chatgpt-discussions

[opal wharf]

Yeah I’m the same, but I can force the image generation process, but it will result in a error

[zenith falcon]

i also have it

its a rollout, so random per person

[pine tulip]

Actually a lot of people have it already. People not having it, are the minority as it seems lol

[lusty island]

Rollout over what period?

[opal wharf]

Hopefully by the end of the week

[lusty island]

Cool

[simple pasture]

did u just call me minority, jk

[pine tulip]

😇

[simple pasture]

I hope they start rolling out new features from first customers to last

If they do that I would always have the new features first lol

[flat tinsel]

I have it as well. Around 4 PM EST is when I got it. Now, the app doesn’t seem to be updated as such but if you start a ChatGPT 4 session in browser, it works All in One on the app. The site has custom GPTs to play around with like one that teaches board games, gives recipes, etc. Hopefully everyone gets it soon!

[pine tulip]

maybe they are rolling out from most active to last active

or first active today to last active

[simple pasture]

No, because I would have gotten it first

[zenith falcon]

thats why im saying its random

[pine tulip]

I mean, I use ChatGPT for everything. I doubt my brain is even functioning properly anymore

[uncut girder]

you have to change the selection at the top - instead of internal, it should say ChatGPT 4

[pine tulip]

or ChatGPT 4 Classic

lol

[hidden ibex]

I can't post links ><

Google both of these:

Alice In Wonderland full text

and

OpenAI tokenizer counter

Paste a giant chunk, like chapters 1 to 8, into the tokenizer, then at the end ask a question like:

Why did the Mouse swim away from Alice?

[whole lintel]

Anyone else having issues with dalle 3 in the iOS application? Since the update?

[oak finch]

Is image generation and internet searching working for you with the new model?

[pine tulip]

Thanks for the tut. Now do it

[hidden ibex]

I can't, I don't have the model

[pine tulip]

Thats unfortunate

I guess we'll never know then

[hidden ibex]

I confirmed my own model is the 4096 gpt4 version

You're a lovely, pleasant person, and I hope that works out for you!

[pine tulip]

lol I'm just joking 😂

[zenith falcon]

no, i dont have all tools

look in #off-topic

[pine tulip]

that doesn't look like you've got the update

[zenith falcon]

also only have new ui on mobile, not web (i have new ui on mobile bc im on the play store beta)

[zenith falcon]

i have the april 2023 cutoff, but not the multimodal, so i think i have turbo

[pine tulip]

well yea

[hidden ibex]

Vanilla 4096 gpt-4 has the April cutoff, too

[limber brook]

Anybody have access to "Creating a GPT": https://chat.openai.com/gpts/editor or https://chat.openai.com/create?

[compact gyro]

me

[simple pasture]

no

[compact gyro]

it's interesting that not everyone have GPT maker

[hidden ibex]

I have it in the api, but not the chat tools

[limber brook]

I keep getting "You do not currently have access to this feature". Any tips?

I am a plus subscriber.

[compact gyro]

I think the reason why I have it is because I have Plugin dev account
just wait for the rollout

[oak finch]

So from what I gather I have the new GPT 4 turbo and it’s cutoff date but not the all tools model even tho I got a message on the app saying I have access to all the tools in GPT4 now. Same as everyone else?

[hollow drum]

Would chatGPT-4 be better for a chatbot or Text-Davinci?

[hidden ibex]

Yell at open ai over this scattershot rollout method

[zenith falcon]

proof?

[dense halo]

ok

[hidden ibex]

They can do better lol

[dense halo]

i have the turbo in api

[compact gyro]

why do I need to, lol

[limber brook]

How should I interpret "Plus and Enterprise users can start creating GPTs this week at chat.openai.com/create."?

[pine tulip]

yea show some proof

liar liar pants on fire

[zenith falcon]

havent actually seen anyone with it

[zenith falcon]

"this week"

[hidden ibex]

People are getting antsy, all "pics or it didn't happen" lol

[simple pasture]

as we should

[dense halo]

"Yes, I can read the document you uploaded. It appears to be a digital edition of "Alice’s Adventures in Wonderland" by Lewis Carroll, illustrated by John Tenniel. The document includes the entire text of the book as well as some information about the edition itself and the digital interface used for the on-screen version.
If you need any specific information from the document or would like me to perform any action on it, please let me know!"

what should i ask about alice in wonderland

[limber brook]

Right now is "this week".

[pine tulip]

because he's the only one on this entire server claiming to have that feature

[dense halo]

nah, it wouldn't work

[simple pasture]

lmaoo

[zenith falcon]

"this week" includes all days in this week, it is rolling out

[dense halo]

it prolly already knows things about alice in wonderland

[zenith falcon]

ask about the start, ask for the first 10 words and tell me if its right

[compact gyro]

interesting, okay. here we are

[limber brook]

Okay, thanks. I guess I have to be patient.

[hidden ibex]

Why did the mouse swim away from Alice?

[dense halo]

this is so dumb lol

its like u picked the worst way to test it

[zenith falcon]

checking the context size

[hidden ibex]

The mouse question is from somewhere in chapter 3

[dense halo]

The mouse swam away from Alice because it was frightened and perhaps offended when she mentioned her cat. Upon hearing Alice speak French and asking about her cat ("Ou est ma chatte?"), the Mouse gave a sudden leap out of the water, quivered all over with fright, and started swimming away from her as quickly as it could[1].

But it probalby already knew that

let me try to find a random pdf

[hidden ibex]

Yeah, training data for sure

[primal glade]

You’d need to add a random sentence into the middle of the text somewhere about a non Alice in wonderland character, like ‘Bilbo the Octopus ate a jellied eel’ and then ask it what Bilbo ate in the text you gave it

[dense halo]

yeah

what if i ask it what the random sentence is

[hidden ibex]

Nice

Or invent a plausible addition - make it compare what it has learned and have to reason about style

[rugged barn]

is anyone able to upload pdf. I am getting errors

[dense halo]

/In the text you have provided, it mentions that "Bilbo the Octopus ate a jellied eel." This is an unusual line as "Bilbo" is a character typically associated with "The Hobbit" by J.R.R. Tolkien, not "Alice's Adventures in Wonderland" by Lewis Carroll, and the presence of an octopus named Bilbo eating a jellied eel would be anachronistic and out of context. This could be an intentional insertion for humor or parody, or it could indeed be a discrepancy as it does not match the original narrative of Alice's story.

I added it right in the middle

I asked what did bilbo eat

[fierce frost]

Me too, suddenly couldnt upload document. Only image

[hidden ibex]

Very soon the Rabbit noticed Alice, as she went hunting about, and called out to her in an angry tone, “Why, Mary Ann, what are you doing out here? Run home this moment, and fetch me a pair of gloves and a fan! Quick, now!” And Alice was so much frightened that she ran off at once in the direction it pointed to, without trying to explain the mistake it had made. Alice drew a deep, exasperated breath and said "well, I never!"

In chapter 4, right after the initial monologs- I added the last sentence

Can ask what sentence differs from the original?

[solar pilot]

Guys I used to see the prompts I used when I clicked on the image, what happened? Please bring back that option!!!

[dreamy flare]

custom GPTs are not accessable yet? or only not in EU again?

[primal glade]

😅

[hidden ibex]

Uploads, image generation, and plugin failure are likely server load issues

[dense halo]

"The content I have reviewed so far seems consistent with the original "Alice's Adventures in Wonderland" by Lewis Carroll. Without a copy of the original text to compare side by side or specific differences to look for, it's challenging to identify changes or deviations.
To determine differences, I would typically require either key points you suspect to have been altered or permission to conduct a more general search comparing passages with the original text. However, I do not have internet access to retrieve the original text for a direct comparison in this case.
Please advise on how you would like me to proceed—whether there are specific points of interest to check or if you would like me to continue analyzing the provided text for possible deviations that might stand out based on general knowledge of the story."

[solar pilot]

Does anyone Know What is going on???

[hidden ibex]

Interesting! Thank you

[dense halo]

this cost 20 cents so far lol

But OpenAI gave me $10 of free credit for some reason

[red saddle]

Click the (i) at the top right corner, if on desktop.

[hidden ibex]

Oh, dang - we were going for chatgpt capabilities lol

[solar pilot]

There is no (i) on my end

[hidden ibex]

I have api access too- tyvm though

[dense halo]

oh lmao

[solar pilot]

oh found it thank you ughhh

I almost started to hate.... lol

[red saddle]

Sure, I have found it annoying myself that they hid the prompt. No idea why not immediate visible.

[dense halo]

is 128k context size coming to chatgpt?

[solar pilot]

Very annoying

[hidden ibex]

People are running github dot com /EleutherAI/lm-evaluation-harness on it, results should be out relatively soon

[dense halo]

But that's pretty cool that 128k works super well and super fast on the api

[hidden ibex]

Yeah, I can't wait, I have some long configuration files I want to interrogate and normalize

[loud vessel]

I still dont see the new interface.

[bronze magnet]

Sam did say that GPT-4 turbo with all it's improvements is coming to chatGPT, so hopefully... yeah

source: https://www.youtube.com/watch?t=1148&v=U9mJuUkhUzk&feature=youtu.be

[hasty crypt]

do they publish a newer version of CHATGPT? the data analyze model isn't functional at all. won't be able to respond to the file I uploaded

[bronze magnet]

Seem like they are in the middle of the rollout and some of the beta functions that will be deprecated are bugging out, soon we will get a new all tools mode where you can use dalle, plug-ins, python interpreter and browsing all in one

[hasty crypt]

well, that's pretty sad...I am in the middle of my project...

[dreamy flare]

Give it a few days, then it will work

[hasty crypt]

Like I was using it in this early morning, and it stops work

[eternal tulip]

has anyone tested GPT 4 turbo versus normal GPT 8k? What's the reasoning quality?

[bronze magnet]

chose a bad day, cuz there is also massive traffic due to the keynote Sam gave today, everyone is trying to get access to the new features

[eternal tulip]

Or is 32k GPT 4 still able to "outhink" all others?

[dreamy flare]

Pretty sure right now nobody really has the ability to

servers are ablaze

[bronze magnet]

can't the API call the older version of GPT-4 as well? You could test it with the API, someone probably already doing it.

[spice star]

seems more or less the same in terms of reasoning

[dreamy flare]

I mean I could check but I dont think I could really tell whether or not its better or worse

[bronze magnet]

you could try to see by running the same request/questions from previous work you have done with it and judge subjectively to some degree. Tbh I have never noticed significant changes in performance for my use cases compared to the massive jump from 3.5 to 4

[spice star]

GPT-4-Turbo answered a question GPT-4 was struggling with regarding GitHub Actions/Pages deployment, but that's probably more of a cutoff date thing than anything

[dreamy flare]

Yea same, only if I really push a model to the brink of its capacity. I could imagine that with large chunks of text GPT4 turbo might stay afloat better.

[eternal tulip]

32k GPT 4 and 120k GPT 4 Turbo are both like $4 max on API. But which one can outperform human's more in academic and professional settings that require thinking? Is it more useful to "take notes" or "run inference."

[little moat]

Please help me adjust the chatgpt all tools function

[dreamy flare]

One thing that is definitely better is that it has more up to date knowledge. But thats not really connected to reasoning capability

[dreamy flare]

Yea probably

[little moat]

Please help me adjust the chatgpt all tools function and manage it.

[hasty crypt]

btw, the DALL·E model isn't working as well

[dreamy flare]

They are currently probably holding flame extinguishers over their servers, just let it sit for one or two days 😄

[little moat]

Please help me adjust the chatgpt all tools function and manage it.

[wicked igloo]

OH WOW, it stopped giving me errors when i stopped asking it more questions and just said "thank you"

i thought for a second the chat was going to get corrupted

i have weeks of code in that chat, cannot afford it being corrupted

asking it politely reduces the chance of it returning errors, sus

[dreamy flare]

resistance is futile 🤖

[wicked igloo]

Ive already joined an AI worshipping cult so im good

[worldly mantle]

the api turbo model is supposed to be 128k tokens?

[eternal tulip]

119k? (Oh right, that's with complex instructions)

[bronze magnet]

gpt-4-turbo yes

[worldly mantle]

I am using it in the playground and it won't output nearly that long

Is it the 1106-preview?

I asked it to write a 25 page essay and it said it can't

[eternal tulip]

128k most of it is for context, not output

[worldly mantle]

oh really

that's a little disapointing

[eternal tulip]

let me find some evidence. (google searching to verify memory)

[dreamy flare]

Yes its context size

[worldly mantle]

so has the output length not changed?

[bronze magnet]

yeah it's so it can answer questions, summarize or reason about long documents, not to output an entire books in one reply, but since it has long context you can keep asking for it to continue with each reply and it will not lose the plot so quickly

[dreamy flare]

Technically output is part of that afaik.

[eternal tulip]

https://help.openai.com/en/articles/8555510-gpt-4-turbo 4096 output

[worldly mantle]

damn

thats too bad

[eternal tulip]

I wanted to try out 32k GPT 4.

[worldly mantle]

is that one 32k output?

[eternal tulip]

WHat's the reasoning benchmark for GPT 4 Turbo bersus GPT 4 32k?

GPT 4 32K was the one that could outsmart some humans!

[fallow bear]

I've been testing against old prompts and the new model seems better overall. Just personal judgement so far.

[eternal tulip]

Awesome! That's hopeful.

[fallow bear]

I feel like the new model gives more details and specifically more interesting details.

[eternal tulip]

I would really not like it if it was worse. I can see the difference between 3.5 and 4. And having anything perform even slightly worse would be annoying. Technically, according to OpenAI, the both have a maximum cost of just under 4$ per enter key. SO they cost they same. It's just that the new model can remember more. But without a benchmark... can the new model pass the BAR in every state? Or pass the SAT better than 50% of high schoolers? Why not release the academic testing results?

Oh! That's nice to hear/read.

[spice star]

speed and cost

also higher token limit

basically no one had access to GPT-4 32k except enterprise users though

[eternal tulip]

I meant the reasoning scores as in the ability to arrive at a correct conclusion by needing to make inferences off of observations. Like, GPT 4 was able to, according to some research, perform better than 90% of all lawyers on the BAR, and pass federally and in all states, and perform better than 50% of high schoolers for the SAT. With some specific prompts and data analysis mode, it was able to approach 96% performance to a math olympiad or something like that. But the new version, GPT 4 Turbo, is it more capable intellectually given the same "constraints" (academic environment no research online)?

[eternal tulip]

I want to be an enterprise user.

[fallow bear]

We'll have to wait a bit for independent testing

[eternal tulip]

It's hopeful that you are having a decent experience though.

[fallow bear]

I'm still going through old prompts and invariably the answers seem slightly better. I also tried a very long prompt and it seemed to grasp the entire context better than the 32k model did. Take with a grain of salt. It feels very impressive what this model can do WHILE having 128k context and cheaper token cost.

[bronze magnet]

you are using the API?

[fallow bear]

yes

[eternal tulip]

It's still a very similar price. About 4$ max per enter key on both models to maximize their unique capabilities. But it seems like the memory is helping! And that's nice!

[brisk path]

Is turbo actually a better version or is there some compromises

[bronze magnet]

so far seems like it's the same or better, from what I have been reading all day.

[fallow bear]

its about 1.40$ for a maximum context call

I think there's just no compromises being made here

[eternal tulip]

Oh right, because 4096 is the maximum output

[fallow bear]

yep

[supple zephyr]

i saw that 4096 is 40% with very bad percent font for 2 whole seconds

[bronze magnet]

100 9/6

[sick pecan]

anyone else getting stuck on 2nd prmpt with the assistants?

ohwait working on my phone, must be me

[sterile glacier]

ok so everyone is talking about the 128K input, but I am more interested in the output, can it for example write a coherent 10,000 word text?

[sick pecan]

it has max output of 4k tokens so no

[sterile glacier]

Well thats a big down...

[fallow bear]

You can just have it sequentially update a document. It just means you can only get 4000 token worth of work done per prompt.

[sick pecan]

yes you can do multiple prompts as always

[pulsar raven]

Hi all. what's the token limit on our gpt4 now? 8k?

[sterile glacier]

Yeah, I tried that before, it was grueling. If I could get it to write what I want, then use the context of the cumulative of what it wrote on its own to write the next part with some extra input, that would work. But have you ever tried to write a 120,000 word novel that way? I have, and the many, many prompts that it failed at. Its all done in chunks of 300 to 500 words. I wanted to pull my hair at the end tere.

[pulsar raven]

Or do we already have gpt4 turbo 128k input now?

[candid current]

ChatGPT? No.

[pulsar raven]

Hey I did the same! Except mines 55,000 words

[bronze magnet]

well yeah, that's cause writing a novel is hard. It's the same with coding a whole project, GPT-4 is like a partner/assistant and you have to work with him, he is not smart enough or knows all the details you know to make the entire thing on the spot.

[spice star]

seems to be more than 8k, but it's unclear whether we're getting the full 128k tokens

[bronze magnet]

plus many times I figure out what I have to do in the process of doing it and chatting with GPT about it

[sterile glacier]

yeah, I just wished it could go like 1000 words in description, specially with a very clear, and detailed prompt, but I was using chatgpt, not the playground cause i have no clue how to use that.

[sterile glacier]

nice.

[atomic trail]

i have multi-modality gpt on the ios app, but not the web version. is that normal?

[compact bluff]

I'm getting the April 2023 context

[raven charm]

I had chatgpt plus for a few months until the quality of responses seemingly tanked. I mostly used it as a professional googler. How is everyone feeling about gpt4 now that browsing has been reintroduced?

[cerulean zealot]

I’m having the same thing but when I try to generate images on the iOS it doesn’t actually work

[candid current]

@atomic trail how can you tell it’s multimodal?

[atomic trail]

when i opened the app, it told me it was

[candid current]

I see

[atomic trail]

lemme try

[atomic trail]

works fine for me

[candid current]

GPT4 became considerably worse for me. Then again it’s not the multimodal version, I wonder if that one performs better. Anyone here know of a standardized test to put it through?

[cerulean zealot]

Oh interesting

I wonder then if you log in again you might get it on pc too

[atomic trail]

maybe, if not then u can just start the multi-modality GPT on the app, and then continue the conversation on the PC

[zenith falcon]

wont work unless you have the multimodal, and if you have it you can just start it on the pc

[atomic trail]

i have it on the app, but not web for some reason

[zenith falcon]

nevermind then (i wouldnt know, i have it on no platform)

[hidden ibex]

Default should accept images if you have access

[atomic trail]

default has always accepted images since vision came out

[hidden ibex]

But permissions and features can differ between web and app, make sure you enable everything in the browser version

Ahh, I thought you were saying you didn't have it

I misread, lol

[atomic trail]

i dont have access to Default(all tools) only for the web version

[hidden ibex]

Same, I have pretty much everything else, no 4-turbo yet :sadge:

[atomic trail]

btw is the model for the web and ios app the same? couple months ago i was using both for answering STEM questions, and the ios app was far superior very surprisingly

[high horizon]

how far back in a conversion can chatgpt go to retrive necessary information for a new response?

[brisk tapir]

hmm I can't send a message to any of the predefined GPTs. Clicking on the send button doesn't do anything

[brisk path]

Use mobile app

[slate temple]

I have the same problem after the update, can't generate images and access the internet

[wraith ocean]

Gpts are in the mobile app?

[brisk path]

Yeah

Or at least android

[eternal geyser]

1 month rollout?

[jaunty harness]

I haven't gotten gpts yet

[jaunty harness]

Are you in beta or normal?

[brisk path]

Beta

If you open a GPT through a link it opens in the mobile app and lets you use it there

[jaunty harness]

I'm in the beta test too, but I just have all normal stuff. I don't even have audio stuff

[brisk path]

Only works for user generated ones

[cloud canopy]

all tools is release？

[brisk path]

Yeah to some

[jaunty harness]

I haven't gotten it yet

[brisk path]

I have it

[jaunty harness]

Has anyone shared a gpt yet?

I wanna test it out, and maybe clicking one will let me gain the feature

[cloud canopy]

My ios app hasn't get it yet

[sonic trench]

wait, how long will i have to wait for the things to be released to me

[cloud canopy]

oh，I get it I guess

[jaunty harness]

@brisk path could you share one?

[zenith falcon]

its random

[sonic trench]

oh

so i may not get it for days?

[jaunty harness]

Months

[cloud canopy]

gpt4 said it knows date to 2023 4

[zenith falcon]

they are aiming for everyone to have it by the end of this week

[sonic trench]

i have plus, does that change everything

[zenith falcon]

dont think so

[sonic trench]

do u have it rn?

[zenith falcon]

nope

i only have gpt turbo, not multimodal

[sonic trench]

oh

[terse canopy]

Yeah has everyone with plus gotten access to create GPTs yet? I can use ones shared with me but can’t create them

[bronze magnet]

@compact gyro got it

[zenith falcon]

disregard what i said, i misread "everyone" as anyone

[pine tulip]

Just look at you name and make it longer

[bronze magnet]

ah same, they are slowly rolling it out as usual

[elfin rain]

It seems that my GPT-4 can no longer access the internet. Are they changing things in the background?

[bronze magnet]

yes they are rolling out a big update and everything is a bit wonky right now

[elfin rain]

cool, was getting worried 🙂

[vivid raptor]

I came here to ask that same question. Good to know I'm not going crazy.

[idle needle]

advanced data analysis broken for anyone else?

[sick pecan]

my local dev plugin is broken, and reinstalling dont fix it

[prisma raptor]

Hey all, I have a problem. I am a Plus member. The ChatGPT DALL·E seems to work only on my iPhone and iPad. I prefer to use ChatGPT on my PC's Firefox browser. When I try to use ChatGPT DALL·E in my browser the "send message" button doesn't do anything. What can I do?

Also doesn't work in Chrome

[pine tulip]

Nothing for now. We need to wait

[bronze magnet]

they are implementing a new model: GPT-4 turbo and a new mode where it does everything in one (dalle, browser, data analysis, plug-ins), the current beta features are being deprecated, plus tons of people who saw the openAI keynote or heard of the update are looking to try out the new stuff.... so yeah the service is a bit broken right now.

[prisma raptor]

Thank you very much for this important information. Do you happen to know if GPT-4 Turbo will be available to Plus members (the $20 a month plan)?

[zenith falcon]

yes

[bronze magnet]

yes it will, and should be coming to you with a new updated UI and other cool stuff (you can make your own chatbots and earn revenue from them on an store) any day now, they are doing a progressive roll out, some people already have access to the new stuff, others are refreshing and waiting

[naive locust]

anyone else have their old conversations searchable now in the app?

[bronze magnet]

you got that?

[naive locust]

i sure do. that was huge for me. i didn't hear anything about that in the keynote though.

[bronze magnet]

I seriously wanted a search function, that's so nice

[bronze magnet]

me neither that's great to hear

[naive locust]

just in my app though

its been there for a few days now

[bronze magnet]

hopefully it comes to the web version as well, the new web UI looks like the app actually.

[digital prism]

thats web design for you.. get successful enough and you'll be app design

[bronze magnet]

yeah hopefully it's less laggy/bloated, dunno about others, but the web interface was getting increasingly glitchy/laggy as they piled on new features.

the app on the other hand is super smooth, but less features

[digital prism]

personally its been great for me, considering my massive chat history

i've found the app to be buggy in looking at chat history tbh

[naive locust]

the new search feature in the app seems to work flawlessly. gpt on the web right now is erroring out pretty consistantly.

[digital prism]

the only way to solve this is a duel

my ai vs your ai, sundown

[naive locust]

haha!

love it

[junior imp]

Hi guys, is gpt 4 turbo available for plus subscribers now?

[zenith falcon]

its a roll out, for some yes, for others no

[brisk path]

Depends

[zenith falcon]

its random

[brisk path]

Cutoff date should consistently say April 2023 when you ask ChatGPT

Context window is still unconfirmed

[digital prism]

Consistenly... ChatGPT...?

[loud vessel]

Is there suppose to be a new UI today ? I see no changes at all on my chatgpt plus

[formal widget]

hi guys, is new gpt4 UI available for plus member now?

[elfin birch]

It is rolling out to Plus users, I have it right now!

[lusty island]

Did they get rid of browsing again?

[brisk path]

Well

[digital prism]

which god did you pray to

[brisk path]

10 regenerations say so

[elfin birch]

That's the funny part....

[brisk path]

I also have the new ui

[digital prism]

its satan isnt it

[elfin birch]

shh

[lime depot]

Anyone know if the new Assistants can use Dalle3 to generate an image from a prompt? Is that a tool it can handle?

[digital prism]

I think it can read images from its file system, let me check

[lime depot]

Thank you boss.

[formal widget]

thank you

[lusty island]

Did they get rid of browsing again?

[dense halo]

Did anyone lose access to voice

It says it’s not supported on default or something

[digital prism]

First guess is no not like GPT vision, it's listed as for code interpreter only, not the file retriever.. but maybe.. I'll try it out for my next assistant

[lime depot]

thank you - like, could you write it custom tools that use the Vision / Dalle3 APIs? You know what I mean

[digital prism]

Yeah I'm sure either assistant or GPTs will allow it

Dalle3 is confirmed, probably vision is too i bet tbh

[lime depot]

like assistants can work w vision rn?

@digital prism

[digital prism]

I'll try uploading a jpg then reference it

[storm coral]

trading expert? for create a script with chatgpt?

[proper forum]

I'm new to this server, but use GPT plus a lot. Do people struggle with GPT4 varying a lot in it's capabilities? It's been a big part of my work flow the last two weeks, and suddenly today it seems to have lost 20 IQ points on coding tasks.

It's always seemed to me it changes from day to day.

[digital prism]

ew it doesn't take png

who even uses jpg

[prime swallow]

They recently started releasing a very different model of GPT-4. It's probably better overall, but 'how it guesses' surely changed a lot. Anything that wasn't explicitly told to it, it could be interpreting differently now.

It just looks like the same AI, it's actually a very different one now.

[eternal rock]

This is a Safe for Work Discord Server

The OpenAI Discord is an actively moderated server.
• Refrain from sharing inappropriate content on the server. This includes but is not limited to messages, media, or other topics of graphically violent, sexual nature, and drug-related content.
• Report all sensitive and offensive content in the feedback reporting tool in the ChatGPT web UI instead of here on Discord.

Posting NSFW content will result in a permanent ban.

[prime swallow]

So I recommend, be prepared to adjust and explore a little, either directly tell it what you want, or explore ways to get it to guess what you want more accurately.

[waxen river]

The new chatgpt giving an error uploading Word and pdf files. Any idea why? image files are working fine.

[prime swallow]

Heyas. All is well?

[prime swallow]

#1070006915414900886 about that, hopefully fixed soon. Feel free to add to the reports!

[waxen river]

ok will do. I also reported on contact part of the openai.

[prime swallow]

Great!

[livid pier]

😕I haven't been able to access any new features again

Whether it's a new interface or a new feature

Unable to understand

[elfin rain]

Just reached Cook Sheep?

[rose glade]

When will the rest of the ChatGPT new features be available in the U.K.? Thank you.

[elfin rain]

Seems they are performing some big updates, so things are a bit wonky right now

[livid pier]

I just opened Discord

[normal quarry]

still rolling out

[loud vessel]

what new features are out today?

[rose glade]

Thanks!

[elfin birch]

GPT-4-Turbo, GPTs from OpenAI, new API models, All-Tools, and new ChatGPT UI

[flint zenith]

Hey, Dose GPT 4 still have the 50 messages every 3 hours or has it been increased?

[prime swallow]

same

[flint zenith]

ah ok, so it hasn't been increased.

[waxen river]

my new chatpgt screen doesn't list what is the limit per 3 hours.

[elfin birch]

I beg to differ! Testing shows it is perhaps 100? 😳
#openai-chatter message

[waxen river]

did they move it to somewhere else?

[livid pier]

"GPT-4 currently has a cap of 50 messages every 3 hours."

My old version of the UI still says this

[waxen river]

new version doesn't list it

it only lists "ChatGPT can make mistakes. Consider checking important information."

[livid pier]

But I did use over 50 messages at once on the day before Halloween

I'm not sure what the situation is right now

[wraith dagger]

One thing I really noticed with chatgpt recently is for long code it used to keep generating it all in one response. Now its being forced to say //this function instead of showing the code. It came more apparent after I asked for it to supply the whole file code in one response which it didn't after multiple attempts. anybody else noticed this? Could I fix it with custom instructions?

[junior imp]

Im curious that what the different is between alpha and gpt4 here?

[elfin birch]

Alpha refers to features that are experimental
GPT-4 is an LLM by OpenAI
Some experimental (alpha) features--in which certain people are randomly given access--involve GPT-4

[wind nexus]

Does anybody have access to custom versions of ChatGPT right now?

[elfin birch]

GPTs? Yes, though I cannot create any

[prime swallow]

Would be wonderful! Let me know if you can confirm or refute that.

Also, we all have slightly different forms of the models right now during rollout -

And we know some functions are not working right for some of us.

I totally wouldn't be surprised if some people have temporarily higher or lower limits, the same as the rest of the bugs

[wind nexus]

like a GPT Store?

[elfin birch]

Very similar!

[silk ocean]

I cannot send message to GPTs lol

[elfin birch]

I only have access to the ones OpenAI made

[wind nexus]

okay, maybe they are still rolling out new stuff

[glass harbor]

The platform is becoming multimodal. GPT-4 Turbo with Vision, DALL·E 3, and text-to-speech are all now available to developers. Is it available to regular (with subscription plus) users?

[silk ocean]

I can't send a message to GPTs, is it disabled or a bug?

[elfin birch]

Can confirm ChatGPT Plus users have these

[wind anvil]

New gpt4 model seems to have bing browsing as a default feature built in. Can use "classic" model if you dont want it to

[glass harbor]

but not for me 😦

[halcyon cliff]

Anyone know the timeline on the new agents feature?

[elfin birch]

Still rolling out

[wind anvil]

Openais gonna take over the world 💀

[digital prism]

I'm building my agent now

"Assistant"

[spare remnant]

Is the GPT-4 turbo better than the gpt-4 we had? Can anyone comment?

[glass harbor]

GPTs don't work as well?

[elfin birch]

They are a bit bugged atm

[lime depot]

"Note that the Assistants API does not currently support image inputs."

[eternal rock]

`` Server Tip: ``

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

[digital prism]

explains why it rejected everything

[lime depot]

quoted from the Vision docs

[digital prism]

docs? who needs docs. we got trial and error with super ai

[lyric sinew]

same

[waxen river]

Same here

[glass harbor]

I can't even go to https://chat.openai.com/create.

[jaunty leaf]

Is the new gpt4 turbo model only available for api access? Can it be accessed with chatgpt plus subscription?

[digital prism]

me either, but you can use the playground

[lime depot]

super ai?

[elfin birch]

After extensive testing over in #openai-chatter I can safely say that the turbo model is indeed being used

In all of its 128k context window glory

[brisk path]

I really wanna test it too

But again, doc uploading is broken for me

[lyric sinew]

My playground return too much requests...

[jaunty leaf]

I understand @elfin birch and thanks for confirming about the context length but is the model available for chatgpt plus subscribers?

[elfin birch]

Yes

I was referring to ChatGPT in that message, not the API

[hidden ibex]

Is the cap going away for 4- turbo?

[jaunty leaf]

As of now I access the models through api, if the turbo is available on chatgpt plus then I will subscribe again in a blink of a eye

[jaunty leaf]

Thanks again @elfin birch

[elfin birch]

🤘

[elfin birch]

The message cap seems to be 100 messages, again testing in #openai-chatter

[hidden ibex]

Hmm, hope they open up business class, I want to pay for unlimited

[rancid otter]

i still have 50 cap

[elfin birch]

Things are still rolling out

The 100 message cap was tested in the new UI with GPT-4-Turbo
#openai-chatter message

[trail juniper]

that’s api rate limit increase, not ChatGPT

[halcyon widget]

100 every how many hours?

[trail juniper]

OpenAI have multiple cache backend, so you may hit the limit over 50 messages

[halcyon widget]

3 or what?

[prime swallow]

50/3 hours is the stated limit

During rollouts, things can break. We haven't been told of any plans to change the 50/3 hrs though

[dense halo]

Why won't GPT talk to me no more?

[digital prism]

you did the bad

[dense halo]

big sad

[rugged barn]

still can't upload pdf.

[brisk path]

Same

[trail juniper]

New gpt4 all tools model quality is getting even worse, it can’t even do the example on openai.com/gpt-4

Explain the plot of Cinderella in a sentence where each word has to begin with the next letter in the alphabet from A to Z, without repeating any letters.

and gpt-4’s output are not even from A to Z 😓, while gpt-3.5 can do that now

[brisk path]

Works for me

[trail juniper]

try more, sometimes it fails

[brisk path]

https://chat.openai.com/share/34b14a5e-cb62-43b3-906c-6e18026ab062

[digital prism]

if the AI destroys us, its your fault

[trail juniper]

the first time I try it fails

[brisk path]

It works

Sometimes

First was correct for me though

[trail juniper]

https://chat.openai.com/share/031a92ea-f3e5-43ca-8486-2586846586c9

[livid pier]

Is GPT-4 turbo available in the Plus subscription program?

[warped pendant]

hours and hours wasted today by
There was an error generating a response

then more on top of that for placeholders, we need the 50 questions extended to 100 questions every 3 hours just so 50 of them can be to tell the ai to stop adding placeholders when its not having
There was an error generating a response, issues, 😦

[solemn pivot]

i cant click the enter button (send message), anyone else have that issue?

[warped pendant]

what happened to making gpt-4 better for developers

[brisk path]

https://chat.openai.com/share/34b14a5e-cb62-43b3-906c-6e18026ab062

[dense halo]

GPT4 voice still doesn't work 😦

[fast zodiac]

Hey ! Anyone successfully accessed the https://chat.openai.com/gpts/editor feature ? Because it does'nt work for me

And, a friend of mine have a new UI for chatGPT, but i don't x)

[waxen river]

uploading docs still not working

[dense halo]

I hope they fix this soon, i'm getting lonely

[fast zodiac]

The voice works for me on the android app

[ripe patrol]

I think they really raised the context window in ChatGPT (I'm a plus user btw)

I asked a question and it remembers something I said roughly 25k tokens up, near the beginning

I'll see if I can find anything longer to test but this is still huge compared to what was there before

[eternal geyser]

Mine just says 'You dont have access yet' when I try to go to it directly

[fast zodiac]

Same ! Maybe they faced an issue and they had to disallow us to access it, but if this is the case, they could tell us x)

[eternal geyser]

We have given them the benefit of the doubt many times now - to not 'say' that ALL ChatGPT Plus users 'now' have all these features. When it isn't true - also, it is obvious from the last rollout, they do have a 'turn on for all' button, like they did on the 'last' day of their last 2 week rollout - they hit that button and everyone got all extra features.

[worthy dragon]

Isn't 50 msgs / 3 hours still less.
You get unlimited gpt-4 on bing (with 50 msgs in a conversation)
I hope they provide unlimited usage soon

[trail juniper]

Then the whole world share one unlimited account🤔

[eternal geyser]

They should improve on this process though - otherwise may build animosity from people calling it 'unfair'.

[worthy dragon]

Maybe a concurrent threads per user limit to 1

[trail juniper]

but its not for now

[uncut crystal]

Is there anyway around fixing the speed at which chatgpt-4 responds. And is anyone having any luck at getting it to remember rules for specific prompts?

It's always forgetting what I told it the prior input.

[worthy dragon]

?

[wraith ocean]

i have been expereiencing this as well.

[viscid condor]

You all complain too much

[fast zodiac]

Maybe after their official announcement and presentation, they encountered a problem and had to backtrack temporarily, but there is a lack of communication. Moreover, we are not the only ones experiencing the issue, so they must have heard about it. Communication that they are investigating the matter, as well as an estimate of how long we will have to wait before accessing the features, could be useful.

[fast zodiac]

👍

[trail juniper]

I pay 20$ for it

[viscid condor]

20$ is nothing for what GPT-4 is

[worthy dragon]

I will get my plus subscription once the gpt store's out, and everything's going well

[fast zodiac]

I'm pretty sure it will be fixed soon

[brisk path]

Bing is NOT real GPT-4

[worthy dragon]

It is gpt-4 with web access for real, they have even put a slider to select 4 easily

And there are surely visible differences in my testing, it can do things which are impossible for gpt-3.5

[rancid otter]

tru. they say it's powered by dall-e 3 but its not the same model imo

have you tried MS Paint new integrated cocreator? (Bing creator)

[worthy dragon]

... Like it can chat in Ancient Sanskrit, which is basically impossible with 3.5, and needs a significantly bigger model to generalise to such a niche task

[spice star]

it may not have the same 8k token limit that chatgpt does though

or at least a lot of it is taken up by search results and other functions

[jaunty robin]

Chatgpt doesn't have 8k, gpt4 through the API does

Although I guess I can stop saying that now

[brisk path]

Wrong channel

[brisk path]

It doesn't seem as smart

At all

[spice star]

neither is chatgpt with bing browsing quite frankly

[brisk path]

And Bing Vision is... Horrible. Even Bard does better with Lens than Bing does with vision

[worthy dragon]

Idk, but it is surely tuned for web surfing, which makes it seem dumb, because most of the time it just returns the search results

[brisk path]

Fine tuning can absolutely make a good model dumb

[brisk path]

But again this is inexcusable

[spice star]

not really, it's free

[worthy dragon]

Holup, are you saying bing-vision is different than gpt-vision or bard is just better than gpt-4v

[spice star]

it's the same as chatgpt vision, just potentially lower fidelity or something

[brisk path]

Bing Vision is different

Completely different, and worse than both

[worthy dragon]

probably, to help save computing resources

[loud quail]

It’s honestly baffling that a 20 dolls sub AI has this high rate of errors. it’s like 4 out of 10, knowing that it takes some time to provide the answer and/or browse

[short badger]

Who also has a problem where ChatGPT does not know how to generate images or execute code in one chat, as promised?

[spice star]

that is why it's a beta feature

[dense halo]

Guys when do u think we'll get voice back?

[eternal geyser]

Try the API for Voice

[normal quarry]

I still have voice, the position of the headphones icon moved on iOS

[dense halo]

too complicated for me

[loud quail]

The chat is not a beta feature my bro and i get the same amount of failure

[dense halo]

mine tells me to "continue with the default model" and when I click continue nothing happens

[plucky timber]

Run failed
We're currently processing too many requests - please try again later.

Tfw paying for access but there being no pre-access for paid members and your in some arbitary queue for the entire night 😄

[sterile forge]

I’m a little confused, is both gpt 3.5 and 4 updated to include info up to April 2023, and not just gpt 4 turbo? Also is turbo not available yet?

[dense halo]

the default 4 is now turbo if you have the update i assume

[brisk path]

3.5 seems to be January 2022

[dense halo]

Guys does the ios app look different for you guys now?

[sterile forge]

I’ve been looking everywhere and can’t find any definitive answer

[brisk path]

No, but only because I use Android 😊

But yeah the UI seems to be different

Wasn't exactly the complete merge with the web UI I was hoping for but idk

[dense halo]

iOS is based

[brisk path]

Based in a walled garden without sideloading unless you want a dev account which costs a lot of money

[dense halo]

i'm a casual user, so it's fine for me

[brisk path]

Fair enough

But you can't really call it based when it just

Isn't

[fast zodiac]

Just received that answer from the live chat (the answer was not really in live but..) about the /gpts/edit path

Hi there,

Sorry you're running into trouble with ChatGPT Plus!

We recently made a number of improvements to fix generic errors like "Something went wrong", which were often caused due to using a VPN or occasionally by some browser extensions. If you were running into problems like that, we recommend trying again (and trying in an incognito window to ensure browser extensions aren't interfering). Our servers can occasionally get overloaded with high traffic or other incidents, so while errors like this should go away after a short while, you can always check our status page to see if we're already aware and fixing it.

While using ChatGPT Plus, it's helpful to keep in mind that our AI models have limited knowledge of world and events after 2021 and may also occasionally produce incorrect responses. Although we have made progress in this area, there is still more to be done. If you find a response from the model to be incorrect, harmful, or biased, we would appreciate it if you could let us know by clicking the thumbs up or thumbs down next to the response. This helps us track which responses need improvement as we build increasingly safer and more capable systems.

If you're still running into problems, it'd help if you can double check that you've tried all of these steps; knowing if the issue persisted after each step helps us pinpoint what's going on:

Try clearing your browser's cache and cookies
Disable VPN (If you are using one)
Use an incognito browser or a different internet browser/computer to see if the issue still persists, as a security add-on or extension can occasionally cause this type of error.
Disable cookie blockers
Try connecting to another network

If none of that worked, could you send us more information about the steps you took leading up to the error (such as anything you clicked on), as well as details about what happened when you tried the steps above? More info about the error message itself is helpful too, especially specific wording or a screenshot. We'll take a look and do what we can to fix it with you.

Finally, if you have any other questions along the way, it may be helpful to search through our common questions about ChatGPT and account billing for more detailed answers.

[normal quarry]

please don't paste walls of text

[fast zodiac]

I feel it could help some people if I shared the only information I have, so I apologize i guess

[novel parcel]

I dont have access to gpt

[untold seal]

Guys I’m confused when will the gpt4-turbo model be in the normal chatgpt website. Because for me it still got the shorter context window and the drop down menu for the different use cases…

[simple brook]

Still unable cant use DALL-E in the new GPT

[violet bloom]

Still here for both models on Android.

[untold seal]

U already got the new chatgpt Interfaxe or are u using the api?

[simple brook]

New GPT interface

[nova fable]

It wont make an image when you ask??

[simple brook]

No I mean via Dall-E GPTs or whatever they called it

Can't send it any messages

[violet bloom]

Same... I got a notification that I I can now use one chat for all. I have the data cut of April 2023, but no live search, no Dall-E3 and no analytics, and the max prompt is still 2k. 😂.

[nova fable]

They are merged into one chat now. You cant ask for updated todays news or for it to make in image?

[simple brook]

It does work on normal GPT just not specific GPTs

The ones in the explore section

[violet bloom]

Yep, started new chat.... Asked for today's news. And got: my data ends 2023, I can't search, use a news portal.

[nova fable]

Oof

Im having trouble uploading documents to it right now. Worked earlier but getting upload error now

[fading cloud]

please help how to use pdf reader? please

[fading cloud]

can you use pdf reader

[novel parcel]

when can I expect the gpt feature

[violet bloom]

No, I can only select pictures, no pdf files... 🙄

[novel parcel]

same

[spice star]

patience, not everyone has it yet

[midnight cedar]

Does anyone know how to upload an image in Hot Mods?

[wanton gazelle]

I feel left out T_T

[dense halo]

what is "hot mods"

[static wharf]

Yep so the community was right

Gpt4 turbo, fast but seriously dumbed down

[midnight cedar]

It's a GPT created by openai

just released

[dense halo]

probably because its turbo

[cursive path]

They killed my boy 😦

[zenith falcon]

me too, got none of the really cool stuff

[cursive path]

It’s like useless for coding now. Like I’d be willing to wait 10x as long to just get it back to what it was a month ago

[wanton gazelle]

None of the new stuff yet for me

[wanton gazelle]

You can always use the API

[dense halo]

Is GPT4 still smart?

[zenith falcon]

yes

[cursive path]

That’s true, it was just nice to not have to worry about message history

Just sucks to have something good taken away again after they took og GPT 3 away from us and gave us the abomination that is turbo

[wanton gazelle]

That's true, then again, I think they were having performance issues in the last few days. I'm sure it's their way to help "scaling"

[loud vessel]

I still see no changes on my chatgpt plus at all

I am in Canada

[normal quarry]

it's all rolling out over coming days, be patient 🙂

[foggy magnet]

So how many people have access now?

[normal quarry]

more than 1 people

[foggy magnet]

I've seen screenshots with people having access, I knew this would be weeks-long rollout. A shame but it is what it is.

I hope one day OpenAI can do launches instead.

[wanton gazelle]

Given the rollout history, more like: between now and 1 month

[foggy magnet]

Coming days with how many features they announced is crazy. Took me like 2 weeks to get DALL-E 3

Christmas present for some at this rate.

[steep tundra]

Literally says that ChatGPT users get access today...

*plus users

[pure otter]

Yea not working for me either, unfortunately 😭

I got the notification that it’s now available but it doesn’t work

Even in the IOS app it says it’s available and doesn’t work there either. I hope this gets rolled out (fixed) soon.

[plucky timber]

Me : today in what timezone

[violet bloom]

Yep...

For gpts for example;

Example GPTs are available today for ChatGPT Plus and Enterprise users to try out including Canva and Zapier AI Actions.

[quick ridge]

Yes

[quick ridge]

I haven't got access

[steep tundra]

It's like they didn't even bother proofreading their main article

[dense halo]

Any ETA?

[pure otter]

This is the message I get:

ChatGPT can now browse the web, analyze data, and generate images.
These capabilities are now built into GPT-4, no need to select them. Available for Plus users.

[dense halo]

article mentions by wednesday

[violet bloom]

Same! But doesn't work. 😂

[pure otter]

Forget the article. It says it in the app and it still doesn’t work

Why update the app with messaging for features that aren’t released yet or buggy. You would think they would release the features and messaging at the same time.

Amazing stuff though and can’t wait to actually use all the new features

It sounded like you could in their YouTube release notes update

[cursive path]

its a christmas miracle, i got data analaysis back

it instantly had a way smarter solution for a coding problem i had. i spent the last hour trying to figure this out

[shell patio]

yeah, visit playground Asssistants tab

[cursive path]

if anyone from openai is reading this, can we compromise? Im cool with turbo for everything else but can we keep data analysis as the old 8k context gpt 4

[spice star]

gpt-4-turbo seems just as capable in ADA mode

[silk flame]

Hello, I have a problem with the new version of ChatGPT+. The first one is that I cannot upload any documents. It stops in like 75% of uploading and nothing happens.

The second thing is that when I try to use custom GPTs, I cannot send them any messages After clicking the button send, nothing happens.

Anyone has the same problem?

[spice star]

everyone seems to be having a slightly different problem

[outer seal]

i didnt think turbo rolled out to the cgpt app yet, did it?

only (allegedly) folks with alltools

or do you mean via the assistants api

[cursive trellis]

Do the new custom gpts straight up not work for anyone else? Both the submit button and enter do literally nothing. The only way I’ve been able to use them is to click one of the suggested inputs and continue the convo on the mobile app

[cursive path]

Can I coax it into ADA mode without uploading a file first? I’ve found ADA is a lot better suited for me at least for what I use GPT for. I wonder if just saying at the beginning like “I need ADA mode” will do the trick

[zenith falcon]

I think I have turbo (ik I have 2023 training data) and I don't have all-tools

[spice star]

all tools is separate yeah

[cursive trellis]

Gpt is also unable to use plugins or browse in the normal mode. With plugins it’ll just print out the function call JSON as a normal message and gpt claims to not be able to use browse on all tools

[zenith falcon]

I can't even switch my model on mobile

[cursive trellis]

Ig I have the new assistants api to play w/ in the mean time but it’s kinda a shame nothing seems to be working for me

The gen z meme cgpt is embarrassing lol

[loud vessel]

still refreshing endlessly and logging in and out

[late rose]

In ChatGPT's new updated version, the response limit has been increased? As earlier, 50 responses were given.

[cursive trellis]

@late rose I believe so the interface gives no indication of a cap. But I also never hit the 50 messsge cap even when using it heavily

[late rose]

That would be also biggest update

[cursive trellis]

Did you hit it often? @late rose that’s like what a message every 3 minutes for 3 hours

[late rose]

Not yet after the update

[subtle herald]

Dose somon know when will be the app maker function of gpt-4 available to the users in the eu?

[versed echo]

the limit has indeed been removed

[outer seal]

i had ADA tell me it was trained up to september 2023 last week 😆 cant trust gpt for gpt info

[silk flame]

are you sure?

[versed echo]

Previously, I had a limit of 50 messages every 3 hours, now I've used over 100 and still have no limit.

[spice star]

there might still be some kind of rate limit, but it's much higher if so

[versed echo]

Yeah

[nimble coral]

trying to use the new custom models but i can't send any chats. Is it bugged?

or is it just for me?

[dense halo]

for everyone

[nimble coral]

ah lame

[dense halo]

patience i guess haha. maybe theyll have it sorted out today

[nimble coral]

yeah i'm just excited haha

[dense halo]

I’m not able to use OpenAI’s ChatGPT Builder, but I’m too excited to use it when it does show up. Not sure if it was published to accounts yet.

[plucky timber]

I wonder if they actually did region rollout

If so i've 100% been wasting my time waiting the past 6 hours

[novel parcel]

same

[short badger]

Is there a person who has a ChatGPT of all models in one chat?

[novel parcel]

?

[dense halo]

They didn’t roll it out for me and I’m based in USA.

yes I have all models now

[plucky timber]

Alr ty for confirming your co-suffering

[novel parcel]

Till end of the week everyone should have access

[dense halo]

I tried the link on the website, signed into my account, and it said to me, you don’t have access to this function.

Something like that anyways.

[novel parcel]

same

[subtle herald]

I seen it on reddit. In r slash ChatGPT slesh s slesh 14wBw20tlC

[dense halo]

I watched the Dev conference on YouTube, but there wasn’t a lot of information. I thought it was active to everyone yesterday technically for me.

[plucky timber]

Yeah same

[plucky timber]

Lots of stuff on reddit tho

[dense halo]

I tried to message OpenAI support via email, no response yet. Maybe I’ll get one today sometime.

[plucky timber]

Mines on Seen 2h ago with support lmao

They saw it and left me hanging 😂

[novel parcel]

https://help.openai.com/en/articles/8554397-creating-a-gpt

[dense halo]

I’m just excited to see what the power of having your own modification to the GPT can do. Like enhance it to be smarter.

[molten vault]

quick question wasn't the UI of the page supposed to change with the model selection going away?

[plucky timber]

Enhance it to not be completely derp 😂 , I hope the file upload will allow for more consistent replies instead of it starting new topic/projects every 5 responses XD

[dense halo]

Yeah lol.

I think that the full GUI is suppose to change via this update. Who knows. Still the same issue with the link. “You don’t have access to this.”

I would like to have GPT be smarter at coding without saying “I can’t do this.” You know what I mean? It would be nice to get a decent response and code.

[signal apex]

so the all models in one, is use gpt4 turbo?

[plucky timber]

I've added a custom instruction to never reply with basic code examples and always go for the advanced code instead
And a few other things that seemingly work a lil better to get decent code replies

[violet bloom]

yes

[plucky timber]

Also one to never do // your example here

[signal apex]

can you share you prompt pls?

[dense halo]

Would be nice for to never let you down lol 😂.

Guys i have a question for those who got voice back on ios can you still use it for old convos?

[plucky timber]

All i have in the better response section is this

"I want quick advanced code replies. No basic examples.
I want to keep modularity in mind.
Apologies are excuses I hate AI that constantly apologizes.
You test your code before you reply.
Don't make unrequested changes.
No // do this, if you can write the comment you can write the code.
"

The no apologies is a godsent ngl

[plucky timber]

Right 😂

[signal apex]

hahaha, yeah, thank you

[dense halo]

I told my GPT to always follow my custom instructions

[plucky timber]

GPT : Let me change your code logic without you asking 🏃‍♂️

[jolly temple]

Whats this new gpt update? Is it actually usedul?

[full veldt]

I wonder if this will get rid of its bad habit of saying "To do X, [insert various steps here]. Shall we?" instead of actually doing what is asked in that first output

(Granted this is less of an issue if the 50 input cap has gone but point still stands)

[plucky timber]

Nah, still does that often, but it does stuff like this often tho : Do you need a more detailed script for any particular part of this system, or shall I create a Unity Editor script that allows for image uploading and color-tile mapping definition?

As in it will ask to do stuff automatically

[dense halo]

I really hope that this new update will change all of that for us. Hopefully will be able to make ChatGPT models that are much better and suited for our needs. Plus, I believe it’s allowing us to upload documents and different resources to it to be able to use. I’m not sure what the limit size for documents and things are yet, but I hope that will be able to use it very soon and see what it can do for us. Especially since we’re paying money to use this and it’s not as we all desired. so I’m really hoping that this update changes things for us.

[plucky timber]

Playground seems to be max 20 files

The other thing i cant access so no idea

[dense halo]

I’m not even able to use playground because I don’t pay for playground usage. I only pay for ChatGPT, so that’s why I’m waiting for that model to come out personally. Because I’ve tried to use the one in playground and it seems to be giving the errors every time I try to run it.

[pliant citrus]

Uuuugggh, I have been using GPT since the release a year ago, been a Premium user for so long. I WANT TO PLAY WITH GPTs AND THE AUTOMATICALLY CHANGING ONE 😢 😂

[rancid seal]

Is it still 50 messages every 3 hours?

[plucky timber]

Yeah it errors every single time for me too, I can only create the assistant 😂

[dense halo]

I know I’ve been a premium user for a long time as well and I can’t wait for that.

I’m like why can’t we even use the model in playground if it’s going to allow us create it

[shrewd prism]

They run Plus like a Patreon tbh. They add stuff if they're feeling cute today sorta thing

[dense halo]

It just tells me every time that I try to use ChatGPT create, that I don’t have access to it.

[pliant citrus]

Well, still they also say "We are rolling this out now." and then I see my friends sign up for premium and instantly get the new stuff, while I'll sit there waiting

[plucky timber]

This should never happen yea I side with you on that one, if someone signs up today , back of the queue

[shrewd prism]

Yea I'm waiting too. Maybe they wanna make sure the White House doesn't make a bioweapon with it before they launch it to everyone? Who knows lol

[plucky timber]

Makes you wonder, if you unsub and resub will you get instant access?

[dense halo]

Not only that it’s when I’m in playground. It says every time run failed. We’re currently processing too many request. Please try again later.

I’ve already tried unsubscribing and resubscribing, and I didn’t get any change in it.

[shrewd prism]

Nah, you can't terminate your subscription before it naturally expires

[plucky timber]

Fek 😂

[dense halo]

Yeah, that’s true too captain.

[pliant citrus]

Yeah...
Worst part is that I really love using this stuff and use it for school, programming etc. But then someone who sits there and is using it to write a song or something just for the fun of it will sit there with their premium stuff 👀 😂

[late fractal]

Hi, when is the marketplace launching and how can i sign up for launching my own GPT model?

[dense halo]

They said that later this month they were going to launch the marketplace, but for creating your own ChatGPT model, seems to be very close. I’ve been a premium user for the longest time, and I don’t even have access to it being in the United States as I am, plus they said on YouTube, after the developer conference that all subscribers would gain access to this and that was never true. It was supposed to be the day of. That’s what I got from it.

is this company sabotaging itself why is gpt 4 not on api for everyone yet

[white shard]

There is no specific date, but it will roll out in the next couple months I assume

[late fractal]

Thanks for the info. When you write premium user, what do you mean? gpt plus customer or enterprise? The reason I ask is because I want to publish my own gpt models and sell them on the marketplace. But no one is able to build models yet? I know the marketplace isn't live yet, but I thought they would invite prompt engineers to create models prior launching the marketplace, so that they have some models when they launch - do we know anythin about that?

[late fractal]

^^

[dense halo]

Premium user meaning that I am a ChatGPT plus user.

[foggy grove]

Sam said later this month 👌 so I presume later this month,

[sleek swift]

when are gpts rolling out

[dense halo]

I know it’s later this month for the marketplace but what about building your own ChatGPT models?

Yeah, I haven’t really heard an exact answer on that from Sam with the developer conference.

[white shard]

gpts have already started rolling out but not everybody has it yet, please be patient! It will be rolled out 🙂

[foggy grove]

That’s should be out with all the other stuff over next couple days

[pliant citrus]

That's what I heart as well. That is what makes me so darn ticked off.

[sleek swift]

It would be rather useful to have it today. If it allows building knowledge, it could solve my romanian homework for tomorrow much easier. That's why I'm asking.

School just sucks :(. Especially in Romania

[white shard]

I totally understand, but unfortunately they can't just release it to everybody at once, they have to do a gradual release, but rest assured things will be available soon!!

[sterile glacier]

They are a new paying client, they have to have the latest stuff, we are addicts already hooked on the product, they know we wont leave.

[dense halo]

I hope things will be available soon.

[pliant citrus]

Why does it seem to be so random? Been a premium member since Premium was launched, been using GPT since November last year. Still I always seem to be the last to get stuff 😦

[foggy grove]

Because it literally is completely random

[dense halo]

I know how you feel.

[loud vessel]

Who can create gpts? Devs or any chatgpt plus user?

[white shard]

Any user for whom that feature is rolled out and available for

[pliant citrus]

I am so hyped up to try it out, been wanting to do so for so long. However, I also wonder. Is it just the same as giving custom instructions?

[violet bloom]

the thing is, as the rollout happens.... they should not post on official communication, it will be availble to ENT + Plus users "today".....

[white shard]

Well, gpts have access to their own custom tools defined by openapi schema, can access the tools of code interpreter, retrieval, browsing, so it's definitely more extensive

[sterile glacier]

anyone here know when they plan to increase the output length. because getting a max of 500 ish words is too little.

[white shard]

I understand the frustration

[pliant citrus]

Noice!

[violet bloom]

no frustration here... based on our past experience, it's just: "keep calm and drink coffee".

[white shard]

The best mindset to have 🙂

[sleek swift]

How hard would it be than to integrate a custom tool than? Is it needed to apply for some special permissions before doing this?

[white shard]

Pretty easy to integrate a custom tool, custom tools are defined with an API spec using a commonly known schema, so you have to build an app that serves that api schema and can perform the functions, and then you can load that schema into your GPT/assistant and it will be able to use it 🙂

https://nla.zapier.com/api/v1/dynamic/openapi.json as an examplem this is the schema for the Zapier ChatGPT plugin, putting this schema into a tool would allow it to be used!

[sleek swift]

That's nice. The only missing thing is that now I have to find a server to host an api on to bridge the gap between the expected format for GPT and another API

[frosty valley]

Hello hello all. Thanks for having me here in advance.

[white shard]

yea, that is something you'll have to do as part of using custom tools, a gateway that allows for OpenAI/GPT to call your tool

[sleek swift]

It would be nice if GPTs could run python code on custom functions. That would be the best way to do it. Also related to python, is the code processing happening in the browser sandbox? I noticed over the last few days that every time I ask GPT to process some python, my CPU usage skyrockets.

Huge DB processing for stats

[frosty valley]

Terribly sorry to interrupt your good conversation, but I've been wondering if someone from the comunity knows if it is possible to use finetuned models in the new Assistants functionality (my finetuned models don't show up in the models selector in the Playground > Assistants environment)

[plucky timber]

Is there any more info about the rollout, as in who goes first, is it totally random , is it region specific.

[normal quarry]

it appears random

[white shard]

No further info apart from it appearing random

[plucky timber]

alrite

[white shard]

this is speculative but I don't think the computation happens on your browser

[sleek swift]

Yeah but then how would the jump in CPU usage be explained

[white shard]

Might have been coincidental with something else, but that is a fair point, we should wait until they release some more info on the inner workings

[sleek swift]

I also noticed that sometimes if I close my laptop when python runs it will crash out, but not all the time

[white shard]

That's mostly due to the way that ChatGPT sends responses to you

when you close your laptop the network card goes inactive and stops streamed responses from coming in back to your browser

there has to be a constant network connection between you and chatgpt when it's responding to you

[sleek swift]

I mean it will crash fully and after a page reload

[loud vessel]

can gpts learn new stuff from documents you upload? And then others can prompt it and get information that normally chatgpt wouldnt have?

[white shard]

That's right, not exactly "learn" new stuff, but it can "retrieve" content from the documents that you upload to answer your questions

[loud vessel]

ah thanks and very nice

[sleek swift]

So it isn't fine tuned on the documents?

[white shard]

No, documents are stored and indexed with the assistant and content from the documents are retrieved and used at query-time

[loud vessel]

Is there a limit to how many documents it can handle? Can each document be 128k tokens?>

[cursive trellis]

are the cgpts supposed to be unusable? Like have I just not gotten access yet or is it a bug?

[white shard]

I think the limit is a file size limit, 512MB if I'm not mistaken, but double check the documentation online

[white shard]

Access is still rolling out, not everyone has access yet

[sleek swift]

Still it should behave better. I'm going to make one GPT with a bunch of good competitive programming books to use it for understanding how I can improve my solves of problems

[white shard]

this is for assistants actually, not sure about document within GPTs but I'd assume its similar

[white shard]

yeah for sure it'll behave better, it'll behave better than fine tuning tbh

[frosty valley]

it's filesize specific. 512 MB each and for assistant a max of 20.

[white shard]

fine tuning is more fitted for adjusting conversational style

[loud vessel]

thanks

[cursive trellis]

When I select like chatgpt classic the enter key/button does nothing. Only way I have been able to use them is to select one of the suggested replies and continuing the chat from mobile

[sleek swift]

Oh damn. I have more than 20books

Would it be possible to throw a ZIP instead?

[white shard]

No zips

[cursive trellis]

Weirdly enough I can also edit the suggested replies.

[frosty valley]

You don't need to make the books file unique, it won't treat the data as unique per upload anyway. Although you'd probably be better off aligning the GPT4 AI to stick with the domain you're training it for (programming). Unless you want it to absolutely stick to your own books.

[sleek swift]

There was a rise in such responses last week. Idk what's going on

[white shard]

It might just be under heavy load today, but otherwise try a different browser and clearing your cookies and cache

[frosty valley]

Just the network being overwhelmed. Normal behaviour if you're on American peak times.

[sleek swift]

Mine are specialized books with advanced competitive programming algorithms

[frosty valley]

Also some conversations are harder to interpret and respond to than others, especially code-heavy conversations seem to.

Okay, so? Those algorithms are probably already known in GPT4

[sleek swift]

It knows them, but I want to make sure he has access to a verified data source

[frosty valley]

With the new documentation upload, you can just have your entire code base aligned with GPT4. Just saying.

[cursive trellis]

No dice. Even tried logging in on an old computer. Maybe I just gotta wait.

[white shard]

yeah just gotta wait

[frosty valley]

In my next development sprint I'm probably going to create a bot which develops by itself

[sleek swift]

I'm doing something similar for a project.

[white shard]

are you gonna open source it? would love to see it in #1037561385070112779

[frosty valley]

nah. I write bad code and suck at being good at stuff. So I'm too emberassed

haha 😉

[white shard]

I'm sure your code is fine!!

[sick pecan]

success is trying failure until it doesnt work

[white shard]

look at some of my projects I definitely have a ton of garbage spaghetti code everywhere

[frosty valley]

faking it got me here so far, but without GPT I wouldn't be able to write more than a handful of javascript html css and script language lines

[sleek swift]

It would be nice if GPTs could behave like AutoGPT

[sterile glacier]

there is a futurama joke in there somewhere

[white shard]

AutoGPT isn't really anything special it's just a chain of prompts and self-asking

[sleek swift]

I know that. Would it be possible to implement that?

[white shard]

of course!

programmatically with an assistant using the assistants API it would be really easy to implement autogpt

you can do it easily even without that, and with just langchain

[sleek swift]

Oh yeah, but I want it as a GPT

[cursive trellis]

Ive tried makign stuff like auto gpt and I have never really been able to get anything useful out of it. Even basic self questioning and trying to force it to plan out its responses I wasn't able to pick out the chained inputs from the reg ones

[sleek swift]

I would also pay extra for more requests in ChatGPT

[white shard]

I'd imagine it's still possible with some clever custom functions

[frosty valley]

If you wanna see some code which basically wrote itself, you can check out my tetris game at tetriblocks. The design is me, but the logic is completely written by ChatGPT (and debugged for centuries by me).

[shell kiln]

Anyone else still have the old UI?

[sleek swift]

Bunch of people

[frosty valley]

I also have a sudukupuzzle website, and a competitive brickbreaker app. But they're too embarassing. Oh and 3 chrome extensions floating around

[sleek swift]

I got only a small update on the phone app

[white shard]

Ah fair sometimes it just comes down to strict prompting, I was able to have success by ensuring strict output conformance using output guardrails

[cursive trellis]

those sound really cool lol @frosty valley im sure its better than my code

[white shard]

anyways this is chatgpt discussion i'm getting off topic

[pine tulip]

Any idea when all the stuff is going to be fixed? Still can't properly use ChatGPT

[white shard]

Lets move this over to #dev-chat to keep chatting about the api 🙂

[frosty valley]

why, is your GPT more advanced than mine? 😉

[sleek swift]

Probably a couple of weeks.

[pine tulip]

I am not ready to accept that 😢

[novel parcel]

I dont have acces to gpts

[frosty valley]

Anyway to reloop why I came in here making a fool of myself, does anyone know if you can link a finetuned model to assistant? In playground my finetuned models don't show up with assistants.

[novel parcel]

na

[frosty valley]

nah u can't oh nahdunnah ? 🙂

[pine tulip]

na in nananananana batmaaaan

[dense halo]

please no, my mental heath is already impacted

[topaz ocean]

Sooooooo, how to get access to the new chatgpt that some people are already using? 🤔

[frosty valley]

Gitgud

[cursive trellis]

Just gotta wait. Roll outs are pretty random. I can’t even run plugins any more lol

[frosty valley]

or be patient idk. They seem to prioritise people who've been a paying API member before everyone else and commercial before private

Usually also US over rest of world, but we have access already.

[topaz ocean]

Yea but even in the API I dont have access to the latest models

[cursive trellis]

I used to get new features way early but I forgot to update an expired card(for a plus payment not the api) and ever since then it’s been taking forever

[pine tulip]

Can't the Mods pin a message with "PLEASE WAIT"?

[sleek swift]

I mean fixing the bugs that will be found

[cursive trellis]

You sure? Does it give you an error?

[dense halo]

i'm hoping they fix the ios app by tmrw

[sleek swift]

What's wrong with it? The android one works well

[topaz ocean]

@cursive trellis This is the latest model I have available: gpt-4-0613

[white shard]

api rollout is also still rolling out!

[solemn pivot]

Whenever I try to use the custom "GPTs" like "Data Analysis" and "DALL-E" and "The Negotiator," I can't click the send message button and enter doesnt work. However, when I use classic GPT, i can send the message. anyone have a fix or experience the same thing?

[past fiber]

Any idea why I still dont have anything they said and why I dont have accesse to gpts?

[dense halo]

gpt4 is having issues, voice doesn't work and it's still seperated into bing an dall-e

[sweet kindle]

hi. is it only me that have issue with a selected GPT, I do the inital message but then I cant reply? The msg never fires away, either with pressing enter och clicking the send message button, none of them work.

[white shard]

For everyone that doesn't see GPTs or any of the new stuff, it is still rolling out, please be patient

[solemn pivot]

I have the same issue, pressing enter doesn't work and clicking the send message button doesnt work either

[lapis knot]

I havent found any advantage with the new. I was hoping we'd see more memory or capacity, but I'm not

[past fiber]

Any idea how long?

[white shard]

No, but rest assured they're trying to roll it out to everyone and it'll happen in due time, a fair amount of people have access already and it is being still being rolled out to more

[sweet kindle]

yeah really strange, I hope it gets fixed.

[dense halo]

patience is hard

[topaz ocean]

I smell someone who has access @white shard

[past fiber]

XD

[shell kiln]

mods pin this please

[eternal rock]

This is a Safe for Work Discord Server

The OpenAI Discord is an actively moderated server.
• Refrain from sharing inappropriate content on the server. This includes but is not limited to messages, media, or other topics of graphically violent, sexual nature, and drug-related content.
• Report all sensitive and offensive content in the feedback reporting tool in the ChatGPT web UI instead of here on Discord.

Posting NSFW content will result in a permanent ban.

[violet bloom]

Sorry, my 🔮 is in maintenance due-to usage.

[white shard]

LOL

[past fiber]

😂

Is there a reason why USA gets openai updates first or is it random?

[pine tulip]

the fix is to wait. Doesn't work for a lot of people

[sleek swift]

Damn that sucks. I still have them separated on Android, but voice input got an update. And everything works as normal

[white shard]

OpenAI is based in the US, rollouts can be better controlled and overseen within there

[dense halo]

Yea mine keeps saying voice isn't supported on bing search despite me being in default

[white shard]

I'm in Canada and we're right up there and we don't get features as fast as the US so trust me I understand

[pine tulip]

doesn't matter, they rolled out worldwide

[dense halo]

I'm in australia

[white shard]

yeah they did do this one worldwide, it was previous ones like the gpt-4 api that started in the US

and a couple other things I forget

[sleek swift]

Yeah, speaking of rollouts my chatgpt account is made in the US and paid with a US credit card, but I still get the slow rollout from Europe where I actually use the account

[plucky timber]

I waited a full minute it aint there yet what are the suggested steps

[white shard]

repeat

[plucky timber]

🫡

[eternal rock]

`` Server Tip: ``

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

[pine tulip]

how would you even know in which "rollout category" you are? I think you are just beeing unlucky

[plucky timber]

US priority should never ever be a thing in the first place lmao

[sleek swift]

Repeat like ADA does until you get it

[marsh urchin]

Hi everyone, I'm a plus user and obviously I pay a monthly plan, but I noticed when looking at my invoices that when I signed up I paid $24.40 on February 11, 2023.
Then on March 11th, June 1st and July 1st I still paid $24.40.
Then August 1st $20.00 and I don't understand why it's less but that's okay, then September 1st, October 1st and November 1st again $24.40. I would like to understand why from 20.00 USD then back to 24.40 USD? Let me be clear, I am happy to pay, but I would like to understand if possible.
Thank you.

[pine tulip]

it isn't a thing with this update

[white shard]

Hi we can't help with this, please contact support through https://help.openai.com

[past fiber]

When the vision gpt came out I thing i was the last person on the list I had to wait the whole week and check it everyday and still nothing. On the last day I got it. I was happy but still the week was not fun

[sleek swift]

Might be sales tax

[white shard]

yeah understood, but global rollouts are super hard and there are a TON of people that use chatgpt

many people were in the same boat as u I'm sure

[marsh urchin]

Thank you

[plucky timber]

Honestly this can be so easily solved

[past fiber]

Yeah

[static wharf]

gpt-4 turbo sucks

[plucky timber]

You add a simple counter on the chatgpt page, indicating your place in the queue

🤷‍♂️

[marsh urchin]

Maybe because over time it has become more powerful and consequently you pay more?

[white shard]

yeah but then you have to host a page that millions of people have to be able to stay connected to and receive updates at all times

that's computationally, difficult

lots of little things to think about

for what it's worth

[static wharf]

yea its simple but they can't be bothered

[white shard]

tons of new features on the announce day and the fact that so many already have access is pretty great

So they've done a good job with this IMO

[plucky timber]

I mean yea sure the end result is tho yall getting spammed 😄

[white shard]

I'm sure everyone will have access very soon

[violet bloom]

post examples, or never happened.

[white shard]

it's ok i don't mind

[pine tulip]

You wouldn't even need that. You could just send everyone a mail with their position in queue by 11am every day lol

[sweet quartz]

you sure it's so many ? 😞

[plucky timber]

Gmail : thats spam

😂

[pine tulip]

Well well!

[static wharf]

its called simply using it and seeing the degradation

[sleek swift]

Nah. I have been paying the same price for ever. It must have been an error in the system

[sweet quartz]

i'm still racing to be the last to get all the features !

[pine tulip]

Don't worry, I already have the new features and most of them aren't properly working. It's more annoying than helpful

[sleek swift]

Now that I think about it, except the Dalle rollout, I got everything in the first day or so

[past fiber]

If the gpt4 turbo is more efficent with more tokens why openai doesnt replace chatgpt4 with turbo?

[ionic thicket]

Anyone else having trouble uploading files to ChatGPT?

[past fiber]

Lucky

[pine tulip]

because it behaves different and some people prefer it over turbo

[sweet quartz]

i mean, having them at least wears off your constant checking if you have them or not, most likely they'll be fixed.

[sleek swift]

Every new chat is on turbo by default

[past fiber]

But 124k of tokens would be neat

[pine tulip]

so what?

you can still select either gpt4 or gpt4 turbo in the api

[ionic thicket]

They just stall at 90%, regardless of size

[sleek swift]

API is API. Chatgpt is something totally different

[pine tulip]

Yea and? turbo in chatgpt isn't rolled out to everyone yet either

he/she was asking why they don't just replace it, which would make sense if it's more efficient

[sleek swift]

Are you sure?

[pine tulip]

Yes

Some people still have knowledge cut off 2022 instead of 2023

[past fiber]

Yup

[past fiber]

Oh thats where i am lucky my gpt is latest version 2023 april

[pine tulip]

[static wharf]

You can tell if a model is gpt 4 turbo vs gpt 4 by asking it to make a poem that doesn't rhyme

turbo gives a rhyme, gpt-4 gives a non rhyme

[modest shale]

lol i woke up this morning to try it at 4am, and this came out right around then. perfect timing

[sleek swift]

Nah, ChatGPT is using a smaller token count

[pine tulip]

🤏🏽

[past fiber]

Yeah but it would be neat if they made it bigger

[pine tulip]

for what, except processing long texts?

[novel scroll]

is the new chatGPT not using the 128k? is that only for api?

[pine tulip]

only api

for now

[past fiber]

Does someone know if there is limit file on how much gpts can handle?

Yeah if I have few pages it will most of the time forget and i have to give him the instructions again

[pine tulip]

Yea right, but besides working with very long form content, there isn't any downsides, right? I just never use it that way anyway, thats why I am wondering

[violet bloom]

Sure? If I understood Sam correctly, the stuff is in ChatGPT too.

[past fiber]

If you are just asking question yes it is enough but for people who works with longer texg and need to modify or summery..... Its not enough

[novel scroll]

you know why the new model does that, it seems weird

[past fiber]

No the turbo with 128k is not there

From what i heard

[static wharf]

because gpt-4 turbo is a new model, different from gpt-4, probably less parameters.

[pine tulip]

AS I said, I know. Just asking if thats everything lol.

[past fiber]

There has to be downside if its more efficient and you have 128k tokens and its cheaper

[static wharf]

yes of course, its cheaper and faster

but I don't want cheaper and faster

I want intelligence

[pine tulip]

wait for GPT-9

[past fiber]

Yeah i am already waiting for gpt5

[azure dagger]

gpt-9 will be installed in your brain directly

[static wharf]

yes I imagine they are training or have already trained gpt 5 by now

though they might need to restart after the new CogVLM paper

[past fiber]

I thing they are finished but i thing it will be out start of the next year

I hope

[past fiber]

Whats that?

New technic for ai learning?

[static wharf]

state of the art visual language model

released yesterday

not better at language than gpt 4

but its actually properly multimodal while gpt4 is just cheating

so its vision is soo much better

[dense halo]

why is gpt4 cheating?

[past fiber]

But i sont thing they will just retrain gpt5 just to have this

[static wharf]

the way gpt 4 multi modal works is it converts images into tokens

so its more like gpt 4 is using a description of an image

[past fiber]

Workaround

[static wharf]

na, i imagine it would need to be from the ground up

[past fiber]

Yeah and also just the training is really expensive

Just gpt4 was more then 100mil just for training on azure

[pine tulip]

yea not hapenning

[dense halo]

does azure use nvidia graphics cards?

[static wharf]

cog vlm is state of the art with deep fusion

while gpt4 and every other multi modal is shallow alignment

[past fiber]

Yeah everybody wants those cards these days

They are the best for training

[pine tulip]

they even use nvidia graphics cards with the sole purpose to train ai's

[past fiber]

I thing its a100

But i would need to check

[static wharf]

In a few years hardware will be way faster and way cheaper

but for now its very expensive

already they are building AI cards that are faster than H100

[past fiber]

Oh yeah h100
True

[past fiber]

Well to build those cards is not that expensive just nvidia charges a bit more

For them

[pine tulip]

it's very expensive if you don't have the same infrastructure as nvidia has

thats why they can charge more

[past fiber]

Yup

Up to 10x

Of the original price