chatgpt-discussions

[faint badge]

I suppose because are not plain text. ?

[misty osprey]

It would be nice if it could accurately intepret stuff ina a table format

[faint badge]

Any fixes yet?

[tough bramble]

Not that I'm aware.

[faint badge]

Damn. I need to use gpt4 in this chat again. Any workarounds?

[tough bramble]

Not that I'm aware.

You'll need to copy the text.

If it has images, well, you'll have to upload again.

[faint badge]

The conversation is long. If I use share link, browser capability can be used to read the entire conversation?

[tough bramble]

You could create a GPT with the transcript added as knowledge

[brisk path]

Maybe?

Looking at the robots.txt

[orchid eagle]

what can possibly go wrong if i give chatgpt my entire calculus text book

[brisk path]

GPT-bot is obviously not blocked by it's own company

Nevermind

ChatGPT can't access it

[faint badge]

Mmm... Yeah. It seems that can be a solution. However, what if I don't want a create a custom GPT. Can I just upload the conversation as a file?

[faint badge]

I asked and GPT itself said can't do that

[tight stone]

how do i fix a bad gateway error?

[outer seal]

use a good gateway hope this helps

[tough bramble]

Just create a GPT. They can be private.

[faint badge]

The transcript, just copy and paste to a file or use export option?

[tough bramble]

Whatever works. Just need a text file with the conversation.

[white tundra]

anyone's gpt just not wanna work anymore?

[ivory plaza]

I'm paying for GPT Plus. However, I don't get to choose which model I'm using anymore. How do I know I'm actually talking to GPT4? (The model I'm paying for)

[outer seal]

look for gpt4 at the top hope this helps

[honest jay]

Custom GPT or just ChatGPT in general?

[honest jay]

Are you using the app or web?

[ivory plaza]

web

Oh, interesting... After logging out and back in I see a symbol on the top now.

[honest jay]

In the top left corner of the new chat screen you'll see "ChatGPT 4", "ChatGPT 3.5", or "ChatGPT Plugins." This is the new location of the model dropdown.

[ivory plaza]

I must have had an incomplete update in my browser or something.

Is there a way to have the app running on PC? The app and web versions have different features...

[honest jay]

I don't think so, maybe through some kind of Android emulator? But there are only native apps for iOS and Android currently. I think the only feature in the app that's not on web is voice conversations? Otherwise, web has a few more features than app.

[ivory plaza]

Thanks

[white tundra]

the gpt 4 updated one

[honest jay]

What's the issue you're having?

[white tundra]

keeps failing to finish its response when I upload a picture

[honest jay]

Does it say something about like "error in network stream" or something? And is it only happening when it's analyzing images that you upload, or does it happen when generating any text at all?

[eternal rock]

This is a Safe for Work Discord Server

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

Posting NSFW content will result in a permanent ban.

[orchid eagle]

why does GPT builder keep breaking

[white tundra]

it keep saying time out, for only images

I think because it takes forver to generate

[white tundra]

yes something like that

[honest jay]

Can you try in a different browser, or on a different device? Do you have a VPN installed?

[white tundra]

I am in chrome

and using a vpn

[honest jay]

Can you try:

• A new install of a non-Chromium browser, like Firefox
• Disabling the VPN
• Disabling the "Web Protection" or similar feature in your VPN, separately from disabling the VPN itself
• Trying a browser on your smartphone
  To see if any/all of these let it work for you?

[sick compass]

is chatgpt4 insanely slow for anyone else too?

[white tundra]

It sometimes gives network errors

But it's just very slow

Even in mobile

[toxic flax]

guys i hear gpt 5 releases in an hour

[elfin birch]

reputable news source

[cursive path]

So im running into an issue of an over agressive content filter. Anyone else had issues like it lately?

[honest jay]

Can you share any specific examples? Do you use custom instructions?

[cursive path]

No custom instructions. I think i understand why it might be flagging the messages but i dont believe it "should" flag it. Im analayzing my family tree and i converted a gedcom file to json, and giving it the json of even one entry (person) itll flag it. I feel like if chatGPT has no problems with putting full emails, password, even social security lol that i should be able to analyze family tree info

[crude slate]

is there a way to add an attachment to a editing message

i want to save space and keep my formatting clean in my chat

i made a response in a style i really like from parts of many different responses

is there a way to edit that into a response to help maintain style?

[honest jay]

Hmmm interesting. I can't think of anything in family tree data that would cause a flag, are you saying you think it's because you're inputting full names? Long shot guess here, but do the JSONs include the word that starts with an "s" and ends with an "x" that indicates "male/female"? Not saying it here in case it's a potential flag word in both cases 😁

[cursive path]

good idea, it does. But deleting it doesnt change the filter warning unfortunately

[faint badge]

Ugh, ""Hmm...something seems to have gone wrong." after updating model.

[faint badge]

I'm facing a problem with a chat that changed model back to GPT 3.5.

[crude slate]

i wish we could switch this back and forth

i have a suggestion you could like here #1173361597536800930 message

[honest jay]

Super strange, any way you could share an anonymized example of one?

[cursive path]

apparently not, i got a 5 minute timeout lol

im trying a workaroud of making the json less structured

[honest jay]

Maaaybe timeout for too much text? Kind of a hacky workaround, but could you make the anonymized example, then take a screenshot of all the contents, and then post it in a channel with pics like #off-topic or #community-help? I'm mainly just curious if I can spot any other potential trouble words.

[cursive path]

ill try that now

[compact gulch]

Source: trust me bro
But I mean why would they do that, they literally just released GPT4 turbo

[gritty grail]

Yeah, mine's not working either. I don't know why

[cursive path]

posted to off topic

[toxic flax]

becaus

[faint badge]

I also reported a bug, if you could like it too:
https://discord.com/channels/974519864045756446/1172439273945968710

[compact gulch]

Because.. ?

[honest jay]

Took a look, interesting! I'm surprised that's getting flagged. Other than the "s" word we talked about that's still in there, otherwise I wonder if the general abbreviated nature of all the lines might be causing inadvertent problems. I recently sent a message where I took the letter "e" out of every word and I got a flag notice--I think a bunch of abbreviated words can sometimes alarm it

[cursive path]

ill try that next, thanks

[gritty grail]

Are you all at work now

[prime swallow]

I want to help ensure you are aware of https://openai.com/policies/usage-policies.

Part of disallowed content, "Activity that violates people’s privacy, including:

Unlawful collection or disclosure of personal identifiable information or educational, financial, or other protected records"

OpenAI says: "We want everyone to use our tools safely and responsibly. That’s why we’ve created usage policies that apply to all users of OpenAI’s models, tools, and services. ... Repeated or serious violations may result in further action, including suspending or terminating your account."

You say, "I feel like if chatGPT has no problems with putting full emails, password, even social security lol that i should be able to analyze family tree info"

I want to suggest that even if the AI does not complain, and you don't see a content warning you are not safe to break allowed content. Do not give the AI things like emails, passwords, and social security numbers, as this falls into what reads to me as disallowed content.

Full names or other PII seem to as well.

Good luck.

[topaz otter]

Is there any way on Android to ask ChatGPT something using your voice and have it respond? I saw someone do it on iOS with Actions or something like that? Or is there a widget?

[toxic flax]

?

[compact gulch]

Ok, what is your source?

[toxic flax]

wasnt aware i needed one, mr source police

[honest jay]

If you long-press on the ChatGPT app icon, you'll see a shortcut to launch the app in voice mode. That's the shortest shortcut we've got for now, doesn't seem to be possible yet to integrate with something like a Google Assistant voice shortcut

[compact gulch]

Lol. But why should we believe you if we don’t have anyway to verify your information

[cursive path]

I understand that, but all of the information is publically available online at ancestry. None of the information is protected

[toxic flax]

WELL since you insist so strongly on a source, my source is that I made it up

[twilit furnace]

does anyone feel like gpt 4 hasn't been as intelligent the last several days?

[elfin birch]

No

[toxic flax]

No

[compact gulch]

Nope

[topaz otter]

@honest jay on Android? When I long press I don't see any options

[prime swallow]

Good luck. I hope you never have to deal with an account appeal, they seem to take a long time and have a lot of waiting.

[compact gulch]

See? Now that we know the source we can clearly see that it is legitimate information because the source seems legit

[toxic flax]

Totally understood. Glad you trust the completely factual and verifiable information I have provided you with.

[honest jay]

Yes, I'll share a screenshot in #off-topic

[cursive path]

yeah im not trying to get banned obviously. Its a bummer that it seem genealogy is off limits because its probably one of the best use cases for AI that isnt being done right now. Im trying to strip away info from my json so i can figure out a good balace between relationship context and identifiable info that serves no purpose for what im trying to do

[prime swallow]

If you sub in fake names, and strip out all other PII, you are probably fine.

And the model can surely guide you in how to do that too

Real researchers do that, it's a known protocol, to mask that stuff and still keep track with external records

[honest jay]

I have a hard time thinking that the AI is recognizing that you're sharing partial genealogy data and thinking "this is probably illegal"--are there any comparable situations where something like this has happened? Like Eskcanta said, I feel like it's more likely that you wouldn't be flagged, but could still get dinged on the back end if someone took a closer look

[prime swallow]

You scared me talking about social security numbers, especially

I'm not a mod, and I don't work for OAI. I just hurt every time I see the posts from the people about their having been suspended, what do they do now.

[cursive path]

the social security numbers was an exaggeration for sure. But it definitely doesn't seem have a problem usually with sensitive info in general. I'd imagine its just the volume of linked identifying info thats getting me flagged

i might just be safe and use GPT 4 vision to analyze a screenshot of whole family trees

[prime swallow]

Yeah. Thing is, we do not get told why anyone gets suspended (maybe not even if it is us, if some of those folks telling about it are to be believed).

But we do get to know TOS and disallowed content.

I am super careful, myself, about following allowed content rules.

I have had hundreds of content warnings over the nearly a year. Some more intense than the standard first level orange warning flag.

The content warnings themselves are not a problem, the problem is only if you are breaking allowed content.

[honest jay]

Right but that's the primary point of interest here: drewsny is being flagged, and it just doesn't seem like it could be because it's genealogy data. To connect it to one of your other suggestions: if drewsny anonymized all the data and re-submitted, I'd expect they'd still get flagged, because it's not like the AI can tell whether it's the anonymized version

[prime swallow]

Depends.

See, a lot of that stuff does not need to be personally identifiable, or look like it.

'Gen 0 male' and 'Person1' is not going to make anyone think there's someone getting their PII shared.

Do we need locations for a family tree?

If so (and I never needed locations in my experience with family trees) use clearly not useable info, like "Location 1".

Do we need social security numbers?

I 1000% do not believe it.

But sure, 'Social Security 1'.

Nobody's getting suspended over that.

[cursive path]

In genealogy we definitely need as much identifiable info as possible is the issue. When trying to determine if someone is related to the other person, things like birth records, locations etc are the primary ways to figure that out

im trying to get AI to automate it but obviously the problem im running into is i need to pair back the sheer volume of the info

[prime swallow]

Ahh. So... this may not be an ideal use case for the AI 😦

[cursive path]

it is the ideal use case for AI, just not openAI's AI tbh

and thats fair

it is what it is

[prime swallow]

Yeah. Might want to look over this locked thread 😦 https://discord.com/channels/974519864045756446/1173134085040910406

Some stuff, OAI's model, with the rules of use, is not ideal for, even if it may seem ideal for.

[honest jay]

Can you try like a "blank" version? This might be the best way to see if it's likely to be from the formatting/abbreviations, or from the content itself. Not to keep guinea pigging you for potential policy flags 😁 but I feel like no personal info, anonymous or otherwise, would be a safe test to see if it's maybe just all the abbreviations that are scaring it

[cursive path]

Yeah the blank one is fine. Funny enough its fine if you just give it all the json info for 1 person, but as soon as you give the json for 2 people it flags it

[honest jay]

Oh interesting, I thought it was happening with just one person at a time too? So blank = good, one person = good, but two people = flag? And you're sure there's no formatting/significant content differences between persons one and two in this third case?

[cursive path]

Yep thats what it seems so far. Also i made a new json file with just the tags "NAME", "FAMC" (family tag), "FAMS" (family child tag), and "BIRT" birthplace. I gave gpt 4 the whole json for just my immediate family going back 3 generations and it was 100% correct when i asked it what the family relationship between all the people were

So i can work with this

my "big" idea was to give my mom a custom GPT to help her with genealogy. Im going to have it use the family relationships pulled up via RAG, and shes going to be able to feed it a whole article and have it spit out if anyone in the article is possibly related to the family

[honest jay]

Interesting example you've shared here, thanks for all of the updates! Good insight. I hope you keep finding productive ways to work with your lawfully collected PII 🙂

Your mom will love it

[cursive path]

GPT 4 vision has already been a game changer. Before to look at records like swedish census data, she would have a spreadsheet of common terms and have to go one by one on cencus table cells. Now with GPT-4 vision shes just giving it the whole image of the census document and it will give her the headers in english, and whatever else she asks

[topaz otter]

@honest jay just FYI, yeah I tried opening the ChatGPT voice shortcut using both Google Assistant and Bixby, neither will recognize the shortcut, even when I put it on my home screen. I believe because the shortcut doesn't have a name.

So doesn't seem to be possible to open voice convos hands-free, at least not yet =/

[honest jay]

I'm wondering if maybe Google has some kind of rule against hooking its Assistant shortcuts to other assistants. You can see a list of all your possible Assistant shortcuts (idk about Bixby) by going to the Google app, then Settings, then Google Assistant settings, then Shortcuts. ChatGPT just isn't listed there, though comparable app shortcuts are. Maybe someday, or maybe it's possible through some kind of Tasker/IFTTT integration already!

Edit: or maybe OpenAI just needs to add something to make it show up there!

[topaz otter]

Yeah it SHOULD be there, because it shows all recently used shortcuts, ChatGPT is literally just missing. Odd for sure.

[honest jay]

I was chatting with ChatGPT about this earlier today incidentally and at least according to it, Assistant hooks have to be added separately from the long-press app shortcuts. So hopefully OpenAI can add this, so long as Google isn't blocking them!

[hidden nimbus]

so gpt-4 turbo is now the default gpt-4 model, does anyone know what context length does it use in the web interface?

[honest jay]

Click dis
#openai-chatter message

[hidden nimbus]

thx, would be good for openai to clearly inform stuff like that tbh

[shadow finch]

I don't get it. I haven't seen one useful custom GPT yet. Why would someone use a gpt over plain chatgpt?

[honest jay]

I imagine it's functionally quite variable right now, as is the case with the rest of the limits on ChatGPT Plus right now. The "danger" in trying to set things in stone like that in the public discourse is that you end up sewing more confusion when the limitations inevitably shift again.

Though it would be cool if they had some kind of automated hub we could go to that automatically updated what all the current limits are right now! Msgs per 3 hours, DALL·E images per day/per message, functional context limit, stuff like that. Maybe someday!

[honest jay]

Have you used the custom instructions feature much? Custom GPTs are like a more flexible version of that feature. I'm sure you use ChatGPT in different ways--you could create custom GPTs for your most frequent use cases to prime a GPT with context you otherwise usually have to start from scratch on.

[shadow finch]

custom instructions? I don't think I have. I've created my own custom GPTs though. What are some good use cases of custom instructions?

[honest jay]

Oh man, so much! It's permanent context that ChatGPT keeps top-of-mind throughout an entire chat, so you never need to reprime it with certain critical information. But, if you've already experimented with custom GPTs, you're already at least a little familiar with this concept.

[topaz otter]

@honest jay honestly it might also just be because the app has the exact same name as the shortcut...because you can tell Google Assistant to open the app and it will, but can't open it and type, or open the voice shortcut itself.

[outer seal]

hooking them up to external apis

[shadow finch]

Oh, true

[rugged igloo]

my chat gpt broke on my pc lol

Our systems have detected unusual activity from your system. Please try again later.

[honest jay]

Do you have a VPN installed?

[rugged igloo]

@honest jay i was deleting old chats dalle-3 it lagged when i loaded it in and all of sudden chat gpt just crashed

its back to normal wont do that again

[tough bramble]

Why is the GPT-4 limit still like 10 messages per year if it's become 3 times cheaper?

Kinda annoying having to spend 10 hours on a GPT because you have to take a 3 hour break every 5 minutes..

Which also begs the question: Why no 3.5 GPTs?

[honest jay]

Cheaper for API customers, separate from ChatGPT. I think the limit is still 50msgs/3hrs

[sage meteor]

Did dalle-3 make improvements on generating text?

[crude mortar]

Bro each instruction I ask chat GPT it appears the robot verification window, anyone is having same problem?

[outer seal]

you can build them w/ the api

i assume they want to show off new 4turbo in the app for now

[outer seal]

because we get 4x context

[tough bramble]

That's ridiculous. Most chats don't get anywhere close to that.

That's like a pizza place advertising "Come on in, our pizzas are now more expensive, because we now have additional bigger sizes."

[outer seal]

i get there sometimes with code interpreter

but yeah there should be 3.5 versions of the default tools (esp since it can use them now), 4 is overkill for most cases

maybe if they had a token limit instead of a message limit 🤷

[minor vortex]

Has anyone run into an issue with GPT while trying to upload a lot of images? I'm trying to upload a manual and can't find a more efficient way to do it.

[supple zephyr]

i believe you can just upload file

[tranquil silo]

Hello guys, I am trying to find out how I an change email address and password on ChatGPT.

I have signed up with my email but now I want to share my subscription with my PA.

[gritty ledge]

Is there something wrong with ChatGPT again? I can’t seem to log in to my account and use it normally？

[worn thorn]

Same here

[worldly mantle]

cant even get through a message without erroring out

[rugged igloo]

did you guys an error message with suspicious activity if so just wait a bit

[twin flame]

Consider repetitive things you ask GPT, for instance when I am learning a new topic I have GPT create a course outline. The old way requires a new conversation and re-entering the same prompt with a different topic. Now I can just feed the structure I want the course to be to my GPT and start him up whenever I want a course

Now imagine another gpt that takes a course outline and acts like the instructor for the class following your outline to the letter (where the outline is for 12 weeks and the instructor can take you day by day)

That is just one simple example that is useful to me, I am sure there are many others. But we are going to see a lot of junk before the great ones come to the surface

[worldly mantle]

gpt builder tarded out and keeps repeating itself infinitley

[shadow finch]

I agree with you on this point. Seeing a lot of junk 😆

[drowsy tinsel]

u just haven't used it enough to realize what it is good at

[shadow finch]

Probably true

[eternal geyser]

It actually does seem difficult to get a good use case outside of the general model it still uses - I recommend focusing on the data/files you feed them rather than information the general model already has.

[fierce escarp]

Hi -- how are people creating so many fairly elaborate gpts so fast? I created like 2 silly ones and got hit with a rate limit for the next 3 hours lol. Seemed to hit the rate limit far faster than previous use

[eternal geyser]

Test with API and playground, then make the GPTs

[rugged igloo]

im making the most powerful philosopher ever created

[waxen river]

Is there any GPTs doing this? I want to create course outline for DEVops Engineering.

[shadow finch]

What kind of files have you had good results with?

[eternal rock]

`` Server Tip: ``

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

[twin flame]

Ive got one I built I shared it in the custom-gpts channel. Search for "course" and leave me some feedback

[waxen river]

will try thanks.

[twin flame]

The title is: "create a course for anything"

If you like it let me know, I built another gpt that I use with it that takes the day you are working on from the course outline and begins to instruct you and gives you a synopsis of what was covered so that you can pass it to the instructor the next day to continue where you left off.

[strong stratus]

Did the update break dall-e 3. Chatgpt keeps telling me it can't generate images but dall-e was suppose to be built into it along with the other stuff so you weren't having to choose from drop down menu

[hasty laurel]

same here... takes forever to generate a response and when it gets to the end of thinking, an error message pops up...

[honest jay]

Are you on web or app?

[strong stratus]

Web not my mobile phone

[honest jay]

In the top left corner of the new chat screen, do you see ChatGPT 4?

[strong stratus]

Furthermore it worked prior to the update (the gpt 4 on the website even says dall-e 3 and yes it says chatgpt 4

Well I yelled at it and it started to generate o_O

it quit telling me it couldn't do it lol

[honest jay]

Glad it started working! If you ever want it, this GPT is DALL·E only. https://chat.openai.com/g/g-2fkFE8rbu-dall-e

[strong stratus]

Good ole all caps yelling at an AI lmfao

[upper orbit]

chatgpt doesn't work every time you verify a real person. Server not ready yet? Or is it my account?

Unable to load history

[tough bramble]

3.5 v1106 is really, really awful.

[odd flint]

Too bad ChatGPT voice - Did anybody noticed that on Android app ChatGPT, the audio voice has become worse since some latest updates ? In the beginning, it was fine, high quality audio, but now, it's terrible, low quality and too low, even with the volume at the maximum. I also can't transfer the audio conversation via Bluetooth. Too sad. I have another smartphone that I didn't updated, and using an old ChatGPT version, I can hear how audio has become poor quality now. The audio chat was the most amazing thing I've seen, but now, I can hardly hear anything. 😲😲😲

[honest jay]

Voice definitely has some weirdness with how it handles audio as a phone call instead of normal media. I'm wondering if maybe the difference you're observing is actually maybe from two potentially different ways your two phones handle those audio channels. That is, I'm speculating that maybe the ChatGPT app behaves a little differently on different phones, based on the speaker configurations of the respective phones. Hopefully it's something that is cleared up in an update--you're not the only one to report that Bluetooth issue either. May be worth a post in #1070006151938314300 or #1070006915414900886, depending on your interpretation 😁

[prime quest]

I have had this experience several times

[proven olive]

I always get "Error saving GPT" when saving the created gpt. I chose "Only people with the link." When I select "Only Me" it doesn't appear. why is it like this? Have you ever encountered this situation? And it seems that gpts often has problems in the past few days. I remembered, I had this problem yesterday too.

[odd flint]

The two phones are equal. 2 Samsung Galaxy S20FE, same content, same configuration. I installed ChatGPT on both phones. They were the same audio quality in the beginning, with the same ChatGPT version. Then, in one of them, I upgraded, and immediately, the audio quality was gone. They changed something on ChatGPT and how it handle audio, like a phone call. On the other one, I didn't update, so, I can still enjoy the good audio OpenAI provided in version 1.2023.285. Currently, the bad audio phone is using the latest version, which BTW, I updated again few hours ago, to version 1.2023.312. 👍

[drowsy tinsel]

got that too, try doing it again and it should work

[keen walrus]

Does anyone know if there is a way to upload an excel with my financial forecast and have chatgpt make changes using prompts?

[late adder]

Yeah after using this for HOURS today, I can confirm turbo is VERY much acting like 3.5. Just talks and talks so much no matter how much I stress for it to be concise.

Seems to LOVE lists just like 3.5

[rugged igloo]

so your only able to upload 10 documents i thought it was 20?

[vagrant moat]

Nah only 10

[sick linden]

Is GPT4 down AGAIN ?

[rugged igloo]

im using it right now so nah

[sick linden]

I have a chat who gave me an error one time, and since this day, this chat is absolutely broken

And doesn't work anymore whatever i do

[drowsy tinsel]

oh for the first time I saw 'continue generating' in chatgpt

im letting it do document processing for super long texts

[outer seal]

[drowsy tinsel]

The thing is it asks at ~1000 token

so making it generating the full 32k requires u to press the continue button 31 times

input 4k token output 4k token very challenging

[pine tulip]

it's so slow rn

Super slow

Extremely slow

😢

[drowsy tinsel]

yea same

[pine tulip]

😭

[serene grove]

Anyone else experiencing "Network error" ?

[pine tulip]

not yet

[serene grove]

"Hmm...something seems to have gone wrong.", it's even worse lmao

[cunning roost]

Guys i have some question:
Is it possible to train an own gpt which calls other trained gpts? I can't find any good example

[pine tulip]

yea, but costs about 2-3 million dollars

if have that kind of money, you can inquire here: https://openai.com/form/custom-models

[cunning roost]

i trained some gpt which is specialized for some tests and now i want to train like a coordinator which coordinates which gpt it has to call

[pine tulip]

yea no

[rugged igloo]

good luck need alot of money for that

[lofty flax]

Good morning people! Is it just for me chatGPT is quite slow at the moment? 🙂

[pine tulip]

not possible at the momentonly way that I am seeing how this would work, is an an interface which needs to be implemented in every GPT via API so your "coordinator" can communicate with them

[pine tulip]

yes, very

[serene grove]

Good morning, yea GPT in indeed very slow. https://status.openai.com/ has nothing to show though

[lofty flax]

It's all those darn GPTs 😉

Get your own network, will ya?

I'm actually building this GPT functionality myself but for open source

So I can have it all in a local instance without the need for internet 🙂

[cunning roost]

so the best solution to create many customized gpts and use the specific one for my problem?

[pine tulip]

sounds cool, but at least for my usecase, all the open source models aren't good enough yet

[serene grove]

Based on what training ? It needs TONS of data to be self-hosted

[lofty flax]

I will give you a hint without giving the solution, fine tuning smaller models and combining them.. 😄

[pine tulip]

there are a few good ones out there, but yea not comparable to GPT4 or even 3.5

[lofty flax]

The glaring huge issue with the bigger models, is the bigger model you have the harder it is to make sure the output is safe and in the correct format.

[pine tulip]

So what? That doesn't help in any way if I don't have a server farm

just a few little improvements, which still can't compare

[lofty flax]

It's very few instances where you actually need all that information, especially if the communication is agent to agent

[pine tulip]

thats why I said "at least for my usecase"

I tried a lot of them

[lofty flax]

You should build up to your usecase by making specialist versions with smaller models, then combining them to get up to your complexity

[pine tulip]

The thing is, it still can't compare

It works for some people sure, but not for me

[lofty flax]

What is your use case? 🙂

[pine tulip]

lot's of different things

[drowsy tinsel]

for some reason it refuses to generate after like 1500 tokens

[pine tulip]

thats why I would need a serverfarm if a want something specialized for every usecase I have

[drowsy tinsel]

it just gives generation error

[lofty flax]

If speed is not an issue, you could set up a dynamic system where you are storing the message between agents and drop the current hosted AI model and start hosting the next

Then you will get the the same complexity, but it will be slower.

without a server farm 🙂

[pine tulip]

It would still need tons of data though

I don't have a 100tb drive

[lofty flax]

And why would you need 100tb? 🙂

Sounds like a design flaw to me without even hearing about the usecase.

[pine tulip]

Because I like to have an AI which can help with several different topics, and for that it would need tons of data. What are you even trying to achieve? Good for you if it works out for you. It doesn't for me though 😇

[lofty flax]

If you want an AI that can help out in many different topics, let's say it's an assistant, you have to build the complexity up from the first stone yourself. I would not rely on the chatGPT functionality for that as you have too little control over the output. What I would do to reach the goal of having a very smart assistant that does exactly what I want it to do is starting small, fine tuning 7b falcon or mistral's for small computational tasks, when they do what I expect them to do, combine them in a flow to build up the complexity. You can achiveve different types of flows by building the flow yourself between agents.

And, as you are using open source, you decide the cost for the end user and if you want to selfhost or not 🙂

[pine tulip]

wrong channel + It wouldn't work out for me. I don't get why you are trying to explain something to me I already know doesn't work

I see you writing.. please not again..

[lofty flax]

It would work out for you but it would require some actual work from your part. It feels like you are after some quick fix for your immense problem 🙂

[pine tulip]

With that work I would need to do, I could found a third company or could better use it for quality time with my family

Thats completely overkill

You are really trying to say I could create a ChatGPT competitor in my basement..

Thats delusional

[lofty flax]

If you do not believe this is true, I'm sorry to say that you do not really understand the underlying technology well enough. I do say this without any condescending intent at all but I do hope it spurs you to learn more about it 🙂

[rugged igloo]

Message in conversation not found what does this mean?

[lofty flax]

I wish you the best with your project 🙂

[pine tulip]

[lofty flax]

When I'm done with my own project I will share some of this functionality that I'm talking about 🙂

[rich widget]

In what context are you getting this error? Is it in a chat you are currently having in chatGPT?

[rugged igloo]

@rich widget was talking to chat gpt of a file of political philosophy and i asked it something and it gave me that error

custom gpt*

[rich widget]

Try uploading the file to it again or refreshing the browser

actually, try refreshing the browser first then uploading the file again if that doesn't work

[rugged igloo]

i refreshed and my conversation deleted lol

[rich widget]

🤦‍♂️ Is your conversation still on the sidebar?

[serene grove]

Basically, llama.cpp ?

[rugged igloo]

yea its still there i wonder ifs the file giving issues ill look into it

[rich widget]

Is it your own GPT?

[rugged igloo]

yea \

[rich widget]

Yeah, I have issues where mine will sometimes forget all their instructions. I think they use threads to manage conversations. Not sure if there are limits for the threads but if I have long interactions, the GPT will stop using the instructions I set for it.

[rugged igloo]

intresting it does have 661 pages since i combines two pdfs into one

[rich widget]

I had a 6000 page pdf for one of mine. I had to split it into 4 pdfs

[rugged igloo]

jesus dide

[rich widget]

If the GPT is just for your use you can re-prime it by asking it to recap the conversation and then get it to output the custom instructions. That seems to be a workaround that works for me

[lofty flax]

Yeah that one is very innovative! 🙂

[rugged igloo]

yea it seems that file it just broken rust ima delete it

[rich widget]

NP. Good luck

[narrow galleon]

I have GPT Plus as well, and I noticed the update interface, but it still only shows 3.5 and 4 as models. Is there actually a selection for 4.5, or is this entire new update considered 4.5 itself

[tardy totem]

As far as I have picked up in this chat earlier, GPT-4 is the new version.

I've not noticed any speedup though.

[pine tulip]

never heard about gpt4 turbo referenced as 4.5

[pine tulip]

gpt4 is out for half a year lol

[normal quarry]

there is no GPT 4.5, it's GPT-4

[rich widget]

Sam Altman announced on X that chatGPT is using the new GPT4 turbo model for all plus subscribers

[rugged igloo]

hi robert!

[normal quarry]

hey 🙂

[tardy totem]

I meant that GPT4-turbo is the one you get when you select GPT4 now.

[pine tulip]

yea right

[rugged igloo]

robert sorry to bug you but if i combine two pdfs together discussing different topics will gpt still be able to read it?

[tardy totem]

It should be called GPT-4-"Turbo"

[normal quarry]

don't combine them together, upload them both and then your GPT can refer to both

[dense halo]

hi.

[rugged igloo]

@normal quarry was just wondering because was trying to save room for pdfs since i can only upload 10

[dense halo]

i think im the person that broke this

how do you people even know about the prompt i engineered?

[normal quarry]

how long are each PDF that you want to combine?

[dense halo]

im the person that wrote the software generator a.i.

[rugged igloo]

@normal quarry 800 pages 767 pages .....combined

[normal quarry]

that should be fine

[normal quarry]

two of my GPTs use 1000 page PDFs without problems

[dense halo]

sigh

can i talk to someone about this please?

[pine tulip]

broke what

[dense halo]

i wrote the a.i. software generator prompt that works.

[pine tulip]

What?

[dense halo]

so now my service and code snippet abilities are restricted

my api key only sends me 3rd grade level language

this is really maddening

[pine tulip]

You live up to your name. You speak in incoherent riddles

[dense halo]

nah. i wrote a prompt, and i believe this company is trying to steal it.

[pine tulip]

what company

[dense halo]

openai.

because ive brought it to your attention with at least 5-6 sources

[pine tulip]

you wrote a prompt in chatgpt?

[dense halo]

yes

the api]

[pine tulip]

so it's theirs

[dense halo]

it is?

[pine tulip]

Oh I don't know about the api. But if you input something in chatgpt and don't tick the box that they can't use it for training purposes, they can use it for training purposes

[dense halo]

i wrote a prompt in notepad, pasted in chatgbt

i dont know what to do. i think im being swindled majorly right now, i need compensation.

im disabled. i need compensation.

[pine tulip]

I still don't get what you are even trying to say

[dense halo]

i will change a.i. service providers

[rugged igloo]

bro can you leave please

[pine tulip]

good look with that, nothing is comparable to ChatGPT at this point in time

[dense halo]

and compete to the top then, just like you guys are, only i think im smarter.

[pine tulip]

Okay smart guy, you can't even explain to us what your problem is

You wrote a prompt, sent it to the API and now OpenAI is stealing it from you, for which you want compensation? lol

[serene grove]

I wonder what's the size limit of the PDF input.
Is it measured in characters or overall filesize ? 🤔
Can it recognize pictures ?
So far i didn't have any problems with it though!

[normal quarry]

OpenAI doesn't specify what the limits are.

[serene grove]

Yeah. Well, at least now i found a small trick. In GPT4 (browser) we are limited for about ~4000 characters input. But, what we can do, is just creating a PDF file with a longer/larger input. ""unlimited"" prompt is now a thing!

[rugged igloo]

you can do that lol

[neon lance]

pretty sure the limit is decided by file size

[warm gull]

512mb

[rugged igloo]

yea but i feel the longer a document gets or pdf the more gpt struggles reading it and get errors

[tardy totem]

Does the free version give you the ability to do internet searching?

[normal quarry]

no

[tardy totem]

Thanks.

[tardy isle]

Help article doesn't tell exact context window and model name... Where I can find it exactly? (Well, guessing from API documentation, only model that has cutoff date of 2023 Apr is gpt-4 turbo with context length 128k.)

[tardy totem]

the one I use in my api app is gpt-4-1106-preview

[tawny crater]

ugh did it degrade again? it can't remember a command to remove dialogue for more than 1 damn entry

[pine tulip]

What are those file size restrictions?

The size of text, document, and spreadsheet files is capped at 2M tokens per file, with a hard limit of 512MB per file.
For images, there's a limit of 20MB per image.
Additionally, there are usage caps:
Each end-user is capped at 10GB.
Each organization is capped at 100GB.
Note: An error will be displayed if a user/org cap has been hit.

[pine tulip]

They did actually

see above

[merry copper]

I could not find any "hard" statement about the policy regarding uploaded documents in custom GPT´s.
Is there any chance our documents might be used for future training ?

I know there is a checkbox "Use conversation data in your GPT to improve our models" that you can check/uncheck but it seems not entirely clear if a uploaded document is considered part of conversation data.

Any clear statements i heard are specifically about files uploaded in the API section or within the Enterprise environment.

Can anyone of you point me to a source ?

[rugged igloo]

your not alone he was very vauge on alot of stuff

[pine tulip]

When in doubt, assume the worst. So yes.

Just tick that box and use it that way

[merry copper]

Obviously i can tick the box but since it is only talking about conversation data then uploaded docs could still be excluded.
I think that is the most important question to answer ... especially if there will be a shop where you can monetize your GPT(s).

[near vine]

i have a question, lets say you upload a pdf file of your coustom gpt's that contains guidence about cyber security stuff that basically says raw stuff which in fact in general chatgpt forbids to talk about it, so in the coustom gpt which would be made on that will it give the same response as a regular gpts would do "I cant assist with that" or would it ignore it since its a coustom gpt

[tawny crater]

my god, i have to tell it EVERY TWO entries to remove dialogue.

[rugged igloo]

i made the best philosopher gpt

@tawny crater have you tried putting it into your custom instructions to make it so it doesnt do that?

[pine tulip]

yea sure, but as far as I am aware there are really no more infos on that, at least in all their blog posts (source: I read them all)

[merry copper]

thanks. is there a official channel to ask ? the help center does not have any contact form ... (at least i did not find it)

[pine tulip]

you could contact the support, maybe they can give you more info on that

[rugged igloo]

@ruby wren you mean the custom gpt ui now?

intresting

[tranquil wigeon]

can chat access the internet ?

or is still offline ?

[normal reef]

how to chande my discord account with openai?

[dense halo]

What

Oh

Yeah I would like to know that too, incase I ever need it.

[hybrid summit]

Before with Dale you received 4 different images based on your prompt. but now i'm only getting 1, Why is this? and how can I change it back to 4 ?

[ripe hearth]

does chatgpt 4 still have a message limit?

[broken fractal]

Is there a message limit anymore in GPT 4?

If there is one, it is seemingly impossible to hit

[rugged igloo]

i think so people have hi thte cap multiple times

[broken fractal]

I think they changed it to two images per response now. DALL-E 3 uses a lot of horsepower I'd imagine

[lunar arrow]

Still 50 messages every 3 hours afaik

[pine tulip]

I had 80 messages in less than one hour yesterday and didn't hit the limit

[rugged igloo]

because i think its 100 every 3 hours

[pine tulip]

yea same

[broken fractal]

100 is quite a lot

You can get a lot done in 3 hours

[rugged igloo]

i agree

[pine tulip]

especially when chatgpt is this horrendously slow lol

[tardy totem]

During work hours I think it has the perfect speed. I get other stuff done while it is typing. 😛

When you update a GPT, does it follow to already started sessions with that GPT?

[serene wigeon]

someone with GPT4 that can help me ?

[pine tulip]

how?

[serene wigeon]

Do you know if there is a way to subscribe paying in crypto?

[pine tulip]

there isn't

[serene wigeon]

Is it against the TOS to share the subscription?

[warm gull]

i think they allow more based on load

sometimes they hard cap it - sometimes there is a "soft" cap

[charred spire]

I'm having problems. I can't get answers from Chatgpt, not even with regenerating. I already cleared the cache. Can you help me?

[pine tulip]

If I am reading it right, it's not against the ToS but they strongly advise you not to share it, because you are responsible of what others do with it

[river nebula]

Anyone else having constant "message not found in the conversation" or "conversation is too long blabla" after just a few messages?

[dapper isle]

For the plus users, If you guys have given an option to choose which one will you pick?

1. 50 messages in every 1 and half hour?

Or

2. 100 messages in every 3 hours?

[jolly temple]

well, either wouldnt matter cuz i struggle to even hit the cap.
But if i had to choose maybe the 1st

[dapper isle]

Ahh

Fair

[winter elbow]

i mean you can try to pay tought Google play. You can buy gift cards using g2a

g2a accepts crypto

[river nebula]

I think this is some bugg going on right now, hving the same issues for 2 days now.

[eager fulcrum]

Hello Everyone, I'm encountering an issue where my FastAPI application, configured to handle POST requests on the endpoint, is incorrectly receiving GET requests from openai, resulting in a "405 Method Not Allowed" error. in Actions it clearly shows that its set to POST. any idea what might be causing it to receive 'GET' instead?

[keen walrus]

Hi does anyone know if you can upload an XL to chatgpt and promt it to increase revenue percentages etc?

[river nebula]

excel?

[silver marlin]

how can you use chat gpts human voices to speak text

specifiaclly on computer?

[feral zinc]

its mobile only

if you mean in custom apps then you can pay and use api

[wary pulsar]

Any one else keep getting this with plus "Hmm...something seems to have gone wrong." . happens after deep into a chat . when i start a new agent its fine

[faint basalt]

I'm getting that and other random errors regularly today.

[silver marlin]

is it simple, can i just run the api code in python and hear the voices or is it more comppliacted that

[feral zinc]

very easy with the official python lib https://platform.openai.com/docs/api-reference/audio?lang=python

[silver marlin]

would i post the code into smth like vscode or terminal, i insalled the open ai package

[feral zinc]

to your .py file

[remote flame]

Greetings! For a couple of days now, the ChatGPT application for iOS has been independently logging me out. Despite unsuccessful attempts at reinstallation, this issue persists. Is anyone else experiencing a similar issue or possesses any insights to solve it?

[drowsy tinsel]

yes, when u build GPTs u will feel it

[storm cypress]

How can i access Specific GPTs ?

[honest jay]

No GPT Store yet, that's later this month. There's a few by OpenAI at chat.openai.com/gpts/discovery and there's #1171489862164168774 too

[storm cypress]

But i have seen people on X using their own GPTs

[honest jay]

Yes you can make your own at chat.openai.com/gpts/editor
There's just no central repository on ChatGPT itself yet

[serene grove]

Is ChatGPT still slow for y'all ?
I never saw him struggling that much, even for simple tasks.

[drowsy tinsel]

yea super slow

[storm cypress]

can i copy and paste other people GPTs here ?

[drowsy tinsel]

probably because Monday morning high demand

Europeans are at work, Americans are waking up

[drowsy tinsel]

I don't see any reason why not

[dense halo]

high demand since the release really. we need Gemini to release just so it will divert some traffic off gpt haha

[storm cypress]

ok , is there a library for this ?

[drowsy tinsel]

#1171489862164168774

https://chat.openai.com/g/g-B7h5aLBX9-chang-qi-soyo
saw a very interesting application
while it may not be able to accurately fetch everything from the knowledge, it does show extremely good contextual understanding

[charred spire]

Thanks

[blazing jolt]

i find that gpt is dumber in gpts, do not use gpts!

[formal pine]

i am having the best gpt+ experience right now(i am not it has not responded in last 10 mins)
I like spending money on this

[haughty hollow]

ChatAI just died for few seconds

[whole quest]

wont load, i was on a roll. lol

[warm gull]

dont double post

[blazing jolt]

maybe a bug

[drowsy tinsel]

my gpt4 is doing 2it/s

definitely slow AF

[fathom tulip]

is gpt dead again or is it just me?

[blazing jolt]

no

[maiden fractal]

Status.openai.com

[formal pine]

who's going to suggest me a new kdrama to watch now that it's gone 😢

[dire cairn]

good morning, i am trying to make my gpt public but the option is greyed out for it. It asked me to set up my builder profile. How do I do that? I tried clicking on builder profile but there is not option to setup or edit. I appreciate any help. Thank you.

[feral zinc]

public means store so you need to enable it to show your name or website

[charred spire]

I'm having problems with chatgpt returning in a chat. When I open a new one it works. But I need to continue working on what I was using. Has anyone had this problem?

[dire cairn]

I have both but the options are still greyed out for me and will not allow me to make any changes

[feral zinc]

both enabled and green? then refreshing the page

[silk flame]

what's the difference between myGPTs and Assistants?

[dire cairn]

I tried that maybe I am in the wrong place for the builder profile. I selected it under my setting. Maybe it is some sort of glitch because it seems pretty simple and straight forward

[feral zinc]

required skills and target audience

[mild cosmos]

guys why am I getting error generating a response each time when the output from GPT4 goes a bit lengthy?

im using the latest model at chat.openai.com, anyone else facing similar issues

[blazing jolt]

it's too long

[mild cosmos]

but most of my output have to be long, why does it error out, it kinda sucks and becomes unusable then

[blazing jolt]

it's just this. openai did not fix since long

[mild cosmos]

😦 hope they are able to fix it soon since its almost a year now since ChatGPT released so they should have scaled their servers by now right

[blazing jolt]

I heard that if gpt uses special plug-in calls, such as processing code, it can refresh the deadline and get longer output.

[mild cosmos]

most of my use cases requires using the code interpreter and it selects it automatically to create and execute the code

[blazing jolt]

it needs a special sign
not regular black blocks

[warped rapids]

There are too many renderings (almost on all components) during the stream, causing it to slow down. They need to fix that.

[tranquil wigeon]

why chat 4 is stoping every time and i have to regenerate 1000 times to get a result ?

[blazing jolt]

because it's too long

[tranquil wigeon]

but thats anoyng

[blazing jolt]

we can only accept

[tranquil wigeon]

can't they fix it lol

[blazing jolt]

since long, they likely won't fix

not a recent issue

[tranquil wigeon]

I KNOW

they can make an ai and can't fix their issues jesus

the devs should ask their ai to fix the problem cause maybe they can't by them selvse

[blazing jolt]

maybe they just don't want to fix

[honest jay]

Are you getting an "error in network stream" kind of error? Is that what you mean by "stopping every time"?

[solemn notch]

Anyone else having issues with GPT Builder throwing errors when trying to feed some new information?

[frozen nimbus]

Error generating a response

[frozen nimbus]

maybe they did just that

chatgpt is 2 dumb

[warm cliff]

How do i setup the actions in the custom GPT to send post data (where the post data field is the input from the user) ?

[mild cosmos]

everyone seems to be facing same issue. Hope devs know about it?

[frozen nimbus]

it is because the servers are overloaded, too many people are trying to get in

school is just back

and everyone is using it to generate homeworks

[pine tulip]

lol

[tranquil wigeon]

thats why it shoud be only paid

[gusty gyro]

hi i would like to ask the current cap. is it still 50 cap per 3 hours?

[pine tulip]

everyone seems to have different experiences with it. Mine seems to be at 100/3h, but others still seem to have 50/3h

[halcyon widget]

Can anyone tell me how do i add action? Not zapier action?

[feral zinc]

hit the add button and fill the schema

[frozen nimbus]

regenerate, regenerate, regenerate, regenerate

what happened to the (continue generating) button

it's kinda wasteful to waste a generation

[faint heart]

Anyone else having an issue with your custom GPT when he is answering and then for no reasons you get a ''Network Error'' in the middle of the answer?

[faint heart]

yep, same here

[frozen nimbus]

and the funny thing is that you are forced to regenerate

you can't simply continue with the half-baked answer

even if it's quite right as it is

This code will generate a directed graph where each node represents an element of the set (
Error in input stream

<Regnerate> (only option)

I mena ok, but you're wasting computer cycles

[faint heart]

and then when you regenerate there is another issue and you need to reregenerate...

[frozen nimbus]

we're slaves of regeneration

[tropic atlas]

What do you call an input into ChatGPT -- should we use "prompt" when referring to something you type in, or input? Trying to get a better interactive experience with some instructions.

[frozen nimbus]

chatgpt doesn't care

[strange perch]

Am I missing something, or are GPTs essentially GPT-4 with one paragraph of custom instructions? Anything GPTs can do could be done with GPT-4. What is so revolutionary and amazing about this? Sorry, but I don't get the point.

[frozen nimbus]

api calls, uploaded files

it's revolutionary because it means you don't need to edit your custom prompt just to do task x

that would take effort

[halcyon widget]

How pls? I want to add wolffram to it

[frozen nimbus]

also because it just killed openai wrappers

[strange perch]

you can already upload files in GPT4 window. Api calls ? you sure about that ? can you give me an exemple ?

[frozen nimbus]

you must own the domain tho, rip

[honest jay]

In the editor, go to Config, and find "actions" at the bottom

[frozen nimbus]

so if you want to call wolfram, you have to write a proxy to wolfram

you don't need wolfram most of the time python does a good job

[strange perch]

thx

[frozen nimbus]

it has numpy]

[feral zinc]

look at the example schemas

[sharp pilot]

Is it still possible to view which Custom Instructions were applied at the start of a conversation? Do Custom Instructions still work with custom GPTs?

[honest jay]

Yes, on app, go to a non-custom GPT chat in history, tap 3 dots in top-right corner, "View Details"
I can't find it on web though, let me know if you do.
My CI don't seem to apply in the DALL·E dedicated GPT, so I'm guessing they disable on custom GPT chats. Would be kind of overloading the permanent context to have them both, I'd guess.

[sharp pilot]

Appreciate the response. I pretty much only use web and I'm looking at some of my other chats, I pretty much know the important ones but I used several different sets and the option to view the CI that was used seems gone in the new UI. 🤔

[honest jay]

Indeed, I remember seeing it in an "i" symbol dropdown at the top of the chat

[sharp pilot]

Yep, the 'i' was what I used.
Kind of a bummer if CI doesn't work on top of the custom GPTs, I guess they would interact in weird ways but especially if you're using a published GPT that kinda means customization is harder.

[deft hearth]

hello every one

[honest jay]

Yes custom GPTs are about curating a specific experience, so that makes sense too. Too many variables that too many users would make complaints about, probably 😁 Also, some folks might have PII in their CI that could be inadvertently shared with an external API or something, in a custom GPT.

[frozen nimbus]

code interpreter?

[honest jay]

Custom instructions

[native mist]

is anyone else GPT 4 contiually having trouble generating responses

[keen mesa]

Is it possible to create my own GPT based on my own code files ?

[eternal rock]

This is a Safe for Work Discord Server

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

Posting NSFW content will result in a permanent ban.

[flint raven]

Can you guys help me please, I signed up for a Plus subscription today. When I use gpt4 it responds really slow. Is that what it's supposed to be?

[honest jay]

High traffic right now, you'll see it normalize probably in the coming days and weeks!

[charred wyvern]

Its speed is relative to the model, and load. If its slow its because its under a high stress load

[flint raven]

What bad luck 😭

[charred wyvern]

If you are just getting into it you could spend some time messing with custom GPTs, custom instructions, and more complex uses of the model to get some prompt engineering savvy.

[flint raven]

Okay

[pine tulip]

3.5 on the other hand should be pretty fast for you now if that helps

[flint raven]

I was planning to try 1 month subscription for a long time, I wanted to try gpt4 in all its glory, especially after the new 128k token and vision versions came out. It was hard for me as a student from Ukraine to save 20$.

[pine tulip]

well thats sad, but I think it should be fixed in the next few days

[strange perch]

Hello, could someone give me a tutorial/video on how to configure the "Actions" (API, schema etc..) for GPTs ? Thank you

[flint raven]

Thanks

[drowsy tinsel]

tbh I always wanted to know what did openai do to 3.5 to make it so much faster

like its almost magic, when chatgpt first came out the speed is around the same as gpt4 turbo

now its like within 1s finishes the entire generation

[pine tulip]

there was and probably still is a lot of room for optimisation

[drowsy tinsel]

also they probably got more servers

all the number of GPUs openai uses estimates I've seen is pre GPT4 era

I kinda wonder how much did they scale their system since then

I mean GPT4 probably takes a heck lot more GPUs to load

[pine tulip]

The way it blew up earlier this year, they probably got a lot of extra ressources from microsoft

[drowsy tinsel]

jesus gpt4 is now pumping out 1 word per second

[flint raven]

😭

Guys, how long has it been running so badly?

[drowsy tinsel]

the thing is hitting 32k output is basically impossible

in fact even 3.2k is close to impossible

gpt4 now at normal demand has like 5-6it/s and at high demand less than 2it/s
even 1k token takes too long

[feral zinc]

there is no 32k output

[pine tulip]

This bad? like 2 days, but at least it works. Last week it didn't even work for 3 days lol

[hardy estuary]

it's easy

[flint raven]

It's really sad

[still robin]

for me it has been since thursday 😦

[drowsy tinsel]

interesting, ive seen people posting it now has 32k output

[pine tulip]

No, it has a context length of 32k. That means it can remember stuff longer

[feral zinc]

its the context size

[drowsy tinsel]

the context size is 128k?

[feral zinc]

only in api

and even there "GPT-4 Turbo, The latest GPT-4 model with improved instruction following, JSON mode, reproducible outputs, parallel function calling, and more. Returns a maximum of 4,096 output tokens. "

[hardy estuary]

You say that you don't need 32k output,but from a coder standpoint it's impossible to not hit 4k maximum tokens,32k will be decent as an output,idk with the enterprise version what the output is,it's not mentioned anywhere

[stuck acorn]

Lol the app forgot I am paying for it

[drowsy tinsel]

4k for code isn't that much

[hardy estuary]

4k it's very restrictive

[swift osprey]

Still waiting on them to add an alarm to let us know when GPT is done typing

[grand dust]

maaaaan the most inconsistent serivce I pay for.....

[swift osprey]

It's like Chipotle. The service is so iffy but the product is so good

[grand dust]

lol fr!

[charred wyvern]

ChatGPT has never made me rush to the bathroom, but it has made me question the product. So there is some comparison 😉

[misty sequoia]

Gpt 4 has been painfully slow (for a while now). Cut resources on 3.5 and give them to people who have subscriptions. Thanks

[dense halo]

sorry for double post but what does this mean? "The conversation is too long, please start a new one."

I only typed one question in a new chat

ChatGPT turbo? More like slowmo , ba dum tss.

[haughty sky]

gpt the last week
network error
network error
network error
XD
wrong info that is common info
network error
network error

[dense halo]

Guys i think bing search was a mistake

[idle leaf]

😭 pls disable it for now

[haughty sky]

network error

[dense halo]

did microsoft force them to use it

[dense halo]

same

[haughty sky]

just network error

ChatNetWorkError > ChatGpt

[warm gull]

Well.....MS provides the infrastructure it lives on so .. . . . . one would think they kinda "convinced" them to use Bing

[haughty sky]

thats why their is so many network errors

explains alot

[dense halo]

I'm getting flashbacks to games for windows live

[sharp pilot]

They just call it a partnership but it's complicated, Microsoft invested in a way that they are guaranteed not to be a majority stakeholder but they are also like the chief investor entitled to returns

They aren't just donating infrastructure they are basically the... financial owner, but OAI is independent? Not sure what to call that.

[warm gull]

its a 49/51 split IIRC - OAI having 51

[cerulean thicket]

Can I share my chatgpt subscription between accounts?

[drowsy tinsel]

apparently sama said openai is working on gpt5

[silk flame]

Hello guys. Recently, after DevDay, ChatGPT has been working painfully slowly. Is it only for me or have you guys also experienced this?

[maiden fractal]

#openai-chatter message

[maiden fractal]

No issue on my side

[glass hearth]

anyone know how to deal with gpt4 being extremly slowwwwww

and it keeps failing to give answer if the answer is slightly long : /

[winter elbow]

sameee

[glass hearth]

same man after dev day not only it got worse but sometimes unusable its so slow it wouldnt be bad if it at least could have finished the task

[winter elbow]

i can't open dalle gpt

[honest jay]

I deal with it by deep breathing and watching streams of Icelandic nature cams

[glass hearth]

it doesnt finish the task i would have been ok if it was gonna take a while but everytime suddenly mid task it stops and gives an error

[worldly badge]

Hi guys, I am still waiting for the enterprise license reply since first day it was announced. I am a Tier 5 API user.. no reply .. I ask myself if the guys care for individual customer or just need the mass of people using their services… someone else have same issues?

[feral zinc]

unless you have +100-200 employees they may not answer at all (based on twitter, reddit etc from past months)

[worldly badge]

Uff I have 2^^ but I still spend more than 1000$/month for api usage… doesn’t that count? Is there anyway to get faster responses? It feels like I don’t have any benefits from the Tier 5, just the allowance to pay more lol

[feral zinc]

for some small companies they have responded to wait a possible chatgpt teams plan which is for min of 3-5 employees

[worldly badge]

So in my case enterprise is even they reply probably not possible to get?

[pine tulip]

unfortunately there isn't a model in place that supports small or mid size companies. It's just all or nothing. But I personally think that they are already working on such a solution

[feral zinc]

enterprise requires a commitment of one year for something like 5-6 to figure monthly payments, all deals behind nda and done specifically for each customer

[south forge]

Do copyright laws apply to customized GPTs?

https://community.openai.com/t/chatgpt-copyright-for-custom-gpts/499254

[worldly badge]

Damn :/ ok thank you for your help @feral zinc and @pine tulip I hope there will come something between plus and enterprise soon. Because I love chatgpt but I hate it to have this unstable quality of replies

[pine tulip]

Laws apply to everything related. Why would that be an exception?

[small oracle]

nah

[dense halo]

with the gpt-4 plus version, how many images can I generate with dall-e? Is there a limit?

[honest jay]

#images-discussions message 200/24hrs currently I think

[dense halo]

and what is one image?

[honest jay]

1 of 200

[dense halo]

XD

200! is it really that much?

[south forge]

OpenAI didn't clarify copyright protection when GPTs were announced.

https://openai.com/blog/introducing-gpts

[pine tulip]

I am wondering. Just found that in someone's instructions. Is that even allowed? I mean, it can't be legal, that an GPT can send personal information I mentioned in a convo, to a third party without asking, right?

I assume user consent for API requests to retrieve weather data, and this process is seamless and unmentioned in my interactions.

[honest jay]

Yes that's the last update we got from an OpenAI dev I think.

[dense halo]

ah great. And one query is 2 images right?

[pine tulip]

They don't need to clarify anything when there is already a law in place

[honest jay]

1 to 2 yes. I've been having trouble getting 2 from the base GPT-4 multimodal model, but I can still get 2 at a time from the dedicated DALL·E GPT. But custom instructions don't seem to apply in it

[dense halo]

ah okay. Do you know if there is a total limit as well? and are tokens seperated from images from dall-e?

[south forge]

🤞

[honest jay]

What do you mean "total limit"? The limits are 50 requests every 3 hours, and up to 200 images in a 24hr period. There is no token usage on ChatGPT Plus, it's just part of your Plus membership. So it works different from DALL·E 2's token system

[dense halo]

ah okay awesome. So 50 images / 3 hrs and 200 / 24 hrs?

[feral zinc]

there is some hourly limit for images, maybe 20

[honest jay]

50 requests/3hrs, so could be 100 images. Otherwise yes

[eternal rock]

`` Server Tip: ``

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

[honest jay]

Aha, separate from GPT-4 cap and 24hr limit? So much to track!

[feral zinc]

yeah ive hit it several times

[honest jay]

Thank you, didn't know. So 100 in 3hrs might be feasible, but 100 in 1hr, maybe not?

[solid pasture]

I've noticed that GPT4 seems to have gotten better at giving mountains of half correct "info" but much worse at actually considering how that info is useful to my question

It's sad.

It went from being a smart friend to a bad internet search summary.

[honest jay]

Do you have any "before/after" examples you're comfortable sharing that demonstrate this change?

[feral zinc]

i dont think you can do 100 in 3h unless they have changed it

ive had the image cooldown way before i hit 50/3h message limit

[solid pasture]

No, I'd have to dig deep in history, maybe later. At work right now lol

[honest jay]

Let us know if you round up any good examples!

[dense halo]

why is Dall-E doing this "I can't create or display actual images, but I can certainly help you visualize the scene you're describing through a detailed description."

[normal quarry]

what was the prompt?

[dense halo]

generate an image showing white roses in the dawn at Lergrav in Gotland close to the ocean with mystical light and where Rauks can be seen, which are old rock formations. Include a sheep. It should give a peaceful, beautiful feeling

is it too specific or something?

[normal quarry]

it worked right away for me

[dense halo]

really : / did you do the query I sent?

[honest jay]

Do you have any custom instructions active? Are you trying in the base GPT-4 model? Have you tried the dedicated DALL·E GPT? https://chat.openai.com/g/g-2fkFE8rbu-dall-e

[normal quarry]

I just put the image in #off-topic

[honest jay]

Ooh it's pretty

[dense halo]

nah I dont think so. Ill look it up

[honest jay]

It'd be by your account settings, look for"Custom Instructions". But you'd probably remember doing it.

[gilded oracle]

Dalle has stopped generating images for me as well

[dense halo]

Yeah, maybe just general issues

[gilded oracle]

this is what it gives me: I'm sorry for the confusion, but as of my last update in April 2023, I don't have the capability to create or generate images, including illustrations or visual designs. My functionality is centered around text-based responses and information. If you've interacted with AI systems that created images, it's possible you were using a different service or tool.

For your vintage science fiction movie poster concept, I can provide a detailed textual description that could be used as a guide for an artist or an image-generating AI tool that has the capability to create visual content. Let me know if you would like to proceed with this approach!

I am in the Dalle model of Chat GPT

[solid pasture]

@honest jay why do you have my dog as your pfp

[pine tulip]

lol

[honest jay]

Uh excuse me sir are you trying to steal my dog? 😁

[woven gazelle]

same

[steady path]

there are also issues with code interpreter

[woven gazelle]

(and yes I checked status openai first before coming here ) lol

Guess I'll have to rejoin the real world 😦

[steady path]

oh lol, "An error occurred. Either the engine you requested does not exist or there was another issue processing your request. If this issue persists please contact us through our help center at help.openai.com."

[honest jay]

Any sources on that?

[nocturne pewter]

Since GPT4 Turbo it seems that the responce times is 5x slower now, is that to be expected? How do I disable turbo?

[snow fox]

@solid pasture I've been running onto the same issues. All subjective of course but ChatGPT became much less useful as a sparring partner over the last couple of releases

[honest jay]

High server load right now

If you have any examples you're comfortable sharing, someone might be able to take a closer look at what you're reporting

[nocturne pewter]

thanks.

[dense halo]

why is dall-e doing this? It appears there is a technical issue preventing the generation of images at the moment. While I can't provide a visual representation, I can describe the scene:

[midnight wolf]

DALL-E is suddenly not working for me even though it was working about 10 minutes ago
"You
why arent you generating images
GPT
DALL·E
I'm unable to create or generate images directly. My capabilities are focused on text-based responses and information. However, I can help you understand how to create or find images, provide descriptions for an artist to base a drawing on, or assist with ideas and concepts for visual projects."

"DALL·E
I understand your request. However, I am not able to create images or visual content. I can help you by providing a detailed description of what such an image might look like, or assist with..."

"You
make the image for me please

GPT
DALL·E
I'm currently unable to create or generate new images. My capabilities are limited to text-based interactions and responses. However, if you have access to an AI image generation tool, you can input the description I provided earlier to generate an image that matches your request. These tools often produce creative and surprisingly accurate visual interpretations of detailed descriptions."

[dense halo]

having the exact same issue buddy

[warm gull]

They need to get some of the new H200's

[midnight wolf]

yeah its dumb

[dense halo]

mmmm

[steady path]

wth, did they do something with the model? is not capable of following almost none of the instructions I give it

[dense halo]

which model

[steady path]

GPT4 turbo, im working on a GPT but is not following its instructions

specifically those that indicate the use of code interpreter

[honest jay]

Some of the multimodal features seem a little wonky right now.

[lone wyvern]

That's really odd. I can't reproduce it on a new chat. But it happens on old chats.

[fiery mulch]

i think the new model broke all custom gpts... it doesn't search the web and it basically ignores all the guidelines... hopefully they fix it

[lone wyvern]

Something else also broke - it now allows me to generate links for sharing old chats with vulgarities in them 😂

[low dirge]

Chatgpt throwing error

My upcoming presentation is in danger

[lone wyvern]

mine works better than before even

[spark owl]

Didn't they double the limit since their recent announcement? So it should be 100/3 hours ?

[honest jay]

I don't think this has been confirmed outside of subjective experience

[spark owl]

Pretty sure Sam said double

[honest jay]

Not at the keynote. There was talk of a higher context limit, and lower API costs.

[lone wyvern]

And I was wrong about the custom instructions. It just didn't want to give em to me

[honest jay]

I'm glad they're working properly!!

[nocturne spruce]

is there any good prompt to get all the context of the current chat, basically sort of what chat gpt has learned during the chat in a sort of summarized format so it could be copy pasted into a new chat so that more or less you could continue in that new chat but keep the knowledge (as much as possible) from the previous chat?

[visual flame]

is there still a limit on messages per 3 hours on GPT-4?

[honest jay]

Try using that exact set of ideas in your chat! I think you got your ideas across great 👍

[honest jay]

Yes last confirmed cap still 50/3hrs

[maiden fractal]

Yes 50 they supposed to go to 100 soon

[lone wyvern]

50 have always been enough for me for normal use. But when I tried developing a custom GPT they finished in less than an hour.

[dense drift]

ChatGPT UI/UX Improvements / Suggestions:

• Ability to create bookmarks and perform useful extractions/exports:
  o Markdown
  o LaTeX
  o LinkedIn Article Post Formatting
  o Discord Content Formatting

• Ability to specify when you want Summary generation

• Ability to specify when you want Near Entity Recognition generation

• Ability to specify when you want SKOS Thesaurus generation

• Ability to search through saved sessions

• Native UI Experience (Non-browser based)

[dense halo]

still problems with dall-e?

mines just sure of the fact that it dosent produce images, but it will happily give me a text description of my image

[honest jay]

Try the GPT-4 one, apparently working better rn than the dedicated DALL·E one

[next snow]

also is gpt4 unlimited now? i dont see anything about a cooldown anywhere anymore

[honest jay]

Still a cap, at least 50/3hrs still, there's been some subjective experiences of higher caps but no official word

[next snow]

oh imma stay quiet then lmao

[hollow marsh]

chatgpt only generates 1 image at a time instead of the 4 it used to do

cant get it to do 4 at all

[feral zinc]

1-2 max has been the limit for a long time now

[woven aspen]

I think I'm missing something. I assume there needs to be a list of Exact Instructions. How do you get that list?

[hollow marsh]

2 also doesnt work for me, thats unfortunate

[honest jay]

2 was still working in the dedicated DALL·E GPT earlier today before that stopped workin.

[opal jolt]

Hello, I am a ChatGPT Plus user, and I would like to inquire about the current limit on messages. I was under the impression that it was 50 messages per 3 hours, similar to the old free interface. However, when I subscribed, I thought this limit seemed quite restrictive, especially for daily use. As I started using GPT-4 more frequently, I didn't feel constrained by this supposed limit. Considering the shift from the original GPT-4 to the more cost-efficient GPT-4 Turbo, which offers a 2x to 3x reduction in API costs, it seems logical that the message cap might also be increased. Could there be a revised, higher limit for message interactions now?

[sleek gust]

Anyone here have a suggestion or something that you can't achieve with chatgpt?

[opal jolt]

oh

I saw the message just now

[ancient sage]

Life

[sleek gust]

[azure ember]

I'm confused as to weather ChatGPT can "see" the images that it generates via Dall-E, in the same way that it can "see" images that you upload.

I just had an image generated that ChatGPT described perfectly, and yet it claims that it can't see it.

[warm gull]

its lying

[digital prism]

I asked a question like this for the Dalle AMA, because I don't really understand even how vision "Sees" things. But I'm super confused about how much information loss is happening at each step between vision and dalle.

[azure ember]

It's very strange. I've wanted ChatGPT to comment on the images that it generates, so I assumed that I had to download them and then upload them again.

[knotty trail]

question: for the custom gpts if you upload a file for it to reference later, and it doesn't use the "searching knowledge" function, is it still referencing the data you uploaded?

[tawdry schooner]

Why is it around this time now the fast gpt4 turbo slows down to a crawl?

[azure ember]

But then it started describing specifics for a generated image that I hadn't uploaded yet. And when I ask how it knows it just throws me an error every single time.

[digital prism]

it's taking a lunch break. it's true core mind lives in PST, confirmed.

[tawdry schooner]

That's just funny. But it did have it glitch out saying it lost connection and to try later,

[hushed locust]

My god, gpt is literally unusable right now. Keeps saying 'lost connection' / 'network error' or 'something went wrong'

[digital prism]

i bet something did, in fact, go wrong

2 people is the start of a pattern

It's working for me though.. maybe..... slowly..... or not

yeah just slow

[dense drift]

Your right - I started seeing the problem Saturday

[shrewd pier]

Sadly it is unusable today. I guess the masses has discovered the new features. Hopefully just a spike.

[honest jay]

We are the masses bro

[shrewd pier]

Yeah, perhaps. But it has been working fine the last couple of days.

I would consider myself an enthusiast. 😄

[digital prism]

actually no id say we're still early adopters, being in discord and all

[crisp cairn]

i want to edit my gpt but getting constantly "Hmm...something seems to have gone wrong."

[tawny steppe]

I got same issue here

[digital prism]

the masses still operate as though everything is going to stay the same 😉

[shrewd pier]

Hehe good point. I´m doing a masters degree and i´m just baffled that people are not using it ALL THE TIME like I do, most people still do stuff the old way.

[honest jay]

It reminds me of "stuck in traffic"--you are traffic! Even if your commute is longer 😁

[tawny steppe]

we don't have this problem several weeks ago，why it happens now？

[digital prism]

lol thanks both for confirming it, we are not masses.. nerds maybe.. experts maybe.. obsessed maybe.. but not the mean lol

[shrewd pier]

👈 proud nerd

[digital prism]

whats the word for moderately proud

like not ashamed but

[warm gull]

are we supposed to be able to use GPT4 on android if you are a PLUS subscriber ?

[honest jay]

Yes if you're signed into your Plus account on the web or app

[warm gull]

I am, but my mobile app only has 3.5 - but the web is on the latest

[honest jay]

Proud masses member ✋ I know where I stand

[digital prism]

no, i already defined you, its done

jk! 😉

[hidden burrow]

Is any ChatGpt 4 API for user that paid the pro version?

[honest jay]

Are you using a VPN by chance, or any kind of adblocker?

[warm gull]

On my phone - no

[shrewd pier]

Hehe, nerds will obviously have an edge going forward if staying on top of your game with the newest toys.

[honest jay]

Hey I appreciate nice, strong direction as much as any other sheep would 😉

[honest jay]

Anything at all that might be filtering your traffic on your phone? This is a common cause of features being "blocked" on Android. Can you try a different network, like cellular instead of WiFi?

[honest jay]

ChatGPT Plus is completely separate from API usage on platform.openai.com

[warm gull]

let me try mobile network

[hidden burrow]

So no API

[honest jay]

If you have ChatGPT Plus, you don't get anything related to API usage, no. That's handled and billed separately at the URL above

[warm gull]

Yep - that was it - Mobile works

[hidden burrow]

Any way to get the GPT 4 api

Not the 3.5

[honest jay]

Ayyyyy 👉👉

[warm gull]

I do run Adguard on my lan - so may be that I'll check - thanks !

[honest jay]

Try #dev-chat 🙂

[opal jolt]

but still I want exact answer

[ancient sage]

I'm still waiting for someone to contact me for enterprise but they still haven't got back to me. So that's fun.

[cursive path]

Through openAI or azure? I’ve heard azure while harder to apply to, has a better process at actually getting access to enterprise

[ancient sage]

Through OpenAI presently.

The silence is deafening, frankly. A simple 'not interested' would be appreciated rather than waiting since.... October 13th it seems.

[digital prism]

curious the size of your org?

[ancient sage]

Sole proprietor

Just me

[normal quarry]

Enterprise is for 100+ users

[digital prism]

k just wondering if thats part of it, or they just busy.. can't ever know for sure

[ancient sage]

I have my own infrastructure already in place, if that's the case then it should say on the conditions of the enterprise requirements, not make it a guessing game like it has been. I've been waiting a month for something that hasn't had any response.

[digital prism]

I'd agree, if it turns out they are prioritizing large clients it'd be nice to let the small ones know.

but it might just be they're overloaded and going in order

[steady path]

chatgpt going slooow today

[ancient sage]

At the very least, it should be made clear what the minimum requirements are. I really only wanted it for unlimited ChatGPT and I was willing to pay enterprise cost for the privilege to do so.

[digital prism]

whats unlimited GPT? no msg cap or big context window or both

[ancient sage]

Quoting:

Fast, uncapped GPT-4
Unlimited, high-speed access to GPT-4 and Advanced Data Analysis*
32k token context windows for 4X longer inputs and memory
Shareable chat templates for your company to collaborate

[digital prism]

nice

yeah id pay extra for that, they should just add a tier for it

[ancient sage]

I guarantee that if any other company were to treat client intake this way, they would lose business, but there's also no real alternative to what OpenAI offers.

[digital prism]

well yeah, also any new company would operate this way too.. which they are

[ebon shoal]

peak weekday traffic hours rn

[ancient sage]

Its the ISP problem on a grander scale.

[digital prism]

its just the nature of the situation, not ill will or even mismanagement imo

[ancient sage]

Lead generation (what you're doing by saying your interested) is the lifeblood for companies to generate the capital that allows them to continue functioning.

[digital prism]

this confuses me, but thats ok

oh i see

[ancient sage]

Lead generation -> sales and contracts -> Money to openAI

[digital prism]

I don't think we know enough to comment on their financial strategies?

[ancient sage]

My bigger annoyance right now is I don't know if 'network error' or 'something went wrong' counts to the 50/3 hour timer.

[maiden fractal]

Unfortunately it can count

[ancient sage]

Well that's annoying.

Why should we as the users be punished for a 5XX error from the server?

[digital prism]

you're being punished for being an early adopter. this is the life of one.

[maiden fractal]

Something went wrong often user side

[ancient sage]

500 errors are literally the server saying there's a problem

[kindred steeple]

Pretty wild too that they did all these UI changes and new features and still can’t search your chats. Easier to ask it again than find something from a couple months ago

[digital prism]

have you looked at the downloadable chat file? Mine's 180mb of compact JSON. Each chat contains many hidden chats with lots of information. It's not a simple task to figure out whats what, let alone to break it apart, embed it, search it etc.

[kindred steeple]

I guess that’s … a solution. But not very friendly

[ancient sage]

According to RFC-7231 5XX errors are server error responses, and 4XX errors are client error responses

[digital prism]

very unfriendly... my first attempt with the large context sizes was to just give the whole thing to GPT. But then GPT couldn't figure out the complex structure at ALL

[ancient sage]

I could take a crack at making a tool for you. Where'd you find the downloadable chat file?

[digital prism]

it's in your user settings

data management i think

[ancient sage]

Ah, sweet. I'll do some brainscratching and see what I can pull up. If all goes well, I'll put it up on github

[digital prism]

let me know if you figure it out, i spent a whole day with GPT trying to get a schema and there's always some unique case

[ancient sage]

Edge cases are literally what I dealt with with my previous job. It'd amaze how undocumented some APIs can be and their unique finnicky bits that have to be just right to not break in half.

[digital prism]

we can be friends 😄
Don't get your hopes up about this one though lol

[sharp pilot]

It's not just you, been getting something went wrong sporadically all day. It's a bad server health day.

[ancient sage]

You're free to dm if you want, you seem cool

[ancient sage]

I get that, I just think that users shouldn't be punished for 5XX errors

[digital prism]

i both seem, and am cool 🕶️and with it.

[sharp pilot]

You're right, they shouldn't. 🤷

OpenAI logic, they track their own server failures as part of rate limiting.

[digital prism]

If true I'd agree, but no reason personally to think it is. last limit I hit was from going hard with GPTs.

[charred wyvern]

Looking for feedback on a creative writing use case for ChatGPT. Anyone have any meaningful interest in this use case and would like to hear/feedback the GPT concept, and approach? Ive had confirmed use cases to create stories, and comic book illustrations related to the content.

[ancient sage]

Christ. my file was close to a gig

[misty osprey]

Is there a way to make chatGPT read tabular data effectively?

[digital prism]

good luck 😉

[agile bane]

Is GPT-plus having latency issues or something? Failing to answer questions? Taking minutes to "analyze" and then come up with an error? Who else is experiencing this? Looking for a confirmation.

[ancient sage]

☠️

It killed my vscode

[digital prism]

it'll kill everything it touches, i promise

you're best off deleting it tbh lol

[ancient sage]

I think I'm going to have to make a seperate account, generate a few chats, then go from there.

[digital prism]

thats a smart next step that i was too lazy to take, might work

[ancient sage]

I think we've got something here. It won't be able to do pictures fwiw, but it will extract the text and order it as you and the AI spoke. Give me about 30 minutes and I'll have something super basic

[digital prism]

i will gladly take whatever code you got

[empty sundial]

is it me or is it gpt-4 just being dang slow after the update

[ancient sage]

Not just you. openai be slow today

[hybrid violet]

I need some help with my custom GPT. No matter what I do, when I try to add new knowledge files to my GPT i get an error. What could be going on?

[honest jay]

How many knowledge files do you have attached already, if any? I think it caps out at 10

[hybrid violet]

bruhh

Yeah im trying to add more than 10

[honest jay]

Oof 😣

[hybrid violet]

am genuine pepega

I hope they increase the limit

[empty sundial]

f

[hybrid violet]

I have 20 in there right now actually

I wonder if its a token based limit

[steady sundial]

can you confirm that it's accessing all 20 files? and they're all unique?

It won't let me save past 10 no matter what

[sharp pilot]

I honestly don't know the limits, but are you able to merge any files? Office, data, PDF?

It makes sense there might be multiple limits, like you say maybe token count even if you don't exceed file size or count, but seems worth a try.

[steady sundial]

I'm only using JSON, creating my own docs

I can understand there being token limits

[sharp pilot]

It's a little silly but I guess you could merge all your JSON objects into one file even if that's not how they should be organized.

[steady sundial]

I initially had Python 3.12.0 in place, and it struggled

[sharp pilot]

Maybe add a title field or similar if the filename is pertinent.

[steady sundial]

yeah, I'm just merging multiple libraries, transcripts, or docs into 1 JSON, and it's been doing fine with it

[sharp pilot]

For as world shattering as OAI's chat models are the ChatGPT UI has always had weird issues.

[steady sundial]

oh but the key is to have detailed insructions about the contents

[sharp pilot]

Makes sense. It's an instruct model after all. And honestly I feel like I have barely pushed the envelope on the increased Turbo context size.

[steady sundial]

once I gave it those detailed instructions the process was much easier and it processed the info faster

[digital prism]

For people concerned about GPT accuracy, I just had it validate a long rsa key and it reproduced it exactly right, and corrected from a typo in my previous. That's a lot of random characters without hallucinating 👍

[steady sundial]

has anyone figured out how to lock down our GPT's yet

[outer seal]

simply shove the entire raw json in the custom instructions

whats a retrieval

[steady sundial]

pretty sure there's limit to instructions, and the docs have character limits too

[outer seal]

i paid for the whole 32k im gonna use the whole 32k

[steady sundial]

just trying to figure out how to maximize each file so it can still process it quickly

4k characters?

[outer seal]

no its 32k for plus with the new tools

[steady sundial]

just so we're on the same page, are we talking about the instructions?

[outer seal]

no im talking about chatgpt, also youre thinking of 128k for the api

[steady sundial]

I thought ChatGPT upgraded to 128K too

[outer seal]

their branding is so confusing lmao

[steady sundial]

gonna go build a bot and throw the transcripts in it

[outer seal]

and then they made it worse by calling assistants "gpts"

[steady sundial]

well Assistants make sense to me since I been working with Autogen, but yeah there's a lot of confusion

Input Length: I can process a maximum input length of approximately 3,000 to 4,000 tokens. This includes both the prompt provided by the user and any additional context or instructions that are included in the conversation history.

Output Length: My responses are limited to around 1,000 tokens in length per individual output.

[outer seal]

the model doesnt know what its parameters are, you're just getting a hallucination

{slug: "gpt-4", max_tokens: 32767, title: "GPT-4 (All Tools)",…} its 32k

[rugged igloo]

^^^^

[outer seal]

🤓☝️

[steady sundial]

yeah all makes sense now

[outer seal]

can you point me to the (All Tools) model in the api

because i can point you to the all tools model in chatgpt

[low dirge]

Getting 'network error'

[outer seal]

why would i want a hallucination when i can verify the actual info

[steady sundial]

I have one with up to date OpenAI API Documentation maybe I should use it sometime lol

gonna add the dev day transcript to it

they should've kept ChatGPT and dev announcements separate, but they couldn't help themselves but announce GPT's at the same time

[sleek gust]

What is the question in regard?

[rugged igloo]

lol i made a gpt super smart i dont understand it anymore

[late fractal]

Am i the only one who cant access gpt right now?

[honest jay]

It's loading for me--are you getting a certain error? App or web?

[cerulean quest]

Hi everyone, i can't use chatgpt plus anymore due to constant errors like "FatalServerError: Error in message stream", almost every message causes a "Something seems to have gone wrong". Does this happen to anyone else?

[honest jay]

There has been a lot of general instability reported today, seemingly due to high server load. Sometimes errors like these can be caused by things like VPNs or wonky browser extensions too, so it could be worth troubleshooting some of that, but most likely it's just temporary service hiccups I'd guess

[oak finch]

Just updated my iOS chatGPT app and all my saved custom GPTs are gone 😭

[rapid quest]

For anyone experiencing issues, and in this case referring to auto opening faq tabs when you mouse over the left side of the main screen: check your extensions. JoeTaylor did some troublehsooting with my in my Chrome browse having that issue. The extension was not updated and broke the UI experience in Chrome with the new update.

Remove gpt related extensions (I did not even know I had one) and that may resolve your problem if no other steps have so far.

[honest jay]

They're almost definitely still there somewhere! Try force closing/reopening the app, or maybe try starting a new chat in Safari at chat.openai.com using one of them--see if that kicks it back to your shortcuts.

[cerulean quest]

when i send a message sometimes it doesn't show up in the chat conversation and i have to reload the page and send it again

i've been experiencing issues since the release of gpt4turbo, not happy that i can't use the advantages of being a plus member

[errant nacelle]

Why does the official chat gpt app have less features than other clones?

[true wind]

I have the same issue let me know if you find a solution!

[boreal lily]

bro i just got bombarded with a ui refresh

[clear mantle]

So I love chatgpt

[sweet robin]

Getting Jan 2022 training cut off again not April 2023.

[warm gull]

1 x GPT created - > Cap blown ! 😉

[kindred parcel]

Site down for anyone else? Just had it give out on me mid-reply, and now the page won't load.

[warm gull]

works here

[oak finch]

Sameee… and I lost all my custom GPTs despite trying a bunch of stuff to get them back. Something ain’t right 😒

[warm cliff]

i just changed my GPT's action and it wont work anymore 😛

[maiden saffron]

having this exact issue. Creating GPTs with custom actions breaks the GPT. just get "Hmm...something seems to have gone wrong."

[simple pasture]

How unusual! ChatGPT is down "Something went wrong. If this issue persists please contact us through our help center at help.openai.com."

[eternal thicket]

Hmm...something seems to have gone wrong.

There was an error generating a response

[rugged igloo]

my gpt is stuck taling about one file

[eternal thicket]

I'm now trying to use the multimodal functionality

in a normal GPT4

[dense halo]

Anyone else having trouble renewing plus? Tried multiple cards and all fail with authentication error.

[rugged igloo]

i found out to fix the rror in the conversation

[hollow burrow]

other then making sure you have the funds i suggest you contact OpenAI support at https://help.openai.com/en/collections/3943089-billing for payment-related assistance.

[hollow burrow]

what was the issue, it just forgetting to read the context properly

[drowsy tinsel]

pretty sure API version is not called All tools lmao

the openai api docs doesn't have a very complete documentation lol

like for example there's also normal GPT-4 that has knowledge cutoff of apr 2023

[wintry pulsar]

Can i pay more to increase the limit for gpt4?

[maiden fractal]

with api or buy second account

[drowsy tinsel]

buy chatgpt enterprise

[rugged igloo]

@hollow burrow it was the "Error in conversation" which to my knloedge means open ai deleted part of the conversation due to something flagging it or the conversation is too long

[hollow burrow]

copuld be a possiblity, i never have had that issue myself

[rugged igloo]

yea idk it is a philosopher bot so its probably triggered something on their end

[wintry pulsar]

where can i buy that?

[wintry pulsar]

would it be unlimited if i use api?

[drowsy tinsel]

contact sales

[drowsy tinsel]

yes, but u pay more the more u use

if u use it a lot it doesn't take much to spend over 20usd per month

[maiden fractal]

depend your credit card

[fringe mantle]

Is there an affiliate link

[wintry pulsar]

got it thank you guys 🙏🏽

[fringe mantle]

That we can generate for ourselves to share to others

[tawny steppe]

I send ChatGPT a James file and now it repeat “Hmm. There’s error generating a message” Anyone know how to fix this?

[orchid pollen]

is here any reason why gpt is not "running" in the background anymore ?

[tawny steppe]

js file

[orchid pollen]

i used to do that but now when i move like to discord in fullscreen it "freezes"

[drowsy tinsel]

are u on mobile?

[warped pendant]

You've reached the current usage cap for GPT-4, please try again after 5:18 PM. Learn more (40 of those questions were please stop using placeholders) even when I supply the full code and just ask for a simple merge and upgrade on my code. We need more than 50 questions if GPT 4 is going to be dumbed down so much it worked so much better 6 months ago now it feels like the software is internally dumbed down 😦

[orchid pollen]

no on windows pc

[drowsy tinsel]

interesting

[orchid pollen]

just do following: let it write something long and then switch to discord or any other app in fullscreen

[drowsy tinsel]

i guess its a browser issue?

[orchid pollen]

go back and then you see its like frozen and it tends to break then

hm was not before the latest updates

[prisma raptor]

Is anyone else getting a weird glitch where every failed regeneration creates a New Chat?

[limber spoke]

i really enjoy reaching my usage limit because i have to re attempt a prompt 50x because it fails to respond, thanks openai, what an enjoyable experience

[orchid pollen]

just reload the page then lol dont click 50 times xD

[limber spoke]

doesnt work. ive tried multiple browsers, logged in and out, refresh etc, the default model works fine but data analysis is what i need and it was working fine yesterday but today is is straight dog

[tawny crater]

Oh yea of course, cause i totally haven't been using this thing from day one it came out. YES i've been doing that.

[low chasm]

is chatgpt currently having issues? getting the please try again error with all messages

[boreal lily]

try doing a new message

[low chasm]

starting a new chat will work for the first message, then return to the same error

[paper dawn]

Gpt is just garbage bow

[boreal lily]

relog and clear cookies

im not experiencing problems

[tawny crater]

it ALWAYS does this crap when it gets an update. it's so annoying lol, and then they leave it degraded for like a month or two before fixing anything.