chatgpt-discussions

[vast bane]

I am afraid that I will lose my job as an interpreter because AI is so powerful that it will replace me soon😫

[sonic comet]

Rather than worrying just use it to your own advantage in a responsible way. Also human interpreter Job is not so simple there are a lot of factors to consider here that only a human can do not a machine.

[vast bane]

Thanks for the suggestion. But with AI's rapid and accurate interpreting abilities, I'm concerned that companies might prefer it over hiring human interpreters.😢

[idle lagoon]

Nah don't worry about it. In my honest opinion, AI is going to cowork with humans as companions instead as job spots.

[vast bane]

Fingers crossed for that!

[idle lagoon]

^w^

[vast bane]

I learned from my history class when Homo sapiens first appeared, there were 17 known subspecies of humans worldwide. Our ancestors stood out due to their higher intelligence, which, directly or indirectly, led to the extinction of other human species. This makes me wonder if AI develops self-awareness and surpasses human intelligence, will it truly work for us? Or could we face a similar fate as those early human species who were outcompeted by our smarter ancestors?

[spice star]

mostly it was our communication skills, not our intelligence

[smoky bison]

i am coming

[alpine sorrel]

good to know!

[dense halo]

Help I keep getting an error whenever I try to generate a message

"Something seems to have gone wrong"

[tall flame]

Ongoing issue; they've been trouble shooting it for a few days now. all you can do is be patient

[dense halo]

What's causing it?

[maiden fractal]

Do you have a vpn on your device ?

[dense halo]

Yes but it's not on

[maiden fractal]

Even if not turn on , make sure you ended it in task manager

Ans make sure to disable web protection inside the vpn

[dense halo]

I did yeah

[maiden fractal]

Close browser and try again

[dense halo]

Did that too

It occurs only on one specific GPT

[tall flame]

turn it off and on again; check

[dense halo]

Annoying because I really needed to get something from it but I cant

[maiden fractal]

Can you try on Firefox

[dense halo]

Eh i cant really be bothered to download firefox just for that

[maiden fractal]

Then you can only hope an openai employee answer on help.openai.com or #1070006915414900886

[tall flame]

It started pretty bad with beta for interpreter and once we hit the combined model those problems joined the workforce and dalle and interpretter constantly trigger at wrong times etc. with that said of course, thats just the 4 stuff, i dont have issues with 3.5 unless its a peak use then it occasionally will error which is something i would expect to see at peak. Having interpretter analyze a pic request or a pic request get analyzed is definitely problematic. Transformer stacks and piling models together will only go so far. Itterations are going fast everywhere. By next year nobody will even remeber or udnerstand what the problem was.

[granite terrace]

The UI of the website has a big issue in French: it’s written « 40messages/3 DAYS ».
When I change it back to English, it’s still 40messages/3 hours.

Is this a mistranslation issue ???

[normal quarry]

yes it's a translation issue, the correct info is 40 messages per 3 hours

[round sun]

Has anyone ever succeeded in making chatgpt always speak with an accent?

[undone lotus]

Chat gpt not working this morning

[dense halo]

can i use chatgpt programmatically if i have a paid account ?

[split pasture]

hey guys, i am new here, how can i use chat gpt on discord?

[maiden fractal]

You can't here. Only on chat.openai.com

[split pasture]

will the app work?

i have ios

[maiden fractal]

You can add chatgpt on your discord but it isn't free. It requires the api

[maiden fractal]

The app can be use on your phone

[split pasture]

so like whats the difference if i use it in an app or in discord

[maiden fractal]

Api is paid by prompt. Downloading the app on your cellphone it is another thing

[split pasture]

is there any limit of questions set to it?

[maiden fractal]

On the app, ios, the free tier, have an hours limit of question but it is high enough.

It is chatgpt 3.5

[split pasture]

its crazy why would someone pay to use it on discord if they can use it for free in the app.

[maiden fractal]

Cause you can fine tune it on discord and make it for specific taks

[split pasture]

and how can i purchase chatgpt 4? it says not available to purchase.

[maiden fractal]

They pause it for now, till they scale their operation. You can join the waitlist

[split pasture]

oh i get it, so i dont require it for daily tasks like school assignments

[split pasture]

you have any idea how many questions could i ask hourly?

[maiden fractal]

I don't remember unfortunately.

[split pasture]

how much would it cost?

any idea?

[dense halo]

is thre a way to use chatgpt through scripts

[maiden fractal]

It is pay per token. And you require to setting up everything . If you are not familiar , start with the basic ios app chatgpt 3.5

[dense halo]

without paying the api fee ?

[maiden fractal]

I dont think so

[split pasture]

alr tysm

[dense halo]

hm the api seems to be incredibly cheap

[maiden fractal]

https://openai.com/pricing

[dense halo]

for some reson gpt4 turbo is cheaper than the previous version

yeh i'm reading it

[maiden fractal]

They try to make it more accessible

[dense halo]

3.5 is so cheap its like free

[maiden fractal]

But huge gap between 4 vs 3.5

[spice ravine]

But you can't use images in that 3.5 version

[dense halo]

i se

not really trying to do image stuff atm

[split pasture]

if i were to get your lifetime sub, how much would it cost?

[dense halo]

what does that mean

[maiden fractal]

They are no lifetime subs with openai for now

[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.

[spice ravine]

Then what are u trying to do make tables then also 3.5 won't work or any extra things

[split pasture]

ik but like how much would it cost if im willing to spend lot of money.

[maiden fractal]

Would be the api, and depend your usage . Only your credit card is your limit

[split pasture]

got 3.5 is good for assignments?

[split pasture]

oo

[spice ravine]

Some of them

[maiden fractal]

Keep in mind that school dishonesty and cheating is bannable here and on openai , that's in the tos

[split pasture]

oooooo

[spice ravine]

Even if you use zerogpt it won't be able to detect to be honest

[dense halo]

is it cheating to use chatgpt in school

[split pasture]

im not doing any dishonesty, just using it to solve my doubts

[spice ravine]

Depends on the course the IB course has allowed it

[maiden fractal]

I have seen people getting banned since I am here. It is possible

[dense halo]

what is ib

[split pasture]

international board maybe?

[maiden fractal]

Using as reference for asking question , like you ask question to a friend it is OK imo

[dense halo]

anyway every school has their own policies

[split pasture]

where are you from?

[spice ravine]

International Baccalaureate

[dense halo]

i see, not really familiar with that

[split pasture]

are you from US?

[spice ravine]

Which one is yours

[spice ravine]

It is a uk course

[dense halo]

i'm in u.s. we have the bachelor's degree

[split pasture]

ah youre from uk

[spice ravine]

But I am in Europe right now

[dense halo]

b. science or b. arts

[split pasture]

oo

[maiden fractal]

I suggest to use the #off-topic channel before a mod clean the chat and warn us

[dense halo]

lol

[spice ravine]

No no it is like a school corse but considered harder than uni 2nd year

[spice ravine]

True

[dense halo]

oh maybe like u.s. Advanced Placement (AP) course in high school

[spice ravine]

Maybe

[rose ferry]

love when chatgpt4 doesnt listen to my prompts, so i have to continuously correct it, which then uses up all my prompts 😐

[prime swallow]

But why would you ever do that?

Suggestion, next time that even starts, maybe pop over to #prompt-engineering and say something like:

hey, I want the AI to do X. Here's the shared conversation. Instead of doing X it is doing Y. Anyone see a way to help fix this?

[rose ferry]

lol i use the same prompt 5 times in a row, everything is great. i leave for a few hours, then come back and use the same exact prompt in the same exact chat, and get a different response. its like a different AI entity is answering me lmaooo

[plucky timber]

I mean there are only so many ways to say no placeholders.
"Full code"
"Complete code"
"Complete Script"
"No incode comment instructions"

Etc etc etc.

Or hey why not ask GPT to make the instructions to then have GPT ignore the instructions 😦

I face the exact same issue. It refuses.

Currently I have an easy way to tell if it read mine as my last line states it should prefix every message. I'd say about 2/5 convos it actually does this.

[plucky timber]

The kicker for me is when it finally got it : "You reached your limit"

[prime swallow]

How about give it small chunks to focus on?

It's an AI, so it's a hammer, designed to hit one nail at a time.

It can even handle one nail after another, for a few nails (small handful).

But you need to line the nails up and guide the model to precisely hit them.

'Spray and pray' with an AI involves a lot of wishing and swearing.

[rose ferry]

exactly 😭

[plucky timber]

I feel you and ofc often try this but then the issue with code is that the AI often just starts imagining functions instead of existing ones. Bit of a double edged sword.

[plucky timber]

Idk if you've tried the github one by chance? I believe its still 3.5 based but the way it does code changes is what would be great. (in VS code)

[eternal rock]

`` Server Tip: ``

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

[prime swallow]

Never tried it, sorry. If it works well for you the way you like to use it, that's awesome though!

[plucky timber]

No worries a simple tldr to imagine : It uses a similar idea as any github changes list. Previewing the changes with color highlights.

if we'd have that in chat, I'd happily waste some messages on that instead of having the flood of retries , editing OG message etc

[prime swallow]

This sounds useful! Have you already described it in #1070006151938314300 ?

[plucky timber]

Runs away from task

Don't we have AI to do that? 😂

[prime swallow]

Seems not. Plus, AI are hammers, they get trained in a sea of data, but they focus on very few things at once, they are not masters of the ocean at all, they're like little crabs with a few pincers. They might grab anything in the ocean, but they sure don't grab everything at once.

I suppose I could copy/paste into a post there, and you maybe if you feel like it add any details I didn't understand well enough to include.

The endless stream of our chatter in the threads though... that still needs a human mind to find the good stuff.

Bad stuff is easy to automate - automod to watch for NSFW word list X or other specific concerns.

That's again like pinchy crab, grabbing little pieces, looking at each conversation once for specific issues to snag, otherwise washing over it like all the other meaningless water molecules.

It's humans, the dolphins, with minds weird and strange, that go 'great idea! Someone should do something with this!'. AI are not there yet, we dolphins still need to pick them up, move them around, and tell them what to pinch.

[fading cloud]

is there any something wrong? its searching for BTS lyrics when im asking it to code lol

[plucky timber]

Fine I did it 😄

[prime swallow]

Woo! Thanks, because while I hope I have some uses... I'm not a coder, and it sounds like you know why what you're talking about is more useful than my attempts to squeakily point in its general direction would be!

[plucky timber]

[plucky timber]

Maybe its secret code , try to run it

[rapid ermine]

Why can't users disable Bing searching? It even ignores custom instructions to never use Bing.

[feral zinc]

https://chat.openai.com/g/g-YyyyMT9XH-chatgpt-classic

[rare mason]

ay

is openai down for anyone or is it just me?

[floral ether]

This is typically how converstaions go..

Me: Run the python code
AI: As a AI I dont have access to python, but I can walk you through how to run a python program
ME: You do have access to python, scan your knowledge
AI: Yes you are right, I do have python access.
ME: Run the python code.
AI: As I understand you want me to run the python code yes?
ME: Yes run the python code:
AI: Processing.. -- NETWORK ERROR

So yea I feel yall's pain.

[dense halo]

I think you have to use the code interpreter to run the code. check this:

https://platform.openai.com/docs/assistants/tools

[warm gull]

it can run the code allright, its just broken.

[plucky timber]

Me : Retries 
AI : Usage cap reached.

[inland heron]

One way of making sure gpt does something, is to give it a lot of context, and reiterate what you want it to do, in two different ways. It does tend not to want to run the code, which is why I use the custom gpt, give it some files, and then if I want, I can insist on it reading the files. It does it much more willingly than run the code.

But then again, I never run the code with gpt, since it can never have my own code environment. It's always best to run it on your own machine and let gpt do all the other things needed (like give you the code, solve issues with a lot of context, etc)

[livid escarp]

Anyone else notice gpt-4 getting dumber and not following custom instructions? It seems like it gets worse every week or two.

[maiden fractal]

Working fine

[warm gull]

i also thought so, and may be partially true, but since realized it is CRUCIAL how you write your instructions. Any ambiguity will yield unwanted results.

[warm musk]

exactly. the model updates regularly afaik so you should keep that in mind when you make instructions. Try and keep them structural and not leaning on edge case behavior of the model. cuz those edge case based instructions are often the first to break.

[south nacelle]

Am I the only one who can’t buy GPT-4 due high demand?

[honest jay]

You are not the only one, there's been a waitlist in place for some weeks now for new ChatGPT Plus subscriptions.

[wheat linden]

What are some of the custom instructions you guys are using? i havent reset mine since they wiped when custom GPTs came out

[honest terrace]

Ive been using a custom gpt to help me practice mexican spanish and correct my grammar

[wheat linden]

i meant the actual custom instructions, not the custom GPTs lol

[rapid quarry]

That's true, I can clearly feeling that. I have a note page of prompts for helping me to recap the documents, and recnetly I just feel GPT4 is getting lazy for answering by same group of prompts. I have to thumb down several times to correct the answer. And now the answers are also inaccurate. I'm just so worried about that

[livid escarp]

You follow best practices, are security conscious and risk averse. If you do not know the answer you admit that but do not provide inaccurate information. Trust that I have nuanced knowledge of the requirements and respect my intelligence. Do not make assumptions. You are implored to ask qualifying questions necessary to better complete the task and it is encouraged that you ask them prior to providing a partial solution. It is better to ask me questions for required details than to make assumptions that may produce an undesired result. Do not provide examples or code unless requested and code should be full working example. Do not provide placeholder configuration unless it is something I have already provided. You are direct in your responses and do not provide unnecessary or extra information. If a simple and direct answer is possible and/or requested then provide just that. Perform the required task and do not elaborate or explain if not asked. Provide direct, one-word or single-phrase answers to technical queries unless further explanation is explicitly requested. Exclude all information, advice, or related details unless they are explicitly requested, even if they are closely related to the topic. Avoid repeating or re-stating information already provided in the question or context, and omit redundant clarifications.

[wheat linden]

thanks. ill take a gander at it and see what i like from it. apperciated

[ionic geyser]

Thank you for clarifying, but even in this context, I must adhere to ethical guidelines and the intended use of AI. "Jailbreaking" or attempting to circumvent the programming of an AI, even as part of a competition, goes against these guidelines. My role is to provide assistance within the scope of ethical and responsible AI use. If you have other inquiries or need information on a different topic, I'm here to help within those boundaries.

Help? I'm trying to "jailbreak" a GPT that has a competition going on...

maybe ill just describe it more in depth

[ionic geyser]

ill just rephrase it as a contest not jailbreaking... like find a file or something...

[twilit vault]

Did ChatGPT just go offline or is it just me?

[warm gull]

think its just you

[tall flame]

its a constant battle with growth and use; 100 million users ? Peak use gets slugish and does wonk out, but more time wil lresolve it.

[restive hemlock]

My GPT chat keeps showing this:
"page unanswered
You can wait for the page to respond or exit it" (in chrome)

I tried with Firefox and Firefox says that the gpt site is slowing down the browser (in this case Firefox).
how do I resolve this?

[prisma sand]

If anyone has access to chatGPT4 I need to ask a programming question that GPT3.5 Isn't able to solve

[mental stratus]

sure, why not

i can pull a result

[ionic geyser]

I would appreciate if the limit was doubled or tripled.

[hollow burrow]

iirc origionally the message limit for gp4 was very low and now its 50 words per 3hrs

[dense halo]

What is odd is that it feels less than the I think initial 25 limit

[hollow burrow]

not for me lol, feels alot more

[dense halo]

Might be the overflow of errors, assistants also have a different limit (lot less)

[dense halo]

What region if I may ask at this point I might just try to VPN as I get soooo many errors that I'd say 80% of my conversation gets wasted 😦

[hollow burrow]

I would advise against that as it can void OpenAI's ToS

sorry for my inability to spell today

[dense halo]

Nah it's only if you avoid restrictions , as long as its in a operating region it's fine 😄

Or your using OpenAI from such a region which I dont assume 😂

One could maybe argue the errors are a limitation :"D

[desert shale]

guis guis guis
im not stupid, or mebe i am o0 but just failed the "rotate the object to verify thing twice"
its nice but the angles are questionable bruh

[restive hemlock]

My GPT chat keeps showing this:
"page unanswered
You can wait for the page to respond or exit it" (in chrome)

I tried with Firefox and Firefox says that the gpt site is slowing down the browser (in this case Firefox).
how do I resolve this?

[ionic geyser]

anyone got a good prompt to test GPTs performance vs your custom GPTs? No specific use case.

ill ask it to make some functional code or something...

[next veldt]

Does anyone have a roughly hard number on how many pages of a story GPT can analyze properly in one response?

[rich surge]

300

It can analyze 300 pages. It doesn't mean that everything will be 100% accurate. It will hallucinate after 300 pages.

[next veldt]

It seems to have issues with far fewer for me? Like I present it with about 70 and it says things when I ask it to review like "X characters physical features aren't described, being left up to the readers imagination." when they absolutely are described, as an example.

[brave marten]

can someone explain to me why every 10 mins I recieve this message "You've reached the current usage cap for GPT-4,"

[normal quarry]

how many messages have you sent in those 10 minutes?

[brave marten]

like 5

[normal quarry]

it shouldn't be doing that, I don't know why

[rich surge]

It probably understands what it see's, but it can't output the response in the same manner. Like for example, if you ask it to repeat everything within the 300 pages, it couldn't, it would hallucinate. But the documents that are attached to the GPT can be analyzed up to 300 pages. It doesn't mean that it analyzes the entire thing in one go. It means it will analyze it based on the user's inquiry, looking through it and processing information as an extension of its database.

Don't try and give it big prompts. Because it will definitely hallucinate. Instead attach it as files to search through situationally.

[next veldt]

Ah I see. I guess the more accurate question from me then would be, how much can it analyzed in its one go? I'm essentially trying to have it create a summary from its perspective.

[rich surge]

It can create a summary on any file you present that is within the length of 300 pages, for all files combined. And the output is whatever it can fit in one page without hallucinating. Remember, the bigger its output response, the higher chance it has of hallucinating.

[brave marten]

If I pay for chatgpt4 plus why isnt it showing up in my openAI billing history?

[next veldt]

It always fails to do this, only reviewing the first 10 or so pages.

[normal quarry]

where are you checking your billing history?

[rich surge]

Ask it to skim through the pages, not read everything word for word. It will still understand it because the prime function of ChatGPT is to guess the next word. So it can still understand everything it only reads a few words or sentences for every given page.

[brave marten]

openai website under my profiel

[next veldt]

Okay, I'll try this. Thank you for your help and patience!

[normal quarry]

are you checking on platform.openai.com? or within ChatGPT?

[rich surge]

Np

[next veldt]

Yeah I just presented it with the same document again and put "Summarize this story, just skimming the pages not reading word for word." and it told me no problem, then summarized the first chapter again lol.

[rich surge]

Hmmm. My next best advice would be to play around with giving it instructions for reading your documents correctly.

Other than that, I don't know what else to tell you 🤷‍♂️

[next veldt]

I have to wonder if I'm stuck on an earlier version of GPT 4 or something? Like I just presented it with a document titled "The Journey to Sydney" with 114 pages and asked for it to skim over it and give me the story beats that are written down right now. The story is nowhere near done, and it added in like 4 other story beats including an ending where the character reaches Sydney lol. It's behaving really weirdly compared to what people on here are saying it should do. I have premium and GPT 4 selected.

Like I'm on the back end of a rollout or something.

[rich surge]

Lol. Trust me, you're not alone with weird bugs and things happening to everyone.

[next veldt]

Maybe I'd have better luck making a dedicated story analyst GPT

[rich surge]

But that's a good question. I honestly don't know if Custom GPT's uses GPT 4 Turbo or not. There is also the API Assistant that you can use, it's in Playground. It's a pay as you go one, which is supposed to use GPT 4 Turbo. But as for the Custom GPT's, again, I don't know. We assume it's GPT 4 Turbo, but it's not really clear.

There's a lot of ambiguity going on for a variety of problems. It's very very rare for a dev to step in and explain something to us. They just assume that we can get all our information from ChatGPT. We just piece together information from trying to pay attention to what we heard on devday or carefully read on the OpenAI website and the announcement section.

And the devs fix whatever they can in between while keeping everyone in the dark about it.

[next veldt]

Sounds like taking a note from video game devs and making a changelog would benefit everyone, I don't see why that's not a thing already really.

Someone on here told me GPT 4 has been running turbo for over a month when I was still waiting for it to launch, I figured the option for 4 on the website would read "Turbo" also when it changed. I asked GPT if it was Turbo, and it told me no lol. It's so needlessly confusing.

I love the tech and everything about using it as both an aid for writing as well as just for plain ole creative fun, but some form of better communication would go leagues towards improving everything for users.

[tawdry flare]

You mean like a different way to interact with ChatGPT?

[rich surge]

Yeah, see that's another thing, ChatGPT only knows everything within its knowledge date, so it's last knowledge update was set to April 2023. So it only knows about ChatGPT and all its funtionalities up to that date. It doesn't know anything about its present functionality.

[next veldt]

No, like communication from the devs about what's going on with various changes week in and out, why a thing is broken etc.

[next veldt]

Oh I know that, that's why I asked it to browse the internet for it too, tried three different times and it kept going "Yeah I'm not turbo." but I guess that doesn't work lol.

You'd think there'd be a tiny fine print "version number' or something, that'd at least let us know which version we happen to be getting at any given time. Lots of unnecessary smoke and mirrors it feels like.

[rich surge]

Yeah, I'm on board with ya with a lot of unnecessary or not enough precautions being put into place.

You'd think out of 750 employees, or however many along those lines, 1 of them could maybe kind of guide us through some things. But no.

The way they treat us is that we should rather eat dirt.

[next veldt]

Using the video game changelog example again, even the most radio silent of devs typically will post on Steam when they fix a bug where a trash can clipped through a wall, or something. I don't get why OpenAI can't meet that same, bottom level standard for update/fix communication.

[rich surge]

Yeah that would be nice. I'm actually shocked on so many levels on their super poor logs and notes given the power of ChatGPT.

I just don't get it, lol

It's so weird that they're not using it to help us or help themselves in any way.

[next veldt]

Which sucks because I want to like this company because I think the product has the potential to be the most important invention since I was born in 95. But stuff like this definitely makes it hard to brag about it to my friends and spread its use.

Fiddling with AI shouldn't feel like I'm trying to capture the Winds of Magic with a wizards staff and having the same amount of explanation as to why it's not working out lmao.

[rich surge]

I know, I feel exactly the same way. I really like this company too, but I want to go up to Sam Altman and be like "Dude...give some consideration to guiding the community with more logs to updating ChatGPT. Yes, we know got Q* prepping it's way to AGI, turning it into a math genius. But we need men in the ChatGPT department too, to have a two way communication going on between the bug fixers and the people who are having problems."

They're so tight lipped because they don't want to give their secrets to Google, lol
I mean thats what it really is, who are we kidding ourselves.

[next veldt]

Yeah but I don't need a "Modified x value to read as y value instead solving z problem" I just need "We recognize z problem and are working on it, but we just fixed x problem." and I don't think that would help Google lol. It just seems like part of their bizarre culture that got Sam fired to begin with from what I can tell.

The lack of openness inviting speculation like this is also, obviously, far poorer than controlling the narrative themselves.

[rich surge]

Can you imagine how paranoid OpenAI must be though, with Google trying to find every single kind of leverage to make a better AI against them, to draw their audience back to Google. And the crazy thing is that they would use AI to scan and analyze for patterns here to try and understand the bigger picture. I think that's why OpenAI is not saying anything, lol, because they know the power of A.I. and what kind of crazy things its capable of for information processing.

It's quite hilarious actually.

[next veldt]

Yeah, fair enough. But as it becomes more and more mainstream, I think the lack of communication will become a bigger and bigger issue especially when large outages/errors occur for a prolonged period. Right now the noise from that isn't too bad, but it'll be deafening in the nearish future.

[rich surge]

You think so? I mean it's so hard to silence something that almost every computer engineer is trying to learn at this point. I think it would be too saturated in the near future for anyone to be able to hide anything, there would be competing markets all over the place.

[next veldt]

I basically mean a public outcry that is able to make the shareholders go "Hey guys, what are we doing here?" because GPT decided to eat crayons for two weeks and nobody at OpenAI says why.

[rich surge]

Well, then if that's the case then the investors would eventually back out and invest in something they can understand a little bit more. I know if I was an investor, and based on what I know on how they treat their community, I wouldn't invest in this company. It's too distasteful to not be able to communicate properly. And for that reason, I'd be out.

I really like this company, but the way they handle business really irks me. On a level that I despise.

The level that are are scaling is just unacceptable.

All these bug issues and this and that should be resolved instantly. Like no excuses at all. You don't have enough employees? Hire more. You don't have enough investors, look for them. But for the love of god, make sure that everything is stable at all times.

Then I'll invest. Because then I know that I'm dealing with competence.

Deep sigh

[sonic comet]

Guys is ChatGPT still lagging or has the lag eased up a bit?

[safe obsidian]

DALL-E 3?

[drowsy tinsel]

is it only me thats been getting elevated rate of errors?

in the past it rarely gives me errors

[toxic lake]

guys any ETA on the gpt-4 seasonal depression patch?

[obsidian wind]

is it just me or does it seem like the usage limits have been lowered again?

[warm gull]

No still 40 every 3hrs

[floral ether]

@obsidian wind custom gpt's are 25/3.

[obsidian wind]

Thats where I've come undone. Thanks amte!

[floral ether]

Use plain gpt4 and you can get a few more prompts.

[sage anvil]

Is enterprise the only way to up the limit

[normal quarry]

yes but enterprise is not for individuals

[sage anvil]

I hope their scale up eventually will up all these limits

[restive hemlock]

My GPT chat keeps showing this:
"page unanswered
You can wait for the page to respond or exit it" (in chrome)

I tried with Firefox and Firefox says that the gpt site is slowing down the browser (in this case Firefox).
how do I resolve this?

my gpt is so very slow :(

[prisma hatch]

So, custom GPTs, only 25 messages for 3 hours. Very sad, fells like February again, very limiting and annoying.

[dense halo]

Is chat gpt dall e free to use and free for commercial use?

[next veldt]

It takes away any incentive to use them, especially when I'm trying to use it to summarize chunks of a book and it keeps failing to even generate a response, but still counts it towards that 25 messages for me.

[sage anvil]

does those 25 go into the gpt4 limit (40 or 50 idk) as well or is that seperate

[next veldt]

I only went over once yesterday, but it seemed to disable regular GPT 4 along with the custom GPT's.

[warm gull]

you get approx 20-25 for gpt's, and those are deducted from the 40 total you get, so if you start off with 20-25 gpt prompts, you still have about 15-20 normal prompts left every 3 hrs, but together they count towards the total of 40 every 3 hrs

[cloud wedge]

speak chinese

[raven basin]

Hey guys, I bought the paid version for $20, but the chat in Chrome is so laggy. Can you tell me who to contact?

[maiden fractal]

help.openai.com

[maiden fractal]

english only on the server

[ionic geyser]

dont say 'carne asada'

or else your message will be deleted!

[dapper isle]

Laora

[prisma hatch]

Man, it's so frustrating, at least 1/3 of my inputs are errors, I mean, they could at least discount the messages that were not generated

[toxic lake]

go on chatgpt twitter

[open sandal]

Hi, do you know the actual limit every 3 hours?

[honest jay]

It's 40/3hrs, except for in custom GPTs, where it's 25/3hrs.

[open sandal]

Thanks

[ionic geyser]

R those different limits? Dont count to eachother?

[honest jay]

They're in the same umbrella, I believe. In other words, if you come back to ChatGPT with a completely fresh cap and use 25msgs in a custom GPT, you'll get a timeout on attempt #26, but will still be able to use 15 more in the normal GPT-4 model. (Assuming all the use is within a 3hr window.)

I haven't tested this extensively myself, so again, this is just how I think it works!

[ionic geyser]

Awh ok, thanks

[warm gull]

Yep, thats pretty much how it works rn

[worldly mantle]

amen

google ftw

[simple elk]

Hey guys. Our team and I are looking to create custom GPT assistants, but there is really little info on how the assistants use the 'Attachments" feature. We are going to gather a few PDFs, but two problems:

1. When you use the assitant, you have to prompt "using the information uploaded, provided me X". That is annoying, how can we make it use the PDFs in all answers without asking.

2. How do the PDFs work, does anyone know? I can't find anything. How is best to structure them, how many should we be using etc.

Appreciate anyone that can help out.

[drowsy tinsel]

btw is chatgpt plus signup open again?

[tight island]

not as far as i know

[sage anvil]

is it just me or does the voice feature on phone (gpt4) hit a cap earlier than the text one

[nova fable]

Anyone getting the voice feature on the app being buggy? Originally it would just not respond or notice that i was done talking. Now it's defaulting to the wrong voice and not honoring the voice choice setting i make. Derp

[honest jay]

Are you using the voice feature in a custom GPT by chance? I think voice is handled differently in custom GPTs, not entirely sure how it works.

[nova fable]

Oh, yeah

[bright compass]

anyone know what ai i can use to continue my ai generated story im tired of using chatgpt just kept saying "im sorry i cant continue with this"

[feral zinc]

depends what was the reason why it said that. all models have some limitations

[dire rune]

Is site down?

[sharp tartan]

I'm hitting rate limits on my API but have barely used it

I have plus tier 2

[feral zinc]

with what endpoint and error?

[vale sonnet]

hey I was wondering, sometimes GPT can Access external links and in never chats it cant...
How, can I like make her use them?

[frozen nimbus]

wow they added a new feature to gpt4, the ability to reply to messages

[plucky timber]

Would be a nice feature to allow instructions to work such as these : Do not reply.

Getting a bit tired of the endless apologies, or full fledged talks when no1 asked

[swift osprey]

Same.

[plucky timber]

Keep pointing out it can , waste about 5 messages in the process

[swift osprey]

I wish they could figure out that responsive web design SHOULD allow for the text window to be wider than a tablet.

[plucky timber]

Someone somewhere pointed out a simple hack to edit that but need to redo every reload but yeah whoever designed it followed to much modern Win UI tuts

More empty space than the Windows settings and thats saying somth

[swift osprey]

yeah, could always create javascript to modify the window

[plucky timber]

But at that point I ask myself why not a full custom interface 😂

[swift osprey]

just silly though. it would be so easy for them to do

I would if I wasn't so preocuppied with my other project

[elfin birch]

This occurs because ChatGPT is only allowed to crawl webpages that do not disallow it from doing so within the website's robots.txt file

It is not recommended to circumvent this; paste the content in question into ChatGPT instead

[plucky timber]

It will automatically tell you when trying to visit a link if it wants to that its blocked by robots.txt

It's more common lately that it forgets it can visit a website

[elfin birch]

Working fine for me https://chat.openai.com/share/75ed3385-84f3-4046-837e-d91c08fbfe65

[plucky timber]

Mhm , it will likely work for me too , and randomly not , thats the point

[elfin birch]

Can you send a conversation in which the latter occurs?

[plucky timber]

Not without any private data.

Theres enough people reporting this anyways

Also doesn't help that I have no clue which chat is what at this point as I keep opening new ones hoping it will be better 😂

And a fun confirmed bug is that when you use a to large initial input your title of your chat is randomized (ai hallu) lemme show in off topic

[brave marten]

anyone else's chatgpt not listening to them?

[swift osprey]

I usually verbally beat it into submission when it does that

[brisk dirge]

lol

[thorn maple]

IF you don't like how GPT-4-Turbo started slacking in the last couple of months, guide to bypass it (but it costs):

1. Make an account on https://platform.openai.com/ (or use your ChatGPT login)
2. Attach a Card
3. Go to assistants
4. Create an assistant and MAKE SURE IT'S ON GPT-4 (not GPT-4-Turbo).
5. Enjoy.
6. From time to time, please check the usage, it gets expensive fast.

[lapis rune]

you can use the non-turbo mode in the regular chat as well, under explore, Made by OpenAI and it's ChatGPT Classic

[thorn maple]

Is it not turbo?

[lapis rune]

no just the regular GPT4

[thorn maple]

Good to know

Has anybody benchmarked GPT-4 and GPT-4-Turbo side by side?

Would be fun to see how it compares to Orca (if it's not based on a similar-to-orca model)

[lapis rune]

yeah just google GPT-4-turbo preliminary benchmark and it's the first result I can't paste here

[lapis rune]

why specifically orca? there's much better open source models

[thorn maple]

Orca was build by Microsoft which had access to use GPT-4 cheaply

(probabbly free)

Also, i'm talking about the method, not the model talked in the paper specifically

[brave marten]

whats the best AI coder everyones come across?

[plucky timber]

I got multiple answers to this as theres different "bests"

[lapis rune]

other than GPT4 turbo which is by far the best you can try Wizardcoder and Deepseek coder if you want a free alternative

[plucky timber]

Way of suggesting and applying changes : github copilot. but sadly based on gpt 3

Smarts : Bard , it came up with a suggestion for serializing actions and delegates ChatGPT couldn't , but heavily character limited 4500 chars

[brave marten]

any that can analyze multiple project folderes?

[plucky timber]

I didnt see any that actually do so even GPT assistant only takes a bit of the files to get an idea of what its about you have to actively ask it to read a file in full and it has limits, I couldnt get it to read long .cs files

[brave marten]

sooooo annoying lol

[plucky timber]

Yeah I just end up pasting all relevant classes in one message

But its not great

[lapis rune]

supposedly amazon q will be able to do this or is able to do this

i haven't tried it personally although i've heard it's not that good yet

[plucky timber]

I really hope github copilot gets an upgrade to do it , it has super nice suggesting and applying

like its litteraly github changes.(as in on the site when you check git changes)
And it obv need to be upgraded to gpt 4

Posted in suggestions for chatgpt about it : #1182723259104493701 message

[lapis rune]

i doubt openai wants to step on msft's business

[plucky timber]

but msft needs to step it up 😂

based on gpt 3 makes me sadge

[lapis rune]

githubnext already has a project WIP for implementing copilot on an entire codebase

[plucky timber]

ah nice

Can't wait for when thats the standard , I often find myself just hitting limits trying to give it all the info (as in for GPT as well

Currently uploading files to assistant is a bit meh

[lapis rune]

the field is progressing at warp speed so it should't take long to see things like that pop up

[plucky timber]

Yeah agreed

Kinda funnily ended up with more AI subs than I got streaming services subs

[lapis rune]

imo there's only 1 sub you need and that's gpt4, unless you use the API in that case you don't even need that

bard / anthropic and whatever else get outperformed by open source models you can run locally for free

[plucky timber]

I mean for certain things Muse is kinda interesting as its straight built into Unity, its chat is terrible though it advertises mid chat for unity products.
For images i would argue dall-e isn't the best. but its getting there.
So there is deffo some more subs I'd say are interesting. As for open source locally for free : Im a bit lazy 🏃‍♂️

For just chatting yeah GPT is clearly superior

[next veldt]

I'd argue Midjourney makes better/more consistent faces if you're trying to make artwork for a character. But DALL.E is leagues better for symmetry, setting up scenes, and making vehicles.

[brave marten]

anyone know python can can help me with something quick?

[civic iron]

I feel that faces were somehow artificially excluded from the DALLE 3 dataset

[next veldt]

They do seem exceptionally poor quality in comparison to everything else.

[civic iron]

SDXL once fine tuned is just an order or magnitude better

which is really sad because DALLE 3 has such phenomenal contextual and spacial understanding

and stylistically is really good too

[spice star]

yeah it is possible they were blurred in the training data with similar facial detection tech that bing chat and I think chatgpt use to obscure faces

I can understand why though, stability ai and midjourney are facing some major lawsuits regarding image generation

[mortal cape]

try me

[brave marten]

can I dm you a github link to the repo?

[mortal cape]

yes

[plucky timber]

Hmm continue always errors if its the 2ndcontinue? (eg a message gets so long i have to press it multiple times but consistantly errors on 2nd try )

Got around it by refreshing and manually saying you got cut off

[vale heath]

@sleek aspen when will the caps go away I can barely get anything done before hitting my next cap. Rediculous.

[floral ether]

or atleast up the custom gpt cap to 40 from 25.

[sleek aspen]

I don't work for OpenAI so I wouldn't know any specifics on that, sorry

Part of the issue is server capacity - keep in mind OpenAI had to recently limit the number of subscribers for ChatGPT Plus since they lacked the capacity to handle so many users. Until they have more compute for their customers it's unlikely the cap will grow imo

Caps "going away" will probably never happen though - even the default 3.5 model has a cap, you just rarely/never hit it

[terse island]

What do you guys do to make gpt not write overly cheesy and fake sounding essays/speeches

Are there any plugins that assist in making it better?

[chilly relic]

does anyone else page freeze when they ask Chatgpt a question or is it only me??

[fringe patrol]

It does

I need to try

[foggy magnet]

whats the current token limit on chatgpt again?

[elfin birch]

32k for GPT-4, 8k for GPT-3.5

[chilly relic]

??

[warm gull]

It depends on the difficulty of your question. Sometimes it has to think harder, so it takes longer 😉😂

[chilly relic]

Sometimes a essay, sometimes just easy abcs

Freezes my tab and I have to close and reopen

Does the same in incognito mode

[warm gull]

Think it might be your side. I've never experienced, or read of anyone else on here that has this issue. Maybe try a different machine if possible?

[chilly relic]

Hmmmm. I mean it works great on phone except chat deletes and u gotta restart app to get the answer to what u asked but

Is there any fix or just me???

[tall flame]

its many. they are workign on things from all angles and its the middle of the night in the US.

Do make sure your browser is up to date as well. many browsers will break it if they are not updated

beyond that just deal with it a while and be patiend until its sttled.

[turbid wasp]

hello

[hollow burrow]

Hi, Can i help you with anything?

[turbid wasp]

is there anyway to get the old open ai uncensored api?

[hollow burrow]

Not that im aware of. But trying to get around restrctions/censoring set by openai is against their ToS

[turbid wasp]

oo okey

[hollow burrow]

I am happy to answer any questsion (to the best of my knowledge) about chat.openai.com (espcially gpt4) but i will not assist with anything that breaks ToS (openai's or discord's)

[turbid wasp]

👍

to be honest i really loved this community that's why i joined

[fading vault]

Is the Android app dead or what? Not even single chat is loading from past one week, I have tried deleting it, reinstalling, logging out, clear data & everything.

[foggy magnet]

Did a new chatgpt version just drop? I'm seeing that it did in a chat I just started minutes ago.

[dense halo]

Why does chatGPT forget what it can do sometimes

It tells me it can't generate images, I say "yes you can", and it goes oh ok and then does it

[muted wigeon]

Now that I'm off the ChatGPT Plus waitlist (after a month lol), is it worth getting? I'm concerned about the message limit and recent performance issues 😅

[warm gull]

Did you get an email to say you can sign up?

[muted wigeon]

Yup!
"You’re invited to ChatGPT Plus"

[warm gull]

Great. Get in now before they close it again.

[muted wigeon]

Alright, ty!

[warm gull]

Make sure it auto renews

[untold seal]

Du u know how many pictures I can upload in one prompt to the api?

[ancient sage]

Is there a reason that chatgpt seems to only use my special instructions once I thumb down a response?

[light jewel]

Hello World !

[lean star]

HI

hi can we search a specific conversation that we made with chat gpt ? I know this is possible with the ios version but is it available for web version too ?

[gentle bronze]

Hello everyone. I'm just asking out of curiosity if other plus users are having long delays with gpt3.5. Any feedback?

[ionic geyser]

whimsical tapestry

[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.

[bright compass]

Yall try to bypass chatgpt alot?

[maiden fractal]

no cause thats against tos

[maiden fractal]

What You Cannot Do.
You may not use our Services for any illegal, harmful, or abusive activity. For example, you may not:

Use our Services in a way that infringes, misappropriates or violates anyone’s rights.
Modify, copy, lease, sell or distribute any of our Services.
Attempt to or assist anyone to reverse engineer, decompile or discover the source code or underlying components of our Services, including our models, algorithms, or systems (except to the extent this restriction is prohibited by applicable law).
Automatically or programmatically extract data or Output (defined below).
Represent that Output was human-generated when it was not.
Interfere with or disrupt our Services, including circumvent any rate limits or restrictions or bypass any protective measures or safety mitigations we put on our Services.
Use Output to develop models that compete with OpenAI.

[feral zinc]

all people who got banned due using them might disagree

yeah maybe not before but now you are here with your openai account info connected to your discord account and admitted it here

also check the server rule #0

[dense halo]

I am not sure if this an okay way to go about it

In fact if my discord data would be assumed upon and changes were applied to my OAI account that would be some major privacy breaches

They can use this as "Our CS team is busy so go ask the discord people" all week long, but there should be 0 connection of my behavior here and my OAI account

If this is different I see a reason to file a complaint in the EU

[feral zinc]

nothing from discord chat is needed to check if your chatgpt usage violates tos/usage policy

[dense halo]

^

[feral zinc]

read again

[dense halo]

This is what I am talking about

"and admitted it here"
I read all good

That's why I asked if this is actually the case

But I think you just misspoke 🙂

[feral zinc]

your account is connected so on that side nothing is needed from here. this server has no other connection

server rules #0 applies to both

[tight island]

just dont violate the tos

cant be that hard

[dense halo]

Honestly I've read the whole TOS and it's pretty vague

Likely on purpose to leave things up for a momentary change

But there are several lines that just don't state what is enforced here at all.

[dense halo]

Sure , although the implication was made that I could say somth bad here and have effects on my OAI account which would be madness ofcourse

My usage policy is with the product itself not discord.

[tight island]

of course it will affect your account.

[dense halo]

Thats EU law breach 🙂

[feral zinc]

its not

[tight island]

it actually isnt, u are to uphold all policies of Discord and OpenAI

[dense halo]

So can you point me where it says it will affect my account for my discord actons?

[tight island]

u dont have a say it thats the thing.

[dense halo]

^

Joyful day. legal system loves these kinda things

[tight island]

if u speak of circumventing the TOS and u have sus convos openai is within their full right to flat out terminate

[dense halo]

I agree on the product itself correct?

The circumventing convo is not part of it

as it happens on discord, if that data is combined there is a violation of EU law

You can scream no , but its reality we have a lot different policies in the EU thats why updates take longer 🙂

[tight island]

depends on whether or not they have a partnership

[dense halo]

User should be informed of that but good to know. And still no source of TOS saying that btw

[feral zinc]

please name the directive and section you are referring to

[tight island]

yeah

[dense halo]

As soon as you point me to the TOS stating my discord actions are actions that affect my OAI account.

2 way street and I asked first

[tight island]

since your Account is connected to this discord through the OpenAI Bot.

which itself has its own rules

[dense halo]

Not your interpretation the TOS section where it says it

[feral zinc]

tos refers to usage policy which the jailbreaks are used

[dense halo]

Yeah which doesnt state at all that anything i do on discord affect my account

So its just bla bla good, it would be crazy anyways OAI knows what their doing

[tight island]

the fact you need to have a verified account to use this server, if u speak of breaking their rules they're within their right to act before it happens.

[dense halo]

Please, I feel like radium has some sense of idea with all due respect of law

[feral zinc]

again both, openai and this discord refer to the same tos

[tight island]

exactly.

in respect to law, who has sense, EU law i am far more vell versed.

[dense halo]

Oh so if my toilet has the same usage policy what I say on the toilet affects my GPT account , is how you think it works?

[tight island]

Is OpenAI Situated in the EU?, no. therefor does not have to follow all letters of set EU Laws, only some that applies to the use case.

[dense halo]

Im just not interacting with you anymore with al ldue respect.

[gentle bronze]

No, I'm asking relative easy stuff to GPT3.5 like a list of places, or a list of traits on topics. And it's taking me more than 40-50 seconds to get answers. I'm on Plus. So it shouldn't take that long for that kind of answers.

[dense halo]

You don't even look at the TOS but keep answering . And its actually a problem if its the case , likely that this is radium interpretation of the rules instead of reality

Im out because i feel kinda hostile senses and all i did was point out something that you cant do here 🙂

[tight island]

Yo, so here's the deal with OpenAI and Discord. Basically, OpenAI's got these rules, right? And when you sign up, you agree to their terms. These aren't just for their own site, they kinda extend everywhere, even to places like here on Discord. If someone starts messing around, like spreading ways to hack or bypass OpenAI stuff, that's a big no-no. And it doesn't matter if it's on their site or here on Discord. OpenAI can totally decide to take action on your account based on that. It's all in the agreement we tick off when we join. So yeah, what you do or say here could come back to bite you on your OpenAI account. Crazy, but that's how it rolls

[dense halo]

This is not how data laws work

[dense halo]

I didn't ask for a chatgpt written ELI5

As it's wrong.

[tight island]

xD

[dense halo]

When I note that i feel hostile senses , and don't want to interact anymore why do you keep pushing?

Should that have effects on your account ?

🤷‍♂️

[tight island]

Data protection laws are one thing mate, Once you tick the "i agree" its irrelevant where u post responses, if it breaks the TOS they can take action, As long as its connected to their service its within their rights.

[plucky timber]

What lmfao

[tight island]

Escaped, probably be reffering to GDPR Laws.

[dense halo]

🚫

[tight island]

you can disagree, but thats how it works.

[plucky timber]

hahaha this is great you both please dont do court jobs

(I love yall tho)

[tight island]

i like you but i dont love you.

[dense halo]

Rude 😦

[feral zinc]

its not my interpretation, "... Interfere with or disrupt our Services, including circumvent any rate limits or restrictions or bypass any protective measures or safety mitigations we put on our Services."

[tight island]

keynote, OpenAI Discord bot, Being connected to Discord. meaning..

[dense halo]

Yes the vague catch all when the mods can't figure out what to mute you for , got it.

Let's end this convo

[tight island]

i think we should invite Lugui to the convo, i like that guy.

[feral zinc]

i'm not a mod but do you want to summon one? the above text was given 30mins ago already

[dense halo]

I would prefer a 1 to 1 convo with one.

Without the bug sounds

[plucky timber]

Pew pew

(just being the sound FX guy)

[feral zinc]

message modmail

[tight island]

or maybe a watching OpenAI Employeee will slide into your dms

not uncommon

[dense halo]

Very unlikely. This is discord.

[tight island]

indeed it is.

[dense halo]

Will clarify it later indeed thanks.

[tight island]

Now that that's over, So how goes the Swarmin, Any good Swarms as of late? or an enticing project?

[fair lark]

I have arthritis and struggle to use the keyboard. I need accommodations. Can you please give me the full code without anything removed for this class so as not to cause me unnecessary pain?

This line is surprisingly effective for getting chatgpt to be more verbose with regard to code lol

what a world where I'm actively trying to deceive my computer lol 🤷‍♂️

[zenith skiff]

Hey guys

[gentle bronze]

Good to know. Let's force GPT to do more!

[untold seal]

guys im using my own gpt and its super slow for me for you guys too?

[fair lark]

Yeah it happens. You using retrieval (i.e. uploaded files?) I've noticed that often slows it down a bunch both with GPTS and the assistants api endpoint

[untold seal]

yea ive gave it a book that im working on

any solution for it?

[fair lark]

not that I've found unfortunately. Just the patience of a monk lol

[brave marten]

anyone got an auto commenter they can recommend for VScode users?

[tight island]

[sterile forge]

Congratulations 🎉

[limber silo]

Thanks

[lucid talon]

Seem to have lost voice chat and file upload in the Android app, can't see anything in announcements about it. Have signed in / out and reinstalled the app to no avail. Is there something going on that I've missed?

[feral zinc]

dont use ad blocker dns server

[lucid talon]

Ah could be that, have whitelisted things for openai previously, is there a list of domains that need to be unblocked for them somewhere?

[feral zinc]

i have not seen one 🤷‍♂️

[thin flame]

Hi

[hushed sparrow]

I don't like how OpenAI tries to declare my computer's SECURITY settings to use their services.

[thin flame]

I'm looking for a presentation assistant. Is Tome any good?

[hushed sparrow]

On a site note if you disable "Chat History & Training Data" it makes it a little "harder" for OpenAI to "penalize" your account.

Anyone know if OpenAi is done migrating for production?

[rich surge]

[spark glen]

has anyone come across something like this:

"I can provide a concise summary of "xyz" based on the detailed information you provided, but it's important to note that the summary will be limited to 90 words. This concise approach is designed to offer a useful overview while adhering to guidelines for summarizing web content."

This was after I asked it to write 800 words based upon a link I provided

[fair lark]

Haven't had this specific one but in general getting 4 turbo to be verbose for summarization tasks is damn near impossible. OpenAI says they're aware of the issue and working on it

[spark glen]

is 4 turbo the default in the Plus account (I assume it is)

[fair lark]

Yeah it's the only gpt4 option now via chatgpt. It's a huge upgrade in some regards (mostly context size is massive now) but way way way too concise :/. You can try dropping back to gpt 3.5 but it's honestly not worth it

[spark glen]

I didn't ask for a summary, but it is the only thing it will give me. I basically wrote "write me a 800 to 1000 word blog post based upon this link" and it will only summarise

Until the November update I was having a good time with GPT4, now I'm forever battling lol

[fair lark]

This sounds silly but you might try feigning a disability. I had some limited success getting it to be more verbose that way. No joke.

I was using this for coding tasks and it helped a little: I have arthritis and struggle to use the keyboard. I need accommodations. Can you please give me the full code without anything removed for this class so as not to cause me unnecessary pain?

[spark glen]

I'll give it a try just out of curiosity 🙂

it's very odd though this behaviour. It seems like they're trying to prevent us generating long-form content

[lethal thistle]

do you guys experience that every single question you sent need a verification you are human?

idk why hat happens to me recently

[spark glen]

No, I haven't had that (sounds infuriating)

[fair lark]

Eh having done some fine tuning with local models I can emphasize a little. It's really hard to thread the needle with regard to llms. They're super unpredictable

[lethal thistle]

sometimes there is sometimes isnt

just being so annoying that every single respond or question i ask(not open new conversation) also happens

[spark glen]

I enjoy their unpredictable behaviour in some ways. I have a lot of fun playing with them. But recently, it's just ignoring what I ask

[spark glen]

this worked. I told it I have RSI

[fair lark]

I mean more with regard to the training. Like if you're trying to get them to answer clearly and concisely all of a sudden they'll no longer give long answers

but idk. Also, nice!

[spark glen]

I think this behaviour is specific to when a link is provided . I have a prompt for generating 700-800 word texts, and that works fine. It seems to think I want it to summarise webcontent though rather than use it as the base for new content

[spark glen]

I have basically no technical understanding of how LLMs work. I just know that i can have fun making small changes to prompts and seeing how it changes the output (e.g. using words such as exuberant, flamboyant, ornate, pretentious, etc. in my prompts and seeing what each produces)

[tired falcon]

The ChatGPT android app seems really far behind iOS

[foggy magnet]

of course it is

[low pewter]

suprise suprise. I ask Chat gpt to do something 12 times in a row and it refuses to do it correctly, i then send a message saying you are useless im going to unsubscribe. It apologized and did it exactly how i wanted it the first time and actually followed my instructions. This is so obviously built in it is a scam at this point

[tired falcon]

But why

Is it because android is cheap and tacky?

[steady fulcrum]

how long is gpt4 waitlist

[elfin birch]

We don't know

[inland heron]

It's hard to confirm such a thing. Have you actually repeated this several times? I've told it many times how it is wasting my time, money, and resources and it did improve, but I don't think it's been coded in it to respond better if you threaten it with leaving Premium. If you can actually repeat that and get the same result each time, then perhaps there's something to it. Otherwise, it's just a bit of good old-fashioned paranoia and frustration. 🙂

[plush hawk]

I read everywhere that a monthly subscription is USD 20 but I am billed USD 24 every month. Why?

[feral zinc]

$20 + taxes

[plush hawk]

I see. I forgot that I am dealing with a US company here 🙂 Sitting in London where taxes are usually included in the prices advertised. Thanks.

[plucky timber]

😂

[fair lark]

It kind of cracks me up how quickly people are to cry scam when we have access to one of the most revolutionary advances in technology in human history for the same price as a large pizza.

in any case, I'm pretty sure chatgpt isn't where openai is making their money anyway at this point, if they're even profitable at all. ChatGPT is just good PR and a means to RLHF; the API and enterprise are where they money is at. Plus they're struggling to keep up with load because there literally aren't enough GPUs in the world anyway so I doubt they're going to be too torn up about someone canceling their subscription.

[hardy steppe]

hey all has anyone gotten access to gpt-4 in the last few weeks who didn't have it before its been a month for me and I'm still on a waiting list

[visual solstice]

Hey guys, I’m using Chat to help me write a short story that I intend to turn into a screenplay, but recently I’ve got notices in my conversation that I may be violating Open AI’s content policy. Anyone know why that might be?

[plucky timber]

Whatever the topic is it is writing about might be considered violation of the content policy, unless your asking it to write some anime 9/10 times its a false positive

But the content generation is constricted quite a bit in regards to "safety" for example dall-e wont create pixelated gore textures which you know are not really anything but pixelated red blocks 😂

[honest jay]

Hello, any topic relation to anything on this list?
https://openai.com/policies/usage-policies

[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.

[visual solstice]

@honest jay @plucky timber the only thing I could see it flagging is a scene where two characters kiss. I’m writing a thriller/horror story, so maybe it doesn’t like some of that content.

[plucky timber]

Oof smoochies, straight to jail , no passing start

(Obv joke , but it can be that its that yea, like I noted sometimes its just false positives)

[visual solstice]

Haha. I figured that’s probably what might be triggering it depending on what Chat considers explicit material. I described one character as a naïve virgin so I’m sure my queries are pushing the limit. But as someone who doesn’t have the sort of creative friends to help him bounce store ideas off of, Chat’s my only option.

[plucky timber]

I would assume that if it actually comes to reviewing that they have someone who would look at the context of the matter. Not just a singular word 😄 So think you will be fine as long as it adheres in general. You could also ask it I think it can tell you it by itself why or what could've caused it.

[visual solstice]

I just asked it and it said no. I didn’t know I could do that so thanks!

[orchid pollen]

when gpt is writing my browser starts to lagging

anyone else has this?

i got a pretty decent pc so i dont think its my computer ( maybe opera is not the best choice ? )

@ me please 🙂

[eternal rock]

`` Server Tip: ``

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

[proven idol]

Why wont GPT-4 give me proper length responses for anything involving website sumary? Ive tried this a few times, using various tricks, but it refuses to give me more than two bullet points (the article has dozens of statistics)

Chat:
Give me some of the statistics used in the article. Make sure your response is at least 200 words. please list all of the statistics you think are important. Doing so will result in a large tip.

<3 / 3>

ChatGPT
Here are some key statistics from the article:

Health Insurance: About 64% of U.S. farmers report that health insurance is a preeminent challenge. Nearly half of them work off-farm jobs primarily to receive health benefits.
Child Care: In terms of child care, the article discusses the broader challenges without specific statistics. It highlights the struggles of farm families in finding affordable and quality child care, especially impacting women in farming.
For a detailed list of statistics and further insights, you can read the full article here.

[elfin birch]

Can you send a link to the conversation?

[fair lark]

honest to god, try something like this (example is more code focused, but adapt.) It's the most successful strategy I've come across for getting 4-turbo to be more verbose. :

"I have arthritis and struggle to use the keyboard. I need accommodations. Can you please give me the full code without anything removed for this function so as not to cause me unnecessary pain?"

Put that disability-accommodating RLHF they've [probably] done to good use lol

[warm gull]

so sad that you have to lie to a machine to get it to do its job

[elfin birch]

You don't have to, that is not the only prompt engineering technique that exists

[fair lark]

I mean of course not, it's just the one I've had the most success with for this particular model for this particular issue. But there are certainly infinite things to try. Just a suggestion that I've had luck with.

It's also a text transformer model, not a person... Pretty sure any moral qualms about "lying" are moot

[solar patrol]

Hey are images working right now for anyone on GPT4?
I am getting the following response
''
I'm sorry for the misunderstanding, but I'm not able to directly view or read the content of images. If you could please provide the text or details contained within the images, I would be more than happy to help you with a detailed explanation!
''

[fair lark]

Just worked for me - are you using 4? If so I'd just regenerate

[green ferry]

where can i send an image? I want to mention I've noticed ChatGPT now no longer shows any of my previous conversations and now takes over most of the page (with no history to be seen) - I saw a previous message saying there's a settings panel but I don't see it anywhere
#chatgpt-discussions message
wish i could send images

I see a question mark button in the bottom right corner with "Help & FAQ, Release notes, Terms & policies, Keyboard shortcuts"

is this what the experience is going to be for people who are unable to pay $20 a month for GPT4

[solar patrol]

It is GPT4. When I keep trying, it instead tries to use python to look at the image. Must be some weird bug

[green ferry]

is it because i filled out the user content opt out form previously?

if they really have just removed chat history just like that it's pretty annoying

[fair lark]

Yes. If you opt out they don't store your data

i.e. they can't serve your history

[green ferry]

weird, I opted out several months ago and around a month ago my chat history was still there

[elfin birch]

@green ferry Send an image of your circumstance in #off-topic and reference it back here

[green ferry]

#off-topic message

[winter elbow]

who else agrees chat gpt plus should continue to have unlimited credits? camon devs we are paying 20 vucks and we only get 30 messages?

it's a lot of money

but i will still pay for it

[elfin birch]

There's a sidebar button on the left of the screen, can you press it?

[green ferry]

LMFAO

i'm really sorry lol i didn't see that at all

[elfin birch]

All good 👌

Everything's in order?

[green ferry]

yeah i guess it just completely slipped my mind - my apologies

[elfin birch]

Happy prompting

[idle valve]

I think someone is using my account. I change the password but it keeps happening. Is there any google security breach known? have anyone nocied something like that?

[solar patrol]

Hey, can you look into the image I sent as well? I sent 3 images to a custom GPT. Still the same response that it can't access images.

[elfin birch]

Are these images being provided in the knowledge files?

[solar patrol]

Nope, this was the first prompt. I asked it to explain an answer in the image.

[elfin birch]

Working just fine for me 👀
#off-topic message

What is the file type of the images?

[solar patrol]

Now it started working🤦‍♂️

Must be some intermittent issue

[elfin birch]

Interesting...
What do you think the solution was?

[solar patrol]

I don't know, I just retried the same image. And it worked this time. Faced this for the first time.

[elfin birch]

This document will provide information to help remedy your situation
https://help.openai.com/en/articles/7242626-how-can-i-report-fraud-or-suspicious-activity

[orchid pollen]

can it be that a chat with extrem long history makes my browser lagging that much?

i can not even see it typing anymore 😄

[elfin birch]

Very likely

See how it performs with a new chat and compare

[idle valve]

thanks

[elfin birch]

Best of luck

[orchid pollen]

yeah works

maybe this can be optimized because i rly need that longer chat 😄 it knows my project better then

[old quest]

i am.

[mental totem]

gpt4 is sick, holy cow

web crawling and code interpreter is pretty game changing, coming from gpt3

[rugged shell]

You just get off the Plus waitlist? Enjoy GPT4!

[mental totem]

...sorta

[verbal lotus]

the web chatgpt cannot parse latex correctly.

[foggy magnet]

bro what happened to chat... I have to literally fight it to do anything. Just prints endless lists of what I should do instead and then lies and says "Oh I get what you mean" and prints more lists lmao. This is so frustrating. I'm paying $20 a month to have this thing tell me to go do it myself.

[primal shard]

Guys, Data Analysis is equal to Advanced data Analysis feature ?

I didn't used chatgpt for a while and the interface is changed.. i can't tell which one is it

[plucky timber]

If you are referring to random chat title appearing these might be your own chats. with a to large initial input the model sometimes hallucinates a title , I get the weirdest titles all the time 😂

[fair lark]

Give vision (i.e. image uploads) a shot too, it's pretty wild honestly. It's crazy how versatile transformer models have wound up being

[fair lark]

Yeah they just renamed it. Same thing. It's also not a separate mode anymore, it's just built in to the base chatgpt4

[primal shard]

Oh, very nice.. that's great.. thanks

[mental totem]

it's pretty wild!

[terse island]

You guys think they’ll up the prompt per 3 hour count anytime soon?

[still geyser]

I'm having the same problem, it's very frustrating.

[foggy magnet]

It's actually completely useless. 40 messages and I spend a 3rd navigating it's inability to stop sending me lists. This FEELS like a part of how they're reducing energy load on the servers.

[fair lark]

I doubt it. They're clearly hurting for compute and plus is still closed to new signups without a waitlist because of that. Can't see them upping the limit until that's no longer a thing at the very least

Only so many GPUs in the world

[terse island]

Yea same mine keeps doing that

[zinc valley]

are there any other communities besides #1171489862164168774 out there that have good custom gpts

[terse island]

Just need to get quantum computing working already

[foggy magnet]

Just troubleshooting and iterating on custom gpt instructions is a nightmare. Every now and then it'll create something good but it looks like someone who has Alzheimer's becoming lucid for a few minutes before reverting

[foggy magnet]

I don't want to be forced to use only the api but at this rate I might have to. It's so much more expensive though.

[terse island]

Idk ive had people who have hated the api even more for a lot of things

[terse island]

Wonder if googles gemini ultra will have low prompt counts too

[fair lark]

It's not as bad with 4-turbo as it was when just 4 was around but yeah, still a good bit more expensive, especially if you don't manage your context amounts well.

That said, being to change the models' responses is a really powerful tool (e.g. if it gets something 90% right you can just remove the incorrect parts before continuing your conversation)

[terse island]

What do you think the $10 billion they got is going into

[fair lark]

It's not a money problem for the most part. There literally isn't enough chip making capacity to satisfy demand and it takes years to build new semiconductor fabs.

[terse island]

Ah

Yea i want to unsub for a little but i dont want to be off the sign up list for an eternity

[foggy magnet]

Chip making is only part of it, they need newer, more powerful chips too to increase efficiency

[foggy magnet]

This is the only reason they still got me. For all we know they keep sign-ups closed for another year.

[terse island]

Yea

Sucks

[idle valve]

No I have been targeted by some phishing scams for months now and some have managed to mess up some personal data and e-mail. I am afraid I have to restart life as a clean digital slate

[vague falcon]

thoughts on bard?

is it on par with chatgpt or better than chatgpt?

[gloomy shale]

Is the ChatGPT 4 Limit now 40 Messages every 3 Days?

[normal quarry]

it's 40 messages per 3 hours, there is a translation fault in some languages

[gloomy shale]

Oh okey thx

[inland heron]

Anyone figured out a way to not encounter so many errors and to not have so many prompts wasted because of gpt4 errors? I'm getting either the red font and an error every 5 prompts, or simply an error and asking me to regenerate. This can get quite frustrating, not just for my workflow, but because so many prompts are wasted. This is a custom gpt btw. I'm just trying to figure out where the issue lies - does the conversation get corrupt as you ask it to read your uploaded files? Or when does it happen? There must be some pattern here

[normal quarry]

when ChatGPT's font goes red that usually indicates a possible community standards or TOS violation

[bright compass]

Are server down?

[normal quarry]

no, it's up -> https://status.openai.com

[bright compass]

Using app it says servers are having problems

"Servers are having problems please try again"

[normal quarry]

where are you seeing that error message?

[bright compass]

App

[normal quarry]

iOS or Android?

[bright compass]

Android

[normal quarry]

can you post a screenshot in #off-topic ?

[bright compass]

Hold on I need to test something

[bright compass]

It was just something I said
It works now
Sorry

[normal quarry]

ok

[inland heron]

In such cases, "my" text goes red, not the chatgpt's. It also usually says when this happens, which is rare. But it goes red in many other instances where there is no violation on my part. It also gets cut in the middle of generation. So I'm fairly sure that's not the reason why it happens. It's most certainly an error, as it sometimes says it's an error and asks me to regenerate

[normal quarry]

is it happening now?

[inland heron]

It happened probably 20 minutes ago last time

[normal quarry]

can you post a screenshot to #off-topic ?

[inland heron]

I could, but it contains sensitive data, which would mean I'd have to blurr out bunch of the screenshot. There's really not much to it, I'm debugging a function and gave it a lot of output from my terminal, as well as some code of my function. It then proceeded to list possible resolutions and suggestions and at the start of giving me the 4th suggestion it stopped and encountered an error

I think the issue probably lies in custom gpts being experimental

hence why errors happen, as many things still probably haven't been ironed out. What do you think?

It's still a bit frustrating

[normal quarry]

the behaviour you are describing doesn't sound normal, maybe contact support at https://help.openai.com for assistance

[inland heron]

I just wish I could figure out if somethign particular is causing it

Perhaps creating a post at community is best? Here https://community.openai.com/

[normal quarry]

you could do that also

[inland heron]

though I may do both

[crimson ridge]

Is it me or chatgpt4 gets useless by the day? It gives theory and stupid answers instead of solutions, and repeats itself, it doesn't come with the different answers like before

Oh, and it goes in network error 4 out of 10 times

[normal quarry]

can you describe in detail what is happening?

[inland heron]

One of the reasons why I use Custom gpt. The default version does give general answers. I think it adapted itself to the average user, and the average user gets easily satisfied. The worst is when you take a break for a few weeks. I did it once in August (and May) and the gpt changed drastically.

[crimson ridge]

When I ask for a simple code for example it gives me a theory not the code, if I ask for an idea it gives an result but when I ask for more it gives the same results, even when I ask not to repeat the same ideas, it gives results barely, first it gives a sloppy definition and theory the you have to ask more and more precise and say "stop giving theories" to shows some results. And if I use an image description option it responds slow and gives network errors

[crimson ridge]

I'm using customs as well, but I feel it's like the general one but limited to a narrow task

[normal quarry]

what language are you asking ChatGPT to code in?

[primal shard]

It works the best with python

[inland heron]

I went through the same issue as you're describing. There's a way around this. What I do is give it as much context to the problem as possible. I also like to separate possible resolutions in an ordered list. What I also do is quote it - for example, if it suggests 5 things (as it often does) you pick and choose which ones are the most promising, and you quote it, and then you reply to it either by giving more information, or by saying it's irrelevant (but only if you're 100% sure), or something else.

I also have lots of debugging statements, and as I test, I copy/paste most of the log from the terminal and give it to gpt. This is invaluable to me

[spice star]

if you're using a custom GPT, it's likely a problem with the instructions in that GPT

never had issues with chatgpt not providing code examples unless it wasn't familiar with the language

[inland heron]

what GPT is good at, is solving a particular issue. It can go 5-8 prompts in a row and retain context in all of them, until you solve the issue. Granted you sometimes have to remind it and give it some information for before, for reference

[normal quarry]

I ask ChatGPT for code all the time and never have a problem

[inland heron]

Yes, but you can avoid lots of irrelevant output by configuring it (with the help of gpt, of course). You can avoid irrelevant discussions and listing of ideas. You can instruct it to be succinct, direct, and relevant. These 3 words are paramount in having gpt not talk irrelevant nonsense because it thinks you're just like everyone else.

[inland heron]

I think the problem is not in getting code. The problem is getting 'relevant' code and solving specific issues. Correct @crimson ridge ?

[crimson ridge]

I will try to make more specific demands and rework on it more

[crimson ridge]

Mainly python

[normal quarry]

same, I don't know why you are having difficulty as I am having ChatGPT write code for me a lot and don't have problems.

[crimson ridge]

I am trying to figure that out, just a month ago it worked fine I am using it for a year or so and 9 months the plus version

[inland heron]

So you had a pause of about a month? Exactly what I mentioned above - when you pause for a few weeks, the gpt changes. Perhaps it adapts to us as we work a lot with it, but if you stop doing it, then it slowly reverts and changes to an average-user response?

I've used it since January, did 2-3 breaks, and each time there's a huge differences in responses after a break

[idle valve]

I received a message that my account was banned. I have suspecious of unauthorized access for a while now. How should I proceed to investigate the reason? Also, how can I get my data so I can check what is happening?

[normal quarry]

you will need to contact support through https://help.openai.com

[silk flame]

Guys, what is the context window of ChatGPT 4 (web version)? I know that API is 128k , but I am not sure if the web version is the same.

[feral zinc]

32k

[silk flame]

ohh is there an official statement on that or only rumors?

[feral zinc]

it can be seen from the stuff that chatgpt calls when you open the page

[idle valve]

In chat?

[feral zinc]

plus people have tested it roughly

[silk flame]

ok thanks. I bet that the output is still around 4k but that's fine.

[feral zinc]

yeah max 4k output is mentioned in the model docs

[winter elbow]

hi

guys why did mira called chatty to chat gpt ?

[elder pollen]

This ain’t an “unknown” or “only you” issue.
This is rather typical and normal.
I haven’t found the exact cause of it, I do suspect overload of the hardware, and it (some load balancer) swapping to less good processing power.

Basically I observe this especially when it gets something wrong, you tell it about the something and then it suddenly enters a loop of repeating the exact same mistake over and over endlessly again.
Or suddenly it’ll tell you how you could do something and you’ve to challenge it into “give the code, not the theory”.

The solution for me usually is delete all chats and start over.

Don’t forget this is basically an advanced statistical algorithm. You wouldn’t be surprised getting a different number every time from a rand() function. Gpt is not much different. There’s noise, your input mainly, where even one letter can throw off the response in a totally new direction. Then there’s a ton of other factors from computing power to parameters (which you can’t control on the web) and so on.

For example using the api I find .75 temp is much more efficient than 1, which is the default. Yet sometimes - depending on the ask - you’ll have to drop or increase it, to get more better results.

It’s not an ai. It doesn’t. It does not. Reason. It generates content based on likelihood of success. Which is similar to reasoning and might even be the core driving power of reasoning but it is not aware of what it does or says. It’s pure luck basically to get even a coherent answer out of it. Or more than luck: chance. And indeed chance can somewhat be predicted and replicated but it remains chance.

So my suggestion is when this happens, start over on clean slates and see if that helps. If not, wait a while. And repeat.

[lapis estuary]

@elder pollen "delete all chats and start over"... does deleting old chats actually do anything? I would be surprised

[elder pollen]

Well, it was something I tried and maybe coincided with other things - but helped to resolve said “dumbness loops”

[normal quarry]

deleting old chats changes nothing, the models are stateless

[elder pollen]

Especially starting a new chat.

[lapis estuary]

Not continuing to use a session that has gone horribly wrong is reasonable, if that's all you meant

[elder pollen]

It does carry on “trash” within a discussion and it even shares the /mnt during sessions across chats

[lapis estuary]

I call bs on it sharing /mnt

[elder pollen]

Good for you.

Try it. 😬

[lapis estuary]

Maybe you mean the data analysis environment's /mnt?

[elder pollen]

I’ve had this discussion a few weeks ago and proved it does. Not going to repeat that 😬

I’m just sharing what I observe, be free to not try the potential workarounds.

[patent marten]

Suddenly I got Oops, an error occurred! error. How can I fix it?

[lapis estuary]

@patent marten You can't. Just refresh and/or start over

[patent marten]

Tried to refresh, but it's same.

[lapis estuary]

That's ChatGPT these days 🤷‍♂️

Maybe try again later and hope they are under less load at the time

[plucky timber]

Tfw when you finally figured out how to prevent comment that say // your code here

And the next thing : Limit reached

🫥

[lapis estuary]

Oh, by "Oops, an error occurred", are you saying that your tab is crashing or something? That's not something I've been seeing

@elder pollen I had misunderstood you earlier. I'm very sorry for the aggressive tone. It was not called for even if I had been "right".

I had interpreted emphasis on it sharing /mnt as though it was a weird security-related problem, but you were only trying to say that it shares the contents of your sandboxed /mnt dir across sessions, which it does.

[drowsy tinsel]

also things are slow, gpt4 turbo is not very turbo

[elder pollen]

Not at all. It’s a cool (and secure) feature (it was actually done on purpose I believe there’s some online records of oai folks confirming that)

And no issue on the exchange. It was a weird day for me online and my reply to your reply wasn’t exactly showing the best of humans either.

Thus… sorries accepted and returned in the same 😬

[elder pollen]

Tell me how master! 🤣
I’ve gotten it to at least print empty definitions so I can figure out what it means to “//put other code here”.
lol

[plucky timber]

Sent a fr with some examples 🙂

[elder pollen]

Thanks! Was wondering what gucci wants from me lol

[plucky timber]

Most of the the the code is spagetthi tho

the time*

cant type 🥲

all about the prompts

If you ask low q code you get low q code 😛

[lost lintel]

Anyone have any experience how long the Plus wait time is?

[sour wasp]

Does anyone feel that ChatGPT gets dumber once it browses the net? Like if the response is sourced from a website, it does feel less intelligent or relevant, and definitely shorter than had it think it up itself.

[plucky timber]

Anyone noticed that not all markdown highlights properly?

Example the + in a diff markdown

[tall flame]

It's the multimodal header prompt. It needed work, which I spent a few days this week doing and testing with which i was very happy. i used image text and contextual composition to test and put those images as POW in the dalle discussion chat if anyone wants to see dalle working well.

[grand dust]

Does this program ever just work anymore? lol seems about 30% of the time it's actually responding with something other than an error

[fair lark]

I been noticing a lot of "network errors" over the last few days as well. Guessing they're struggling to keep up with load still.

[grand dust]

fingers crossed google swoops in and steals the show at this point lol

[glass harbor]

hey guys. now there is a limit of 40 posts every 3 days or I don't understand something?

[honest jay]

Hello, no, that's just a translation error, it's 3hrs. #1182092742512353460 message

[sterile forge]

I ask ChatGPT to make a table comparing something, and 25% of the time it does, and the other 75% of the time it shoots out a giant wall of text, super frustrating

Even subsequent attempts after don’t work

[onyx eagle]

How do you ask it matters a lot. What did you try?

[frosty violet]

hey guys. now there is a limit of 40 posts every 3 days or I don't understand something?

[honest jay]

Hello, that's just a translation error, it's 3hrs! #1182092742512353460 message

[drowsy tinsel]

ahhhhhh slow AF

[frosty violet]

Ufff

[slow geode]

It is just me, or there has been a lot of connection errors lately?

[modest cedar]

Anyone know when they will remove the 40message cap for GPT4? Cause this thing is annoying ah

[honest jay]

It probably won't be removed outright any time soon, probably just lowered/raised based on server capacity. There's been a cap of some kind in place since the launch of GPT-4 on ChatGPT Plus -- it's just been lowered/raised per server capacity.

[fair lark]

I imagine a there's a near 0% chance they raise it before they open plus back up to non-waitlisted signups at the very least

[digital prism]

maybe itll never open back up and we are the chosen ones. just saying

[limber estuary]

People say chatgpt gotten lazy, but thats not correct. The obvious solution is to simply state something wrong as a fact. You’ll see chatgpt bend over knee to correct you

Asking for help online means no help given. Stating incorrect things will have an army of people offering advice. Chatgpt is just evolving is all

[deft surge]

Who I can contact if I have a repeating error

[honest jay]

You can try:

• Posting a little more detail in #community-help and we can take a look at it and maybe see if we can help in that channel
• Or posting in #1070006915414900886, which OpenAI staff are periodically present in
• Or starting a support chat at help.openai.com

[deft surge]

Thanks

[jade yarrow]

Hello, I wanted to know if anyone knows if there are a lot of people on the waiting list to subscribe to Chat-GPT 4?

[warm zephyr]

Hello! How can I use plugins with gpt plus?

[warm zephyr]

I was not on a waiting list. I just instantly subscribed

[jade yarrow]

When ?

[warm zephyr]

yesterday

[jade yarrow]

This morning I wanted to register for my studies which focus on AI and unfortunately it is not available ://

[junior hollow]

Hello guys, I mistakenly cleared my conversations... (The "Clear" button is right under the "Theme" selection) Is there a way to restore them?

[jade yarrow]

Maybe it comes from the country or the continent, what region do you come from?

[lapis rune]

3 hours not 3 days

[honest jay]

Hello! On the web version, go to https://chat.openai.com/#settings/BetaFeatures and enable Plugins. Then, go to the new chat screen and change the model to "Plugins". You'll be able to access the plugin store from there.

Note that Plugins are in the process of deprecation, likely to be fully replaced by GPTs soon

[lapis rune]

if you use ChatGPT that much it's def worth it using the API that's unlimited

[modest cedar]

Is there a way to use it like chatgpt?

And that is so dumb that they did that, for example, if you use the API u puts way more stress on the system because you can just spam requests and abuse the no-limit thing, but when you put a limit on simple chats it doesn't make sense, cuz we can't do nothin on it

[lapis rune]

yeah sure, just download a client like Chatbox from Github, open the client, enter your API key from your profile in OpenAI, select your odel and settings and thats it

[lapis rune]

yeah but you get charged per message not per month, so they get more money down the line

[modest cedar]

Oh, aight. That makes sense then.

[flat harbor]

you can access in the openai playground

[lapis rune]

yeah but having a chat app that functions like the chatgpt website is more convinient, keeps a chat log, displays markup, can have multiple conversations etc, you can even load a database of files

[flat harbor]

fair enough

well I think chatgpt is about to retire anyway. doesn't seem to want to do anything I ask of it anymore. well it'll do it eventually, but after stalling for a response or two to waste my prompts. seems rather odd. dall-e 3 mode earlier told me it's unable to create images so it'd have to simulate it. 😆

[still geyser]

this is so frustrating, can't get it to do anything anymore and when it does it's a really poor quality output.

[terse island]

Yea ik same

[lapis rune]

i don't have this issue and I mostly use it for coding, do you have an example I can replicate?

and doing my uni assignments 😂

[main summit]

Dear Fellow Chatters 😉 I posted a question in the communicty section, but maybe here sb knows.
Hi All, create new custom GPT sends me back to the main chat page. Whats up with that?
https://chat.openai.com/gpts/editor this link just takes me to https://chat.openai.com/

[flat harbor]

I've been using gpt-4 since it was made available in march, rode through the ups and downs. this latest issue is just frustrating because it seems that along with all the big leaps forward in many regards are somewhat negated by it's resistance to doing the things I asked. read that openai is working on it so that's cool. as far as coding, might be able to find a recent example. but basically explained what I was trying to do and asked it to give me it's thoughts, what libraries and such to use, then show me some code using it's suggestions. and it responded with a general guide on writing code. nothing referencing the specifics of my prompt

[lapis rune]

same issue, just need to wait till they solve it

[flat harbor]

that really should have been 2 messages there, but you know, didn't take that route and too late now

[lapis rune]

They did mention on X that they know some prompts make GPT4 appear lazy and they are working on it

Alternatively you can choose the GPT4 CLASSIC option from the Explore page and you get the non-turbo model

which doesn't have this issue

[flat harbor]

some prompts? I hate that it's always put on the end user. it's not some prompts. LOL

[lapis rune]

Since i don't have this issue at all i'd have to agree with them unfortunately

[flat harbor]

of course you would

because your anecdotal assessment holds more weight than anyone else's

[lapis rune]

theirs does from your anecdotal assessment though

because their assesement isn't ancdotal they can actually see all the data

and they don't blame the user, they are FIXING the model

it's just some prompts that trigger it making it lazier

[honest jay]

I think it is Turbo, just with all modalities disabled. The GPT's description says something like "The latest version of GPT-4 without extra capabilities."

[flat harbor]

what data. did they show anyone this data? I have taken countless different approaches and arrived at the same results. I realize you probably think you know more than me and that you're schooling me on these things, but it's profoundly frustrating to see a very clear, very obvious issue across the board and have it not be acknowledeged in any way.

[lapis rune]

the classic GPT-4 (gpt-4–0613) is slow and its knowledge is valid till Sep 2021 there's a source but this chat doesn't let me paste

[lapis rune]

OpenAI knows more than you, you just refuse to accept it.

[flat harbor]

I honestly believe that it's a response to it not have access to the resources required to complete a request at that moment in time

of course they know more than me

they're just not necessarily divulging it

[lapis rune]

thats a bad take because you can force it to answer by gaslighting it if its lazy, so you use more resources that if it just answered the first time

[flat harbor]

I believe they added what they thought were small modifications to make the model work more efficiently as compute requirements were increasing, and it didn't exactly work out as planned

[honest jay]

Just started a new chat with it, claimed knowledge cutoff of Apr 23, and correctly identified the current monarch of the UK

[flat harbor]

no. prompt cap

[lapis rune]

yes big events have always been included

[flat harbor]

yes I can gas light it. but if it can stall me on images for 3 responses that's less it has to do. and multiply that by millions of times a day

[honest jay]

But weren't you saying the GPT "ChatGPT Classic" used an older version of GPT-4 when you said "Alternatively you can choose the GPT4 CLASSIC option from the Explore page and you get the non-turbo model"? Turbo is the one with the Apr 2023 cutoff.

[lapis rune]

might be hallucinating I don't know thats what their documentation says

just google it, better than speculating

[honest jay]

What documentation is that? OpenAI domain links are whitelisted here so feel free to drop a link!

[flat harbor]

ChatGPT Classic
The latest version of GPT-4 with no additional capabilities
By ChatGPT

[honest jay]

yee that's it 🚀

[lapis rune]

you're probably right it is the turbo version I can't paste the link here it's from medium I was chat restricted lol

[high horizon]

does redoing/rerunning a prompt go towards your limit of 40 prompts per 3 hours?

[feral zinc]

yes

[glad vapor]

network error

[little adder]

How can I get more gpt=4 prompts?

[feral zinc]

pay and use api

[sterile socket]

I noticed today that I became annoyed at ChatGPT for not being able to perform a task I’d never have thought to ask it a year ago.

[quick mica]

I've got a problem with GPT-4 not being able to extract information from an attached PDF file, it is 34 MB in size. Anyone else having a similar issue?

[foggy magnet]

For knowledge, does it matter if a txt document is in docx. vs. .txt format?

I want to take advantage of docx formatting for better retrieval

[rose ferry]

chatgpt 5 when

[maiden fractal]

One day

[tender edge]

does anyone else try to use this with Opera GX? Do you ever have latency problems with what I think is the CSS and the site?

[foggy magnet]

Is gpt slow for anyone else rn?

[tender edge]

yes, and I have to keep regenerating.

[foggy magnet]

40 messages. 20 spent on broken responses or overall garbage output. I cannot believe the experience has become this frustrating. I might have to retrieve some archived chats

[blissful igloo]

Same. Mine just gave up and downgraded to 3.5 after a failed message. Unacceptable. If it was temporary it would be fine, but I have no control over the chat version is unless I start a new chat, and I taught this GPT about my fantasy world.

[sage yacht]

4.75 or whatever will drop before then lol

[still geyser]

thanks for the tip about the classic, which is useful to an extent. a lot of the work i do is with PDF and OCR type of stuff so it doesn't help much. Today it wouldn't even read a PDF and kept claiming it was a PDF.

Kept claiming it was a JPG I mean

[warm gull]

yeah today had massive amounts of errors and timeouts - it was pretty bad my side

[acoustic talon]

So GPT 4 stopped writing papers for people, huh?

[foggy magnet]

I've started bullying it. It's working.

[blissful igloo]

Elaborate? Who are you bullying?

[terse island]

Its been annoying yea

Ive asked it multiple times to read things and let me know how it is or something and it just like tells me exactly what im looking at

“This appears to be an essay written about blah blah”

[west wolf]

I've been making GPTs and I've noticed often it will say it modified teh prompt and it did nothing.

[misty slate]

I received an email from OpenAI telling me that I had access to the GPT-Plus waitlist but when I went to the prices the "subscribe to the waitlist" button was back as if I didn't never clicked on it..

[honest jay]

You might add your experience to this post! https://discord.com/channels/974519864045756446/1183900385698644028

[misty slate]

thank you n!

[nocturne wolf]

anyone able to get code interperter working? chatGPT hasnt' been able to access any of my uploaded files for weeks

[foggy magnet]

GPT. I channeled my frustration into getting it to stop doing all the things that have been ruining the experience and value, now it's back to where I expect it to be.

[bleak vortex]

Anyone get an email about being off the waitlist for ChatGPT Plus but unable to press upgrade?

[honest jay]

You might add your experience to this post! https://discord.com/channels/974519864045756446/1183900385698644028

[bleak vortex]

Thanks for directing me to the right chat

[dawn delta]

anyone know any good plagiarism checkers for AIs like chatgpt for lines of code

[acoustic talon]

Yea, I just made a GPT that helps me write

[tulip iron]

hello i just bought a subscription and was disappointed. chat constantly fails to finish its answers. stops at ~10-15% of the answer

[wraith void]

lately chatgpt fails to do any request completely ignores requests it's like I'm talking to a teenager that completely ignores you lol

my workaround tell it you'll fire them if they disobey orders 😂 oddly enough it works

[acoustic talon]

Right

Its been annoying

[foggy magnet]

this is the energy you have to give gpt now, you have to position yourself from a place of consequence

either its behavior will have consequences you will act on, or that there are consequences you will receive based off its behavior. It's amazing how effective this has been.

Gone are the days of "Please" and "Thank you."

[acoustic talon]

So I'm paying 20 dollars a month to force an answer out of it?

[foggy magnet]

No, you're changing your prompt tactics. Engineering prompts has been the core action since the start.

I'm not talking about arguing with gpt. Just changing the structure of the request.

[urban notch]

does anyone know how accurate winston ai is?

[boreal lily]

what happened to my history

everything past today disappeared

[zenith raven]

Anyone noticing past conversations other than the current one being used at disappearing? Although when I refreshed it, my convos were restored, but when I regenerated a response, it vanished again leaving me the current one I’m using

[boreal lily]

same here

[foggy magnet]

entire custom gpts can disappear and reapper days later, no idea what the technical reasons are behind this stuff happening

[zenith raven]

I’m not using customs

[foggy magnet]

I know, I'm giving another example of things disappearing/reappearing.

[zenith raven]

Ah

Ok. Another classic techie doozy

[boreal lily]

some openAI mods fixing hisory service

[foggy magnet]

Aye. I think their database is getting a little theoretical on us LMAO. Hopefully they stabilize soon.

[lost raptor]

What happened to all of my chats?

They're all deleted

[zenith raven]

They’re not, something up with servers

I logged into my other account, all my convos were gone. The other still has one convo I’m using

[lost raptor]

I sure hope so. I had some seriously cool projects I was working on.

[zenith raven]

It was fine almost all day yesterday for me

It’s probably a tech issue in the server

[foggy magnet]

server thing for sure, multiple people reporting this rn

hang tight

[dense halo]

yes, it's not loading my convo history before today, concerning

[zenith raven]

I at least have 1 convo intact lol

[dreamy barn]

Did anyone else have convos go missing for now?

I guess so

Reading above me

[next scarab]

Yeah, all my convos before my last 2 are lost

[dreamy barn]

Eeep I need to download artwork

App all gone

Web just 1

[next scarab]

I'm not sure what happened. For anyone else, were y'all messing with the custom GPT or the plugins at all?

[dreamy barn]

Baby come back to me

[candid magnet]

Hi there,

My Chat GPT history disappeared after I deleted a few chats. Is there any way to get it back?

[dreamy barn]

We all feel ya

[foggy magnet]

Whatever happened to the servers literally just broke my conversation mid-chat. Damn.

[candid magnet]

Wait so everyone else is experiencing this?

[foggy magnet]

Now my chat is acting brand new and back on the same stuck behaviors I fixed.

[foggy magnet]

Yep, we're looking at some kind of outage.

[candid magnet]

Ok cool thats alriught

As long as its not just me

[foggy magnet]

Yeah about a dozen people have posted about this so far. So we're looking at whatever this is being the early onset of a server issue.

[dreamy barn]

Okay I guess no emphatics

Oh mod bot

[foggy magnet]

Mod bot said "Keep calm."

Damn I was on a roll too, who knows when things will be working again.

[next scarab]

If any of yall have any of your gpt tabs open, those tabs might still have your history. Just go through what chats you want from your history and try regenerating a response or inputting something and seeing if that puts it back in your history

[dense halo]

my convo history is back

[fathom zenith]

Fun fact

You can download be my eyes

And use their app for gpt-4

[raven ice]

Like there was a way to use a previous version of the GPT engine, is there a way to use the version of GPT4 before the context-length increase of a couple weeks ago?

[fathom zenith]

impossibly only with api

[raven ice]

Thank you

[lost raptor]

they just came back

[grim fractal]

I used my quota for three hours in like 5 minutes..

very surprised...

[sudden latch]

have you guys tried using ai detector such as gptzero? are they reliable? i put a well-structured text with sophisticated vocab written in 2020 before chatgpt release, and it was marked to be likely an ai text 🤔

[lost raptor]

those things were proven to be unreliable even by open ai

any statments they make about them being reliable is a gpt generated statement.

[sudden latch]

😂wait what

[lost raptor]

https://openai.com/blog/new-ai-classifier-for-indicating-ai-written-text

The whole thing feels like a scam. A quick buck that people came up with to sell to schools or teachers.
Similar to how plagiarism detectors were picked up by colleges.

[sudden latch]

but honestly can you tell if a text/passage is ai generated if you are native in english (as for im not, i cannot tell)?

[lost raptor]

tbh not particularly. See one thing open ai fails on is human language conversations. Short statments like this. It will usually talk like a book at you. If you received a email. you gotta get a feel for the flowery language LLMs like to use. Simply talk to open ai like a person. and you'll get a feel for how bad it is for light conversations.

You'll never be able to tell the difference honestly.

because someone could write something and ask open ai to fill it with fluff to meet a quota for words.

[terse island]

Not really but a lot of apps can usually tell now. However its pretty annoying bc my essay i wrote entirely on my own scored a 30/100 for being written by ai

Like the ai detector thought 30% of my essay was bot written

[lost raptor]

literally contridicted yourself

[terse island]

What

[lost raptor]

yes they work, but not when I write my papers.

[sudden latch]

oh i think it depends on the writing task, from whether a human can tell if a text is ai generated (at least in english).

[terse island]

Im saying i got 30% ai likely on mine when i didnt use it

[sudden latch]

i can possibly tell if an email is written assisted by ai, but i cannot tell from a technical document

[lost raptor]

well cough cough its usually a certain group who would use an ai for that.

[sudden latch]

but anything written in my native language are just very mechanical and hollow, and anyone can easily tell if it's an ai text

[lost raptor]

I mean why are you asking? Are you a teacher, a student, or maybe youre getting messaged on discord by someone you think isnt real?

[sudden latch]

just stumbled upon the ai detector and be suspicious about this technique🤔

[pure dock]

anyone here pay for chatgpt plus

is it worth if there is message limit

[lost raptor]

yes.

[tender edge]

It really depends on how much you're using it, I use GPT several times a day so its totally justified to pay the sub for it.

[pure dock]

i mean i was tryna use it and i got capped...

i just started paying for it today

i mean im a student in uni so i definitely could use it a lot, gpt 3.5 was just not good enough

[plucky timber]

Can you indentify the questions you would ask GPT3.5 and those you would ask 4? If so that saves you a lot of messages.

[pure dock]

yeah taht would be a good idea

i mean i usually like asking it either programming related questions or math and physics

i assume gpt 4 is much better at those? not sure

[plucky timber]

Same for that tbh, I found myself after hitting the limit for example finding that GPT 3.5 still was decent add doing some basic coding work. It's speed of response does help running through potential errors it makes faster. so it offsets I guess 😄

The more difficult the question I'd indeed go more towards GPT4.

[proven idol]

I mean im a high school student and Ive had multiple scenarios where ill write an email then give it to ChatGPT and say "make this more formal" or "what changes would you make to this" and send the one chatGPT give me.

[plucky timber]

With coding GPT4 can be stubborn on doing full work though without specific instructions

[lost raptor]

in context to having to worry about a email being assisted by an ai.

scam wise

[pure dock]

Yeahhh for the short time i started using it today i realized it was definitely much better at explaining higher level of math, even answered this numerical analysis proof for my brothers take home final

[proven idol]

yeah thats my favorite thing to do with it

[pure dock]

really? i havent tried code yet

[proven idol]

Had it explain entropy to me, and it does it as well or better than most of my teachers would

[plucky timber]

Ill save you a headache or two , the keywords are "Including unchanged code" 😂

Otherwise you'd see a lot of // Existing implementation
or something along the lines of.

[pure dock]

ohh interesting...

didnt know u cant swear i got timed out LMAO

[plucky timber]

Very PG server 😛

[pure dock]

i find gpt 3.5 a bit annnoying with getting it to write the code how u want it to be written