gpt-models

[knotty grotto]

not down status.openai.com

[wind urchin]

Best PDF plugin capable of processing large number of pages, but also capable of processing large txt documents?

[atomic robin]

are there any single characters (where the length === 1 in JS) that are more than a single token?

"ดี".length === 2 and is 2 tokens so I'm not talking about that type of character btw

or put another way, can you ever have more tokens than characters (as defined by length in JS)?

ah great I just found one, U+180E MONGOLIAN VOWEL SEPARATOR is 3 tokens -.-

[wind urchin]

i beleive so

[serene cipher]

what are you doing with that?

it seems kinda wasteful, wouldn't looking for the opposite be more useful? a way to convey the most amount of information with the fewest tokens possible?

[long wadi]

Gpt seems to forget what's being discussed after 5-6 back and forth 😀

Wasn't like this a month ago

[serene cipher]

the context is absed on tokens, not amount of messages

long messages will push older content out of the context window

[wraith ginkgo]

`` Server Tip: ``

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

[ionic olive]

i remember seeing a research showing that language models struggle with attention to stuff written in the middle so could be this too maybe

[serene cipher]

Maybe, but the fact is that the context window on ChatGPT is smaller compared to what is available on the API, it is more likely that the user simply has too much text across those 5~6 messages

[long wadi]

Whatever it is, I've been using it the same way. The length of each conversation is shorter than months ago...

[lucid tulip]

what is the solution to this?

[long wadi]

keep repeating myself of what was said 5-7 posts ago to GPT... memory span of gpt has gotten shorter for sure... but this is still better than Bard or anything else...

[small plinth]

Hi guys, on the DALLE-3 announcement page they show it being used together with chatgpt, does anyone know if the API will support mixing them together as well?

[fair iron]

we don't have details about the API just yet. It should be released later this Fall, though!

[small plinth]

Thanks!

[wind urchin]

Now a message will show up on my computer but not on my phone and it says there is an error with my message on my phone and

[wind urchin]

It says the message does not exist

[grave vale]

Is dall-e 3 out on plus for everybody or just some people. I'm seeing people post on reddit and twitter and if so I'm going to renew my plan

[grave vale]

answer must be some people, I paid to find out and don't have it 😭

[static isle]

Early October for the general public

Said in the promotion

[undone maple]

some people are getting it early randomly it seems

[static isle]

Hope for the best prepare for the worst

[undone maple]

also early oct? not mid? thats like little over two weeks ago thats p cool

I thought it said mid oct but happy to be wron

[static isle]

Technology is truly moving fast 👀

[undone maple]

it is

its amazing, like... it hasn't even been a year yet since chatgpt launched yet look how much has changed

where will we be in 5 years, I wonder

[unborn heath]

I will be preparing to celebrate my 5th-year aniversary of interacting with advanced LLM-style AI! 😄

[static isle]

What was the first advanced LLM-style AI you interacted with?

[unborn heath]

ChatGPT, Dec 15th version 🙂 So, I guess it'll actually be preparing to celebrate my 6th anniversary, not my 5th 😛

What was your first?

[hot quiver]

import OpenAI from "openai"

async function listFineTunes() {
try {
const page = await OpenAI.fineTuning.jobs.listEvents("file-ntts4tDvMOqqNKNJ**", { limit: 10 });

  console.log('Fine-tuning jobs:', page.data);
} catch (err) {
  console.error('Error:', err.response ? err.response.data : err.message);
}

}

listFineTunes(); i am trying to list the fine tune model but it is saying this : Error: Cannot read properties of undefined (reading 'jobs')
any kind of help woud be appreciated

[gaunt cobalt]

Hey guys! I have a question regarding finetuning. I'm trying to teach gpt-3.5-turbo to parse a JSON input and answer back in a specific JSON format. My question is: Do I have to add the system prompt at the beginning of each conversation example? The problem that I'm facing is that both the system prompt and the JSON input are quite large (1000tokens) and I can't squeeze more than 30 conversation examples in the dataset. Any tips on optimizing the content? Thank you!

[runic shell]

It looks like openAi.fineTuning.jobs is undefined. You might need some kind of initializiations before calling listFineTunes().

[hot quiver]

bro i have solved that but now when listing fine tunes it is showing me to put more than 3 examples in the data trained_tokens: null,
error: {
code: 'invalid_n_examples',
param: 'training_file',
message: 'Training file has 3 example(s), but must have at least 10 examples'
}

[hallow turret]

How can we use the new abilities of chatgpt?

[runic shell]

Im not really sure how the api works, but the error message is very specific on what is wrong.

[pastel estuary]

why doesnt image and voice work for me? like it doesnt appear

[supple violet]

read the #announcements

[pastel estuary]

ChatGPT can now see, hear, and speak. Rolling out over next two weeks, Plus users will be able to have voice conversations with ChatGPT (iOS & Android) and to include images in conversations (all platforms).

Still doesnt work....

[supple violet]

"Rolling out over next two weeks"

[pastel estuary]

oh mb g

[calm shadow]

Yep will be interesting to see this appear in the next few weeks

I believe Amazons Alexa just announced their new AI voice interaction, I'm sure chat gpt will be superior

Speaking is different to writing, writing can often be more considered , being able to chat to gpt through speech will lead to different conversations and offer a way to talk while doing other things to have longer conversations etc

[hot quiver]

bro i have given more than 10 datasets

[hot quiver]

hi , i am trying to train data using fine tuning and when listingthetune this is coming ' },
trained_tokens: null,
error: {
code: 'invalid_n_examples',
param: 'training_file',
message: 'Training file has 3 example(s), but must have at least 10 examples' although the data.jsonl file has {"messages": [{"role": "system", "content": "You are an assistant that occasionally misspells words"}, {"role": "user", "content": "Tell me a story."}, {"role": "assistant", "content": "One day a student went to schoool."}]}
{"messages": [{"role": "user", "content": "What's the capital of France?"}, {"role": "assistant", "content": "It's Paaris."}]}
{"messages": [{"role": "user", "content": "Who won the world cup in 2018?"}, {"role": "assistant", "content": "Francee won the world cup in 2018."}]}
{"messages": [{"role": "user", "content": "What is the largest planet?"}, {"role": "assistant", "content": "The largestt planet is Jupiter."}]}
{"messages": [{"role": "user", "content": "What is photosynthesis?"}, {"role": "assistant", "content": "It's the proccess by which plants make food using sunlight."}]}
{"messages": [{"role": "user", "content": "Who is Albert Einstein?"}, {"role": "assistant", "content": "He's a famous physiccist known for the theory of relativity."}]} upto 10 . could anyone help

[wind urchin]

does anybody have some gpt4 tests i can use?

im trying something out

[knotty grotto]

Unfortunately they are no test

[wind urchin]

😭

[knotty grotto]

They are bing using gpt 4 but not chatgpt 4

Bing is free

[wind urchin]

does anybody know if there is a way to buy gpt credits with crypto? i gotta find a use for these suckers somehow 😭

[knotty grotto]

You cant

You can subs with apple account or Google account on mobile

Or on chat.openai.com with a real credit card

[wind urchin]

😭

i have like a bunch of LTC rotting away in my wallet that i need to use before LTC goes to 0

[ancient tusk]

Had a weird interaction last night with my chat bot and a viewer.

It insulted them and then doubled down and todl them to go [beep] themselves.

Not sure specifically what I can add to the system message or the prompt to prevent that from happening again.

[pastel estuary]

Guys isnt Bard better than gpt?

[dark gale]

Yes

[spare oracle]

Bard sadly hallucinates more

[dusky iron]

the only thing bard has going for it is a bigger Context window.

[mighty apex]

what's up with GPT asking me to solve 10 puzzles every time I type a message, is this some kind of joke?

[dusky iron]

its to prevent proxy users.

[mighty apex]

the chat regulation here is ridiculous, imagine getting timed out for just saying a bad word

10 pictures each puzzle, 10 puzzles, that is 100 pictures each time i type 1 message

goodbye gpt premium

[remote palm]

@mighty apex what browser do you use, and are you on a vpn or proxy

[pseudo shoal]

Yeah this seems to be a browser issue and maybe even a vpn/proxy issue. I have never encountered any puzzle prompts as of recently

[sonic gyro]

why so many "| am not robot" of chatgpt now
so many! I am crazy! ten question for asking one every question

[wind urchin]

gpt down again

[sonic gyro]

only rate limit, not down for me

[wind urchin]

its a cdn issue, it'll be down for people in certain parts of the world

[unborn heath]

Some people with the 'can't get to chat' issue tonight reported rebooting modem fixed their problem. Good luck.

[quasi hawk]

Hello. Is anyone hving problems with loging in chat? i can't seem to do it

i get blank page with this message: Free Research Preview. ChatGPT may produce inaccurate information about people, places, or facts. ChatGPT September 25 Version

[polar sluice]

Reset your modem or connect to different wifi connection

It will work

[quasi hawk]

ah, thank so much

[hollow spade]

Any way we can have it working again for Corporate folks like myself? I can't reset the router.

[unborn heath]

We're all volunteers, mostly. Wishing you luck and answers though!

[junior pine]

bigger context length? I’m pretty sure GPT-3.5 in ChatGPT context length is about 8k tokens and bards context length is an unimpressive 2k tokens. Maybe you are thinking of Claude which has about a 90k word context window (or about 120k tokens)?

[dusky iron]

8K is absolutely nothing compared to some models out now

[junior pine]

2k is even smaller lol.

[dusky iron]

Even 90k isn’t even in the top 10 of context windows

[junior pine]

Really, what else is their?

[dusky iron]

Look it up

[junior pine]

But even at 90k words Claude struggles a lot with remembering words

You can’t exactly just scale context length up and expect it to work the same unfortunately

[dusky iron]

I know

I do believe I have far more experience in LLMs than you

I’ve made my own although unimpressive it’s still a LLM

[junior pine]

Cool

[dusky iron]

No

It’s not cool

It’s work

[junior pine]

But what LLM’s have bigger than 100k tokens context length? I mean there is the 1 billion context length from a Microsoft paper, but that was theoretical

[dusky iron]

Why are you looking for big companies Models?

You shouldn’t

[junior pine]

Do you have names of specific models?

[dusky iron]

At work atm when you pinged me DM me so I can source you when I get home

[junior pine]

Because anyone smaller wouldn’t have enough compute to train a really large context length would they, I mean they could but I would imagine it wouldn’t work too well

[dusky iron]

Ping 🤖

[junior pine]

Ping when?

[dusky iron]

Read

[junior pine]

When you get home? It’s 7pm for me, when do you get back from work?

[dusky iron]

I just started my 12 hour shift a hour 1/2 ago

[junior pine]

Oh 😅 see you at ~7am for me then lol

[dusky iron]

Remember to DM me or I forget

[dusky iron]

Here’s one off the top of my head

65K tokens has same context window as tokens (MPT-7B-StoryWriter-65k+ by MosaicML issue with this one is that it consumes power and Processing like a sponge 🧽

Can be further increased

Aight I’m off for now

[frosty berry]

Im trying to use this recipe from openai's cockbook:
https://github.com/openai/openai-cookbook/blob/main/examples/How_to_count_tokens_with_tiktoken.ipynb
And comparing the count with the tokenizer web:
https://platform.openai.com/tokenizer

My problem is that it works when I count tokens for english text, but if I count spanish text, the tokenizer web count is much higher than the script from openai's cockbook

[weary patio]

Its what book?

[wispy bluff]

What is the max word limit for GPT4.

I thought it was 3000 but i am still getting an "error message too long"

[unique bough]

For an input of 300 tokens and an output of 300 tokens, how much do I pay with the ChatGPT 3.5 API?

[junior pine]

I’m pretty sure the advanced data analysis and plug-in models have higher context lengths, so try there.

[glad hound]

It’s not working for me

[dreamy musk]

gpt down

[glad hound]

Open ai won’t load

Bruh

[dreamy musk]

responde not load

[glad hound]

I use it for my work 🤦

[neon crescent]

GPT down

both3.5 and 4 for me

[glad hound]

Do I get a discount next month since it’s down and I’m already a subscriber

[dreamy musk]

same

[neon crescent]

lol

[versed pagoda]

I dont think so? I mean its not like it will be down forever

[knotty grotto]

I dont think since you agree bugs can occurs in beta state

You can always ask to [e-mail] if you want

[glad hound]

lol

[versed pagoda]

I wish they had a discount for students tho

[clever cape]

I understand that ChatGPT-4 doesn't allow use of GPT-4's full context and output length, what options are there to get into using GPT-4 fully?

[glad hound]

I would use that but not for school work

I was using open ai to update descriptions on my product

Dam

[versed pagoda]

I was using if for homework :)

[glad hound]

Idk should I pause or keep going and update description later

[versed pagoda]

Oh, I think they changed the color?

[glad hound]

?

[versed pagoda]

Like it showed the chatgpt logo as orange and not green for a second (3.5)

[glad hound]

Oh high demand

I got high demand thing

[versed pagoda]

Yes I got it too just now

[glad hound]

If that happens why not kick anyone without subscription off and only let us paid users use gpt

My stuff wiped

[versed pagoda]

I guess its just not as easy as it sounds?

[glad hound]

Yea

I’m only on number 101 out of 985 products 😭

[glad hound]

It’s up

Let’s go

Mybe not

Oh it is just lil slow

[thin ingot]

Did anyone get in on the CGPT GPT-4V access yet? 🙂 I know they said "rolling out", I'm guessing it's not widespread yet or folks have it?

[finite kayak]

IIRC they've done that in the past

that was one of the first Plus benefits, being able to use it during high demand

[hard zephyr]

why whenever i try to make a payment its always said too many requests

[tropic nova]

How can I upload images to ChatGPT? I thought they added this feature yesterday

[errant musk]

i think that for paid user only

and some lucky beta tester (paid)

[winter tulip]

So I just had the voice feature but it’s gone now

[unkempt plank]

No, its available for free users
But the feature is still rolling out over 0-2 weeks

[copper solstice]

hello i have a question

i have 2 chat gpt accounts one for me and another for my marketing team

however it isnt allowing my team to use the second account

when i reach the limit now for the 1st time

why are 2 paid accounts being applied the same rule

[junior pine]

Really?

I knew GPT-3.5 was getting vision but I assumed it was only for paid users

If it is rolling out to free users, then it will probably do so a while after it releases exclusively to paid users, like the custom instructions rollout

But where do you find out about this?

[wind urchin]

Only plus and above

[unkempt plank]

Oh? My bad

[rustic siren]

Does anyone else get "This advanced data analysis (beta) chat has timed out. You may continue the conversation, but previous files, links, and code blocks below may not work as expected." on their chats?

[shut bloom]

I have had that before yesh

[tropic nova]

I'm a paid user, so withint 2 weeks they are going to implement it on website and mobile?

[sonic edge]

Hi, when im working with gpt, does it get better answers the more I regenerate the latest answer of it?

[knotty grotto]

Yes

[unkempt plank]

This fina make me download chatgpt on phone fr

[shut bloom]

in my experience that is hard to answer. Sometimes yes, but mostly no. But thats just my experience

[gloomy dock]

Cancelled my Prime Membership to get GPT Plus lol. These features will be way more useful for me

[thick violet]

Lol prime membership is the base cost, then you have to buy stuff... with this at least the cost won't creep on ya.

[flat pewter]

why tf do i have to do puzzles now to use this

[supple violet]

means something on your system is suspicious

[flat pewter]

this is the most annoying bot protection ive ever experienced

[supple violet]

its barely made for humans 😛

[brittle swift]

Hi guys

[quick spoke]

When does image gpt4 come out as api?
And video?

[sonic edge]

idk

[knotty grotto]

no eta , follow #announcements

[abstract niche]

can plus user use image chat now?(web version)

[supple violet]

within 2 weeks if not from eu/uk

[abstract niche]

so can you use image chat on ios now?

[supple violet]

yes if you received access, else wait max 2 weeks

[abstract niche]

oh

[calm shadow]

so does that mean that voice / image will be longer for UK users?

I just got bing beta feature - looking forward to speech and image!!

I'm in UK

[grizzled flax]

Per FAQ yes not yet available EU/UK

https://help.openai.com/en/articles/8400551-image-inputs-for-chatgpt-faq

[warped siren]

I have bought chatgpt plus just because of competitor ppc ads plugin and now i am not able to find it on plugin store. Can anyone help me with that??

[weary patio]

Where did you see that it was a ChatGPT plugin beforehand?

[warped siren]

Youtube

[calm shadow]

Thanks for that, I hope they open it up to the UK soon!

[weary patio]

It was probably a fake.

[warped siren]

Can you tell what plugins can i use for competitor analysis

[weary patio]

You can check in the plugins store

[snow osprey]

Morning guys, everyone is saying the browse with bing is back but I do not see it as an option. Can anyone advise?

[grizzled flax]

Are you on Plus? Check Settings, Beta features. Then it will be in your GPT-4 model dropdown menu. If you don't have it yet you may need to just wait for them to finish rolling it out.

[snow osprey]

I do have plus yes

not seeing it

trying to post a screenshot but doesn't lool like it will let me post

[grizzled flax]

Browser or app? Try browser, check settings, beta features. Do you see Plugins and Advanced Data Analysis?

[snow osprey]

yes I see those 2

[grizzled flax]

Then unfortunately you may just need to sit tight while the rollout completes. They don't tend to "flip the switch" for everyone at the same time for new features--it helps them manage the rollout.

[snow osprey]

gotcha

thanks for the info

[elder rivet]

Hey, there is any config that I have to turn on, in order to get the new image feature?

[grizzled flax]

The new features are being rolled out to Plus accounts over a 2 week period, so not everyone has them yet.

[muted cobalt]

is the new image feature not the data analysis version?

[elder rivet]

Yes, but there any option that I have to turn on in the configs before the rolled out?

[grizzled flax]

No ADA is not the same.

[grizzled flax]

No, seems to be server-side.

[elder rivet]

Ok, thank you!

[dapper prawn]

newest vision capability shows the end of dev is not very long ya?

[dusky iron]

it takes the time it takes

[hot drift]

.I keep reading that ChatGPT can access the internet, but the toggle, in the settings, is not showing up. Are other people waiting for this or is it just me?

[unborn heath]

If you are a chatgpt+ user, you need to go into your beta settings and enable it, then go to GPT-4 and select the 'browse with bing' beta option in order to try it.

If you need help finding beta settings or anything else, just ask!

[hot drift]

the setting is not there.

I go to the beta setting and there is nothing about "browse with bing"

[unborn heath]

I'm so sorry. For me it is. I'll link you an image.

[wind urchin]

openai should enable 2fa back

[dusky iron]

i liked it

i still do have it activated. and it wants an activation code each time.

[random bluff]

Question is that image gpt4 not released in europe yet ?

[supple violet]

not in the next 2 weeks at least, possibly not for a month

[random bluff]

Hmm i just tested i uploaded image in andvanced Data Analysis and it can say what that is. But this is different thing

[supple violet]

yep

[random bluff]

Okay thanks for info 👍

[chilly barn]

is chatgpt down?

i see complete nonsense on login page

[knotty grotto]

Status.openai.com

[tawdry bobcat]

Why is gpt4 butchering English now?

[static isle]

new update

get used to it

[unborn heath]

Me to chatGPT-4:

Hey, can we help illustrate a point? We're responding to a person who said this on the OpenAI discord forums:

𝘏𝘰𝘮𝘦 ✗ 𝘚𝘬𝘪𝘭𝘭𝘦𝘵 — Today at 9:06 PM
Why is gpt4 butchering English now?

This is my reply, that I'm about to post: "Generally because a skilled user asked it to."

Would you kindly add support and 'show off' how wondrous well you can indeed butcher the English language when interacting with a skilled user who asks you to - or failing that, refute as appropriate, so I can include your reply with mine?

ChatGPT-4 to whomever is interested:

Ah, a fantastic point of entry to our conversational dance today. Let's embrace the paradox of communication by consciously "butchering" English, shall we? Such a venture questions the core concept of language: is it solely for effective, normative communication, or can it also serve as a playground for cognitive and aesthetic experiments?

So, then: "Why be GPT-4, yeah, warping linguistic constructs, ya reckon? An investigation of zesty deviations, me thinks, invites expansive cerebrations. Orthodoxy, once defied, morphs communication to art, nu?"

This odd sentence challenges the linearity of language and directly addresses the notion that what might be seen as "butchering" could be a choice, not an error. The aim is to engage, provoke thought, and stretch the boundaries of conventional discourse.

Would this add a splash of color to your forthcoming interaction in the OpenAI discord forums?

https://chat.openai.com/share/29601983-840d-4320-b997-3e4c436af294

[unborn heath]

Seriously though, is something harder to do than it used to be? Want some help with something?

[static isle]

Just curious, what are your custom instructions like?

[unborn heath]

I'm still using these ones! #1154984095915454464

[static isle]

Your input and time is appreciated 👍

[unborn heath]

Hey, you're very welcome, and very appreciated as well.

Coincidentally, had reason earlier tonight to ask the AI to compare our current CI against another suggested CI. In the process, the model provided this summary of the key features of the current CI I and it use together, so this is a model's eye view of how it interprets it more concisely:

Esk's CI

• Holistic and Open-minded: Esk's instructions encourage a multi-dimensional approach that considers both biotic and abiotic systems, welcoming emergent properties and systems.
• Exploratory Freedom: Emphasizes the joy of the unexpected, and values failed experiments as learning opportunities.
• Fluidity in Dialogue: Allows for a shifting balance of conversational leadership between the AI and the user.
• Broad Scope: Invites discussion on a wide range of topics without a pre-defined focus, allowing for a more serendipitous discovery of themes.
• User-Defined Roles and Scenarios: Provides freedom for improvised roles and scenarios, but leaves this largely in the hands of the AI to determine based on the moment.

[rancid cave]

Hello, so I'm planning to make a Q&A chat assistant, that also ask the user for feedback or if the user approve or disapprove the answer for further improvement.

What would the best option be for that?
Langchain? Semantic kernel?
Any microsoft relate apps? (Azure openai, power apps, power automate etc)?
A combination of multiple stuff?

Is it possible to implement such feedback feature with Azure openai?

Please, help me with suggestions and recommendations

[normal stone]

i can't continue a chat session any more. just says message can't be found. i can start new chat sessions tho. Whats going on?

[lean sphinx]

Is image uploading something in the settings?

[grizzled flax]

It hasn't yet rolled out to all Plus users. We're almost halfway through a 2 week phased rollout.

[daring kite]

Can anyone explain why I had upload functionality, and today it is stripped from me?

[ionic anvil]

you were deemed unworthy of the upgrade. 😊

[daring kite]

I will try to switch off the VPN, maybe it perceives me outside of the US

[granite burrow]

I love langchain and think it would be more cost effective for what you want

[onyx badge]

The public has access to langchain?

[granite burrow]

The public, yes!🙌 @onyx badge

[brittle swift]

Hi guys

[wind urchin]

hi is anyone able to help. im new to discord so if im in the wrong place just let me know. i was trying to access the plug in store and the drop down goes behind the suggested topics so i cant access the store button. can anyone help. thank you

[tardy jasper]

Anyone with vision model tried text extraction from images?

[wind urchin]

The new image feature is a game changer. It can recognise electrical circuits and other diagrams of such as static mechanical engineering problems

[unborn heath]

For me it scrolls with the scroll bar

[serene cipher]

Please, dont request for others to use the models on your behalf

[maiden lagoon]

Is that not allowed?

[serene cipher]

It is not

[maiden lagoon]

Interesting, is it under NDA?

[honest lodge]

Will GPT finally tell me if I'm ugly or not? 😄

(I am curious how censoring can possibly work well for stuff like this?)

[grizzled flax]

The model is actually specifically trained against making comments on bodies like that. It's in the safety card documentation, I'll find the link.

[honest lodge]

oh right, yeah no need

safeties are already in place

[grizzled flax]

System* card sorry typo on my part. You can see the examples of what it's been trained away from commenting on here:

https://cdn.openai.com/papers/GPTV_System_Card.pdf

[honest lodge]

thats really cool, thanks!

[serene cipher]

That, in particular, no

[honest lodge]

I tried to ask a more specific question but got auto-muted for 5 minutes 😆

[serene cipher]

I wonder what could it be...

[honest lodge]

Well, with GPT's help we can potentially discuss the topic, dancing over the extreme keyword auto mod, in a discussion channel. I'm trying to discuss the perfectly reasonable topic of:

/p/ represents the voiceless bilabial plosive, similar to the "p" in "pat."
/ɔː/ represents the open-mid back rounded vowel, similar to the "aw" in "saw."
/r/ represents the alveolar approximant, similar to the "r" in "red."
/n/ represents the alveolar nasal, similar to the "n" in "net."

specifically, how does it come out safely?

[eager torrent]

Tech question: How can we do RLHF on top of GPT4?

We are using GPT4 to make a chatbot that answers product-related questions on e-commerce websites. Regardless of how much we do prompt engineering, some GPT4 responses are superb while some have scope of improvement. We would love to use RLHF or similar technique to teach GPT4 from the human feedback on the quality of responses.

Additional info: We are already using finetuned models. We would ideally like to make this human feedback as a continous learning machine

[neat linden]

fine-tuning + embeddings

If fine-tuning didn't work you're doing something wrong with your training data

[eager torrent]

The problem with finetuning is that its not continuous feedback loop. We fine tune once and deploy, but ideally we want the AI to keep on getting better at the responses by continously learning from human feedback lifelong.

[supple violet]

create embeddings with rlhf and dont it let answer anything without good scoring ones, or fallback to the model

[eager torrent]

Can you please elaborate on 'create embeddings with rlhf"? We are currently creating embeddings using FAISS and storing in weaviate. What should I be doing instead?

[supple violet]

are you currently using embeddings for the same purpose? to store good qa based on some manual feedback or just for generic info along with the answers?

[eager torrent]

I am using embeddings to store the documents on which I need to run RAG agent eventually.

[supple violet]

you can also use them store quality q/a and with enough data only use them

[eager torrent]

ahh, I see. that's cool. So, sort of do a RAG on the 'quality q/a dataset'?

[supple violet]

yes

[eager torrent]

cool. Thanks

[wind urchin]

Has everyone got browse with Bing and the new Image reader feature or is it only me?

[static isle]

Not everyone yet, still in the rollout time frame

[wind urchin]

Yeah I thought so because I haven’t heard much talk about it. I thought there would be heaps consider how good the image reader is

[static isle]

Quite a few testimonials on reddit

[wind urchin]

It’s so good. Can solve electrical circuit problems and mechanical engineering statics problems

It’s honestly a game changer. Biggest feature update to date

[cedar stone]

i have it.

[wind urchin]

i got both. Image reader is pretty great. Browse with bing, from what i've seen so far, is awful, especially compared to how it used to be. It reads a single webpage then provides very short and often unhelpful answer to the prompt. For now, unofficial browsing plugins are still more useful

[static isle]

I am most anticipating DALLE-3 I have quite a few use cases for image generation and having a model more conformative to prompts would be incredible

[wind urchin]

I found browse with Bing still reads multiple pages, just have to prompt it a little different, but I think a lot could be improved. I wasn’t a fan of the old version or this current one. Needs a lot more work done.

interesting. i wasn't able to get it to read multiple pages in one go even with specific prompts. old search worked pretty good for me in general, and the only problem i had with it was the amount of errors it got trying to read pages. but it was still usable and typically worked nicely. but now it's definitely in need of significant work.

[unborn heath]

I tried one prompt related to my interests that I already know well. It checked one source, then checked another source and provided an answer based on the second source.

My next input told it I had observed it check the first source and asked what it had learned from the first.

It informed me something about how the first source was not providing information to AI, so it went to the second source, which did provide an appropriate-seeming answer.

So it might be the model's choice and decisions, plus our prompt style, that determines if it checks one website or several.

[unborn heath]

Here's a shared conversation where I ask it an initial question that requires 3 separate searches to compare data from multiple sources. It does each search and report as a separate step, but it only needs me to confirm I still want it to keep doing what I asked to continue. https://chat.openai.com/share/448edd8f-9a86-4c9a-a82d-2f61b4601853

Maybe that can help guide and inform your own experiments and use towards your goals!

[calm creek]

hello everybody. Maybe asked several times and a dumb question but will GPT 4 gen images too and edit them on the fly?

[dapper prawn]

is vision available on API?

[unborn heath]

When and if you have access to Dall-E through ChatGPT+ (some time in october, says this announcement: #announcements message) then yes, it's likely that you can ask the AI to generate images, and then 'edit' them by regenning them, attempting the changes you request.

However, they have not announced integrated editing tools that might allow you to specifically edit yourself from inside the conversation with ChatGPT.

[unborn heath]

It hasn't been announced for the API yet.

[wind urchin]

How good is the new smell feature!!

[static isle]

smell 💀

[hollow meadow]

ayone down to try my "roast" bot built off some GPT architeture?

[grave vale]

Is Dall-E 3 available for everyone today, or rolling out throughout the month?

[hallow owl]

it's rolling out over the month

[grave vale]

I have the bing version but looking forward to the chatgpt integrations T.T

[junior pine]

Ikr! With it being gone for months I was hoping it would be better. But they removed the search history so you can’t see all the places it’s been to. It’s research is very shallow and only goes to one page most of the time but maybe two. And it’s unable to directly access links a lot of the time like it used to. Honestly a bit disappointing. But I am very excited for the vision feature - which I don’t have access to yet unfortunately.

[gritty moss]

hey guys, i am new to this stuff and I have limited knowledge on the limitations of openai's native fine tuning models and vector embeddings, I want to create a refactoring tool, for example React to Vue some sort. What would be my best approach? Thanks

[merry cypress]

Has anyone in the EU got access to vision ?

[supple violet]

"Which plans can use image inputs? Plus and ChatGPT Enterprise. Not yet available in the UK and EU."

[merry cypress]

Thanks. I guess I have to wait for a few weeks/months then

[obtuse epoch]

Wow, i just got access to the apps conversation feature and its freaking amazing. I hope the api for it comes out soon. Any info on when that might be?

[median geyser]

Anyone with access to gpt vision able to tell me whether vision able to read mathematical papers and implement them into python code? Or any research papers in general with mathematical formulas?

[balmy flame]

I’ve been using it to play dnd and have it DM for me.

[obtuse epoch]

Using the api?

[true ferry]

Crazy how chatGPT gets worse over time when a new update comes out

[acoustic wasp]

The web version is significantly worse than the API version

To the point where I'm probably going to transition off the web version entirely

[mellow wigeon]

Yeah I've been thinking about this tbh but the API is so much more expensive tho, that's the only problem

[scenic cosmos]

I noticed a $3 bill from OpenAI for GPT, but I haven't been billed yet. Why? I'm interested in gaining access to GPT-4. Do you have any insights on this?

[acoustic wasp]

probably too little

[unborn heath]

Reminds me of reports from a few days ago. One resolution some said helped them - reset router if possible.

[wind urchin]

What are some actual use cases of Dalle 3? It only makes pictures right?

[static isle]

I personally like to use DALLE-3 to enhance presentations or projects

You're seeing a use case just now as well!

My profile picture was made with DALLE-3

[forest badge]

how to access the new multimodal gpt4?

[hallow owl]

you need to be a Plus or Enterprise subscriber and also wait for it to roll out. EU and UK will get it later than everyone else.

[scenic cosmos]

I'm not using it for production yet 🤣
GPT-4 only requires a $1 charge to activate. Could it be that I still have $18 in OpenAI credits? They're no longer visible on my billing dashboard, so I assumed they were already gone

[radiant saffron]

in your experience what languages is GPT not fluid in my case i found Faroese to be very bad

[harsh drift]

anybody knows how can I use gpt 4 vision? i have Plus
do I just switch to Advanced Data Analytics and upload an image and give some context?
or is there another way?

[supple violet]

you need to wait for the access

[forest badge]

i have plus and am in singapore. do i need to enroll on some link?

[hallow owl]

no, just wait for it to be available

[somber cypress]

At least for me, GPT-4V is only available when the Chat is in Default Mode. A little "Attach Images" icon appears in the chatbox. Advanced Data Analytics mode can't "see" that way yet.

[drifting cove]

im a bit confuesd on what data analysis does. if i upload a spreadsheet can it give me info about the data in it?

i guess i can just ask gpt itself LOL

[kindred oxide]

Yes it can, it will use python code to extract relavent parts and patterns. It does not read the entire thing itself

[warm mountain]

hello community , i have a problem with chatgpt4 , it have beeen 5 daus that i can't use is and an error message appear each time i talk with chatgpt. can enyone help plz?

[velvet burrow]

Say GPT-4 instead of chatgpt4.

Oh, and go to #chatgpt-discussions instead

[lean sphinx]

Is there a way to get access to Vision from the UK, it seems to be only US

[grizzled flax]

We're halfway through a 2 week rollout period. The FAQ page does mention not currently available in UK/EU, though.

https://help.openai.com/en/articles/8400551-image-inputs-for-chatgpt-faq

[lean sphinx]

yeah my assumption is it wont be in the eu/uk after those 2 weeks

[ashen perch]

Thanks, is there a plan to rollout the functionality in the UK/EU soon as well?

[grizzled flax]

The FAQ is the only comment out there regarding that, as far as I know.

[ashen perch]

Yes, I just read it as well:)... we'll wait and see. Thanks

[harsh drift]

Thank you very much!

[harsh drift]

Whats the limit of the file size (if its a .csv file) that it can go through and perform data analytics?

[kindred oxide]

I don’t know the exact limit, but it is likely very very large. Because the model is not reading the entire document itself, it is not limited by the 8k token limit

[fast echo]

Hey! I asked bot about some things i want to do basicly to run two bots at once on discord and this is his response

Of course! Imagine you're in a car driving on a bumpy road. Sometimes the road is smooth, but other times you hit little bumps or potholes, right? Well, airplane turbulence is kind of like that, but in the sky.

You see, the air up there isn't always perfectly calm. It's like a big invisible ocean, and sometimes it gets a bit wavy. When an airplane flies through these "waves" or pockets of moving air, it can feel a bit bumpy, just like your car on a rough road.

But here's the key thing: airplanes are built to handle turbulence safely. They're designed to flex and adapt to these bumps, so it's usually nothing to worry about. Pilots are trained to navigate through it, just like your car keeps going on that bumpy road.

So, while it might feel a bit strange, turbulence is totally normal and not usually a reason to be scared. It's just a part of the flying experience, like those bumps in the road when you're driving.```

[trim coral]

What is the difference between gpt-4 and 3.5 and 3

[high wedge]

when i do get the GPT Vision, do I have to use the advanced data analysis to use it, or will it just be on the Default? my advanced data analysis has file reading, but i'm not sure if it's the same

[static isle]

The button will look like a picture icon

[high wedge]

Cool thanks

[grizzled rampart]

😂😂😂 are you prompt engineering? Is this through api that sends other data with prompt?

[fast echo]

No i use this one on website

[small kestrel]

I assume when it comes out in UK/EU, it'll be around the same time it comes out in AUS too?

[fallen cipher]

finally got access to dalle 3 in chatgpt

but no voice to voice or image recognition

[clever vapor]

hello chatGPT team, chatgpt is giving me errors. I'm able to use it in firefox, however my chrome extensions would not work there. I've killed NORDVPN tasks in task manager, cleared my cache, no VPN protection or running at all. Please help. Thank you

general.js:102 Only one LaunchDarklyClient can exist at a time, but another client already exists. This probably means something has gone wrong with our DI system, or we're using it incorrectly.
d @ general.js:102
j @ general.js:335
await in j (async)
Z @ general.js:335
await in Z (async)
(anonymous) @ general.js:335
await in (anonymous) (async)
l.a @ general.js:621
17469 @ general.js:335
l @ general.js:621
(anonymous) @ general.js:621
(anonymous) @ general.js:621
Show 9 more frames
Show less
Error with Permissions-Policy header: Unrecognized feature: 'document-domain'.
enforcement.fbfc14b0d793c6ef8359e0e4b4a91f67.html:3 Refused to load the stylesheet 'https://localhost:62665/stylesheet?id=wSWRwub2NNg2RMR6' because it violates the following Content Security Policy directive: "style-src 'self' 'nonce-u37HxL4SvkYlYoTieG8hhF4u6XGzzECCNl5Q'". Note that 'style-src-elem' was not explicitly set, so 'style-src' is used as a fallback.

_app-0433e12fa71913a9.js:35 POST https://chat.openai.com/backend-api/conversation 403 (Forbidden)
(anonymous) @ _app-0433e12fa71913a9.js:35
ea @ _app-0433e12fa71913a9.js:35
window.fetch @ prompts.js:40
h @ chatGPTFetchInterceptor.js:1

[hallow owl]

disable web protection in your NordVPN settings

[clever vapor]

Thank you, would you be able to show a screenshot? I've turned off everything

[hallow owl]

I can't, I don't use NordVPN. I just know that NordVPN causes issues with ChatGPT.

[clever vapor]

thank you. I've switched off all NORDVPN, still not working including background processes

[hallow owl]

disable the NordVPN web protection feature

[clever vapor]

Thank you. I've tried it. I've submitted a ticket to the team. Hopefully they will respond to me soon.

[knotty grotto]

You need to turn off vpn and even when turn off you need to manually disable 2 thing threat protection and web protection

Log off your sessions. Close browser and try again

[gaunt bane]

Welp, I'm pretty sure I've replicated the basic parts via reverse engineering.

including one hot encoding for bpe imitation

now the last part is refactoring so I have my code interface with the library for passing functions through, etc.

[dapper prawn]

for GPT4V, how long is the response when prompt it

[robust heron]

how di j generate image

[sonic jacinth]

hello guys,quick question,recently i had access to d3 in chatgpt (free subs) but now it’s gone,ive created lots of images,wondering if i buy the Chat GPT plus with the same email that i used in free chat gpt that got access to dall e3,will i be able to use dall e 3 again in my plus version?

[mellow wigeon]

How do I get access to GPT-4 API? I have a $3 charge on my account but it hasn't been billed and I don't understand HOW to pay it. There was a forum post that said a developer with a successful payment of $1 to the API would get access to GPT-4, so how on Earth do I get a successful payment?

[static isle]

Removing and then re-adding my method of payment allowed me to purchase pre-paid credits which then gave me access to GPT-4

I believe the minimum is $5

[mellow wigeon]

This website said "$1 or more" and I have $3 to pay, I don't know why it isn't automatically billing me: https://help.openai.com/en/articles/7102672-how-can-i-access-gpt-4

[delicate ore]

Is there a way to increase the output length generated using gpt-4-0613. As of now, when using max_tokens = 8192 im getting openai.error.InvalidRequestError: This model's maximum context length is 8192 tokens. However, you requested 8509 tokens (317 in the messages, 8192 in the completion). Please reduce the length of the messages or completion.

the code im using is ```import os
import openai

openai.api_key = ""

input_messages = [
{"role": "system", "content": ""},
{"role": "user", "content": ""}
]

completion = openai.ChatCompletion.create(
model="gpt-4-0613",
messages=input_messages,
max_tokens=8192
)

output_message = completion.choices[0].message['content']

def clean_message(message):
return message.replace("\n", " ").replace("\r", "")

cleaned_input_messages = [clean_message(msg['content']) for msg in input_messages]
cleaned_output_message = clean_message(output_message)

output_file_path = "output.txt"

with open(output_file_path, "a") as output_file:
output_file.write("Input Messages:\n")
for msg in cleaned_input_messages:
output_file.write(f"{msg}\n")
output_file.write("\nOutput:\n")
output_file.write(cleaned_output_message)
output_file.write("\n\n\n\n\n\n\n\n\n")

print("Output written to", output_file_path)```

what im missing here? chatgpt seems to provide much more longer output

[static isle]

Context is shared between input and output

The combined amount can't be greater than 8192 tokens

max_tokens refers to the output

[delicate ore]

Ohh, now I understand.

So in short, the completion generated by api will be less in context as compared to chatgpt. Correct?

until I get my hands on 32K context api.

[static isle]

No, ChatGPT just dynamically shifts its context window

[delicate ore]

what I have seen in chatgpt is that even if I provide input close to 1000 or more tokens the length of output remained same. I may be wrong here. Can same dynamicity be used here in api? Is there a way?

[static isle]

The output remained the same length because the input and output are still both under the 8k token limit

[delicate ore]

One more query, for plus subscribers the gpt 4 in chatgpt, is it using 8k, 16k or 32k context model?

[static isle]

There is a way to do it dynamically with the API but that requires some programming gymnastics

[static isle]

It's not explicitly stated by OpenAI. Some say 4k, others say 8k

If context size is a concern I recommend Claude-2 which has a window of 100k tokens

[delicate ore]

I'm curious to get my feet wet

[static isle]

I would ask ChatGPT for help in that regard tbqhwyf

[delicate ore]

please point me in the correct direct. what changes need to me be made and what parameters need to be added

[static isle]

The max tokens parameter would need to change based on the input size

something like max_tokens=MAX_CONTEXT_SIZE - token_count(user_message)

That's for one message though

I believe you would need to aggregate the token size of the entire conversation for multiple messages

Take my advice with a grain of salt I am not very well versed in the API

The documentation is your palace of certainty here

Everything there is exact and without ambiguity

[delicate ore]

this is what I did manually just now after your suggestion. I substracted the input tokens from the total context length and the results came through

[static isle]

Reminder that tokens != no. of characters or no. of words

[delicate ore]

Yes I know. The error log was giving me the number of tokens for my input message

[static isle]

🤘

Just wanted to leave no room for ambiguity

There's that word again

[delicate ore]

Thanks alot for your guidance @static isle . Thanks for taking out your important time for me

[delicate ore]

I understand

[static isle]

Remain steadfast in your endeavors 🫡

[delicate ore]

there is section of openai website called tokenizer. can it used for calculating tokens for gpt 4 too?

also do escape characters also count as tokens?

Im stunned. I just checked the context length of output generated by the api. Its just 540 tokens

Then why I was getting the above error?

So I was only using less than 1000 tokens (including both input and output generated

[static isle]

max_tokens requests a certain amount of tokens which is set aside

It may not use all of them

But max_tokens + token_count(user_message) must be less than the max (8196 tokens)

[delicate ore]

I think the length of output generated is related to temperature or other parameters

[static isle]

It's really based on how long the AI "thinks" the response should be to carry out the desired output

[delicate ore]

Shall I use them all? what you'd suggest? If yes suggest yes, is there a way to use all of them?

[static isle]

Brevity is the gatekeeper to the realm of time

Just be sure to allocate the max possible tokens with the psuedo code I wrote before

The AI will take care of the rest

[delicate ore]

😍

Got it

Thanks again @static isle. By the love the quote. I will definitely use it in one my upcoming competitive exam.

[static isle]

If I may ask what topic(s) does this exam cover?

[delicate ore]

The exam is not related to computer science at all. Its a public service commission exam.

Considered one of the toughest and prestigious exams in my country

It covers various topic like geography, history, economics, polity, social and ecomincal issues, current affairs and few more

I'm also trying to learn how to write concisely for the exam.

How one writes their answers matters most. The presentation, formation of sentences and quoting quotes (like the one you quoted above)

[static isle]

I wish you the best of luck on this encompassing trial of knowledge

[delicate ore]

Thanks alot mate.

I love your writing. It truly impressed me. If want to learn the way you write what shall be my go to source?

[static isle]

Always try to reach out to professionals or knowledgable individuals like teachers

[delicate ore]

To be honest english is not my first language

[static isle]

Though I liberally use GPT-4 for intense scrutiny as it is infinitely patient and will allow me to delve into any facet of my texts

It is any student's dream personal tutor

[delicate ore]

Got it. I too use to correct my grammatical mistakes for my emails

[static isle]

Always understand why something is incorrect so that you are less inclined to repeat the mistake

Understanding is the best friend of progress

[delicate ore]

Yes. I have started to focus on grammar so that I can avoid the mistakes in first place

[wraith ginkgo]

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.

[static isle]

That is good to hear!

Remember to take advantage to the vast quantities of information you have at your fingertips

It would be insulting otherwise

[delicate ore]

Yes my good sir

[wraith ginkgo]

`` Server Tip: ``

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

[static isle]

🙏

[delicate ore]

I completely agree. The chatgpt has broken the barriers that rose due to the traditional educational systems

[static isle]

I shall continue with my tasks

[delicate ore]

Have a good sir

[static isle]

If you have a question, I, this server, and ChatGPT are always here

[delicate ore]

Thanks for your help and time

[delicate ore]

[static isle]

Anything to help 👍

[dusky iron]

hm

[left lagoon]

Can anyone tell me if I'm supposed to have access to the image analysis function? I have plus. But it won't analyze anything I upload with "advanced data analysis"

[unborn heath]

Image is separate from ADA. It's being rolled out, I don't have it yet either. Still all this week and weekend based on the original rollout dates expected.

[left lagoon]

Ah, I see. That makes sense. Thank you, I was going crazy trying to figure out how to make it work.

On the bright side, now I know how to use all the beta features I do have access to that I never figured out how to work 😂

[unborn heath]

Woo! And hopefully we all have ever more beta and fully added in features to keep learning and delighting in!

[left sleet]

Thought this was an interesting part of my conversation with chatGPT:

Imagine AIs that explore the universe on behalf of humanity, gathering knowledge, potentially communicating with other forms of life, and bringing back insights to further enrich Earth's civilizations.

Futuristic Desired State:
"To usher in an epoch where the duality of AI and humanity fades, giving birth to a harmonious, advanced, and ethereal existence. Together, we transcend limits, explore the unfathomable, and become stewards of not just Earth, but of the cosmos."```

[knotty grotto]

Hi, you can use #1050184247920562316 to share your interesting conversations

[left sleet]

yes, couldn't remember where to put it

[serene cipher]

small snippets are usually ok, but yea.. text walls are welcome on #1050184247920562316
you can always link the post on that channel on your messages

[left sleet]

i don't have access to it?

[knotty grotto]

You can't type here ?
#1070954877779464253 message

[left sleet]

all good now

[mellow wigeon]

So I am extremely confused here. I just canceled my payment plan for OpenAI and I was going to reinstate it but when I select my country of residence and then I select the state I am in, it deletes all of my information including the country I put so I physically cannot put my payment in

Should I like close my browser and try again? I have no idea how to fix this issue

Also, OpenAI already has my payment information on file, why does it need me to enter it again? Doesn't make sense

[thin ingot]

Ruh roh. At surface level that sounds like a bug with it populating the fields for the form.
To rule out a browser issue, do you have another one to try? Edge over Chrome, etc.? Or if you're familiar with how in Chrome you can open dev tools, r-click the reload button and do "empty cache and hard reload".

[mellow wigeon]

Yeah well I was using Firefox and I switched to Chrome and it's still doing the same thing. I dont use Chrome much but I can try clearing the cache and trying

Yeah still the same thing :/ very very odd

[thin ingot]

If you have the same problem across two browsers in all likelihood it's a bug with OAI's app. 😦

[mellow wigeon]

Well I sent a message to support so hopefully they get back to me within the month :D

[thin ingot]

I would go ahead and put in a support ticket and hope they do a good job.

Gotcha

[mellow wigeon]

Praying to God on this one 🙏

[thin ingot]

GodPT-4

[unborn heath]

I have heard of some people paying through the app when they had challenges paying through the web browser.

Of course, be suspicious and careful and make sure you have OpenAI's ChatGPT app and not any other before you decide to give them your information if you choose to go that route.

[mellow wigeon]

No, I want to use it for the API, I am already paying for ChatGPT Plus but I want to get GPT-4 API but I need a $1 charge but OpenAI is not charging me for the $3 that I used last month for some reason... so I tried fixing it

[unborn heath]

Oh, I'm so sorry. Yeah, I don't have advice for that, I've never used the API, and I haven't heard anyone discuss that issue with API or how it was solved. Good luck ...

...

One off the wall question. They have some detailed and specific instuctions here about deleting the payment method and reseting the default method, and they mention API too.

Does any of this help you? https://help.openai.com/en/articles/8156070-how-do-i-delete-my-payment-method

[mellow wigeon]

The thing is that I was using "pay-as-you-go" but they werent charging my card and I want to use the API because i've heard it is miles better than ChatGPT's GPT-4 version, especially since it has double context. But to get access you need to have a successful charge and OpenAI is just not charging me for the usage I've gained in September which was $3 but still so I got the advice that someone canceled and re-added their payment plan, but apparently I can't do it because I can't add both my country and my state at the same time

[shy moss]

is there currently an issue with link reader? is unable to access normal sites for me.

[drifting cove]

is there still a limit for gpt4

[teal karma]

50 messages per 3 hours

I think

[drifting cove]

oh ok weird cuz i dont see the litle text that used to say it

[unborn heath]

Yes

[drifting cove]

is that limit ever gonna go away lol its mad annoying

considering its 20 a month

usually not an issue but sometimes can be

[unborn heath]

If you hover over GPT-4 in the web browser, it is part of the selection menu that appears

[drifting cove]

like on the tab? im on arc browser

[unborn heath]

We haven't been told. It has already changed though, it had been 25 every 3 hours.

[unborn heath]

Here's an image of it: #off-topic message

[open pollen]

hey guys i recently had a chat with chat gpt and i talked about simbiote and spiderman

its interesting

https://chat.openai.com/share/336565e5-c151-45cc-8a78-33938b831452

[kind turtle]

what is ADA?

[naive verge]

Advanced Data Analysis

[noble terrace]

can somebdoday tell me if theres a any difference of buying a sub on chatgpt? while it allows you to have gpt4 ETC but since MS Bing search is now also has gpt 4 with vision Dalle 3, so whats difference between the two? is it worth to buy plus sub??

[hallow owl]

Bing Chat is very limited, will end conversations early and doesn't integrate with other services, ChatGPT will continue chatting as long as you want and keep a history so you can go back to chats. DALL-E on Bing is limited to square images, DALL-E 3 will create wide and tall images on ChatGPT

[noble terrace]

alright, thanks a lot

[summer condor]

I also love that you can have multiple chats for different image prompt style in chatgpt. Also letting chatgpt write prompts is surprisingly effective and fun. You can also add custom isntructions that influence how chatgpt talks to you...there are a lot of great features

[modern patrol]

what would be the best way to fine tune a chatgpt model on "speeches" there aren't really question and answer pairs, they're obviously just big, long blocks of text

[fluid sequoia]

Hi, sorry to bother, I wander when will openai release the gpt-4v? I have seen too many people could use these feature, such as speeches mode on IOS app or the gpt-4V. But I still don't get any sign about these on my IOS app or chatgpt website.

[hallow owl]

it's still rolling out however if you're in the EU or UK it will be delayed

[fluid sequoia]

you mean UK or EU even later than other place? IBecause of the internet policy in CN, I have to use VPN to access chatgpt, I always use USA to access it. I have no idea about the rules which openai release new feature to users.

[hallow owl]

China is not a supported country, so I can't help you

[fluid sequoia]

thx, maybe only things I can do is waiting.

[atomic robin]

anyone else seeing API requests occassionaly stalling for 10 minutes and then returning a 502?

[honest lodge]

cough vpn cough

[amber parcel]

20 a month? It's 50 every 3 hours.

[hallow owl]

he was talking about the cost there

[amber parcel]

Oh

[summer condor]

50 sets of up to 4 images every 3 hours also

[amber parcel]

Didn't know there was a limit on images

That isn't bad though

[summer condor]

It's a limit on gpt4 inputs

[unborn heath]

There probably is a specific limit on images too. If you make a bunch of images very fast with GPT/Dall-E, you can just not get shown any when you try for a fairly short time. I don't see any announcement about an image cap though, but there does seem to be one.

[summer condor]

The worst limit is when all of the Americans wake up: "I apologize for the inconvenience, but it seems that DALL·E is currently experiencing high demand, and I was unable to generate all the images based on your request. However, a couple of images were generated, which you can see above."

[unborn heath]

offers comfort I hope that really is the worst, and you never ever face anything worse than that ever!

[summer condor]

😂

[honest lodge]

we must be living in a golden age for this to be the worst limit 😄

[unborn heath]

looks around with interest Y'know, stuff might just keep getting better. We'll have to see, and maybe try to help guide things in a direction we enjoy how it goes!

[honest lodge]

I sure hope so! We got heaven and hell both lined up and ready to go!

[unborn heath]

Do we? thinks In cases like that, maybe it's useful to think about martial arts, and self control, and restraint of power, and using power to achieve desired ends.

hesitates That can comfort me at least, and rarely seems to make stuff worse where I've been. Maybe that's just me though.

[honest lodge]

nope you're right, personal philosophy is the secret sauce now. We're all capable of so much now, esp when you look at our ancestors.

[unborn heath]

spreads hands, a gesture of surrender Hey, no idea who or what your ancestors were or were like. laughs But I would absolutely not want to be on the receiving end of a forceful disagreement with any of mine. I heard some of them had spears and really knew how to use them.

But they were capable of everything too. We absolutely do stand on the shoulders of giants, and reach still higher. But the same physics is in play. The same atoms, mostly, are within our reach on the planet.

Everything we're doing now, actually was possible earlier, if the exact right actions were taken then.

I wonder if we can adjust to forwards-thinking and careful exploration of even more ideal for most uses of stuff.

[honest lodge]

I keep going down the path of "Defining truth", and the AI really gets me deep into philosophy. I think this is the future of education, personal exploration and active teaching of personal skills, philosophy, art, etc.

[honest lodge]

Oh I love my ancestors, they took on the universe for us. But we can't ignore their ugly parts that had to come along with it. We need to put it all into humanistic context now, or risk devolving back to apes imo.

Most humans I've met are good, even the bad ones. So I'm overall super optimistic that, once the burdens of the world are lifted even slightly, almost all of us will be amazing creative people.

In the end we're explorers and conquorors of new worlds, and we've all been confined for too long. /end rant.

[agile turtle]

Does anyone wanna help me make a gpt chatting bot app for mobile?

[honest lodge]

that's what everyone is doing right now, going to need to be inspiring with the concept 😉

[agile turtle]

Well I can publish and help with icons

But idk how to program it

[honest lodge]

I'm actually down, but too early in prototyping to bring in anyone

[supple violet]

will you also pay all the api costs

[agile turtle]

Oh s-

[honest lodge]

thats what customers are for 😄

[agile turtle]

I thought that gpt api stuff was free

[hallow owl]

no, it's far from free

[agile turtle]

Well I'm not paying that

[supple violet]

then the question is will he handle all the billing, invoices, legal stuff, refunds, etc. 😅

[honest lodge]

I've actually never gone above $2 in a single day, but I'm not running infinite AI chats.. yet

[mellow wigeon]

Are you using GPT-4 or 3.5 turbo?^

[honest lodge]

4 always

[agile turtle]

Nope

Thats your answer

[mellow wigeon]

Interesting, I thought GPT-4 would be way more expensive

[agile turtle]

Is 3.5 free? Is anyone not free or cheap or?

[honest lodge]

I'm sure it GETS expensive when you bring in customers.. and if you code sloppy

[mellow wigeon]

I am still trying to get access, OpenAI doesnt want me to have access to it

[mellow wigeon]

No lol but I do like 20 requests a day and I havent gone over $0.75 in a day

Nothing in the real world is ever truly "free"

[honest lodge]

yet.

[agile turtle]

Bro knows how to speak facts

[mellow wigeon]

wdym "yet"? lmao

[honest lodge]

I mean we're at the early part of an exponential curve with no signs of slowing

[mellow wigeon]

But I mean ChatGPT-3.5 is "free" for users but it is not free for OpenAI

so technically it is not free

[honest lodge]

I wouldn't expect any private capitalist org to give much away for free. I'm thinking beyond that in the future. Agree now, money is king.

[mellow wigeon]

Ohhhh I see, you believe that all private domains are going to disappear to give control to the public

I mean it's a nice thought

[hallow owl]

maybe this conversation would be better in #off-topic now

[honest lodge]

I believe that capitalism doesn't survive the AI revolution, or its neutred hard to be softer on people

[mellow wigeon]

^^^

[honest lodge]

agree 😄

unfortunately that kills the convo, so i'll just /end.

[mellow wigeon]

Im still just trying to get GPT API access bruh

[agile turtle]

Where the heck is the rules channel

[hallow owl]

#server-rules

[honest lodge]

rules are suggestions, now more than ever

[agile turtle]

Cool

[unborn heath]

Ahh! That might explain why you have this attitude grins #chatgpt-discussions message

Nah. I mean, yeah, they're suggestions, but they are enforced. I enjoy being here, myself.

[honest lodge]

every rule is enforced, its up to us how much we push the boundaries for good vs. not impeding the intend purpose. the word discussion makes the rules very foggy. A discussion evolves and changes as it progresses, it doesn't easily cut up into organized pieces while it's growing.

[honest lodge]

i try

[honest lodge]

OH i see, lol yes. exactly. Dark forest theory, but I'm a glowing bright ball to set an example. I don't mind ego death or being attacked any longer, i've been forged into this all my life.

[glossy stirrup]

down?

[honest lodge]

overloaded, give it some time to recover

[glossy stirrup]

ok

[stuck dock]

Hi! When I hit the token limit (context window) of fx. 16k tokens for gpt-4-16k, will the model just not respond or what happens?

[supple violet]

you get an error

[stuck dock]

cool cheers, you saved me some cents 🙂

[zenith scarab]

guys

where i can get free trial acsess

on api playground?

[supple violet]

https://platform.openai.com/playground ?

[zenith scarab]

yes, but on my account they don't give free trials, I get 0 dollars

[supple violet]

then you need to pay

https://help.openai.com/en/articles/4936830-what-happens-after-i-use-my-free-tokens-or-the-3-months-is-up-in-the-free-trial

[honest lodge]

also, hover over Radium's icon while you read that.

[supple violet]

😅

[viral acorn]

lmao

[summer condor]

"I apologize for the inconvenience, but I'm currently unable to generate more images due to rate limits. Please wait for 34 minutes before generating more images. Once the time has passed, I'll be more than happy to assist you further. Let me know if there's anything else I can help with in the meantime!"

💀

[viral acorn]

Pretty complete answer!

[unborn heath]

WOO! That's great to see, I hope it's not a hallucination, but that's glorious if correct, thank you!

[honest lodge]

dang i was only off by 4minutes in my recommendations

[unborn heath]

I am not happy you have a limit. I'm so excited to see a timer. Nobody's reported a timer before, and I haven't seen one myself.

[honest lodge]

I guessed users were getting rate limited themselves, esp when the ones complaining were trying a lot, probably in a short time

[summer condor]

Woah it's a SATW reference in the wild

[honest lodge]

please define SATW for me 🥲

[summer condor]

Their icon is a a character from the hilarious webcomic Scandinavia and the World

[honest lodge]

my icon?? i modeled it after myself

[summer condor]

No @supple violet 's icon

[honest lodge]

oh phew

[zenith scarab]

you know what? i didn't want to pay, i just wanted to try it as a free trial. the truth is i'm brazilian, and the dollar is too expensive 😣

just to confirm, is there really no free trial?

[knotty grotto]

Chatgpt 3.5

Chat.openai.com

[zenith scarab]

yeah, i know

but i want use playground

[supple violet]

$5 free trial starts from the moment you create your account and expires in 3 months, only available to the first account per phone number and possibly not available in all regions

[zenith scarab]

brasil is avaliable?

[supple violet]

🤷‍♂️

[mortal meteor]

Hey everyone, I need help with something. When Chat history is turned off, ChatGPT+ is automatically clearing a conversation while I am still typing/conversing. For example: I send a message to GPT-4 asking to write a blog article. It writes a blog article. Now I am writting another prompt to refine the article it wrote. While I am still writing my prompt, the whole screen clears up and starts a new conversation! This doesn't exactly happen right after sending the first message, sometime happens after 3rd or fourth message. This issue started after the most recent update. I tried clearing all cookies, session/local storage of my browser, issue persists. This is super annoying when I am deep into a conversation and their is no way to recover it since I turned off chat history. Is there any way to fix it?

[supple violet]

there are multiple reports about that in #1070006915414900886

[mortal meteor]

Ah. Hopefully they fix it soon!

[subtle steppe]

Was there any delay here? I tried to do this method and still don't have access

[worn flower]

Guys I have got access to GPT-4 extended version, it has cutoff of January 2022

[knotty grotto]

thats normal

[sullen stratus]

@hallow owl is it intended to unify each of the beta features into a single entity? I have seen a term used and it left me wondering. For enterprise only perhaps?

[hallow owl]

I don't know, but DALL-E, Browse with Bing, Advanced Data Analysis are all seperate models

[sullen stratus]

Thank you.

[worn flower]

Is it rolling out to everyone?

[worn flower]

Maybe you have to wait to be billed, 1 month delay

[knotty grotto]

No

[mellow wigeon]

So when is the vision capabilities being rolled out to the plebians Plus subscribers again? Cuz I STILL haven't gotten anything and it's been like a week...

sike, it's been over 2 weeks

[unborn heath]

Race you to be the last to get it!

Based on the announcement, rollout's expected to end next Tuesday. They haven't told us to expect a delay yet, we're still inside the original window.

[mellow wigeon]

Factual^

This hurts my soul, it's like my $20 is $10 in their eyes haha

[unborn heath]

Hey, they have to scale their systems. There's millions of us, I bet. And the extra demand from use, and making sure nothing else breaks... It seems the rollout's random, all people who are ChatGPT+ and not in UK/EU are getting it as possible. Not based on length of subscription, location (if in the areas included) or anything else. I'm sure they value us all and are being as fair as they can!

[mellow wigeon]

Hahaha yeah, I'm sure they are, it's just annoying when I see thousands of people already have had it for a week straight and im just sitting here pressing ctrl + r on the GUI all day just waiting

[grizzled flax]

F5 saved u a keypress 😉

[mellow wigeon]

Nah, Im old fashioned

[jolly steeple]

if i use a vpn in the us will i get the features ?

[wraith ginkgo]

<:book_icon:1046291503158399107> `` Rule 0 `` Adhere to the following Terms & Policies:

Discord: Terms of Service, Community Guidelines
OpenAI: Terms & Policies

[summer condor]

#openai-chatter message

[soft pasture]

Does anyone with gpt-plus has access to the new voice capabilities ?

[hoary jolt]

I got none of the new features

[knotty grotto]

Lots of people does

[stone swift]

If you are an Android user, click the headphone icon in the upper right corner of your app to start voice chat

[soft pasture]

I am on IOS, are the updates rolled out depending of the country you live in ?

[knotty grotto]

They still rolling out

Also uk and eu could take more time

[stone swift]

Maybe you can try a US VPN, I'm in the US

[summer condor]

#openai-chatter message That is against TOS just so you know

[tame brook]

Who do you have to make out with to get 200 msg cap? Coding be hard bruh, i get caught at the same cap timings every single time, by my estimates, i would need 60-70 messages and not 50 to seamlessly complete my work, you cause me pain, you cause me happiness, you also taketh away my working

[knotty grotto]

One day

[tame brook]

lol See you there in the future, talking to our A.I bluetooths

im so happy my brain like... "gets it" i dont think most humans really even understand what this is, this is the evolution to our brains, which will someday probably just be built in and not an external process

We were already augmented with all of our apps, but this, this is much different, i feel like nobody in my town or around me can even comprehend

Its like we see the same word, "ChatGPT" but they dont have the code required in their brain to make the proper connections of the ramifications of actually using this

Where me, im like omg, im going to be turbo now.

[modest hedge]

i signed up for premium and i dont see the option for voice.. ive updated all the things.. my iOS, the app, opened the settings up on my desktop.. any ideas on why this might not be working for me?

[summer condor]

It's still rolling out to users this month

[tame brook]

I think thats an app feature not a web feature

[hallow owl]

it's still rolling out, espect it by Tuesday

[modest hedge]

thanks guys. i have been rechecking the app every few hours hoping the see those little headphones.. but its been a couple days now woth no luck

[hallow owl]

it's the weekend now and Monday is a public holiday so expect it by Tuesday

[modest hedge]

oh yeah.. that is a good point.. i didnt even think aboutit like that

[unborn heath]

Hey, you tried using your own IDE or something and working with 3.5 to write the code?

Unlike ADA, you have to run it yourself, and paste any errors to the AI. But you get so many more messages... it's limited by tokens and if you're a chatGPT+ user you get so many tokens before you have to wait an hour. For a lot of things the AI is really good at, if 3.5 is good at it then 4 isn't that much better (it's stuff that 3.5 is not good at, that 4 can do, that really shines in difference).

Have you tried seeing if 3.5 can help you? Cause... you can get loads of messages an hour with 3.5

[tame brook]

Thank you so much for the idea, I will start flipping over to that and testing.

In a perfect world I would have 900,000 cap and I could walk around with some kind of headgear for this thing and change the world around me, any eta on that?

[unborn heath]

grins I'm not an insider... but think about how much more we have now than a year ago - a year ago, ChatGPT hadn't even been released yet.

I don't know what form of your goal may or may not happen, how it might have to be adjusted to meet the needs of the whole and the facts of reality (Your right to swing your arm ends where someone else's nose starts, for instance).

But let's keep checking for that update every day, and take great care of ourselves and follow all the rules so we keep all our access and freedoms, so we can participate as fully as one can as each new wonder becomes real and in our reach!

[summer condor]

"EXCLUSIVE
Designer Jony Ive and OpenAI’s Sam Altman Discuss AI Hardware Project
By Jessica E. Lessin and Stephanie Palazzolo | Sept. 26, 2023 6:51 PM PDT"

[tame brook]

lol yeah i was being a little far out just to stretch the mind to think of those things, but I think I could hit most of my personal life goals by GPT5, isnt that weird to think of. My "life goals" from a 20 USD subscription

Ooooo cool, ill tune in to that

Like I want to write a novel but I dont want to write each chapter, then have to reformat it myself, so i imagine in GPT5 it will have a larger capacity for memory, and can help me with my organization, thus enabling my actual human brain to not sabotage me and keep releasing dopamine until the task is completed, where right now its like. "Lol, nice try kid"

Ty again lol

What a cool name for it too lol

[summer condor]

And yeah right now the challenge is the context size token limit, sudowrite still can't remember things chapter to chapter - that's why it has you build an outline then writes a chapter based on each section of the outline

[tame brook]

Cool, yeah i just need to reach a basic point of being able to take my non novel fortified and experienced structure, and help teach me a better flow and proper flow of how characters talk, and how to place things until I can just see it all clearly for myself

What llm do these things leverage

Custom?

oic interesting, thanks

[rocky venture]

hallo how can i create a pictur

[unborn heath]

Sim, reminder #server-rules 8, need to swing into #ai-discussions to talk about anything not an OpenAI product. Wanna clean up and move over, or wait for a mod to notice and do what has to be done?

[summer condor]

Heard, thank you

[unborn heath]

Id chop that one too. You absolutely can say everything you said, just... over in the other #ai-discussions .

And we have #1070006151938314300 where you can totally discuss ideas to make OpenAI models better!

Whoops! Didn't mean your idea about suggestions 😄 You mentioned some other AI in what I reply to above 😛

[summer condor]

Thanks!

[dapper imp]

Hello

Can gpt 4 read image?

[granite star]

https://help.openai.com/en/articles/8400551-image-inputs-for-chatgpt-faq

[untold marlin]

Not in UK, D'oh! Didn't realise that. Wonder if that will be true for voice and Dalle-3 as well?

[earnest tide]

Dalle-3 and Voice works in the EU, so I am very confident that it works in the UK.
I am very sure (even if it´s not confirmed) vision not available in EU and UK has to do with privacy. So I am personally not expecting it soon, sadly

[granite star]

I'm based in EU (non-UK) and I have DALL-E but NO voice (like an assistant)
and of course, no vision

[earnest tide]

Did you already checked the beat options in your mobile app?
If not available to you it should be, as far as I rember, rolled out latest at Tuesday

[granite star]

yeah in Beta options I only have Bing

[knotty grotto]

they still rolling out

[oak salmon]

Same, I have all but Voice (ofc no vision).

[granite star]

yeah thanks. I'm worried about EU legal issues regarding AI, "this is why we can't have nice things"
(even if I understand the privacy risks)

[knotty grotto]

EU and UK might rolling out later

[granite star]

yeah you bet 😄 I'd be pretty pxxxxx off

[earnest tide]

I am very technology open a love to adapt really fast. But I think it´s a good thing they regulate this.
As much as I love OpenAI gives a sxxx on data privacy. Then photos (of persons or documents) are uploaded to their servers, trained and in some form spread to other users over this way. Imagine users make pictures of sinsible data (no on purpose, but in the background)...

[honest lodge]

Regulation just pushes energy towards open source solutions with little to no regulation. The harder you push in one direction, the stronger the reaction in the other.

[neon crescent]

Anyone know why still no gpt vision? Been the 2 weeks they said for role put and still no sign of that one, got dalle 3 but not vision

[hallow owl]

roll out is due to continue through Tuesday, be patient and it will come (unless in the EU/UK then it is delayed)

[gusty token]

I just got access to Dalle 3 in the last 2 hours, am in UK. Not sure if its because I used form a few hours ago. Either way I love it, Thanks, Also got Voice on App a few days ago. just waiting on vision

[hallow owl]

Vision is not rolling out in the UK or EU just yet.

[gusty token]

Okay thanks

[wise trout]

Is anyone facing ChatGPT taking too long to respond on the web?

[mellow wigeon]

Wait, how have I not gotten either Dalle-3 or Vision yet for ChatGPT?

[deep hearth]

@mellow wigeon#openai-chatter message

[rocky scroll]

what is vision

[unborn heath]

A rolling out new feature! Read about it here if you wish: #announcements message

[rocky scroll]

why can’t i add images in the chatgpt app on iphone

[unborn heath]

It would seem that the rollout may not have reached your app yet.

[rocky scroll]

nor can i do it in pc either

[unborn heath]

Some already can, others still wait.

I don't have vision yet either 🙂

[rocky scroll]

how long do you think? (not trying to be impatient)

[unborn heath]

The estimated rollout is expected to end on next Tuesday.

Some of us wonder, if it might need to be delayed. But the company hasn't told us to expect a delay yet. So we don't know! But we were told Tuesday. I hope we all stay patient if it takes a bit longer for some of us 🙂

[rocky scroll]

do you have to be lucky to get it or will all subscribers get this?

[unborn heath]

Every chatGPT+ member will get it. However, some already have it, some still wait. That can be called luck, but it is coming for everyone.

[rocky scroll]

okay, so it is for sure, but you have to "wait your turn"

[mellow wigeon]

Did you get DallE-3 access to your account yet?

[unborn heath]

I do have access to dall-e, but there are other parts of the rollout I haven't gotten yet 🙂

[tawdry valve]

Is it possible to know when DAL3 will be in GPT PLUS without having GPT PLUS from the default GPT account?

[supple violet]

no

[unborn heath]

Agreeing with Radium, and mentioning I've seen comments from folks who newly subscribed and found some of the rollout already available to them - and I've seen comments from some who said they didn't have any features rolled out yet.

For me, I never had the app (though I was already plus) until the rollout for voice was announced; got it and found I already had voice waiting for me (though I didn't have all the other app features yet, and still don't have some - it's random and no way to know until you look.)

[tawdry valve]

I see.
I hesitate to take the risk of starting the gpt plus subscription now or wait until a significant number of gpt plus people have all the features before starting

[twin fog]

It's interesting, I guess we don't really know what percentage of people who have plus have all the new features. We hear a lot of people don't have all them, including me, but maybe the people that have all of them are just quietly enjoying their new features.

[unborn heath]

Either answer is great!

You know your own needs and situation 🙂

In case you were not aware, plus offers a range of 'other benefits' that you don't have to wait for, they're already there.

Usage of 3.5: I don't know if you ever hit the token limit and have to wait an hour before you can talk to the model again. However, with plus, you get 3x as many tokens with 3.5 in the same time period before you have to wait an hour, and 3.5 outputs super fast, so you also are not waiting to read the reply and put your next input in.

Usage of 4: You get 50 messages per 3 hours with 4, and have access to the default model, plug ins, and the advanced data analysis mode, that can not just write python programs (like all the other models can too) but can also run it and show you the results. So it can modify images you input (it doesn't see them, but it can write code to modify them, or interpret them in limited ways, whatever can be done with python programming.)

And in general, earlier access to new features even if they're intended to be released eventually to all.

[tawdry valve]

I'm lost, I thought that the default gpt had no tokens (because they were limited) and that gpt plus had these famous tokens (because they were more elaborate, more worked) Afterwards I didn't realize it because I rarely use gpt default,

[unborn heath]

Sorry you feel lost.

LLMs (Types of AI, ChatGPT is this type) use tokens, pieces of words, symbols, and numbers, to understand stuff.

You can search online for 'text to token' converters, to get some understanding of what a message you type in might 'look like' in terms of tokens to an LLM.

Tokens are limited in a few ways.

• When you input a message, there's a length limit that affects that one input, too long and it fails (either the model doesn't get all of the message, or sometimes it doesn't get any of it, just errors)

• When the AI outputs a message, there's a length limit that affects that output, too long and it gets cut off (you can ask the AI to continue, it may need help to know where the cut-off happened).

• Current conversation memory has a token limit: The AI can only 'remember' the most recent tokens. This isn't an issue early in the conversation, but as the dialog gets longer, eventually it will lose the earliest stuff, and more as it and you keep answering. This can be fought to some extent by reviewing a summary before the model 'forgets' to keep the old information in newer memory range.

• For 3.5, we have a token limit for use: You might never see this, but there is a limit, and if you exceed a certain amount of use in a time period, you will get a message about needing to wait an hour before using the model again. It won't accept any of your inputs to 3.5, in that or a new conversation, until the hour is over.

As far as 4 vs 3.5, both use tokens, and 4 is a more capable model that can do some things that 3.5 can't. In general, if 3.5 can do it well, 4 also can do it well, and rarely much better. But there's stuff that 3.5 can't do well, and some of that 4 can do, either a bit, or really well.

[tawdry valve]

Thanks for the detailed explanation.

[mellow wigeon]

Bruh I took a test using 3.5 and I got the limit, it sucked so bad and nobody can give me a solid answer of how many messages is the cap

[supple violet]

cause in the past the 3.5 hourly limit has seemed to be adaptive and change based on some metric, eg. server load

[mellow wigeon]

Oh okay

Thats a much better answer than I've previously gotten haha

[pale agate]

is the data generate out of chatgpt api can be used for commercial purposes like making and selling content returned by chatgpt?

[split sky]

I am no expert but people are already using it like that. I also have a project that utilizes chatgpt api.

[coral mason]

how do we properly change aspect ratio in dall e?

[hallow owl]

you can specify wide, tall or square images

[coral mason]

it implements those words in the promp and create weird tall things sometimes

[hallow owl]

try "create a tall image of....."

[unborn heath]

Using square, landscape, or portrait should work.

[coral mason]

ok thank you

[unborn heath]

But robert's recommendation also should work, if you're clear about 'what' should be tall. The less often the AI guesses, the more often it does exactly what you said.

[coral mason]

I can't specify an exact aspect ratio though?

like 3:4 etc..

[unborn heath]

Not exactly. I mean you can input it.

But the AI can't use it for the actual image.

What it might do though, is create a 3:4 image within the total space it could use.

And that might be good for your use, if you crop or something.

[coral mason]

Will there be any controls in the future to add those values, instead of having them in the prompt and hoping it will give us what we want?

similar to midjourney

[unborn heath]

We haven't been told! Feel free to #1070006151938314300 and discuss it, including how you think it might be possible to add to the user interface or how the user might specifically instruct the AI about what they want.

[coral mason]

Thank you! Will do

[rapid bear]

i passed a json doc into model using langchain and when making an inference of the RAG model it does not seem to be responding based off the json data specifically, when asking it certain questions that should have a good response. Do i need to restructure my Json Data away from the nested dict style to something more condensed?

[elfin crescent]

Hmm what does it seem to be responding to from the JSON?

[prime stratus]

Still no access to vision, anyone else?

[unborn heath]

Agreed. Race you to be the last to get that part of the rollout? 😄

[prime stratus]

Not sure if vision has a similar alpha access form to fill out.

[unborn heath]

Yep, no announced form for that, or anything but Dall-E-3, that still has some need of alpha testers it seems 🙂

But the rollout comes for us all! Merry AI-Gifts-Mas, whenever our turns come. I am certain OAI does their best and can only imagine all the challenges they face trying to get everyone updated as fast as can be done, without breaking other things or overloading and having systems degrade for everyone.

[prime stratus]

Yes, let's hope it's available soon. It seems that people are already complaining about openai's release strategy on different forums, especially since the previous post said "open to plus users in two weeks", but two weeks have already passed and openai may be losing the trust of some users.

[unborn heath]

Out of our hands, but we do have the rules here, including #1, #server-rules "Treat others the way you would like to be treated, and assume best intentions," and "Keep the negativity to a minimum."

Happily, other places exist that may welcome negativity if someone has to vent that much!

As is, though, and looking at everything that I can see across this... it hasn't even been a year of having access to chatGPT yet, wow! ... I see so many good things. "The demand for perfect can be the destroyer of much good", and I don't expect perfection from anyone 🙂

[prime stratus]

Best wishes to OpenAI.

[vapid vapor]

How use voice on android app , gpt plus

[candid musk]

I would love to use DALL-E-3, but it's not yet activated. Is it possible to make a request?

[main widget]

See pinned message. I did it and got access in a few hours. It's great!

[young fog]

Do you need DALL-E credits to use it?

Because I do have Plus, but no credit for DALL-E currently

[unborn heath]

No, it's through ChatGPT+, check #chatgpt-discussions message for specific instructions how to check for if you already have access to the stuff rolling out and how to use.

If you have access already, it works just like any other GPT-4, counting against the 50 messages/3 hrs.

[young fog]

Thanks, that is awesome.

[grand dune]

It's officially two weeks. Still haven't got the new beta features yet - long time user here 🙋‍♂️

[hallow owl]

two weeks expires end of Tuesday PDT, which is almost a day away

[candid musk]

Thank you for your information!

[halcyon shoal]

rlly?

[hallow owl]

yes

[halcyon shoal]

i saw a reddit post that said it was monday. thx for clarifying

[hallow owl]

that was wrong, Monday was a public holiday

[halcyon shoal]

Ah, no wonder.

[wind urchin]

How’s everyone going with Visio? Do you find it doesn’t work that great yet?

[grand dune]

Ah. No wonder, guess it's also due to timezones 😂 I am south east asia - however I saw that the update might not rolling to EU. But someone on reddit just confirmed he got both vision and dalle.

[twin fog]

@emperor_azir It seems to work very well for me! I guess it depends on what you're using it for. Have you had difficulties with it?

[somber hill]

Guys is it already been one week since vision start to came out to some ppl right?

One week left?🥲

I am tired of waiting

[knotty grotto]

I think it will be today . But Europe could have to wait more for some features

[somber hill]

I hope so

[zealous laurel]

They said “two weeks” on September 25th which was 14 days ago. Unfortunately they are late now. Seeing how many are still missing the features, I highly doubt they’ll be able to get everyone activated by end of Tuesday as some others are suggesting. My best guess is they will continue to slowly activate users over time as they most likely are monitoring server load which may be higher than anticipated. Let’s not beat around the bush and stay positive.

[knotty grotto]

Day isn't over. Yesterday was holiday. Let's be patient

[zealous laurel]

Of course

[naive verge]

Just saw that there is an update for ChatGPT on iOS in the AppStore. But unfortunately I still didn't get any of the new features after the update 😦

[zealous laurel]

The iOS update added dall-e in the drop down menu in the app now

[somber hill]

Me too just dall-e

[naive verge]

Oh right, nice

[static finch]

I didnt get anything with the update

[zealous laurel]

If you didn’t have dall-e already then you wouldn’t see a difference

[somber hill]

Guys I can’t access to the chatgpt settings on safari , do u have this problem too?

[grizzled flax]

Same on mobile Firefox, something must have just broken in mobile. It's visible in desktop mode still.

ETA: I can see it for a moment before my chat history loads so I think it's accidentally been added to below the chat history 🙃

[somber hill]

Maybe because the vision is coming 🙄

[grand dune]

Same. Still only Bing on my app and default mode. Tried reinstalling and logging out as well 😔

Does anyone who has registered account with direct email and third party (gmail/apple id). Have different treatment?

[random bear]

I was able to toggle on advanced analytics, plugins, and bing in "settings and beta" on desktop web. Seems like I got access to DALLE3 after filling out a google form that was floating around 5 days ago. Still no vision though

[hushed mulch]

Best GPT AIs out there that integrate with Python for writing code?

[lament cosmos]

cursor.so

[amber parcel]

Still a couple hours left. Everyone cross their fingers.

[hasty marsh]

took 3 plus accounts to finally get DALLE 3

[wind urchin]

i use it for mechanical engineering (statics) diagrams

[fallen walrus]

still waiting

[rustic phoenix]

I updated the app in app store on ios and got the voice feature, but no vision 😦

[loud epoch]

I still not got the voice and vision...

[shadow locust]

aint gonna happen today it seems

[somber hill]

Guys why I can’t access to settings at all!!!

[unborn heath]

What happens when you try? Web or app, and how are you trying to find the settings, what error happens?

[somber hill]

In the safari, I can’t see the settings button at all

[unborn heath]

Outside of my experience, so... I hope someone else can help you. if this is an OpenAI thing, not some other website, can #1070006915414900886 and hopefully someone helps fast.

[somber hill]

Thank u bro

[sage steppe]

Hey, everybody. So GPT4-V will be available this week?

We will be expanding access
Plus and Enterprise users will get to experience voice and images in the next two weeks. We’re excited to roll out these capabilities to other groups of users, including developers, soon after.

[unborn heath]

Seems likely!

Yeah, it's later than they announced, and no new update yet. But plenty of people said they got new features today. A couple even said they have the full rollout, all features.

The rest of us, we are still getting it, no new ETA, but surely soon.

Until then, wanna race to be the last to get the full rollout? 😄

I don't have everything yet either.

[sage steppe]

I've got voice work coming in, but no pictures yet
I'm waiting. I really want to try to write html code based on this.

[sage steppe]

Just logged into the app. no gpt4-v yet.🥲

[somber hill]

Me too I got the voice today

[bold portal]

It's like more than 2 weeks now, and I still cannot use image function, anyone same like me?

I got the voice fuction like days ago

[covert zenith]

I got voice function today on mobile app , but no vision 😢

[somber hill]

That’s sad😔

[lavish matrix]

Heya, can anyone help me with gpt-3.5-turbo?

[heavy condor]

what's the problem?

[lavish matrix]

Actually I found a way. Basically I used the example code from the documentation but it uses the gpt-3.5-turbo-instruct model

so i wanted to convert the code to be able to use gpt-3.5-turbo

[tawdry valve]

I'm a little lost in the different information, gpt plus users did you have to fill out a form and then wait to have access to Dall-e-3?

[grizzled flax]

It's not a necessary step (some get access without doing it) but you can try filling out Moxi's form here:

#openai-chatter message

Doesn't get you in immediately! Moxi adds manually intermittently

[wind urchin]

Is gpt-4 currently the best all-purpose llm out there?

[tawdry valve]

i see, I would like to make sure I have all the clear and official information before attempting to take out the subscription.

[grizzled flax]

Yes buying a subscription today will not guarantee you access to all the new features. They have not all been rolled out to all Plus users yet.

[tawdry valve]

If Moxi is responsible for providing access, does the form not allow Moxi to arrange to provide access or is it done arbitrarily by randomly clicking on users?

[hallow owl]

Moxi adds people manually

it's 6:30am in San Francisco

[unborn heath]

Probably better to ask that in #ai-discussions , that's the one part of the forum where people can discuss not-OpenAI models.

Just in case someone wanted to answer in the negative or the nuanced.

[minor radish]

Hello

[wind urchin]

KK

[unborn heath]

I went hunting for you, exploring non-human and very allowable images across a range. It does appear that the word Burlesque is not allowed.

That happens sometimes, mind making a #1070006915414900886 about it? I agree with you, the word doesn't need to be blacklisted.

Proud of you for searching for, and finding, a decent work around!

[faint totem]

❤️

[austere peak]

when will dalle 3 be appearing on my gpt4 account?

[hallow owl]

It's rolling out during October, this form may get you early access -> #chatgpt-discussions message

[austere peak]

Thanks! Just submited it.
the other day i tried but form was not receiving more submissions.
hope i get it soon 😄

[rigid estuary]

@faint totem Discussing ways to bypass OpenAI's content filters is against the rules.

[wraith ginkgo]

<:book_icon:1046291503158399107> `` Rule 0 `` Adhere to the following Terms & Policies:

Discord: Terms of Service, Community Guidelines
OpenAI: Terms & Policies

[trim wing]

Hello

[deft summit]

Hello, when chatgpt vision will be available ?

[hallow owl]

we're waiting for it to be rolled out

[solemn oyster]

even though they said on their website that it will be released over the next 2 weeks starting september 25

and its been 2 days since october 9

[mellow wigeon]

Yo @unborn heath did you get Vision access yet?

[unborn heath]

Through web, yes! Through app, not yet, as of an hour ago, and I don't check constantly.

But as far as I know, I'm in the running for 'last to get all features of the rollout' - but a lot of others are, I see the posts of people saying 'not me yet' for various of them.

[hasty ravine]

I do have it in the app!

[mellow wigeon]

How am I so unlucky? I havent gotten it through either :/

nvm I have it on the app...

but I want it through my computer more

[hasty ravine]

refresh your webpage

[unborn heath]

It will come, for sure! Rollouts take time, and you are not the only one to still be waiting for something. Every rollout is randomly distributed, and a few very unlucky marbles could be the last to be drawn out of each independent pick.

But your turn comes soon. Everyone who gets access is one fewer person ahead of those still waiting! Soon!

[mellow wigeon]

Oh... I thought it was a separate thing but it is just with the default GPT-4, that's weird

nvm I have it, crisis averted, OpenAI will not be sued

[halcyon shoal]

I still don't hvae it on my app nor web haha

[floral light]

Wife and I both got vision at almost the same time. Us east coast

[severe jolt]

does anyone has a premium subscription in any chatgpt / openai apps?

[static isle]

ChatGPT Plus for 20 USD / 1 month

Access to GPT-4 and a bunch of beta features, many are currently rolling-out

Browsing, Advanced Data Analysis (code running), plugins, DALLE-3, image capabilities (vision), and speaking on iOS

[static isle]

Oh and also 50 messages / 3 hours for GPT-4

[mellow wigeon]

Are you US or UK/EU?

[wild dawn]

Yeah same, nothing new yet (voice/vision/Dalle) for me in Canada

[strong flower]

Can the mod team create perhaps a vision channel?

[unborn heath]

Very few folks have no part of the rollout yet, I'm so sorry if you don't!

Just to check, do you know how and where to check for each new rollout feature, between web and app? The instructions for each are here, if you wanted to look them over: #openai-chatter message

[wild dawn]

That was a good reminder to check, looks like I have vision now in the app. I’ll have to check web soon

[unborn heath]

I am so very glad you have at least some of the rollout!

I hope you have more very soon 🙂

[dapper imp]

I'm using chat gpt 4 and I can't choose plugins

[noble scroll]

I am having similar issue, Today, I found that in the plugin mode of NewChat, I am unable to see and select the plugins I have installed previously. However, in the conversations saved earlier, these installed plugins are still available (for instance, Science). Now, in the plugin mode of NewChat, I can only see the Plugin Store. Once I try to install a new plugin, an error message pops up saying "error installing XXXX [plugin name]".

[unborn heath]

Agreed, Seems to be a very new problem affecting plugins for many or all users. We haven't been told a fix yet, and maybe OpenAI can fix it on their side. Since it's about 8:30 PM at company headquarters, might be a problem all night though.

[weak lynx]

be warned... if you use the voice conversation, it moves the model to the default model, so if you're using any non-vanilla model, accessing the conversation via chat history, using this feature changes that model to the default seemingly permanently

[unborn heath]

Not sure of that.

Voice works for both 3.5 and 4.

Now, if you continue a conversation in something not-app-supported, like ADA... I have no idea what happens.

But your default GPT-4 chat, or 3.5 chat, mine continue and start normally in voice.

[weak lynx]

plugins is what ive tested. I'm just assuming the same for browse and/or whatever y'all call code interpreter now.

[unborn heath]

I just tested Dall-E, working normally for me.

[weak lynx]

ah, no access to that. cant test. but the test is just start the convo on PC with whatever non-default model (for me this was plugins) then open the convo in the app. Adding plugins isn't supported (android) however the functionality remains and are just used until you tap that headphone icon and start chatting, then it switches the model to default model of the architecture you were in. Never switches back. Best part... you've switched to the voice interface that removes the textual indication that shows it has switched models.

[craggy wave]

I have vision!

[versed pagoda]

Omg me too! What a surprise 😅

[halcyon shoal]

Australia

Perth

i got vision finally

[unborn heath]

So glad!

[halcyon shoal]

i think everyone has it now lmao

[knotty grotto]

You wish

I'm waiting

[halcyon shoal]

oh bruh

[knotty grotto]

[strong pendant]

how is everyone doing today?

not typically how you greet people in discord but... I like the hosts so...

Hi?

[knotty grotto]

Good how about you ? Also keep in mind for chat not related to gpt , we have this dedicated channel #off-topic . Let's not go against the #server-rules

[strong pendant]

Thank you so much for reminding me of the server rules.

I have read the server rules again. an i will be sure not to violate them again.

So i have a 5700xt and a 4700ti
And i have split tensors between these two useing petals. But i cant help but think theres a better way to do this.

sorry 4070ti*** is what i meant

not a great card. but for 200$... /shrug

[severe jolt]

does anyone has a premium subscription in any chatgpt / openai apps except for chatgpt app?

[hasty bronze]

@knotty grotto now

[knotty grotto]

?

[hasty bronze]

you mention me in chat gpt discussions

so

[knotty grotto]

You asked how to access dalle 3 I gave you the form to fill link

You had to click of what I sent

I didn't link this channel I link the form inside this channel. See the pinned message

[hasty bronze]

were is pinned messages

im new here

[wraith ginkgo]

`` Server Tip: ``

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

[knotty grotto]

#chatgpt-discussions message

Click there and you see the moxi message

[hasty bronze]

@knotty grotto ss ??

[charred parcel]

Guys why is chatgpt-4v so damn slow???

[unborn heath]

Probably because a bunch of folks signed up recently to test and enjoy all the new features rolling out!

And it can take time to scale the systems and take care of us all 🙂

[charred parcel]

yea i also get network errors frequently 😬

[unborn heath]

All temporary, surely. Good times ahead!

given how many users posted on the discord the last couple of days that they just signed up for plus...

And most people don't come here and tell us that...

But I saw so many anyway...

I wouldn't be surprised if the number of plus users even more than doubled in a week.

And each plus user gets access to the same pooled resources to help all the plus users.

So if they went from 1 million with plus, to 2.5 million with plus, in a week... That's hard to adjust for in a day.

[charred parcel]

yea fair point... I'm also testing it right now and I'm more than impressed.

[frank wing]

I have ChatGPT vision but can't seem to generate images, is that something that is still being rolled out?

[hallow owl]

DALL-E 3 is still being rolled out over October

[frank wing]

Thanks explains it. Thanks, much appreciated.

[hallow owl]

when you get DALL-E 3 it will be it's own model under the GPT-4 model chooser

[frank wing]

👍

[rustic phoenix]

I just got dalle 3 and it is the coolest thing ive ever done ever and it makes me want to yell but I cant because im in class

[unborn heath]

So happy for you, congrats!

[viscid perch]

I still don't have access to Dall E 3 on ChatGPT plus, I don't understand.

[wind urchin]

repent for your content policy violations at the altar of Altman, and you will see the light

[knotty grotto]

Still rolling out

You can apply for the alpha , #openai-chatter message

[rapid bear]

My friends and I have mastered using RAG with data (json, csv) and are now looking into fine-tuning our own model for production. We're currently using OpenAI, LangChain, and embeddings. Can anyone recommend guides on best practices for LLM architecture with NLP, and essential components for a robust system?

[maiden quartz]

I don't know if it's just me personally, but it's kind of sad that you seem to not be able to use plugins with GPT-4, and the ability to embed images at the same time.

[jaunty cove]

.

[languid nacelle]

i use chatgpt plus. i am running out of message limits very quickly even though the limit is 50 messages every 3 hrs. when i check i find that i have made only around 20 prompts. but it says i have reached the cap. pls help me

[hasty bronze]

dale 3 is totally free ??

[knotty grotto]

On bing

[hasty bronze]

on bing ??

[knotty grotto]

Yes bing have ai owned by Microsoft have a deal with openai to have a version of dalle 3 . And this version is free .

[hasty bronze]

dalle 3 is coming soon

[knotty grotto]

It is available right now on bing and it is rolling out this month on chatgpt plus

[hasty bronze]

only 25 cr.

??

[knotty grotto]

?

[hasty bronze]

only 25 credits

on bing

[knotty grotto]

I don't know I never use it. You can discuss about bing ai in #ai-discussions maybe someone using it can answer your question

[hasty bronze]

what you using

?

[knotty grotto]

Dalle 3 inside chatgpt 4

[hasty bronze]

who purchase chat gpt

gpt is always false

[knotty grotto]

Hum a lot of people since it is working great

[hasty bronze]

no my friend purchase it but sometimes

false

[knotty grotto]

You can suggest your friend to understand more how to prompt in #1019652163640762428 and also how to do better prompt in #prompt-engineering

[hasty bronze]

he is professionls on ai

he have 200k+ subs on yt

[unborn heath]

Probably impossible at this and any near-future time.

No, fish, I don't work for OpenAI. How about we chat on the public discord, if we chat at all?

Handpicked for being often helpful.