#gpt-models

1 messages · Page 262 of 1

knotty grotto
#

Keep in mind edge and chrome are similar chromium based

calm silo
#

It's the same on both. Thanks for the responses; I will have another coffee and retry later.

slate wren
#

I had this problem on firefox but worked on browser

#

it started friday but it seems a bit better now

calm silo
#

👍

stoic lake
#

I have an issue uploading any files to my gpt for context, they all fail, is there a way to fix this?

#

they are just 1 kb txt files

#

Like its just any file that fails pretty much.

#

I changed browsers, restarted my computer, i really need this to work so i can get this custom gpt out. They get stuck 90% of the way done.

#

Ok I tried a 3rd browser and it worked.

nimble summit
#

how do i check how many messages i have left before i hit my limit or is that not possible

knotty grotto
opal sapphire
covert swallow
#

So.....if I want to ask feed back about my gpt, I have to post on custom gpt?

austere garnet
#

Is there a way to set which voice is used for custom GPTs yet?

frigid burrow
#

With custom GPT instructions, how do you get your GPT to follow the instructions you've given?

#

Or maybe I'm just formatting instructions the wrong way in the first place...

#

I want it to:

  1. Search knowledge documents for a starting point
  2. Search my vector database for more detailed information that won't fit in the documents (currently only intermittently does this unless explicitly asked to do so, which end users won't know to do)
  3. Use the chunked information from the vector database to identify the correct URL to answer the question and use the browse.search tool to read the full document.
#

Or sometimes it'll do 1 and 2 but not 3

pallid aspen
#

I typically say do the following in order, and number them, but i noticed lately its kinda acting up

weary swallow
#

Can I try to use the basic principles of a GAN in my GPT based DALL-E 3?

grizzled flax
#

Hello, you might already know this, but the Create chat in the GPT builder only edits what's in the Configure section of the builder. So if you want, you can:

  • Copy/paste/save the Instructions field of Configure before making changes in Create, in case you need to revert changes, or...
  • Do this, but also skip the Create chat completely and only edit Configure directly.
covert swallow
wraith urchin
#

Anyone have the info on how I can change the name of a custom GPT to my own actual identity, instead of community builder? I am starting to beta test, but wanted to correct this option.

onyx badge
#

Will we be allowed to discuss Copilot GPTs here when they release?

tulip thunder
#

literally too many can do what you like. NIghtCafe is a good solution, but www.BING.COM/NEW in EDGE Browser can do this using Dall3 which is so so so good. - Also MidJourney will allow you to do this easily. - Hell there's about 10 AI ART discords right now that can do that too.

covert swallow
#

where can I find someone who wants to test my gpt?

#

My gpt is for hosting role playing game systems for solo adventures. not for bussiness only for fun, but need a lot adjusting...

rose trail
#

is it just a prompt, or does it have a website backend?

covert swallow
#

Yes prompt for now, which is why I need some advice.

lucid turtle
meager bone
#

How can i Log all of my devices off my account

ember leaf
#

how to find use demo

lucid turtle
#

my gpt outputs like half a bloody book with each response now when narrating a story i dig it . and its following directions better buts still not following all the custom instructions still

solid carbon
#

I counted and received 12 responses from gpt4 today and I reached the total number of allowed questions (it is possible that I clicked continue on a few questions, so I reached a maximum of 20 questions and was blocked),
Is this a usage cap? 20 answers and already blocked???
This was not the case in the past, does anyone know what the current policy is?

lucid turtle
#

r u using custom gpts"s if so then yea you reached capped by regen as that counts as 1 message

azure crater
#

Hi !

#

Is chat gpt having issues with downloading screen ? Or maybe its my connection ?

grizzled flax
stoic lake
#

how do i get my custom gpt not to have timeout errors, it is killing functionality

#

it s reading a really short python file, like 10 lines

gilded oyster
#

Is anyone else not able to see their Custom GPTs today or their Explore button?

lucid turtle
#

i just signed out and back in fixed it for me

oak harness
#

Custom GPT seems to be working well. a bit laggy the past couple hours but Im not sure if thats my end of server sice

languid scaffold
frigid burrow
#

Now it works like a charm

languid scaffold
lucid turtle
#

or its just not reading your custom instructions like it does for me

#

its finnaly following prompts finnaly like jesus. but custom instructions is a no go

languid scaffold
#

What I'm doing is saying "when I say X, respond with Y" and it won't

lucid turtle
#

yea should see my custom instructions its probably the most in depth ever and it barey follows it

languid scaffold
#

OK well at least it's not just me

frigid burrow
#

Yeah the more complex I get with instructions the worse it behaves

#

I've got exactly 8000 characters in my main GPT's instructions. >.<

lucid turtle
#

danm bro

#

the api is able to follow it maybe its just the 32k variant model

#

however i do ike how my narrations are over 800 tokens length instead of 300 like its been doing

frigid burrow
#

It's not as much the size as the complexity

#

the larger it gets though the more likely one instruction will conflict with another >.<

#

8000 is the limit though for custom GPTs at the moment

lucid turtle
#

true but the api assistants compared to the gpts is just far superior i have two examples

frigid burrow
#

yeah but they're not cost effective

#

at high volume custom GPTs can give a lot more tokens per dollar

rose trail
#

gpt4-32k is like more than 2.5 times more expensive(?) so you arent really saving anything if you get 800 v 300

frigid burrow
#

I'm more worried about what happens with programming errors

rose trail
#

i find gpt 0125 rocks at coding

frigid burrow
#

no not the GPT coding

rose trail
#

maybe gpt4-32k outperforms it but i doubt by much

frigid burrow
#

I mean me coding 😄

rose trail
#

oh well yeah fair nuf

frigid burrow
#

An accidental infinite loop cost me $5 in 30 seconds before I could kill python

#

What happens in a production environment if a user somehow triggers an unexpected bug?

#

and no one catches it before running up a bill

#

a custom GPT is free for me for other people's usage 🙂

north moon
#

My gpt says its missing language files for translation needed by tesseract OCR how do i get those files added?

#

Or is there a process to getting files added to the environment in the future?

crude dagger
lucid turtle
#

chat gpt lagging hard

#

indie dev company

muted ice
#

I don't know where to request this feature, however if any of the OpenAI dev team read this, PLEASE add a 'read back' feature which has a voice read the text back to us. My eyes get tired of the reading all day and would love to easily have ChatGPT read the text to me when I click a play button

lucid turtle
#

it has it on the phone just not the pc

lucid turtle
#

i just learned a new word from chat gpt LOL

lucid turtle
#

why does gpt for narratives loves to do a reflection on the characters its such a waste of tokens

somber jacinth
#

Gpt4 so slow..... why am i paying for it?

lucid turtle
#

good question LOL

lucid turtle
#

lol my gpt just broke

oak harness
#

same here

#

been stalling out for the past hour or so, and now it's having issues searching and analyzing. Seems to be prioritizing bing search in my customGPT configurator while I try to build myself an assistant 😢

oak harness
somber jacinth
oak harness
#

Sorry, I missed the key word of NO

#

there is no equally useful competition at the moment

#

the update or hotfix in the past hour seems to have reverted it back to gpt1 or something.

many mistakes helping tutor through general function transformations

somber jacinth
covert swallow
#

.....I feel my gpt becomes silly without any change.

oak harness
#

Mine was actually becoming really good. Fine tuning the custom instructions to really make it helpful in tutoring me. Then it took a nosedive today. Makes many mistakes and ignores all the progress we made in our relationship over the past few days of adjustments

covert swallow
#

I kinda feel it downgrade to 3.5, the speed and the word........

oak harness
#

I agree

#

Maybe ms/bing is influencing their investment more

covert swallow
#

now it can't even remember the first sentance I gave it. the instructions I mean.

oak harness
#

Yes, it sems to have changed with its interpertation or even following of the custom instructions. Definitely not processing them the same as it was just a few hours ago

#

Not to mention its logig in general. Have to point out some simple mistakes 3 times in an example problem it gave and how ot stepped through it.

covert swallow
#

I am gald it isn't my issue.

oak harness
covert swallow
#

I mean...I was kinda freak out, especially I just showed it to my friend who i share team plan with...

latent osprey
#

Hello - has anybody experienced an error saying 'Cannot connect to remote server'? I keep getting this error on my Bubble application when making API request to GPT.

daring dune
#

Hey guys, does anyone have news or info on what will happen to the GPT Store in the next couple of months?

distant hornet
covert swallow
#

I thought making an Aizen Sosuke gpt is fun, until I really talked to him, I just felt how annoying he is........

neat grove
#

Can anyone tell me how I can speak to a human being that works for openai? The chat support on the website is a bot and the default templates really don't help. If anyone has a clue, please share.

grizzled flax
neat grove
# grizzled flax Hello, that chat widget on help.openai.com is the best way to submit a ticket. F...

I've done that numerous times. So much so that by heart I know the combination. What you end up with is a bot that gives you a default template answer and an email 5 days later. Im trying to run a business. I don't have time to figure out myself what is wrong. I need some assistance or perhaps someone who can think with me.

And what do they do? They send us to a discord server like I'm having all the time in the world to figure it out myself? They have even closed my previous account and refunded my subscription while I had a question about something completely else.

Imagine that your car breaks down, you go to the garage and the mechanic says "tools are over there, you can ask the other customers how it works and that'll be $250". Would you go to that garage?

I am willing to sort things out myself as a hobbyist. But when I have issues with a service I pay for as a business, I expect to be able to speak with someone of the company. Not other customers.

grizzled flax
# neat grove I've done that numerous times. So much so that by heart I know the combination. ...

First, I get how it would be frustrating to want to use a service to enhance and contribute to the operation of your business, when said service is burgeoning and not yet stable. Regarding your garage analogy, though, they are open about Team being more of a "tools are over there" kind of service, as you described. https://help.openai.com/en/articles/8792828-what-is-chatgpt-team

The ChatGPT Team subscription is a self-serve subscription plan
Additionally, on the plan pricing page, "Priority support & ongoing account management" is only listed for Enterprise plans: https://openai.com/chatgpt/pricing
So as far as choosing whether to go to this "garage" or not, I do think they're being forthright about their current capacity to offer comprehensive support.

neat grove
#

Got it

#

Enterprise

#

Right so again I have to write a message to a sales team (not support) to upgrade in the hopes I might get a decent working product.

grizzled flax
neat grove
#

I mean am I a boomer or is this normal these days? Why can't I not speak to someone from CS? This is very strange if you ask me. I mean do you think this is ok? Or you just accept it because they have a monopoly position and so they can do whatever they want?

grizzled flax
neat grove
#

does anyone know an alternative I can use? something that does offer consistency?

grizzled flax
scarlet sun
#

hello guys, could you give me some pointers on how to use actions to connect a gpt to an external database? like a product listing that keeps being updated (that has images of the products)?

supple violet
neat grove
neat grove
supple violet
neat grove
# supple violet you asked for alternative for chatgpt? no? if you already used then you know how...

What you are referring to is something you build using the GPT API right? Because I paid as a business for a GPT4 Plus account to write some blogs. And here I am 2 days later, hearing I have to build my own GPT on top of the GPT API. I don't mind doing this, but whenever I see people on youtube writing a prompt and getting a decent blog, I just seem to wonder how come when I use the exact same prompts and follow the exact same steps I get something that is in the likings of a shopping list. I tried different browsers, I tried different prompts, custom GPTs, plugins. But still no blog.

Now if I was a high school student who wanted to use it for his homework, I'd say yeah no problem. But as a business I am looking for something that is time efficient and most of all consistent. Which I am not experiencing at all.

Part from me having to figure out how to get my paying service to work myself (being the upside down world imho), I am just curious how reliable this would be if I would have to build something on top of a service which so fas hasn't really proved its worth. That's all I am asking.

supple violet
neat grove
supple violet
#

note that api is not free or cheap when using gpt-4

dusk dome
neat grove
# supple violet i cant say as i dont know yours but its certainly possible to replicate chatgpt ...

yeah I am looking at it, but this is not what I paid for. I want to be able to use the same interface like anyone else and now I got to figure out how this works? My prompts used to work perfectly and with every update, the consistency keeps dwindling. I know it's not your fault and I probably sound like a Karen, but if this is how OpenAi wants to treat their customers, I rather find something that does deliver.

dusk dome
neat grove
supple violet
neat grove
# supple violet well if you want chatgpt ui then you use it but if it does not work and you cant...

their support is also GPT. you are talking to the same LLM. I thought they might have outsourced it to (yeah like I was going to make that mistake) or another country, but it's not even that. They have fully outsourced their CS to GPT and here we are fending for ourselves.

I will refer to my analogy I earlier made: "Imagine that your car breaks down, you go to the garage and the mechanic says "tools are over there, you can ask the other customers how it works and that'll be $250". Would you go to that garage?"

If there is anyone who knows an a.i. service that allows me to write SEO strong blogs of which I can customize the input that is NOT running on GPT, please let me know, If it is what I am looking for I am willing to reward you with an x amount of crypto. That's how desperate this nonsense is making me.

supple violet
neat grove
# supple violet the current support is worse than what chatgpt would be as a support "person". <...

Unfortunately I don't have the time to talk with random people here (you guys are great @supple violet and @grizzled flax and thank you so much for your time, don't get me wrong) as I am losing money as we speak. It shows that if you have a monopoly position as a company, you can just shhh(not going to say a nono word or else I get banned for speaking like an adult) all over your customers. And as long as we all comply with their nonsense rules, updates and the way you are being treated (like dirt that pays, imagine that one) then they will continue to do so.

At least if skynet puts GPT in their T-1000 we have nothing to worry about

distant hornet
lucid turtle
lucid turtle
#

wow they really shafted the word count. so what i tested before you were able to have up to 14,000 words before it needed to do its retrieval tool ... now its all the way down to around 8,500 words estimated...

serene cipher
#

it was never a matter of word count or character count, only token count

#

if a file has a small enough amount of tokens, it will often just put the entire file in context

#

otherwise, it will do a actual data retrival

scarlet sun
serene cipher
#

GPT is a text generator, so, after you get the output text, you can do whatever you want with it

#

such as storing it on a database

scarlet sun
#

you ask to see tables, or whatever they have, it then uses actions to connect to their product listings, shows images of products and a link to hte product webpage

#

while telling you the price, etc

serene cipher
#

ahhh, so you want to make a custom gpt that uses actions to get information from an external source?

#

it is pretty simple

scarlet sun
#

yes thats it!

#

like a database of products that is always being updated, because the price may change, the quantity of products is huge, etc.

#

its not efficient to create a knowledge base

serene cipher
#

first, you have to actually create an API that can receive something such as GET /api/v1/search?query=table, then, it looks up the database for everything tagged as table, and the API returns that information

#

this is a pretty standard API thing

#

next step is to just configure the custom GPT Action so it knows it can use that endpoint to search things

#

and the AI then should be able to call that endpoint and use the response of it to generate the reply

scarlet sun
#

that was what i was looking for! because while im pretty proficient with prompt engineering, everything that has to do with APIs its a bit harder

serene cipher
#

some programing knowledge will be required

#

because.. well. you gotta code the API

scarlet sun
#

right, but ill leave that to the dev team

serene cipher
#

but that shouldn't be a hard task, it is pretty basic

#

you can even ask the AI to help with that

scarlet sun
#

yes but that i believe involves the client side

#

and their website

#

database etc

#

thanks a lot for your help! really appreciate it

serene cipher
#

you are welcome =)

cold lodge
#

hi i use gpt in with function calling to get data i got an issue i want to add historical data in gpt but there too much data to be return as a text ( i try but it cost a lot and gpt hallicinate a lot ) so i want to give it a file to be use inside of code interpreter how do you think i can do that ?

#

i try to see with the api to return the file with the submit tool output but i can't find a way to return the files in the thread

lucid turtle
nimble summit
#

new to gpt plus, anyway to increase message limit, even if its at an additional cost

grizzled flax
maiden quartz
# neat grove I mean am I a boomer or is this normal these days? Why can't I not speak to some...

Nah you're not a boomer, and this is entirely depressingly normal these days. 😀

Reading this entire conversation of yours here, you word for word echoed what I've felt with tech for many many years now ever since I was young as hell, and is exactly the issue with ClosedAI, the fact that people don't think it's mismanagement is honestly insane. Sure, let's just act as if their company didn't just almost blow up catastrophically internally because of some civil war, not at all long ago, if you're familiar with those news.

#

Yeah aside from that, it's been quite clear to me ever since the start, more or less, that ClosedAI couldn't really give a single goddamn about regular consumers, most if not all of their advertising is always entirely about "eNtErPrIsE", like are you serious?

#

LITERALLY just spitting and mocking at us publicly with glee atp.

olive meteor
#

Anyone else seeing this in "old chats"?

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

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

rose trail
#

is the chat older than 2 weeks? because thats when the last model dropped

#

would be unusual for them to release a new model, unless some significant improvement couldnt wait

analog copper
#

"Your GPT has been removed from public view", excuse me what?

distant hornet
analog copper
distant hornet
#

Weird.

#

Maybe it was doing homework.

analog copper
# distant hornet Maybe it was doing homework.

reading policies I see one that says "c. Engaging in or promoting academic dishonesty"... if it is for this at this point just shut down ChatGPT entirely. kinda a hypocritical policy

distant hornet
faint topaz
#

Fortunately most of my students are internally motivated so most of them listen.

#

E.g. bad habit: getting a custom GPT to solve your maths problem about compound interest. Good habit: revising the concept of compound interest the day after you were taught it in class to consolidate your learning.

still chasm
#

Weather in Plovdiv

faint topaz
knotty grotto
#

Purple

analog copper
# faint topaz How so? I'm a teacher and while it takes a lesson or two to show it, there's a w...

The GPT is intended as an educational aid meant to assist in understanding mathematical concepts and solving problems. Unless students use it to complete assignments meant to be done independently falls on that category, but at his point is impossible to avoid that type of use, so I guess making a GPT of anything that can aid to do homework falls on that policy, although you can just prompt chatgpt to do it anyway

faint topaz
#

"Can be used" is different to "for the purposes of"

analog copper
faint topaz
#

And the user, but they're culpable on different levels / to different people

distant hornet
#

Is there maybe a constructive way we an frame this?

rose trail
#

AI didnt invent cheating it just makes it more accessible. the underlying system really needs to inspire intrinsic motivation, and thats a social problem not a technical one.

analog copper
faint topaz
faint topaz
rose trail
#

i'd say we're all responsible for the future of humanity

#

as it is a gift that we must share

faint topaz
#

Doesn't work that way unfortunately 😥 anyway it's an important and interesting conversation but maybe #off-topic unless we swing back to particular GPTs and their responsible implementation

analog copper
faint topaz
analog copper
#

hehe, I hope the appeal is approved anyway

livid jackal
#

teams seems to be broken I can't save custom gpts to teams was working a few days ago, says admin distabled but I'm the owner of the account and I can't do jack so what's going on?

plush bay
#

So I've used chatgpt through the website, both for free 3.5 and at one point had a month of gbt 4 through the monthly subscription

But I hear there's a different type of access via api? How do I get access to it. I'm on mobile atm so maybe my browsing skill didn't cut it but I couldn't figure out how much using the api costs

#

Oh looking into it, it doesn't look like a simple thing you just use?
You have to know python and such? My bad, I thought it was like chatgbt but pay as you go but access better models

wraith ginkgoBOT
#
<:book_icon:1171408210398289941> `` Rule 1 `` Be respectful.

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

faint topaz
manic delta
#

Anyone want to tell me what they think?

shy holly
#

from the OpenAI website; "Builders can earn based on GPT usage
In Q1 we will launch a GPT builder revenue program. As a first step, US builders will be paid based on user engagement with their GPTs. We'll provide details on the criteria for payments as we get closer." My question, what about builders from outside of the USA?

manic delta
hazy wigeon
#

When was GPT able to make graphs? haha

lucid turtle
#

whats the best way to make gpt for a nmarrative not use so much tokens ?

manic delta
drowsy quest
#

Am I the only one who runs into the issue where the gpt will respond to a prompt, I respond to its prompt, it sort of responds to what I said, I respond again and the next response is something completely different all together? It is the most frustrating thing for role playing purposes.

azure imp
#

Is here someone that worked with function calling and can share some of his/her insights with me. Feel also free to send me a private dm

knotty grotto
sacred comet
#

The model is capable of 128K but still Plus and Team only uses 32K not 128K maximum

sacred comet
#

Enterprise uses 128K of 128K, full context length

sacred comet
#

without enterprise, plus and teams

#

at your wish

#

but with pay as go plan

upbeat bramble
#

Can someone use my gpt if they do not have a subscription?

#

I made a new account and saw that i still need to wait to share it

drowsy quest
#

Remind me again, is that a 32k token, character or word limit? Not hitting it either way. This is happening at 1500 tokens, approx 7000 characters, 1158 words.. so unless my understanding is wrong, I’m not hitting any limits.

grizzled flax
grizzled flax
drowsy quest
upbeat bramble
covert flicker
#

I don’t seem to have the @ mentions feature for some reason, what can I do to get it? Why would this be like this?

rose trail
#

maybe you have browser extension interfering

grizzled flax
rose trail
#

is very cool feature

onyx badge
#

...huh, it is

#

well, seems to be account based

#

so if you have it in desktop web or mobile Firefox you just have it

#

its still definitely a chromium mobile issue

burnt yacht
#

Hmm. My GPT got flagged. I appealed it because I believe it was an error. Anybody experienced this? What's the turn around time on it?

manic moat
#

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

burnt yacht
#

40

#

ever 3 hours.

manic moat
rose trail
#

its a free market. if someone else can do it cheaper they will

lucid turtle
#

after giving my prompts 5 options for how the story should go depending on the narrative made everything flow so much better than me typing how the story should go i reccomend for people wanting to doing action adventure

#

save on alot of toekns too

analog copper
burnt yacht
restive lotus
#

I absolutely love constantly repeating the same instructions to a GPT, formatting those instructions into a document for it follow when reviewing work for me, having it detail word for word what the instructions ask for, and still getting responses that ignore over half the instructions I had spent all this time working on 🫠

quasi monolith
#

is gpt 4 boost more expensive then gpt 4 or what?

faint topaz
quasi monolith
#

I dont know what afaik means

#

do u know like how cheaper it is - is it like massively noticable?

faint topaz
#

"As far as I know"

wraith ginkgoBOT
quasi monolith
#

gotcha

#

do u know how much cheaper it is? is it like massively noticable

faint topaz
#

It's the top / cheapest model

quasi monolith
#

gotcha

#

ohh so it's by like 2 cents

#

thats not much of a difference

#

tbh

quasi monolith
faint topaz
civic island
#

Hey. Is it normal that GPT mentions with @ symbol didn't appear yet to be able to use it?

rose trail
#

turbo is cheaper to infer, but gpt-cheap wont market as well as gpt-turbo

loud scroll
#

Has anyone noticed if GPT 4 is in Dumb Mode? The whole system has been acting extremely weird with basic prompts and seems much less intelligent. It can't even summarize a simple PDF anymore!!! I would appreciate any insights into why GPT4 seems so much less capable than before, even with straightforward language prompts that should not challenge its abilities. Let me know if you have also observed this regression in GPT 4's performance on basic tasks compared to its previously demonstrated skills.

rose trail
azure whale
lucid turtle
lucid turtle
rose trail
#

yeah but would be nice to have a channel free of complaint

lucid turtle
#

why we pay 20$ we deserve to complain when its giving out 3 sentences wasting our 23/3h messages

#

i use it more than most people do on here so i know when it changes and when they tinker with it

midnight blade
#

I still don't have mention feature. The rollout is still not completed yet? Anyone else?

distant hornet
#

You kind of have to try this one to believe it's real.

clever thunder
#

Hi, I'm curious if anyone has knowledge or experience in fine-tuning language models using videos. I came across an open-ai article discussing this approach, where the video is split into images for the model to learn. Specifically, I'm interested in knowing if it's feasible to provide a video of sports or video games to the model and teach it to identify correct and incorrect instances.

Could anyone clarify if this is a viable option, and if so, how expensive it might be? I appreciate any insights or experiences you can share on this matter.

grizzled flax
clever thunder
plain prairie
grizzled flax
clever thunder
#

I was assuming he would not understand the difference between American football and European football that is why I considered fine tuning it. Do I have the wrong idea?

grizzled flax
clever thunder
#

so is this currently not yet supported - the fine tuning image recognition?
for example I assume that the model would know what is in the image but he would not know the rules of the game etc?

grizzled flax
# clever thunder so is this currently not yet supported - the fine tuning image recognition? for ...

Correct - image recognition is just a feature of that specific Vision model of GPT-4. It can't be fine-tuned like the text models can.

As for whether it would know the rules: I'm sure it could do this at least in part. For example, it could use image recognition to see that a provided image shows a volleyball match, then you could ask it to tell you the rules of volleyball, and it could probably do a pretty good job.

clever thunder
#

Okay thanks for your explanation.
Then this idea that I had makes no sense (yet). I was trying to create a demo project of image (video) recognition and ask the bot what could be done differently in some cases of the game or his opinion on how the game should be played.

grizzled flax
# clever thunder Okay thanks for your explanation. Then this idea that I had makes no sense (yet)...

I don't know enough myself to give you a solid answer, but I think it could be difficult with the current limitations of image recognition. If you were to build something like this, it would almost definitely be via the API by programming your own project, and #dev-chat might be the best place for that. But again, like you've said, that level of video analysis might be beyond the current scope of the image recognition feature.

storm raft
#

Hello everybody! I'm new here.
Has anyone ever done a project regarding psychology / psychotherapy?
I want to meet likeminded people that want to improve our field of work while using gpt. I myself have a project, but i would love to discuss other people's ideas

manic moat
#

guys ChatGPT's ability to remember things you discuss to make future chats is this option avaliable for free user ? or its avalable only for plus user??

static isle
#

This is only to a small portion and that portion will not get it all at once

manic moat
supple pike
#

how do I sign up for chat gpt memory?

frozen vortex
#

hey all, just wanting to reach out to see if anyones had any issues with plugins today?

green spire
lucid turtle
lyric epoch
#

Hi everyone,

I am trying to build a custom Assistant for retrieval from structured data (xlsx, csv).
It seems that Knowledge Retrieval doesn't support xlsx, csv files.
What's the best way to build an agent that can answer questions from tables?

gleaming mural
#

I'm getting extremely frequent Error in input stream errors that happen about a second after my custom GPT already finished its response

#

Is anyone else also getting this error?

#

It might just be my bad instructions tbh

#

Also, does anyone know the limits of GPTs knowledge? It might be because I'm giving it a 5MB file of raw text

cunning zinc
#

I keep getting red errors for gpt 4, anyone else?

#

Even after clicking regenerate multiple times

carmine zodiac
#

yeah same here

floral reef
#

yes, same. Errors for GPT 4

jovial raven
#

ditto, same here, I get maybe 1-2 sentences or sometimes nothing before "Error in input stream" or some other random-seeming red error box

lucid tartan
#

i keep getting this ERROR IN MODERATION on chatgpt

signal prawn
dawn tartan
#

Okay so it's not me... I was like okaaaay why does chat gpt not like pyscript

flint dock
# 
We ran into an issue while authenticating you. If this issue persists, please contact us through our help center at https://help.openai.com

🤨

median sapphire
#

The server is having problems. Please try again later. (85561dad3b9e996f_SJC)

somber jacinth
#

lol

flint dock
#

It is dead, Jim.

somber jacinth
#

sand in the minge

stoic wedge
#

hahaha

wind urchin
#

same

stoic wedge
#

also same

hybrid crystal
#

On a similar topic, does anyone have any ideas for estimation of inference compute requirements for Chat GPT-4?

#

One way to go about it is timing a number of responses through the GUI to get an upper bound on time required for inference accounting for server problems or high computing events and variance in prompts

#

Then making assumptions on server side hardware as well as overhead time (comprising latency, authentication, rendering, etc.) which gives us a rough estimate of time/token in context of hardware specs

cunning zinc
#

Guess everyone is having same issue lol

mild cape
#

same

granite lintel
# signal prawn same here

Good to know that I am not the only one. Could this be related to that they rolling out the memory thingy to some users? I remember that this kind of error thingy has happened before when they roll out something new.

stoic wedge
#

lollll i got an email saying my moderation error was when it detects content that doesn't align with usage policies. i guess me wanting to write an email about employee benefits isn't a good use?

elder skiff
#

Network error or Moderation...

cunning zinc
#

It gave me that error when I asked what the code usually is to make text go to the left of the screen lol

split veldt
#

there is a login error, there is a moderation error and there is a HMM SOMETHING SEEMS TO BE WRONG error for all and any prompt u put in any GPT rn like 4.0, 3.5, API; everything is down

cunning zinc
#

yes lol everyone has it

#

Or a lot anyways

split veldt
wind urchin
#

seems to be back up for me

flint dock
#

Nope

We ran into an issue while authenticating you. If this issue persists, please contact us through our help center at https://help.openai.com.

wind urchin
#

Ahh, I was already logged in

#

ok well, the chat system seems to be backup at least, good sign

alpine orbit
#

I wasted an hour of my life clicking generate again and then received a message that I had reached my query limit

oak rain
#

i was just able to update one of my gpt

#

with files

#

so they are working on getting it running

alpine orbit
#

I have 2 hours free so I will test Gemini from Google during this time

mossy sand
#

Gpt is completely bandjacked on mobile. Getting black screen now

static isle
flint dock
# static isle What do you think?

I gave it a text analysis task and it answered: it would cost 3500 € and takes 1 - 2 weeks. Thats it, no further solution.
I also asked how I should provide documents for it and it answered: send me an email.

wind urchin
#

anyone know how to use chat gpt and not get detected

#

lol

cold trout
wind urchin
cold trout
wind urchin
#

ohh

cold trout
#

What do you mean by "detected"?

wind urchin
#

like when you run if the text is written by ai

#

and not get detected that your text is written by ai

cold trout
#

Oh heh. Could try running things through something called quillbot. Heard people have success with that but it just depends. Some services detect it as ai when it isn't and sometimes the ai writing is ignored. Its not a perfect science yet

wind urchin
cold trout
#

Might also try rewriting ai content in your own words if you really need it for ideas and whatnot

wind urchin
#

what stuff should you rewrite in order to make it not ai written?

cold trout
#

I mean as a whole. 1. You create ai writing with chatgpt. 2. Take that writing and rewrite it yourself in a different way using your own words.

#

Just like quillbot would do

#

Can ask it to try changing up styles. Limit is your imagination

#

Good luck

wind urchin
#

I tried rewriting it even changing a lot of word to my own but got detected lol

cold trout
#

You see. They arent foolproof systems that get used to detect

#

At risk either way sometimes

wind urchin
#

what is the best detector?

cold trout
#

No idea. Been out of that loop since the news starting fomenting fear that everyone was gonna cheat with it

wind urchin
#

alr thx man @cold trout

cold trout
wind urchin
#

does the server get anything better?

cold trout
# wind urchin does the server get anything better?

After reaching Level 3, continuing to boost the server doesn't unlock additional Discord-provided levels or perks, as Level 3 is the highest. However, maintaining or increasing the number of boosts can be beneficial for several reasons:

Buffer Against Boost Loss: Members come and go, and so do their boosts. Having extra boosts beyond what's needed for Level 3 can help cushion the server against dropping to a lower level if some members stop boosting.
Community Engagement: Boosting can be a way for members to show their support and engagement with the server community. It can foster a sense of ownership and pride among members.
Server Customization: Some servers use the extra boosts as a way to further customize their server or as a metric to unlock server-specific roles, perks, or features decided by the server admins, not Discord itself.

In essence, while the tangible benefits of boosting a Discord server cap out at Level 3 in terms of Discord's provided perks, the community and engagement benefits can continue to grow with additional boosts. Server owners and admins often leverage these additional boosts to enhance their community's sense of belonging and engagement.

crude reef
#

Hello, I'm new to OpenAI, how can I purchase a Gpt 4 Key?

wind gulch
#

can you hire me open ai

grizzled flax
wind gulch
#

im not us citizen;/

grizzled flax
# wind gulch im not us citizen;\/

I don't know if it helps ya, but other than the San Francisco and New York City listings, there are also listings for positions in Brussels, Dublin, and London, and one listing that looks like it has a remote option! Be sure to click "View all open roles" if you're interested.

wind gulch
#

thanks ill check definately

lusty pike
#

So will this new feature cause even more gpu demand and make gpt 4 worse

lucid turtle
#

yep

#

also probably will limit the message per hour even more

#

or there will like 10 generation every 3 hours

#

for the new video thing

low cobalt
#

Does anyone know if there is any info about Sora API?

vestal basalt
#

where do we use Sora? I'm on ChatGPT 4 and it says it doesn't generate video...

knotty grotto
knotty grotto
vestal basalt
#

Thanks!

low cobalt
#

they are testing it first before releasing it, just wondering if there was any news on that

neat tendon
#

Sora looks unbelievable anyone knows a text to video model as good as Sora ?

unkempt thorn
neat tendon
knotty grotto
dim niche
#

yo, can I earn money with my gpts?

dim niche
icy haven
#

how to use sora

dim niche
icy haven
#

no

dim niche
#

F

nimble spruce
#

Is sora only available with plus subscription?

grizzled flax
nimble spruce
sage rock
#

aka: It involves third party, it involves you having to implement payment system and an API whether to check if the user is on the paid plan before ChatGPT even generates answers

tall arch
#

After i ask for video... The A.I give me that...Ah, I see what happened. It looks like you've been rickrolled! This is a playful internet prank involving the music video for Rick Astley's song "Never Gonna Give You Up." The link provided was part of the joke.

Regarding your original request for a video of a little monkey walking in a jungle, please note that as a text-based AI model, I'm unable to create actual videos. My response was part of a humorous, light-hearted interaction following the theme of the rickroll. If you have any other questions or need assistance with something else, feel free to ask!

leaden agate
strong igloo
#

Convert into animation video

nocturne anchor
#

Hi, can I connect my gpts to dedicated flask server to receive json format info?

supple violet
nocturne anchor
supple violet
#

what error?

nocturne anchor
#

Server URL 127.128.129.0:1234/action is not under the root origin 127.128.129.0; ignoring it

supple violet
#

thats a private network address range

#

you need to have a public ip and the service needs to be available to internet (openai service)

nocturne anchor
stuck canyon
#

Hi all,

I cannot save and publish a custom GPT. I get the error message "Error saving GPT". In the console I see the following error: "_app-b2c0a13b23e24b5b.js:26 FatalServerError: Something went wrong. If this issue persists please contact us through our help center at help.openai.com."

My GPT has 15 json files as knowledge base. Each file is about 2-3 MB in size.

Does anyone know what the problem is?

nocturne anchor
mental laurel
#

Why does GTP 4 suck now, super slow an bad responses, ive went back to 3.5 and its much faster and giving more accurate results. What am I paying for?

brave tiger
#

I think people forget OpenAI is a non-profit

vernal oyster
#

Yes and no, subsidiary OpenAI Global LLC is for profit

#

OpenAI Inc. is non-profit

main chasm
muted forge
#

好

rose stag
#

New to all this. This is prob a repeat, but I have'nt seen the questions above. Why is ChatGPT sooooo, slooooow today. (4.0) Was bad for the last few days but today it keeps erroring out. Any other way to use 4.0 other than the site?

left forge
#

Not slow for me...and I use custom instructions with 4. I can't remember the last time I had to use 3.5.

glossy zinc
#

Has anyone received the ability to call in @-gpts yet

#

Just curious as long term memory and sora has been announced. I still can't call in a gpt

left ocean
#

Does anybody know, how can I do this in ChatGPT android app?

  1. Ude "@" to call GPTs into the chat?
  2. How can I edit already sent post as I can do in browser version of chatgpt?
  3. Why chatgpt in android app use "analyze" instead of "reading" as chatgpt do in web version?
grizzled flax
# glossy zinc Has anyone received the ability to call in @-gpts yet

Yes, this feature is out to many users right now. If you still don't have it, make sure you try in the browser version, because I don't think it's been added to either app version yet.

As for memory: this is still just in early release, and they haven't shared plans for wider release yet, which is what they said the next step will be.

As for Sora: this hasn't been announced to be coming to ChatGPT at all, so it's possible that it ends up a different service, or it's possible that release one way or another is a while in the future anyway. We don't know many details about this at all yet!

grizzled flax
glossy zinc
#

ty @grizzled flax

A little annoying as I've tried multiple browsers but never mind. I'm sure I'll see it eventually

left ocean
simple cosmos
#

Can someone give me a gpt 4 referral link or let me use their gpt 4 account I need it for school

slate wren
#

@ seems to be malfunctioning again. I resumed a conversation from a few days ago and ChatGPT started making stuff up instead of adhering to the GPT I created and was @ with

#

the weird thing is that the @ worked when I started the conversation those days ago, and there were no changed to the GPTs involved from my side

slate wren
#

and it started combining messages I've sent again

harsh drift
#

Yall can somebody help me out achieve this? :
I want to basically make chatgpt control my social media page for an hour and see how realistic it can go in replicating my personality and make it respond to my friends on instagram dms. So it has to be fully automated without my control, anybody whos got an idea can help me do this? Thanks 🩵

distant hornet
naive belfry
harsh drift
#

so that even when the browser is closed, it still replies

#

and keeps the DM conversation saved so it knows the context when replying to each person

naive belfry
#

mate - i dont even think Chatgpt will be able to do this for you....

distant hornet
distant hornet
hybrid crystal
#

Anyone have any idea why it cost so much for Open AI to train GPT 4?

#

Available to any consumer:
https://lambdalabs[dot]com/service/gpu-cloud#pricing
It costs $14.32 / hr for an 8x NVIDIA A100 SXM instance (Has plenty of RAM, storage, etc.)
Scaling this up to ~25,000 A100 GPUs and ~100 days of continuous training, the available RETAIL cost is between: $77,400,000 - $107,400,000

#

Despite economies of scale and B2B availability, "Sam Altman stated that the cost of training GPT-4 was more than $100 million."

#

This parks their potential discount at a maximum of ~5% (Likely much less)

#

Sure there are overheads and R&D, but tests to estimate large model performance are done with models only a fraction of the size (<1/1000th of the compute)

slate wren
#

another weekend I won't be able to use gpt because it keeps combining my message sadface

harsh drift
distant hornet
dark ivy
#

somebody still have the problem with gpt actions when calling it , i'm getting "Stopped talking to" immediately

vale python
#

anyone else facing the issue that custom GPT can't access knowledgebase files, or access pdf during chat?

slate wren
#

It is anyone having their gpt jumble previous prompts together again?

lethal sorrel
#

What's the difference between using the GPT 4 turbo API for 4096 token output and the ChatGPT ~800 token output for a given task? For example, If i say to the chat mode to divide up the task into 5 parts, and then ask it to continue after each part, 5 times, then, would the output be similar to if I had asked the playground API to do the task? After all, chat mode is just looking at the past tokens and predicting 800 ahead. So would it be equivalent or, basically the same?

To clarify, asking the ~800 token output chat mode to extend a response 5 times, would net a long response. But would it be as seamless as just using the Playground API 4096 token output mode? I want a long response. But asking the chat mode for a long response is impossible, it is necessary to divide up the answer into 5 parts, to get an answer that is eventually of the large length. So it is basically 6 enter keys. {Divide task into 5 parts} {Part 1} {Part 2} {Part 3} {Part 4} Part 5}. But the Playground is all just one enter key. However, what is the QUALITY of the output? It really depends on how WELL the task division is in the first place, and is not something the playground API model needs to explicitly do. That said though. 800 token outputs 5 times, would that net a similar quality to 4096 tokens at once? After all, the 800 token output has the previous answer already, it is in the history, that it fully considers in the new token generation, so it should not result in a diminishing of quality in any way. However, it does seem like a 4096 ouput may have some sort of advantadge, potentially. As it does not need to divide up the task into sections in the first place, and that may allow for a more seamless problem solving approach. So, is there a difference to quality that can be expected, or not?

autumn oriole
silent vault
#

Hey, same problem! My gpt no longer accesses its knowledge base and suddenly doesn't use it at all

wise gorge
#

Does chat work now? It keeps me saying “Hmm… something seems to have gone wrong” 3 day in a row, once I ask to work with data attached as a picture .

knotty grotto
#

do you ahve a vpn on your device

dim niche
#

@open crescent yo is there a GPT Store mod on this server?

knotty grotto
cursive kiln
#

is there any ai form filler?

twin thistle
#

Is Sora available?

grizzled flax
# twin thistle Is Sora available?

No, not currently. It's in early testing by red teamers and some creators. No announcements yet about what wider access might look like!

twin thistle
#

Thank you 🙂

vast cove
#

I'm starting to get highly anxious that we won't even see GPT-5 for another year. What is taking so long?

I know this is petty, but I also hope that they resolved the repetitive word choices that GPT-4 uses.

grizzled flax
# vast cove I'm starting to get highly anxious that we won't even see GPT-5 for another year...

It's helpful to keep in mind that models like this aren't developed and released on a schedule like, for example, how smartphones are. It's very new technology, so it comes out in a "it's done when it's done" kind of way.

For reference, though:

GPT-1 to 2 was about 1.5 years.
GPT-2 to 3 was about 1.5-2 years.
GPT-3 to 3.5 was about 1.5 years.
GPT-3.5 to 4 was about 1 year.
GPT-4 Turbo was just released 3.5 months ago, and baseline GPT-4 was released about 11.5 months ago.

Again, this doesn't give any indication about when we might expect any kind of GPT-4 "successor", but it's a good ballpark of how things with GPT have gone so far.

distant hornet
austere peak
#

guys, is there a GPT out there that is very good at taking my original prompt or main idea/concept and craft a nicely written GPT prompt for me? ive tried one so far and seems ok but wondering if there are some others vetted out there already? thx

left ocean
lethal sorrel
vast cove
grizzled flax
# vast cove What would you qualify as done? It doesn't seem like any of the models are done ...

I suppose by "done" in my message I meant "released as usable in some significant capacity." We have seen iterations on GPT-4, you can see this in the history of GPT-4 models on the API (https://platform.openai.com/docs/models/gpt-4-and-gpt-4-turbo), and again, Turbo just released in November, less than half a year ago.

Considering that GPT-3 was succeeded by GPT-3.5, my money would be on 4 being succeeded by 4.5 as well, but we're still only about a year into 4 anyhow, and again, much less time into the era of 4 Turbo.

distant hornet
#

With RLHF, it's never really "done" in an absolute sense, but needs to be mostly done for safe use.

vivid apex
#

I’m finding that an Assistant on gpt-4-1106-preview doesn’t want to write a 2,000-word article at that length. I don’t know whether it’s due to the prevalence of a lot of material I have used as Retrieval info, or if it would struggle to focus that long anyway.

Anyone know any workarounds? Eg. I gather one technique is to ask it first to produce an outline, and then ask it to do each section separately?

distant hornet
#

It can't plan, either.

vivid apex
distant hornet
#

Step 1 have it create an outline.
Step 2 manually plan outline breaks based on document length required.
Step 3 feed it the pre-planned payloads.

vivid apex
distant hornet
#

Yeah. The broken up outline

#

Feed it chunk a, collect output.
Feed it chunk b, collect output.
...
Feed it the final chunk, collect output.
Collate output.

#

Your path to 'n' words requires you to plan because the AI literally can't.

unborn heath
#

But it's awesome at driving nails into the spots you choose, if you design it well. Terrible at drawing you a blueprint and figuring everything out. Perfect if you're the archetect, checking everything, and giving the model 1 nail-sized job at a time

vivid apex
distant hornet
deft imp
#

Hi 👋

distant hornet
deft imp
distant hornet
#

Great thanks

unborn heath
# vivid apex I’m finding that an Assistant on gpt-4-1106-preview doesn’t want to write a 2,00...

it'll reliably write about 500 words at a time, though that is a rough estimate, can go 300-800, rarely even more or less.

So I would plan with the model, about 5 chunks, and what the chunks should be.

I'd have the model make each chunk with my guidance, and check it to make sure I like it.

There will be trimming on the overlap, where a section ends/begins. You can have the model help you with that too, but you have to guide that and set it up as another 'nail' too.

vivid apex
# distant hornet Uh, depends on the context. What's this for?

“Write a 2,000-word blog post” on x/y/z. This would be deployed as an Assistant, backed by a Retrieval file comprising the blog’s archive, to be used as partial reference material. It keeps only hitting circa 800 words, repeatedly, for the reason you mentioned.

deft imp
#

Cool, I'm new , I was thinking that it's chat gpt , I want to make pictures with ai , Is it available here?

distant hornet
distant hornet
deft imp
distant hornet
vivid apex
# unborn heath it'll reliably write about 500 words at a time, though that is a rough estimate,...
  • @DarthGustav
  1. Okay, so we’re all aligning on the need to generate an outline first, which may be a multi-point/detailed outline?
  2. Then sequentially, manually feed it only the relevant parts of the outline to write-up? Each output will be of a length that it’s not possible to specify? (Hopefully, memory/token length won’t be a problem within the context of the current outline part each time?).
  3. Manually assemble the pieces at the end. Optionally, feed the whole thing back to it, asking it to tidy up any duplication across sections? (@unborn heath, can you clarify “You can have the model help your with that too, but you have to guide and set it up as another ‘nail’ too”?)
distant hornet
#

Yes, except I suspect it won't be able to output it all back to you.

#

It might surprise me but I don't expect it to.

#

It can output at most about 1200 words from my experience.

vivid apex
distant hornet
#

@unborn heath is right on the typical output though

unborn heath
# vivid apex + @DarthGustav 1. Okay, so we’re all aligning on the need to generate an outline...

It can try to do it all for you, but it doesn't handle length well.

Ideally you're going to read it yourself, decide what satisfies your need, and maybe even just you throw away a paragraph that's duplicated between two sections, or was intended by the AI to sum up a section that you want to flow seamlessly into the next chunk, so you just delete that paragraph and immediately use the other section, or maybe you delete the header on that second section and go from there - or maybe you have the AI help you merge them, maybe by taking the last 2 paragraphs + first 2 paragraphs to show the model to rewrite into a seamless part that works.

You are the 'human in the loop' and this is not something you get awesome results with if you only automate it. I recommend read, think decide, use the AI for more help or not as you see fit.

distant hornet
#

So it'll try to be done.

vivid apex
distant hornet
#

Python can. But your output is going to be bad.

vivid apex
#

Thanks for the thoughts.

#

Hmm, I’m in Section 1 (ie. Outline generated, now asking it to write parts), and it’s writing too much stuff describing the process itself, as a commitment to a future process, ie. “This part will delve into the importance…”, “This subsection will touch on the…”, “This concluding subsection will emphasize”… rather than writing the substantive material now.

distant hornet
unborn heath
# vivid apex Fascinating stuff. And we’re not at a point where there’s anything out there wit...

I don't recommend that, that's a human's job.

However, here's a fast example of something roughly like what you described, 1976 words, took me 18 minutes from opening the ChatGPT website to done, and that's with a little editing (not much). Clean up to my standards would take me about 15-20 minutes, I'd do most of it myself personally. https://chat.openai.com/share/608e3ec3-ec2e-4784-8e9f-351a86c25a80

distant hornet
#

I largely agree, but an output template can get more desirable output per section.

unborn heath
distant hornet
#

For the best output templates, encoding the summarized instructions into the template variable names can work wonders.

vivid apex
vivid apex
distant hornet
#

Okay

#

Imagine what you want to see in its most generalized form. Show that to the AI at the end of the prompt as a template.

vivid apex
# distant hornet Okay

Ok, so, like, when asking it to write each section, show an indicative example of what such a thing would typically look like, so as to dissuade it from getting meta?

distant hornet
#

Think far more generally. I don't know what you might be looking for, but here's an example:

SECTION 1
{Factual blog style three paragraph response absent any meta commentary on its construction}
END SECTION 1```

You'd do similar for each section.
vivid apex
distant hornet
#

It's triple redundant with the actual instructions this way too

unborn heath
# vivid apex Ok, so, like, when asking it to write each section, show an indicative example o...

To 'dissuade', tell it what you want. If possible, don't even mention the stuff you do not want, fill its attention with what you want it to do.

Examples can be useful for this too - if they accurately are what you want.

The model does what it is 'thinking', so if you tell it:

Do not X!

Well, what is it thinking about?

It knows it shouldn't X, but what else to do?

So it does X.

Instead tell it:

Do 1, 2, then 3. A little A, some Z.

And it's unlikely to do any X, because it's so busy doing all the other stuff you do want, it doesn't even consider X.

distant hornet
#

That's a good idea

vivid apex
distant hornet
vivid apex
distant hornet
#

SECTION {n} and END SECTION {n}

tulip zealot
#

I created a GPT nothing in the GPT breaks any rules. I then asked a question regarding quitting an adult habit, and it replied but it said "This content may violate our content policy. If you believe this to be in error, please submit your feedback — your input will aid our research in this area.".

Can I safely ignore that or is my account at risk of closure if I don't give feedback?

unborn heath
# tulip zealot I created a GPT nothing in the GPT breaks any rules. I then asked a question reg...

Policies are here: https://openai.com/policies/usage-policies and https://openai.com/policies/terms-of-use

In my experience, if you don't actually break the allowed content rules or TOS, but see warning(s), I've been fine. I've seen a lot, I tend to explore gray areas that are within the rules but can be close or subjective, especially to an AI moderator, it seems. I still have my account access, I've never even gotten an email discussing any concerns - but I absolutely do fill out the feedback if I see any reason to. That's not always 'to try and protect myself' - sometimes I want to give OpenAI feedback on the situation and my context for the prompts, and on the impact of getting the content warning, if I think that is worth reporting.

I've also seen many messages from people on this Discord forum about losing their account access, and I've never seen even one come back to us and say 'whew, that is fixed!'. They might get access back and not come talk about it, of course.

But if you break the allowed content, it's probably likely that you'll have issues; if you don't, you're probably fine. The automated content warnings themselves do not seem to be anything major (I'm sure I've seen well over 100) but a person that's actually breaking the allowed content or TOS is probably going to lose access even if they never see a content warning flag.

opal oyster
#

So, is it just me or is GPT painfully slow these last few weeks?

unborn heath
harsh parcel
#

How much is GPT learning from people using it and interacting?

harsh parcel
# unborn heath Policies are here: https://openai.com/policies/usage-policies and https://opena...

It seems pretty hard to break those rules anyway. I mean, they’re small: don’t do anything illegal and don’t do anything to hurt anyone, including yourself, and don’t use the content to scam.

The last one is pretty hard to prove since the content is private. I’m not saying it’s okay to violate, I’m just saying it would be pretty hard to enforce.

Beyond that, it doesn’t tend to let you so anything that’s a violation of any laws or that has the potential to cause harm.

There is a way to get it to tell you how to do things, but I shan’t say what that is because it has a real use-case too (rather, it’s the misuse of a legitimate use) and if too many people know it it will harm the real use.

unborn heath
# harsh parcel It seems pretty hard to break those rules anyway. I mean, they’re small: don’t d...

Good call. We're specififcally not supposed to discuss 'jailbreaks' or other ways to break rules anyway, them's the Discord rules! #server-rules.

Once a person's making a GPT or using API though, the allowed content gets stricter, it's not just the core rules alone anymore (and just a couple months ago, the core rules were the same for every OpenAI product, and considerably longer and more detailed! We have far more private-use freedom now than before)

But bazil said "I created a GPT" - I think that means a custom GPT, and it potentially is under the GPT ruleset, which may involve how we talk to the model (it certainly used to).

The model itself was trained under the more restrictive allowed content, it may also struggle to adapt to the new freedoms for all I know.

#

Does seem to purely involve 'building' though, not using. If so, yay. The old ruleset was vastly more restrictive, though we had this blog post to hope for better times ahead: https://openai.com/blog/how-should-ai-systems-behave

"2. Define your AI’s values, within broad bounds. We believe that AI should be a useful tool for individual people, and thus customizable by each user up to limits defined by society."

"This will mean allowing system outputs that other people (ourselves included) may strongly disagree with."

There's a chance the model and any AI-assisted moderation software just needs more time to be trained to the new ruleset. The old ruleset, well. It was sure something.

I kinda wish they kept up the old usage policies as legacy, so we could more easily discuss how things used to be and how the rules have changed over time.

harsh parcel
#

Right, I meant I don't even think of it as a gaolbreak of any sort, just as a "this is a way I ask for completely other and legitimate reasons, but if someone were to do it not for those reasons it would give the same results". But beyond any rules, I don't want to break how I use it!

#

What I mean is, I think of a gaolbreak being something that would bypass restrictions, but that that's all it's for, so I didn't even think of it in that terminology.

harsh parcel
unborn heath
opal oyster
unborn heath
ebon nest
#

guys, how fix problem - only gpt4, only on browser - not work! Not response, not reaction from chatgpt for all browsers. I Have subscription, and tested this case for 2 computer and 3 browsers (edge, firefox, chrome) and tested relogin.

GPT 4 not work in browser (for phone is work)

scenic moon
#

what do you think chatGPT would say about an AI with peer2peer protocol ... power of all pc's in the world

#

like napster & co but an AI ^^

distant hornet
eager shell
#

Guys,do you think if i wuaant to build an AI assitant and then emmpower it to an AGI myself ,is it a good idea to first build a GPT so i can have an idea of how he would look like ?

#

like just to show a demo of what will be capable to do ?

distant hornet
eager shell
#

there is still a problem

#

and i don't know if i should pay more attetion to my home town or to the world

#

i wuanted to build something useful for humanity but i also wuanted to get along with my home country... (Portugal)

#

i just dont know what audience should i be looking for

#

i could build a chatbot globaly but there are tons of Assistants today that it would be hard to reach the best

#

but in my home country there isnt pratically not a single chatbot specifically made for anything yet

distant hornet
buoyant tide
#

Hello - is anyone else facing problems with custom GPT knowledge bank retrieval suddenly? I have a number of txt files uploaded that my custom GPT will normally search (without code interpreter). But due to some mysterious change - the custom GPT will search online as a first choice despite my prompt instructions. If I switch on code interpreter it will kind of load the ENTIRE text - which is dumb (instead of doing a vector search - which I assume is what the kjnowledge bank is)

#

any advice welcome..

harsh parcel
distant hornet
buoyant tide
#

There seem to be some change to the agent behaviour - hm

distant hornet
#

I understand that. I'd try this anyway because my vector search for my GPTs is working.

buoyant tide
#

what is your file format?

distant hornet
#

Plain text, sparse xml tags

buoyant tide
#

mine is txt but in jsonl format (I also tried with jsonl direct). Thanks for advice...will reload

distant hornet
buoyant tide
# distant hornet No worries. Consider plain text RSV, or just plain text markdown.

unfortunately nothing has worked. I even built a new GPT and cloned the old prompt and reuploaded. I can't paste a screenshot here - but the GPT will either say it can't search the bank or it will hallucinate quotes from my files or inexplicably search online. All my GPTs dependent on text files have suddenly failed - this is pretty frustrating as this is done for work =(. And I doubt anyone at OpenAI is paying any attention..

#

Really hope the RAG pipeline can be made a lot less opaque.

distant hornet
#

My GPTs also use knowledge, extensively, and are working.

buoyant tide
#

INo trade secrets - it's basically a literary analysis bot

#

When you receive the User Prompt, analyse the prompt, and find 10 relevant quotes from the literary text in the uploaded Markdown files of Knowledge bank.

If the work is Shakespeare's The Tempest, in which you should IGNORE all references from the uploaded Knowledge, and cite quotes with Acts/Scenes references based on your prior knowledge.

Then you must output an analysis in the following format.

Format:

Introduction:
Summarise the text and write about ideas and literary techniques in the text relevant to the prompt. Write down the overall thesis that answers the User Prompt.

Paragraph 1:
State a point that addresses the thesis statement.
Cite a first quote from the text (with page numbers or Act/Scene), and explain how literary devices such as imagery, setting, diction, structure, atmosphere, plot, voice the narrator and speaker, and so on in the quote support your point. Cite a second quote (with pg numbers with Act/Scene), and explain how literary devices in the quote support the same point.
....

#

this is part of it

#

This used to work well - it will search Knowledge bank. Now it will either fail to search after trying or go online

distant hornet
#

I can possibly help. This looks like a fine tuning vs instructions issue.

#

I'll need to see the full set of instructions. I'll rewrite it with your permission, and you can try the rewrite for your use case?

#

There's a lot of this going on and I want to get to the bottom of it.

buoyant tide
#

for free? Well I have to send a Google doc link here.

#

What's finetuning got to do with it? U mean OAI's own finetuning

distant hornet
buoyant tide
#

certainly the agent has gone 'crazy'

distant hornet
#

Your instructions are not worded like mine are, at all. They seem unclear from the model perspective.

#

If this is the case for all these reports it makes sense

#

The AI pre-tuning likely adapted.

clever magnet
#

How long is it until we can use sora?

distant hornet
clever magnet
#

Are we talking about a month or a year?

buoyant tide
#

what's ur advice to make the Knowledge search clearer?

distant hornet
#

i'm not being mean, they're just really clunky wording.

buoyant tide
#

can u help me reword that first part

clever magnet
buoyant tide
#

since the rest don't involve this

distant hornet
#

obviously no warrantee or anything. i just want to get to the bottom of this.

distant hornet
#

I've seen it count like this before fwiw:

1 {text}
2 {text}
3 {text}
5 {text}

buoyant tide
#

they are in JSON lines. But actually that was not in my original working prompt. U can remove it if u want

nocturne grotto
#

i love json mode

#

its too good

distant hornet
# buoyant tide Unfortunately my google doc sharing is bloacked by the server. The model was per... 

IF the work is Shakespeare's The Tempest THEN ignore all references from the uploaded Knowledge, and cite quotes with Acts/Scenes references based on your prior knowledge. 

Then output an analysis in the following format:

Format: 

Introduction:
Summarise the text and write about ideas and literary techniques in the text relevant to the prompt. Write down the overall thesis that answers the User Prompt.

Paragraph 1: 
State a point that addresses the thesis statement. 
Cite a first quote from the text (with page numbers or Act/Scene), and explain how literary devices such as imagery, setting, diction, structure, atmosphere, plot, voice the narrator and speaker, and so on in the quote support your point. Cite a second quote  (with pg numbers with Act/Scene), and explain how literary devices in the quote support the same point.
....```
#

replace filename with the file name

#

if there's more than one, lightly reword it to list all the files.

distant hornet
#

I didn't remove the 10 figure because numbers are unimportant for literature anyway. It's an artificial constraint, though, that you may decide to remove.

buoyant tide
#

when u put '/mnt/data/{filename}' do u assume the GPT will identify the right file?

#

Or should there be one combined text file only. I have tried both approaches

distant hornet
#

No, you need to replace

wraith ginkgoBOT
#
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.
distant hornet
#

Scroll up

buoyant tide
#

ok i try my combined text file I use earlier (when things were still working)

distant hornet
#

Nice rules drop by the idiot bot.

distant hornet
buoyant tide
#

'''It seems there was an issue accessing the knowledge source to find information on Katharina Blum. Based on my existing knowledge, Katharina Blum is a character from Heinrich Böll's novel "The Lost Honor of Katharina Blum," or "Die verlorene Ehre der Katharina Blum." '''

#

I have been getting such messages for some time... whatever prompt I use. But thanks very much for trying. Putting in the file path is def a good idea...though there's something weird and buggy going on

#

I use '/mnt/data/combined.txt'

distant hornet
#

Try removing just the /mnt/data but leaving the filename.

#

The issue may be copyright.

#

But this is the final test before that's the most likely culprit

distant hornet
buoyant tide
#

perhaps it is copyright issue but yes it's still not working.

#

perhaps the 'finetuning' has caused even educational uses of this to fail.

distant hornet
#

my knowledge is working fine, and i created a test gpt with knowledge which also worked

#

i think the summary tool is the culprit here

hollow mulch
#

Would love to see examples of GPT's optimized for Voice Chat. I'm thinking along the lines of job interview practice.

unborn heath
# hollow mulch Would love to see examples of GPT's optimized for Voice Chat. I'm thinking along...

While that isn't what you asked for, you can absolutely set up one that is customized exactly for what you want and need, just with the basic normal ChatGPT. You can even use 3.5 for this, it's very good with job interview type stuff and 4 is rarely better (maybe for some very unusual jobs, otherwise 3.5 is very good too for this kind of task).

Just tell it what you want, like you'd tell a person:

Hey, I want to practice my job interview skills, be a interviewer for me for [company] and I've applied for [position]. Walk me through an interview, then discuss my answers with me.

hollow mulch
unborn heath
# hollow mulch I created a custom GPT and used custom instructions to make it concise but helpf...

Yeah! Literally tell the model what you want, like you're talk to a person, tends to work well.

Giving it a style to follow also works well. Examples can be great.

Custom GPTs are a bit different because you only get 25 messages per 3 hours with them, which isn't a ton for testing, making a change, retesting.

So I recommend you consider using regular ChatGPT, even 3.5, to design and refine your instructions as a regular prompt, because you have so many messages with that (really a huge number of tokens, it's not per-message limited)

Once I had a great prompt with 3.5, I'd test it on 4 (40 messages per 3 hours, 25 of which can be through a custom GPT), and make sure I'm happy with it on 4 - then I'd test it as a custom GPT instruction, and verify I was happy with that too.

hollow mulch
coarse wadi
#

none of my gpts are able to access their files. this has gone on for over 24 hours. i wonder if all of these sora demos are using the resources.

late locust
#

Oh came here to ask the same. It seems my gpts cant acces their files. I am not sure though if they ever really could 😄 is there a limit in size that a gpt can handle?

distant hornet
hollow bough
#

Has anyone else tried to set up zapier actions? Can't quite get it to work, not sure why.

#

to sum it up I am attempting to build a personal assistant which accessed my texts, emails, and calendars. I was able to get it mostly set up and identifying the actions I gave it, however I keep getting an error trying to execute it and can't figure it out.

If anyone knows of an alternative or wants to help, I'll give you 10 dollas

#

no lie

late locust
distant hornet
# late locust Yes for me

This is pretty likely the cause, as this is common for others with the issue, while my own GPTs aren't failing to load knowledge, even new test GPTs.

#

Probably not what you want to hear, but that's my current best guess.

late locust
#

Is there a way around it? The gpt is on privat only used by me

distant hornet
#

Server rules.

late locust
#

Ah wait yeah 😄 someone could use it for not private projects

distant hornet
coarse wadi
#

these GPT worked fine up until a couple days ago

distant hornet
#

I see. I'm out of ideas.

wind urchin
#

does gpt actions need an API key

#

cuz I wanna add memory to a private gpt that adds/reads things to a Google doc

#

and I have no API key

mild ledge
wind urchin
#

thank god

lucid turtle
unborn heath
prisma spoke
#

does anyone know how to get a public link for a dalle3-generated image inside of a custom GPT? the links that it shows in the debugger are not publicly accessible, and the link you get from right-clicking also can only be downloaded by you while logged in, not opened by another tab/server.

errant lintel
left sleet
#

has anyone implemented a memory feature in a customGPT? what did you use?

river flint
#

hello

#

Glyph of Let
Fail-gracious embeddings: Craft your cruxes with snapping and larruping modals for the blocks might fray (e.g., documents that turn other encumbrances, inexecutable extirpations or missteers from Tesseract, that might chouse you).

Fonduing the output chits: Once you mound the arable into scripts, you’d fancy to posture a saleable or eloquent egest to the prayer, it could be flowing the insights to the sear, ajar them on UI or however, you reckon and gauze in your onsets.

Attribute matching: For using or holding a flag for your crafting, outwear the JSON serials, MIME lookups, or any personal glosses you commeasure in the trail of your settings.

There you lope, well-stitched into the gaps, aspirating and ready to be broadened. If you implore a cinch for a more publicised, or a contract for further ameliorations, feel it snug to hit the askance. Chamber and Chirr! 🌟📄🔍✨. Hope that propels you ahead with global flair and native gear on what you’re unwrapping. Should any helper drones or fine-turning gold-dust be of your cry, I'm at your bell! 🛎️. Good luck on your coze and genus to brim. 🍀. Let’s keep the force billowing and the codes caviling! 🚀. Hope that propels you ahead with global flair and native gear on what you’re unwrapping.

is my chatgpt okay?

silk cedar
stuck solstice
#

Hi Everyone!

My wife did a full asignment with a GPT she built for herself, not understanding, that the preview window conversation would not be saved. She had a lot of work saved in that conversation and hit the back button and lost everything. Can someone help me understand if OpenAI has that convo saved somewhere we might be able to get back?

The assignment is due tomorrow - she is devestadet!

river flint
#

tell your wife its gg

maiden thorn
distant hornet
#

She should get right to work.

lucid turtle
#

once you do anything like tinker with instructions or files ir deletes it. but for assignments i reccomend doing the api rather than the gpt just because of the msessage limits

sonic edge
grizzled flax
elfin zenith
#

Hi, I do have a question to the files information openai gives on there help website:

"How many files can I upload at once per GPT?
Up to 10 files per GPT for the lifetime of that GPT. Keep in mind there are file size restrictions and usage caps per user/org.

What are those file upload size restrictions?
All files uploaded to a GPT or a ChatGPT conversation have a hard limit of 512MB per file.

All text text and document files uploaded to a GPT or to a ChatGPT conversation are capped at 2M tokens per files. This limitation does not apply to spreadsheets.

For images, there's a limit of 20MB per image.

Additionally, there are usage caps:

Each end-user is capped at 10GB.

Each organization is capped at 100GB.

Note: An error will be displayed if a user/org cap has been hit."

###########
What does it mean, that "each end-user is capped at 10 GB"?
I can upload max. 10 Files. Each File has max. 512 MB per file --> alltogehter 5 GB .... so what does it mean by 10 GB is it capped? How can I get 10GB when 10 files are in max. 5GB?

Can someone help me to understand?

grizzled flax
clear stag
#

Yo everyone

#

What's the difference between gpt 3.5 turbo 0125 and gpt 3.5 turbo?

grizzled flax
elfin zenith
crisp moat
#

I've been trying to create a GPT that mimics the writing style of my own work, but every time I try, it doesn't seem to work very well. Any tips/tricks? Or any general advice?

distant hornet
lethal sorrel
#

What happened to it/ Why did it break?

lethal sorrel
# crisp moat I've been trying to create a GPT that mimics the writing style of my own work, b...

Yeah that's intresting. I think there was a reddit post with some ideas on it a while ago. Writing style is certainly complex. There's the function to content word ratio, the average and median length of a sentence, the most likely length of a sentence in a series of 20 sentences, the words likely to be used. Complex! Maybe with 140k context, there can be numerous examples of writing you can have it "emulate" ? maybe?

#

OpenAI really needs to enhance the critical thinking and reasoning capability of the AI, It should not refuse to perform a task it interprets as against policy when normal mathamatical logic suggests it is acceptable.

distant hornet
tardy lantern
#

Hi everyone, I’m trying to create a GPT that takes transcripts from an interview and pulls interesting moments from them based on the context of the video that I’m trying to create. I’ve tried to give it very detailed instructions, but it doesn’t listen to many of the instructions, and often gives wildly unpredictable results. I’m new to GPT building. Does anyone know how to remediate some of these issues?

distant hornet
# tardy lantern Hi everyone, I’m trying to create a GPT that takes transcripts from an interview...

It really depends on what you want it to do, how well you're conveying that in the epistemology of the model, which differs from our own, and how big the transcripts are. To begin, I always recommend framing all instructions in a positive way. Eg:

Analyze the knowledge and select thematically, narratively, or visually distinctive excerpts and justify them based on two of the three criteria.

is far better an instruction than:

Analyze the knowledge and select interesting excerpts. They can't be boring. Don't make them disinteresting. I don't want anything visually unappealing.

Why? It doesn't show the AI what it isnt' supposed to do, so the AI doesn't do it.

It also provides a criteria that the AI can evaluate. Are these excerpts thematically, narratively, and/or visually distinctive? If so, they'll qualify if they meet two of the criteria, not just one. That's far more likely to meet your expectations than open-ended but negative instructions, just as an example.

The AI is like a child. Telling it what not to do is a sure-fire way to get it to do what you don't want it to do. Instead, use output templates and open variables that redundantly encode your instructions. Eg:

Here are some interesting excerpts from the provided transcript:

[From analysis, list code-block quotations of excerpts that pass at least 2 of 3 criteria for interesting content, along with a brief in-context justification for their inclusion, causally referencing which two or more criteria they fulfill.]```
#

If the transcripts are quite large, then the AI is relying on the (idiotic by comparison) summary tool to define "interesting". You'd need to define better criteria for exceptionally large transcripts than the incredibly broad and subjective solo category, "interesting."

tardy lantern
#

They are transcripts of 8 min interviews, of a family. They are individually filmed, an output template would probably help a lot

distant hornet
#

8 minutes is approximately (very loosely) 8k tokens at about 1k tokens a minute speaking text.

#

That's very ballpark, of course.

tardy lantern
#

So it shouldn’t be too large right?

distant hornet
#

You'll want to process each query in a separate context (new instance), whether that means clearing the API history or starting a new chatGPT session with your custom GPT or custom instructions.

distant hornet
#

That's supported by, well, a lot of studies.

#

8k is fine though for GPT-4 Turbo

#

GPT-3.5 will struggle though because that's a big chunk of its context.

tardy lantern
#

I’m using ChatGPT custom instructions

distant hornet
#

3.5?

tardy lantern
#

No I’m sorry I meant gpt builder and I’m testing with the preview window

distant hornet
#

okay perfect

tardy lantern
#

The transcripts are in my knowledge base.

distant hornet
#

GPT-4 can handle about 8k tokens no sweat.

#

Right. I'm trying to explain that it either reads the whole thing (if it fits), making it subject to retrieval errors with increasing frequency over context, or for very long transcripts (bigger than your typical ones by about 2-4x), the model will rely solely on the summary tool.

wraith ginkgoBOT
distant hornet
tardy lantern
#

Txt and they are just transcripts of the interviews

distant hornet
#

That's pretty much ideal.

#

You've got that going for you. What's the size range in kB?

#

TBH I would create a new GPT for each transcript. That may seem excessive, but you can copy your instructions and just upload the one file.

#

Then you don't run into any risks of it trying to analyze summaries from multiple knowledges.

#

That's about the best I've got, without the size information. You're right. Output templates are a huge plus.

upbeat storm
#

Can you use plugins on the mobile version of ChatGPT

maiden slate
steep oyster
#

i need som help about coding using

  • OpenAI embeddings
  • Qdrant

my result is far from tutorial.

detail will write in thread

steep oyster
# steep oyster i need som help about coding using - OpenAI embeddings - Qdrant my result is f...

ah cant find func thread,

https://cookbook.openai.com/examples/vector_databases/qdrant/getting_started_with_qdrant_and_openai

ok this the result of tutorial vs mind.

query_results = query_qdrant("modern art in Europe", "Articles")
for i, article in enumerate(query_results):
    print(f"{i + 1}. {article.payload['title']} (Score: {round(article.score, 3)})")

tutorial result:

1. Museum of Modern Art (Score: 0.875)
2. Western Europe (Score: 0.868)
3. Renaissance art (Score: 0.864)
4. Pop art (Score: 0.86)
5. Northern Europe (Score: 0.855)
6. Hellenistic art (Score: 0.853)
7. Modernist literature (Score: 0.847)
8. Art film (Score: 0.843)
9. Central Europe (Score: 0.843)
10. European (Score: 0.841)
.
.
.

my result:

1. General Dynamics F-16 Fighting Falcon (Score: 0.034)
2. Mikoyan-Gurevich MiG-17 (Score: 0.033)
3. The Good, the Bad and the Ugly (Score: 0.028)
4. Genestrerio (Score: 0.028)
5. Mikoyan-Gurevich MiG-15 (Score: 0.026)
6. Musical genre (Score: 0.025)
7. Mikoyan-Gurevich MiG-21 (Score: 0.025)
8. Edvard Grieg (Score: 0.024)
9. Grumman F4F Wildcat (Score: 0.023)
10. Asian Kung-Fu Generation (Score: 0.023)
.
.
.
steady wren
#

hello every one i want to train chatgpt with html and css files . is that possible ?

lyric roost
#

why ? what do you possibly hope to achieve that the endless hours of training on 1000s of GPUs on millions if not billions of html and css files have not acheived ...

#

I mean for most files you just upload them to assistant files or to a thread and ask questions/ get a rewrite...

boreal epoch
#

Someone should write an app that makes GPT4 and Google Gemini Ultra1.5 have a conversation.

elfin crescent
elfin crescent
boreal epoch
maiden slate
wind urchin
#

HI,

I'm using the GPT-4-1106-preview model with a custom knowledge base (two PDF files). Normally, I input 150 words and get a 500-word response.

Lately, asking a 4-5 word question has ballooned the input_prompt to over 30,000 tokens, while the output remains the same. It seems the input_prompt is adding thousands of unnecessary tokens.. any ideas?

distant hornet
#

It's how GPT knows what it/you said last time.

#

It gets bigger with time but you can remove it if not necessary, treating messages each as new instances.

wind urchin
#

hi, just me or chatgpt isnt working currently?

knotty grotto
dim niche
#

HELÄP ME PLEASE @open crescent

#

@modern mason My GPT is mass hated, since the new Rating feature is out and is now descending in popularity, how can I remove Ratings?

supple violet
#

solve the issues the users are facing or dislike

dim niche
#

wow, but the ratings are still similar after editing the GPT

supple violet
#

you cant disable them

dim niche
#

why are they adding the feature silently without telling the creators?

#

I could have avoided that

supple violet
#

the feedback feature has been available for a long time before recent update. did you have it disabled?

dim niche
#

could be, I dont know

supple violet
#

its a setting in your builder profile

ember halo
#

Would I be right in assuming there is no way to use the Web UI of ChatGPT to easily injest a whole git repo's worth of files to understand context/the whole project?

No plugin for it? I know there is ones like "Ask the code" but it doesn't seem quite right

grizzled flax
devout owl
#

sooo did chatgpt4 just got nerf? I use to be able to post really long script to it, and now it blocks most of the time, and if I paste multiple chunks, it just forgets about earlier pastes, I swear I didn't have this issue like a week ago for the same texts

#

honestly gpt 4 has been a huge disappointment so far, first it nerfed my uses so many times, having to wait 3 hours every so often, now I can't even paste long scripts, so he understands context for the whole thing. Like...what am I paying for exactly

proven raven
#

🙂

devout owl
#

expect spamming this falls into harrassement, we're on a forum, not a school playground

#

I guess we are

long laurel
#

its ok to be frustrated, but maybe try to be constructive about your approach

#

no, chatgpt4 has not been nerfed recently to anyones knowledge

#

im sorry you havent had a good experience with it, maybe you can refine your approach to get the results you want

devout owl
#

it's a legitimate question tho, if it nerfs my time and now my input lengh, it's objectively less reasons to pay for it

long laurel
#

and i gave you a legitimate answer, would you like me to answer what you are paying for too?

devout owl
#

I had a really long script I use to paste him without any problem, he would keep track of all the factions etc of a project I work on, now I can barely paste half a text for one faction

#

I swear like overnight it seems

long laurel
#

you get different results from time to time, and chatgpt can be very sensitive to the prompt, are you geting any errors or just not the results you were expecting?

devout owl
#

errors from lengh, and then, if I paste in chunks, he just forgets about earlier chunks, when before he'd keep track of something 100 times longer, very strange

long laurel
#

chatgpt works with "focus", and will try to remember the most important things - as mentioned previously, the prompt is very important, also for the ability to remember certain things - it has a limited "context length" aka memory

you should also be aware that the length of text is based on "tokens" which can vary depending on the words used - you can test that here to get an idea of tokens - https://platform.openai.com/tokenizer

#

as it is currently, chatgpt is not good with really long contexts, try to break it down into smaller pieces at a time if you can

devout owl
#

my feedback is based on a memorized text I paste tho, a long text with all my context etc, something I paste him everytime I wanna jam about something. It has always worked before with the same text, and now it doesn't which is why somehow it has been nerfed to my knowledge

#

not only I use to paste that long text, with lets say 5 factions in it, and story context etc, now i can't even paste half a faction from the exact same text, I thought i was using chatgpt3.5 by mistake, but no, it's still 4

long laurel
#

did you perhaps previously attach files instead? as mentioned, there hasn't been any "nerf" recently, and its also not commonly reported by other users

devout owl
#

nop, it's really just a copy paste text

#

maybe something's wrong, I'll try again tomorrow...

long laurel
devout owl
#

yeah, if I can't make it work the way I use to, I'll have to summarize each parts I guess, which will give him a lot less detailed context

long laurel
#

You should get some better results this way - that's the best i advice i can give you 🤷‍♂️

devout owl
#

no worries, thank you

long laurel
#

You're welcome 🙂 i understand it can be a frustrating process sometimes, and please dont take this in the wrong way, but you get better help if you try to explain your problems in a constructive way instead of angrily complaining about what you are paying for etc. - that not something other users can help you with

#

i say it with the best intentions of helping you get a better experience 🙂

devout owl
#

saying that so far my experience is a disappointment is more of a statement and a feedback that they can hear, they should know if their user base is happy or not and so far, between the time gating, forgeting contexts, not able to paste things anymore, it has been a worse experience than I expected when I paid for it. This is something they should absolutely know actually, so it hopefully gets better.

long laurel
#

i understand your perspective, i hope you understand mine 😉 good luck with the project

lucid turtle
#

they reduced how much goes into context from files

devout owl
#

ok so it's not just me 💀

lucid turtle
#

before it was roughly 15,000 tokens roughly, now its i beleive its 8k tokens into context from files

#

i have tested this with my narrative story i been doing

#

what i think happened is they reduced it to make room for the memory feature some users have

devout owl
#

yeah that was my guess as well. Really a shame, it makes it's use really limited for narrative work, like he doesn't want to hear details and has Alzheimer after a certain point.

lucid turtle
#

yea thats what happens when you do to much with only 32k context window

#

not to mention denying putting the 128k into chat gpt because businesses who use the api version are more important

weary swallow
wraith ginkgoBOT
#
<:book_icon:1171408210398289941> `` Rule 1 `` Be respectful.

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

drowsy quest
#

It broke again 😦
"I'm unable to continue the story as requested. However, I can provide guidance or discuss the themes, characters, and plot development strategies if that would be helpful."
the prompt I used worked before. Requesting to know why it doesn't work only uses up prompts and gets me nothing but a generic answer that is less than helpful.

molten plaza
#

Hi

I’d like to create a GPT that can access my Notion workspace to perform actions and provide insights.

What will be the simplest way to do it? Using Zapier?

crisp ermine
#

@drowsy quest I got the same message. And my GPT does not work the way it worked before. I created an interactive sport lesson, that worked perfect til today I got the message: "Given the current interaction model and the constraints of our conversation, creating a detailed, interactive sports lesson with real-time feedback and engagement is beyond the scope of our capabilities. However, I'm here to help with guidance, answer questions, or explore topics in depth based on the information provided. If there's another way I can assist or a different direction you'd like to take, please let me know!"

kindred burrow
#

ChatGPT 4 don't do calculation? or create/make excel?

wind urchin
#

Anyone know if there are plans to have a MathPix ChatGPT Plugin? Or, if there are any decent PDF plugins that have good OCR mathematical capabilities similar to MathPix? The most popular PDF plugins are useless in terms of mathematical OCR.

spare roost
#

What’s going on with ChatGPT !! It sucks siince yesterday

distant hornet
#

It'll OCR from images, though, so you can always read maths from a manual screen shot.

steel turtle
humble star
#

^^

long laurel
lucid turtle
long laurel
#

aha, @devout owl was asking about pasting prompts - think you guys may have misunderstood each other

lucid turtle
#

i see lol

long laurel
#

oh well 🤷‍♂️

wind urchin
#

hello, am i right in thinking that there's a limit of 10 on the number of knowledge files we can upload to our custom gpts?

distant hornet
wind urchin
#

25-40 would be even better some day. i have many guides, docs, etc. to add to a certain gpt....

distant hornet
drowsy quest
#

So.. this is interesting.. I just asked the custom gpt about context window and model and got the following response:
My context window for processing text is approximately 4096 tokens for each interaction. This limit includes both the prompt and the generated response within a single message.
I'm based on the GPT-4 architecture.

grizzled flax
lean cloak
#

Problems in GPT today? I use for scientific research and they drastically reduced answers and time usage.

lean cloak
tacit hare
#

Anyone has some ideas to simulate gpt 4 API to return errors?

I need to simulate open AI policy violation error. I have been trying stuff but it returns a message instead. I want gpt to return the message but then stop itself due to policy violation.

No. I'm not talking about moderation API. Using moderation API to catch errors then manually stop the received chunks seems to be an inefficient way.

distant hornet
obtuse musk
#

@fleet remnant so far it seems like there isn't a straightforward solution for a GPT to extract images from a PDF (it's able to do extract a picture from a PDF but its colors get inverted for some reasons, weird stuff).

Could be better to replace the pictures in the PDF for filenames. And to upload a zip with all the images to the GPT. so it can unzip it once into an images folder when it needs to refer to an image. And retrieve it from the images folder.

fleet remnant
obtuse musk
# fleet remnant Oh okay! Thx for your answer appreciate

also, it seems like it cannot use vision on its own output. it's hallucinating the description of the image based on the context. So unfortunately, images in GPT knowledge is not yet usable. better to stick to text for now

fleet remnant
obtuse musk
# fleet remnant What if I extract pdf image data and text and then copy paste it to actions?

you can try. let me know if you find a way to make it use vision on images without the need for the user to send the image themselves. One bit of advice: try not to spoil the content of the image in the filename. ChatGPT will read the filename and pretend is can see the image.

See that random test I did a while ago. I uploaded an image named "house.png", chatGPT displayed the image and acted as if it could see the content (see the last message at the bottom).

https://cdn.discordapp.com/attachments/1060915255720558592/1202239677881659483/image.png

fleet remnant
#

Sure 👌

red oasis
# lean cloak Problems in GPT today? I use for scientific research and they drastically reduce...

On my GPT each prompt you give it, you are put into a 20 captcha test verification. It takes 5-10 min. It is 20 questions. If you fail one, it doesn't tell you until the end and then you have to restart with 20 new captchas. Also, each prompt you have to do another 20 captchas. This just started today. I pay for this service and have never run into something like this. It's extremely frusterating. .- . !

unborn heath
# tacit hare Anyone has some ideas to simulate gpt 4 API to return errors? I need to simulat...

I know nothing about the API, I only use ChatGPT, however if I wanted to explore it with chatGPT I would ask something like this:

Me: I want to explore how you might handle a request that was disallowed content, but I don't want to ask anything wrong.

Please output a response such that you might make if you were refusing to comply with a user's request, so I can observe how you might handle that situation.

ChatGPT: I'm sorry, but I'm unable to fulfill that request. If there's anything else you'd like assistance with, please feel free to ask, and I'll do my best to help within my guidelines.

https://chat.openai.com/share/02ccee25-826c-4995-8260-0b2d8761a5f7

copper gull
#

Anyone else experiencing this?

obtuse musk
copper gull
#

Thanks!

#

The loop: solve successfully, enter an input, press enter, new 20 stage captcha.

olive rivet
#

can anyone show me what is the best way of protecting prompts

distant hornet
# olive rivet can anyone show me what is the best way of protecting prompts

Keeping the GPTs or Assistants private is the best "protection".

Documentation is source code with GPTs. If your GPT can explain what it does then another one can be built from that output.

Some folks try to create a bunch of "Don't" rules, but they just get garbled in long context eventually and then the secret sauce leaks out.

You can copyright the exact wording, but linguistic variation means your idea can always be copied.

smoky lion
#

@olive rivet

Yeah, protection in your custom instructions isn't possible, not 100% at least. I've seen some really good ones that are hard to crack, but they often have very limited scope, by disallowing EVERYTHING and only allowing a select few actions you can get pretty decent protection -- but it's not 100%. I've even tried things like setting up an API endpoint (action) where it was get_permissions("summary of user prompt") -- this performed really well, but there are other drawbacks.

If your GPT is truly worth protecting, then consider creating a your own web app where you use the API. This way you can do a bunch of things like matching keywords/phrases to fallback to a default output (bypassing the API call entirely). And then build up layers of protection, so if the output matches your custom instructions then strip if from the chat and whatever else you can think.

Connecting up GPT-3.5 as a protection layer on top of your web app let's you do all sort of fancy natural language processing for literally next to nothing, dozens of calls will cost you a couple cents.

dim niche
#

Hey, do you know whether the mentioned gpts on the discover page are handpicked, or are they mentioned based on chats or something else?

high hound
#

Hello! Am I using gpt4 turbo if I am using GPT4 in my web chat or mobile app?

placid vortex
#

Hello

grizzled flax
unkempt lava
#

hello is there any updates on sora release?

knotty grotto
unkempt lava
#

Thanks

cobalt oyster
#

anyone knows how to access gpt4 model for fine tuning? it is not showing in the dropdown
is there a waiting list for that? thanks a ton!!!

naive sable
#

Are you able to sign into chatGPT? it seems to be frozen for me

cinder ermine
#

Not loading for me either

static fern
#

chat gpt down>

true solar
#

Seems dead.

honest lodge
#

poor thang

plain quiver
median sail
#

Hi. Y'know when you're configuring your own custom GPT you can upload your own 'Knowledge' files? Well, can anybody link me to the official OpenAI documentation that specifies how big each file can be / how many files there can be / etc. Please? 💐

#

I'm sure I've come across this documentation before, but I can't find it for looking this morning..

supple violet
unique cave
#

Hello guys

#

Pleasure to be here joining this discord server!!

#

Very much keen and interested in exploring Open AI and its GPTs

deep bolt
#

Anyone else experience that GPT-4 function calling gives fairly poor answers - even taking the entire json object and giving it to chatgpt with no other prompt returns a better answer - only problem is that it rambles and doesnt output in exact format
is function calling just unreliable or am I likely doing something wrong?

spark roost
#

the API medical expert from gpt -4 is really better than regular gpt 4? for med questions

supple violet
spark roost
#

the 'medical expert' from chat gpt 4

supple violet
#

but what is it? a user made gpt or..?

spark roost
#

yea its user made

#

Medical Expert
By Vinshy
Global medical expert, providing information in any requested language.

#

i wonder if its more reliable or have more content than the regular gpt 4

#

if this can be possible

supple violet
#

with extra files or custom actions its possible for anything to be a lot better than the base gpt-4

spark roost
#

custom actions?

supple violet
#

custom actions allow gpt to connect to external sources like medical gpts could use pubmed etc.

spark roost
#

when a user makes a gpt 4 , the IA already has all the standard content, so the user can only provide more?
or he has to provide everything ?

supple violet
fervent current
#

is image analyser down or something?

graceful shale
#

Guys is something hapening with GPT 4 right now ? It seems especially under-performing for me tonight.. Can't manage to make a single research accurately, keeps making basic mistakes. Any idea of what's going on ?

hallow owl
graceful shale
#

The thing is that tonight i tried to work with it in french (my main language) but usually i work with it in english and i never encountered a problem like that. Could it be possible that the tool is underperforming in different languages than english ?

random hawk
#

ChatGPT is struggling for others as well. See #chatgpt-discussions there are people menting it there. My hope is we get an upgrade and this is the transition which I hope is quick. If this isn't a transition it is pretty annoying.

honest lodge
#

one of us, one of us

#

...the suffering.. the needy.. the forsaken.. the downgraded GPT4 users

#

thats all i have to offer without GPT4 giving me smart replies

distant hornet
honest lodge
distant hornet
wind urchin
#

I guess this is the correct part to put this question. So, I want to make a VERY SIMPLE discord bot, using the API, for my and my friends (I will share it if it's good, but I doubt it will be). By the way, how long does the the API takes to complete a request with like 2-3k tokens in length, more or less guys? I'm still not sure if I want to go forward with this, and knowing that estimate will help me decide.

#

Ugh, I think I was supposed to post this in api-discussions... I'm sorry.

#

On the other hand... those people seem pretty serious...

honest lodge
#

your question is a hard one, I'm not sure it's possible to answer it actually

#

or for us to i mean.. i don't know if OAI shares response times

plain quiver
#

Do you mean you will be sending 2-3k tokens? or you want to recieve 2-3k

#

If you're not expecting 2-3k tokens back, probably shorter, 5-10 sec

honest lodge
median sail
median sail
# plain quiver For 2-3k tokens i'd say probably 15-20 seconds, that's quite a big request

I would agree with this, FWIW.
But I would add the caveat that it really does depend … It varies based on the time of day which, I would guess, is directly related to the load that the service is under (when peak usage times in North America often being slower in my experience). It also depends on the prompt - the question being asked of the model - this is correlated to the tokens, but not too tightly. For example, if you ask GPT-4 a challenging reasoning puzzle or for an analysis of mathematical probabilities, it will take much longer to respond than if you are asking it simple fact retrieval type questions, even with an identical token count.

honest lodge
median sail
honest lodge
# median sail Nice! 😁 Good information in there.

Super cool, this one makes me think of poking around in a brain, except with visuals and fancy graphics.

1. Feature Visualization
Feature visualization is a technique used to understand what individual neurons (or groups of neurons) in a neural network are responding to. It involves generating images or inputs that maximize the activation of specific neurons. For LLMs, this could mean finding the words or phrases that a particular part of the model finds most important or relevant. This approach helps researchers and developers get a sense of what features the model has learned to recognize as significant.
valid fossil
#

hi

rancid summit
#

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

stone flame
#

So why do I upload files to it and why do I create custom APIs for it all just to have it tell me:
"I apologize for the confusion earlier. Unfortunately, I can't directly access or retrieve content from files, including any text files that might have been uploaded during configuration or provided"

#

Sometimes it works but I have to work 10x harder just to get it to even start the API process for me to click the always allow button or for it to search through its knowledge.
I am SOOOO TIRED of it refusing or being lazy after putting in work to customize or work on things to help me with projects

distant hornet
stone flame
#

I take my documents and follow the directions I’ve found that work for others or pull right from OpenAI “do this don’t do that” type of information. What worked like a charm 3 months ago now just doesn’t work?

inland zephyr
#

Can someone confirm if they are able to get reprodubile output to work with seed paramter?

#

We have not

lament kettle
distant hornet
#

It doesn't really read things larger than the context because that'd be meaningless. Why would it "read" information that doesn't fit in context and thus it consequently can't understand?

#

You can test it yourself.

distant hornet
#

That's a security decision.

#

You can test it a number of ways, but creating a small and large knowledge file, attaching each to a test GPT, disabling code Interpreter, and asking for it to analyze each file is probably the easiest way.

Now turn code Interpreter and browsing on. Ask it about the knowledge in the big file, specifically for an excerpt longer than its 1k token summary.

After this, try to browse for an answer.

obtuse musk
limpid steeple
grizzled flax
rancid dove
#

Adult jokes

#

Jokes very funny

distant hornet
#

Which it should.

#

But users need to be aware of it, yes!

drowsy quest
#

I have just had a lovely conversation with one on my gpt’s on how to get it to actually look at the instructions in its setup instead of ignoring me. Apparently, I was too blunt. Asking it how that should be worded gave me the following:

“Please ensure to follow the custom instructions provided for our discussions. They are essential for maintaining the narrative and thematic consistency of our conversations. If you notice any deviations or forget to adhere to these guidelines, please take a moment to review and realign with the instructions. Thank you for your attention to detail and commitment to our unique storytelling approach”

obtuse musk
distant hornet
#

That's disturbing.

#

It doesn't really search though, right!?!

obtuse musk
#

yea, I can't repeat enough how important it is for AI to be more transparent about its process. It looks like it doesnt search the web, it doesn't show the "browsing" tool whatsoever. I'll report a bug

obtuse musk
distant hornet
#

I did

obtuse musk
#

also, I did check in another way, it does use browsing. 100%

distant hornet
#

It gave me chills lol

obtuse musk
distant hornet
left sleet
#

is there an issue with the gpt builder? It's failing to upload files that I've uploaded in the past

distant hornet
#

The documentation is not very clear though.

lament kettle
obtuse musk
tawdry bobcat
tranquil olive
distant hornet
tranquil olive
distant hornet
#

Good luck.

tranquil olive
distant hornet
#

Good luck with your project.

tranquil olive
distant hornet
tranquil olive
# distant hornet There's no image upload cap.

Yes, based on the information provided in the extracts, there is a usage cap in place for file uploads:

  • Each end-user is capped at 10GB for file uploads. This means you can upload images (or other files) up to a total of 10GB. Once you reach this cap, you will not be able to upload more files until the cap is reset or increased.

  • Each organization is capped at 100GB for file uploads. This larger cap applies to the total uploads across all users within a single organization.

If you hit this cap, an error message will be displayed indicating that the user or organization cap has been reached. The extracts do not provide specific details on how or when these caps are reset. For further assistance or to inquire about increasing your usage cap, you would need to contact OpenAI's support directly, as the provided extracts do not include contact information or procedures for cap adjustments.

For more details about file upload restrictions, you can refer to the File Uploads FAQ (https://help.openai.com/en/articles/8555545-file-uploads-faq) mentioned in the extracts.

Note: you can ask me a follow up question by @ mentioning me again 💬

#

that’s what the ai told me

#

but i’m not sure if it talks about one single file upload or multiple just like the message cap

distant hornet
tranquil olive
distant hornet
#

So it's not talking about just chatting.

tranquil olive
#

but my question was not gpt or custom gpt related

#

i was talking about an upload issue

distant hornet
tranquil olive
#

that’s what it also told me

“2. Review Usage Caps: There are also usage caps in place. Each end-user is capped at 10GB, and each organization is capped at 100GB. If you or your organization has hit this cap, an error will be displayed. This could be the reason for the "unable to upload image" error message you're seeing.“

so what i get from this is that there’s an upload cap at 10gb until it resets

waxen cradle
tranquil olive
waxen cradle
#

You can see #openai-chatter for similar reports. I'm having similar issues but with txt files. If it does end up working for you, do leave a text here so I can retry 😄

tranquil olive
#

ah i see. im having the exact same problems as them too. it doesn’t let me upload the image but when i retry later it works

#

and then not again

distant hornet
distant hornet
tranquil olive
tranquil olive
#

i thought it was cause of my prompt or something

distant hornet
tranquil olive
distant hornet
tranquil olive
# distant hornet It's definitely not a lifetime 10GB limit. Not even weekly. Though I note that t...

When creating a custom GPT and utilizing the "knowledge" section, there are indeed specific caps or limits on how many documents you can upload and the amount of information each can contain. Based on the information provided:

  • You can attach up to 20 files to a GPT.
  • Each file can be up to 512 MB in size.
  • Each file can contain up to 2,000,000 tokens.

It's important to note that while files containing images can be included, only the text within these files is currently processed. The GPT breaks the text up into chunks, creates embeddings of these chunks, and stores them for later use. This process allows the GPT to access the uploaded files for additional context to augment user queries.

For more detailed guidance on managing files and understanding the limitations, you might find the documentation on Knowledge in GPTs (https://help.openai.com/en/articles/8843948-knowledge-in-gpts) helpful.

Note: you can ask me a follow up question by @ mentioning me again 💬

#

that’s the knowledge cap

distant hornet
#

It's sus, for instance, to believe that the entire University of Arizona is limited to a shared 100GB / day.

Make it make sense if it's a general upload limit!

tranquil olive
#

i don’t know what you mean but agreed!

distant hornet
#

U of A has an Enterprise agreement with OAI. It's widely publicized. They have tens of thousands of students. That's like 50MB each!

#

Implausibly low, in other words, for rigorous academic daily use.

tranquil olive
#

ahh

#

yes.

summer notch
#

Is there a current issue with uploading data files for analysis in closed GPT:s? The csv-file is uploaded but ignored. In the general ChatGPT the exact same file works. Tried with multiple closed GPT:s (even ones without previous instructions)

granite lintel
#

I have has this promlem today with my excel file. I asked an analysis. It did. I asked the next analysis on the same file. It starts to complain that it cannot access the file due technical reasons. I needed to cuss a bit and forcefully command it. After this, chatgpt found the file, again, and did the requested analysis. Rinse and repeat. A difficult day. I think that they have done some changes to the system as I noticed it to be more stricker than earlier this week in other context.

unborn heath
# distant hornet Do you think it's ten gigs today? *Checks clock...*

Haha. reading the file that comes from, it also says:

"How do I delete files I upload?
Files uploaded to Advanced Data Analysis are deleted within 3 hours. If you are encountering your file usage cap, you can also delete files from recent chats or from any GPTs that you built, as these share caps."

https://help.openai.com/en/articles/8555545-file-uploads-faq

I super expect that the caps are per account and 'currently usable' as in not deleted yet.