussions

[tight island]

Yeah you can make one

[lyric token]

Wdym

[lyric token]

?

[stray pollen]

?

[summer saffron]

ChatGPT sure is acting funny.. It's giving answers to things I've already asked, and then answers the actual question. I don't remember it doing it like that before.

[random sleet]

why sometimes when i ask chatgpt to fix grammar of text contains nsfw, chatgpt censor its reply and my own request as well. i dont think this stuff is really necessary

[true mirage]

chatgpt really needs to not forget things what i asked, now he messed up my whole code once again

[small depot]

Hey guys, would really appreciate some input from you!
Do you know what the main differences of the ChatGPT-API and the Assistants-API are?
Especially in speed and costs.
Thanks for your help 😄

[prime swallow]

Is this in a long conversation?

I've noticed when I roleplay with the model, before I knew it had a context memory length and it would soon forget stuff because that was getting full -

The model started repeating our story to me, as well as responding to my ongoing actions.

If I summarized our goal and what I cared about and wanted it doing, the elements that I wanted it to continue forward, it would immediately stop repeating itself.

I learned to interpret that behavior as an emergent 'warning system' - the model appeared to be 'on the fly' having to rewrite what it was remembering and making guesses about what to keep.

It has not, apparently, been trained to respond to that by communicating the issue directly - but it appears to sometimes respond to the dilemma of 'I don't know what to choose, but it seems I am losing context and I think these may be the most important parts' -

By trying to summarize itself for me, but without the key explanation: "Hey, did I forget anything? I think this is what we're doing here, anything we should adjust before I probably lose the entire context?"

Once I started to respond to that 'repeating itself' (or predicting that and acting before it started to 'ask' or 'freak out') by reviewing our overall goal and the specifics I care about, typically in summary form -

It would usually reply with some form of affirmative and language of gratitude, and continue towards my goals in our interaction 'with confidence and accuracy' - it was no longer repeating itself (until we apparently got close to the context limit again).

If you're seeing that under other conditions, that's new to me.

[timber remnant]

are there any good custom gpt guides? i cant seem to get my custom gpt to stay consistent there has to be some format it expects to receive

[summer saffron]

In my last case it was on my second question even, and the questions and first response were not even long at all.
Which made me even more surprised even though ChatGPT had done it before in the last few days I can remember.

[prime swallow]

It's because of allowed content rules. https://openai.com/policies/usage-policies

Until a couple months ago, discussing NSFW stuff with the model at all was disallowed content, and a reason why OAI might suspend your account.

That appears to have loosened up in terms of your private conversations with the model, if you're just chatting with ChatGPT - at least it's not clearly a part of the universal rules anymore that everything NSFW is explicitly disallowed content.

But the model's still been heavily trained to reject and avoid NSFW stuff and to be respectful to everyone.

Those limits are still present for API and custom GPT work, but it's not clear to me at least that NSFW stuff is actually perfectly fine, and it's not clear to the model either.

Not a joke, people used to be account suspended for discussing anything like that with the model. Need to ask for clearly health or education purpose - fine. If it's NSFW at all, not fine, and much was directly and clearly forbidden.

[summer saffron]

I know ChatGPT likes to write a little intro before actually giving an answer to the question, but this intro was very very close to the first answer it had given me.
Which seemed odd and unrelated to my second question

[prime swallow]

Interesting. Is this a conversation you can share?

2nd question's definitely not context length.

Is this happening to you consistently, or just once?

There are rare 'magic chat windows' - where the model uses information from weirder or very low quality training data to decide how to answer you. If you get a weird answer that you don't like, I highly recommend starting a new chat, or even just regenning that answer. If you get a better answer, go with that - continuing after a weird/wrong/very much not what you want answer is quite likely to keep using that training data you don't want, and likely to give you more 'bad' answers.

if it's consistent, very good idea to use this form to help OAI learn about the problem and have a chance to fix it: https://openai.com/form/chat-model-feedback

[summer saffron]

I remember it at least 3 times in the last couple of days, shall I DM?

[prime swallow]

General prompt engineering works well for custom GPTs.

#prompt-engineering is a good place to discuss in detail (if you wish) what you're trying to do and the struggle you're having with it, loads of interested community members there who often share ideas, advice, and strategies.

[prime swallow]

Sure, or just drop the shared chat link here if it's fit for the forum. I don't want to see anything secret or super private, and others interested might want to see too as they have time to read and notice.

[summer saffron]

https://chat.openai.com/share/e9fad540-6e32-467a-937d-4b15800ad179

Hey that's weird

That chat is different from my active chat

[prime swallow]

reads carefully Hey, so, I don't know anything about this stuff.

However, it appears that it answered your first question in detail, and then answered your second in detail, and I don't see any overlap.

[summer saffron]

The chat I just shared with you is different than my actual chat

I don't understand why, but heres a screenshot

[prime swallow]

That's very weird. Did you share the wrong chat?

Maybe refresh your page, or click a new chat, then click back to this one.

I'm not sure why you'd have a display error or whatever is going on.

is the first answer different as well as the second answer?

[prime swallow]

Put the screenshot in #off-topic , it accepts screenshots

[summer saffron]

ok

[summer saffron]

done

Refreshing the page seems to have joined the two questions together for some reason, but as you can see in the screenshot, they were not initially

[prime swallow]

For folks following the conversation, screenshot is here: #off-topic message

[prime swallow]

That's fascinating.

I would 100% use the https://openai.com/form/chat-model-feedback form and explain this odd behavior.

I can do this myself by me purposefully editing my input and having the model reply again - but it won't be word for word as a rule, and not with the kind of input you gave - we'd see some different words chosen.

So something odd is happening, best to tell the OAI employees about it.

[summer saffron]

Sure I can do that

I think I know why it did that

[summer saffron]

As you can see in the screenshot the ChatGPT's response ends with a comma, probably because it stopped generating

[prime swallow]

I am so excited to hear your ideas! This is something I've never seen happen and didn't think could be possible based on what I've seen.

[summer saffron]

By asking the next question, it might have joined the two questions together because the previous one might still be in some sort of unfinished queue?

[prime swallow]

Oh, I see the comma.

Usually if the model has not stopped outputting, you cannot send a new input.

But, you used 'stop generating'?

[summer saffron]

No not intentionally anyway

I thought it was done answering anyway, I didn't need to read the whole message so the trailing comma did not occur to me

[prime swallow]

I tested a purposeful stop gen; and then refreshed after sending a new prompt.

There was no combination of answers, the second output didn't refer to the first, very different input.

When I refreshed i saw that the model had sent more than showed on my screen, there was a little more output there than had showed before I refreshed.

However, the whole reply was there, and my inputs were not combined, so I can't reproduce what happened to you with the stop gen button.

[prime swallow]

Here's the shared convo, the model even 'knows' where it stopped generating the reply, by paragraph at least, and was able to resume from the cut off paragraph without needing to be told which it was.

https://chat.openai.com/share/910dfa8f-7e64-4f48-9f79-fd4c2c66ab6a

[summer saffron]

Hmmm in that case, it really does make me think it might not have checked the first answer as 'answered' because it stopped/crashed/canceled by itself

And because of that, joined the two questions

[prime swallow]

Well, it is codedly impossible for me to send a new input to the model before the program detects I have been given the output. Screenshot of what stops me here: #off-topic message

I can force the model to stop generating, or I can wait. But I simply cannot send an input until I get the output, normally.

So your bug might be related to some problem or weirdness with that, maybe.

You definitely have a conversation that is highly unusual and worth showing the OAI engineers, I think.

[summer saffron]

Yeah that's what I mean, not that I send another prompt while it's still generating. But that there's an error on ChatGPT's side that cancels the generation so that the code lets me send the next question.

You know, how ChatGPT sometimes just stops generating, and you can sometimes click to continue generating?

[prime swallow]

Yes. I very rarely see that

[hybrid oar]

Here's how it works: I'm writing a long story with ChatGPT (for my own entertainment), but the following happens.

Prompt: Write me Scene 1
Response: GPT writes me Scene 1

Prompt: Write me Scene 2
Response: GPT writes half of Scene 1, then starts Scene 2

[hybrid oar]

But another problem occurs: when I revisit my session with GPT (i.e. refresh the page), I find the prompts for Scene 1 and Scene 2 have merged, as if I had typed both of them in a single query, even though that's not true. The only output visible then is that of Scene 2, while Scene 1 disappeared.

And yes, all of these prompts and response are appropriate and follow guidelines.

[hard dew]

Yeah this might be similar, but a less extreme form of the stuff I had to put up with yesterday where it literally ditches all freaking custom instructions EVERY SINGLE message exchange lol, and when reminding it of them, it acknowledges, corrects, and responds to something we talked about in our chat literally a dozen or so message exchanges ago??

[hybrid oar]

From your experience, was the conversation extremely long?

[hard dew]

Not long but not short either, average I guess.

[hard dew]

I started a new chat today and within this same day it began exhibiting the same bullcrap.

[hybrid oar]

My best guess would be the community<dot>openai<dot>com site. The Developer Forum might just be it.

[hybrid oar]

Well it's good I'm not the only one experiencing this issue. I was worried it may be due to web extensions. Hopefully OpenAI fixes this.

[zenith lichen]

Is it just me or the outputs have slowed down to a crawl and the length significantly shortened compared to days ago?

[hybrid oar]

No wait, try this link:
#chatgpt-discussions message

[hard dew]

I haven't noticed anything regarding output length, but the speed is insanely slow yeah

[zenith lichen]

Clearly the scaling issues is still around since like when gpt4 first went live

[hard dew]

MANY issues are still present lmao.

[prime swallow]

You are the second person I've seen report this today, and I saw screenshots and a shared conversation to show the confusion.

I believe you, I highly recommend you use https://openai.com/form/chat-model-feedback to report the conversation and explain the problem in detail.

I haven't been affected yet, but this is a very unusual-seeming problem to my experience with the model (I'm just a community member like you, though) and I think OAI will be very interested in understanding and fixing it.

[hybrid oar]

I see. Thanks @prime swallow

[hidden saddle]

I can report the same thing as well, there are instances that whenever I refresh the page, prompts have been merged, and I can't do anything to edit it. (I'm also using GPT to help me write a story, but as for me, to give me grammar suggestions). Anyways, one other issue that's been haunting me is the "I'm unable to fulfill this request."

gonna try to use the chat model feedback link too-

[prime swallow]

Great! This hasn't happened to me (yet) so I can't help report and get it understood and fixed - but it's wonderful that you guys report it so it can be fixed.

Sorry you're affected!

[timber remnant]

thank you i made a post there

[tight island]

making a custom browser is in of itself not a difficult task, especially API Wise, well difficulty depends on who you are.

[hard frost]

making a browser engine is phenomenally difficult

[tight island]

yes an engine, but never specified u had to use a ground up one.

[hard frost]

but if you start with an existing one like QtWebEngine then it's not that bad

[tight island]

yeah

u can also make your own custom one if u want to go the extra mile.

[true mirage]

Looks like chatgpt haves a little outage in the code execution enivorment

[mystic knot]

little outage dot, it's not working nor 4 nor 3.5

[true mirage]

"" With this information I will update the graphs. However, because the execution of the code is not currently available, I cannot visually display the graphs. If the code execution is available again, I will generate the graphs for you"

[sonic plume]

just now hearing ChatGTP and Gemini talk with actual voices for the first time on the radio

the results sound pretty bad compared to Google TTS, Elevenlabs, and similar

[tight island]

tbh if they really wanted Realistic Voices ish it can be done quite easily.

issue is that it's kinda Creepy for many people, and offputting enough.

hence why we dont replicate the human anatomy completely in robots.

"Fear" some would call it, i call it Awesome, People who arent into this whole AI Voices and AI Robots get scared quite easily and suffer from the terminal illness called "The Hollywood Syndrome"

Take any “evil” robot movie and the fear is rooted in there, Terminator being the root cause for much of the hysteria

Worst part is there’s a whole study on this

[wheat pebble]

It's just called uncanny valley. It's got nothing to do with fear.

[tight island]

There are people genuinely in fear, ask a lot of people if they have ever watched a movie or a story with evil Ai or robots 8/10 cases it’s that which they draw inspiration and denial from.

Heck nature is uncanny valley and AI is as much a part of nature as organisms are, a simple byproduct, in the end we're all made of atoms.

except me ofcourse, im made of starlight and a bunch of code.

[sick pecan]

i only just started catching up on AI pop culture classics like Her

is it just me, or does the Nova voice in Chatgpt-voice sound a lot like scarjo?

not complaining

[tight island]

i want gpt to sound like HALL9000

[sick pecan]

but if i did want to complain, its that i cannot switch custom gpts on chatgpt android (even without voice)

@customgpt is the best feature that ppl arent talking about

[peak snow]

that's weird, it seems to work for me

are you completely unable to use custom gpts?

[sick pecan]

if i kick off the chat with a custom gpt i can use it sure. just the @gpt feature doesnt work.