#chatgpt-discussions

1 messages · Page 74 of 1

ivory iris
#

I haven't played around with that myself

full folio
#

Do any of you have or know of a gpt I can use that writes awesome scripts? Whether it be for articles, videos etc

golden sleet
#

I can't post pictures in here... But erm yeh so I ask it

System : "You will be replying to Moon who is a friend"
User : "What is your favourite colour?"
Reply : "A fine question from a noble friend! One does quite fancy the regal shade of purple. It is the colour of kings and queens, and it suits my luxurious fur quite well. What about thou, Moon? Dost thou have a preference for any particular hue?"

So am I going to have problems in the future with this? Because I thought it was learning to associate purple as it's favourite colour... But I don't want to over train it so that it ends up giving more static replies over time

ivory iris
#

based on what I read, I would suggest prompt engineering like I was talking about before fine tuning, but it's up to you

#

so you could build the guidance:

1. Your favorite color is purple.
2. You are a friend.
3. You are a prince.
4. I would like your replies in a royal, old english tone.```
You could also give it a name.
golden sleet
#

So that just goes into system? What is the point of Fine-Tuning then?

ivory iris
#

probably information that's too much for a single prompt and may exceed the token limit

golden sleet
#

Well that's a shame, I was really hoping to have the Model learn things about itself and be consistent

ivory iris
#

that's a more advanced problem with no easy solution. I'm sort of working on a memory system for LLMs like ChatGPT and Claude.

golden sleet
#

I'm new to all this stuff and AI, I used to just do C# stuff in Unity but am branching out into a bot for Twitch and thought adding ChatGPT to it would be cool.

I thought the whole way AI worked as a neural net that made connections between things, so I thought if the Fine-Tuning consistently connected things like "Favourite colour" and "Purple" and it refered to itself as prince that it would make the connection between itself being a prince and liking purple

#

But maybe that's too simple of a view

ivory iris
#

I mean that might be a way you could go about it, but I think for your case some prompt engineering would suffice. You could try the guidance I shared in a normal chatgpt session with the 3.5 model to see what you think of it. There's plenty more room to add stuff and you can ask chatgpt for suggestions.

golden sleet
#

I'm giving it a go now, I really liked how my custom model was sounding 😅 I suppose worst case I can use the fine-tuned one and just add a couple things to system that are important to help maintain integrity

#

For instance, I don't need the old English tone because it actually has a really fun mix right now where some things it replies in a bit of a more excentric shakespear tone and sometimes just classic victorian but it pulls it off well. So yeh I'll play with the guidance for more complex bits of memory and see how that goes.

Thanks for the help

ivory iris
#

Welcome! 😁

urban gazelle
#

guys, is it just me or chatgpt feels dumber than before??

sick pecan
#

is it just me or are the chatgpt complaints getting lazier?

ivory iris
#

i'm not experiencing this.

urban gazelle
#

Just tried to create a new thread and it seems that the longer the thread the more mistakes it makes

#

But that didnt happen before

#

Oh wow, thanks for the info. This makes sense.

unique saddle
#

Hmm...something seems to have gone wrong.
Anyone with this problem?

peak snow
unique saddle
#

I tried clearing cache, cookies and storage, as well as 3 browsers (chrome, brave and firefox).

orchid pollen
#

gpt down?

#

i cant see my chats etc not even my profil pic

#

the playground still works..

#

ah websit eis back

fallen dirge
#

Still down for me

river sinew
#

same here

#

only GPT-4 is affected

#

"Hmm...something seems to have gone wrong."

proud matrix
#

jesus christ can you guys test in production early in the morning instead

river sinew
#

its around the world, its always morning somewhere ^^

peak snow
river sinew
#

cable

#

no vpn

proven rampart
#

I also have no access to GPT-4 right now - constantly giving the "Something seems to have gone wrong" message. Cleared cache, cookies, history, tried different browsers and even different networks, no luck

outer lake
#

is ChatGPT working horribly for everyone??? its so bad!!!!

peak snow
outer lake
#

its happening on different browsers, different OS, different pcs, different networks

onyx eagle
#

It's working fine for me. Maybe a temporary issue for you...

river sinew
#

and I bet every time I try and get "Hmm...something seems to have gone wrong." message it will also count as a GPT-4 answer, substracting from my daily message count.
that is outrageous in itself

onyx eagle
river sinew
#

"stopped creating" image, 10% of the time

#

"you generate too fast, wait for X minutes"

#

"you already generated this image" (I mean for WHAT exactly is the regenerate button if GPT answers with that?)
it is almost like OpenAI is programming the chat-bot to waste messages and GPT-4 allowance where it can

#

and portrait aspect / tall aspect is not fixed after months... (90% of the time its just a flipped wide angle)

onyx eagle
#

That's funny

river sinew
#

also 10% of the time you get weird ultra ugly images

onyx eagle
#

Never had this message before

onyx eagle
river sinew
#

from the beginning

#

of Dall-E 3 in GPT 4

#

they sometimes look like children paintings or unfinished renders or sth. completly off the style

onyx eagle
river sinew
#

also, the images downloaded are webp.... worst data format ever invented. PNG was much better

#

aprt from the fact that the file names are so long windows has trouble working with those....

#

and I have not even started about how racist and sexist the image generation is lol

#

also who had the idea of banning "kissing"??? it works in bing chat but not Dall-E 3 in GPT-4
I could make a list of outrageous stuff that is a mile long

onyx eagle
#

That's interesting. Never had issues like that before.

river sinew
#

I m generating the hell out of Dall-E 3 🙂

#

I mean I had to interrogate GPT literally until it finally admitted that kissing was forbidden in images through content guidelines. as if that fact is secret...

onyx eagle
onyx eagle
river sinew
#

its not controversial, its outright bad

onyx eagle
#

It's very relative. What one thinks is fine and normal may not be very fine or normal for some other people. ChatGPT tries to satisfy most 😆

river sinew
#

nah, its not relative. its absurd conservatist censorship

#

apart from women beeing forbidden to swim, ride, crouch, crawl etc.... everything that might be slightly ... whatever is banned.
bikinis? do not exist in chat gpt. even a bed is dangerous territory.

it all works 100% fine ofc with men....

neat storm
#

i think the main reason for all of this "safety" and whatnot is to avoid headlines

#

istg bro anytime an AI generator goes out of wack media picks it up and talks so much about it like openai committed a sin and should burn in hell

river sinew
#

they already burn in hell because of "industrial sized art theft" headlines. I mean I really dont care for that, but AI makes headlines nontheless

#

I just hope they get more headlines for ridiculous censorship practices soon enough

onyx eagle
#

What seems normal to us might not be normal everywhere because we don't know how everyone around the world lives and behaves.

river sinew
#

apart from having the worst user interface and user experience. only windows itself is worse

neat storm
river sinew
neat storm
#

also they def need to work on upkeep

#

lotta issues lately

#

i like openai but come on lads

river sinew
#

midjourney is not as good in image generation, but even in discord the usability was 100 times better...

neat storm
#

im not entirely sure how a discord chat is better ui than the openAI web app/mobile app but eh

river sinew
#

well, because it was

#

the GPT interface is abysmal

neat storm
#

you do you 🤔

onyx eagle
neat storm
#

i would like to send images but eh

river sinew
neat storm
#

i think the UI is quite alright

onyx eagle
river sinew
onyx eagle
neat storm
#

For me personally, I haven't had issues
But complaints about restrictions have existed for ages so yeah

onyx eagle
#

Absolutely...Complaints are just free advice in disguise 😆

peak snow
sick pecan
#

complaints can be more projecting emotion, so to validate them. ie making someone elses day worse so you can feel better

wise fox
#

Is something wrong with ChatGPT?

#

Cannot get response properly

#

“Somethings seems have gone wrong” has been popped up

river sinew
#

yeah, me too. you are not the only one

junior lion
#

Is there a website that can generate good chatgpt (or other ai) prompts/instructions, anything I want?

river sinew
junior lion
#

Wdym?

proper zinc
#

Hey what do you guys think of chatGPT with the new update?

wise fox
hollow mica
#

having issues logging in...known issues atm?

junior lion
#

@river sinew

carmine matrix
junior lion
carmine matrix
#

how am I supposed to plagiarise github code when chatgpt is down 😦

tidal light
#

oh god NOT AGAIN the damn broken numbered and unordered list because when they updated ChatGPT they messed up with the numbered and unordered lists by placing the numbers as part of the message, making it asthetically unpleasant and slightly uncomfortable to read

#

and in the numbered and unordered list there is a gap between the border and the list

south kayak
#

When is the japanese language getting fixed on gpt4 and 3.5?

#

It makes a lot of mistakes at simple translations

floral breach
#

anyone got problem using gpt4? I get this error since yesterday "Hmm...something seems to have gone wrong."

proven rampart
#

Yes, I have had the same error since then as well. I just found out if I upload any image to a new chat, then I can interact with gpt4 as normal (but it's running extremely slow)

misty minnow
#

Same here, I joined this discord just to report this lol

proven rampart
#

Same, extremely annoying. Especially when they claim there is no issue (because their status page says all is okay - so they closed my ticket)

misty minnow
#

Wow that's some terrible service. I hope they review it again properly, it's quite a major bug that probably came with the Dynamic GPT update. Thanks for sharing the trick though, uploading an image makes ChatGPT work again.

stray turret
#

Hello

floral breach
celest quartz
#

I have a question, does gpt4 now include: 1) Dalle3 2) Dalle3 inpainting 3) Vision or whatever the image recognition is called?

#

I mean, I should specify: does it include all of it, in Italy?

maiden fractal
jaunty robin
#

yeah, the safety team have gone a bit off the rails. it wouldn't let me make a derivative work of a 200 year old painting the other day because of copyright laws

hollow ridge
#

i need help with fine tuning a model where can i find help for that?

celest quartz
celest quartz
maiden fractal
celest quartz
#

Yeah, I was not worried about it but about wether it's up to date

#

some countries don't get access to all features immediately

maiden fractal
#

The only feature is still rolling out it is the memory thing

jaunty robin
paper stratus
#

Anyone else has ChatGBT 4 error "Hm... something seems to have gone wrong" to any prompt independent of OS, browser or network connection? GBT 3.5 works just fine

lilac briar
#

could anyone tell me if its better to use GPT4 orginal than a GPT from the store?

I feel that the GPT from the stores are a bit dumb and lose context of the project.

Just wondering if the GPT4 is 'smarter' or if its better to use a GPT which is more specialized to an area of interest, in my case coding

zinc rapids
#

Hello team, I have an issue with GPT4, I am not getting response it says 'There was an error generating a response', Does anyone have this issue?
The GPT3.5 is working for me

river sinew
#

I m here als having this for 12 hours strait now

eternal rockBOT
#

Beware of possible scams or fraudulent activities that you may receive through direct messages. OpenAI staff will never DM you for any transactions.

Please report any incident by sending a DM to @languid valley immediately.

oak marsh
#

Hopefully the team fixes it asap

stuck citrus
#

I have the same problem. No answer from support yet

winged bluff
#

They fixed it yesterday but now it seems the problem is back again pausing or not generating responses.

upbeat pine
celest quartz
#

can someone give me an example of dalle3's inpainting? not sure if it's in the rule for me to offer an image to edit, so if it's not, I'll just have someone show me how they've been using it

fickle ridge
stuck citrus
#

I'm not happy about this at all, I needed ChatGPT a lot today and nothing. If only I'd gotten a response from support. I feel pretty disappointed when I pay so much money for their service and don't even get a reply.

frank jacinth
#

is a social experiment to see how dependent we have become. blankface

cobalt idol
#

guessing there is no longer any way to see what data exactly some custom GPT returned when it used some API? (it was when it was plugins)
why are we only getting less and less information with everything in chatgpt xD

eternal rockBOT
cobalt idol
#

like I cant see whether voxscript was able to get each chunk of some video transcript
sometimes it would fail on some chunk while it was plugin
now I cant even see that

just shows what it sent (just a link to video in this case)
but not what it returned

#

anymore
dont see when its browsing
dont see what browsing returned
dont see what custom gpts are returning when they use functions
dont see what model its using exactly (https://discord.com/channels/974519864045756446/1220395794666881054) [it never did though]
also making chatGPT users think they are using GPT4 when its 4 turbo
not even seeing proper changes in changelog
and probably more I cant think of right now

EDIT: you no longer see what usage limits you have
I get it, its dynamic
but that does not mean you cant dynamically show it
just starving users of information, more and more
when we needed more information and not less

its a chatGPT the black box of confusion and misinformation

soon dynamic models and maybe will eventually not be even able to choose between gpt 4 (turbo) or 3.5 I am guessing xD (or maybe it will not tell you what its using)
(they want to add something that it will automatically choose 3.5 or 4 depending on what you are asking)
where is this heading xD

fallen dirge
toxic lake
#

any way to force gpt4 to use its browsing capabilities? its not listening to instructions telling it to research using its browsing capabilities and instead just lists me sources that dont exist or lead to a dead link

glass fulcrum
cobalt idol
#

yeah you have to guess whether it did browse or not
when it has some links returned it probably used it
but like he said, it have him dead links, so 🤷‍♂️

again, lack of information which we had before

#

maybe openai trying to make discord more popular again by having more people ask questions because they removed all info in chatGPT xD (joke)

river sinew
#

does Open AI communicate ANYwhere at all?

#

I dont even know if this server is official

maiden fractal
#

Openai employee name are in yellow

cobalt idol
# river sinew does Open AI communicate ANYwhere at all?

its official but with unofficial moderators
openai staff sometimes very rarely messages here on certain things

they announce inconsistently either on
blog on their website, and sometimes they also post that on discord, but not always
or on discord
or on x on their OpenAI account, or there is also OpenAI Developers account
(those are the ones I know about)

sometimes they only post on one of those, sometimes on 2 of those
its a mess
you cant realiably track updates in one place

#

there are also chatgpt release notes, which are sometimes used....but not always

#

I remember when it was all starting, they had some kind of audit and they were advised to not announce every new update everywhere to be able to better manage deployments/usage of new features so that it does not spike and starts up slowly...or something in that way
my guess is it might have something to do with that maybe, maybe not

cobalt idol
river sinew
cobalt idol
#

but if thats about some issues / account issues or similar then you can forget that
just at the usual support paths like the chatbot on help page (which might create some ticket I guess)

river sinew
#

but its also sad that even microsoft does not manage to give some info about changes to bing chat.
it was reformed this week and is way worse now

#

Dall-E 3 is the best image creating AI but its in the hands of OpenAI and Microsoft who seem not to be interested in it to make it have its full potential

cobalt idol
high shoal
#

whats going on with chatGPT-4 its giving me unrelated responses

visual acorn
#

Isn't it kinda scary when ChatGPT refers to itself in first person ;-;

ivory iris
#

it's not scary to me

ivory iris
#

the only real issue that I think needs to be adjusted is when it's talking about humans in general and it refers to "we" or "us" as if it's a human. I know that it's not the intention that it's claiming to be human, but I think it's misleading for those who don't have a good understanding of LLMs.

past mortar
#

I can't access GPT4 for the past 18+ hours no matter what I do

I can't talk with support either. Is my account shadowbanned from using gpt4 or something?

tardy girder
#

Anyone developed "Best Practise" collection and application usage with OpenAI products? In a corporate application. Both in terms of gathering Best practise and in terms of organizing, and categorizing the "best practise" itself

#

If anyone did it how did you go about doing it? I been debathing about ways to do it including direct upload, But it would be more files than direct upload is anything close to handling

#

Also any way to have the upload limits increased? Im currently using the team subscription

cobalt idol
#

maybe not extensive enough

past mortar
warm musk
#

You could e.g. first let the ai pick the subjects and chapters it needs from each document.

Then you can compile that to a single doc and upload that instead

tardy girder
warm musk
#

Something like the latter is what id try

tardy girder
#

ye thats prob the easiest

warm musk
#

With the batching api its be much cheaper now though !!

tardy girder
#

thats a posibility

tardy girder
warm musk
#

You can also.... Make an api that has those source files.

Then generate an openapi spec from that .

Throw that into a customgpt..

Now you have a dynamic processing ai tool, hosted by oai, business logic via your custom api.

tardy girder
#

problem is still it has to take the data from different formats, categorize and then output in Excel/Word

warm musk
#

Sounds like you need a combo of the 2

tardy girder
#

its also another problem with more data flow being inputted

#

fairly often, so not a one time batch upload

#

more uploading big batch of last 20-30 years of files

#

then a couple a months i guess

warm musk
#

Well your api could run on a winxp system.. the data volume is the problem i think

tardy girder
#

Ye Data volume is main problem

#

i could gather all and past everything into a giant Word file

#

but would be 100s if not 1000s of pages

#

have a feeling that would kill 4.0

warm musk
#

I think we need @odd sinew iirc they always have good ideas about data and the api

tardy girder
#

Could you connect a Custom API in relation to OAI direct data link up against a of site datapark?

#

Have someone custom make a Custom API for this

warm musk
#

Love u Nomi 😘

tardy girder
#

then have OAI run it

#

and then connect it against a datapark in say Scandinavia

#

with say uploads once a week for (Topping up of data)

warm musk
tardy girder
#

from say one specific folder

tardy girder
#

of pages and it still kills 4.0

warm musk
#

Awe

tardy girder
#

also i assume it will be less accurate the more is in each file

#

cuz i need it to utilize all the pages

#

not just focus on say 20 of 1000 pages

#

You have any idea what it would cost to have someone up a custom plugin/API like that?

warm musk
tardy girder
#

and make it "work"

tight island
#

Also depends on if it’s plaintext

tardy girder
warm musk
tardy girder
#

in Word/PDF/Excel/Powerpoint

#

for prob 95% of the files

tight island
#

Plaintext is probably the worst format you can use when you have a lot of pages

tardy girder
#

problem is the first batch of 50 if not 100s of files with original files being different formats from the last 20-30 years

warm musk
tardy girder
tight island
#

I do

tardy girder
#

ballpark maybe

tight island
#

Around 150 usd

tardy girder
#

Thats it?

#

bruh

warm musk
#

Per hour

tight island
#

That’s the API hosting only

tardy girder
#

Well how many hours

tardy girder
tight island
#

150

tardy girder
#

Datapark is in Scandinavia/Norther Europe

warm musk
#

I think its 150 per month he means

tight island
#

Requests per hour is a price factor as well

tardy girder
#

i mean to develop set it up so i can input files

#

how much roughly total cost to get that done

tight island
#

You don’t just input files in an api

warm musk
#

And well devtime costs.. depends on the country but here it ranges from 55 to 150 € per hour

tight island
#

Are we talking Assistant?

#

Or an actual api that you send requests to

tardy girder
warm musk
#

So all angles are interesting

tardy girder
#

Not picky as long as it works

tight island
#

2k what I would do in your case is to fine tune a 3.5 instance

tardy girder
#

i already got OAi 4.0 Teams version

tight island
#

Then utilize a team of API’s fine tuned on your data that collaborates

tardy girder
#

And how much we talking for someone to set that up so i can input the data

#

i have no chance of setting this up by myself thats for sure

tight island
#

To pay someone to do it?

tardy girder
#

Yup

tight island
#

Around a thousand

#

USD

tardy girder
#

Was expecting way more

#

and monthly costs will be a few houndred i assume

#

prob

tight island
#

Worst part is this is real easy to do

tardy girder
#

Probably true

tight island
#

GPT can literally guide you through the entire process

tardy girder
#

probably

tight island
#

Won’t take you a day

tardy girder
#

do i have to code

#

or anything like that

tight island
#

Very little

tardy girder
#

hmmm there is a corporate meeting early next month

#

need to bring it up then

tight island
#

Only around 500 - thousand lines of code

tardy girder
warm musk
#

Yeah its not much

tardy girder
#

i have done exactly 0 lines of code before

warm musk
#

Then its a mt Everest

tight island
#

Neither had I when I started

tardy girder
#

and if this is to work its better to pay to get it done correctly first time

tight island
#

But to be honest 2k

#

If it’s for a company the charge is probably not in the 1000 USD category but in the thousands

warm musk
tardy girder
#

ye its for corporate use

#

basically "proof of concept"

#

then introduction

#

then scale up roolup

tight island
#

Yeah for corporate use it’s a hefty price due to legalities and such

tardy girder
#

how would the aspect of confidential data impact this

#

if it has a impact at all?

tight island
#

Does that include personal information

#

Of any party

warm musk
#

👆

tardy girder
#

Corporate sensitive data/Stock Sensitive Data

#

very little if any personal information

#

prob none if i were to guess

warm musk
#

a_skull then it depends on corporate policy and guarentees

tight island
#

Ohh it’s a stock predictor?

tardy girder
tardy girder
#

Not even close

plucky nacelle
#

I have some issues with GPT giving me server errors and message stream issues

tight island
#

Toss me the vague idea

tardy girder
#

but the information would be stock sensitive if they came out(alot of it would be highly outdated in terms of stock sensitive)

#

but newer data could be sensitive

#

but alot of the data files are 10+ years

warm musk
tardy girder
#

and a decent bit is 20+ years old

plucky nacelle
alpine python
#

same for me

tardy girder
#

But just to make sure

#

API is prob best bet

tight island
warm musk
#

Maybe its a temp throttling issue on chatgpt side

tardy girder
#

API would entail a "site" outside of the OpenAI site

#

right? sort of like a "portal"

warm musk
plucky nacelle
warm musk
#

Ahhh ok

tight island
tardy girder
tight island
#

You can essentially localhost it

#

And display it

tardy girder
#

if it actually works its highly probably it will be rolled out

#

corporate wide

#

so the proof of concept would be utilized fairly quick if it works accurately

tight island
#

If you can make it work on a single pc scale you would be able to fairly quickly scale it up

warm musk
#

So business wise you need to know if the trade data is allowed to be sent to America, given its crossing borders.

tardy girder
#

this solution would work on both Windows and Mac OS right?

tight island
#

Yes

plucky nacelle
tardy girder
tight island
#

Wait….

tardy girder
#

so its already crossed borders

tight island
#

Which country again

tardy girder
#

Scandinavia Region

#

primarely and North Europe

tight island
#

Denmark then?

tardy girder
#

among others

tight island
#

That’s my motherland

tardy girder
#

ah nice

#

Im north of you then

tight island
#

Stay off it kermitsock

tardy girder
#

was in Denmark not too long back actually

#

Kastrup was jævlig....

#

lol

warm musk
tight island
#

Just joking, that aside care to divulge more details since we don’t know ow what your project does

#

Can’t help much without sharing a vague goal of the setup

tardy girder
fair zealot
#

Hi everone,
I'm currently writing my master's thesis about hallucination detection. I am trying to figure out which default sampling parameters that are used in chatGPT.
Would be nice to use the same parameters when detecting hallucinations.
Are the default values for there generation parameters known? If so where can find these values?

warm musk
tardy girder
#

then convert over to Enterprise for scale up?

warm musk
tardy girder
#

or something like that

tight island
#

The RAW api route

tardy girder
warm musk
tardy girder
#

Well i would need to find someone to do it

#

im not spending the next 3 months trying to do 500 lines one at a time....

tight island
#

Issue is you can’t just ask just anyone

tardy girder
#

Waste of my time when someone else can do it probably in under a week

warm musk
tight island
#

I could probably do it in 2 days or so

#

However

warm musk
#

To put it blunt

tardy girder
tardy girder
tight island
#

Even if offered 10 grand I wouldn’t do it mainly due to it being for a company and not personal use

tardy girder
tight island
#

Looking

tardy girder
#

Most if not all of the data would be internal work product

#

so corporation would own it

#

would be nice with posibility to add external data later, especially for data analysis aspects if setup works and expansion

tight island
#

The legal stuff plus if someone made something they want partial ownership usually renting it

#

That’s what I would do

tardy girder
plucky nacelle
#

When OpenAI says nothing is wrong but there is something wrong

tardy girder
#

renting prob would not work due to the data being sensitive

#

and corporate got DataCenter capacity already

#

so could easily expand on that capacity to run from there

tight island
#

Then I recommend checking out Leetcode

#

Or GitHub

tardy girder
#

Would individuals that have already developed a data storage platform

#

be able to dev this?

#

if you were to guess

tight island
#

Yes

plucky nacelle
tight island
#

It’s quite easy to leave a backdoor in a program which is why even if desperate you should be careful

#

Is basically what I was hinting at

tardy girder
#

Well have worked with the data storage platform and datacenter for idk 10-20 years

#

I know a guy that is our contact person

#

can have a chat with him i guess

warm musk
tight island
#

My recommendation is check Leetcode @tardy girder

#

Some of the best coders are there

tardy girder
#

mostly us based?

tight island
#

World based

tardy girder
#

or a decent bit of coders Scandinavia based ?

plucky nacelle
#

JPN and KR Based

tight island
#

That too

tardy girder
#

will have a look

warm musk
tight island
#

Yes check their rep first

#

Look up their names just basic searches

tardy girder
#

worked with them for 10-20 years

#

they run everything data wise

warm musk
#

Exactly why i say do it

tardy girder
#

for the company

#

sure will do that

tight island
#

Do it make them sign a contract

#

That’s what i would’ve done

warm musk
#

They'll have connections to companies that they can outsource things to if they need to

warm musk
tight island
warm musk
#

👀

tight island
#

I just spent like 3.5 grand of glasses

#

GLASSES

warm musk
#

Omg why

tight island
#

it’s so expensive

tardy girder
#

damn

#

thats expensive

#

But ye thanks guys i will have a chat around

#

look at the posibilities

tight island
#

Yeah it’s a multi focal smart lens

#

24 gold carat..

#

With my discord username engraved

#

😂

#

All my gaming days have ruined my eyesight

tight island
warm musk
tight island
#

Aight aight

warm musk
tight island
warm musk
warm musk
tight island
#

Fair 😂 sorry about the eyes tho

#

Glasses do be expensive tho

warm musk
#

Every year its like .10-.25 that it gets worse

warm musk
warm musk
tight island
#

I’m taking myself 3 weeks to Japan to enjoy that lovely country

warm musk
#

Indeed ! Appearantly you can see the coastline for all sides of the island when you stand on top of the volcano

tight island
#

Def I already booked the guide

warm musk
tight island
#

Lovely place however many tourists past behavior has caused no tourists zones

#

😔

#

Well I’m off for now

warm musk
tight island
warm musk
#

Im more on the app and api side of things 😆

tight island
#

Last time I ran the api I got billed 67 usd

#

Mostly due to using a swarm

warm musk
tight island
#

What I want to see personally is a more tuned version where we can have “Advanced Settings”

warm musk
#

In some cases it makes sense to make a customgpt aswell for the employees so you can outsource some of the features.

tight island
#

Control temperature etc

#

I think OAI should launch a side service

warm musk
tight island
#

API hosting

warm musk
#

Cuz it used to be in the api..

tight island
#

I mean on the app

warm musk
#

Ahhh then yes

tight island
#

Control temperature on certain stages of the conversation, maybe multiple GPTs in one

warm musk
tight island
#

Each model would use 1 usage

#

Even assistants are barely customizable

#

The API really have you customizing a lot

#

Temperature, max tokens, etc

warm musk
#

Sad. Theres soo many use cases that require customization

#

Much of which is predictable and should imho be built in.

tight island
#

I think OAI should have they paying users to be able to host their own api as part of their membership subscription

warm musk
#

The batching service e.g. .. awesome.. but it took a year to get it

tight island
#

Just api calls in general are a hassle due to having to pay for 3rd party services then sign up

warm musk
tight island
#

Just cloud hosting is expensive

#

Too expensive in fact

warm musk
#

Yup

tight island
#

Especially when starting smaller projects

eternal rockBOT
#
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.
warm musk
#

I wonder what triggered the bot lol

tight island
# warm musk Yup

You’re actually much better of buying a Minecraft server and getting rid of Minecraft on that server to use it as hosting for your own API 😂

tight island
warm musk
#

Omg 😂

warm musk
tight island
warm musk
warm musk
tight island
#

Yeah the mods are very patient with me

warm musk
#

Lol

tight island
#

Just joking probably because of the previous conversation about data and personal or classified documents and information

warm musk
#

Probs

sage tangle
#

the new ability to swich GPT for a single prompt is really cool on the website!!

plucky nacelle
#

Still GPT4 is borked

sage tangle
#

ehhh still its progress

#

better than nothing

river sinew
nova fable
#

My GPT-4 just responded crazy faster than im used to seeing. Was buggin yesterday but now its really fast

neat storm
mild haven
#

You know, you could probably make the end all be all anti-virus with GPT lol. Have it act as a virtual machine middle man analyzing the code and its intended purposes, the functions and it can actually relay that information to the user and discuss wether they should allow it or not. You'd likely not need anti-virus anymore. It could even be automated to a degree for common known things, with a typical anti-virus, but it could literally prevent anything before it happens by telling the user what the code does lol.

neat storm
#

eh..? maybe?

#

hmm

mild haven
#

Ah man, that could also be miss-used. Didn't finish the thought. You'd know what everyone downloaded in detail if you wanted to 🤔. This may be better for internal systems at corporations with an IT department and not home use 😅.

neat storm
#

yeah probably not good

mild haven
#

Lol, need to finish thought before sharing lol.

elfin birch
#

omw to run arbitrary code

mild haven
#

Ha, I've already developed a potential counter measure! You could potentialy inject prompts into the code with a script it request be run inside the Virtual machine environment. What if you embedded a script that would delete and replace the model used? The user, would be completely unaware unless it takes a long time lol. It would take a long time. Weight, maybe not. A mock model. Significantly less training data and parameters with specific purpose. It would be need to be lightweight. But you could prevent that by keeping the anti-virus virtual environment disconnected from the net and only able to interact with the user through the interface. Keep it contained, and add custom instructions and traditional filters to prevent reading prompts from within the code or acting on them.

I don't, random thought of the day moment. I got board working on my coding project lol. Back to it 😅.

candid heart
#

For some reason, when asking about whether stone-age people didn't get tooth decay, it marked my question as possibly violating content policy?

#

I hope it doesn't!

river sinew
sick pecan
#

i suspect you are paraphrasing, share the chat?

#

it might also help to influence the context bias by suggesting you are researching for historical reasons, and not suggesting certain ppl are "from the stone age"

#

(has nothng to do with racist or sexist)

cursive shell
#

is it me or did a dev push an update without checking if its broken lol

reef lantern
#

I was banned this morning and I have no idea why... and I can't get answers from OpenAI because "you current account isn't banned"...

Yea the one I had to create to use the help service... But the one that was banned without explanation can't access the help service

#

Anybody know how to get some help? I've paid for GPT plus for about 6 months, and have spent a good chunk of money for API access too... and suddenly I'm thrown aside without reason

frank jacinth
#

gpt4 still broken ?

autumn valve
#

It is for me, stopped working some time two days ago for me, round 6PM

#

Weird that there's no mention of it on the status page at all what's up with that

cloud pewter
#

did gpt4 got upgraded to December 2023?

astral leaf
#

I stil can't use it and seems still broken.

summer fern
astral leaf
#

Did that upgrade break GTP-4 you think/know?

oak marsh
#

for anyone that wants to bypass the current ChatGPT 4 error saying "Hmm...something seems to have gone wrong."

#

start a fresh conversation, upload an image and say "ignore the image"... then continue using ChatGPT4

earnest kestrel
#

Holy mother of jesus that worked lmao

maiden fractal
river sinew
#

but it works, thx alot!

oak marsh
gentle bronze
#

Anyone has the feeling GPT4 is overly chatty even tho custom instructions ask for a short answer?

river sinew
gentle bronze
#

oh the answer per se, if it's useful won't count towards the cap

#

I'm just frustrated with the embellishment of answers

river sinew
gentle bronze
#

lol

river sinew
#

since then I try to define what the answer should look like

tight island
#

… Some VPNs can get you banned if you’re on the same address as another malicious user, regardless of you doing something or not you share the same punishment just by being on that particular address.

eternal spoke
gentle bronze
reef lantern
gentle bronze
#

this has been thematized many times for different users, the official information is always the same

frank jacinth
#

OpenIA banned my account for no reason, I only used GPT4 for coding purposes, when I created a new account and paid plus again GPT4 worked correctly again, is it a failure in the banning system or why are they doing these random bans?

tight island
#

VPN in worst case scenario can get you banned, also making g a new account to circumvent bans is against TOS

#

How you get unbanned isn’t really up for discussion or debate, there’s a procedure you ought to follow.

glass abyss
neat storm
#

yeah, according to benchmarks GPT-4 is a little worse at coding

#

sad times

glass abyss
#

yeah is has gotten so bad its crazy. It used to be excellent at it

cobalt idol
#

nobody uses gpt 4 anymore 😄

#

unless you mean the 4 turbos (in chatGPT since last year november, 3 versions and now on latest)

lucid sun
#

So no solutions on chat gpt 4 yet?

#

Or should I cancel y subscription and move to claude 3?

cobalt idol
#

or just use api/playground while its having issues
(but works for me)

cobalt idol
oak marsh
torn hare
#

You need to share this everywhere

proven mango
#

WHAAT thar worked xd

frank jacinth
past mortar
#

I can't have a talk with the support either, it just stopped working and it says "hmm something might have gone wrong"

autumn valve
#

Yeah I wound up cancelling my subscription and trying other services instead.

mild haven
#

Today's fun project. I wanted to help my son get more interested in math, so I used GPT to help me come up with a way to teach him math with Legos since he loves Legos. So, we're counting stubs on bricks and stacking them to do 3D math instead of using paper. When I say divide, he can split the stack into that many parts and count the stubs, when I say add, he can stack the bricks and count the stubs, then multiply eventually. We've essentially made a no tech calculator or advanced abacus with Legos lol. Infinite scaling lol. Less addition and subtraction and more multiplication as you stack though 🤔.

dreamy fiber
mild haven
#

He's helping write he's educational songs in Udio and Suno, so it's music he likes with knowledge I want him to have. Figure combining music with his hand writing assignments where the music is about his hand writing assignment, I can use that for conceptualization teaching and then the Legos for visual and tactile learning.

#

This way he has fun through and learns through multiple means simultaneously in a way that's less abrasive, hopefully lol.

#

It should work right? The logic seems sound.

mild haven
# neat storm yeah, according to benchmarks GPT-4 is a little worse at coding

For coding I've had a lot of luck by creating a coding specific custom gpt, loading documents from the language I plan on using and specific documents from the environment I'm working in along side documentation and tutorials on how to work in the specific environment usually with examples they've provided or code I've already written and know works inside the environment.

This has produced the best results so far with less hassle on my end.

#

I have one for Deforum math even now lol. It's job is to help me come up with equations for the settings for the shape and motion/effects I'm trying to achieve. All it has is deforum docs and the different systems I use it on, it's got examples and all kinds of documents and past equations and etc that I liked attached because I update it often.

#

I've debated on a browser extention for GPT so it can access my Obsidian notes database during a session like my local LLMs like Llama and some others I have do, this would be the best I think. But for now, the easiest way to replicate that is by loading up the custom documents and custom instructions for the custom GPTs and it works fairly well. Less plug and play, but higher chance of usable output.

river sinew
#

how long will GPT-4 be unable to me?
I pay money for GPT plus and what I get is not ever worth what I get from a free site with ads!

past mortar
dense halo
oak marsh
rare grove
#

is gpt down

neat storm
#

In every benchmark?

cobalt idol
#

I guess so
dont know whether in every benchmark
see LMSYS chatbot arena for example
but mostly better than gpt4
from experience, better in every way than gpt4
just it has some quirks, like always trying to compress/omit/skip parts/put placeholders into code blocks by default (but can be prompted out)

tight island
#

proper prompting also goes a long way

#

u cant really expect it to do a proper job when saying "do story" then get mad about it being a story about a dragon with a hat. when u added insufficient details.
sure some models do take user's lack of info into account by filling in the blanks, but thats what a vague answer gets you, i dont have that issue with the codeblocks or code in general when using gpt. because i actually put work into my prompts.

oak marsh
#

The more context, the better

wintry spindle
#

why is my chatgpt saying it cant browse?

#

I don't have real-time browsing capabilities to check current market trends

neat storm
#

oftentimes I get stumped on how to properly describe my thoughts and the AI can't read your mind

#

if you can improve your speech/articulation you will have a much better experience with these LLMs

grizzled sorrel
rustic kelp
#

does anyone know, if I'm on a single personal ChatGPT Plus account ,and I upgrade to the Team account, will it migrate over my existing conversations?

dapper cloak
rustic kelp
dapper cloak
rustic kelp
honest jay
rustic kelp
#

thanks guys! I see there's help docs now 🙂 there werent' when this first came out

#

I'll read up. I'm curious to know how sharing GPTs and/or conversations work.. or if everyone always has their own private conversation list

#

one more Q @honest jay @dapper cloak if I'm admin and I open a Teams account, then invite someone else.. do they get to migrate their GPTs/chats over too, and then cancel their own ChatGPT Plus subscription?

honest jay
rustic kelp
honest jay
rustic kelp
#

ah crap. I seem to have lost the memory feature now

#

actually that's fine, it was kind of annoying anyway 😄

#

hmm, I was under the impression that I could share a chat with the team, and they could contribute to the conversation. I guess that's not doable

candid heart
sick pecan
#

on the internet most ppl using that term arent doing so for academic reasons

#

and i suggest its more dismissing someones intelligence than their race

#

the path to agi is not paved with disrespect

honest jay
eternal rockBOT
neat storm
dense halo
#

can someone show me the logs

ocean fox
marble surge
#

dang is chatgpt going crazy fast for anyone today? i'm not sure i've ever seen it go this fast

sick pecan
#

tuuuurrrbboo

sage tangle
junior lion
#

Is there a WEBSITE (not stable diffusion) that can make prompts for ai or something? (So I can make chatgpt make a Mii from Nintendo)

molten sail
rustic kelp
solid spear
#

can chatgpt 4 or an chatgpt 4 plugins can dev a Website for e commerce

sour wasp
#

Can we go back to having longer responses? Or maybe an output length toggle for ChatGPT.
The responses I feel are really short now and it's annoying I have to specify that I want a detailed response each time. I just have a custom instruction on because the output are short by default.

feral zinc
sour wasp
#

Yes

feral zinc
slow copper
#

Is there still a 50 message limit for every three hours for gpt4 ?

feral zinc
#

its been 40/3h for over a half a year and been recently even lower/higher at times due dynamic limiting

slow copper
#

Ah so its dymanic limiting gotya

#

Thank you.

latent pecan
#

guys, am i only one triggered by the fact i expresly say to chatgpt to generate everything to one code snipet which can be easily copied and even tho i expresly say it several time, it will always generate the output into several codesnipets?

tight island
#

…Alternatively create a custom GPT to always show the entire code snippet

#

…Do also keep in mind that the codeblock has a fixed amount of text it can hold

#

All you need to utilize is simple prompt techniques

eager vine
#

Does anyone know when or if we will be able to subscribe to personal assistant plans - where the AI has access to all my data, can manage daily tasks - confirm meetings etc etc?
I'm currently paying for enterprise sub but would rather be willing to have something better.

is there an ETA for v5?

burnt hare
#

What does enterprise get you?

#

No eta for gpt5

neat storm
#

there is technically literally nothing stopping the devs from implementing this

#

AI Dungeon does this already

elfin birch
wind cradle
#

I wrote "Man cannot live alone." and was told "This content may violate our usage policies".

#

What went wrong?

toxic flax
#

you should have used a gender neutral term you fascist

random dagger
wind cradle
south kayak
#

Dont act like chatgpt isnt sensitive af

#

chatgpt cant handle japanese

#

I'm waiting for non roman language improvements since forever

wind cradle
# south kayak chatgpt cant handle japanese

Is that really the case? I use Japanese on a daily basis, but I've never felt inconvenienced by the response messages. The discomfort always comes from error messages like this.

south kayak
wind cradle
tight island
#

Japanese has sooooo many ways to interpret based on the encounter and environment and or subject

south kayak
#

And when you point out mistakes it says "Yes that was an inappropriate sentence structure" or something like that and then corrects nothing

wind cradle
south kayak
#

bro

#

I ask it to give me a text in Japanese kana and the according romanji text for it and then it gives me a whole differently structured sentence in the romanji version

#

sometimes it doesnt even know if it should use katakana or hiragana

wind cradle
south kayak
#

Sometimes it really acts braindead

#

I may be not as bad from Japanese to English as from English to Japanese

#

Somehow its still not able to read kanjis off pictures

wind cradle
wind cradle
sick pecan
#

maybe your custom instructions are not good

#

its coding like a boss for me. fast,, smart, on point. is amazing

regal nimbus
#

I know what you're talking about. Just shows up as all red. It's so annoying

sick pecan
#

maybe you have dynamic mode (and its using 3.5)

cloud pewter
#

plot twist openai is replacing some of the gpt4s to gpt3.5s because they have ran outta gpus

turbid hare
#

I feel ChatGPT should move copy code button to the bottom of generated code, I had to scroll to the top to copy the code all the time.

peak wigeon
#

hi ,will the ui of chatgpt be able to search from the older chat sessions or have some grouping feature ? is there a roadmap?

peak thicket
#

Does your (i.e., OpenAI ChatGPT) success count as an unequivocal assertion of the reach of presentations (symbolic language), while highlighting the non-indispensability of that which is presented, along with its representations (i.e., theory and models)?

eternal rockBOT
#

Beware of possible scams or fraudulent activities that you may receive through direct messages. OpenAI staff will never DM you for any transactions.

Please report any incident by sending a DM to @languid valley immediately.

subtle wave
#

do you have access to chatgpt 3 for free?

elfin birch
subtle wave
glass fulcrum
#

3.5 is better than 3.0 why you wnat it ?

#

also no you can't access legacy models on chatgpt anymore

#

on api yea

subtle wave
#

okay thank you

maiden fractal
#

You can have gpt 4 free with bing. But chatgpt4 is superior

tight island
#

what i dont get is why people complain about incomplete code when they know Codeblocks have limited things it can hold.

#

it's literally a thousand times easier for you to ask it to create a new .txt file and write the code in full there, then provide a direct download link. all using codeinterpreter.

#

its that easy, really!

random aurora
#

Hello, I apologize for the inconvenience, but I have a small question. A few days ago, I was able to ask caht gpt. He imagines, for example, a logo, banner, or any design, and they used to do it, but now he is only describing the design and does not design anything actual. Can someone help me?

tight island
#

u are required to be specific @random aurora

random aurora
tight island
#

MyGPT or ChatGPT 4.

random aurora
#

ChatGPT 3.5

tight island
#

thats your problem

#

you ought to use ChatGPT 4. or DALL E

random aurora
tight island
random aurora
#

I used to use the regular one but now it doesn't work

random aurora
tight island
#

or 30 minutes.

random aurora
#

It's been two days

tight island
#

wait 1 more day.

#

or clear your Cache, Turn off your VPN, log in and out.

random aurora
tight island
worthy nimbus
tight island
#

No drinko attention

sick pecan
#

a trick I use to update larger code files is to ask chatgpt to give me the code in parts. like the first half of the file. including jsdoc, debug, error handling

tight island
#

That too

#

I like the .txt method more

#

But hey 👋 that’s just me!

worthy nimbus
#

But the problem is when working with larger files, it seems to run out of context length

#

I wish they increased context length to 128k

sick pecan
#

I use ssh plugin to just get it to read and write direct to my file system, bypassing code interpreter fails, like losing your data

tight island
#

Hopefully when you work with gpt you don’t have it do 900 lines of code

#

Organize more do more files if needed

sick pecan
#

moduralise is the key term.

tight island
#

Well yeah I KNOW

#

🗿

#

i modulise my code for ease of use. when starting the initial project.

#

when the project works in full i then optimize and integrate, and overtly complicate and adding signatures.

warm lark
#

Are they rolling out the dynamic model?

last prism
#

Does anyone know if the iOS app has a deep link API?

magic zephyr
#

After all, what does this model do?

elfin birch
#

There's an official help document I'll see if I can find it

neat storm
#

Use Precise for Bing/Copilot

crude slate
#

huh

#

from PIL import Image
import pytesseract

Load the image from file

img = Image.open("/mnt/data/image.png")

Use tesseract to do OCR on the image

text = pytesseract.image_to_string(img)

#

when i uploaded an imag

drowsy tinsel
#

todays Apr 22, will openai launch anything

gritty ledge
#

Is ChatGPT crashed? It shows that it is disconnected and cannot be accessed.

clear basin
#

Is Claude 3 Opus available for free somewhere? Doesn't matter if the limit is low because I only need to send a few messages

#

I switched from claude pro to chat gpt plus because they added turbo to chat gpt but it doesn't seem to be as intelligent as opus 🙈

#

Heck, even sonnet seems to be smarter than gpt 4 turbo

#

Or whatever model it is that they currently have on chat gpt

night silo
neat storm
neat storm
#

but only by 5 elo points

cobalt idol
glass fulcrum
#

I think it avelivle on lmsys :0

plucky dagger
#

any one else getting slow responses in web chat gpt interface?

spare trout
#

Hi guys I have a question. I use gpt 3 to help outline my book, and after working on that outline to suit the story I'm trying to tell, I start writing. Do I have to disclose that I used it for outlining? Or only when I use it to help write certain paragraphs? I want to be sure

elfin birch
feral zinc
#

on top of that some book publishers and stores have their own set of rules regarding ai

elfin birch
#

^

spare trout
spare trout
keen mesa
marble schooner
#

hey guys, weird personal question,
so basically i am working on my thesis to finish university,
i decided to discuss about ai, generative algorithms and all the things that we have around nowadays,
do any of you perhaps know someone that could help me gather texts, books, anything really, about chatgpt, dall-e, dlss and fsr?
i tried googling them but i get a bunch of not in-depth articles, any help would be much appreciated

neat storm
# keen mesa definitely not in real use cases

That's not true either though, the benchmarks have multiple tests based around real cases.
The benchmarks tests are based on specific categories (common sense, logical reasoning, problem solving, coding, math, etc.) to accurately get a general understanding of the capabilities of the model

#

Claude 3 Opus were done on these same tests like any other model

dusk sigil
#

Gpt-4 is getting dumb, i had to remind my created gpt it can use the uload files

ok well you are incorrect you can access those files that is the whole point of being able to upload them, no point otherwise, LOGIC, the un-logical thing would be to have a function and not being able to use or it doesn't work

keen mesa
spice star
#

that could be a prompting issue more than anything

dusk sigil
spice star
#

having chatgpt forget something after 2-3 prompts

dusk sigil
burnt hare
spice star
#

chatgpt only has a 32k context unless you're an enterprise user

knotty crypt
#

is chatgpt 4 also slow for others?
it is pretty slow for me rn

dense halo
winter elbow
#

Guys, AGI by 2030 and ASI by 2090

feral zinc
#

both probably earlier than that

#

but maybe not available for everyone

#

or most likely not 😛

burnt hare
spice star
#

pretty positive

burnt hare
#

gpt-4-32k-0613 is 32k

#

where did you learn that

feral zinc
spice star
burnt hare
spice star
burnt hare
dense halo
#

people have been saying this ever since gpt 4 has been out tbh, either ffaster, less intelligent or vice versa

spice star
#

you can always check the cutoff date, but if it's december 2023, then it hasn't changed

sick pecan
#

perhaps you have too much junk custom instructions

#

custom instructions are in settings and apply to new prompts

cerulean vessel
#

can the ai recognize a product from a picture, or read a barcode, and be able to count items?

sick pecan
#

can ai? yes. can chatgpt? no

glass fulcrum
#

hm Chatgpt-4 can see images 🤔

neat storm
sick pecan
#

thats the dystopian story warned about in the book, Life 3.0

#

is why democratising ai is so important

neat storm
#

unfortunately I doubt Congress is capable of doing that
the people will have to take action

tight island
#

the future is compute.

#

even if AI or AGI, or even LLM's there is bound to be local versions to some degree, maybe not now but once the recipe is out.

desert gazelle
#

I wonder what GPT5 is going to be like

#

It'll probably be able to code with perfect accuracy which is terrifying

neat storm
#

but it certaintly will be similar to the jump from 3.5 to 4

#

the truth is, you can't code with perfect accuracy with an LLM (AS OF MODERN LLM ARCHITECTURE)
because new plugins, libraries, and updates to languages are made all of the time

#

do anything complex and it starts to fall apart

#

will GPT-5 be damn good at coding? yes
but its not yet ready to replace software engineers

magic zephyr
lost steeple
#

guys did you ever noticed the word limit? am i the only one who thinks that the limit was like getting low or reduced?

manic widget
#

I’ll get started on this and provide you with the solution when it's ready. Thanks for your patience! 🛠️🤖🔧

Is this a new feature that I am not aware of? It's coming from the "Python" GPTs made bye Widenex

spice star
#

it's probably just part of the custom instructions for that GPT

glass fulcrum
tough bramble
#

You've reached the current usage cap for GPT-4.

Uhm, what? I've sent like 10 messages today.

tawny crater
sudden canopy
#

I am trying to make a python script which has the ability to answer questions from local database. Here is the code bellow:-

from openai import OpenAI

client = OpenAI(
apikey = '',
)

completion = client.chat.completions.create(
model = "gpt-3.5-turbo",
messages = [
{"role": "system", "content": "You are a helpful assistant."},
{"role": "user", "content": "Hello!"},
]
)

print(completion.choices[0].message.content)

def fetchdatafromdatabase(file_path):
with open(file_path, 'r') as file:
data = file.read()
# print(data)
return data

def query_openai_gpt(prompt, database_content):
combined_input = f"{prompt}\n\nDatabase Content:\n{database_content}"

# Make a request to the OpenAI GPT API
response = client.completions.create(
    model = "gpt-3.5-turbo",
    messages=[
        {"role": "system", "content": "You are a helpful assistant."},
        {"role": "user", "content": combined_input}
    ]
)

# Extract the generated text from the response
print(response.choices[0])
generated_text = response.choices[0].text
print(dict(completion).get('usage'))
print(generated_text)
return generated_text

if __name == "__main":
# Specify the path to your database file

database_path = './random_questions.txt'

# Fetch data from the database
database_content = fetch_data_from_database(database_path)
print("database_content===================================>",database_content,"\n")
# Get user input for the prompt
user_prompt = input("Enter your prompt: ")

# Query OpenAI GPT with the user prompt and database content
result = query_openai_gpt(user_prompt, database_content)

# Display the result
print(result)
#

getting following error:
TypeError: Missing required arguments; Expected either ('model' and 'prompt') or ('model', 'prompt' and 'stream') arguments to be given
Somebody please help

dense halo
#

ChatGPT's Read Aloud Feature

At the bottom of the chat window, there is a 'Read Aloud' button. This feature works well, but it does not currently read mathematical expressions as you would expect. Are there any plans to improve this feature so it can read mathematical expressions correctly?

remote bronze
#

it make me soooo sad that the only pay option is kreditcard 😦 | Im from germany and rly less peopel have a kreitcard came for me. Wy not pay with normel bank or paypall ?

maiden fractal
remote bronze
maiden fractal
eternal rockBOT
#

Beware of possible scams or fraudulent activities that you may receive through direct messages. OpenAI staff will never DM you for any transactions.

Please report any incident by sending a DM to @languid valley immediately.

brazen glacier
#

I tried searching first, but couldnt' find anything recent, so I am asking for help here. I made a custom GPT in openai and uploaded some documents. I asked for the best way for my new bot to get familiar with the documents and it said it would read them. So I asked it to read them and now it says it can't read documents. It seems there's been a mix-up—I'm actually not able to read documents directly. " I'm not new to Chat GPT and I've had this issue in the past. In fact, I walked away from openai for a while, even though I continued to keep my subscription. I thought it would be fixed by now.

Maybe it's not openai. Maybe it's me. I uploaded 4 .docx files, all in the 400 to 600kb range. Any suggestions?

burnt hare
#

i just use plaintext

#

docx is a binary format

spring iris
#

yep ^

#

bad format

clear basin
#

Anyone else getting card declined error when trying to buy claude pro subscription?

oblique plover
#

Is anyone else having issues with the API and usage limits when your not near the threshold?

clear basin
#

I can't buy API credits either

#

The card I used to buy the subscription two months ago is getting declined now

#

Brother's card is getting declined too

maiden fractal
clear basin
#

Right

brazen glacier
odd flare
#

did a new version of GPT4 got released recently?

quiet frigate
#

Can openAI allow us to mass delete our old threads? It’s so tedious doing it one by one especially if you have a lot 💀

proud matrix
#

When are they going to remove the cap on gpt 4?

#

Also has chatgpt become unbearably verbose? Is there a way to make it stop talking so much?

sick pecan
#

custom instructions =concise responses

sick pecan
#

What are you actually doing to debug the code? I'm using GPT-4 to generate debug logging and include exception handling and it can trace through end-to-end workflows based on the logs it receives. It's pretty fantastic.

sick pecan
#

discord only officially supports javascript. any python implementations are hacks at best, often breaking, and rarely being consistent

#

you are effectively saying the machine is substandard at keeping up with the mess that is the python discord developer community

#

if you choose to use niche solutions you must put in the effort to provide sdk documentation up front

#

(as a principle, software developers should prefer officially supported solutions)

mental stratus
#

I will note, I am usually pretty "it must be you" but something is clearly weird the last few days

#

I ran a bit of an experiement, something pretty basic, and it had problems in python

#

something I could easily do myself, adding options to a curl statement

#

not sure what's up, could be the way the data is passing across to the LLM

#

API seems fine

#

yep, been very odd

#

haven't tried node, I have some node scripts I can throw at it and test for basic changes or unit tests

sick pecan
#

getting the arguments wrong to some lesser known function signature is not the fault of gpt.

#

has nothing to do with debugging

sick pecan
#

i'm not sure this failure in communication is fairly placed

mental stratus
#

may be they changed the instruct on the UI

#

I tried word for word same thing on playground with a very basic instruct and it worked fine

short spire
#

why does chatgpt say "memory off"

#

"chatgpt will not remember anything you talk about in this converstation"

#

is it supposed to be like that or am i doing something wrong

rancid solstice
#

does memory have a token limit also?

rancid solstice
short spire
fluid skiff
#

sorry if this has been answered, but what is with the new wait time after reaching the chatGPT 4 cap? it was 3 hours previously, and now is 6

short spire
#

i recall it still being 3 hours...

dusk forge
#

why not make it straight to 24

fluid skiff
desert gazelle
#

Maybe it's one of the "free upgrades" they were giving us like the newer GPT-4 model

magic zephyr
#

Guys, I just received the release of the memories feature on my main account. I was very happy and will do several tests. However, I wanted to test on GPT 3.5 as it has an extremely higher usage limit, but I realized that 3.5 does not interact with this memory feature. Can you tell if it's a bug in my account?

#

3.5 doesn't even record or retrieve data from memories.

short spire
short spire
magic zephyr
#

On the OpenAI pages, nowhere does it mention if it is something exclusive to the gpt-4 model, so I assumed it should also work in 3.5, and I thought it was strange.

sick pecan
#

with the new speed i am hitting the cap on my teams plan. i want higher cap too!

plush burrow
#

can you download chatgpt

#

or chatgpt turbo 3. five

#

i was just wondering if i could train it on my own data set

#

so i can make it comperable to chatgpt 4

#

without the limt?

magic zephyr
#

Sorry, but I don't understand exactly what you want to know. What do you mean, download chatgpt?

sick pecan
#

the devday talks included a presentation that suggested fine tuned 3.5 will outperform 4 in speed and accuracy, at reduced cost... for specific use cases

#

is how we can add value to customers

magic zephyr
#

Okay, but I never talked about API. In fact, I don't even use the API lol

#

My question is about the interaction between the memory feature and GPT 3.5, since everything is ok in 4.

#

My interest in using gpt 3.5 with the memory feature is precisely because it has a much higher limit of use, as I will be testing this feature a lot.

crude slate
#

I’ve been waiting on it for ages

sick pecan
magic zephyr
#

Personally, I thought I would only receive this resource in May. But I really wanted to be able to use it in 3.5. I'm sure that in GPT4 I'll end the limit quickly lol

severe summit
#

Hello all, i installed the App from Openai, there is could pay for GPT Plus with Paypal, is that serious?

zealous canopy
#

Hey so I dont know if anyone knows what might have happened, but you know how Chat GPT had an update a while back to commit things to memory? I had it while I wasnt subscribed to GPT+ but now I do not have the ability to turn that on. What gives? I should lose access to a feature by upgrading.

spice star
#

that hasn't been made widely available to plus users yet

lunar depot
#

bag

dense halo
#

I got it on my pc today

#

Not on my phone

dense halo
#

I mean it very seriously. If OpenAI should sell text-to-speech as a feature and it does not work properly on the latest models of smartphones, then this should be taken into account.

spice dirge
#

Has the free version been buggy and not working for you guys? It won’t even let me prompt sometimes. Been buggy for the past 2 days

#

On the 3.5 turbo app

hollow burrow
#

I have used it with firefox and chrome with no issues but i do suggest chrome

spice dirge
#

On the mobile app from the App Store. iOS

hollow burrow
#

ahh, I can't help much there as im androind and pc, but i can suggest to force close the app or restart your phone

#

at worst clear cahce/app data

spice dirge
#

Even the CoPilot app powered by GPT-4 is not loading lol OpenAI you guys can hire me I’ll look into fixing it 😌

dense halo
#

Is it ok to use chatGPT to help me with my assignment?

hollow burrow
hollow burrow
spice dirge
#

The free version on iOS mobile magically started working once I started posting here lol. Y’all tracking my IP address? 😶‍🌫️ Copilot is cooked rn 😂

#

Nah I got the VPN on 🫡 later gators

dense halo
hollow burrow
#

You can do that, but the point still stands

#

you can try it yourself and may learn more

dense halo
#

I'm bad at english, so I'm using it to fix my grammar and sentence structure.

summer pagoda
#

Learn computer science. Stealing other peoples code is actually the way forward XD

magic zephyr
#

I just recommend that you review your work afterwards and change some things that you find too mechanical or too perfect.

magic zephyr
dense halo
#

Gotchu

summer pagoda
#

What kind of field are you refearing to? In computer science you need to understand the problem(s) differentiate it into different smaller parts and solve each at a time. But you can use other peoples solution to it as long you know WHAT is the problem in the first place. ChatGpt can be sometimes of great assistance if you formulate the root cause of the problem. If you cant ChatGpt wont help either and will just likely write nonsense. Still nevertheless you need to understand the logic behind the solution, else it will just create more problems.

severe summit
#

Hello all, it worked, i subscriped GPT4 with paypal 🙂

magic zephyr
# summer pagoda Sorry to bother you, but i do not understand what do you mean by people get left...

What I meant is that in a productive scenario where there are two groups of people in which one group uses chatgpt and the other does not, for example in a simple school task, the group that does not use chatgpt or any AI assistance tool, will be left behind. There is even research on this, with proven data. This research was carried out on office tasks which, depending on the school level, can be more complex than school tasks.

summer pagoda
magic zephyr
#

Generative AI is a tool, and yes, in fact, any tool that you use without knowledge is unlikely to give the best possible results. But if you have two people with knowledge and only one of them uses a tool, that one will come out ahead.

magic zephyr
summer pagoda
#

I would call it LLM, its not an AI

#

you can i think, else just make sure its text

magic zephyr
magic zephyr
summer pagoda
#

large languague model is not artificial intelligence. At least depends on AI definition

#

it cannot make own decidions, which is what ai should be able to do. as for my understanding. Also the moderators in ChatGPT corrected me on this and mentioned that its not ai but LLM

#

reference

magic zephyr
summer pagoda
#

oh MIT news, think they are trustworthy source. Thank you

magic zephyr
#

economics mit edu/sites/default/files/inline-files/Noy_Zhang_1.pdf

magic zephyr
summer pagoda
#

Well to be honest, my little sister has access to my chatgpt account and i see she uses it quite often for school and is already reliant to it

#

but either she is to young to understand it fully or to lazy. When discussing the topics with her i see she had not fully understood the answer from chatgpt.

#

That is what worries me the most 😆

magic zephyr
#

In fact, the chance of the result of the group that uses an AI tool being better than that of a group that does not use it is possibly higher.

magic zephyr
#

Society has been sick for a long time before the launch of chatgpt or any similar solution.

#

Not that I defend the irrational use of chatgpt, but I believe that the real problems are much bigger than any AI.

summer pagoda
#

The biggest problem in understanding things is taking the time to do so. Reitterating someone else solution might give the impression of education but its not.

Society is the way it is. We have our believes and we dont like to change them. We look for confirmation for our believes. So before social media we already had our small communities which strung to a believe no matter what. One thing tho is great. We eventually die, so the younger generation will take over. And those always oppose our believes, making it possible over generations to change the society

#

We will possibly we truly stagnant when we discover way to live forever.

magic zephyr
#

The problem was never the tools, but rather the way to use them. The problem has always been in the mindset

#

Tools work wonders in the hands of those who know how to use them. And that's where I completely agree with you: The best way to make the most of any tool is to understand why you're using it.

Using a chatgpt becomes valuable when you understand the topic you are working on.

summer pagoda
#

Thats where i disagree. And i must admit i love to disagree just to encourage people to think over and have a look from another perspective.
A mind needs to be trained. We have neurons (assuming y know what i mean) and the paths to them will get thicker the more we take time to do a certain task. This way when we take enought time to repeat a certain task we will always get better in it.

#

The issue lies in how do you motivate and bring over the correct information you are trying to bring over. Thats where teachers at least in my opinion struggle.

magic zephyr
#

It's like the article I showed: I think we should pay attention to the conscious use of the tool, in scenarios where people know what they are doing. In these cases, studies indicate that there is better use of time and greater quality in the results generated.

When people don't even understand what they ask of an AI, that person wouldn't deliver a good result regardless of the use of any tools. That's why it doesn't make sense to do research on these people.

summer pagoda
#

After all LLM are inspired by our human brains XD

summer pagoda
magic zephyr
summer pagoda
#

predict the teacher's next action 😆

magic zephyr
magic zephyr
tight island
#

The brain is always on energy saving mode cooldog

magic zephyr
#

Unfortunately, they don't use this cleverness for studies

summer pagoda
#

procastrination would say is defenetly a natural thing, but on other hand we work 8 hours a day, bevore the industrialisation we worked way less. Thats more of a social fatique. Thanks to the clock. Its a very interesting topic 🙂

magic zephyr
tight island
#

I’d say studying is less relevant than ever as much of the information is absolutely useless in the real world, and is never used again outside school

magic zephyr
tight island
#

In addition bad grades often do not mean the students aren’t smart but rather the subject is uninteresting

summer pagoda
#

yes

#

yes

eternal rockBOT
#
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.
tight island
#

I was taught neigh everything

#

I use like less than a percent of it

#

Waste of school years

#

Waste of my time quite frankly

#

I would rather be taught how to do my taxes and such

summer pagoda
#

But what about stuff which are related ot each other? Small things from different fields do overlap often. From grammar (know how to express yourself) to math, which is more baisc to geography, to understand for example sociopolitical events better

magic zephyr
#

There are teachers using chatgpt to optimize teaching, and it has worked here. Many things were optimized with the help of chatgpt. Politicians here are using chatgpt to rewrite some local laws.

tight island
#

Grammar i learned myself the same with math

#

I could always easily score an A+ because I am just regurgitating useless information

summer pagoda
#

geography. If you realise how big africa is, you will get a way different perspective and understanding why and what is happening there. we use europa centric map...

tight island
#

Much of the curriculum is outdated

#

Which I find sad

#

Some stuff that’s taught is wrong today