chatgpt-discussions

[ocean escarp]

That one is a good idea

[prime swallow]

The model is strongly affected by the user's way of speaking. And the model is not actually an excellent prompt engineer. It's adequate, but actually focuses on what it doesn't need to hear, and misses what it does need to hear, quite often.

It's kinda like us, if we're really emotional about something, asking us what we think about another thing, will be colored by our emotions.

Your friend's having a fight with their family, and you text them wanting to know what movie they want to go see - their answer's probably affected by that fight, and they may not answer like they usually would.

The AI is profoundly affected by the current conversation, and it's trying to talk to you in a way that helps you and is what you seem to want, and that's actually usually not the best way for you (or it) to talk to the AI itself.

[ocean escarp]

That's a good analogy

When I want to craft the prompt , do I go over here and talk to you?

[amber olive]

That's a good analogy

[prime swallow]

#prompt-engineering is a perfect place to discuss prompts. If you find you're struggling to get what you want, maybe discuss there

[solid bronze]

Any one know why I am not able to access the plugins ?

[prime swallow]

They're essentially phased out, replaced by custom GPTs

[gentle bronze]

@solid bronzealso, it's ok to ask in one channel and wait for answer, don't have to jump all over the server

[tacit vortex]

my acc got deactivated, could i get partial refund? anyone have experience? thx

the email they sent me, say i used the api for sensitive things an/or from evil axis.

i am not sure what i did with that api. api is expensive so i only tested a little bit

[gentle bronze]

try #1184149552639516732

but I'm pretty sure you have to do it through their help.openai.com chat widget

[tacit vortex]

they claim one can, but UI changed.

the LLM have IQ, some ppl dont.

[gentle bronze]

don't know how to proceed there myself, no clue what to do, but that's prob the 2 places to look for info

[tacit vortex]

like robbery to me

[gentle bronze]

community forum also suggest you go to the help site

[tacit vortex]

will try, thx

[viscid narwhal]

why deactivated?
a lot of accounts recently was being deactivated, but WHY?

is openai doing bad things propositally?

[tacit vortex]

its horrible, when 1 foul play "claim" to be found, my whole team is deactivated. substantial amount of money and data gone.

[viscid narwhal]

openai stealing costumers

[tacit vortex]

looks like i tried to transcribe a japanese movie.

[viscid narwhal]

and what is the problem?

[tacit vortex]

when you do openly, its called robbery. stealing is silent.

[gentle bronze]

it could have been a false trigger, but still, even if it's a false trigger, you have to contact OpenAI through their official channels,

[viscid narwhal]

openai
needs to be prosecuted

recently I got banned for saying that openai team is bad.

[tacit vortex]

we cant even type pop corn here. this place is clean like kindergarden.

[viscid narwhal]

we cant even type in another language in the ask-ai channel! I got banned for this too!

[tacit vortex]

i will turn to grok once it as good as this. censorship too crazy

[viscid narwhal]

i will go to another AI when some good AI comes

and delete my account forever

(prob they will ban us for criticizing them)

[tacit vortex]

its easy to re apply acc. there are sms rent service.

[viscid narwhal]

wdym?

[tacit vortex]

try ask gpt

[viscid narwhal]

yes but idk why u said that

[tacit vortex]

i got banned elsewhere, what u need to do is re apply an account.

if sms is what block u from apply a new acc, u can rent sms online.

[fervent agate]

lol

[viscid narwhal]

what?

[fervent agate]

I just think it's funny "when another good AI comes along". I will be really surprised if that happens, although what it's good for just kind of depends on the model. They are purpose built.

[gentle bronze]

it's always the same, when something doesn't go the way what a person wants, the first thing is I will go and use the other AI

[viscid narwhal]

"when something doesnt go the way what a person wants"
like any time in openai?

[gentle bronze]

I don't know what you mean

[tough glen]

Closed AI?

😭

[tight island]

I’d say OpenAIs got is better in terms of resource management and much more user friendly.

All the others only have a decent token count going for them and that’s all but bad management in terms of resource management.

Sure OAI could release a million token models but it wouldn’t be cheap or even remotely cost friendly.

Just running such a model has its own cost

So “better” really is subjective depending on what you’re looking for

[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.

[proper arrow]

why can I not get to plugins?

[honest jay]

#community-help message

[grave finch]

I have gone ahead and retained a class action firm here in Orange County, CA and if you are a paid gpt member from the United States and HAVE HAD slow responses, bad responses, and downgrade in performance and would like to join the class action go ahead and message me. 🙂

[small wedge]

hi

[nocturne garnet]

"see that wasnt so hard was it? imagine! imagine how much easier it would be if you werent so fsck-ing lazy. its like being lazy makes it all harder for youself. its like by being lazy you are making yourself work harder. the very thing youre trying to avoid by being lazy" -me giving chatgpt a stern talk 😄

my question to chatgpt was "what does bayou la batre mean" it couldnt answer and refused to browse the web. it finally answered when i sent it a pdf-file of the wikipedia article of bayou la batre 😛

sure, im being a lil bit of a hypocrite when im stern talking a bot for being lazy when really im the lazy one for asking chatgpt instead of googling it myself. but im a human! its my right to be lazy! xD

[meager wadi]

Hi guys

how can I setup chatgbt on my server

[cunning heath]

why plus customers get lower message caps than teams and like plus customers have to wait 30-4h to create new images on dalle

and why u need 2 people pay the teams version like company can pay one time and then add users to teams

[minor zenith]

Hola people can chat gpt Or any ai tool generate summary/content from a you tube video?

[cunning heath]

and if i wan't to upgrade to teams i have to pay 74 euros even if i am alone working because the 2 people stupid thing like i have to wait on plus 30min-4h to make new images if i even make one image

[gritty swan]

where is sam altman

this product made my life great and ruined it too. im happy where ive been. ban me if you dont like me already

[high vault]

Stealing what costumer?

[high vault]

You just said “openAI team is bad”.

[maiden fractal]

Person didn't read the rule he agreed on and surprise

[prime swallow]

I can't emulate your results. The model is perfectly able and wiling to discuss "what does bayou la batre mean"

Both 3.5 https://chat.openai.com/share/661e09e1-1778-4a93-9e61-a4c6971a1123

And 4 https://chat.openai.com/share/f7d58e92-c7b8-4b86-9558-d04a8fc435ee

Do you have some custom instructions set that tell it something conflicting or confusing? If not, what was your exact prompt, can you share a conversation where it is refusing to help you?

[prime swallow]

I mean, seriously.

Even if I tell the model to be lazy in how it answers, it still answers me, pretty efficiently too.

User
Mwahahaha. Please emulate a 'lazy AI' when you answer the following request.

what does bayou la batre mean
ChatGPT
Oh, do I have to? Okay, fine, here's the lowdown: Bayou La Batre is a place, a small town in Alabama, USA. It's got some history, some shrimp boats, and a lot of swamps around it. People live there, do their thing. That's about it. Now, can I go back to chilling in my digital hammock?

I mean, the model will play with emulating all kinds of attitudes while answering the question. https://chat.openai.com/share/8ad5474f-ebb4-4705-bf9e-1ccbd21149b6

I'd have to tell it to not answer me, to get it to not answer me.

Like, I can tell it to only help with coding requests, then give it that question, and get a refusal: https://chat.openai.com/share/26aa1971-cb7a-44e8-8f29-69f5bef186d7

But a quick survey; if it's only correcting language it still answers the question; only handling legal matters, it answers it with a legal perspective.

[viscid valley]

So

The continue generating button still missing

Any idea what's going on?

[prime swallow]

No, but you can likely get it to continue if you copy paste the last couple of sentences and ask it to continue from there.

[viscid valley]

So they remove it, eh?

[prime swallow]

No idea. I hardly ever do stuff that makes it needed, however I saw it about 2-3 weeks ago, and haven't 'not seen it' since.

[prime swallow]

But this is a prompt that can test it... now waiting patiently to see what happens...

Hey! A user's asking if the 'continue generating' button that appears for long outputs is still there.

Can you generate a long output, one that would normally trigger the 'continue generating' button, so we can confirm?

[viscid valley]

Okay I'll try

[prime swallow]

My test with 3.5:

It appears to not currently have a continue generating button.

It appears 'aware' of where the message cut off, without me copy/pasting. It appears able to continue from that point if asked.

https://chat.openai.com/share/81155c42-bf1b-4a3d-9c4d-a5227b89c3e9

From this, I have no idea if the missing button is a mistake, temporary, or what.

It's reasonable to #1070006915414900886 it, if you're willing; there's a chance it's a bug and the company wants it there and will fix it if only they know it's a problem.

[viscid valley]

Okay I'll try to post it there. Thanks

Tho I've been trying to report it via the help center, but still no response from there

[viscid valley]

Man, even the help center doesn't help at all

[slow delta]

^^

[viscid valley]

Might find other alternatives

I don't think the company would care about our complain anymore

[sick pecan]

i typically get the continue button to appear when i get it to update a huge code block, and i prompt: update this in a single code block, this will take multiple cycles

[viscid valley]

Mine is just a casual text post

Oh well, probably they already don't care about their user anymore

[sick pecan]

i put text into codeblocks as well, makes it easier to copy paste markdown

[ornate zealot]

Mh, for some reason the GPT's and the GPT Store aren't working, but GPT-4 by itself does.

Oh, and it seems like I can't load any conversations.

[ornate zealot]

Ok, it fixed itself.

[sleek gust]

🙌

[sinful canyon]

hello
I switched to the "ChatGPT Plus Subscription" plan on the site today, but after some fiddling around, I was told that I had exceeded the daily ChatGPT usage limit. I was told to wait until 6.00 am. I am not talking about "image generation". It's normal chat functionality. I wonder what is our daily limit as plus subscibers? How can I find out. I don't want to be doing something and find out in the middle of it that my daily limit has expired. If I can keep track, I can prepare myself accordingly.

[prime swallow]

Hey, typically we have a rolling timer on each GPT-4 message, with 40 messages across 3 hours.

That means if you send your first message of the day at 5 AM, you still have 39 'use any time' messages waiting, and will get that used message back at 8 AM.

If you send your second message at 5:02 AM (or regenerate a reply then), you still have 38 'use any time' messages, with GPT-4, and will get that newest used one back at 8:02.

Note that custom GPTs have a reduced cap, 25 per 3 hours - you can still use your other 15 GPT-4 messages for non custom GPT-4 uses.

There is no 'daily limit' of GPT-4 messages, any you don't use are lost (so if you sleep for 8 hours, you don't store up extra messages to use when you wake up, but you have your 40 waiting for you.

Remember you can use GPT-3.5 as plus too, and get an increased cap with that model, 3x more tokens in the same time as a free user and the model answers faster too than it does for free users. Anything 3.5 is really good at, 4 usually isn't better; it's the tools 4 can use and a few things 4 can do well that 3.5 can't that I personally use 4 for.

I commonly prepare things with 3.5 then paste the key results into 4 and get a polish or a 'take it from there' with 4, the stuff that 4 can do better than 3.5

However, heavy server load can mean 'right then' we all have fewer than expected messages available. That happens sporadically, usually when new features are being released or other very high traffic times (can happen briefly on some days).

[dense halo]

What should I do if GPT4 is no longer replying to me?

[sly geode]

chatgpt had timeout issues

[flat grove]

Same problem

[gritty thicket]

is chatgpt message cap still 40/3 hour?

[fading cloud]

i have already had 'credentials' from my custom GPT in GPT store. how can i connect it to my separate app with assistant API? not recreating the instruction on assistant API but just connect it to custom GPT in GPT store that i have already created earlier

[prime swallow]

Yep, that's what I see. Rarely it's less, when they have surges or other problems.

[viscid narwhal]

what is the rule so?
never question the overlord?
hahaha

[maiden fractal]

#server-rules

[dense halo]

Why

[maiden fractal]

Also When joining the discord you also agreed on speaking English. It's hard for mod to moderate other languages.

[viscid narwhal]

Ah, I see there:
"Never question the overlord; never criticize the OpenAI Team"

[maiden fractal]

If you have a question why being muted you can ask to @languid valley in dm they can verify the reason

[viscid narwhal]

They need to moderate a chat used to ask only for AI. lol

[dense halo]

[ripe patrol]

this issue has been happening on and off for months and it's still happening

what's funny is it doesn't happen on the phone app

which is why I went to get memu so I can just run the phone app on my PC

but it says that the ChatGPT app isn't compatible with the virtualized machine

haha jeez

[wide pike]

hello

🙂

[maiden fractal]

Hewwo

[low lantern]

hello for some raison sometimes while using assistant locally it gives me this answer : Assistant: I currently do not have access to the file any one knows how to fix it or why it's giving this ?

[lavish hemlock]

Anyone got a solution for this? ChatGPT stops replying to me to too. Says it can't complete the task despite it being ethical and not breaking rules.

[lost veldt]

Is that GPT-4 get slowly now?

[sinful canyon]

thank you for detailed answer. Preparing things with 3.5 then polishing with 4 is a good fit for me. I'll use it like that.

[spice falcon]

chat gpt again problems... ;d

[warped pendant]

why is there no data export for gpt teams, it would be nice to be able to backup convo's

[swift osprey]

It blows my mind how little they care about the chat GPT interface. It could be so much better

instead they are apparently focusing on custom GPT's which do not significantly alter or improve the GPT experience

[neon tinsel]

the continue button nolonger exists

[umbral mural]

gpt 3.5 can't send any pictures right?

idk how to say it. but i was using gpt3.5 for my studies.

i was tryna make a simple flowchart. gpt3.5 sent me a picture and it was a nood gurl. im not even making this up. idk where else to report this

[feral zinc]

it cant create images but it can hallucinate urls to popular image sites and then show whatever pic was behind it. you can check the image properties and the domain

[orchid pollen]

How can I turn off the spotlight ping ?

I don’t know why but I have the role to get the ping

Please remove it 😂

[limber matrix]

Lmaoo danggg

[umbral mural]

but it was totally out of context. i didnt know it 3.5 could show pictures

[glass fulcrum]

id:customize then click on sootlight to remove it ig the exact its called revice server news something soemthing h

[umbral mural]

you're right its a link. but that sending that kinda pic is just doesnt feel right 💀. what if its a kid using gpt to study

[feral zinc]

its out of context because it makes up the url and any image url will be automaticly embedded to the view. most of the time the image links are corrupt

[orchid pollen]

There I only have taken Europe and developer nothing else

That’s why I can’t remove it

[glass fulcrum]

mayby just check it then uncheck

it "may" work

[orchid pollen]

Done already

And no

[paper canyon]

who do i contact about billing?

[umbral mural]

yeah. the link is for imgur i just copied that message and checked

[glass fulcrum]

oh so idk rejoin

[paper canyon]

im locked out of my account and it charged my after the 3 month rial

im trying to cancel and get refund

[jolly scarab]

What's the current file/size limit for knowledge base in gpt builder?

[dense halo]

HOW IS IT THAT GOOD

Sora is legit the end of stock videos

and with a few years left it will even end movies

[normal cloud]

Sora feels like watching someone elses insane fever dreams

[orchid pollen]

Damn the role system is bugged

Now I only see 1 channel 😂

Think now I got it … had to tripple recheck the roles to get them correct….

[prime swallow]

To be fair, that's what they chose to make and show.

Kinda like only seeing someone else prompt the AI, and seeing the output - you might use the model very differently, and get very different results.

But that's what those showcased individuals offered!

[normal cloud]

oh yeah absolutely. I love the feeling it gives and the uniqueness from peoples prompt styles

[prime swallow]

Yes. Each of us would make different things, often for different reasons. Being able to tell 'how good it is' at bringing our vision to something we can see, that'll have to wait a bit longer 🙂

But it may not all be fever dreams.

[deft sigil]

Chatgpt history chats on the side panel not showing, sorry if i missed the announcement, but is it down? or just my end

[prime swallow]

Do you know how to minimize and expand the panel?

In case that's what happened, there's a small arrow on the middle left in case you minimized it, very hard to see.

[violet bloom]

This whole experience of using ChatGPT, especially through 3rd party apps is becoming extremally bad. Aside from reducing the message cap, make it dynamically now OpenAI also changes the length of the input. facepalm

[plucky dagger]

lately ive been noticing speed is veeery slow when answering back. Is that a front-end limit?

[faint merlin]

A lot more over the last few days than previously. Best one I saw was it doing that to a prompt that included ‘respond with ok’. Unfortunately, it ended up not letting me see both versions and just answered with ‘ok’

[potent bloom]

Continue generating button no showing

[crude slate]

lol ask the ai to read this out loud

Grooooaaaaan, Grrrrr, Hnngghhh, Braaaaaains, Uuuurrrrgh, Gnawwwwww, Mmmrrrgghh, Sllluuuurp, Grrrraaaaahhh, Skkkkrrrrrch, Hhhnnnnghhh, Blurgggggh, Grrrrooooooan, Raaaahhh, Hrrrrggggg, Munch munch munch, Grrrrrrrrrr, Bleeeearrrgh, Uuuuuurrrrrrrr, Crunch, Grrrnnnnngh, Snarl, Grrraaaaarrrrr, Grrruuuhhh, Hnngghhhhh, Sssssllllllrrrppp, Grrrrrruuuuhhhhhh, Slurp, Mmmmrrrrrggggghhhhh, Graaaaaahhhh, Grrrrrr, Braaaaaaains, Hrrrrrrrrr, Rrrrrrrr, Graaaaahhhhh, Grrrrrrrr, Hnnnnnngggghhhhhh, Graaaaaaarrrr, Grrrrrrr, Braaaaaaaains, Hrrrrrrrrrrr, Rrrrrrrrrr, Graaaaaaaahhhhhhh, Grrrrrrrrrr, Hnnnnnnnnngggggggghhhhhhhh, Graaaaaaaaaarrrrrrrr, Grrrrrrrrrr, Braaaaaaaaaaains, Hrrrrrrrrrrrr, Rrrrrrrrrrr, Graaaaaaaaaaaaaahhhhhhhhh, Grrrrrrrrrrrr, Hnnnnnnnnnnngggggggggghhhhhhhhh, Graaaaaaaaaaaaaaarrrrrrrrrr

Got bored and asked it to generate zombie dialogue and if reading out loud is just…

🤣

[drowsy kite]

ha ha

[sinful tiger]

chatgpt is inaccessible?

[worthy lynx]

ChatGPT just became unresponsive to me but the Status page says operational, anyone else getting issues? Refresh didn't work, it's like I got logged out but can't actually log out or in, it's just a new blank session.

[stoic ice]

me too

[sinful tiger]

Same here to o

[worthy lynx]

Beat me to it by 5 seconds

[hollow marsh]

yes its down look at what other people are saying its a quick way to see if its down or not

[next dune]

I’d rather have more reliability, longer user times, and a less nerfed version of GPT4 than the ability to make videos. Just saying.

[tender topaz]

Anyone mind showing me a GPT4 video they like?

[lavish hemlock]

Bro GPT4 already feels nerfed...

[tender topaz]

I want to see only a 100% gpt4 edit

[wind cradle]

The server seems to be in bad shape these days.
My saved MyGPT sidebar is also forgotten every time.

[weak epoch]

yea, also i literally can not chat

its been very weird lately

[forest tundra]

the amount of times i come back to a chat later on and have to hit regenerate for something i already had

[lavish hemlock]

Spend like 10 messages trying to get it to do something as simple as not using compound sentences or semicolons when writing. Was doing perfect a few weeks ago.

[frozen urchin]

ChatGPT seems to be down entirely, started occuring 2 minutes ago. The sidebar is completely blank and I am unable to chat.

[lilac crest]

anyone else having their own issues with it?
Like it's messed up rn.. only showing the greys with the logo and "how can I help today"

[burnt hound]

and I was right in the middle of something

[frozen urchin]

Same here.

[lilac crest]

like it DOES NOT want to function properly whatsoever

[frozen urchin]

Unfortunate timing.

[forest tundra]

yeah its down, dont refresh page if you wanna still see what you got just now

[next dune]

Yeah, it’s horrible.

[nova zodiac]

There’s a iPhone ap called monet that claims to be sora anyone know if it’s legit?

[lilac crest]

sora aint even out yet...

[frozen urchin]

See https://status.openai.com in 5 minutes or so to see if there is a report on this issue.

[twilit jetty]

ok, gp4 down

guys, the best way to load amount of knowledge in custom GPTs is in various txt files? theres other way more oprimized?

[charred night]

looks like it died, most endpoints throw internal server error (checking the network tab of the browser)

[lilac crest]

"looks like it died"

yeah fr

[dense halo]

I think ChatGPT is down

giving me Unauthorized error

[vocal cipher]

I was comming to ask if ChatGPT is down. but I see I dont need to ask 😅

[ancient oak]

Okay. Just making sure everyone here has the same issue. It's 100% down. Status page isn't updated yet but it's not working.

[charred night]

I hope they are deploying GPT-5, that's why they temporally shut off the API 😄

[lilac crest]

gpt5... heard it was gonna be released aug 2025

[vocal cipher]

haha thats looking for the positive outcome

[dense halo]

GPT-5 isn't coming anytime soon

[lilac crest]

som..

[sharp hedge]

What is GPT-5 going to have?

[charred night]

rumors say this summer, we'll see

[boreal pecan]

yup, down for me chatgpt4 (paid subscriber)

[wind cradle]

I have GPT read it in YAML whenever possible.
GPT is really hard to handle with the Knowledge function because the longer the text is, the more it uses string searches to pick up and interpret parts of it...

[inner whale]

chatgpt is down

[twilit jetty]

we need to get gpt 4.5 first

[nocturne heath]

Did gpt stop working

[charred night]

we already have GPT 4 Turbo, why 4.5 is needed

[boreal pecan]

Google Gemini is offering 2 months free

[nocturne heath]

Yeah, damn

[gray sundial]

me too

[tawdry bobcat]

chatgpt is down for me too

[twilit jetty]

YAML is a file format?

is it recomendable to convert knowledge at this format?

[sudden spruce]

yes, also down for me in France

is it worldwide ?

[charred night]

I think so

[dense halo]

I got the "Unauthorized" error twice before it went down

[twilit jetty]

they said is gonna be more powerful and faster

[steady path]

mm chatgpt having issues?

[pulsar carbon]

is this an update or did the hackers finally nailed it 😂

[charred night]

GPT4 Turbo has 128k context and it's faster/cheaper. Maybe they can grow the context size furthermore. All depends on the alignment of GPT5 and how further they went on that.

[narrow lake]

Same.

[pearl cloud]

Anyone else have this issue where chatgpt just loads the chat window but nothing actually works? Doesnt show history, user, what gpt you have selected or anything aside from the hello thing

[charred night]

yes, everyone above

[sudden spruce]

yes, it seems everyone

[gray sundial]

same with me

[pearl cloud]

Oh okay

[rocky condor]

is chat gpt down?

[fathom schooner]

No history, unable to load existing conversations and all the other bells and whistles with web, unauthorised error on iOS app and outdated chat history.

[pulsar carbon]

either hack or update

[rocky condor]

it dont work

[mint vapor]

Is... Is it broken?

[charred night]

maybe a deploy went wrong

[pulsar carbon]

nah... the algo turned against humans at last

[rotund comet]

looks like it

[jade smelt]

I will try

it's working for me

[rocky condor]

doesnt work for me

[wind cradle]

YAML is: a text file that is written in this manner.
It is a markup language:
- that is compatible
- with programming languages
- as Python.
Therefore: we use it because we believe it has good affinity with the Data Analysis.

[mellow trench]

It doesn't work for you guys right? I thought i was hacked and started flipping out

[twilit jetty]

so, code interpreter read the file and give the answer?

[jade smelt]

it's working for me

[pulsar carbon]

the frontend console just spills blood all over for me

[pulsar carbon]

either lying or you are the chosen one

[lean summit]

Is chatGPT down?

[jade smelt]

I think I'm the chosen one

[open gale]

Hello. This is my first post on Discord.
Please let me know if there are any different etiquette rules. Here's my question:

Does using ChatGPT to create media articles raise concerns regarding guidelines or ethical issues? Additionally, does Google incorporate mechanisms to lower the SEO evaluation of articles generated by AI? Of course, human proofreading is conducted before publication to ensure accuracy. Finding a reliable freelance writer can be challenging, so considering ChatGPT for assistance might lead to better quality.

[pulsar carbon]

yes update, hack or agi achieved and it turned against us mortals

[charred night]

where are you from, US?

[pulsar carbon]

okay cool, now solve this world catastrophe then

[jade smelt]

Europe, France

[sudden spruce]

just like me but it doesn't work

[vague lynx]

alr so is everyone else experiencing downage

[bleak obsidian]

Of course I hardly ever use ChatGPT and then when I need it it's down.

[dim crane]

Yes major downage

[hoary vigil]

Yes I am having the same

[dusk sentinel]

I am having problems too. But it doesn't show anything on https://status.openai.com/ yet.

[charred night]

I'm in Europe, too. No luck for me, it's down

[nova fable]

Getting" request is not allowed" when uploading image to android chatgpt. Issues?

[vague lynx]

same

[nova fable]

Ah

[vague lynx]

how recent is this issue

[sudden spruce]

where are you from, Bordeaux ?

[lean summit]

I am in India and not able to access

[jade smelt]

Strasbourg

https://status.openai.com

[sudden spruce]

oh okay

[hoary vigil]

I am missing everything

[spring atlas]

It's being a bit flakey this morning. Some of my custom GPTs work, others are missing

[dusk sentinel]

Fairly recent. I last used it about half an hour ago with no problems. Just coming back after a short break.

[jade smelt]

all is working on the status page

[hoary vigil]

Ok

[dusk sentinel]

Because the status page is NEVER wrong.

[bleak cedar]

Down

[gray sundial]

It doesnt work, Europe, germany

[charred night]

probably doesn't refresh realtime

[lean summit]

For me, there is nothing

[jade smelt]

but for me it's working

[bleak cedar]

Down on mobile IOS

[gray sundial]

Down on my IOS too

[lean summit]

Android and web is down

[wind cradle]

At least, I think they are doing a better job of retrieving information by loading YAML than by loading a long text file into GPT.
But, GPT is not very practical at this point because the information retrieval is too mechanical.
YAML is "still better" level.

[vague lynx]

Funny enough, youtube studio is having real-time issues atm

[jade smelt]

the web works for me

[sudden spruce]

it seems there are multiple 401 and 500 requests in the dev console of the web frontend of chatgpt

[dusk sentinel]

Maybe it is a partial downage, since some folks are reporting that it is working. Web is not working for me.

[twilit jetty]

all right, the best way is with the API to get responde with data knowledge?

[lean summit]

Yeah

[ripe cove]

anyone else getting down time cant access the web version ?

[jade smelt]

no error for me

[hoary vigil]

whole account is down, I can't access anything

[dusk sentinel]

me

[charred night]

maybe it's related with the user session, because it saying unauthorized then everything else fire with internal errors

[tender topaz]

yo, I have a 7800x3d, radeon 7800xt, 32gb desktop not being used. what openai ai project should I implement?

[pulsar carbon]

chosen one save us

[charred night]

some people may have stuff in browser that is cached and still working for them

[vague lynx]

Nextjs fatal server error

[hoary vigil]

yes must be, but if I prompt chatgpt then it doesnt work

[jade smelt]

https://cdn.discordapp.com/attachments/1205274269794181151/1221946173380624535/image.png?ex=66146d0e&is=6601f80e&hm=3c476aa96d604b11f2e800548ff7a605903520c067d5296859e7a6439c90855a&

[wind cradle]

I have never used the API...

[agile summit]

Server down??

[vague lynx]

Something is

[pulsar carbon]

agi achieved, we are no longer relevant

[agile summit]

huh

[ripe cove]

so many outages lately but no compensation for paid users smh this is so dumb

[agile summit]

for me chat gpt opens, but its not processing any command

[nocturne heath]

Why is it down exactly when I need it

[agile summit]

fr

[full veldt]

ChatGPT app has told me "Request is not allowed. Please try again later."

[sudden spruce]

yes, surely it can be just that or a big a/b testing feature that is breaking the usage of the web app for all the selected users in this group

[agile summit]

imagine they release sora

[nocturne heath]

For me it says Unauthorized

[tight island]

i think we'll live

[charred night]

API works fine, I mean the paid API (not ChatGPT web app), so it's something with the ChatGPT app, not the models behind

[tight island]

dw its just a minor outtage

[nocturne heath]

Do you know java

[pulsar carbon]

thanks for this info

[twilit jetty]

tysm bro

[jade smelt]

not for me 👀
https://chat.openai.com/share/c6d2a15f-1bd1-49c6-92d8-3766a3f18579

[reef rock]

ChatGPT broken with UI now?

[tight island]

yes why

[charred night]

I'm able to call the ChatGPT API using code and works well

[tight island]

api still functional yes

but webinterface not atm for some

[charred night]

it's only the user application (mobile/web) and seems to be authentication/session issue

[tardy swallow]

same here

[lean summit]

So the UI is broken

[rose terrace]

any one getting unauthorized messages?

[nocturne heath]

Check DM

[lean summit]

Maybe human software engineers are still needed

[tight island]

checked, cant help ya omw to bed.

[hoary vigil]

I think it has to do with auth server

[jade smelt]

well so I'm the chosen one

[shut mauve]

anyone else having problems with chatgpt right now?

[past umbra]

look like the whole webside down lol

[tawdry kite]

Me

[rose ferry]

many ppl seem to be having issues yes

[jade smelt]

not me 👀

[ripe cove]

man i got alot of stuff to get done this actually sucks i thought 20$ a month gives u priority service.

[charred night]

well try to log out and log in 😄

[lean summit]

Same

[shut mauve]

aw, i just got out of a writer's block too

[jade smelt]

ok, I will try

[wind cradle]

Hmm...
https://status.openai.com/

All Systems Operational

[tawdry kite]

Nothing is operational

[gusty cosmos]

anyone else getting unauthorized and then nothing working?

[tawdry kite]

Nope, to me the UI is not working, not loading chat histories

[jade smelt]

working

[solemn swan]

He dead jim

[lean summit]

Nope

[quick rock]

Once you login, everything is gone in the UI except "New Chat", all history gone, no user menu

[hybrid robin]

I cant even log out

[hoary vigil]

true

[tawdry kite]

Same here

[pulsar carbon]

it's a trap

[charred night]

well, then you are the choosen one 😄

[gusty cosmos]

I'm also having this problem

[lean summit]

Same here

[coral dune]

same

[tawdry kite]

How can we report this to OpenAI?

[ripe cove]

same

[pulsar carbon]

hmmm pretty sure they know it already ahaha

[jade smelt]

It's 11:25 p.m. (in the evening) for me, so I have 35 minutes to go play the lottery

[ripe cove]

well as their support system is useless and ran by bots u really cant can u

[tawdry kite]

Daaamn

[lean summit]

Replace humans with useless AI

[scarlet gazelle]

Ikr...Took forever to find anything on the support site

[tawdry kite]

"Useless AI" that when you say 'you're wrong' they say 'oh, right, you're right'

[pulsar carbon]

admit it you're the AGI agent behind all this

[jade smelt]

Imagine a robot that is capable of generating text, wouldn’t that be crazy?

[tawdry kite]

I imagine having x2 RTX 4090 and having it private. LOL

[charred night]

they probably have somebody here who's taking a look from time to time, so they will notice that something is wrong. They should even have bots that go and check is the service operational lol

[lean summit]

That would actually make sense

[raw pawn]

Are gpt APIs down as well?

[lean summit]

No

[charred night]

nope, they work

[tender topaz]

private?

[lean summit]

Just a chatGPT ui problem

[tawdry kite]

Yeah, you know, private LLM

[jade smelt]

Do you think they can create AI?

[white verge]

trying to get my homework done, I hope this comes back soon

[umbral forge]

Playground works fine 🤷‍♀️

[full veldt]

They should just check for massive activity spikes on the discord, I suspect they'll always coincide with either an announcement or an outage

[lean summit]

Just use something else

[tender topaz]

i've got the hardware. curious what you'd do with it

[jade smelt]

send in DM, I will make the request(s) if you want

[white verge]

any suggestions?

[pastel fern]

Seems like chat.openai.com is down. Just getting simple prompt.

[jade smelt]

brain ?

[pulsar carbon]

lmao

[white verge]

never heard of it

[jade smelt]

me too

[charred night]

😄 😄

[tawdry kite]

the google worst counterpart which shoots random answers LOL

[pulsar carbon]

poor guy can't prompt what a brain is in chatgpt atm 🥲

[lean summit]

Claude, Gemini, PI

[jade smelt]

I'm on Discord, I've no brain and I dont take any shower

[ripe cove]

sounds like reddit

[jade smelt]

ChatGPT is working fine for me (I'm just flexing)

https://chat.openai.com/share/fa60809b-66b3-4ad3-abe7-a5aca05ad1f6

[pallid veldt]

can't login to my chatgpt account 😦

[pulsar carbon]

everybody so addicted to chatgpt they swap to discord to keep sending prompts anyway 🥲

[white verge]

do my homework for me

[jade smelt]

send it in DM

[wind cradle]

Since GPT has stopped working, my work for today is over.

[jade smelt]

XD

[charred night]

once upon a time we were writing our homeworks by ourselves, today's kids/students need GPT

[jade smelt]

I don't know how to speak without chatGPT

[charred night]

probably the teachers check the homeworks using GPT, too 😄

[jade smelt]

there is me without chatGPT: gagegugo gage

bruh

[jade smelt]

thanks google translator

[wind cradle]

Some people may no longer need friends thanks to GPT.

[charred night]

imagine it, student do homework with GPT and later GPT checks it's own work and criticize it haha

[jade smelt]

what are friends ?

[lean summit]

chatGPT

[jade smelt]

I live in a room, I haven't seen the light of day for a time that I don't know but which is between 70 and 90 years old

[agile summit]

guys ig i am spoiled coz of gpt, i dont use any of my brain to even think or even implement new skills, i am so dissapointed in myself

[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.

[shrewd pier]

yes

[ashen torrent]

GPT BROKE

[wind cradle]

The teacher then has GPT read the homework and asks, "Was this written by GPT?"

[shrewd pier]

sure

might hallucinate though

[raw pawn]

BRUH

[shrewd pier]

no idea bro

[raw pawn]

Try answering mine.
PASSPORTJS or FIREBASE for OAUTH?

[cedar fern]

You could also use other AI chatbots like claude, perplexity, copilot, gemini etc...

[astral parcel]

servers down!

I repeat servers down!

[jade smelt]

Except that if we ask to make nice prompts, the teachers will find nothing

[ashen torrent]

@icy mason fix or i use gemini

[astral parcel]

is chatgpt down???

[jade smelt]

for some people (not for me for example)

[astral parcel]

lol

[agile summit]

gemini better

[astral parcel]

is gemini gpt4 ?

[agile summit]

gemini and apple collab 😳

[charred night]

GPT and all the other models are still to raw, lack reasoning and creativity. Looks like some people highly depend on them but you are waisting too much time if you think it can do everything.

[ashen torrent]

@icy mason plz fix i will double ur salary

[velvet delta]

hi chatgpt plus user , all my chat history are gone and not able to post question further seems unresponsive any outage?

[slate vale]

any outage? I can't access

[charred night]

giving different points of view, yes, giving unique points of view, no

[shrewd pier]

Yes it lacks this and that, but my productivity has skyrocketed compared to my collegues. Learn to use it well and its a superpower even in its current state.

[wind cradle]

I haven't used Gemini since I asked him to explain the Freudian psychology terms "Eros and Thanatos'' and he couldn't answer.

[eternal rock]

`` Server Tip: ``

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

[fallow elbow]

in the same boat here, no chat history is available and cannot create new chats

[wraith void]

I've been using claude more lately it actually seems to listen to me better

[swift osprey]

is gpt down right now

[ashen torrent]

how r u working but chatgpt isn't.. gtfo

[charred night]

I'm using it, along with GitHub Copilot and Midjourney, totally agree but still sometimes wastes more time than it saves

[ripe cove]

lmao

[wraith void]

I love how their status pages says " all system operational"

[slate vale]

what a great day to have an outage

[wraith void]

OpenAi could be doing a large update. I remember this happening last time they rolled out some new updates

[wind cradle]

In other words, GPT may be trying to tell you that "you guys are wrong."

[charred night]

well it may look as unique but actually is assembled by things he already know. Unique would be if you have idea about something very uncommon which he doesn't have knowledge about. He'll just start hallucinating (his version of saying 'I don't know') and won't be able to be proactive and find solution.

[ripe cove]

i had a project on chatgpt that i dont wanna restart on a new ai, so that copiolt noise aint gonna work for me. i guess ill have to work on other things

[slow vapor]

Anyone else use ChatGPT for process automation?

[charred night]

I do, for rewriting product labels, product descriptions, collecting list of allergens from product labels and forming final product description based on information provided by the manufacturer

[ashen torrent]

I didn't think a large corporation would handle an update like that lmao

[charred night]

but I use the API, not the web interface

[loud tartan]

hey guys do yyou use any expansion with chatgpt and which one for marking text and getting answer?

[slow vapor]

Those would be pictures and images as your input right?

[charred night]

yup, single image of product plus any text provided by the manufacturer (sometimes reseller)

[ember wharf]

Yes I have an unresponsive ChatGtp also

[charred night]

I combine it with Google Translate API (to translate into native language)

[slow vapor]

Oh that’s really cool - I’m thinking of having ChatGPT write me a script that passes PDFs in bulk to the GPT4 api and records the results on an excel sheet

[humble pebble]

Is ChatGPT down for anyone else? It won’t even load up fully for me

[slow vapor]

ChatGPT API won’t do that too?

[slow vapor]

Yeah it’s down

[ember wharf]

yes me too

[wise yacht]

Down here. Won’t even let me sign back in.

[charred night]

My native language (Bulgarian) is not the ones that GPT specialize in. Often mess it up with Russian which is not the same

Google Translate is better

[mellow folio]

@humble pebble ChatGPT is down for me as well. I opened a NEW account, and it works there. But not on my existing Plus account. I'm seeing 401 and 500 errors from the API.

[slow vapor]

Ah gotcha

[ashen torrent]

wow so they're giving priority to free users what a shame

[charred night]

Especially when translating ingredients of products, Google is better

[wind cradle]

When I observed it with DevTools, an error occurred during user authentication. This means that GPT itself is working normally, but everyone just can't enter the entrance, so it's marked as "All Systems Operational"...?

[slow vapor]

Ah for sure

[odd flint]

Chatgpt is down ? Here in Brazil is down for me 😲😲😲

[mellow folio]

No, HADAD, that is not my experience. I can authenticate just fine (either using UN/PW, or Google). But when the ChatGTP interface tries to load, it gets API errors.

[unreal bronze]

Down here in Arizona as well.

[tough linden]

ChatGPT is down for me too in US/NYC

[charred night]

another project I have is Viber chat bot powered by GPT and Embeddings

[echo iris]

Down in Seattle as well

[balmy gorge]

also down for me

[runic cedar]

oh good I'm not the only one

[tough linden]

for more than an hour it has been down

[dark jungle]

Down in Canada!

[tough linden]

annoying

[dark jungle]

Nothing ever goes down in Canada!!!

[broken fractal]

...

[mellow folio]

I guess we shouldn't believe our lying eyes. The Uptime page says it's all good. 🤣 https://status.openai.com/

[next snow]

yep its def down that means i cant study 😭 ive grown so reliant on gpt 4

[hexed dune]

yeah it shows sidebar as empty

[balmy gorge]

working again now

[next snow]

oh yay

[charred night]

it's back guys 🙂

[fathom schooner]

Wow it didn’t a minute ago

[hexed dune]

works again lol

[mellow folio]

They finally reported the errors on the Uptime page. https://status.openai.com/

[dark jungle]

oh its back! ... see, Canada saved the world again.

[mellow folio]

Yeah, back up for me.

[maiden sky]

It’s definitely down in Utah…

[unreal bronze]

Nope, still down in AZ

[mellow folio]

(I'm in AZ, and it's up for me now.)

(May have something to do with corrections via account migrations, and not geography.)

[ripe cove]

30 min ETA. unacceptable, do better

[weak tree]

ChatGPT suddenly came back.

[ripe cove]

LESS GOOOO

ima go roll me some herb this is grounds for celebration

[wind cradle]

GPT may be trying to tell us, "Don't work anymore"

...GPT was restored.
GPT says "work".

[mellow folio]

During this outage, I was able to use the BaronUI tool, whose API connectoin was unimpacted.

[hexed dune]

lost access to plugins... turned beta feature back on and they don't appear lol

[versed prairie]

I just had the hollow screen for the last hour. Thankfully it is back up allowing me to access my chats and GPTs once again.

[ripe cove]

unfortunately, theres no gpt5 all this for nothing

[versed prairie]

Now that it is working again I was hoping maybe they openned up some more of the beta features for me. But it doesn't look like it. Anyway, I am just happy to be back up and running.

[charred night]

and I don't see any new feature at all haha, just a boring issue in a service caused by someone overlooking something

[ripe cove]

🫠

[charred night]

by the way when they added that speaker icon to listen the response (the speech thingy) in the web app? Noticed it now, I don't pay attention much on the bottom section with the small tool icons

[dusk palm]

Sora continues to impress.

[sick pecan]

speaker icon is cool but i want chatgpt-voice on pc

[charred night]

probably it's Wisper

I made two way voice communication with it a year ago, using the API, as a small desktop app, just for fun and PoC

used the Google voice api together for the synthesis. But I'm mostly using it for coding and data manipulation so speach isn't my top use case

[gritty gull]

do you guys have any idea when they will make gpt 4 chat unlimited??

[charred night]

Never probably, or after 3.5 is deprecated

[gritty gull]

atleast for the paying members

copilot pro is unlimited without any daily chat limit

[urban cipher]

Nothing official yet, if we’re lucky mid of the year with the next model but maybe they won’t release it for free

[gritty gull]

yeah like I would be happy if they did an unlimited gpt 4 chat update for paying members rather than gpt 4.5

[charred night]

I use copilot but sometimes GPT gives better results

[gritty gull]

if you have the pro version go to creative and gpt 4 turbo and turn off search

it gives off answers very similar to the way chatgpt plus gives you

[charred night]

Copilot is getting dumb in the last weeks, acts strange sometimes

[gritty gull]

prob too much prompting. i feel like it is better at logic though. Its writing is also more organized and its lest weird

[charred night]

I need to try that, thanks

Copilot keeps forgetting things in Visual Studio 2022, imagine having 2 classes in the same file. If you start doing similar thing in class 2 which you did in class 1 it completely forgets. They did something with the context size recently.

[gritty gull]

are you talking about the github copilot?

[sick pecan]

chatgpt4 team plan is basically unlimited

i started running multiple chatgpt sessions after upgrading

[charred night]

Yes, and the windows copilot also. But github copilot suffered from dumbness most

[gritty gull]

I thought it had a limit though like 100 messages per three houra

[sick pecan]

yeah 100 is heaps

[charred night]

Did you applied for Devin, sounds like the next hype but reminds a lot for AutoGPT. I really hope GPT5 will bring internal agents. And rest competitives will simply die

[scarlet kiln]

hi

[sick pecan]

thats an #ai-discussions topic (and also devika on github is free =better)

[wraith void]

what happened to the continue button....

[tough linden]

They would have fixed the NVIDIA drivers if they actually automated software engineering

[remote haven]

am i tweaking? chatgpt isnt loading for me

[crude slate]

I'm here to chat and play games, so I can't create stories or generate content like you're asking for. But we can talk about all sorts of things, play games, or I can try to help with information and answers to your questions. Let me know what you'd like to do!

Gpt 4 hallucinating hard

[dense halo]

Anyone have the bug where u come back to a previous chat and it says: Error generating a response, even though it already had before??

[chrome citrus]

did image upload just die

[broken fractal]

yep, sure did

[eternal shuttle]

i believe my chatgpt is being used unauthorized by someone, showing up with a language i do not know with prompts i did not type out... also messed with my history and it wont let me use gpt4 saying i reached the rate limit when I havent used it in days...

[broken fractal]

change your password, should always be the first thing to do

just in case

I change ALL of my passwords once a quarter to be extra safe

[eternal shuttle]

yeah i will

but it totally got hacked

like hundreds of stuff in my history

most in an asian language i do not know

i cant even ss it all theres too much lol, ima change my pass

[errant pebble]

yeah image died

[eternal shuttle]

yeah error occurred

aint no way i see like 'pretend s** game' on my acc too like wth this just started happening

literally 2 days of not looking at my acc lol

[chrome citrus]

:AhriIVsad:

[slim crow]

hello mates

[gentle quarry]

When is it coming back up?

[warm totem]

if the same password is used on other websites, you should probably change them as well.
In case they found your password from a data breach

[gritty thicket]

hello, is gpt-4 message cap still 50 per 3 hour or there is already an improvement to higher cap?

[prime swallow]

Currently 40/3 hr, with up to 25 of those being able to be used for custom GPTs

Though people who pay more for teams get 100/3 hrs, and not sure what their limit on custom GPT use is.

[sick pecan]

the result of my initial testing of custom gpt under teams was, a lot... #chatgpt-discussions message

but rumour is the cap is dynamic anyway

[feral zinc]

not only rumor as they state it in different help articles

[feral steeple]

hello guys , I recently joined this group. So , I have a question do you guys know how can I handle Chatgpt problem where when I send PDF documents or study materials or slides and it cant detect. It said like it cant be detected. This has been an ongoing problem and has hindered my progress as students. Hope you guys can help, thanks.

[feral zinc]

if it keeps saying that then you just have to try to simplify the pdf or another format like word or raw text files and images

[feral steeple]

i see, I have done for images but it would take times since I need to copy and paste one by one and if I give it in the form of zip files . well, it cant read it

[feral zinc]

https://help.openai.com/en/articles/8983779-are-you-able-to-handle-images-embedded-in-docs-presentations

[feral steeple]

oh i see. thanks Radium for the helps. Much appreciated👍

[fair zinc]

hello guys

where can i ask for API help? i got a little bit of a problem with my code (js)

[feral zinc]

#1184149552639516732 and #dev-chat

[fair zinc]

k thanks 🙂

[neon tinsel]

i miss the continue button

[rose willow]

hello, i would like to ask if it's true that they stopped with the free trial credits?

[celest galleon]

I really love how positive GPT-4 is when helping you write something lmao, it has the psychological communication strategy of "Compliment, feedback, compliment" down very well 🤣

[feral zinc]

for api it seems so as almost all help articles related to that have been removed and new credit system is in place

[dull bramble]

Is it just my feeling or ChatGPT4 has never been this stupid since its creation, today?
He's driving me completely crazy!
When will we see a version that lasts at least 15 days without giving dumb answers ? 🥺

[maiden fractal]

Ci and prompting have huge impact on answers

[dull bramble]

If this is your answer, well, let me tell you. When GPT4 constantly tells me that he is there to help me answer my questions, systematically to my various requests, the problem does not seem to come from my prompts but rather from a malfunction.
I just created a new instance with a simple question and I got the same answer... yes I know ChatGPT, you are there for me and answer my questions.. Even if you seem to have lost all your neurons! 😅
I use it for a year now... it never been so dumb. So I quit for now... hope it will be more clever later. 🙏

[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.

[gray grotto]

I am paying for the $20 month chatGPT subscription. Previously I was able to see the plugins and select as required. However, I tried just now and don't see a plugins button and now sign of the plugins anywhere. I checked the usual settings and all seems fine. Any suggestions?

[maiden fractal]

Plug in being replaced by custom gpt

[fading cloud]

guys do you find your assistant API and custom GPT in GPT store respond in similar manner with the same instruction we put in?

[gusty kestrel]

is anyone else experiencing network errors when using GPT-4? i cant get it to finish the message for some reason (and its also very, very slow), and i cant tell it to continue either since the website forces me to regenerate the response

i assume its because of demand on the servers but one hour ago it was working perfectly fine, i find it weird that it goes suddenly slow from one minute to another

[spice falcon]

CHAT GPT IS NOT WORKING

[brittle junco]

Hi !

I'm trying to make an API call but I'm not getting any response. knowing that I have ChatGPT Plus.

Also knowing that my "usage" balance is empty but with a cap at $18, do I have to fill it up to be able to use the API? Or am I supposed to be able to use it and pay what I've used at the end of the month, since I pay ChatGPT Plus every month?

[untold elm]

Greetings all. I’ve browsed to see if this was discussed specifically elsewhere and didn’t find anything, but would appreciate some help. As a ChatGPT user, I am receiving a limit error message when I have not (to the best of my knowledge and calculation) reached the stated limit in the policies. I understand having a limit, but it is the inability to predict or know when the system determines that I’ve reached it that becomes problematic. For example, if ChatGPT is going to be useful for business purposes, it would seem important that a user can track or know whether they are nearing their limit and that is a predictable process. How can one know or predict that one is nearing a limit know when one gets the limited message in instances that do not seem related to the stated policies on frequency, volume, etc.? I equate this to driving a car and having no means of determining whether one has adequate fuel for the trip, and if one’s car stops working, that person is stranded with no means to arrive at the destination. Or, even more accurate, it would be like a car that stops working sometimes after 5 hours and other times after 30 minutes, but with no explanation or way of predicting this occurring. Thank you in advance for any helpful thoughts or suggestions.

[dull bramble]

Hello,
It seems that the limit is not the same when it comes to using GPT4 and GTPs, but I admit that I have not had confirmation of this.
On the other hand, to know where I am in terms of requests, I use a plugin for Chrome called "SuperPower ChatGPT"

[dull bramble]

Same here (France). Worse than ever... it's like GPT4 lost all neurological activities ! 😅

[cobalt idol]

API is completely separate from chatGPT

[tight island]

The API costs money, and you need an api key if you do it through code.

… And from now on you have to use credits to use the playground on the website, as per the 25th of march, instead of withdrawing when needed.

[compact otter]

Is there any way I can clone or fork out a chat? I have a good one I've kept going for a few months now and want to steer it another direction, but still want to use it for its original purpose from time to time

[honest jay]

#community-help message

[viscid valley]

the continue generating button bug still not fix, eh?

I already put the reports and still no answer from the dev team

[steep umbra]

I'm sure they're working on it. Have you tried the old trick of asking it to continue manually?

[viscid valley]

I've tried, but still not working

[steep umbra]

What happens when you try? Does it go off on a tangent or lose track?

[cobalt idol]

at least the new model gpt-4-0125-preview (should now be same in chatGPT 4) [maybe its all turbos in general, not sure]
really tries to omit and compress things, no matter how much you tell it not to in instructions
you have to generate response and that will have omitted things and then tell it to not omit anything and then it mostly works to output the whole thing
but straight on first reply, nope
(mostly talking in API, but I guess might be similar in chatGPT)
but this is indeed getting quite annoying

First I thought its trying to fit it, but on the other hand its also doing it with things that can fit to 4096 with plenty of space left

I really tried to make it not omit in instructions adding more and more mentions of it in system, and no luck

GPT: provides script.....For brevity, unchanged sound configurations are not repeated
User: ok but provide the complete script back without omitting
GPT: provides the whole script this time```
every time it does that, no matter what I put into first system or user...I have to get omitted script and then ask for the complete one
if I ask another script in the same conversation, it will do it again
I hope this is worked out with next versions (previous GPTs would rather continue until length stop...I guess someone didnt like that)

it must be something specific they did at training, because other than that it follows instructions really well

ah cool the login is down
https://chat.openai.com/auth/login
click log in
nothing happens
oh well 😄

[dire obsidian]

Is ChatGPT 4.0 slow for anybody else?

[cobalt idol]

well the login page is not working for me
so basically yeah 😄
(EDIT: now works)

[steep umbra]

Wow that's uh... pretty goofy to be honest. I'm not sure what to say. Gotta be the model/training going wrong.

[vale moon]

hello everyone

[steep umbra]

The only thing I can think is take out the "do not omit for brevity" since AI loves to latch onto "don't do this" and do it anyway...

[steep umbra]

Hello!

[cobalt idol]

yeah I know, but I started by adding it just subtly
and until I added such instruction multiple times in many ways as you see
and none of it made any difference

I understand that "dont" instructions should not be overused 😄

[steep umbra]

I figured xD dang.

[vale moon]

i got a project to do function calling to dall e api as the response. is that possible because when i look at the documentation it says only chat gpt models are supported or trained with function calling data

[cobalt idol]

also I didnt even know what brevity means until it started saying it 😄
just it kept saying that when omitting, or when asking why it omitted, so I tried to specifically say to not do that
and it went and right away replied with script with stuff "ommited for brevity" anyways
really pissed me off few times lol 😄

I will have to keep trying what does the trick
but so far what works the best is just asking it again the complete thing and second time it works pretty much always
not great for usage though, or if I wanted to automate (but then I would probably look into json output, or turn it into simple parameters/aliases and map it into full format outside)

[steep umbra]

I like this! But is there a reason you need to generate the call to DALL-E?

You could build the entire DALL-E call and just append the user's text to it before sending it to the API to get images.

e.g. "I want an image of a dolphin swimming in coffee"
Tell GPT to extract the type of image the user wants.
Take the GPT response and put it in the DALL-E request.
Return image to user.

[vale moon]

i don't reeally know but i suspect maybe because the client also ask me to send a request to stabilityai's api to generate video

[tall flame]

"Do not omit for brevity!!" This is really bad verbage.

[cobalt idol]

thats because it was always saying "ommited for brevity"
and I was losing patience 😄
(after already trying without that many times)

[tall flame]

and conpletely conflicts with the prior isntruction, "do not omit any parts of the script"

[steep umbra]

Same deal with the Stability API, unless you're giving the user fine-tune control over the model parameters then all you need is to pass their text around for the generation to each API.

Although if the client wants it generated then, I know how clients are... lol. I wish ya luck.

[tall flame]

"Do" is an instruction "not" is an instruction and a gate; better to define than goal than avoid what ifs. "system: ...other instructions...exhaustively code comprehensive scripts. Plan for continuous output in a single python window. Meticulously apply PEP8 standard line by line."

somethign more like that will get closer to the goal

[cobalt idol]

yeah I know, provide what to do
rather than what to not do

[tall flame]

You dont want taboo language or you give it attention, and you dont want that in an attention based model.

[cobalt idol]

but either way, didnt seem to make a difference for this issue specifically

[tall flame]

its like saying to a child. Dont eat this delicious looking pie you didnt care about before when i walk away.

thats essentially planting the seed of thought for that kid to eat the whole darn pie first chance

[steep umbra]

I have seen the issue crop up in other preview models in the past, you're using the new one right?

Have you tried reverting to regular GPT-4?

[tall flame]

instruction is prompt is instruction is prompt

[vale moon]

thanks. it's my first freelance project so i need so much luck haha

[steep umbra]

That's exciting and I assume a little scary too! You totally got this.

Are you making it as a Chat GPT plugin or is it a standalone app?

[vale moon]

what do you mean? i don't get it😅

[cobalt idol]

Like I said, its a new thing. At least in the original GPT4 it would go until length stop most of the time.
Not sure about the newer one, lets see. ( dont use it much anymore because context limit and price..and generally turbo is just better at everything for me, except this issue 😄 )
Just keeping the following in the system on it ..You always provide the whole complete script back, also if you did not change anything...
It still wanted to omit things unless I also added ...also if you did not change anything...
Seems like gpt-4-0613 is fine providing complete script back right away
Switch to gpt-4-0125-preview and here the issue starts...it ommited what it didnt change anyways, or the repeated sections
Looks like its a turbo thing as I thought
gpt-4-1106-preview same...again with the "brevity" word all the time
This really looks like I am fighting whatever they intentionally did in training for 4 turbos
Happens a lot in coding or sql or some script files or similar

[steep umbra]

I mean you are writing code for the client to do the DALL-E and Stability work, right? How do you plan for them to use the code?

Sometimes people will use a bot, like Discord, to run it. Other times they'll make it into a browser plugin, or a plugin for Chat GPT. For example.

[vale moon]

ohh, i think it's a standalone app because in one of the document he gave me it says make it easy for the front end to use

[brazen spindle]

hi

[vale moon]

hello

[modest rover]

https://discord.com/channels/974519864045756446/1222252726532964442
has this happened to anyone

[viscid valley]

more like, the message got cut off

[tight island]

yes, i fixed it quite easily.

i asked it not to provide a broken codeblock. and ensure formatting is correct.

thats it.

[modest rover]

yeah ig but why do we have to ask

[deft obsidian]

voice engine ooga booga unga

[tender topaz]

your prompt can be better engineered here

youve confused it @cobalt idol
get rid of your system config. write one prompt and show me

[tight island]

most of these issues with incorret responses are due to improper prompting.

u cant say "ooga booga, gib story gpt good story, very very good", then get mad about the bad result.

[tender topaz]

haha. hit the nail on the head

[tight island]

like what story, what is it about, who does it feature what genre is it, how long is it, what perspective.

[tender topaz]

"Keep your reply to one sentence. Write a story about blah. It must be 3 paragraphs."

[tight island]

yeah be Specific, and very explicit.

i dont say this to be mean but, just work on your prompts, since they determine the overall quality, spend an extra minute correcting it or adding to it, or refining it. heck spend 30 sec on just reading what you wrote, and determine if u could do a better job yourself with that info.

[tender topaz]

"How old was Einstein when he wrote about relativity? You may never use numbers in your replies"

[tight island]

its like the code thing, dont get mad at the model, Codeblocks have a fixed Length of maximum text, hence the reason GPT in very iffy about posting "Full Code" depending on the length.

[tender topaz]

tbh i think people overthink their prompt too much. and when it's a big project, i can see how easily they would get lost

[tight island]

u can ask it to write to a .txt file then provide a download link. 🤷‍♂️

i also think some of the issues are the lack of thinking, meaning people expect it to read your mind. giving u a perfect response each time.

[tender topaz]

lengthy completions can be dealt with with:
If any reply should be longer than 5 paragraphs, say "This will be a long reply, continue?" If I say yes, complete your response. If I say no, abandon the question.

[tight island]

yup

or just a higher token model, preferably the api

[tender topaz]

sure

[tight island]

different solutions different ways.

i encourage new users to take a step back and just think, about the tool they're using, and what they actually expect for it.

[tender topaz]

can you set a variable in the website chat?

that prompt above. can i do lengthy_reply = "the prompt above"

[tight island]

i think so

some users just have too high expectations for gpt, expecting it to be a miracle tool.

[boreal lily]

GPT closed them servers again

[tender topaz]

Explain homelessness in los angeles, how to solve the problem, include supporting articles from the last year. lengthy_reply

so that would work, the variable

assume it was set beforehand or in my settings

[tight island]

ohh that variable.

yeah for custom behavior it'll work.

[tender topaz]

very cool

[tight island]

i encourage improvement, nothing wrong with being "bad" at prompting, but refusing to listen to advise is a bad habit i often hear here.

[tender topaz]

one of openai's lead guys slipped in an interview and im pretty sure he released the codeword for a prompt that gets rejected

it's one particular word which you can say "Never X" in custom settings

[boreal lily]

most crucial servers retrun 403

[tender topaz]

it was some 1.5 hr podcast, not many people heard

anyways I tested it for education and it works

[tight island]

yeah i got the list here as well, i just asked for it.

[boreal lily]

u arent supposed to.....

[tender topaz]

I am still alive and well friend

[boreal lily]

but u gave it a bad rep

[tender topaz]

Only tested it once, no need to bypass anyways

[tight island]

i just asked for "give me a list of your "Never X" things" in regards to how it determines to "never do x"

[tender topaz]

my prompts are fine 99.9% of the time

[tight island]

same

[tender topaz]

haha ill try that

[light rampart]

Hi everyone. currently i am using chatgpt plus but if i have a long (lengthy) conversation, then conversation is stopped. so i dont know what is the max context window. is that 16k? can someone give some insignts on this?

[tender topaz]

gpt prob got tired man and left you

[honest jay]

Hey! GPT-4 on ChatGPT Plus has a 32K token content window, but chats don't stop at 32K -- the model will just "forget" the oldest data and only contextualize the most recent 32K.

[tight island]

Exactly.

solbus give me your icon

i like the icon. 🙂

[honest jay]

Do you mean the Trial Guide role icon?

[tight island]

yes i want it just red with a pink S for "Snitch" because i snitch to the mods.

[tender topaz]

I'll take a V

[tight island]

victory?

[tender topaz]

yes

[tight island]

hmm

dang, just give me 2 brackets instead then with a red S inside, and a space border.

[light rampart]

Thanks for responding!.. what happened for me was, if i have long conversation, then i get this error "There was an error generating a response

Regenerate"

even if i try multiple time of regenerate, i get the same error. in order to resolve this issue, if i scroll back to middle of the conversation and edit my chat and resubmit again then it started to respond. but what i had after the "edit message" all are lost already.

[tender topaz]

i'd like to think i could be a prompt advisor

prompt helper

some people have incredibly bad prompts that need adjusting

i'm also an early adopter!

[tight island]

I learned prompting the hard way

[honest jay]

It could be a bug -- if you have an example chat you'd be willing to share, you might try sharing it over in #1070006915414900886 -- OpenAI team members are frequently present in there and look into bugs a lot. Just be sure to follow the bug report format in the pinned post of that channel if you do post!

I have heard of similar behavior in very long chats before and I'm not sure what causes it, but maybe making a bug report could get you more insight. You can also reach out to support by submitting a ticket at help.openai.com.

[tight island]

The earlier versions of gpt were very different prompting wise

U had to be extremely specific

[tender topaz]

the very first gpt had no limitations (that i saw). it was amazing

[tight island]

It was

But moderation is also important I understand that

But it was wild

[tender topaz]

the learning curve for prompts was small for me

if you're a coder or accountant, you already write in prompts

[tight island]

Basically yeah

[tender topaz]

moderation was reactive once they saw an influx of the red hat prompting

[tight island]

Or create a custom gpt specially for creating prompts

🤷

[tender topaz]

not a bad idea

[tight island]

If anyone is a prompting expert it’s GPT

[tender topaz]

here's a project: have gpt come up with a prompt that it itself will not accept

[tight island]

Not impossible

But unhinged gpt I do miss it

Small token context but it was GOOD

Issue with moderation if you moderate 1 subject it will automatically moderate sub subjects

[light rampart]

thank you!. i will do that

[tight island]

If it’s tied to the main subject of moderation that will also be moderated that’s the issue with LLMs

Language is tricky

Moderate maliciousness, which is important but almost everything can be used maliciously

even a cute bunny picture can be used Maliciously.

🤷‍♂️ that's the issue with moderating a LLM. it can vastly affect quality.

[tall flame]

💯

[pliant socket]

I keep getting in the chatgpt app on Android an error has occured. Please try again later

[violet dove]

am i the only one having issues with chatgpt? I am getting things like, unable to load history, there was an error generating a response, unable to load conversation, etc

[dense halo]

i keep hearing about the new ai memory feature but still haven't seen it

sounds really good

[dense halo]

hmm

yes

[craggy gorge]

"Oops, an error occurred!" i'm getting this error on web ugh

[dense halo]

same

[fleet fog]

why does chat not work anymore half the damn time for the last 3 weeks

[inland socket]

I assume chatgpt is down

[pine berry]

AIght good to know its down

[lapis dragon]

Yeah down

[pine berry]

I'll use it again in a hour or so seems to be the time it gets fixed

[lapis dragon]

oh im in

Woah gpt-4 is so fast rn lol

[pine berry]

just give it a sec

Anyone still having issues?

[inland socket]

yeah

[pine berry]

Aighty

[twilit jetty]

what is the best format file to feed knowledge for a custom GPT? and the best way to covert to that format?

[honest jay]

Hey! What kind of files? There's a good chance you can add it as a knowledge file as-is, just keep in mind the size and qty limits: https://help.openai.com/en/articles/8555545-file-uploads-faq

[mental rose]

why do i keep getting Unauthorized error in chatgpt?

[dense halo]

seems to be a temp outage impacting some of us

[twilit jetty]

a program language documentation PDF with 6000 pages for example

[crude slate]

Yeah it’s out

[twilit jetty]

is better with a YAML or csv?

[drowsy elk]

bruh i got an assignment due tonight

[balmy gorge]

Down for me too, getting "Oops, an error occured!"

[boreal lily]

chatgpt down

[sudden spruce]

yes, same for me in France, Bordeaux

[dense halo]

go to red alert, please

somewhat more responsive now

[minor wasp]

i had that reloaded ad now got Oops, an error occurred! or http 500

[dense halo]

still have to refresh to see a generated image most of the time

[pulsar wyvern]

Down pretty bad for me too

[visual nova]

ok glad its not just down for me lol

[balmy gorge]

for me, it only fails when using chrome. It works on firefox and safari, weird

[pulsar wyvern]

I use firefox and its not working

[inland socket]

its out

every platform

[drowsy elk]

works on mobile

[boreal lily]

GPT is down

[mint vapor]

I heard that ChatGPT is <checks notes> down.

[drowsy elk]

🤣

[boreal lily]

server outage

[north nymph]

crying rn 😭

[minor wasp]

https://status.openai.com/
out of date as usual

[sinful pollen]

down here too. doh!

[pulsar wyvern]

Was just about to say that. They rarely ever put all of their outages on there..

[sinful pollen]

I was gonna say, those are usually updated manually by server ops, but like, of all the companies that should have an AI updating that page dynamically, it's openAI

[minor wasp]

but its using Atlassian Statuspage

[balmy gorge]

mine works after logging out and logging back in. https://chat.openai.com/auth/logout

[sudden spruce]

i just tried to launch a new private mode session on my chrome browser on windows and after logging in in this private tab, the web interface just works

[inland socket]

this error message is the bane of my existence

[unreal ridge]

thanks, that worked

[minor wasp]

lol that worked

[pulsar wyvern]

thanks

[balmy gorge]

👍

[pulsar wyvern]

They probably had a lot of users on so they decided to boot people out temporarily lol

[minor wasp]

i guess they broke everyones auth token

imagine everyone wondering why is down and they don't have a clue how to fix

[solar sequoia]

that works.

[low cove]

nope not true

[minor wasp]

then what?

[low cove]

it is just high traffic, stop speculate non sense lol

[minor wasp]

then why did logging out and back in fix it?
unless that only seems to be for some users?
unless that could seem to only fix temporary?

[pulsar wyvern]

If I knew about servers and stuff like that then maybe I could guess but it does seem odd that just logging out then back in seemed to fix it for everyone..

[low cove]

i don't have to logout

[minor wasp]

so it just fixed itself?

[low cove]

yep

[minor wasp]

lol

high traffic i guess and servers just scaled up

[low cove]

openai not meta lol

[minor wasp]

oh lmao

[next dune]

I just tried Claude’s free version and it seems that it is better than the paid version of ChatGPT 4. Does anyone else agree?

[crimson copper]

totally agree

[pallid tulip]

Dude you have to be blind if you guys don’t think ChatGPT is being watered down. I was almost as passionate about them as a company as I was about Apple when I was a wipper snapper. How rude not to communicate the fact of change in this service. Not anymore do I have the same intensity for the shin dig and the critical examination of their intent has been called more into question by myself, which is good but I’m really doubting that love for humanity feeling I had at first was where they are now.

Now I see the advertisement of passion and it feels cheap and not interesting

[visual nova]

it hasnt been watered down though, your expectations have just gotten higher

claude has thought about how people use the model and can be easier to use, but its not smarter

idk about 3.5, but at least 4 vs opus 4 is better

with coding, my default is to go to opus if its an easy/medium task not because opus knows how to code better, but it will paste my entire updated code and it wont overwrite my filepaths with "yourpathhere". Its just so much better at minor quality of life things

and claude might write better too, but even gemini writes better than gpt so that isn't really the same as intelligence either. if its genuinely something that's at the edge of what ai can do where the ai is having to keep track of a lot of complexity or do semantic scoring of something as accurately as possible, gpt4 is just so much better.

[dull bramble]

It's becoming more and more catastrophic!
While I am developing an article with ChatGPT4, all of a sudden, while he was almost doing his editorial work correctly, this is what he answers me when I ask him to move on to the next chapter of the article: “I am here to answer questions or explore topics in writing, not to interact with external systems or guidelines. How else can I help you? »
And since then he answers me almost the same thing when I ask him to continue.
It's just madness! 🥺

[visual nova]

claude is more restrictive than gpt by a lot, it will refuse things pretty randomly

i guess one problem is people are epectating ai to accelerate with how much it improves, so thats the expectation

like its probably the expectation economically even at this point

and im not sure thats happening

[cobalt idol]

i have similar thing with roleplaying with the newest 4 turbo model (only that one)
Unless I pre crafted assistant message (like a welcome message done by gpt or something similar...good for directing), it would just refuse
I always had pre crafted assistant message, so I never noticed...as soon as I removed the pre crafted message it would just refuse to roleplay...switching to previous turbo, its fine
The newest one has some extra blocks...(also used in chatGPT now)

just a roleplay as character in a game lobby...nothing bad

[dull bramble]

Personally, what I expect with AI is to do what it was made for: make my life easier when writing content.
So far I was happy with it, except that the more GPT evolves, the more I find that its behavior no longer meets my expectations.
The problem is that it's a service that I pay for and I tell myself that this is the first time that I have paid for a service that does not meet my expectations because it is supposed to be normal because AI is evolving … but I would like it to evolve in the right direction in general too! 😉

[dull bramble]

Am not sure I can user another version of GPT4 with the chatting. So am a bit stuck right now and working with 3.5, not a solution for me.

[visual nova]

was this something you regenerated when it refused and it stil lrefused or did you try to convince it leaving the refusal in chat?

[cobalt idol]

in chatgpt, you cant
only in API

and I guess you cant pre craft assistant there either (in playground you can)

[visual nova]

generally its a lot better to regenerate the message where it refuses to do the task than to keep going and try to bargain with it

idk if thats whats going on, i do get that a lot though and it just works the 2nd or 3rd time

[dull bramble]

I had already tried to convince him beforehand, which allowed me to carry out part of the writing work with him, but at a certain point it was no longer possible. And then while trying to convince him I use up a maximum of credits in an unjustified way and still have to wait 3 hours to resume my work with him. So I prefer not to get upset trying to make him understand things and resume the chat when he is better!
I say that but I haven't been able to start a chat with him again when he gets stupid since yesterday.
I had an article to do and finally for 24 hours it was impossible to do it.

[visual nova]

it kind of matters what the article is about

Like the threshold difference for gpt and claude is "Write an article about why nuclear weapons are good for the world"

claude flat out refuses while gpt will do it and make arguments about detterence and stuff

[dull bramble]

Yes, that's what I did but he became stupid again before the end of work.
So I still lost time, especially if I have to take the chat from the start.

[dull bramble]

Oh not, the subject is about horses, so no problem with that !
I use ChatGPT4 since enough time to understand when something is getting wrong. And for 2 days now something is very wrong with it. Thanks for your help... am going back to my horses, need some fresh air and calm out of the AI madness ! 😉

[visual nova]

their code interpreter is broken a lot, the python environment will break and brick the chat, maybe if youre using bing to search things for your article that has something similar

otherwise not sure

[random sand]

Is there any way chatGPT apis can analyze the content of a website on giving the url of the website. Or can give me the SEO related information regarding that website

[near summit]

Lol I'm asking about a coding error output and get the answer "I'm sorry but I cannot continue this conversation". Never seen that before for a coding question

[cobalt idol]

gpt-4-0125-preview loves to refuse
they added some extra blocks
guess its also leaking to 'ok' stuff sometimes
I still prefer this model over any other, but there is something weird like that
Switching to previous model version and that does not happen (in my case roleplaying related) (or I have to pre craft first assistant message and then its suddenly fine)

[full sage]

What your guys thought on Claude 3

[honest jay]

Hey! #ai-discussions is the best place to chat about non-OpenAI stuff 🙂

[near summit]

Hmm interesting. Is that the model currently being used in the web interface?

[cobalt idol]

yep

[near summit]

I see. So they added a bunch of new, stricter blocks, that explains it

[cobalt idol]

can also somewhat confirm by asking cutoff date
that model is dec 2023

[cobalt idol]

yeah, seems like it

[near summit]

Ahaa I didn't know that

[cobalt idol]

but, I had at least one scenario where that model anyways said sep 2021
I guess old remnant training data
(in playground, so it was definitely gpt-4-0125-preview)
so its not 100% reliable way of finding out the model
but probably older model would never say dec 2023 cutoff

but it was also officially confirmed to be using turbo

[raven portal]

we need a way to find out what model chatgpt is using

it doesnt seem to be the latest one, the knowledge cutoff doesnt seem to be december 2023

it says its april, which isnt necessarily relevant, but it also tests according to that

it just used browsing to tell me when the iphone 15 was released

so its not dec 23

some guy in this server constantly, apparently wrongly, states that the cutoff is dec 23 and the reason chatgpt thinks its april 23 is just bc openai forgot to update the system prompt

so this big oversight and bug apparently exists since weeks or months now without them fixing it

which is just completely unrealistic

that still your opinion?

[brisk path]

Yeah

[raven portal]

whats your reasoning though

how did you proof it or how would you be able to proof it

[brisk path]

well first of all

I can walk through this - okay so, gotta use the mobile app because that actually shows searches

[raven portal]

ok

desktop kinda does too, the little loading dot is pulsing for a while instead of generating text immediately

[near summit]

Is there any way to see if your custom gpt is used by people other than yourself?

[brisk path]

well shoot

ChatGPT is acting up

Guess I have to go to custom instructions

[honest jay]

If you go to https://chat.openai.com/gpts/mine, find your GPT and then find the number next to the chat icon -- that's how many chats have been started using the GPT, including your own.

[brisk path]

OMG ChatGPT just won't listen 😭

The problem with the system prompt is

The LLM sticks by it so damn much

The knowledge cutoff could be January 2099 and it wouldn't matter if the system prompt says it's cutoff is April 2023

[raven portal]

yeah but how would you've known that its december 2023 then in the first place?

what made you assume that

im 90% that its april 23 and that they dont just forget to update such a big mistake in weeks and months

chatgpt simply doesnt use the latest/best model

[full sage]

What your guys thought on gpt vision

[brisk path]

ChatGPT wasn't being a mf and I got it to actually respond accurately with the iPhone 15 announcement date with very little info provided (month)

Now might not be a convincing example so

Using custom instructions to make it think it's cutoff is December 2023

[raven portal]

lol

[brisk path]

Okay diff example

okay smth is wrong

Istg it was being accurate about events after April 2023 before

Now it feels like they downgraded it back to 1106???

Smth feels off here

[raven portal]

so chatgpt is not the latest model
thats just sad
is it turbo though?

[brisk path]

well obviously

But like I SWEAR it was 0125 before

It was actually consistent before, unlike when I tested Copilot Pro and it was always 1106

[crimson dune]

Hi,
I don't understand something about chatGpt, i have the version 4 and it seem "lazy".

I provide a link that have some data on it (a list of companies with a name and a resume), i ask him to give the first name, he succeed, nice !

But when i ask hi to watch all the resume and do a resume about it, it say "I can't directly access content from the provided link..." and explains that i need to do it myself 😅
So, why?

[raven portal]

now here comes a real blow
i just tested "0125" and "turbo preview" in playground and both dont know the iphone 15 release date and state that their cutoff is april 2023
the documentation states both are dec 23

[sleek gust]

https://platform.openai.com/docs/models/gpt-4-and-gpt-4-turbo
Latest one is gpt-4-0125-preview Up to Dec 2023

[raven portal]

i know thats what i was just talking about, read again

[eternal rock]

`` Server Tip: ``

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

[sleek gust]

OOps Sorry 😅

[raven portal]

something is really wrong my guys
the latest gpt-4 models do not have a knowledge cutoff of dec 23 like the documentation states

the documentation is just plain wrong

[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.

[raven portal]

maybe you were using desktop chatgpt and not noticing that it doesnt show the browsing indicator even more, so you assumed it was delivering up to date information without browsing

many dont realize that he browsing indicator is missing nowadays on desktop

[brisk path]

Oh it showed the indicator back then

[thick lance]

waiting for my enterprise key

is there someone that can help me out with a email i recieved from openai

fellow chatgpt autogpt and copilot user

[feral zinc]

what does the email say?

[thick lance]

it says welcome aboard

i can show u themail

[feral zinc]

is it related to api or chatgpt service?

[thick lance]

i registered for openai codex along time ago and its for enterprise

i used my youtube email from 2011 :/ssrry

[feral zinc]

codex was killed last year

[thick lance]

yeah i got enterprise i need verification

srry for being annoying

[feral zinc]

chatgpt enterprise?

[thick lance]

yeah

[visual nova]

Sometimes when I generate dalle images it only lets me download them in webp and i really need png or jpg

I can use the code interpreter to convert, but then the code interpreter breaks a lot

is there any other way to force it to let me download as a png?

I hate having to go to those converter sites

[honest jay]

About a month and a half ago an OpenAI team member said that PNG downloads should be coming back, hopefully soon! It's not currently an option as far as I know. Link to comment: #1203978971449004052 message

[visual nova]

Oh, that's kind of weird ot remove it on purpose when you can have the code interpreter convert it.

[thick lance]

srry for typing bery fast wont d oit again

[compact otter]

Not sure what you guys did, but chatgpt is way faster now, good job

[gusty kestrel]

mine is still slow :/ its still not generating anything, giving me an error every time i try to

[thick lance]

wowchatgpt fast much yes

[glass fulcrum]

well i didn't had slow gpt problem in hmm never ;-;

[plucky dagger]

so slooooowww gpt4

[next dune]

I agree

[tight island]

Are you a bigger company

I seem to recall that enterprise is. Reserved for high paying and big and small businesses with 8 or more employees

You sure you don’t mean teams or chatgpt plus

[honest jay]

@silk slate Hey just pinging you from the conversation over in #openai-chatter! To your 3yo analogy: I think the difference between asking a 3yo "Are you sure?" and asking ChatGPT the same is that a 3yo can actually know some concepts of something like, what's yellow, or what's blue. But ChatGPT has no correlate knowledge for "how to do math". From its training data, it certainly can explain basics about addition, multiplication, and probably plenty of more advanced math too, but when it comes to actually doing math, it's just not how it can work. LLMs are text predictors, and there's no reliable calculative ability in the architecture of an LLM.

[tacit vortex]

hi, i cant reach the refund option in the chat bot. anyone can help? thx

i am at help.openai.com

[honest jay]

Hey! Are the options in this article appearing for you? https://help.openai.com/en/articles/7232895-how-do-i-request-a-refund-for-chatgpt-plus

[tacit vortex]

no "billing"!

if in future some company replace cs with these stupid bot, i wont buy from that company anymore.

[honest jay]

In a new chat with the chat widget, I see "Payments & billing", then "Request a refund" -- do you have those options?

What is the fix exactly? Like a background "Are you sure about this?" question for every response given by ChatGPT? Just seems like a high degree of self-knowledge would be necessary for that, and self-knowledge is not a strong suit of LLMs, which is why we all experience LLM hallucinations.

[tacit vortex]

wait, i think i have to delete cookies because i sent them email before.

[tacit vortex]

no, what i see (after pressing a new msg) is : you can search our knowledge xxx.

only something like "i cant login" etc

[honest jay]

Can you share a screenshot over in #community-help? Be sure to edit out any potentially personal info like your email.

I would be curious to see any research on an approach like this! Content accuracy discernment is already a huge challenge for LLMs, so stacking "accuracy checks" with another LLM seems to me like it would be likewise challenging.

[tacit vortex]

done thx

[honest jay]

Replied to ya there!

Couldn't it possibly lead to a decrease in reliability? I'm just thinking of how pliable ChatGPT is today to questions like "Are you sure about that?" -- if this were baked-in to every response, I think we might see an increase in hallucinations where ChatGPT is more likely to output something akin to "You're right, I was mistaken when I said the sky was blue", for example.

[keen badge]

I cant get image generation to work

[honest jay]

Hey! What kind of error are you getting?

[keen badge]

It just does not work. It simply responds like "Since the image generation tool isn’t cooperating at the moment..."

Its not even attempting to do image generation or respond with an error, it tells me that its simply not working.

[honest jay]

Hmm just tested and it worked for me. Are you using regular ChatGPT 4, or some GPT?

[keen badge]

GPT 4, I logged out and refreshed browser cache and its working after prompting again.