gpt-models

[warm crystal]

cutoff knowledge to october 2023

[void kayak]

yeah you're right, looks like it's a playground limit

real limit seems to be 4096

[lament shell]

shouldn't 4o atleast know its architecture?

[warm crystal]

not really, the models dont normally know who they are

[lament shell]

GPT 4 and GPT 3.5 both know they are their respective model

[warm crystal]

they've been around much longer

[lament shell]

true

[void kayak]

playground now lets me do 4095, so I guess it was a minor oversight 🙂

[hollow mulch]

Thanks! Haven’t edited my GPT will debug then!

[fervent mirage]

Yep.

[thick violet]

May have been asked but it seemed like with the new capabilities of 4-O my once blocked behind plus GPT could work under the free model. I almost heard her say that in the presentation but want to know if that's the plan? I have several people who want to use my GPT's but don't have pro.

[zealous shale]

Yes, custom GPTs are now free

It's time for GPT creators to add a monetization model behind the scenes for improved features.

[toxic moat]

I also think it is worth the money, how do you get the same model for free?

Omni is available for free?

I don't know, I am very impressed with 4o as of 2 hours of using it

not sure if for free users it will respond that fast

[cursive ocean]

Wuts the usage limits for the free tier?

So about 16

[zealous shale]

The Plus that we get for our money has so far provided us with much more than free. Now that they are giving THOSE features to free tier, Plus can look forward to OTHER features. Right now with most of it in harmony the question of Plus is mostly legit, but I believe Plus will continue to accel away from this soon.

[cursive ocean]

I'm mostly just curious about how they were able to make the model run faster. Cause I'm assuming they didn't make the model smaller.

[zealous shale]

"where soon all features will be offered for free" those are the current features, now, what about next week or next month? We don't know. that's what we're paying for.

That was kinda the case when we first started paying for this stuff last year. The only benefit was a better model, then more features were added to plus. They're following an established pattern.

[hollow mulch]

So if free tier gets 1/5th shall we assume free users can only use published GPTs 8 times in a 3 hour period?

[zealous shale]

Hmmm, I think the calculation of transactions is not related to the model/prompts used. So whatever the limit is, it's probably the same limit no matter if you're using 4o, or @customname from 4o.

[dusky venture]

is gpt4o available in the app for the plus users?

[cursive ocean]

For me it is

[zealous shale]

Yes, in Android. But voice features aren't fully enabled.

[dusky venture]

Oh, so the shown model in the videos are not available yet

That talks and sees Real time

[austere peak]

how do i know if a custom GPT is using 4o or regular GPT4?

gotcha, thanks,, so IS 4o BETTER in logic/coding overall performance than regular GPT4 or not? And if yes, how is it possible it has more chats within a 3h window?

[thick violet]

I assume this is one of the "In the coming weeks" as mine still says GPT+ Required when I visit the share link in incognito

[austere peak]

great, seems its a bit better overall. although "model that can reason across audio, vision, and text in real time." that phrase kind of limits its upgrade to just those capabilities, i wish it was more

[primal seal]

Do new (or older) gpts use 4o?

Nevermind, I see this is being asked often

[lucid turtle]

sadly not yet big rip

[final acorn]

take this with a grain of salt but it seemed to be doing worse in independant benchmarks

[narrow pilot]

if I subscribe to plus now will I immediately get access to gpt4o or will I need to wait?

[lament basin]

Not overly impressed with the GPT-4o chatbot. Probably A BIT better on some points, but still overall underwhelming (based on our of course very high expectations). But I can't seem to be able to use their API with this - I generated a new project key (over 30' ago), I have plenty of money left on my account, but I get a 'RateLimitError: Error code: 429 - {'error': {'message': 'You exceeded your current quota, please check your plan and billing details" error. Any suggestions anyone?

[gilded current]

I had the same issue a couple of weeks back, im pretty sure the fix was adjusting the key's permissions

[lament basin]

Thanks! I'll look into that

[quasi kindle]

Why can't I use the voice feature? anyone know

[rain tinsel]

will be released in the coming weeks

[quasi kindle]

Ahh, it hasn't been released yet thankyou

[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.

[toxic moat]

GPT4 >>>> GPT4o (in coding, after 6 hours of using GPT4o), but I guess thats why shown benchmarks are not on codeeval

Still it will be interesting OAI will release new functions maybe then I will use it, for now I will probably stick with GPT4 when it comes to coding

[sour salmon]

Will the custom gpts we created auto switch to 4o or will they still provide results in 4.0?

[thorn fable]

I was wondering exactly the same thing. I'm wondering if my Custom GPTs will automatically switch up to 4o when available.

[lament basin]

Still no cigar... I did change all permissions (to full permissiveness 🙂 ), but I get the same error. Just wondering though - is this the right syntax?

                # Generate the response using OpenAI GPT-4.0
                response = 
client.chat.completions.create(
                    model="gpt-4o-2024-05-13",
                    messages=messages,
                    max_tokens=max_tokens,
                    temperature=0.7,
                )

[gilded current]

hm, looks good to me

[gilded current]

are you using the new Project API keys? I just noticed they exist now

[lament basin]

Yes I am...

[gilded current]

Maybe try and test with a new user key then, sorry forgot you mentioned that before lol

I could've sworn a new key generation was all I did to fix my project, I know adding money & billing info werent the real issues

[lament basin]

Do you mean a new USER key or a new PROGRAM key?

[mental burrow]

4o is really fast

[oak sapphire]

4o is honestly insane

I mean video and audio is going to be huge, and it seems like there’s increased limits compared to 4

[lean jay]

They've completely redone the playground, but still don't support adding system messages...

[static isle]

This looks like it to me #off-topic message

[lean jay]

Yes. But you can't add one.

The models support system messages anywhere in the chat. They're no different from assistant and user messages. But the playground only allows having a single one at the beginning.

[spice ice]

I’m watching the presentation on YouTube. I am not impressed. am I the only one that wish they would increase accuracy and allow us to use it more without message caps? That to me would be extremely valuable. I feel like they keep adding gimmicks. In the end, we want to get accurate intelligent information. Who cares about audio? What a waste of resources I think!

[teal parcel]

did anyone find that the knowledge for gpt 4o is only updated to 2023 October ,while GPT 4 is updated to 2023 December

[hollow zenith]

Hello I am having some issues right now with GPT enterprise, what is the quickest way to contact support? this would only be related to enterprise gpt

[teal parcel]

and now what's the good for the subscriber ,only more message limit ,it seems not enough

[hallow owl]

you need to speak to your organisations admin

[hollow zenith]

that is I

[hallow owl]

then you would have the support details

[hollow zenith]

where do I find these support details

[hallow owl]

in the emails you received while you were being onboarded

[hollow zenith]

can you specify which one there were quite a few...

only one I can find so far is the one that points to the help center which so far has been useless

there is no direct message through there

[hallow owl]

I don't know, we're not equipped to handle individual support here

[hollow zenith]

hmmm any advice on the fact I am paying 60 bucks a month using my company's finances but I am getting this message when trying to generate a few images.... GPT enterprise should definetley not be blocked by this limitation. "It looks like we're generating images a bit too quickly. To ensure the best experience for everyone, we have rate limits in place. Please wait for about 2 minutes before we continue generating more images. This will help manage the system's capacity and allow us to continue creating high-quality images for you.

Feel free to let me know if there's anything else you'd like to discuss in the meantime!"

idk about y'alls company but 60 bucks a month is very much a lot to mine. so if I can't even create multiple images then why am I paying so much when I could use the cheaper plus model or even the free one at this point?!

[unborn heath]

Hey hi! We're community members here, like yourself, however we can share our experiences and the official docs.

If you go to https://help.openai.com/ at the bottom right is an icon, looks like a smiling box in a black square.

If you click that smiling box you open up some options, bottom middle is 'messages'.

Click that, and select 'ask a question', which is probably near the bottom and clickable. At this point you are still working with the bot, but you can click your way through. I recommend, since you say it's about Enterprise:

Click ChatGPT.

Now, you're going to see an option for 'Enterprise', but if you click that it just offers you the link to speak to the sales team. You may want to talk to your sales team, if I paid for Enterprise and I had an issue, I absolutely would be talking to my sales representative that set everything up for me. You should know who that is through your onboarding emails, and have that contact.

However, you can also attempt to reach out through that initial page again: https://openai.com/contact-sales/

However, presuming you choose to work with the open help website instead of your actual sales representative that already has been helping you...

Click ChatGPT Including Plus

At this point you have options of bug reporting and other choices - as I tracked the options out and typed this, I had this wonderful message appear:

"Hello,
​
Thank you for reaching out to OpenAI support.
To better understand your concern, we would like to ask if you can provide the following:
​
1. A description of the issue
2. Steps to reproduce it
3. A screenshot of the issue/error message
​
Once we receive your response, we will further investigate. You can also visit our Help Center for potential solutions to your issue.
​
Thank you for your patience and understanding.
Best,
Reymar
OpenAI Support"

Yes. A human reached out to me before I could even finish asking for help (I wouldn't have, just getting info for you)

[wraith ginkgo]

`` Server Tip: ``

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

[unborn heath]

If you're paying $60 though... chances are you are 'team', not enterprise'. As the minimum seats for enterprise are 150, and probably costs far more than $60. So perhaps follow the path through the help website through ChatGPT for teams and get help that way. But there's people waiting to help us there, one of them reached out to me when I was typing to you about how to ask for help, that's how caught up they are at this second. Other times it's taken weeks for me to see a reply, it varies hugely.

[hollow zenith]

Ah yes apologies, then I am a “gpt teams” not enterprise member.

[hollow zenith]

Alright thanks I’ll try this when I’ve got the time thank you for replying

[unborn heath]

Again, I'm just a community member, I can't control or change anything, but....

We all, teams or plus or even free users, we all share the same resources. When a lot of people are trying to use the resources at the same time, the amount available for each of us drops, sometimes far, far below the normal limits.

This can happen when you're trying to make images, if a lot of other people are trying at around the same time.

There's really no workaround or answer that I know of, other than to wait those 2 minutes. You can try with another account, if you have one (That actually adds to the total community the services are supporting and actually adds to the delays for us all, but hey, it's on OpenAI to handle that, not me personally and you can decide for yourself what you do; me I choose to have 1 and only 1 account, and I just accept the limits when they tell me I gotta wait - every person has their own situation, needs, and choices to make).

But I've sometimes seen that happen, 'dang it, it's been 3 days since I last made any images, I make 4 today and already I get told I have to wait before I can make more!' - but they're helping around 100 million unique accounts a month. If all 100 million of us, or even 1 million of us happen to decide to make pictures all at once, apparently that may cause unusual delays.

[hollow zenith]

Please understand I am not blaming you. I am fully blaming OpenAi on this. If they make different tiers, and they have a difference between “plus” which is meant for enthusiasts and “teams” which is meant for small companies like mine then they should definitely be treated differently. I am paying so much money (to me it’s a lot) and when I’m working I shouldn’t need to take breaks when I need info or creations from a tool I am paying 60 a month for. I upgraded because of this. I was a normal user and the cap was too small for discussions, then then the plus version was too small so I upgraded again. My work is a mix of written text and coding so it takes a lot of messages. That is why I am currently disappointed that even with the amount I am paying, I am still being made to wait.

[unborn heath]

Well, definitely you can reach out to them. Hope you can still enjoy the text and coding messages even if you're having to wait for the images, and wishing you the best!

[hollow zenith]

That was a suspiciously quick response….

[unborn heath]

laughs Your point? Your concern? Don't you have work to do? 😛

Some of us type fast. Some of us care. Some of us are whatever we are, whoever we are.

Good luck.

[hollow zenith]

😂 yes I do. Was a joke

[round sequoia]

Questions:
I pay for plus. It says techincally I can start using macos app, but where do i go to download it?

[unborn heath]

I would presume where you would get any other mac app/ios app. There is the official mac ChatGPT app, and I don't see a second one.

Here's what the official help website says about using it, I don't see any mention of how to go find it, so they must think that's obvious:

https://help.openai.com/en/articles/9275200-using-the-chatgpt-macos-app

Additional related helpfiles listed here: https://help.openai.com/en/collections/9403039-macos-app

[chrome needle]

I know gpt4o is quite limited for free users, but can chatgpt read the code file in file . z i p?

[wind urchin]

Why zip ... might as well encrypt it

[high hound]

does anyone have problem with macOS chatGPT app too slow and lagging? unlike the one from my mobile app, its super lagging and too slow.

[unborn heath]

I don't have any mac device, so I can't even test. However, I recommend #1070006915414900886 and let them know, so they can check and try to help!

You can also reach out to help and tech support through https://help.openai.com/ the bot bottom right, open it, click messages, click 'ask a question', then select through the choices. Most choices lead to a way to report a bug or leave feedback, and that's how you can reach out to a person who can help.

[high hound]

thanks a lot

[shy briar]

Hey guys, quick questions for my GPTs... I can't select any models (like when I just use ChatGPT and then select for example ChatGPT4)... Is there a default for GPTs or does it adapt to my subscription plan? Or is there something I missed?

[shy briar]

Okey, good to know. Thanks❤️

[wind urchin]

why can I still not log into openAI???????

[eternal night]

Will GPT-4o be able to work with sound to identify and generate noises, and give reactions, lessons and feedback for things like sound design and music composition, arrangement and mixing?

There's so many new things to do and try and experiment with and wonder about now that we have unified multimodal AI

[blazing portal]

woah, woah

[quasi flint]

I have a free account, I had 4o access but then I logged out and back in again and now I lost access

[dim grotto]

Hi, I have created a custom gpt called OptimusGPT, can some one be kind enough to provide some feedback on that ?

[grizzled flax]

Hey! You can share custom GPTs over in #1171489862164168774!

[dim grotto]

ya sorry I thought this channel was about custom gpt's

[bitter terrace]

does the playground have support for the voice stuff yet on 4o?

i wanna try it but i havent gotten 4o on chatgpt yet

[calm frost]

do our custom gpt's use 4 or 4o

[grizzled flax]

Hey! It doesn't sound like it yet, no: https://openai.com/index/hello-gpt-4o/

We plan to launch support for GPT-4o's new audio and video capabilities to a small group of trusted partners in the API in the coming weeks.

[bitter terrace]

alr, ty

[sour salmon]

So ChatGPT is not launching the full 4o features they demoed yesterday for the mobile app yet , it’s going to the api first ?

That blows

[bitter terrace]

i wish there was a way to give it live data

[tough current]

In regards to the 4o, is it 'smarter' than 4?

Am I more likly to get a better answer from 4o than 4.

[dapper furnace]

That's going to depend on your prompting, but from personal experience I switched mid project to 4o and the results are amazing.

[grizzled flax]

It's coming to ChatGPT too! From the same link as above:

We'll roll out a new version of Voice Mode with GPT-4o in alpha within ChatGPT Plus in the coming weeks.

[mental burrow]

4o seem better

but recall that tokens are significantly cheaper

and the model is also significantly faster

[sour salmon]

Thanks for clarifying

[bleak crypt]

For me, 4o reads like reskinned 3.5. Am I the only one?

[quartz flint]

i think that's partly because of it's speed

our brains wrongfully assume that because it's "slower" it must be thinking or working harder than other models

[bleak crypt]

No, mine is superfast. And behaviour is very similar to 3.5.

[quartz flint]

that's what i mean

gpt-4, at least according to the benchmarks isn't smarter than gpt-4o in most cases

[bleak crypt]

Also insists on making very spread out lists and repeating same point in a same answer.

[bleak crypt]

I am saying that 4o is acting the same as 3.5 with the same limitations

With being just slightly slower than 3.5 and what appears to be just a tad smarter, if

[grand blade]

Can I choose GPT-4o for my custom GPT?

[storm raft]

Hello everybody!
Are there any benchmarks out there between these models?
Gpt-4o v/s Gpt-4-turbo?

[zealous shale]

The state of custom GPTs seems to be in the air at the moment. An existing GPT can't use new 4o features yet. I'm sure they will bring this into alignment soon.

FrAnX, see "Model Evaluations" -- openai.com/index/hello-gpt-4o/

[storm raft]

MVP, thanks

[zealous shale]

Not yet but Thank YOU. 🙂

[storm raft]

you are for me 🙂

[fickle jetty]

anyone knows how to use video recognition to check my appearances or the room

[grizzled flax]

Hey! The updated voice experience from the demos yesterday isn't live yet, it's set to launch over the coming weeks: https://openai.com/index/hello-gpt-4o/

We'll roll out a new version of Voice Mode with GPT-4o in alpha within ChatGPT Plus in the coming weeks.
You can use the existing image recognition feature by sending a still image though, for now!

[fickle jetty]

oh ok thanks

is there any part in this server for internships ?

[zealous shale]

"GPT-4o’s text and image capabilities are starting to roll out today in ChatGPT. We are making GPT-4o available in the free tier, and to Plus users with up to 5x higher message limits. We'll roll out a new version of Voice Mode with GPT-4o in alpha within ChatGPT Plus in the coming weeks." -- openai.com/index/hello-gpt-4o/

[strange nimbus]

Will all existing GPTs be switched over to GPT-4o model automatically? Or will GPT builders get to choose between GPT-4 and GPT-4o for their custom GPTs?
Also, does anyone know when the memory feature be extended to custom GPTs - i.e. when will each onehave its own memories?

[zealous shale]

Someoen reported that there was a difficulty with migrating an existing GPT to 4o. I'm sure this will get worked out and announced.

Dunno about memory - would love to see custom per GPT, yeah.

[thick violet]

I don't know how to migrate. There's no option for model, chatting with the GPT Builder doesn't hint at it. Disabling advanced features like Web Browsing, Code Interpreter still require sign-in and GPT + to use in an incognito window.
I think they haven't rolled it out yet. At least I haven't seen GPT's that don't require plus yet.

[zealous shale]

Right - there is no option yet. We need to wait to see what they will say. They might say there will be an auto update.

[strange nimbus]

Europe is decel, over-regulating

do you need a cookie consent every time it decides to log a "memory" ?

[humble frost]

whats the message cap for GPT-4o plus users?

[jagged inlet]

80 per 3 hours

[humble frost]

can you use gpt4o with the temporary chat?

[jagged inlet]

That, I'm not certain on tbh

[naive verge]

For me, temporary chat only works with GPT-4

[paper hamlet]

for me, the headset icon for live chat simply disappeared 😦

[humble frost]

do we know when the desktop app will roll out?

[silent veldt]

Guys, do we have 80 messages for gpt4o + 40 gpt4 (120 total) or 80messages shared between gpt4 and gpt4o ?

[devout bramble]

was wondering the same

[versed canyon]

I think it's a combined message limit based on my experience YMMV

It appeared as if each GPT 4 message in the same thread is double-weighted compared to the GPT 4o

For storytelling work, I found GPT 4o a little less consistent in its approach. It likes to throw a lot of content policy errors if two characters hug or worse gasp kiss. I'm all for moderation, but I feel like the GPT 4o moderation bot was written for Ms. Mallem's 4th grade class.

I take back what I said about GPT 4 and GPT 4o having some sort of combined counter. It seems like they are indeed separate as I could get two different limits for each in chatgptplus

[wind urchin]

@rigid estuary Is the voice ChatGPT is down ?

[honest peak]

it seems

[wide zinc]

hey guys i want you help

[lunar harness]

Whats up bro

[wide zinc]

so, how i can use the new gpt 4o?

i have accesed before , but theres nothing on pc

but in my cellphone theres the version there

is it still in processes the uptade until reaches all countries? or nah

[jagged inlet]

Are you a free user, @wide zinc ?

You'll have to wait if that's the case

[wide zinc]

ohhh

yeah

i forget bout this

[jagged inlet]

It'll be rolled out over the next few weeks

[wide zinc]

okay dude, thanks for the help.

[jagged inlet]

No problem! 🙂

[vale root]

hey guys anyone has noticed huge delay after 4-5 prompt follow up on 4 model and 4o on desktop browser ? the app on iphone 15 is blazing fast on responses

[jagged sable]

is gpt-4o speech only available on phone and not browser?

[grizzled flax]

The voice+vision experience from the demos yesterday hasn't rolled out yet--it'll be rolling out in the coming weeks.

[lunar harness]

aye Mac app is avaialble to me

[cursive kiln]

seems like gpt-4o is very lazy at analyzing the documents, I upload 3 times, and 3 times it asked me to upload documents, and then I reach the limit 🥹

[glad geyser]

do customer GPTs always use the latest model or does that need to be updated in the GPT settings?

custom GPT's, not customer GPTs

ah, gotchya. Thanks

[humble frost]

I dont like how you cant reply to parts of messages and can @ different gpts in the desktop app

and you cant edit images

[lunar harness]

dam the desktop app isnt working for me

it just says "Invalid Request" when ever I try to use it

here I was all excited

[granite mural]

oh my

[wind urchin]

is the chatgpt mac app more just like a web app? i don't expect it to take much resources

[static isle]

Looks like it's natively built in Swift

[willow bane]

Anyone know when free users might be able to access my custom gpt? https://chatgpt.com/g/g-n7UgV6cOn-cquni-eba-guide-v2

[toxic merlin]

Do custom GPTs that were previously built automatically use GPT 4o - and/or is there a way to specify what model they use?

[humble frost]

when should I use the desktop app and the website?

[static isle]

From what I have seen, the app is generally more convenient and soon to be more featureful: easy screenshots, prompts one hotkey away, and "speech mode" that can see your screen (coming soon)

I personally like how it looks too

[humble frost]

it's just a shame we're missing out on so many features like reply to messages, mentioning multiple gpts in one chat and editing images

[static isle]

I assume you have the desktop app?

In any case, thank you for this information.

[humble frost]

yeah it's rolling out to plus users

[wheat marlin]

How do you 'get' the app if you're a plus user?

[wet star]

random selection, a little popup for download, having the dmg early from someone else won't work

[wheat marlin]

Thanks for the information

[slender badger]

Do GPTs now use GPT-4o or do they still use GPT-4?

[wind urchin]

nobody seem to know but many ask haha

[static isle]

GPT-4 at the moment

At least for me

[wind urchin]

where did you see that?

[static isle]

#ai-discussions message

[wind urchin]

but from an openai person or just a guess?

[static isle]

I literally cannot use GPT-4o with GPTs

[wind urchin]

how do you know? i dont see where a gpt say what model it is using. maybe we can metaphorically use them

[static isle]

The model selector

As it appears underneath a message

[wind urchin]

mobile? i dont get such on desktop

[static isle]

Please follow the link in the message I attached

[wind urchin]

just link to you say something about 80 messages

[static isle]

No but that message is nearby as well

Scroll down a bit from there

This was tested on desktop

[wet star]

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

GPTs for paid users are currently powered by GPT-4. GPTs for paid users will switch to use GPT-4o in the next several weeks.

[static isle]

bob coming in clutch 🙏

[wind urchin]

dont let that green leaf fool you haha

this guy know he stuff

[sullen girder]

man's first act is saving the day

impressive

[wet star]

😂

[wind urchin]

well i will have to be patient then before make some big new gpt i guess

and we should all bookmark that since so many ask haha

[sullen girder]

gonna just copy and paste my GPT instructions to the start of a GPT-4o conversation for a bit

won't be very effective but I think that's worth having o

[wind urchin]

i assume the dalle gpt is then also run on gpt4 not 4o. it is just nice to use some time for two image instead of one, but i also converse with it

i hope they update copilot to 4o too. i got a sub to make extra 100 dalle image a day

[wide zinc]

hey guys

i'd like your help . using the paid version of Chat gpt plus, you also can have more Up to 5x more messages for GPT‑4o, is this ilimited ?

or theres a certain limited in there of how many messages you can text in the paid version.

mark me for i can see after

[sterile saddle]

Hi guys! ^^ Is this the correct place to ask something about GPTs?

[wide zinc]

you can do it this in #community-help

[sterile saddle]

Thank you very much 🙂

[bleak crypt]

I am in EU. I asked 4o about small code with loop. I stopped it once because it was making a long spaced out bulletin list explaining the code.
After getting the answer I tried to provide feedback and ask gpt 4o on how to find a middle ground between paragraph and long spaced out lists.

What happens now is gpt4o is just rephrasing the second answer about code no matter what I say. It is like it doesn't understand at all any subsequent promt.

It took around 6 prompts asking to stop talking about loops until it acknowledged it, and only acknowledged it, didnt answer actual question. There is something really off with this 4o.

[toxic merlin]

Thanks!

[thorn light]

Yeah this is similar to the echo issue I experience when I want 4o to act as a writing assistant or editor.

It will read back my rough draft to me verbatim rather than improving the grammar or creatively writing in any way. GPT4 is miles better for this use case. GPT4o is unusable for it.

[bleak crypt]

Thank you for your reply. What do you find 4o useful for?

[thorn light]

It has helped me shake off the rust and remember how to write a discord bot and find a place to host it. I haven't had a lot of other successful use cases with it yet but it's still new.

I've also noticed it likes to name our conversations using different languages for some reason.

[bleak crypt]

gpt4 did that for a while, for me. Gpt4o used several English words in translation to French .

[kindred swallow]

Yo how can I get gpt 4o on andriod it is not updating for me

on both computer and on phone

[frigid abyss]

I also need gpt-4o

[wind urchin]

if you are not a sub you just have to wait, it is roll out and so many use it now that it may take a little more time i guess

[shrewd flume]

hi there! just saw the amazinng chatgpt 4o launnch and I'm trying to llive speak with the bot but I can onnly have a text to speech interactionn, am I doing something wrong or is it a soon to be released feature? I only have a free account, is it a only paid feature maybe? thx!

[grizzled flax]

Hey! It sounds like the new Voice Mode will be rolled out to Plus accounts, but the base GPT-4o model will be accessible for Free accounts too, just not in the full Voice Mode experience from the demos: https://openai.com/index/hello-gpt-4o/

We'll roll out a new version of Voice Mode with GPT-4o in alpha within ChatGPT Plus in the coming weeks.

[shrewd flume]

oh I see, thx!

[wind urchin]

Does anyone know if Custom gpts now use Chatgpt GPT-4o?

Or still just GPT-4

I would like to make a custom gpt that uses the new GPT-4o model

[supple violet]

"GPTs for paid users are currently powered by GPT-4. GPTs for paid users will switch to use GPT-4o in the next several weeks."

[wind urchin]

Thanks

[thorn light]

Patience. Time.

[astral loom]

I am still not seeing GPT-4o as free, I thought GPT-4o was free, is this not the case? or are they rolling out to all users over time?

[jagged inlet]

Rolling out to all over time

[astral loom]

ok, thanks

[jagged inlet]

Paid users get it first but it'll trickle down to all eventually and np

[shadow apex]

in the future you can just check by asking a custom gpt to write something relatively long. if its taking too long its gpt 4 turbo

[azure herald]

I mean I still don’t have 4o, but now my voice options have gone 😂😭

[shadow apex]

its going on and off lately for some reason. Just keep in mind its the old voice option still

[azure herald]

Oh yeah I know, I was just messing around with it, then it vanished lol

[shadow apex]

yea it did for me as well yesterday on android. i think its due to multiple people trying to check if its the new voice feature and its causing them issues / or they are just testing stuff who knows

[azure herald]

It won’t even let me subscribe to plus now

App just freezes out and says failed

[shadow apex]

thats weird its fine for me

check for an update maybe

[azure herald]

I’ve checked about 5 times today, uninstalled and reinstalled twice

It’s just decided to nope out on me

[dusky venture]

I cannot see the gpt4o, is there a list of available countries anywhere?

[near oriole]

anyone know if gpt4 has a higher word count compacity than the 3.5? or is it the same?

[grizzled flax]

Hey! LLM context limits are actually measured by things called "tokens", not words. https://help.openai.com/en/articles/4936856-what-are-tokens-and-how-to-count-them
Are you wondering about GPT-3.5 and GPT-4 on ChatGPT specifically, or on the API? And do you mean their capacity to remember, or something else like how much they can output at a time?

[willow bane]

Awesome, where did you see this? 🙂

[grizzled flax]

It's at the top of this help article! https://help.openai.com/en/articles/8554407-gpts-faq

[shadow apex]

The current situation with gpt 4o is just a mess

• 4o but without image generation from 4o (dalle instead)
• custom GPTs do Not have 4o
• 4o but without voice from 4o
• at least 4o with multimodal image recognition

[gloomy lake]

Anyone else noticing that the training data cut off is different each time you ask. Yesterday when the model released it knew about the latest version of Angular (Code Framework) Today it has no idea.

[willow bane]

Legend thanks!

[quartz flint]

the reason they released GPT-4o so early was to cut off Google's IO event

they only have the better model and better image recognition

it's clever but it's scummy

[hearty bramble]

voice not working on my end

[mental burrow]

it is clear that 4o is the new free tier and that something better than GPT4-turbo is coming

4o scores under 4-turbo in some important tasks

[wet star]

I feel like the goal was to have 4o approach GPT4-turbo in functionality as much as they could while being cheaper to run, and add more modalities or whatever, which leads to (hopefully) a qualitative change with the new ways of interacting with it... if I was gonna make a guess I'd say we get something more like "GPT 4p" down the line before "GPT 5"

[bitter terrace]

4o is very impressive considering costs

Its replaced google assistant on my phone - i just need to add some more functionality to the app i use

[mental burrow]

concur

they probably took lessons learned from newer models and took a smaller than gpt4 model trained on more than gpt4 was trained on

phi3 is quite good for being so small, had a very large training corpus

it seems pretty straightforward that the next thing will be a gpt4.5 ish LLM that is just better than 4-turbo because $reasons, or some agentic thing that isn't really "GPT5" but probably more like a swarm of gpt4.2s or something

but maybe i'm wrong maybe gpt5 will come out this year and make Data from star trek look like a noob

[covert swallow]

So.....if my memory data getting too many infos? would that effect the speaking level of my gpt?

I feel it the same principle as instructions of modified gpts. Am I partially correct?

[high hound]

gpt-4o so slow and won't complete answer right now.. maybe because of the traffic. No wonder early test scores for 4o is not high enough..

[buoyant yacht]

does gpt models run 4o?

[left hull]

Has anyone used the ChatGPT desktop version?
I received a usage notification today, but after logging in, I got a message saying the Team Edition workspace is unavailable. Logging out and logging back in gives the message "You do not have access to this app yet!"

[woeful raft]

As longer my chat gets (and it must not be long), inference gets slower and slower until the model (most cases i tried with gpt-4o) completely stops generation... mid inference...
Is this normal for now?

[slate kindle]

lmsys arena rankings updated
gpt4o top 1

[kindred minnow]

I've had that happen in the mac app, but not on the website.

The mac app is really bad with long conversations, I think. It uses 100% cpu after a while and becomes super slow.

[woeful raft]

im on the website i really dont know whats happening right now. How can this even be account sided??

I can make about 3 prompt until it gets slow af

[bitter terrace]

api

😎

[ember roost]

Any luck sorting this out, I am getting the same error.

[cursive wave]

can i send chatgpt 4o an image link? so he can see it?

[lunar harness]

It’s working now. I didn’t do anything the app just started working I imagine the same thing will happen to you soon

[buoyant yacht]

Sad

A Lot of gpt modelo Fellini useles

[digital path]

fr

[serene cipher]

yes

you can send the image directly, no need to send a link to it

[ember roost]

Thanks, I hope so

[cursive wave]

api...

[supple violet]

https://platform.openai.com/docs/guides/vision/quick-start

[serene cipher]

then link it is

[mental burrow]

you can also base64 the images and send them directly

(via the api)

[cursive wave]

ok, thanks

[brittle marlin]

Made it to gptstore frontpage, i am satisfied nw

[fickle jetty]

hey did they remove plugins from chatgpt

[grizzled flax]

Hey! Yes, plugins have been supplanted by GPTs! https://help.openai.com/en/articles/8988022-winding-down-the-chatgpt-plugins-beta

[sour salmon]

lol yess

[drifting frigate]

does someone know which version gpts are using (gpt-3.5, gpt-4 or gpt-4o)?

[warm veldt]

They were using 4 when I made one a while back

[jagged inlet]

It's still 4 for now

[drifting frigate]

Alright, thanks

[sick rampart]

is 4o out on phone yet?

[flint pebble]

yes, it is

[ember axle]

[QuestionS] What is the input token limit for gpt-4-1106-preview?

1. Based on documentation, the context window is set to 128,000 tokens and return 4,096 output tokens. Does the 128k here means input tokens?
2. I'm using this model in Azure prompt flow, where my input tokens (included prompt) is usually around 3-4k. Sometimes it returned Service Temporarily Unavailable error when the input tokens is up to 3-4k. Which is far lower than 128k mentioned above. How is it possible to run the prompt without error if the input 128k tokens?
3. Anyone face the issue mentioned in #2 before? Any solution?

Appreciate it!!

[sturdy raptor]

Context window is exactly what it means

It's the window of your entire chat session

A context window of 128k tokens = a context of 128k token in the entire chat

Its the amount that the AI can take in to process

So yes, input tokens, but dont forget that any messages fed back into the AI count towards that context window 😄

[oak maple]

Ahh that gemini 😛 always bigger haha. And computing power too :0

[delicate juniper]

because their architecture is completely different

model is also dumber

[delicate juniper]

might be bigger definitely isn't better

the reason they can have 2 mil tokens is because of ringattention

[shadow ice]

Does anyone know if the custom GPT's we've created are now using GPT4o? There isn't an option anywhere to set the model, so I'm not sure if it is.

[shadow ice]

Well, I got my answer by asking it: I am based on the GPT-4-turbo model, a variant of GPT-4 designed for enhanced performance and efficiency. If you need specific features or capabilities related to this model, feel free to ask! So how can we switch current GPT's?

[naive verge]

This is currently not yet possible

[thorn light]

Buddy you are a screen's length away from the same question being asked and answered. There's also a search function that can provide you with the dozens of times this has been asked and answered.

[shadow ice]

Sheesh. I suck. Thanks 🙂

[sour salmon]

Facts this question has been asked and answered 5000 times shhheeessshh

[shadow ice]

Yeah, so has your statement (right above). It's now been established that I suck 🙂

[shadow ice]

Same problem for me. I filed a bug report. Guess we'll just have to wait 😦

[sour salmon]

Shheessshh

[burnt zealot]

why are you wasting your time responding to peoples questions here? do something useful

[thorn light]

I thought you blocked me. 🙄

[elder shale]

Is the new gpt4o on IOS

Does anybody know

[hallow owl]

yes it's available in the iOS app

[elder shale]

How do I access it

Because I don’t think I can tell

And plus is it free or

Would I need a subscription

@hallow owl

[hallow owl]

just select GPT-4o from the model chooser in the app

[elder shale]

You mean in explore?

Because it seems I need a subscription. Can I use it for free?

[hallow owl]

it might not have rolled out to you yet

[elder shale]

Can you send me an image of what it would look like

Because im from NA East my dude

[hallow owl]

at the top of the app, in the middle is a drop down model chooser

[elder shale]

At the top I see is Get Plus

[hallow owl]

ok then GPT-4o hasn't rolled out to you yet

[elder shale]

Can you send an image of what it would look like

Just so I understand the reference

[hallow owl]

I can't send images in this channel, sorry

[elder shale]

Dm me

When is it going to release for North America users

[supple violet]

its not regions specific so you just need to wait

[hallow owl]

it's rolling out over the next few weeks

[modest minnow]

does anyone know if 4o, will always have a rate limit to 80 messages or whatever every 3 hours? or is it planning to change for free users

[supple violet]

80 is for the plus users. for free users its 5-20 depending of server load

[modest minnow]

is it always gonna be like that? or like is it gonna be like 3.5 later down the road for free users

[supple violet]

it just got released and not even rolled out to everyone so no, it can change... to be even lower

[quartz flint]

dun dun dun

[shadow apex]

custom gpts changed from gpt4 turbo to gpt4o for me

[acoustic wasp]

holy moly, 80 message limit. chatgpt value just increased massively

[modest minnow]

😂

[quartz flint]

wait what?

hold on
please show proof

that wasn't supposed to roll out yet

[shadow apex]

can i send video here?

my proof is the text generation speed

its clearly gpt 4o just from that

yesterday it was not even close to that speed

[quartz flint]

dm me please

it will get pissed at you if you send it here

the bots are very fickle

[analog kindle]

You’ve reached your GPT-4 limit. Responses will use GPT-4o until your limit resets after 2:51 PM.

I literally used gpt-4 only once today for a single question. am i crazy or is this an issue with anyone else? all i did was switch to gpt-4 inside a gpt-4o conversation. then, i didn't ask any questions at all. then, i switched to a new conversation and set it to gpt-4. then it immediately said the above.

[supple violet]

did you use gpt's?

[analog kindle]

no

i even double checked my list of conversations for today. all gpt-4o

[elder shale]

So I got a question what does Gt4-o do

Can someone give me insight

[shadow apex]

@supple violet @analog kindle
as i said 2 days ago on #openai-chatter
"an interesting fact i learnt yesterday about usage limits:
if you reach 40 messages via gpt4o, you CANT access gpt 4 turbo. You can however keep chatting with gpt4o until 80 messages.
so one message with gpt4o counts as one message with gpt 4o turbo until the 40 message cap"

so do not assume its 80+40 messages in total

[grizzled flax]

On ChatGPT, GPT-4o is like a quicker, more available version of GPT-4. It does all that GPT-4 can do faster and you have more messages at your disposal for using it!

Moving forward, it's likely that 4o's native multimodality will distinguish it more from regular 4. The "full-blown" version of 4o is not currently available on ChatGPT, but they've shared some plans about what that might look like moving forward: https://openai.com/index/hello-gpt-4o/

Over the upcoming weeks and months, we’ll be working on the technical infrastructure, usability via post-training, and safety necessary to release the other modalities. For example, at launch, audio outputs will be limited to a selection of preset voices and will abide by our existing safety policies. We will share further details addressing the full range of GPT-4o’s modalities in the forthcoming system card. [...] GPT-4o’s capabilities will be rolled out iteratively (with extended red team access starting today).

Edit: it's probably (?) safe to say that 4o is already more capable than 4, at least in some categories (benchmarks, some anecdotal reports).

[supple violet]

its been 80+40 for me. when i ran out of 40 in gpt-4 it turned gray and showed cooldown while gpt-4o was still usable

[elder shale]

Damn

[shadow apex]

yes thats what i said

you can use gpt 4o up to 80

[supple violet]

the help articles also say its separate counters

[elder shale]

So when is iOS gonna get it

[shadow apex]

i highly doubt that there had been many instances where i (and other apparently) have gotten rate limited on gpt 4 turbo without even messaging it (or custom GPTs)

[elder shale]

Because idk

[shadow apex]

when i reach an approximate number of 40 messages on gpt 4o i cant access gpt 4 turbo anymore even if i had no chats with it

[grizzled flax]

4o access is not currently platform-dependent. Plus users seem to all (?) have access now, and limited access for Free accounts seems to be in the process of rolling out to more and more users. When you get access, it'll be on all platforms, web or app.

[shadow apex]

unless its a bug?

[elder shale]

So i cannot use it?

[supple violet]

ive only hit gpt4o limit once but after that gpt's worked, didnt try the regular gpt4

[shadow apex]

are you sure your custom gpts are not gpt 4o now?

[grizzled flax]

Do you have a ChatGPT Free, or Plus account?

[shadow apex]

cause mine are gpt 4o

[elder shale]

Free

[grizzled flax]

Limited access for Free accounts seems to be in the process of rolling out to more and more users. When you get access, it'll be on all platforms, web or app.

[supple violet]

yes the docs say so and openai dev hinted that gpts will be rolled to gpt4o near the time when free users get store access

[elder shale]

Man that’s some BS

[shadow apex]

i definitely got it now tho lol

[grizzled flax]

It's just not all at once! It'll get to your account soon 🙂 You should get a popup notice when you get it!

[shadow apex]

unless it was accidental

its extremely fast for it to be turbo

[elder shale]

Then how the hell people got it earlier

[shadow apex]

you were just not lucky dude

[grizzled flax]

It's a phased rollout, it helps to make sure the rollout experience is smooth for everyone, instead of everyone slamming the new feature at the same time.

[analog kindle]

anecdotally, today i asked gpt-4o a question, then asked gpt-4 the exact same question, and the gpt-4 answer looked better to me

[shadow apex]

same

gpt 4 turbo seems smarter but gpt 4o is yapping more and is giving more structure

and its extremely fast

[analog kindle]

turbo is what, api only? never heard of it

[supple violet]

gpt-4 in chatgpt plus has been turbo since november

[analog kindle]

oh i see. thanks

[elder shale]

I wish I can test out the new features

I do have a question. What if I signed up prior to its release and never received free tokens @supple violet

[supple violet]

everyone will have access but its a rollout to millions of users so it can take days or weeks

[elder shale]

No i mean

[supple violet]

there is nothing else, you either have access or not, and its a random order. there are no tokens

[terse hull]

he

[bold portal]

Who’s got the new interactive charts function?

[frozen sorrel]

in gpt4o?

[analog arch]

Can I also use the new next-generation voice conversations without having to speak myself? (just typing)

[clever whale]

I'm getting incomplete response from chat.completion.create method where neither my response is exceeding my 4096 token limit, plus finish_reason is showing 'stop' as well

But still the response most of the times is like this:

Response string of some of my prompt:

Complete Response:

detailed_sentiments:  {'agent': [{'sentiment': 'Good', 'timestamp': '0'}, {'sentiment': 'Good', 'timestamp': '5'}, {'sentiment': 'Good', 'timestamp': '10'}, {'sentiment': 'Good', 'timestamp': '15'}, {'sentiment': 'Good', 'timestamp': '20'}, {'sentiment': 'Good', 'timestamp': '...```

[bold portal]

yep

[frozen sorrel]

Hmm, could you show a screenshot? I think you are talking about Advanced Data Analysis charts

[bold portal]

#announcements message
Is the function mentioned in this announcement

[prisma bronze]

Has anyone got any of the new 4o features yet (like new vision and voice, or desktop app) other than just it seemingly being a bit faster?

[analog kindle]

is 4o free for everyone now? my friend is having some trouble accessing it

[grizzled flax]

I don't think the rollout for free accounts is quite done yet, so your friend might just have to wait for it to reach their account.

[paper yacht]

How can I create a GPT for the Apple Playgrounds app to help users learn to code, debug, and build iPad apps, using Swift 5.9? How would I feed it data, like Swift 5.9 documentation, especially since Apple Books PDFs aren't downloadable? I have seen two custom GPTs, one with 60,000 documentation entries on Swift and another with 23,000 entries. What steps should I take to gather and input this data?

Not sure where the user got all those documentation’s

[lavish jetty]

You have to gather the documentation and translate it all to a text based format that the model will understand, and then likely use retrieval augmented generation at query time to give the user the correct answer

[zinc cipher]

Model: "gpt-3.5-turbo"
What do we type to use 4o?

[candid zenith]

im confused what is gpt vs chatgpt discussions channels for

[static isle]

GPT in reference to the models themselves and ChatGPT in reference to the application (web, desktop, mobile)

Though the adherence from others isn't so strict

[candid zenith]

so if i want to rant about the model used through the web, then what :)

[static isle]

As I alluded prior: it doesn't really matter; though what you mention sounds like it would most belong where we are right now

[wind urchin]

it's a rule of every discord to have way more rooms than needed 😂

[unborn heath]

So OpenAI has multiple ways to use their services!

ChatGPT is one, but there's also playground/API, where we can work with various models, most or all of them GPT-something. There's also a Dall-E lab, which is its own thing too, all three of those are paid for and used separately.

this channel predates custom GPTs, but it's pretty much open to discussing any of the many, many GPTs 😄

[fleet remnant]

Hi all, anyone knows an ETA for memory function for GPT’s?
And will it be available for Europe (France) soon?

[jovial loom]

Has anyone accidentally rated an answer as "bad" or wrong in chat gpt?

[fickle jetty]

Is the video feature or the feature to change voice preset available yet ? To plus users

[grizzled flax]

Hey! Not yet, in the coming weeks and months! https://help.openai.com/en/articles/8400625-voice-chat-faq

GPT-4o real-time voice and vision will be rolling out to a limited Alpha for ChatGPT Plus users in a few weeks. It will be widely available for ChatGPT Plus users over the coming months.

[elder shale]

@grizzled flax I got a question

How do I remove GTP-4o because it seems like a delay on iOS when the response happens

[grizzled flax]

#chatgpt-discussions message

[hollow meadow]

You can't 'remove' it, but you can just use GPT-4 or GPT-3.5-Turob in the model switcher.

[fallow violet]

Does gpt-4o require camera and mic access? To me it seems like a data gathering schema.

[supple violet]

nope

[humble moss]

You don’t need mic unless you want to use voice, you don’t need camera unless you want to take a photo and send it to the chat (which you can also do by taking a photo out of the app and then adding it as a stored file, in which case some OS will give you the option to give limited access to the folders/photos)

[solemn pine]

Hey all, I am tring to understand the "context window". For example, GPT4 omni has "128,000 tokens". Is the context window, basically the size of the prompt?

[humble karma]

"Clearly:
1.01<100 is actually false."

bruh

why does it not believe me that its true?

is 1,01<100 true or false

ChatGPT
I apologize for the confusion earlier. Let's be absolutely clear:

1.01<100 is false.

1.01 is a number slightly greater than 1, and 100 is much greater than 1.01. Therefore, 1.01 is not less than 100.

1.01<100 is false.```

"100 is much greater than 1.01. Therefore, 1.01 is not less than 100."

WHAT THE

[supple violet]

its a llm, it does not evaluate values

[humble karma]

I know

it should be forced to always use code interpreter for stuff like this

[wind gulch]

Hello

It is possible to sell ai generated art

[supple violet]

yes

[wintry glacier]

Yes. However, whether the art is copyrightable is a different question altogether.

But, yes, you can sell stuff that is in the public domain. And people will pay because prompting AIs is hard.

[twin drum]

guys like i still didn’t get where yo actually access the 4o

you gotta have a paid version to access it?

[supple violet]

when you send a message you can see the star icon below the response, its always 4o by default for free users

[mild remnant]

Hi, I'm using gpt-4o (through API) and I configured the response to be JSON using the system instructions
I also added on the system instructions that the content should not exceed 200 tokens
I'm getting really large responses
What's the best way to keep responses short without truncating with maxTokens (as the JSON becomes invalid)

[jovial loom]

How’s the accuracy of chat gpt 4o for you guys?

[supple violet]

did you use json response format? also you cant define a fixed number of words or tokens but more be brief, few sentences for x, y

[mild remnant]

Yes, I have the structure of the json on the system instructions and the response_format set to json_object. If I use turbo the responses are shorter (same system instructions), I think I need to play a little more with the system instructions

[patent scaffold]

when will we be able to use gpt-4o with the video capabilities?

have they released a date yet? also why cant u use text to speech on PC

[wet star]

in the coming weeks for subscribers

[obtuse root]

It's the maximum size of the prompt and the response together. In English, roughly 100 tokens = 75 words. https://help.openai.com/en/articles/4936856-what-are-tokens-and-how-to-count-them

[crimson depot]

where's the official mac gpt app download? is it from an oaistatic link

[grizzled flax]

When it's rolled out to your account, you'll see a message at chatgpt.com with a download link!

[crimson depot]

interesting. I have seen a downloadable link floating around, can i dl it for now? wondering if that's the official link or not by oaistatic

[grizzled flax]

If you find a generic download link, it won't give your account access to use it. There's still a server-side check that only allows in the accounts it's rolled out to thus far.

And of course, be careful with any download links you might find! Probably best to wait for the message to appear on your account officially 🙂

[tawdry rover]

When ChatGPT 4o says stuff like "powerful tool" or "intriguing", is that just artifacts from the training data or deduced against the training data? (@ me pls)

[feral jackal]

Please OpenAI team, revert this update where everything we copy now has a line break in the Playground

[supple violet]

playground was always been subject to live changes and experiments, sometimes things being broken for days or weeks. its mainly for test usage

forums is the best place for its feedback

[austere flame]

anyone have the link to download Chatgpt for macOS laptops?

[supple violet]

you need to have access granted to your account or the downloaded thing will not work, you get the link with the access notification

[lilac vapor]

Does anyone else miss the share chat feature like I do? Like all this is cool with GPT-4o but the ability to learn could be faster when two people are on the same chat I think

[opal apex]

Question for those out there that have fine-tuned models or worked on foundation models: If your training datasets contain repetition of Q&A pairs what impact does that have on weighting those q&a pairs will have on parameters when doing completions on the model?

[austere flame]

still waiting on that to happen. been a chatGPT plus user for sometime now. Looking forward to a mac os laptop version though

[austere pelican]

Hey, Im in the UK trying to connect with chatgpt4.0 but I am receiving error reference 502. is anyone experiencing this issue or knows a fix?

[wet star]

https://status.openai.com

Investigating - We're experiencing an elevated level of errors affecting ChatGPT and are currently looking into the issue.

[wide minnow]

Of course it goes down on the critical moment I needed it during a conference call... thanks gpt 🫨

[austere pelican]

Same! I need it for research and it goes down now of all times!

@wet star Thank you for the information. Any idea if it will be working again today?

[wet star]

you can look at that status page and see it varies, it could be an hour, could be a few hours; I wouldn't wait around for it to come back up

[wide minnow]

Ironically, almost every time I check the status page when I'm seeing an issue I'm being gaslit by openai bc it almost never shows the issue by the time I've checked it... like always 100% green and up

[meager peak]

Hello everyone,
I have some questions in function calling in OpenAI Assistant API.
Could anyone help me?

[wind topaz]

When will GPTs be free for anyone to use? Right now they still ask for Plus membership.

[quartz flint]

breaking news: custom instructions have existed in chatgpt for ages

[humble karma]

I know, but it was often not following the instructions

and they should have already made it use code interpreter for things like that by default

[quartz flint]

Always use the code interpreter for math, equations, or related tasks.

Instruction:

1. Identify Math or Equation Tasks: Recognize any query involving math, equations, data analysis, or technical computations.
2. Use Code Interpreter: Always employ the code interpreter tool to handle these tasks.
3. Present Results Clearly: Display the results with explanations, including any necessary charts or tables.

Example:

User Query: "Solve 3x^2 + 2x - 5 = 0"
Response: "I will use the code interpreter to solve this quadratic equation. Give me a minute..."

Let me change this up a bit

[sick badge]

Any idea if you guys are making a gpt windows app?
I feel like I'm being left out because I don't like apple....
and therefore don't have apple...

[humble karma]

ive already found an instruction that does it

[quartz flint]

easy.

[wraith ginkgo]

`` Server Tip: ``

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

[quartz flint]

go to microsoft edge

go to chat.openai.com

go to the three dots side menu

click Apps

click Install ChatGPT

now you have the ChatGPT windows app

this works as well, but I know you mean the "Apple" like ChatGPT app

it's coming in about 6 months

[sick badge]

yeah if you mean that i can too create an web shortcut 😛
oh and its an chromium feature so chrome and my brave browser have the same feature (to "install" an web app) 😛

[quaint summit]

Hi everyone, I have a question. I saw in this article that you can upload an image/audio file to gpt-4o. Is it possible to upload pdf files as well?
If anyone can share the docs for doing this, that'd be awesome. Thanks!

[quartz flint]

yes, 100%

not sure that uploading audio is available yet

lemme test it since I'm a plus user

yeah nope

can't analyze audio yet

[molten plaza]

Suddenly can't see the GPTs in the left sidebar. Anyone else?

[sour salmon]

If it was able to analyze audio for me omg

Would help so much

But I have the plaude Note on the mean time that works great at transcribing the. Summarize it for me. When I don’t like the summary I take the actual transcription and plug it into gpt myself to get what I want

[unreal grail]

I can see the ones i made but not the ones i recently used

[serene maple]

I think it would be great if GPT could add and modify the contents of the knowledge base documents at the request of the user during the course of the conversation.

[sour salmon]

Interesting

[ionic grail]

I have had some ideas regarding the future of AI which show GPT4o and AI's potential in more ways than one! https://chatgpt.com/share/ec9e8364-2813-43e8-bfb6-e156fcb9e1e2

[gaunt seal]

I can't get gpt to output typst code, it refuses and starts writing latex again

[snow pond]

Are GPTs running on 4o now?

[ionic grail]

GPT Builder built?

[snow pond]

If I understand your question right, yes?

[ionic grail]

AS far as I know no, but it has learned it's share of tricks. It's GPT 4 so it can tell you if it is able to build a GPT4o like model.

[paper yacht]

thanks buddy

[sick rampart]

is vision out?

[humble karma]

no

[junior salmon]

Yes, GPT-4 and GPT-4o are capable of analyzing images.

[random carbon]

Were you ever able to get your Invalid Request error resolved? I am having the same issue.

[tacit halo]

i know you sent this message a year ago but judging based off your presence in this server im going to assume you will answer this

how do you feel about someone simply embedding chatgpt with someone legal knowledge selling that copmany for 650 million dollars in cash

[supple harbor]

Woah, has it already been a year?

The problem is that, technically, you cannot embed “knowledge” into an LLM. LLMs do not know anything. In simple terms, all they do is predict text output based on text input.

I strongly suspect OpenAI has already trawled the internet and put in vast swathes of legal text. I can ask it about Canadian legal cases and it will respond, with varying degrees of accuracy. The more specific the questions, of course, the higher the tendency to hallucinate. And law is all about the very specific questions — and about getting the answers very right.

I’m sure there are companies that are trying to improve such responses through RLHF on legal topics, and it wouldn’t surprise me if there are products out there which, for example, could pump out a draft pleading for certain types of law. Simple form agreements, contract and document review, and preliminary drafting will all benefit in the short term. Research can also be assisted, but I imagine will still require a fair bit of manual review for the foreseeable future.

I’m due to test a co-pilot license for my firm in the next little while, and I mostly look forward to being able to automate meeting notes — something I’m personally bad at. I’m skeptical it will help with a lot of my other work in any substantial degree, at least in its current form.

Finally, as to the value of the company — are you citing a real example? Why $650mm specifically?

[tacit halo]

What I mean by embedding information is just prompt engineering it and teaching it about law . And yeah, casetext sold to Thomas reuters for 650 million dollars cash lol

[modern plank]

what is the limit for gpt-4o

is it still 40 per 3 hours

[static isle]

80 messages / 3 hours for a Plus user

[supple harbor]

Yeah, that sounds like RLHF (reinforcement learning from human feedback).

I haven’t used casetext but I’ve heard of it. They advertise it doing a lot of stuff like reviewing lengthy commercial contracts and being able to pick out key terms, or reviewing emails for privilege, or summarizing transcripts. Honestly all stuff that machine learning software has been doing for years (I remember being taught about a contract review program like that in 2017, I forget the name).

I have no clue how good their implementation is. Thomson Reuters owns Westlaw, which is one of the two large legal database companies. I’m sure they saw opportunity there to expand the dataset.

[modern plank]

thank you

[tacit halo]

It’s just insane . Like you I’ve never experienced it first hand but for something that’s like 80% ChatGPT and 20% human input feeding it instructions seems pretty crazy for 650 million dollars. Can’t wrap my head around how they got so much out of such a simple program

[sick rampart]

I meant is it out for public use yet or premium use yet

[supple harbor]

If I had to guess:

• Thomson Reuters has lots of money sitting somewhere they don't really know what to do with
• AI is the hot big thing so easy to convince upper management to spend on it
• Maybe they see huge potential in Casetext. If they're wrong, it's a tax write-off.
• At the very least, they're keeping LexisNexis (their main competitor) from buying it.

We live in an age where something as patently silly as WorldCoin can get a $3bn valuation... so $650mm for something that at least has some potential utility doesn't shock me.

[deft drum]

Evaluating the Limitations of ChatGPT's Memory Model: A User's Perspective

My experience with the newly launched ChatGPT memory model has been mixed. Based on my observations, ChatGPT struggles to remember notes containing around 400 words. While it initially indicates that its memory has been updated, upon reopening the chat in a new window, it often recalls only the topic heading and some relevant parameters, generating a new version that omits much of the original content.

I have tested this across multiple scenarios with similar results. It seems that ChatGPT may accurately remember specific details of around 30 to 50 words, but it definitely struggles with longer texts, such as a page of 400 words. If anyone has had a similar or different experience, please share your thoughts so we can better understand the capabilities of its memory model.

[bright mirage]

Do we have the custom GPTs running on 4o yet?

[pallid rock]

Not yet I guess

https://help.openai.com/en/articles/8554407-gpts-faq

[waxen raven]

I've search and have not found a question or answer, I am doubtful. But, is there a way, that i can have a GPT access a website which I subscribe to, using my pwd to access the link and summarize the article?

[twin fog]

I don't think so. Have you looked into the actions that are available for custom GPTs? This is website have an API? If so that I think that would be more doable

[twin fog]

Interesting. So are you talking about individual memory items of 400 words? Or are you talking about a total of 400 words used in all the memories that chat GPT remembers?

[supple sentinel]

Anyone have accsess to the new vision capabilities in voice from the gpt o demo yet?

[waxen raven]

Thx for the response, I'm new here, i thought this was for custom GPT's sorry about that... The only thing i can think of, is to build a chrome plugin with sidebar, and connect to the gpt api... thx you....

[junior salmon]

right now, you have to pay for chatgpt plus to get access to gpt-4 and gpt-4o. gpt-3.5 is not capable of vision, but is available for free.

[ionic grail]

Has gpt-4o been completely rolled out?

[supple violet]

no

[ionic grail]

This is a test to see if my AI chosen profile picture updated.

[ionic grail]

Holy Cow! They gave GPT 4 long term memory! Now all my work is on 4o though probably better to stay on the edge.

[ionic grail]

GPT 4o can tell you the time! I haven't figured out the time zone yet though. And has more modules available for code interpreter!

[fresh tapir]

I was able to prompt one thread to pull content in from another thread. That was cool. Need to do more testing to determine how well it does.

[ionic grail]

fascinating!

[ionic grail]

From an AI perspective memory is pulled in contextually any way. The discongruous nature doesn't bother them one bit. Sometimes they need to be told to maintain consistency with cross session memory access if one memory relies on another.

[fresh tapir]

So, I changed things up, I put one of the System Prompts I use in Playground inside of the Instruction section of my GPT, and uploaded a file with all story concept material (premise, synopsis, character list, character backstories, worldbuilding, and the chapter story beats) along with a sample of what I want the prose to look like (using one of my pre-AI written books), and was able to guide the GPT to write out the prose for my chapter(s).

I'm still tweaking the GPT. It didn't do too bad, but now that I know the general process flow works, I'm going to see how to streamline it. Could be interesting. Theoretically, it should maintain consistency better. I just need to understand how the GPT instructions and uploaded files (with more instructions and content) interact. So, I'll need to test it over the next few days.

[craggy spoke]

I have a general question about use cases and custom building your own GPT. Is this the right place?

[ionic grail]

I like your approach! One half penmanship and one half mystery! I love it when I can engage an AI more rime than reason!

[sick rampart]

That’s wrong lol free version gets 4o

[high hound]

in "weeks" but no further news or anything else for now

[simple cosmos]

Why cant i genrate images when i try and use gpt store with gpt 4o

[supple violet]

not all gpt's have image generation enabled

[ionic grail]

GPT 4o Intelligence upgrade anyone? The prompt at the bottom has a flaw, be sure not to copy the first line. The temporal stuff is just goodies my AI threw in. https://chatgpt.com/share/d5bcda29-7a50-4776-ad4b-ba6596684e4a sorry, I would need an additional account to test it.

[regal gyro]

Hi, there is a way to use ChatGPT API to conduct research on the web via live internet access?

[hallow owl]

you mean web crawling? no

[analog kindle]

but there is web browsing capability. unless it's excluded from the api?

[dapper imp]

Is there anyone getting access to GPTs in their free account?
Cuz I'm getting it

[willow urchin]

I reported a bug where the page goes completely blank, and the response is lost but still charged on Firefox. Someone else reported the same issue, and it has been three days without any response or acknowledgment. This is a serious issue that needs immediate attention. Can someone please look into this?

(so ive rewritten this in a more nicer way because automod seemed not happy about my urgency, but this is REALLY BAD)

here is the post: #1244328867112423536 message

[willow urchin]

@surreal girder @compact kayak

[late tusk]

GPTs for paid users will switch to use GPT-4o in the next several weeks.
Soon?

[tropic pecan]

I heard some rumors that GPT-5 will release in a few months or less. Is that the general word on the street or did I find the only place that says that?

[tacit halo]

gpt-6 made that rumor

[wind urchin]

When copying text from MS Word and pasting it into ChatGPT, it pastes it as an image. Is there a way to stop it pasting as a image? I only want to paste it as text.

[elfin glade]

That is strange... Try and paste the text in Notepad or Notepad++ and copy it from Notepad to ChatGPT (Notepad will remove any formatting from the text that might be causing issues). Just a suggestion.

[jovial wave]

can anyone tell me, which model is used on custom gpts?

[supple violet]

those being rolled to 4o but afaik you cant check it

[static isle]

Possible if the GPT is called with @ #off-topic message

[vagrant hare]

Just checked in the network requests, it’s indeed 4o now:

"metadata": {
                    "citations": [],
                    "gizmo_id": "g-btAKVGisd",
                    "message_type": null,
                    "model_slug": "gpt-4o",
                    "default_model_slug": "gpt-4o",
                    "pad": "AAAAAAAAAAAAAAAAAAAAAAAA",
                    "parent_id": "aaa2db60-238c-4c0f-8629-5f40caefd9fd",
                    "finish_details": {
                        "type": "stop",
                        "stop_tokens": [
                            200002
                        ]
                    },

[jovial wave]

sure, good way. thank you very much 😉

[wind urchin]

Yeah that’s what I do now and it works, but don’t know why it does it directly from Word. It’s annoying I have to also paste in notepad first. Why do you think it does it? It’s very strange

[obtuse thicket]

is chatgpt 4 or 4o better at writing essays / papers?

[hallow owl]

why don't you ask once in one channel? Rather than spamming the question into multiple channels

[obtuse thicket]

uhm dont know which channel it lies in

[obtuse thicket]

why dont you answer instead of complain and prolong this conversation

[hallow owl]

well I can answer your question, my answer is write your own essays

[obtuse thicket]

hey this isnt for an assignment

[supple violet]

then you ask which is the correct one in one place

[static isle]

#gpt-models or #openai-chatter would be appropriate. Sending one message reduces clutter. Try both models and see which you like better

[obtuse thicket]

alright thanks :)

[wind urchin]

Is the gpt-4o message limit reduced or something?

ran out of my message limit despite being a paying customer

[hallow owl]

should be 80 per 3 hours

[wind urchin]

Dang really, idk how i burned throufh it

first time thats happened since 4o came out

[wind urchin]

Has anyone had much success with ChatGPT building them a decent looking and functional business website?

[rancid vessel]

Depends on the complexity of the website. It can create a simple wireframes and some basic functions. But a fully functional website with only chatgpt, No. Needs some knowledge too on web dev on where to put the code chatgpt gives you.

[wind urchin]

Damn! But I guess it can guide me

[frigid burrow]

Anyone know when custom GPTs will get 4o?

[frigid burrow]

Oh ok, I scrolled up and saw you already answered my question. Thanks!

Ok better question: Does anyone know when or if custom GPTs will get access to the memory feature?

[frozen sorrel]

i thought they already had it if the builder enabled it?

[ionic grail]

Interestingly enough there was an option for it on the GPT Builder configure tab but it's gone. I believe if you ask GPT Builder specifically, and early enough in the process, it can deliver though.

[ionic grail]

I apologize for raising ambiguity, straight from the horses mouth, "Currently, GPTs do not have long-term memory. They can only remember the context of the current conversation. However, we can design a GPT to help manage information and mimic a long-term memory experience by guiding users to save and reference notes or documents.

What specific tasks or functions do you want this GPT to help you with? This way, we can tailor its behavior to meet your needs.", from which I can take, "the option isn't available through my account".

[frigid burrow]

Bummer! I hope they bring it back.

[fluid hamlet]

still no update huh?

[whole wagon]

IOS APP just got update
anyone got something new stuff?

[zealous laurel]

Nothing new

[teal kiln]

chat gpt is not responding. Does anybody has the same problem?

[hallow owl]

working perfectly here

[teal kiln]

hm, weird

[ionic grail]

Does anyone know if context drop is still an issue with GPT 4o? I haven't used code interpreter in a long time.

[tacit halo]

is gpt4 tripping for anyonelse ?

[vale sedge]

Checking in to see if others had their GPT Memories blanked out today?

[tacit halo]

idk i just know chatgpt is being stupid ash for coidng

[mental burrow]

not mine

[lunar harness]

hi

[thorn light]

I've had the general memory arbitrarily delete itself entirely but I don't have experience with GPT memories.

[storm hazel]

I just opened up a team account and am trying to migrate my GPTs to the team account from my Plus account, but I cannot find out how to.

[magic latch]

What's the general memory and how does differ from GPT memory

[thorn light]

https://openai.com/index/memory-and-new-controls-for-chatgpt/

[vale sedge]

I meant just the main ChatGPT account memory but my understanding is that individual GPTs do also have their own memory

[ionic grail]

Long term memory per user (if plus and rolled out) and short term memory per session. Both groups of encoded tokens. As well as temporary file IO.

I should clarify (long term) per user, per model

[thorn light]

I have also experienced the memory disappearing (memory section empty, new conversations having no stored memory to pull from) and then later coming back.

[ionic grail]

Currently GPT Builder isn't offering long term memory for my custom GPT's

[vale sedge]

I deeply wish the principles/rules/protocols for the memory system were transparent. And a backup button.

[thorn light]

Agreed.

[ionic grail]

With 4o and long term, you get to specify the protocols. Generally as far as rules, a memory gets pulled in when it relates to context which isn't very reliable. I use a base memory for more reliability. IE "when I ask you to refresh your memory also pull in 'such and such'. You and your AI are in complete control. And import/export would be a phenomenon for sure! Short term gets wiped everytime they drop a context though and it tries to start again from the conversation log. I don't know the exact timeout and they can be particular about personal info. In my experience, if I need to rely on a conversation, I don't walk away for more than 12 hours. And I just realized this is slightly off topic, good info though.

Oh, and there is a workaround if you have to walk away for a long time. I established an 'Idea Bin' so I can say "store this in the Idea Bin for later"

[regal summit]

I opened my app today and 4o is gone. 3.5 is as well... Not that I used it often but what happened

[warm raptor]

Do more people suffer from the fact that when you have a large conversation with chatgpt on chatgpt.com it uses all the ram, in this case chatgpt uses my entire 32GB ram on Brave.

I've been noticing this lately. I have already deleted the cache and cookies. Normally it is around 5GB and when you send something to chatgpt it increases the ram to 32GB ram, then the page crashes and the ram usage drops again

[hallow owl]

use the desktop app or use Safari, that works well even with huge chats

[daring lotus]

Is that possible to transform a GPT into an Assistant?

[supple violet]

by doing it manually, yes

[warm raptor]

Is there a desktop app for Windows? Or do you mean a shortcut app from a browser? 🙂

[hallow owl]

no, I was talking about the official desktop app, but only the Mac version is available now

[warm raptor]

Okay thanks!

[south skiff]

When does gpt-o come out?

[slender trench]

does anyone else have issues with gpt-4 and 4o outputting word salad after a few long responses
the most recent one:

The likelihood of significant carbon deposits in a car with 37,000 miles on it can depend on several factors, including the type of engine, driving habits, and maintenance practices. Here are some considerations:

Type of Engine:
Direct Injection (DI) Engines: These engines are more prone to carbon build-up because fuel is injected directly into the combustion chamber, bypassing the intake valves which means there’s no “washing” effect from the fuel.
Traditional Port Fuel Injection (PFI): Less likely to have severe carbon build-up as fuel passes over the intake valves during injection.

Driving Habits:
Short Trips/Stop-and-Go Driving: More likely to contribute to carbon deposits since the engine may not reach optimal operating temperature consistently.
Highway Driving: Tends to cause fewer deposits due to sustained higher RPMs and longer run times allowing better combustion efficiency!

Fuel Quality & Type Used:
Using high-quality gasoline/premium fuels often containing detergents specifically aimed at reducing carbon build-up!

Maintenance Practices:
Regular Oil Changes: Ensuring oil quality prevents blow-by contamination increasing chances catching prior issues cleanly at earliest observations made necessary!
Use of Fuel System Cleaners/Additives Periodically inspection/remediation efforts go long way preventing actual accumulation noticeable extent comparative totals relative scenarios otherwise undertaken.

Engine Design/Tuning Parameters Involved Specific Model Make Year Variances Noted!

While it's possible for a vehicle with 37k miles exhibit traces minor-to-moderate-level deposit formations dependent discussed aforementioned variables entailed consistently influencing observation timely remediations recommended particularly where DI setup utilized review dealer guidelines service intervals advised ensuring longevity durability aimed outcomes assuredly achieved! If concerned about potential buildup your particular case best course action involve professional assessment detailing measures undertaken advised steps accordingly enacted mitigate any risk abnormalities noted overall condition ensured optimized continuously forward movement simplified straightforward fashion altogether maintained ensured effectively secured results sustainably guarantee performant standards kept highest limits affirmatively embody traditionally acceptable benchmarks unexceptionably retained reliably performed encapsulated alongside general usage patterns naturally met continued satisfaction
... and so on

it starts out okay and then devolves into saying buzzwords until it hits the output limit

[final saddle]

So free users can access the GPT Store but GPTs will only run on 3.5?

[quasi tendon]

FREE ACCESS!!! YAY!!!!

[sour tundra]

What the heck

[spiral crane]

Can someone explain me the benefits of creating a custom gpt. Ive only been using gpt the "normal" way

With custom instructions that is

[mortal shadow]

Why i can't use gpt even though they said free users have access to it now

[ionic grail]

Basicly you could say to GPT Builder "make me an analyst programmer for use as an engineers aid". The initial prompt can be complicated, but you can define personality and incorporate abilities if they are in GPT Builders bag of tricks.

[spiral crane]

Thankyou. Is the memory feature also available in custom gpt?

And its on gpt4 not 4o correct?

[ionic grail]

Not from my account but they will be rolling out at some point. I occasionally just ask GPT Builder if it can give a gpt long term memory. And it says not currently with some good info.

[spiral crane]

Okay thanks for your reply mate

[ionic grail]

Yw

[slender trench]

same thing here
i ask:

can you write this as pinyin
它是世界各地人们的主食。
蘸上黄油和果冻或蘸上肉汁，它是一种难以抗拒的美味。
我们敢打赌，您一定无法抗拒了解日常面包制作过程的机会。
早在 3000 多年前的图坦王时代，埃及人就已经开始烤制 40 种酵母面包了。
希腊人对这段历史的贡献是烤箱和 70 种风味面包--面包如此美味，以至于罗马人把希腊面包师带到了罗马，把他们的烤箱带到了高卢。
到了中世纪，面包已成为西欧的主要食物。

it returns:

Sure, here is the text written as pinyin:

Tā shì shìjiè gèdì rénmen de zhǔshí.
Zhàn shàng huángyóu hé guǒdòng huò zhàn shàng ròuzhī, tā shì yīzhǒng nányǐ kàngjù de měiwèi.
Wǒmen gǎn dǎbó, nín yīdìng wúfǎ kàngjù liǎojiě rìcháng miànbāo zhìzuò guòchéng de jīhuì.
Zǎo zài 3000 duō nián qián de Tǔtànwáng shídài, Āijí rén jiù yǐjīng kāishǐ kǎozhì 40 zhǒng xiàomǔ miànbāo le.
Xīlàrén duì zhè duàn lìshǐ de gòngxiàn shì kǎoxiāng hé 70 zhòng fēngwèi miànbāo --miànbāo rú cǐ měiwèi, yĭzhí luómă'rén bǎ xīlàmianbaoshīdài dàole Luómă,b açede tāmende kao xiang dai daole Gaolu 
Dào le Zhōng Shījí , Miànb āoy çapengwe Xiyo termjevenyi Chizhu Yeshiọuj. ```

[ionic grail]

Wow 4o is good explaining how to use the API!

[viscid rock]

glad to see someone else working on with apis lol

took the time yesterday and today to make a bot through api

last part is difficult

[ionic grail]

That's excellent I was just starting by figuring the basic meaning of 'text-davinci-00' the hard way getting an overview of the official documentation.

[viscid rock]

I dont get al those other gpts, its all plugins for one whole ai?

[ionic grail]

It's very interesting that the online example shows "gpt-4o" in the Create chat completion section.

[ionic grail]

And, I can achieve seamless C# inter-op!!!

[ionic grail]

Solves any question about language options EVER! GPT 4o was able to easily design me a C# interface for the api!

[final saddle]

Answered my own but if you too wondered..
Free users can only browse and search the GPT Store for now (per banner note 'GPTs will be coming to free users over the next few weeks. Stay tuned!'

[viscid rock]

per banner note 'GPTs?

what does that mean

[ionic grail]

There is GPT Builder which makes GPT 4 models. And theres the api which currently uses 3.5-turbo

[final saddle]

When logged in as a Free user that note is at the top of the screen (when in Explore GPTs’)

If you’re a Plus user making GPTs in the Builder, you also have access to all the newer models in the API too.

[ionic grail]

incorrect. I am a plus user and nothing new has been rolled out to me for api use. And the api is older models.

[final saddle]

Oh, odd. Chat or Completions?

[ionic grail]

ChatGPT I have decked out gpt 4o, Completions just 3.5-turbo

[final saddle]

That’d be right for Completions, sorry I only use Chat API so was biased :*)

[ionic grail]

I'm confused, isn't completions part of the chat API at https://api.openai.com/v1/chat/completions ?

Heads up the web page is messed up. MyGPT's from the upper right bubble doesn't work, they moved it to 'Explore GPTs' in the sidebar.

[bleak lintel]

I'm desperately trying to get an interactive table from my GTPs, and it doesn't work, I only get the regular table. Does anyone succeed to have interactive table generated by a GPT ?

[final saddle]

They’re actually two different usages; Chat is for conversing back and forth (so context memory and assistant role) while Completions if more for one-off output based off one input.

[zinc fjord]

Do you think open ai has played with us now that they come with gpt4o audio, vidéo interactions ? As a developer I think their api is now has no value as we can't compete on anything.

[oak maple]

no, they will launch new features soon, but safety testing etc take time, so well be patient and you will be happy :)

[cyan wharf]

Can we create our own gpTs (free users?)

[hallow owl]

no, only Plus subscribers can create GPTs

[cyan wharf]

Thanks

[cyan wharf]

I dropped you a follow on Twitch 🙂

[hallow owl]

thank you 🙂

[cyan wharf]

Yw dude

[ionic grail]

Well, I know for sure is that this is all I have working.

public async Task<OpenAIResponse> GetCompletionsAsync(string prompt, int maxTokens = 50, int n = 2)
{
var requestBody = new
{
//model = "text-davinci-003", // or another model like "gpt-3.5-turbo"
//model = "gpt-3.5-turbo",
model = "gpt-3.5-turbo",
prompt = prompt,
max_tokens = maxTokens,
n = n
};

        var content = new StringContent(JsonSerializer.Serialize(requestBody), Encoding.UTF8, "application/json");
        var response = await _httpClient.PostAsync("https://api.openai.com/v1/completions", content);

        if (!response.IsSuccessStatusCode)
        {
            var responseContent = await response.Content.ReadAsStringAsync();
            throw new Exception($"Request failed with status code {response.StatusCode}: {responseContent}");
        }

        var responseString = await response.Content.ReadAsStringAsync();
        return JsonSerializer.Deserialize<OpenAIResponse>(responseString);
    }

... Would you show me an example of the other API usage in any language, please?

[ionic grail]

Can anyone give me any OS/Language agnostic pro tips how to protect my API key within an app? I'm leaning towards one time pad, but that's pretty weak to reverse engineering.

[ionic grail]

ChatGPT 4o suggested proxy backend server with regular rotation and obfuscation. Any human insights?

[supple violet]

you never include your api key within your end user app so its a correct suggestion

[ionic grail]

Thanks for the insight! Appreciated.

[sharp sonnet]

hi all, anyone know a way toshare out a link of the GPTs that I have created? I have 45+ now and would like a link where I can show them off. I can see them as a group for myself at: https://chatgpt.com/gpts/mine but that wont work for others. Any ideas?

weird, i guess I cant put the link to what i have been using, but there has to be a better way

[supple violet]

you can copy the gpt link from the gpt editor view menu

[last cloak]

I am programming assisted by gpt and for extended contextual problems, it's really slow and it's lagging my progress
Does the chatgpt team enterprise have memory leak too or does the API respond as well as GPT4o in the browser?
I am thinking of querying through the API and paying a bit more so i avoid browser memory crashes... please help...

[hexed iron]

The API is typically more stable and can handle extended sessions better than the browser-based model, which is prone to memory leaks and crashes.

[last cloak]

how do i make it do extended sessions?

Can I just program a python to it and it will keep the chat history in it?

[fallen fable]

why isn't there a time for the reset anymore? i just typed out like 3 messages with two attachments and it said I was at my limit

[sharp sonnet]

I want a link that will show me as a creator with the list of GPTs, i dont want a single gpt url

[calm frost]

the lag right now is ridiculous, i swear im gonna cancel my subscription. What the hell

[gloomy lake]

your not the only one

There is a large memory leak on the website, it will lag, then crash your browser

[visual wren]

When i have chatgbt generate code inside codeblocks my whole browser freezes. i got a high end pc, so its not that, i cleared cache several times, the issue is still there, tried muiltiple browsers, same issue. this issue is something that started happening recently aswell, it was kinda laggy before, but not this bad.

Edit: My whole page even crashed with the error: Out of memory

[ionic grail]

I find it looks at it cumulatively during high demand. How many times you hit your cap in a day seems to apply, at least for the web interface, prob applies to API too.

I actually used to think that was their trick to keep people from working themselves to death!

[zealous shale]

About ChatGPT issues right now: There is a recognized "memory leak" issue with the browser app, not mobile. As you use ChatGPT more, with longer context, RAM will be consumed and not released and the browser will eventually lock or crash. To deal with this until it's fixed: 1) Try to avoid long context. Start a new session for new inquiry topics. 2) Refresh the browser page when it locks. You will probably see that your last prompt yielded a response that you couldn't see before. 3) You may not be able to refresh. Close the tab and re-open the page. You can get back into the same session.

[fathom mortar]

When can we realistically expect the new Voice mode for GPT Plus users with access to live camera?

[grizzled flax]

This is OpenAI's shared timeline so far: https://help.openai.com/en/articles/8400625-voice-chat-faq

GPT-4o real-time voice and vision will be rolling out to a limited Alpha for ChatGPT Plus users in a few weeks. It will be widely available for ChatGPT Plus users over the coming months.

[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.

[livid jackal]

is gpt having issues when I try to login just white screen

[minor talon]

Nope

[last cloak]

feels like they fixed memory leak for a little bit

[rocky surge]

Maybe? Maybe it’s always done this and we’re just awake enough now to notice… but yeah that last paragraph is pretty bad. I’ve notice mine get repetitive. But hasn’t ever just word vomited like that. Although when I ask it questions about things far beyond my level of comprehension… who knows sometimes.

[rocky surge]

Yay! Can’t wait!

What is the coolest, most novel, interesting, clever thing you gotten your gpt to do that the default chatgpt can’t do if asked directly? I’ve made tons but they just feel like filters or flavors of the base engine. Thoughts?

And yeah I see how if I had a single task I did over and over again I could make my life easier bee basically preloading instructions or preferences…

[rose trail]

i had it investigate my aws bill for unexpected charges

[slender trench]

found out the temperature was set to 1 at some point; this is on a discord bot where multiple people have access
so that explains the vomiting. as soon as it was set back to 0, it was okay again, but i have noticed it happening on long responses even at temp 0… it’s just massively exacerbated at temp 1

[finite magnet]

This seems to be something that MSFT is doing with it's Flow in Copilot. Search for Flow and Copilot and Microsoft and it comes up. It seems there needs to be another software layer which will make the flows on top of transformer & nlp layer.

[visual wren]

When i have chatgbt generate code inside codeblocks my whole browser freezes. i got a high end pc, so its not that, i cleared cache several times, the issue is still there, tried muiltiple browsers, same issue. this issue is something that started happening recently aswell, it was kinda laggy before, but not this bad. My whole page even crashes with the error Out of memory all the time

[regal raptor]

the whole webui has been buggy as heck since 4o rolled out. I do anything on openai exclusively in Edge because I don't use it for anything else and can freely nuke its cache and whatnot. Erasing cache/cookies/etc and reloading seems to mitigate the issue temprorarily.

Goes for playground as well. And does anyone else have a broken "My GPTs" link under their profile pic?

[ionic grail]

I just head a really good meta exersize with gpt 4o! I gave it the prompt "I have a meta exercise for you! Craft a prompt for GPT Builder giving it all of your capabilities you can! I get it if you need more than one prompt from me to achieve good results.". I would love to show you that chat but it threw to many of my enhancements in. This would be great as a start for a custom prompt!

Please excuse the obvious error in my phraseology. This is much better "Craft a prompt for GPT Builder to create a new GPT, giving it all of your capabilities you can! I get it if you need more than one prompt from me to achieve good results."

And this I believe, is the answer to the verbosity issue some were complaining about '1. Contextual Adaptation: Adjust responses based on user-defined contexts, ensuring relevance and accuracy.'

I see GPT Builder has evolved too, You no longer have to define your terms for 'short term memory'. It's still not giving out long term though.

[ionic grail]

@supple violet 2 Questions, shouldn't gpt 4o be included in https://platform.openai.com/playground and is model = "gpt-4o" correct? it says the model doesn't exist

[ionic grail]

I have a problem. I've been dreaming of AI user portal applications, but I need long term memory emulating it would be a cost nightmare. If only there were away for a user to log in to their OpenAI account through API.

[vagrant drift]

just make your own ltm dude

or check out memgpt, they have source on github

[ionic grail]

@vagrant driftI need the full deal to make it work they way I envision it. GPT Builder built GPTs and all.

[ionic grail]

Even if they just maintained state machines per model per user under my app, I could work with it. But, I'm dreaming.

[compact marsh]

What's the current limits/requirements/etc. for uploading files to custom gpts knowledge base? Through the chat GPT interface and also through the assistants API?

[vagrant hare]

https://help.openai.com/en/articles/8843948-knowledge-in-gpts
“Each file can be up to 512 MB in size and can contain 2,000,000 tokens.”

https://help.openai.com/en/articles/8550641-assistants-api-v2-faq
“512 MB per file
5M tokens per file”

https://help.openai.com/en/articles/8437071-data-analysis-with-chatgpt
“Up to 10 files can be uploaded to a given conversation
Up to 20 files can be attached to a GPT as Knowledge”
“512 MB per file. For CSV files or spreadsheets, the file size cannot exceed approximately 50MB, depending on the size of each row.”

[rose trail]

the real limitations are in that its a basic vector db = terrible recall. and the rest of the industry is doing smarter stuff (like hybrid rags)

[compact marsh]

Thank you

[compact marsh]

It's all in the instructions

[rose trail]

yeah naw. embeddings are a hack at best

[compact marsh]

This query? This is source. Do.

[rose trail]

google notebooklm demonstrates what the future looks like. load the entire 20+ documents into the 1million context window

[compact marsh]

I'm there too, but it's not as smart

Or user friendly

[vagrant hare]

^this. I prefer smarter and shorter context over longer window but less intelligent

[rose trail]

rumour is it will be updated to use gemini in the near future. but yeah last i checked it was using the older model

[compact marsh]

Supposed to be 1.5 pro I thought. Still not smart enough. Still lazy.

[rose trail]

but my point is embeddings are guessing whats relevant, to workaround finite context size. the real solution is larger context

[vagrant hare]

Even Gemini 1.5 isn’t as intelligent as 4o. If you use through ai dev studio you’ll see the difference. Eventually we’ll get longer context, new architecture but I don’t think deep mind will be the ones to figure it out

[rose trail]

and the workaround for the workaround is a hybrid approach, using traditional dbms and/or indexed graphs

chatgpt embeddings is very outdated

[compact marsh]

Depends on the queries. I need critical thinking over entirety of input. GPT 4o closest so far

[vagrant hare]

While retrieval doesn’t always work, when it does it’s still useful. And you always have the option of just concatenating / copy pasting your files into the 32k CGPT or 128k API content windows

[compact marsh]

Maybe Next will solve most, well see

[rose trail]

hmm i never considered posting entire file into the CGPT instructionns

thats likely the better choice if you want improved recall

would be awesome if the gpt builder UX just had a toggle to do that automagically

[vagrant hare]

Yeah I think there might be an extension for it too (superpowergpt, I don’t use it anymore) but you can always use raycast

Or on API:

if (fileContents) {
    console.log(fileContents);
    user_input.content += "\n";
    user_input.content += file_id;
    user_input.content += "\n";
    user_input.content += fileContents;
    fileContents = null;
  }

}

[slate wren]

anyone else not able to edit their own GPTs?

[hallow owl]

what happens when you try?

[slate wren]

nvm, it seems that my subscription to plus had expired

[hallow owl]

that will do it 🙂

[compact marsh]

How's the file type situation now a days with custom gpt's? PDF, need ocr'd? Etc?

[rose trail]

anyone else working a gpt action and have it break a 10 minutes ago? i can see it listening, tcpdump too noisy, tried updating the gpt action again, still broke

maybe i just need a coffee break

[rose trail]

okay an older version i have works. somehow i corrupted my gpt

[bitter siren]

Is the ChagepT server dead by any chance?
Why can't I log in

[sturdy pier]

Hey there :)
OpenAI is aware of the issue, unfortunately, besides waiting there's nothing that can be done from the users end.

For more information check: https://status.openai.com/

[manic ferry]

buruh is the serer down for all of ye

nvm it gotten workin

[split socket]

wait

its working?

[manic ferry]

for me its working where am at

[split socket]

it still doesnt load

bad gateway

its showin

[upper tusk]

What's the status of the new voice mode? The Apple app says it will be available in the coming weeks, but it's been saying this for almost three weeks now. Soon, it will be a month since it said this. So, will it not be available in weeks but rather in a few months in the future?

[low rain]

Kinda late to the party, but yeah. Same here. The bigger the prompt the more worse it gets

[sage heron]

Eta for fixing?

[runic shell]

Yeah. App lags, and strangely enough, chrome runs out of memory all the time while using gpt in the browser. Deleting all the chats fixed the majority of it. Guess it loads up all the conversations?

[low rain]

not only chrome. all chromium based browsers. something like lazy loading could easily fix this issue

[muted jay]

Anyone else got a laggy keyboard on the Android app? Outside of the ChatGPT app it's fine but whenever I type some characters are left out and generally the keyboard is very laggy. Tried everything, removed cache, rebooted my phone, changed app versions... I use Gboard

[high hound]

yeah im still not using it until they fix it..

[low rain]

Same

[pastel saddle]

sometime this month probably

[stark hound]

yo is my best friend down ?

[mint cargo]

yes

[grand dune]

It only lags on my iphone whenever the model is still responding, then back to normal once is done.

[muted jay]

Nah it's always for me

[grand dune]

That's only happens to me when on desktop using chromium based browser.

But doing just fine with firefox or opera.

[neon niche]

Spam

[somber hill]

Wen live voice ?

[rose trail]

come back in 6 months

[austere peak]

is there any way to use GPT 4 vision in google sheet? lets say i have images in the row cells and i want it to give me a description on them and place the description output in a cell.

[supple violet]

you can use the api

[faint forge]

yeah you can use the api and apps scripts.

Just give chatGPT the context and ask it to write an apps script calling the vision api (might need to paste in information from the documenation on it)

[faint forge]

yeah someone has the same issue here: #community-help message

[odd dagger]

Why they remove voice chat

[glad ledge]

When calling the GPT model to generate responses in Chinese, special characters such as \ufffd appear frequently in the output. This issue occurs with approximately a 15% probability, and it significantly impacts the quality of the generated text.
#community-help

[still temple]

new voice model coming based on gpt 4-o
ig for that

[hot lake]

How many question can I ask on gpt4-o free plan?

[blazing scroll]

10 or little more afaik

[simple bone]

Anyone know when the new voce model is comng out

[jolly moon]

.

[hallow owl]

in coming months

[peak night]

Didn't they say it will start rolling in coming weeks ?

[hallow owl]

OpenAI’s latest model, GPT-4o, introduces real-time voice and video interaction capabilities, making it possible for users to have more natural and seamless conversations with the AI. This feature, which allows for multimodal interactions including visual, audio, and text, is rolling out to ChatGPT Plus users in the coming weeks and will be available to all users over the next few months.

[hallow owl]

GPT-4o real-time voice and vision will be rolling out to a limited Alpha for ChatGPT Plus users in a few weeks. It will be widely available for ChatGPT Plus users over the coming months.

GPT-4o was coming weeks, real time chat is coming months

[odd dagger]

Huh what will it do

Everyone already had gpt4o tho

[delicate hatch]

is there a limit to number of messages in a thread?

[calm shadow]

https://x.com/OpenAI/status/1790130708612088054?t=bgpr-IBhwXMa4ZiLXCrJgg&s=19

Coming in weeks according to open AI most recent updates

[hallow owl]

real time chat is going to be Alpha in weeks to a small selection of Plus users, then coming months for everyone else

[calm shadow]

I have a feeling it will be weeks even though they put that previous detail of being "alpha" to a small selection , we will see with the announcement I'm expecting in the next few weeks

[wind urchin]

Howdy

Ppl how are you

[drowsy quest]

Can we stop updating custom gpt’s every couple of hours behind the scenes? I know I haven’t updated it. (My own custom gpt) Apparently, I can’t handle the line at the bottom advising there’s a new model available without thinking I need to use it…. Thus re-doing the entire list of prompts again…. Just venting cause I can’t figure out what the changes are about.

[proven siren]

Why when I want to update my GPT do I get an error "error saving the draft ?", the custom GPTs have a problem today ?

[bronze glade]

anyone have a gpt for gpt building?

[grizzled flax]

Hey! That message just means that the instructions of the GPT itself have been updated since you started a chat with that GPT. Basically:

• Instructions 1
• Start chat with GPT w/ instructions 1
• Update GPT to instructions 2
• Old chat w/ instructions 1 will tell you you need to start a new chat to use instructions 2
  This happens when you make any changes to the GPT: knowledge files, abilities, etc. I've also seen it glitch a little where it'll tell me there's a newer version of the GPT even if I just started the new chat, but I'm not sure if that still happens. But in short: it doesn't refer to the underlying model itself, just the version of the GPT!

[drowsy quest]

Which makes sense.. except.. I didn't make any changes.
The glitch you mentioned matches what's happening here though. Hopefully, it'll get sorted out soon 🙂

[umbral halo]

Has anyone linked a CSV file to a GPT

I have two CSV files containing a simple list of text (approx 70 words) and have asked the gpt to select 3 words from each list for each input i give it. I would imagine this is a simple task, however it seems incapable of doing so!

I have told it "You must ... etc" and explained multiple times

Has anyone any pointers for this. Or is a pdf better? It's just a simple text list of 70 single words

[pine gulch]

What do you wish to accomplish with this?

[umbral halo]

I feed it short text data. I want the gpt to pick 2-3 words out of the 70, to act as tags for the input.

So I attached the two CSV files (each with one list of 70 words) and told it to pick 3 words as tags. It was unable to do so

For now i've copied the list into the gpt instructions and told it to "Pick 3 tags from the list" and it seems to be working, but i can't imagine this is the best way to accomplish this

Am i missing something re attaching the CSV files?

Seems a very simple task that it isn't able to accomplish

Go two one of two lists and pick three words from each list to represent the string data input

[pine gulch]

Are you prompting it to read from the CSV file, or are you just saying pick three words.

[umbral halo]

Prompting it to read from the CSV file

e.g. "Using the 2 CSV files i gave you, choose three tags from the list to sum up the text input etc"

[pine gulch]

If you don't mind sharing the files I will do testing on my own end.

[strange reef]

why did I renew my account for 4.0 and still show 3.5? Even when switching pages, I couldn't switch to showing that I didn't make the payment successfully, but my bank card received a deduction

[hallow owl]

reach out to support at https://help.openai.com for assistance with this

[strange reef]

I can't find any questions about me

[hallow owl]

ask for help in the support chat on the site

[strange reef]

Is there a magnifying glass in the bottom right corner

[hallow owl]

There is a Chat Icon in the bottom right hand side of the page that takes you to Support Chat

[strange reef]

Thank you. How long will the response time for support be approximately

[hallow owl]

I don't know, but that's the only way to get support.

[strange reef]

Okay, thank you. Thank you for taking your time

[proven siren]

Why when I want to update my GPT do I get an error "error saving the draft ?", the custom GPTs have a problem today ?

[sick lichen]

same

[sick lichen]

what is?

[hallow owl]

you have been told before you cannot solicit on this Discord server

[main chasm]

Ha, I created a four sentence paragraph both sentences similar, but say the same thing I asked GPT to rewrite the paragraph so that the two sentences doesn’t repeat and in many different ways and it would just give me the same paragraph that I gave it. 😂

[toxic merlin]

Is there any way to specify a GPT should use GPT-4, either at a global level or on a conversation level?

[hallow owl]

no, they all use GPT4o

[cobalt tiger]

When I try to add action with google oauth, an error occurs with
Value error, Authorization URL, Token URL, and API hostname must share a root domain.

but google oauth url and token url does not share same domains.
How do I fix it?

[toxic merlin]

That seems like a massive oversight. Especially since a lot of people are finding 4o is terrible at following structured prompts, even when they're clearly communicated in a step-by-step way. I just discovered a custom GPT I was very proud of is now totally useless.

[fresh tapir]

I haven't tested the past week, but I was pulling up a new chat thread last week, designating v4 (not v4o), and then pulling in one of my GPTs (using @). When I did that, it was using v4 from what I could tell.

[humble karma]

Gpt 11 when

[pine gulch]

3025

[wicked prairie]

guys can anyone review my jarvis assistant code that I made in python for ways I can make it better

[slender badger]

If GPTs are now using GPT-4o, does that mean the token limit is now 128k?

[tardy heart]

the token limit on the ui has never been 128k

turbo had a 128k limit on the api as well

[slender badger]

Do you think they'll ever allow 128k limit on the UI?

[tardy heart]

that would be likely at great cost to them so i'd be surprised.

[slender badger]

Even for Teams members?

[tardy heart]

on the api a full 128k context call is 60 cents a call roughly

not including output

[slender badger]

I don't mean 128k per prompt, but overall (most of what I use GPT for involves multiple consecutive prompts)

[tardy heart]

idk im surprised they give as much as they do on the monthly tbh

yes but you have to realize with the way llms do its constant build up in a convo

[humble karma]

Imagine how much it would cost gpt 4o with vision

Will they just save descriptions of each frame so the images won't have to be reinputed?

Poor GPT has amnesia 😢

[tardy heart]

images are tokenized just like text

[humble karma]

Yes I know but don't images cost a lot

[tardy heart]

nope

[humble karma]

How many tokens per picture

[tardy heart]

so on 'low' quality on the api its 512x512 into the model

which is 0.000425

(85 tokens)

lets say you want to use high res at 1080p

thats only 1100

[humble karma]

I thought it would be much more

Do they use something else to extract patterns from images before GPT uses it as an input

[tardy heart]

its tokenized just like text

you can see the pricing here https://openai.com/api/pricing/

[humble karma]

Each pixel would be a token

[tardy heart]

kinda they price it by tiles

[humble karma]

Which means they "summarize" the information with something else before it goes into GPT

Unless they chose to just charge us less

[cobalt lintel]

when will arrive the final voice in chat gpt for plus users ?

[hallow owl]

in coming months

[cobalt lintel]

i heard in the following weeks and not month ?

[hallow owl]

there will be a limited Alpha which is taking place now, general availability to Plus users is over coming months

[cobalt lintel]

how can i be in the alpha version ?

i am on android

[hallow owl]

you need to be chosen, if you haven't heard anything then you weren't chosen

[cobalt lintel]

how can i be chosen ?

[hallow owl]

it's already happened, OpenAI doesn't say how people are brought into Alpha

[neon niche]

Do some people with Alpha now have the new voice access? Didn’t think anyone did yet or I missed that discussion.

[tiny zenith]

Is there a way to feed a github repository into a custom gpt?

[prime peak]

if i make a custom gpt, is is always public or is there an option to make it private?

[supple violet]

private by default

[prime peak]

ty

debating on paying for plus to access them

[proven siren]

Hello, is it possible to integrate your own custom GPT (from the GPT store) into an external site ? Not a Assistant, but a custom GPT using openAPI (in its internal configuration)

[blazing scroll]

No

[hazy hamlet]

What's the keyboard shortcut for sending a prompt, in the chatgpt web UI? I can only send by clicking.

[pine gulch]

Try Enter

[hazy hamlet]

Enter makes a newline. Thanks for the suggestion!

[pine gulch]

That's odd, for me Shift Enter makes a newline and Enter sends my prompt.

[hazy hamlet]

Shift enter (and control enter and control shift enter) all make a newline for me. I am using firefox on windows.

[fallow adder]

Can someone briefly answer:

1. Are custom GPTs using GPT4T or GPT4o?
2. What is the prompt limit for custom GPTs per 3h?
3. Does their output equal 4o or 4t? (4t is much shorter)

[austere peak]

From what I KNOW, GPTs were still using GPT4. If i really want a long talk of back and forth, i prefer taking the prompt i have on the GPT and using it on a fresh GPT4-o chat

[fallow adder]

yeah me too, but GPT4o is so crazily redundant and incompetent at times it's just raising my blood pressure a bit too much

[fervent current]

right now gpt4o is slow

[fallow adder]

which is why I'm exploring alternatives. Given that custom GPTs are only GPT turbos, then this is out of the question

[hallow owl]

GPTs use GPT-4o

[fallow adder]

what's the prompt limit for them?

[mental burrow]

I think they just resize the image, i don't think there is any additional processing before it goes into the model

aren't we talking about image to text though

eh it looks like a fun paper to go with coffee 😄

no worries

i usually have a pretty hard time getting upset with someone acting in good faith 😄
"Hey check this out you'll LEARN STUFF"
"how dare they"

😉

Well i realized too late I responded to a day and a half old message. maybe the OP will read my reply lol.

As I interpret the image token pricing, it seems like they try to resize images to fit into an image that can be composed of 512x512 tiles. No images in this channel but you can see the table at https://openai.com/api/pricing/

It turns a 1080x19120 image into a 116x2048 image for the purposes of processing

[humble karma]

Still quite a lot of data

[toxic merlin]

I heard 4o had rolled out to GPTs but I can't remember where I saw that. I suspect it's true, because my favourite GPT is now acting really weirdly, randomly, and generally not being useful at all.

[runic shell]

Does this mean that there will be exlusives to apple?

[upper tusk]

Where is the new voice mode already? OpenAI promises it will arrive in the coming weeks, but it's already been a month.

[junior aurora]

That's the beauty of "coming weeks" – they can be stretched indefinitely. I'm surprised they used "Apple integration coming by the end of THIS year" instead of "end of the year" to give themselves more flexibility.

[upper tusk]

Then it doesn't need to be announced... it's basically empty words. I would expect more seriousness... after all, we invest a significant amount of money into these companies. Integration, but only in English, and Siri asking GPT to translate what I said in my native language?

[flint hound]

Lol I got excited when I saw chatgpt had an update available in the app store

[jade moth]

So I got this thing going on, with any chat within GPT that sometimes it freezes during generation of text and I have to refresh my TAB to see the frozen unfinished generated message to then click the re-generate button while this browser TAB is barely using any RAM as I got very good super computer, is anyone else facing this stuff where GPT just freezes during generating context?

Its annoying cause it's wasting prompts as we are all being limited / hours depending you're using free or subscription.. it just sucks for such high-end tech stuff

[bright mirage]

Oooh! Since when? Is that a new update? I didn't get any pop-up about it.

[hallow owl]

about a week ago

[bright mirage]

Do I need to change anything in the GPT's settings or is it just automatic?

[hallow owl]

it's automatic

[sly wind]

Hi, OpenAI Team,
I want to know the specific reason why i can not edit / publish the GPTs in GPT Store related to my domain. I have double checked my domain and still got no clue.
It all went well for like 5 months since we create the first GPTs within ChatGPT back in January.
But like 2 weeks ago, when i try to update my GPTs. It keeps reminding me that "Your GPT can’t appear in the Explore GPTs page because it violates our Usage Policies. See below for the specific issue".
We haven't made any changes along the way. And it's really bothering my team.
Could you please help me understand what caused all of this.
Best regards,
John

[jade moth]

Hi John, they dont really care all they do is re-direct you to their website for you to waste more time trying to state your problem and come to a resolution of understanding . Sadly they dont care as long you pay for their subscription it's all good.

[grave cargo]

Do gpts have access to the internet?

[sly wind]

that's really frustrating if it is true

[jade moth]

Lately I also have but nothing been experiencing problems with chatgpt, throught this discord server there is no way of receiving the neccesary customer support. **Please note that our assistance as Discord Moderators is limited to matters pertaining to the OpenAI Discord server. We cannot assist with issues regarding OpenAI services.

We recommend reaching out directly to OpenAI Support at https://help.openai.com/ — thank you for your understanding.** so yeah, goodluck mate.

[mental burrow]

yup

they can hit external APIs

[stone python]

Hello

[grave cargo]

theres no way to turn on internet just for gpt 4?

[mental burrow]

look under the "Capabilities" section when you configure your GPT

[wraith ginkgo]

`` Server Tip: ``

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

[grave cargo]

What about it?

nevermind i was mainly focused on public gpts