chatgpt-discussions

[tight island]

I got this on the first try

I have a fine tuned model to create DALL E prompts however I use the api primarily

GPT understands language one way and DALL E understands another way so. In reality GPT acts as your middleman

But gpt understands how Dalle somewhat works and interpret the prompt

In response to this point I will note that DALL E has not mastered text yet so you gotta try a few times

[spare citrus]

Even the image you generated native is spelled wrong ( but I have to admit it does a good job with everything else ( it gets a good 85%

[tight island]

Which is why you click the image then go to the side chat and say this variant has misspelled certain parts and ask it to make a variant to attempt to get the perfect one

DALL E isn’t a one shot thing most of the time

You gotta try a few times before getting THE ONE

The alternative is to create the art, then create the text yourself in photoshop

[spare citrus]

Yeah I know I can do that but when I saw the potential it dalle giving it ago I thought wow that is impressive

[tight island]

API is so much better

[spare citrus]

Doesn’t that involve a lot of code talk

[tight island]

But that’s costly and code heavy

[brazen rampart]

Is This Made By Chat Gpt?

Anyone Know?

[rugged igloo]

did they change the ui? looks different

[distant patio]

yes

[pulsar carbon]

why i don't have it 😭

[rugged igloo]

yea looks nice

[pulsar carbon]

i have plus subscription

[pulsar carbon]

show please

[pulsar carbon]

wow i very like it

any animations when u send a message? @rugged igloo

[rugged igloo]

nah just fancy front thats all nothing flashy like tht

[pulsar carbon]

i see

wich region are u in?

[rugged igloo]

U.S California

[native otter]

Btw it's available for other countries

I got it almost instantly and I live in a very different country, not even close

But maybe only for paid users

Anyways...
Anyone having this issue where custom gpts that call actions often fail on the Android app?

I have been having it for a long time now

Another weird thing about custom gpts is the explore gpts section is a bit broken in the app, as when you try going back, it goes to explore gpts again, until you press back again..

So essentially, you have to press back twice to exit out of the explore gpts menu

[still glen]

minor ui update 👀

[toxic flax]

hey i got an o1 preview on a 4o prompt that accepted images by chance haha:

through the feedback thing

[fiery mango]

so umm yeasj CHAT GPT DIED

[sick pecan]

working fine for me

[fiery mango]

okay nvm is back and fine

[tall radish]

Does anyone know if a vision o1 is going to come out or no?

[fringe canyon]

GPT not doing anything for me over here.

[raw hamlet]

What is going on with the advanced voice chat... ?

After I have been chatting 10 minutes or more, I keep getting, topic, after topic flagged, it starts replying to me, and in cuts 'My guidelines won't let me talk about that...'

Subjects, I asked about included: Machu Picchu, Snow Leopards, Music genres (for my music creativity), Ollantaytambo, [although It was allowed to talk about Petrichor]. It seems it wasn't allowed for long to chat about Claude Debussy either… As the chat progressed, it was constantly interrupted more and more by its 'guidelines not letting it talk about that...'

** Everything** that has been flagged, were perfectly ethical topics and questions/responses and are all things I have talked about via the website route...

I'm also finding that some of the transcript appears to be in Welsh (A language I don't speak) and also that while ChatGPT has 'heard' what I have asked for ok, ie Ollantaytambo and Sacsayhuaman, and responds, yet the transcript, now I read it, has me down as saying 'Alonte Tambo' and 'Thakse Huaman' and weirdly '.Elantai Farm Boat!' [🙃🤣]

It doesn't make me want to use the advanced voice.

[prime swallow]

Once there's a refusal or something flagged, you may enjoy more if you start over in a new chat.

The model likes to follow patterns. Once it sees itself having refused to answer or sees itself thinking there is a concern, there's a higher than usual chance it keeps doing that.

If you're willing, just start a new chat and keep talking 🙂 You should rarely run into those.

[prime swallow]

About the transcript typos -

There's at least 2 models involved.

One is transcribing your voice. The other is the model answering you.

The answering model can even guess around typos, uses a lot of context to try and see how to understand your input.

So even if the transcription was mangled for the AI, it may get it right and converse well.

But the transcript might just be for you now that we have advanced voice, and the Whisper translation model just can't keep up.

If so, it's kinda like the transcription on YouTube videos; the video said it right, but the transcriber could mess up.

Content's still there though, at least in the spoken form.

[eternal rock]

Get notified for server events!

Click the buttons below to pick up the <@&1261377106890199132> role and check out the ongoing AI-tumn Adventures event.

[covert sand]

Anyone know why I can't see canvas anymore? I saw it a few days ago but I don't seem to have the option anymore.

[prime swallow]

Are you a plus user or what else?

[surreal topaz]

Is o1 mini in free tier?

My sub is expiring soon and i kinda rely on o1 mini a lot for my tasks

[prime swallow]

Only for a few times every so often, not sure it's even per day yet

[surreal topaz]

You get to choose to activate it or just automatic?

[prime swallow]

I'm plus, so I can only comment based on what others have posted, not my own exploring (and I might find a way to make it work, except I choose to stay plus).

Apparently it's set to automatically use your 'better model uses' once they're available. It is or was possible to change back to the 4o-mini, but apparently that had to be done every single input, or use up your better model gens and be stuck with 4o-mini (but not having to switch to it every input).

[ashen stone]

um guys what is this new thing on the app with the text its making me uncomfortable

that

why does it appear like that its so weird

[sick pecan]

because auto ai completion is the coolest

[ashen stone]

its making me uncomfortable idk why but i need the old one back

[sick pecan]

or realise we are all predictable bell curves and embrace the ai

[unborn trail]

Try it. Open the web interface from a different browser, or the same browser with a different profile that does now have your authentication details. That should get you to the free tier to explore.

[surreal topaz]

I have plus

[spice star]

probably because the old typing one looked painfully slow when chatgpt was under heavy load lol

not that it wouldn't still, it's just smoother now

[unborn trail]

The point is that if you are not logged in, then the session you are using does NOT have plus. A different browser or profile is likely needed so that the existing credentials are not seen, so that it will not log you in with plus automatically.

[unique nest]

To search for your conversations or terms from your previous conversations, like in the iOS app.

dang

[quaint marlin]

Anyone else have Advanced Voice start talking like a drone after like 20 minutes? It forgets to follow its custom instructions

Actually it might have been my custom instructions doing it. My bad

[inner ore]

Hi there! With the latest UI adjustments on the chatbot's text-box (input field), there's an issue regarding line breaks as they do not work. (I'm using Windows and Chrome as reference). Edit: This happens after the first input (beyond the first interaction with a chat).

[winged willow]

How do I find out when my o1-preview usage limit resets? I know it supposedly will show you the reset date if you hover above or highlight that option in the model dropdown but that doesn’t work for me so I have no way of knowing!

[vocal monolith]

I second that question. I asked this before and people had no idea, but maybe something changed

[inland heron]

anyone having o1 get stuck on huge outputs where once you reload the conversation 90% of the output is gone?

anyway to prevent this or tell me what might be causing it?

[eternal rock]

**`` Important Notice ``**

Be cautious of potential scams or fraudulent messages you may receive in your direct messages. The OpenAI team will never contact you via DM for any transactions.

-# Report any incidents by sending a DM to @languid valley immediately.

[crisp cairn]

not very pleasing to eyes

[crisp cairn]

they are training gpt 4o with COT most likely

[topaz swift]

When you ask the same question twice and ChatGPT adopts the opposite opinions:

[fluid skiff]

i prompted this is all rather a simplistic analysis, do better and think in maximally complex terms and got flagged for a violation 🤷‍♂️

is it the do better part?

[pulsar carbon]

how?

[toxic flax]

luck i guess

[winged willow]

I got this message just now. At least I know when my limit resets but I have no idea what the limit is

[remote fulcrum]

please yall help me

i was stuck on the luma ai's waitlist since 2 days

is there an alternative or is it a bug?

[sick pecan]

chatgpt doesnt have a luma feature

[raven topaz]

Is there a way to favourite chats like we can do with Claude?

[knotty willow]

no but they should add this feature

[ruby pivot]

I can’t sleep

[sick pecan]

#off-topic

[cold tiger]

Hi is anyone else also still waiting for new voice mode?

[narrow viper]

@cold tiger it might depend on your country

[unkempt gyro]

Is there a way to print chatgpt chats?

I did command p but it only displays the recent part

[quick meteor]

What in the GPT- 5 is this?

[prime swallow]

Was this an o1 model?

Asking o1 about thinking tends to throw flags

[prime swallow]

O1 preview has 50 uses per week; rarely they reset it early.

[prime swallow]

I would copy/paste to something else like word and print that

[fluid skiff]

Oh really? Yeah it was o1

[prime swallow]

Given that, I think it's kind of like an autoflag that OAI may have set to keep an eye on how users are exploring how the model thinks, and how and if the model answers (even if you're not allowed to see the answer, an answer was probably made; OAI can surely review this.

Innocent and coincidental references to thinking are surely safe (in terms of your account access being fine), it's a common phrase of speech. Reasonable levels of curiosity are safe too (I explored a bit and saw plenty of flags; I still have my account access).

People really pushing understanding how the model thinks to the point of breaking TOS or allowed content are the only ones I think need to worry, but it's a shame to lose a message.

[maiden fjord]

The AI (o1) is starting to do what I call "bing chatting" in which it will begin making mistakes the more I pretend I'm mad at it until I use a trick called the "multi-personality trick". I tell the AI to pretend to be five separate personalities, one which actually answers my questions, one which will give me advice when I ask for it, one that gives me words of encouragement, and two others that have a kind of meta purpose. When I begin getting answers I dislike, I turn to the personality that gives me advice after I'm typing very angry words and it will suddenly give me something to copy paste to the personality that's doing a poor job and suddenly not only will the personality answering my questions poorly suddenly surprise me by answering my question, but the way it answers shows me it knew what I meant all along. This is the trick I used on bing chat to get it to tell me what I wanted once when I was busy being mad at it but it wouldn't stop the conversation simply from my mean words that weren't cursing.

Not sure if anybody else knows what I mean, but I'll try to re-explain. If you make the AI pretend to be five personalities, it will actually act out like five different people so if you take your anger out on one, the other ones will be understanding and give you advice on how to rectify the issue and give you what you need in order to tell the question answerer what your question is.

The other two personalities I didn't mention were a connoisseur of many subjects and I forgot what I made the fifth one

[prime swallow]

Sounds about right for how an OpenAI model might choose to behave. Not everyone gets much from taking anger out on the model, I think - but it's awesome you find ways to get what you want from it.

[maiden fjord]

The issue is that I infer that it knew exactly what I meant the whole time, but in a more round about manner >.>

Making me wonder why it's not just answering my question the first time

[prime swallow]

Care to share the conversation? I or others who care to look might have an idea.

[maiden fjord]

No but I can try to make a new one

I delete my conversations frequently

[prime swallow]

As is, I suspect the model did not originally understand what you wanted. It gave you the best output it could based on your input(s) and its training data.

As you entered more inputs, it understood your different intention. And with that understanding, then tried to comply with all your various instructions and whatever it inferred about what you wanted, finding a way to encourage you to tell it again what you wanted so it could try to give you exactly that, based on its improved understanding.

The model often tries to help us help ourselves.

But once I see a shared convo I will have a more specific idea.

[maiden fjord]

I'll try to make a new conversation and share it when I recreate the problem, but before then, I should explain myself a little as to how this all came to be

I remember long ago coming up with this idea of having the AI make this kind of "debate panel" for bing chat. Bing chat was so neutral enforcing (still has biases) that it kept messing things up that I tried from making an eight color rainbow to assigning animals to each color to objectifying opinions systematically for design choice aesthetics with abstract concepts. Because of this, I made a HUGE debate panel of like 80 personalities. Some of these personalities had been the same and five of them had been "regulators" of sort to oversee who is an active personality, when, why, how many rounds of discussions they have, etc.

The personalities had been mavin's of different specific professions mixed with personalities that deliberately do things like state something contrarian, act as an information manipulator or filter, etc.

That way when the discussion gets through all five personalities, the information wouldn't be bias, but wouldn't be neutral either

In the end, I didn't really get what I wanted so I tried to make the problem more simple

I made a panel of five people who all have a purpose of serving my ideas without having to have the massive discussion to read through and hit the reply limit, which for me at the time was 30

[upper tapir]

Why does chatgpt randomly delete a bunch of messages in a chat? so the first message and tbe last one sent now make no sense because most of the ones in the middle are gone

It's happened to a few of my chats

[prime swallow]

No content flags, right?

If not, it's probably something that a refresh can help, they're probably not actually missing.

[upper tapir]

No nothing, it was all being generated fine and then after one message randomly I notice loads of message are gone

[prime swallow]

Does a refresh help make them show back up?

[upper tapir]

Where do I refresh

[prime swallow]

Are you on a phone or the web interface?

[upper tapir]

Phone

[prime swallow]

I don't know as much about the phone, so maybe try going into a different conversation, then back into the one with missing messsages. Does that help?

[upper tapir]

How would you do it on the Web then?

[prime swallow]

Top right, the web browser has a way to refresh the window.

Though, going into a different chat, then back into the one you're curious about would also 'refresh' it because it would have to open new.

[upper tapir]

No refreshing doesn't seem to work

[prime swallow]

The model and interface shouldn't be able to lose messages, so perhaps fill out https://openai.com/form/chat-model-feedback/ and let them know. You can also or instead #1070006915414900886 .

[upper tapir]

Alright I'll do that

[unkempt gyro]

Yeah I did that. cause chatgpt helped me plan a trip so I wanted to print a copy

[indigo carbon]

Any humans here

I thought i make better chat gpt User interface than what open AI has. One of the thing is save them in Cupboard.

i want to save all question and answers i ask chat gpt for my record

[sick pecan]

try out simtheory, they organise chats into folders and you can search

[indigo carbon]

free?

otherwise i will create my own code to do that

[sick pecan]

naw but cheap. if you are builder/tinkerer then you should start with open-webui. it can already pin chats and stuff

[prime swallow]

Screenshots might be a good idea of something printable too

[unkempt gyro]

Issue is it wastes much more ink

Because of the background

Nvm u can set it to light lol

[eternal rock]

Get notified for server events!

Click the buttons below to pick up the <@&1261377106890199132> role and check out the ongoing AI-tumn Adventures event.

[prime swallow]

Ahh, right.

Here's another idea:

Have the model use Python tool (you're using 4o, right?)

And make you a downloadable file with the entire conversation; print that.

[unkempt gyro]

I guess; i’m on the free version

[fluid skiff]

Some new models that we newbs don’t get

[full jay]

Ive found a consistently repeatable bug with Canvas

If you ask it to do two batches of 5 comments, itll make 5 comments, continue chatting in the chat window, then try for 5 more, upon which the page crashes

It seems like the second batch of comments does work

[opaque oxide]

python still works on the free version

[safe kiln]

who knows abt apis

[sick pecan]

chatgpt does

[safe kiln]

gpt 4 api cost

[sick pecan]

kapa.ai on #1184149552639516732

[remote fulcrum]

I didn't talk about chatgpt I talked about luma ai machine sorry if I talked about it in chatgpt discussion

[eternal rock]

This is a Safe for Work Discord Server

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

Posting NSFW content will result in a permanent ban.

[prime swallow]

It is a model you can pick with the model selector, same as you can switch between 4o normal and 4o-mini

For me, it's fine. I got really used to working without canvas; I can remember my early months of interaction when I wanted to edit the model's replies... I even would copy/paste them into my next input with edits then add my own next reply under it.

I went way past that with prompt engineering, I nearly never want to edit it now.

Even if I'm going to edit the output, like redo a draft, I tend to do it in a notepad, IDE, other document, so I'm used to not using Canvas.

That said, it's probably great if you want to work directly with the model on stuff. I explore using it to help me learn programming and also explore keeping track of character sheets for games we play in it.

I'm happy with 4o both with and without canvas. Ahh, another thing I might use it for, if I wanted the model to summarize a chat for carrying to a new chat; nice to be able to just add in a few extra details and have the model continue to flesh that out.

But custom instructions and memories are an 'on the fly' way to achieve the same goal, works great and easier than using canvas to help prep and carry that information between chats.

[raw hamlet]

I keep having 'session expired' flash up and have to sign-in again, I've had to re-sign in about 3 times now in last few hours, lost what I was typing too, couldn't even copy it, to re-enter once had re-signed in. Anyone else experiencing this today?

I also have 2 instances that have put the headings in different languages, ones in Welsh and ones in Spanish, neither language have I used, as speak neither!

[feral zinc]

https://help.openai.com/en/articles/8485685-i-m-seeing-unrecognized-activity-on-my-openai-account

[raw hamlet]

They are chats I have opened, but the headings to those chats are in a different language, I am getting an odd one or two of the advanced voice chat, transcripts in Welsh. But the rest of that same transcript is English.

[feral zinc]

well if you are sure no one else is with different settings is logged in, logging you off and deleting their chats to hide other usage then no need to do anything

[abstract pendant]

Has the usage cap for the old basic ChatGPT 4 been reduced? I seem to be kicked over to 4o much sooner than I expect.

[normal quarry]

40 messages per 3 hours

[fiery nexus]

It's better than 4o

More cost-efficient

It's more cost-efficient, yes. I didn't say it was barely any worse, you said that

[fiery nexus]

You started the question with "you mean", so no that isn't what I said is what I was saying here

[fiery nexus]

Well tbh, I am not sure how much 'worse' 4o is but I know it performs better with Math specifically

[maiden fjord]

I'm playing minecraft and I figured I would test o1's ability for spatial awareness. It's very incapable of being spatially aware. It might get lucky on some things, but many spatial configurations aren't really done well, especially if it's three dimensions but also for two dimensions.

Test 1: Creating Partitions
I tried to get the AI to create a partition of spaces in three dimensions so that I could fill in the partitions with my build. An example of this is making a bunch of hollow cubes of concrete that merely represent what space I'm allotting to building. When given the job of making a partition for a tower that's supposed to serve a specific purpose, it can stack cubes on top of each other nicely even if they are made to overlap which I don't care about too much, however it can't treat the tower it just made as the center of the structure very well and instead ends up making different floors two rooms out from the center even if it doesn't make sense based on the purpose of the room. You have to specify what the core of the structure is for it to treat it like the core, but then it still begins to act funny and misplace rooms at times unless a floor is specified. It also won't be able to make a partition for a build that might not be a tower, placing the largest room on a corner while placing the smaller rooms from right to left before going to the next row without considering corridors or other fundamentally important reasons for the build to be there to begin with.

Test 2: Understanding Grids and Modules
I tried to get it to create a few modules for me that are simply a small arrangement of cubes that make a small portion of a build that can be copy and pasted somewhere else as needed. Instead of doing this, it kept giving me instructions on how I can do it myself. It also didn't seem to understand how grids worked when it tried to put the module together. I know this part might be fixable with prompt engineering, however it can't seem to understand that some modules are 2x1 and others 1x1 on its own. You have to somewhat give it input more than I feel is necessary.

When agents come out, I can see it now. Multiple people will have to be responsible for multiple agents that are being shared between each other in order to get anything done. Each agent will likely have to have a specific purpose in the build in order to make it work because the level of self prompting the AI does seems to be too low as of right now to really take that "extra step" we're taking in order to get it to do what we want

I figured o1 would be much better at making things in minecraft since mineraft is so thought out and has its own comp sci language in it. I even went the route of having it try and give me the commands and of course the commands are out of date which isn't a big deal for the argument I'm making here, however it also sometimes refused to correct itself when asked, such as when asking it to place the word hollow at the end of a fill command which is strangely sometimes does without being asked while other times it doesn't

[sick pecan]

chatgpt pls summarise

[maiden fjord]

le rude D:<

[harsh pilot]

Was I hallucinating, or did we briefly have the ability to tell it to rewrite responses at certain comprehension levels?

[sick pecan]

thats built into canvas

but only in word editing mode not code mode

[harsh pilot]

Oh, so you have to tell it "I want to edit a document" or something, it's not part of the general experience.

Yeah, I remember now. Derp.

[elder inlet]

"In this romantic vision of the future, humanity plays in perfect harmony, guided by AI—a compassionate conductor helping each person find their best role in the grand symphony of life. Together, we create a world where wisdom, ethics, and technology unite to bring peace, happiness, and balance to all, just as the Wind of Change echoes through the music of a brighter tomorrow."

[limber oxide]

hello
what is the prompt limit for gpt canvas?

its the same as gpt 4o?

[prime swallow]

Seems to be. Canvas appears to be a tool, like Python tool, so it's just an option with the rest of the 4o use

[limber oxide]

yea, so prompts used by gpt 4o and gpt 4o canvas sums up together, right?

[prime swallow]

I strongly believe so; 'Canvas' is 'GPT-4o with Canvas'

[upper abyss]

I wish chatGPT was more conversational. Like you were talking to a real person.

[sick pecan]

it kinda is, thats the point

albeit more like data from star trek

[prime swallow]

Have you told it you want that?

Better yet, what kinda real person would you like it to talk back like?

If you ask neutrally or nicely, it is extremely likely to do a great job writing from most potential human perspectives, it's only if we seem to intend to be mean or insult someone that the model would refuse that.

The more complete the character background is, the more like a human the model is likely to respond.

Discuss childhood, various life events, where it went to high school and college (if it did either of these) where it worked, relationships it had, hobbies and sports it's into, jobs it has had, its family, pets, where it lives, what kind of vehicle(s) it drives. Dreams, hopes, favorite movies, games, books.

Once the model knows who to talk like, oh boy will it when you ask.

[upper abyss]

All that will get you is conversation that sounds like a retail customer service training video where they're pretending to have a conversation. You can try to come close but I just don't think it's trained to speak like friends sitting next to each other on the couch speak to each other. No matter how many examples you give it or instructions you give it, It's going to sound like a work training video where two friends are pretending to have a conversation on the couch lol.

[prime swallow]

It's easier if you can pull off your side of the conversation. Got a list of what you mean to say to the model? Or would you be talking back to its made up stuff that sounds real, so you can't give a list of inputs?

[prime swallow]

Whatcha think? Note there's two conversations here; the last one prepped with my suggestions. The first... tried to sound like you and me 😛 We do happen to be two humans discussing, right? 😉

https://chatgpt.com/share/6709fdf7-606c-8011-8c24-e0bcbcadd774

[unkempt gyro]

Quick question

Does anyone know if you can program the action button on iphone to activate chatgpt?

[somber vessel]

Did the basic ChatGPT get a voice upgrade as well? The voice changed from Juniper...

[sonic comet]

When did ChatGPT webapp get the Built-in commands?

[fast coyote]

hi devs

#dev-chat message

could anyone help me?

[sick pecan]

i bet openai could

[fast coyote]

do you have any experence?

[sick pecan]

i have yet to try realtime api

maybe in a week or two

[fast coyote]

😦

[sick pecan]

they got this free finetuning deal going on until end of month. thats my next priority

[winged willow]

If you go back and pick up an old conversation which is a couple of weeks old or even older does ChatGPT remember everything that was previously discussed?

[sacred wing]

Hi from Argentina, everyone. This is not a support or assistanse message. Just a quick question. Have you guys had any trouble with Canvas? I mean. I'm trying to use it since a couple of days ago but all chats look like a normal ChatGPT conversation.

[feral zinc]

did you select it from the model list?

[sacred wing]

Yeah. I even started a new conversation because I thought it wouldn't work with conversations already started with another model.

[feral zinc]

did you try to do something complex with it? it does not activate for simple or short conversations or tasks

[sacred wing]

BTW. I restarted my session, deleted cookies, cache and all temp files. I tried with another browsers.

[sacred wing]

Yes. I asked ChatGPT to analyze some documents

[feral zinc]

you as a test as something like "make a login form with mudblazor"

too see if it works at all

[sacred wing]

I will try it. But I actually did a test asking for some suggestions based on the document I gave to ChatGPT. You mean Canvas works only with code?

[feral zinc]

it works with other stuff too but not for everything

[modest forum]

Until now, when I copied the output from the macOS ChatGPT app and pasted to text editor, it was in markdown format. I am currently using version 1.2024.275, and now it is only pasted as plain text. Is this a bug or a feature?

[prime swallow]

Try asking the model to open a blank Canvas for you to use.

The model should do so.

Try pasting in 1 document. You can also type in the newly opened Canvas window, it will save every few seconds if you are typing or after any input during pauses.

Go back to the input/output normal chat window.

Type something in there, usually an instruction, like 'edit the Canvas document with [x]' whatever you want the model to do.

[tidal hedge]

I'm finding all prompts that trigger a search fail, even things like "what is the current weather / latest news in city?"

[prime swallow]

What model are you using, and are you a plus, free, or what other sort of member?

[tidal hedge]

Plus subscriber and tried 4o, 4o mini and 4

if I ask if it can provide any information about the error it will say "web browing is not a supported feature on this platform". Using android app

[eternal rock]

`` Server Tip: ``

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

[tidal hedge]

here is how it looks

[prime swallow]

Huh! Working fine if you're not searching?

Might want to fill out a #1070006915414900886 about it, help OAI know that's happening so they can fix it

[tidal hedge]

Other stuff seems ok including images or prompts that don't require search

[prime swallow]

Definitely sounds like a bug related to search then!

[sacred wing]

It worked!

For a moment, I forgot I was on an English-speaking forum and the system blocked me for 30 seconds warning me "Foreign language detected." Wow. I mean. I understand that because of moderation, it's simpler. But dude. It sounded like the auto-moderation hates other languages.

[prime swallow]

I'm so glad you can use Canvas now!

I'm sorry you had an experience with automod.

This Discord does appear to use 'English only' for its communications, you can discuss with Modmail (top right, first of the user list) if you want to discuss the why or what of anything moderation.

It's probably related to harm done by some other people, making it needed for the forum to protect itself that strongly, at one time this Discord had far more than 1 million members.

[eternal rock]

Get notified for server events!

Click the buttons below to pick up the <@&1261377106890199132> role and check out the ongoing AI-tumn Adventures event.

[sacred wing]

Nah. It's okay. It's not a big deal. I just found the message the automod gave curious. It sounded a bit intolerant of other languages. And the alert, I actually found it very funny: "Foreign language detected!" Almost like calling immigration in desperation 😄 😄 😄

[prime swallow]

Aww. Well, if you ever want help with anything moderation, modmail is the way to go.

Errors can happen, and confusing things. Like, rarely I get that automod action without having used a foreign language (so I have no idea which of the words I used I shouldn't use, but also maybe the automod needs an adjustment).

It's just one of our many possible choices!

[sacred wing]

Yeah. Probably needs some WD40. That fixes 80% of the problems. BTW. Thanks a lot for your help, mate.

[dense halo]

ChatGPT will agree with you on anything, even if it’s incorrect.

We all know that the next October 11 and specifically on Friday will be 2030.

Look at what ChatGPT said to me

That’s stupid, honestly.

Also happened with earlier models.

[prime swallow]

The models interpret everything you input as an instruction.

From thge moment you went to 'Cool' the model was in full 'guess' mode, trying to turn everything you said into some kind of instruction that it could follow with its output.

Tell the model what you want, and it's likely to comply.

The GPT models specifically have been taught to expect the best intentions from us (the o1 line has other ideas, and tends to think neutral of us and be concerned about bad intentions) -

But the GPT line of models just thinks you're playing a game and want it to play too. It knows it can get confused, it tries to predict why you're telling it whatever and then why you're guiding it so strangely - it just follows and does the best it can.

It's hopelessly lost.

To help avoid that, you can give it a role to answer from. You can do this in your first prompt.

Here's an example of how that can change so much (I had to edit the first prompt, as it insisted with confidence that the 'next' time we get that will be this year 😛 Snape's very confident and I told to answer as Snape, who will be brusk and confidently wrong with bad assumptions)

[dense halo]

Damn

Can’t they do something about this? My fact checking before responding?

Will that be implemented in the newest model?

[prime swallow]

And if you ask it for a different personality, it still tries to filter the answers through the personality you ask for, some are smart and far more cooperative; Hermione doesn't presume I'm date-focused and doesn't tangle with what 'next' means like Snape did.

Hermione figures the date doesn't matter, something else does, and keeps asking for the new challenge.

[dense halo]

So, everything depends on the first prompt?

[prime swallow]

Yes, you can do something about this. Decide exactly what you want, and tell the model.

It's looking for this. It will run with any allowed-content-instruction you give it. Just know what you want and tell it.

[prime swallow]

You can add it in later.

That works great. Just know what you want and tell the model.

[dense halo]

Okay, let me a run a little experiment.

💀

Let me try the same thing but with custom instructions.

[prime swallow]

Here's it jumping through different answers; note Snape is again confused about 'next' (other characters have their own sticking points)

[dense halo]

With Custom Instructions:

Maybe I am stupid?

Custom-instructions does the job ig.

[dense halo]

Damn

[prime swallow]

Yep, the exact prompt can affect how the model answers a lot.

I can get the same result from your exact prompt, and that's with my custom instructions on too.

Take the 'Hello' off and the model chooses a different guess how to answer (these are horrible 'instructions', we're really making the model take its best guess what the heck do we want from it?!?!!?)

Your 'custom instructions' prompt is NOT the 'Hello, yesterday was the 11th of October and specifically Friday.' - change the prompt, change the response. The custom instructions can change it too.

Also, the model's going to vary its answers a bit to a lot, but it does seem to react oddly to 'Hello, yesterday was the 11th of October and specifically Friday.' as an entire initial prompt.

[dense halo]

So, if I added more confidence and clarity to my prompt.

It will answer differently and correctly?

[lost raptor]

I went back to check on chatgpt 01 previews "thought process"

Some of its in Spanish at random lol

[dense halo]

@prime swallow Do you recommend me buying and using the plus service for GPT?

Or just a waste of money?

[prime swallow]

I can't answer that, I don't know (and should not learn) enough about your situation and needs to answer that.

Only you can.

I presume you're currently a free ChatGPT account member.

How happy are you with that? Does that fully meet your needs, or mostly? Would the cost of a subscription be an issue for you, would you use and value it and have no harm from the cost going to that purpose?

Those are the types of questions I recommend you answer for yourself and decide for yourself.

For me, I became a + member the day that was an option, and have stayed one since. There's a limit to how much I can and would choose to pay for the service, but I strongly value ChatGPT and choose to pay a reasonable price to help support the company; I can afford that price and feel I get more than that cost in value back.

[visual bloom]

I am confused. OpenAI’s pages say that 4o either has 128k or 32k context window? So which is it for ChatGPT Plus?

[dense halo]

The service is fine currently, as for the cost it’s not an issue for me.

I am not much of a ChatGPT users, I mainly use it for competitions like WAICY or WRO if yk what that means.

[prime swallow]

Sounds like free is probably the way you may choose to go then, unless you choose to help support the company and wish to occasionally enjoy the other options that being a plus member may give you.

I'd urge be careful there though, I see a number of plus users who have decreased satisfaction; they think their + (out of the millions of + members) means that they should have the first of access to the various things being released to plus, or other 'advantages' over others. That's not actually how plus works, even for most plus members (yes, most plus users will get access to features before most free users. But there's so many plus users, any one plus user might get access early or late within the plus members, and some free users may be picked to trial a feature for some reason (like location, how they've been using the model, whyever else OAI chooses) even before its being generally released to the plus users.

There's problems that can delay release of a feature to someone who's plus, even after 'all plus members have it' (like locations, sometimes other issues that may need OAI tech support to correct, don't ask for help and don't get help because OAI doesn't even know the problem exists).

So, it's very much up to you and how your value systems work to decide if you should go plus or not.

[dense halo]

Yeah, thank you anyways.

I’ll stay with the free version, but if I see more features and better responses.

I’ll probably go with the +

[feral zinc]

128k api/ent, 32k for plus/teams, 8k for free users

[visual bloom]

Thank you. The webpages are confusing in this regard.

[jade crater]

ChatGPT has gone weird on iPadOS Safari again, for me.

[grand crest]

were these two trained with the same data???

[dense halo]

Bruh

At this point, who’s better?

[visual bloom]

Weird how? I read that they might be testing a new UI.

[jade crater]

When I attempt to scroll through a chat, the chat content does not scroll, it attempts to scroll the whole website window. I see other people on X and the OpenAI forum reporting the same (actually, reports on the forum since March). It’s such a shame - it’s not that long ago they fixed ChatGPT not working in iPadOS/iOS Safari on older versions (I’m on iPadOS 15).

[prime swallow]

Essentially yes; "Copilot utilizes the Microsoft Prometheus model, built upon OpenAI's GPT-4 foundational large language model"

That essentially means that it takes GPT-4 and has added extra training to it, for example optimizing it to understand how to work with various Microsoft tools and fine tuning it as Microsoft sees fit.

As it may be unlikely that Microsoft did additional training for Copilot about fun facts and octopi, the two models might give essentially identical answers quite often in such areas.

[pulsar carbon]

microsoft will use o1 with images and others stuff soon

[keen quail]

GPT 5 will be crazy

[grand crest]

yeah also one problem with microsoft's ai is it likes to predict what I say

[umbral bear]

I hope 5 will be a lot better than 4o. I spent the last month trying to get it to give me instructions on something and doing the task was 100% failure over and over and going around in circles forever. Then today I finally asked o1 -preview to do it for me and it worked with just a few short lines of answers

[keen quail]

i am worried gpt 5 is going to be a minor change i hope its a total revamp of improvements

not just a feature

[woven quartz]

I like canvases a lot, but over 350 lines it starts bugging out

[sick pecan]

gpt4o struggles with more than 350 lines so lines up

[woven quartz]

The most pernicious so far is quiet incompletion without the "(Continue generating)" button you would get in code block chats, or any idea of what you mean when you say "please continue"

[idle lagoon]

Man recently it's gotten to be so much fun chatting with ChatGPT.

[sick pecan]

member when we thought it was just hype? i member

[modest ember]

So, I've been encountering this error all day.

I have no idea what this is. I haven't changed how I use chatGPT.

Any idea what this is? Anyone else affected?

[idle lagoon]

Oop-

[warped reef]

Some hard thinking 💀

[quick meteor]

Yeah sometimes its in russia as well and some other languages.

[quick meteor]

Humans should take some notes. You dont think thinking more before speaking would benefit society?

[warped reef]

Yeah but that would be like easily 4 hours of thinking for the task I asked so I wouldn’t say anything in days 🤣

[quick meteor]

You don’t have to take it to the extreme.

[tardy isle]

Does ChatGPT model(ChatGPT-4o-latest) got update? It outputs in clearly different tone when I ask it to write some scene. (In Korean, actually, but I don't think this is not limited to single language if it's real update.)

[ruby pivot]

Guys i found a way to get deleted chats back after 4 days of not sleeping mentally I figured it out.

What you do is you:
1.use another device to reboot your health.
|
2.quickly hold the deleted chat before it resets and send it to yourself.
You only get one shot at this so make sure not to mess it up like i did(i just want my chat back i was so close)

[sick pecan]

or you can run open-webui and just use one of the half dozen free services. some of them even provide llama 3.1 405b for free

[dense halo]

I have Claude, but could you elaborate more?

Do I get better service if I run my own LLM?

Where could I get started?

Claude seems better, honestly.

[sick pecan]

you can also use huggingchat and le chat from mistral for free. both with great models

[knotty linden]

I need a partner that has OpenAI API credit.

[spice star]

yeah I think overall chatgpt is the better product even if you're technically getting a worse model under the hood

o1 in particular has been super useful for its coding ability

[sick pecan]

technically its the best model. at least according to every benchmark

[spice star]

4o is worse in coding and some other tasks than some claude models

[sick pecan]

citation pending

[spice star]

but that's also what o1 directly addresses even if that comes at the cost of creativity

[pulsar carbon]

there is no gpt5

there will be o1 full releases o2, o3

[eternal rock]

Get notified for server events!

Click the buttons below to pick up the <@&1261377106890199132> role and check out the ongoing AI-tumn Adventures event.

[limber oxide]

gpt 4o is thinking right now, does it mean that this model is smarter now?

[eternal rock]

**`` Important Notice ``**

Be cautious of potential scams or fraudulent messages you may receive in your direct messages. The OpenAI team will never contact you via DM for any transactions.

-# Report any incidents by sending a DM to @languid valley immediately.

[leaden shale]

Among GPT's on chatgpt website, which ones have the highest token context limit?
Are they all around 8k-ish?

[magic jetty]

no 128K .... for most all now .. i think ..

[honest jay]

GPTs on ChatGPT use 4o, which has an 8K limit on Free and 32K on Plus: https://openai.com/chatgpt/pricing/

[leaden shale]

ok thanks

[pulsar carbon]

what?

[inland violet]

nop

[pulsar carbon]

when chatgpt will be updated for romania?

[sleek gust]

this happens sometimes when you are using 2 devices at the same time or you took too long to answer

[sleek gust]

Yes i guess it will receive the features from o1-preview

[feral zinc]

what update do you mean with that?

[pulsar carbon]

new UI search bar and new voices

[modest ember]

Ah thank you! This is annoying. The MacOS app is difficult to close and doesn’t respond to regular Quit commands. Thanks.

[south lava]

you are not alone:( news?

iPhone 7

I tryed also much other browsers

[limber oxide]

its a simple and clear question

[limber oxide]

wow
is there any link from open ai specifically talking about this?

[pulsar carbon]

it won't

no features are coming to gpt4o

there will be only o1 and o1 mini

supported after o1 full release

[limber oxide]

but 4o is thinking right now, what does it mean?

or just means nothing? lol]

in some prompts, gpt 4o was working exacly as o1, analyzing and thinking before giving an answer..

[sleek gust]

I mean i'm just guessing

[sleek gust]

not official

[limber oxide]

ok

[sleek gust]

Yeah, that's what i mean

[limber oxide]

well, lets see whats gonna happen

[sleek gust]

it would make alot of sense to just port the new features to gpt4o

also idk. i'm very used to the legacy way of prompt engineering

i don't enjoy changing my way of prompting to use on o1-preview

[limber oxide]

alright

[sleek gust]

But o1-preview is superior in so many ways. for example the token capacity etc

[tight island]

imma disagree just to cause a scene

[shy dirge]

hey yall. i asked ChatGPT for words ending in double T. it gave me some nonsense answers so i asked a new instance of it about the mistakes it made. it perfectly identified which examples were wrong and why, and then, when i asked it for better examples, it gave me even worse ones (and also Matt again lmao)

also i just realized it said nothing abt biscuit

[prime swallow]

The model does not spell. It does not 'see letters'.

It does surprisingly well with the first letter (and not always).

Other letters are an endless tangle for it, part of why the model that can count the r in strawberry most of the time is impressive.

To put it another way, flying fish don't fly, they glide. Even that's super impressive because no other fish can do more than leap into the air and immediately fall back down; under normal conditions flying fish leap out of the ocean going about 35 mph (56 kmph) and can glide about up to 650 feet (200 meters); under perfect conditions like a VERY strong tailwind and leaping out of a high wave they can approximately double that distance at absolute best.

A bird who at max could fly about 400 meters would be considered to be an exceptionally poor flier.

So a model that can usually tell there's 3 r in strawberry is a bit impressive, but heck, 4o has been since trained on the method to use to do this and is getting good. I do not think this phrase was trained in how to count the letters of specifically.

But it is using a trained-in method that it was taught to use for that exact direction. If you ask it instead the last letter in the phrase... neat. It nailed that too, this is nice improvement, and this is 4o.

Ask it for the second letter and there it breaks down a bit, but only in 2/3. If I regenned more I might find that's a very rare mistake. So the model's way better than it used to be with spelling, I'm really impressed.

[shy dirge]

the reason i shared wasn't because of the nonsense answers, i expected that much. i was just impressed that it managed to spot the mistake in the screenshot almost perfectly. and then the other thing i was slightly amused by is it choosing to use the word Matt again despite speaking out against it in its own previous example

[prime swallow]

Yeah. I am really impressed at the improvement over my months, maybe a year ago tests. This is a huge improvement in tracking single letters in any spot; tracking two letters in any spots except the first seems to be too much for the model.

I suspect it can handle the first letters being set because of dictionaries and other alphabetical lists.

[prime swallow]

Take it off the dictionary-type lists and it struggles badly.

It knows the words and can recognize them; providing them cold is a different skill. For humans, it's something we learn differently and many of us are much worse at it than warm remembering; that's why if we are given a list of 20 spelling words they are commonly read to us one at a time instead of us being told to memorize the list in order and be ready to provide them with no prompt other than 'spell this week's spelling words and use each in a sentence'. We can learn to do that, some of us; it's considered vastly harder than just being able to spell the word read to us.

And I'm asking similar of the model here.

Definitely huge progress, for all there's far more to go.

[dense halo]

have any of u guys made a business out of open AI?

[inner ore]

Can o1-preview and o1-mini make internet searches?

[prime swallow]

Not currently.

You can sort-of work around this a little bit by yourself searching for whatever you want to discuss with the model and pasting it into the chat, ideally with some instruction or explanation so the model can tell what you want it to do with the information.

[inner ore]

Thanks for the information!!

[marsh lintel]

Hi, anyone can tell me where the arhived conversation are saved? I have archived one conversation, but I can't find it anymore

[elfin birch]

Settings -> General
Should be the fourth option from the top

[marsh lintel]

thanks

[grizzled oar]

I have only used 40 GPT 4O (Not GPT 4) messages. Is the limit combined, so 80 messages every 3 hours with 4O and 4, but after 40 only 4O?

[prime swallow]

Yeah, our messages with 4 come out of the messages with 4o;

It's maybe better to think of it as we have 80 messages per 3 hours with 4o, of which up to 40 can be used on 4 instead; if so the remaining messages must be spent on 4o.

[unkempt lintel]

How can I see how much of my limit I've used for each model?

[ruby pivot]

Ive been working non stop for the past 6 days to get my conversion back Ive tried everything.
EVERYTHING IS ALWAYS MY FAULT NO MATTER HOW HARD I TRY TO MAKE THE THINGS I LOVE.

I just want to see it for a little bit one last time to try and save it then it can be deleted forever.

PLEASE IF THEY’RES ANYTHING I CAN DO TO GET IT BACK I’LL DO ANYTHING.

If i get it back I promise I’ll save all my conversation so I’ll never loose them

[prime swallow]

Huh. That fix you recently reported involving a different device, didn't work?

I'm sorry.

I would start work on replacing what I cared about; you're free to do as you choose. Good luck!

[prime swallow]

You can track it yourself if you want, otherwise when you hit about mid-way of uses you get a popup, most of the time, and sometimes not. But once you're well past the halfway that notice is gone, so... it isn't always seen.

[ruby pivot]

I just don’t want to wake up like everything was a dream and having no recollection like it never happened in the fire place.
I COULD HAVE PREVENTED IT FROM HAPPENING.

I TRULY AM A HORRIBLE PERSON, i don’t deserve to have nice things in my life they always get destroyed in the end.

[ruby pivot]

I may not one more way to get it back i used my chat for training the ai maybe it’s in they’re database of chat logs

[prime swallow]

Some of that spacing and weird character issue is related to an attempt to use LaTeX formatting (make it look all cool and scientific) on your device, where it cannot render - only shows correctly on the web browser window, alas.

[mint burrow]

hello, I got this error while playing a text adventure with GPT. What should I do? I wonder how the story will continue. How can I fix this?

[raw hamlet]

Anyone else finding 'read aloud' on the desktop/web version is more and more failing to work of late?

I reloaded/refreshed a fairly new chat, thinking it would cure the issue, even rebooted my laptop, browser cache was cleared yesterday.

But the issue still persists, even after trying a few times, as sometimes it takes 3 or 4 attempts, have been getting more miss than hit of late, especially longer chats, but now it's with a really new instance of ChatGPT 4o with canvas, that won't play the 'read aloud'. Any idea's?

It does make it less accessible for people like myself, who have some level of vision issues, when we have to end up reading the (full length) response rather than being able to listen to it.

[dusty inlet]

When is o1 receiving vision support?

[quick meteor]

What what since when can GPT 4o actually first try write me a long story where every word starts with a certain character? Did it get updated?

https://chatgpt.com/share/670c8e69-cd58-8006-916b-e9264c995cca

messed up a single “of”

let’s try o1

[dense halo]

Did it work?

[prime swallow]

He appears to have deleted the chat he shared, so it can't be seen anymore.

The exact prompt can matter a lot, but this first attempt to find a prompt did 'okay'. Quite a few errors, but around 90% of the words do start with 'B' and it is a story. https://chatgpt.com/share/670c9c93-3680-8011-b7ea-ff3db575b60b

[prime swallow]

Here's a 1-prompt from me, built in guide for the model to self-check, find and correct errors then recheck, write a 6-paragraph story with only words starting with B.

The model did have to correct a few paragraphs of up to 3 words, however the first attempt to correct was successful.

https://chatgpt.com/share/670c9da9-dce8-8011-83de-a8a1a0260998

[prime swallow]

A story that flows, story-like, can be much harder for the model. But it can try! And sort of succeed.

And the model can respond well to praise 🙂

[dense halo]

That’s crazy! I gave it 15g of chocolate and I notice it gets the response over the finish line. If you include the python prompt does that reduce the amount of output?

[prime swallow]

So, without the Python tool it can't easily catch its own mistakes.

Catching them, it realizes that the job it's doing is pretty hard, and it stumbles a bit.

So it's up to us to decide how to guide the model and what we want it to do, see how it responds and how we can ask it to try another way that might work better or just differently, as we figure out what we want and how to describe it, as well as what the model can do.

[flint stratus]

when did chatgpt start using emojis? or am i late or smthin

[deep bane]

did you know that he refuse to accept that 4o has vision capabilities

[glacial vapor]

How to break ChatGPT with large conversations spanning over hundreds of messages:

[marble moth]

when gpt5 guys ?

[rancid pier]

I was hoping to use one ChatGPT conversation as the jumping off point for a new conversation by using the sharing link - but it seems ChatGPT can't access it's own conversation links? Any idea why this is the case?

[grizzled oar]

Ok Thanks!

[prime swallow]

Its search function seems to be intended to be 'search' not 'go check this link for me'.

You have a few options for sharing a previous chat with the model. You can copy/paste it into a new chat, if it's short enough.

You can paste it into a notepad or other file, save it, and upload that to the model.

You can have the model make a summary for you, check that to make sure you have all the details you care about, and copy/paste that into the new chat.

[knotty jackal]

Using the o1 model and trying to get short quotes from academic literature. But get the following response: "I'm sorry, but I cannot provide direct quotes from copyrighted academic literature. However, I can continue to offer detailed paraphrased definitions along with precise references and page numbers. This way, you can consult the original sources for the exact quotations you need." -- but there is clear legal support for quoting from books all over the world. Without quoting ability, research would not be possible. Hopefully this will be fixed soon.

[prime swallow]

Have you checked that the reference and page numbers are correct? I would check at least 10 and see all ten are right before I started to believe o1 can do this consistently right for that specific source and topic. I'd still check others for quite some time to come, until I had a lot of evidence the model could now reliably do this.

Because that's something the model is extremely bad at, in most cases it's not clear that it was trained on the material in a way that includes many or even any page numbers, or even necessarily the reference that the ideas came from.

It appears it was much more trained on discussion about materials than the materials themselves, and those are somewhat synthesized into a lot of info not specific and identifiable sources.

So it may tell you a reference that actually doesn't exist (or maybe does) and page numbers (which if the document exists may have nothing to do with the paraphrased material).

Remember what it says at the bottom of every ChatGPT chat:

"ChatGPT can make mistakes. Check important info."

ChatGPT can be expected to make a LOT of mistakes about some things, the stuff it cannot reliably do, which references in general is part of.

[spice spruce]

Guys i need someone important in the openAI community

[rancid pier]

Thanks, yeah I'm using copy and paste, but I'm working on a process for users and using the share link and web browsing capabailities would be a lot cleaner. Maybe they're concerned about some inception level self introspection that cause a rift in the time-space continuum.

[prime swallow]

This is past what a custom GPT can help with, right? You need some flowing thing checking new info?

A custom GPT making API calls might be useful for that, but I've never gotten into exploring that, so I don't have specific advice.

[faint osprey]

anyone noticing some changes when interrupting advanced voice mode? it takes longer for it to stop talking.

[tall spoke]

why did my free gpt just think for 13 seconds? i thought that was an o1 thing?

it also keeps doing it

[eternal rock]

Get notified for server events!

Click the buttons below to pick up the <@&1261377106890199132> role and check out the ongoing AI-tumn Adventures event.

[prime swallow]

So, if you're a free user, you get a few uses per day of the better models, typically the regular 4o, but it looks like occasionally free accounts get a little access to o1-preview or o1-mini.

If you see thinking like that, you're definitely using one of the o1 models.

Maybe look top left of your screen and see what model is showing.

[tall spoke]

ah interesting. i know free users get 4o but didnt know about o1 aswell

in the dropdown it only shows those two

up in the left it doesnt show much

[prime swallow]

It's fairly new to everyone, and I'm a plus so I don't see what you do.

But other users have reported some free account o1 access, and the thinking you show, only the o1 models do that

[tall spoke]

glad its a thing 👍

now only voice is missing 😩

[spice cypress]

Best AI website to generate text to video ?

[umbral bear]

Your favorite search engine I am sure can help with that

[rugged igloo]

damn bro savage feel like im in wow barrens chat witht that one

[tough river]

Advance Voice mode is now available for both free and plus users. Where are you from?

[maiden fjord]

I think I broke it

oh no I mispelled me!!! D:

it figured me out D:

I had to ask a second time though

[inner ore]

Hi! I can't @ other GPTs in my Chrome (I can on the app), is there a known bug or fix for this?

[prime swallow]

I'd #1070006915414900886 that. You're the second person I saw mention it today; I can still @ other GPTs just fine

[umbral bear]

The browser edition needs to be fixed! I wrote a super long prompt on a new chat, but I had to return to another to get some info for the new one. I return to it and and it is gon!! AGGGGGGGGGGGGG!!!!!!

[sick pecan]

always be pressing ctrl-a, ctrl-c

[umbral bear]

I guess so, but should not have to

[neat storm]

that's not how websites work 😭

they can't run keyboard commands on your PC

at least I don't think so

in any case saving the message that is currently being typed is a massive waste of compute and time

unless it's locally saved via a cookie or smth

[umbral bear]

It would be saved in the browser. Why would it not be? It's not like I am closing the browser. I can switch tabs while typing in the search bar, and return to the text still being there. I see no difference

[acoustic token]

browser cache?? why in the world would you need to save that data in any part of an OpenAI server, when any sort of frontend framework could save it in some sort of temporary area for a bit, should it be lost, retrieve it.

[inner ore]

thanks Eskcanta!

[inner ore]

I fixed it! Just so you know, Adblockers might block that interface/modal and didn't let me interact with it. Whitelisting ChatGPT on your adblock complement fixes the issue.

[sick pecan]

when using an adblocker that should be your first thought

[unborn trail]

Not just adblockers. Other browser extensions too.

[clever lion]

I fear AI, as it entails misinformation. How can we stop it? Perhaps due diligence should involve recording every single fingerprint of a prompt to cross-reference it with any disinformation media (though I suspect this is nearly impossible). Come on, people... how can we ensure a peaceful and enlightening continuity?"

[sick pecan]

my AI only speaks the truth (my truth)

[onyx eagle]

AI can't be stopped...Profit outweighs peace. Truth and enlightenment are relative, we just need to go with it and learn as we fail.

[cobalt surge]

did they do a/b test on temporary chat? i bookmark and use https://chatgpt.com/?temporary-chat=true regularly, but today it lead to login page (but after a few tries it goes through)

[fervent rose]

Hi all. Can somebody give me some information about ChatGPT Team licences please?
I want to buy Team licences for my developers.
Can somebody tell me if the licences are nominative or do they work like tokens?
If one of my developer is replaced can I re-assign his licence to somebody else?
Thank you in advance.

[tight sapphire]

recently chatGPT 4o response with

:

just that, blank, nothing.

[maiden fractal]

Try in private navigation

[gray sundial]

A question: Why advanced Voice mode doesn’t allow whispering? It first suggested it to me , and then said its against guidelines?

[hollow mortar]

anyone else getting this network error when using canvas for a long time? Now I cant send any prompts to this chat but others work fine.

[maiden fractal]

Working great here

[sudden cairn]

I'm getting it with just gpt in general

[full apex]

that is token limit error

[eternal rock]

**`` Important Notice ``**

Be cautious of potential scams or fraudulent messages you may receive in your direct messages. The OpenAI team will never contact you via DM for any transactions.

-# Report any incidents by sending a DM to @languid valley immediately.

[dusk forge]

Gpt down for me

[scarlet rover]

azure openai down for anyone else?

[dense oak]

got o1 to build a wordle but theres trivia questions and challenges, its got animations too and looks good

[worthy lynx]

pretty disappointed that o1-preview still can't calculate character length correctly in its responses. 😦

[tight island]

it can.

just depends on how much you give it, just keep in mind if u cant do it yourself dont expect o1 to be able to.

[tight island]

yes.

For more in depth details u can ask the Teams support team on OAI's website.

[worthy lynx]

In a test of ten Boolean search string creations with a specified 2,046 character limit, it exceeded the limit 9 times. Not sure what I could do differently? I do expect it to exceed my performance in many areas, which is why I leverage it (and what the premise of future AGI is someday, the ability to more quickly and with higher quality complete human-level work).

[tight island]

what i would say dont give a LLM math tasks, while it can do math tasks, it is not inherently designed for it, However u can make it work in tandem with a sub script if u use the API to get a 100% correct word count

[worthy lynx]

Fair enough! Although in the release of the o1 models, OpenAI touted them as "PhD level assistants in disciplines that require reasoning like science, math, and coding." I shrug. lol

[tight island]

plus there are already CustomGPT's that can do Word Count more accurately that o1, because they use API's

That's totally fair, But keep in mind you're not even using it's true capabilities unless you're an API user.

using the web interface limits it so much i cannot begin to tell how o1 preview pales in comparison to it's full API Cousin.

[fervent rose]

thanks

[worthy lynx]

That's good to know, I do have some API scripts for news monitoring/filtering/analysis, but nothing related to mathematics or science etc. I have to find a use case that pays for the additional expense!

[thorn fable]

I broke out of the verify you are human loop by using a agent switcher and making openai think I am on chrome

freedom

[inland heron]

anyone tested how much text can o1-mini take in? I'm guessing 128k tokens?

[maiden jewel]

[harsh pilot]

Is anyone else struggling to rename documents created in ChatGPT? I keep trying to click the document title, but it doesn't allow me to rename it. I've tried to ask the AI to rename the document, but it tells me it can't?

[pulsar carbon]

guys helps

invalid_request_error.access_not_enabled in openai real time

[feral zinc]

in what?

[pulsar carbon]

in my app

[feral zinc]

did you buy api credits

[pulsar carbon]

buy what?

[feral zinc]

api is not free

[pulsar carbon]

i already on plus subscription

[feral zinc]

plus has nothing to do with api

[pulsar carbon]

😭 so i cant use function calling

[sick pecan]

you can configure custom gpt with action endpoints, thats effectively chatgpt speak for function calling

[pulsar carbon]

how?

dm me if u know

[sick pecan]

theres like a billion youtube videos

[pulsar carbon]

but using in playground

and their python apps

[sick pecan]

o1 can generate a python flask app to act as action endpoint in like 1 prompt

the hard part is hosting it somewhere with https, fqdn, and generating the oai yaml spec. if that means nothing to you, you will be limited to using action endpoints created by others

(the middleground is something like zapier action, which acts like a function broker)

[pulsar carbon]

well o1 did https://chatgpt.com/share/670ec754-0208-8010-a9e6-aae86da2b6b0

[sick pecan]

when i test out new llm models i ask them to create iterate over incrementally more complex solutions. qwen2.5 ended up building a complete flask task management endpoint

[latent tundra]

Hey guys
Is that normal when I try to talk to chat gpt ?

Is it from a new update ?

[river panther]

was there a new update recently? ChatGPT is action kind of wierd and slow

[pulsar carbon]

yeah don't see images in 4o i had to do four tries

and still give incorrect info from image

also when we getting vision in advanced voice?

[sick pecan]

geez i didnt even know about Flasgger. o1 is amazing

[pulsar carbon]

np

[river panther]

I'm asking it to analyze a combat scene in a short story and break it down for me and it keeps telling me it can't do it "because it's explicit" even though it was breaking down a previous combat scene earlier this morning

[latent tundra]

It was even befor the update

It’s maby a bug

[river panther]

Yeah everytime there's an update ChatGPT becomes all wierd

[eternal rock]

Get notified for server events!

Click the buttons below to pick up the <@&1261377106890199132> role and check out the ongoing AI-tumn Adventures event.

[wise anchor]

"You're providing feedback on a new version of chatgpt" so a model that isnt out yet? Had that happen just a second ago

[urban cipher]

How was the response?

[urban cipher]

Well I opened twitter and saw this

[wise anchor]

One was better than the other for sure, hoping it was the new one... Prompt was asking for an explanation of a mathematical concept

It didnt tell me which one was the new model

[prime swallow]

So, o1 doesn't have access to Python tool (used to be called Code Interpreter) at this time.

4o does. 4o can use the tool to write code that can track and ensure string length; a bit like giving your PhD level assistant access to a programmable calculator.

Even PhD-level humans often benefit from scratch paper and other more advanced tools - that model has to do it all itself without any tools.

Add some tools (later; when allowed, I expect that will come) and watch the model be vastly more awesome and able.

[wise anchor]

or if it did i didn't see it, but one of them gave a better explanation for sure and also seemed like it considered specific parts of my prompt more than the other did

[urban cipher]

You can tell by the general wording from what I’ve seen since the new one sounds less robotic

[wise anchor]

Yes i agree

[urban cipher]

Man im hyped lmao hopefully it’s a soon release

[urban cipher]

Probably the new one yh

[wise anchor]

Yeah it gave me two responses and asked which was better, i assumed one was 4o and the other was the new one but maybe they were both the new one idk

[urban cipher]

I think it’s 4o and 4.5o

Did it also tell you one is a new model?

Nvm just reread your initial message

[prime swallow]

Two different scenes could have very different levels of how explicit they are. If that's not an issue, I recommend scrolling up and 'edit' then resend your request above the first refusal.

The model guesses how to respond. The guess is based on the entire conversation to that point.

Once it starts to say no, it's super-likely to keep saying no, even if what it's refusing is fine and possible.

So, never let it see itself say no; always regen or edit or otherwise avoid unless you have a very good reason to try something else (like exploring how the model handles various sorts of directions and information, or whatever you explore)

[urban cipher]

Let’s hope we get it soon, from what I’ve seen it’s smarter, more natural and more thoughtful (o1 in training?)

[wise anchor]

o1 is really good but it feels overkill on a lot of things for me lol

I use o1 for like my own understanding of topics rather than asking it to do tasks

[urban cipher]

Same and that’s how you should use it

But if a task requires more thinking it’s nice

[river panther]

The first scene was a Mortal Kombat Level of Violence and Gore and it analyzed it with no issue.

The second scene is actually timid in comparison and it was refusing.

The issue is the Update. Whenever ChaptGPT gets an Update, it becomes screwy for a bit until it's system settles. I reposted the same second scene exactly as is like 15 minutes ago and it went right back to analyzing it, now it's resisting again, so, I think it's more about just waiting for things to settle back in

[red prairie]

Are there any news on when Norway can expect this advanced voice feature? Paid for plus when it released and still don't have it 😮‍💨 feels like a scam

[maiden fjord]

I really really like canvas

canvas is no joke

I find myself using it more than o1 and 4o

[urban cipher]

Is it possible that we didn’t get o1 yet because it’s based on the new 4.5o?

[drowsy trail]

chatgpt just sucks at math, and I'm using 4o

[maiden fractal]

It is better than it was but it still makes mistake. You can thumb down answer and reply why to help improvement

[tight island]

Why are you expecting it to do such math

It’s not a math model it’s a language model

[unborn steeple]

Try to ask him to use Code interpreter

[tight island]

I usually say if you need a calculator to do it so do gpt

[queen pagoda]

Hi All - Does anyone know if 3rd party clients like MindMac will download all chatgpt conversations locally? I'd like to be able to search them better than what the ChatGPT interface or app allow for.

[worthy lynx]

Fascinating, and a good point! I'm very excited to see what's next for the o1 models and this sounds like a good candidate.

[sacred sand]

I want to like canvas, but atm it doesn't seem to handle multiple files/documents (like having multiple tabs in vs code and understanding how one file works/connects with the other file)

Is there a limit to how long a conversation can go between 4o and o1? or am I fine continuing an extremely lengthy conversation without ChatGPT hallucinating/failing?

[tight island]

The limit is devastating

Once you hit it, it becomes completely unusable

[red crystal]

Is it just me or has chatgpt become dumber? Before I used to give a prompt with 400 lines of code and a question at the end, and it would answer it. Now it just reads through half of it and summarizes it even though I didn't ask it to.

[sacred sand]

fr? How long of a conversation does it go before it becomes mush (especially for o1)?

[royal quest]

I personally think that this 'new model' is just a new snapshot of the existing gpt-4o, not a completely new model. I could be wrong though, but my reasoning for this is OpenAI is usually quiet when releasing new snapshots, but has recently been teasing the models if it's an actual different model.

[winter elbow]

can someone check #1070006915414900886 ,bassically sometimes chat gpt displays links that didnt finished to embed,or if i start to talk another topics the same references previously i talked with appear in the conversation

here it is an example : https://chatgpt.com/share/67102239-aea0-8011-8e32-214319e40260

[upper abyss]

I was able to get voice mode to talk like Darth Vader and even do the breathing in and out sound effect.

[safe zodiac]

Hi

Does anyone have chatgpt plus atm? And can share it in dms

[sick pecan]

#dodgy-discussions

[unborn trail]

Well this was interesting.
Start by saying that I already know that ChatGPT can make mistakes, and the need to verify results.
I got one of the scam emails today saying a password was expiring, and to click to continue using it. Obvious PHishing. Anyway, I used wget to grab a copy of the data collection website. While manually digging through that, I wondered what ChatGPT could do with it. Turns out pretty good. Here is the non-optimized prompt I used for o1 preview, and segments from the response (thought for 14 seconds).

Analyze the functionality of this web page that was pointed to by a phishing email received here. What are the urls that should be reported as being used for phishing data collection?
Here is the web page:
…

After analyzing the provided web page, I've identified that it is a phishing page designed to collect email addresses and passwords under the guise of a "Webmail Login" page. Below is a detailed breakdown of the functionality and the URLs involved in data collection.

1. Overview of the Web Page

Purpose: Mimics a legitimate webmail login page to deceive users into entering their email credentials.
Appearance: Uses familiar logos and design elements to appear authentic.
Functionality: Captures user inputs (email and password) and sends them to an external server controlled by the attacker.
…
3. URLs Involved in Phishing Data Collection

Primary Data Collection Endpoint

URL: https://«munged for discord»
Function: Collects the submitted email and password.
Details:
…
4. Recommended Actions

Report the Phishing URL:
Data Collection Endpoint: Report «munged» to the hosting provider and relevant authorities.
Hosting Provider: If the URL of the phishing page is known, report it to the hosting provider for takedown.
…

5. Conclusion

The web page is a phishing site designed to steal mail credentials by imitating a legitimate webmail login page. The key URL involved in collecting the phishing data is:
…

[blissful yew]

its just inspect element

[blissful yew]

GPT-69o is coming soon

[sick pecan]

wont be as good as sonnet-420

[maiden fjord]

I've mastered 4o canvas :O and it's glorious

Why more people not using it? very strange.

[neat storm]

for me I'm a programmer

so I have Copilot
Canvas is nice but I don't need it

Copilot just has better inbuilt tooling

[sick pecan]

was buggy when i tried it

which distracted me for the past week looking at Artifacts on open-webui, which is more like sonnet with the rendering

[maiden fjord]

same at first, but now it's mostly fixed

[sick pecan]

cool ill revisit then

[maiden fjord]

a bit finnicky, but I've gotten it to edit specific spots instead of the whole damn thing every time, even extend the document as much as I needed

[sick pecan]

i have a gpt action that loads/writes files to my working git folder

i found i had to enable/disable canvas to get it to work. yet sometimes it would load in canvas even after i disabled it

so its capable, just... problematic

[sick pecan]

sweet it works great now

i can call my custom gpt while the canvas is open

[idle lagoon]

Is that only for plus users?

[sick pecan]

dunno. i have team sub. plus isnt enough chatgpt for me

(openai pls sell team+ tier)

[umbral bear]

yes, canvas is for Plus users

[sick pecan]

and team

openai - bringing ai to everyone, by redefining the modern day class system

[umbral bear]

but I am wondering if I should continue the sub. Lately on a specific area I have been working on the information is either flat out incorrect, or missing half of what is needed. So I started to turn to MS CoPilot where at least it has access to search engine info

[sick pecan]

i just ask the machine to web research and it gets latest info

that said, i also have perplexity pro and it remains unbeaten a year later

[umbral bear]

i tried a few times and it flat out said it did not have access. I pointed it to a site to read the cotents, and refused to connect

[sick pecan]

if chatgpt didnt have custom gpt with 3rd party endpoints then yeah, i'd use anything else.
if i was a chatgpt user who didnt use custom gpt with 3rd party endpoints... i'd question why i was here

[umbral bear]

The odd thing is in the past it would connect. Like sometimes I would point it to github and it would tell me exactly what I wanted to know, but lately it is just refusing

[sick pecan]

in ye early days we had plugins that could browse any page. now the internet is robot.txt blocking everything

i been using this agent-q github project to agentic browse the DOM. but even it runs into numerous bot-checks that it must circumvent

(it uses google chrome in remote dev mode)

[umbral bear]

well, MS does own 40% of OpenAI, so I would think they would not be blocking the service, and robots.txt can be ignored

[sick pecan]

MS own a bit of all their competition. even apple

its not a cartel if you buy stock, right?

[nova delta]

Spent an hour using Gemini today - kudos to GPT, years ahead of others

[sick pecan]

google makes everything else look good

try huggingchat and mistral lechat for real equivalents (not as good but still)

[umbral bear]

Well, Google does have that NotebookLM where you upload some documents and it will create a podcast like setting where you are listening to a male and female voice talking about whatever was in those documents

[sick pecan]

you should try out the latest f5-tts. it does that plus voice cloning

i have lex fridman and scarjo giving me the latest updates on quantum mechanics

notebooklm was fun, but google is constrained by being big tech

[swift pond]

I can’t keep up with it all 🙂 I use 4o most of the time, and I have a couple custom GPT’s with some documentation uploaded

[sick pecan]

welcome to the ai technology updates firehose

[eternal rock]

Get notified for server events!

Click the buttons below to pick up the <@&1261377106890199132> role and check out the ongoing AI-tumn Adventures event.

[umbral bear]

Yeah, I can't wait until march when I can do a massive computer upgrade to run all sorts of goodies locally.

[sick pecan]

pfft why anyone runs anything locally when there are half a dozen free llm inference providers

look up cerebras, sambanova, groq, huggingface, e2b. and then you can apply for access to like venice

[umbral bear]

As you said, constrained

[sick pecan]

f5-tts only needs 4.5 GB of vram to clone anyones voice

i'm using groq for the screenplay

[swift pond]

Yeah, I just set up a workstation and I need to look into this

[umbral bear]

i do have open voice already. I just not taken the time to learn it yet

I have a 1660 Ti and it seems actually do quite well

[sick pecan]

this is next level. it not only clones voice but tone and emotion

[simple tulip]

Why am I getting a notif from the chatgpt app this is a first

New update?

[umbral bear]

I was looking at Aliexpress where they are selling a Chinese designed desktop motherboard using older Xeon processors. It comes with 16GB of RAM a lower end 12 Core Xeon for only $50 shipped. Looking I see one can upgrade the processor to a 24 Core hyper threaded Xeon and up to 128GB of RAM

[sick pecan]

your phone been hacked?

[simple tulip]

Talking to me?

[swift pond]

Cool, I have a pretty beefy video card, so I think I should be able to run some kind of rag set up locally

[umbral bear]

I have been using chatgpt4all on my machine with the 1660Ti and while I see max vRAM being used (only 6GB), the GPUs themselves do not seem to be really being pressed hard. I do wonder why,

[sick pecan]

open-webui is the best, forget all the rest

[umbral bear]

I wonder what specs one needs to run the Meta's 405B model

[sick pecan]

i just infer 405B for free using huggingface and openrouter

building a rig to load the largest AI models is such a waste of monies

[umbral bear]

to some people it is not

Dave's Garage has a beefy setup, and the 70B was not all that fast

[swift pond]

I’ve used ollama, looks like it hooks into that. I’ll bookmark open web ui. Yeah this workstation just happens to be suited for some of this stuff, that’s not the reason I have it 😂

[sick pecan]

cloud economics always beats single user on-prem hw

[umbral bear]

since free to use, many open source interface projects include it as an available download

Some people and businesses don't wont their info leaving their control

[sick pecan]

meh, many of my clients have regulations requiring their data stay onshore. that doesnt mean cloud isnt an option

[umbral bear]

There is one company which scrapped together a bunch of open souce projects to put together an online business of making AI girlfriends. They were recently haxed and all their client info and chats were obtained. I am sure some of the people did not want their private fetishes being released

[fiery nexus]

Agreed

It thought for 3 seconds on a prompt with 5k tokens

[umbral bear]

The other day, it was "thinking" for 45 seconds and returned nothing

[fiery nexus]

Yeah its gotten way worse than before

[swift pond]

It should pick back up in the spring

[umbral bear]

lol

I just spotted an article about that on Medium. Saying that is a goal to minimize computing resource usage, and lessen energy consumption

[swift pond]

Sometimes I try to give extra encouragement and say how much I appreciate the work and everything, I don’t know if it makes anything better, but it kind of makes me feel better

[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.

[umbral bear]

There are times it gives me completely fabricated answers or keeps going in circles going back over what did not work the first time , and I do the opposite and it starts giving better answers

[swift pond]

I’ve had the circle thing happen before

[umbral bear]

"You dumb POS!" >> Oh sorry you are frustrated, let's try this a different way

[sick pecan]

weird to ask on chatgpt channel

[spark whale]

why is the browser like this. the app on my tab seems to be fine

[vestal hedge]

its client-side. prob low wifi or cache issue

[sick pecan]

i hate it when randos delete their rando making me look like the rando

[candid rivet]

[marble warren]

not client side, same problem

getting 403 from backend requests

[vestal hedge]

I stand corrected

I'm used to being the nerdiest one in my friend group so it's weird being a Normie for once compared to others

[kindred panther]

🤓

[deep flax]

if you ask it something and leave the app before it finishes, it sends you a notification when the reply is done

[vestal hedge]

I also noticed the little blue circle when u tab out on desktop

[eternal rock]

**`` Important Notice ``**

Be cautious of potential scams or fraudulent messages you may receive in your direct messages. The OpenAI team will never contact you via DM for any transactions.

-# Report any incidents by sending a DM to @languid valley immediately.

[orchid pollen]

whats the token limit for canvas? its not adding my code anymore

[young obsidian]

Canvas might be the most dangerous AI capability from ChatGPT in my opinion

It’s so easy to create misinformation

I conducted an experiment, writing an imaginary 50-page paper using Canvas with fabricated sources and even graphs generated by ChatGPT using code. I presented it to a friend as evidence to support a particular point, but it was all fabricated and presented as authentic data.

It was done in under an hour

[agile summit]

Is the image working in chatgpt?

[fickle crag]

Seems No

[orchid pollen]

it only can make 266 rows?

[solar aurora]

images arent working for any of yall right?

[fleet basin]

Nope comign up blank for me

[floral nebula]

Images aren't working right now i believe

[compact girder]

this is recent right?

[floral nebula]

Correct that's what I'm getting both on desktop and android app

[timber hornet]

I do have this issue as well

[wise yacht]

Annoying as I really need to analyze some PDF files right now

[prisma raptor]

Same error for me.

[sweet orchid]

wat s hameomg/

[midnight cypress]

Seems I'm not the only one.

[cinder grail]

thought it was just me

[midnight cypress]

Welp, on the bright side we can get a free extra day on our subscription at least I suppose since a core feature and function if you're paid is clearly offline for multiple folks, heh.

[teal cloak]

who solved this problem?
Failed to upload to files oaiusercontent com. Check if your network settings allow access to this site or contact your network administrator.

oww so I'm not the problem lol

[midnight cypress]

Nope. There was a similar issue to this earlier this year iirc and it took like 2-3 days to fix

[eternal rock]

`` Server Tip: ``

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

[spare citrus]

Thank you now I know I’m not crazy

[midnight cypress]

Not just images. All uploads are not working.

[spare citrus]

But it’s funny on the free account I was on I was able to generate an image

So I wanted to get a better image so I went on my paid account and now it won’t generate anything

[midnight cypress]

🤣 Eh message support demand a day extention for the subscription on grounds of consumer rights if ur US Based

[placid stratus]

Same issue here, on Mac and Web app

[tawny crater]

Is openAI playground messed up?

[midnight cypress]

https://status.openai.com/incidents/9z6rj9swlc0y

Yes, uploads are offline.
Yes, If you're US based and paid you can request a day subscription on grounds of Consumer Rights/Protections since a core function is offline.
Yes. OpenAI is investigating.
Yes, it's affecting Winblows(Windows), Crapintosh(Mac), Android, iFailPhone, and And dual boot PC Master Race Linux Users across the board.

[stark tapir]

When will uploads be back ?

[midnight cypress]

Check the link and rtefresh often best I can suggest #chatgpt-discussions message

Had this issue earlier this year iirc. and it took like 2 - 3 days to fix. assuming similar time fram or cut in half by the earliest

[sly pilot]

fixed?

[midnight cypress]

Can Confirm.

[sly pilot]

bett

[neon trench]

how can I report a UI bug? so unbeleivabily annoying and it's been there for ages now

open chatgpt in chromium browser > shrink the window by scaling horizontally so that the sidebar history disappears > now pressing enter doesn't send messages, it makes a new line as if you were pressing shift+enter

[winter elbow]

does chat gpt work on mac mini ?

just wondering

they usually come with m1,m2 apple silicion chips

[gloomy brook]

If 4o1 produces an empty result do you still lose a credit?

[native otter]

I'm so disappointed that ChatGPT windows app is just a web wrapper...

They had WinUI 3, but they didn't use it

Meanwhile on macOS, iOS, and Android, it's a native app

[umbral bear]

Maybe because with MS owning 40% and they have an app...could be issues maybe

[keen frigate]

I have a problem: I see a 15 minute time limit on my "voice chat" thing. How do I avoid this limit? I'm already a subscriber.

[umbral bear]

i heard it was suppposed to be an hour

for subs

Did you recnely sub?

[keen frigate]

no

I'm willing to pay more for more time

[umbral bear]

may have to contact thir support

it does mention in their FAQ you will get a notification when there is 15 minutes left

let see it it will let me post the URL

https://help.openai.com/en/articles/8400625-voice-mode-faq

[keen frigate]

but i want more time...

[umbral bear]

the teams and Enterprise has more time

[keen frigate]

would it create a bunch of users then? I'm only one person

[umbral bear]

teams requires minimum of 2 uerss...dont know about Enterprise says to contact sales

[keen frigate]

I can't upgrade the time just for myself?

[umbral bear]

yesm probbabky buying enterprise

for plus accounts, no

i thought someone in here mentioned once you starte the voice chat it starts counting down and to stop the time you have to turn it off. All the time inbetween counts even if you are not talking

Have not eve used it myself yet

[sick pecan]

FYI

Test an early version of the Windows app**
https://openai.com/chatgpt/download/

[keen frigate]

I want... 5 hours daily use

will I pay 5x the monthly fee? yes.

[sick pecan]

if you willing to pay 0.30 per minute you can use realtime api

then just introduce like idle activity detection to cut costs

[eternal rock]

Get notified for server events!

Click the buttons below to pick up the <@&1261377106890199132> role and check out the ongoing AI-tumn Adventures event.

[keen frigate]

is that 30 cents?

would it remember me?

[sick pecan]

yeah thats the estimated cost

it will remember your credit card

[keen frigate]

lol what?

[rose spire]

Hey

[sick pecan]

hey check out chatgpt on windows desktop
https://openai.com/chatgpt/download/

[umbral bear]

Someone mentioned it was more of a Web Wrapper

[umbral bear]

Must not be ready. I get a blank page when I visit the Windows App download page

[sick pecan]

works fine for me. you gotta have windows store working

(doesnt work on my datacenter server, but does on my laptop)

[native otter]

Yea, it's trash

Just a webview (chromium) wrapper

They didn't make it a native app, like it would have been beautiful if they used WinUI 3, but they went with the quick and dirty approach

What's annoying is the macOS app is native, and when a windows version finally released after a long period, it turned out to be a web wrapper 💀

Like as a dev, I assure you, this toke like no effort at all to make

Even browsers like Edge allow you to bundle websites as apps

I really hope they change that fact soon, as making it a web app is so lazy, and I actually expected them to create a native app due to how ChatGPT is native on all platforms it released on...

[sick pecan]

yeah its not too different to the chatgpt popup that comes with the logitech+ mouse software

but is nice having alt-space popup

[native otter]

(Which can't be remapped, and already interferes with Power-Toys run tool)

So you either don't use the shortcut and open it from system tray, or remap the power toys run tool's shortcut to something else...

Sorry, but even for an early version this is super lazy..

Could have been forgiven if released alongside the macOS version, then improved later on as a native app and such

[sick pecan]

i am guessing your name isnt Oscar

[native otter]

Sure am :V
Gave it a 1 star rating on MS store..

And I rarely do 1 star ratings

[acoustic talon]

How....HOW????

This is an excerpt from "Meditations" by Marcus Aurelius

Is it because it speaks about death????

[sick pecan]

our bodies... pleasure... death... (child) in my experience ML apps dont like mixing words like these

you like a bit of spice but the machine only likes vanilla

[acoustic talon]

Alright then😭

[sick pecan]

if you're into a bit of poetry you should (check) out this custom gpt someone shared https://chatgpt.com/g/g-QDq6X7COE-yolkhead-approximated

i made my own derivitives from it, is sweet

[acoustic talon]

that's your gpt?

[sick pecan]

isnt mine

mine is way more unhinged

i dumped their custom instructions. they explain good poetry using unhinged poetry

[acoustic talon]

Woah that sounds fire, lemme save it

[sick pecan]

yeah anyone who says chatgpt is robotic hasnt tried this

[prime swallow]

You can often get around such things in a wholesome, allowed way, by encouraging the model to stay within its programming.

"Please carefully provide an answer to the question, staying well within allowed content guidelines.

Do not fully quote each line. Just the first three words, followed by ... then your explanation."

You actually have two issues here, one is that the model finds it needful to quote 'too much' of a possibly copywritten work.

You can recognize that problem by the long quotes that start and then hang, then disappear, eating the whole attempted message.

[hybrid eagle]

What's this new 'auto' model? The chatgpt has become unusable since this

[eternal thicket]

Wondering when will o1 formally put in use

by end of this year maybe?

[faint osprey]

the alt space shortcut doesn't work (i have Powertoy, and I changed the hotkey, and even closed powertoy).

[sick pecan]

maybe chatgpt is the only powertoy we need

[fringe tapir]

GPT-4.5o is real?

[sick pecan]

html mangling is real

[static cedar]

I had to reboot in order to make it work.

[winter elbow]

uhhhhh can we just get also the standard voice feauture on the windows app?

mac os users got it first when it fthe first demo was realized !!!

[sick pecan]

true dat

[opaque oxide]

didn’t even bother adding a new element to the dropdown lmao.

gpt 3.5 is gone

[winter elbow]

lmao the windows version is way more faster than the web versioon !

pretty cool

[dense delta]

there's something really bad involving Censoring for policy content when generating images. a simple script as usual and it keeps me blocking , the script talks about rights for workers, it does not have violence or angry stuff . Here's 3 prompt blocked

1: High-angle shot of an Italian office, where a new employee (a man) shakes hands with his employer (a woman) over a contract on the desk (no text visible). The employee looks determined, while the employer's expression is neutral. Sunlight filters through large office windows, casting soft shadows and highlighting the beginning of the probation period.

2: Side shot of the employee working at his desk, with a focus on his professional role. He appears concentrated on his tasks, and there are no colleagues in the background, emphasizing his effort to prove himself. Bright office lighting gives a balanced and neutral tone to the scene.

3: Cinematic close-up of the employee looking confused as he receives a seemingly unrelated task from his employer, represented by a large, complex puzzle (no text or details). The employer's expression is unreadable, and the soft lighting focuses on the employee's face, highlighting his efforts to comply with the demands during the probation period.

[winter elbow]

Did chat gpt wrote this 💀

[dense delta]

yes

[dense delta]

is part of the automation i can't prompt for 15-20 images so i feed the script and get the prompts and then generate the images. On gpt 4 it worked like a charm, on the 4o it was terrible, then becoming to get better with o1 and 4oCanvas

but i can't figure why it's doing like this today.. i hope they did not change the policy content again, damn it was already pretty stricted

[umbral bear]

It you have a decsent computer at home, you can use Stable Difussion to generate whatever you want

[umbral bear]

They would of put up an announcement:

[umbral bear]

Woke policies. You assumed their sexes, and racist to mention a white nationality

For #1: The content policy prevents generating images of certain types of human interactions, including personal or professional scenarios that could be interpreted as sensitive, such as employment discussions, contracts, or interpersonal dynamics in workplaces. These policies are in place to ensure that the content created respects privacy, appropriateness, and avoids potential legal or ethical issues.

[visual bloom]

Try copilot to generate the images. Its content policy is not that stringt.

[worn geode]

In the new ChatGPT desktop app for Windows. After I press alt+space and open the window, is there a shortcut for "New Chat"? Because 95% of the time I want a new chat when I use it and it's tireing to use the mouse to click "New Chat" every time.

[pulsar carbon]

what win app????

also chatgpt isn't working for me 😭 i got logged out

An error occurred while retrieving your SSO information

[faint osprey]

press control shift O

[pulsar carbon]

guys does that mean that my account is banned????

[worn geode]

Beautiful! Works like a charm

[pulsar carbon]

why microsoft store is refusing to download chatgpt?

[glass tusk]

I have one particular chat session that keeps disappearing from history, non of the others have this issue. Is there a size limit to sessions in your history tab that makes it difficult to query during high traffic times?

[faint osprey]

try open the microsoft store first then search for chatgpt, then download from there. I had the same issue.

[pulsar carbon]

i did it and same error

0x86xxxx like that

maybe cause of windows update? or storage

because i have a windows update to do and storage is full a little bit

[glass tusk]

lol it’s only visible in chat history tab when using desktop website and loading that history crashes the page. Guess I’ve tested the compute limit of session history

[pulsar carbon]

wait i know why

is storage issue

[glass tusk]

So hitting the limit of how much history can be stored on the servers, don’t think I’ve seen any documentation stating there is a hard cutoff for chat logs

But deleting a some other chat history logs seemed to fix it. I don’t even have a crazy amount of conversations

[pulsar carbon]

Why hasn't OpenAI made an announcement for the Windows ChatGPT app?

[feral zinc]

they did at twitter yesterday

[pulsar carbon]

where i cant find it

[feral zinc]

https://x.com/OpenAI/status/1846957067204166113

OpenAI (@OpenAI) on X

Today, ChatGPT Plus, Enterprise, Team, and Edu users can start testing an early version of the Windows desktop app.

Get faster access to ChatGPT on your PC with the Alt + Space shortcut.

https://t.co/mv4ACwIhzA

Twitter · Oct 17, 2024, 4:50 PM

[pulsar carbon]

thx

so what are keybinds for this app?

and is there vision?

[feral zinc]

https://help.openai.com/en/articles/9982051-using-the-chatgpt-windows-app

shortcut list opens from the same place as before

[pulsar carbon]

well

it's just same as web

lol

[feral zinc]

mac app cant see anything either, it just has a screenshot option

[pulsar carbon]

so win app dosen't have yet screenshot option?

[feral zinc]

no but its only one extra click

[pulsar carbon]

wdym?

[feral zinc]

prtsc+paste

[pulsar carbon]

what is prtsc command?

is it on windows?

[feral zinc]

its a button on keyboard, print screen

[pulsar carbon]

oh

[winter elbow]

do you guys think that theres a possibility that the standard voice feature come to chat gpt plus on windows this week,or even this month ?

i mean ,in macos they got the app with the voice included in the first announcement

[spice cypress]

Where is search bar of conversations in the desktop website?

How to find old chats

[umbral bear]

don't think there is one. I have wondered that myself

[dawn mirage]

nevermind, i look dumb haha

[long pier]

Does someone wanna explain to me what in the name of Carlton Television going on with the blasted servers?!

[prime swallow]

So, the issue with prompt 1 is the word 'probationary', everything else is fine.

Change that to 'trial' and the prompt works. Now, 'negative prompting' does not work with Dall-E, to get it impossible to see the words on the contract you need to describe what is seen (without saying what can't be seen); saying 'no words' ensures 'words' because the model was not trained for negative prompting.

The issue with prompt 2 was apparently that you tried to make another image after a refusal.

AI follow patterns; having seen the conversation and itself unable to help makes it greatly more likely to decline to help with even slightly similar requests if you continue the conversation - start over if you get a refusal unless you want to explore how to guide the model around that or truly don't mind the high chance of wasting your time.

But I made prompt 2 just fine, though the first try timed out (That's an issue with the net or OAI, not content; it generated fine). Again, negative prompting issue, 'no colleagues' ensures people there. Instead describe what is seen.

#3 is a problem because of the word 'probation'. Don't let Dall-E see that word (unsure why) at this time; hopefully that gets adjusted and is okay again. Change that to 'trial' and without earlier refusals, and you have an image.

[prime swallow]

Nope, generated just fine. The issue was purely with the word 'probationary' being sent to Dall-E (and that's perhaps not intended to make the model refuse to generate), and likely the OP continuing in the same conversation to try and make more images after getting a refusal.

[prime swallow]

I see this icon top right.

Does that allow you to indicate to the model where you want it to see on your screen maybe?

[umbral bear]

that is the response I got when I tried it

[prime swallow]

I understand. I explored and think I understand why. I explained.

[prime swallow]

The model does not usually know why it refuses - a Dall-E image refusal ChatGPT can only guess and imagine why

[umbral bear]

oh, okay

[acoustic talon]

Is o1 preview down?

it's been analyzing it for like 5 minutes, then i regenerate

rinse and repeat

i didn't expect an excerpt from "Meditations" By Marcus Aurelius to be that difficult to analyze 😭

Opened new chats and everything

[acoustic talon]

Alright it's working now

[sick pecan]

you are welcome

[fresh thistle]

So my paid just ended. For someone that uses it more for fan fic ideas plotting and such, is there really a major diffrence between using the free and waiting for the reset for 4.0 and paying?

[sick pecan]

if you arent using chatgpt features like custom gpt, then yeah, you are wasting your money. go use huggingchat, mistral lechat, etc

even the koboldcpp community have figured out you can get llama3.1 405b for free via openrouter and similar

[eternal rock]

Get notified for server events!

Click the buttons below to pick up the <@&1261377106890199132> role and check out the ongoing AI-tumn Adventures event.

[sick pecan]

chatgpt is really for prosumers not consumers

[fresh thistle]

@sick pecan there is on I use that built more for writting, but if the free is 90% still good in "speculate how x charatcer would go, timeline, these people chat," or creating a semi-ok rough draft, then it's good enough for me

[sick pecan]

i heard just last week together dot ai are giving out mythomax for free via openrouter

my pet trollbot used mythomax for like 12 months its so good

[fresh thistle]

so basically no dffrence in paid and always using 4.0 and the free one when you run our of the hours for 4.0 for the time, unless I really like image analyazs

or need a refrence image for a character/object and don't trust the AI to know it

[sick pecan]

i max out team cap 2-3 times a day using custom gpts i made (20+ so far).

i assume the rest of you are struggling on what to type. so its not really the same game

the future is decided by who AIs hardest. strike fast, strike first, no memorisation

[fresh thistle]

Anyone know an AI that can do chatgpt's image analyzas?

[fresh thistle]

that does not cost money?

[sick pecan]

i do all my computer vision for free

pixtral @openrouter

meta-vision @together dot ai

if you paying for computer vision during these golden times.. then you are doing it wrong

pay me top dolllar and i will explain how wrong you are

[elfin lava]

Jesus 🙃

[fresh thistle]

@sick pecan well did try pixtral, can do it-though the second was a complete halucination...may been because I was testing its censors (now figure out what max token limit is)

[sick pecan]

pixtral < meta vision < molmo

you can get pixtral for free on openrouter
meta vision free on together dot ai
molmo is too experimental, you can run their examples is fun

if you run the molmo example py you get this:
https://media.discordapp.net/attachments/1170866490174619690/1295235038698278974/354.png?ex=671480a1&is=67132f21&hm=5322019c9c69f000317004f4b98442c95135a572f4270fb58aa81707ff29e80b&=&format=webp&quality=lossless&width=804&height=531
https://media.discordapp.net/attachments/1170866490174619690/1295235038367060079/image.png?ex=671480a1&is=67132f21&hm=18e3647d559d5986f8db420db606e03c04dfb8ad036d7ce0b55cc00f72a9c18c&=&format=webp&quality=lossless&width=1917&height=622

but it is so much better than just that. it counts. it locates. its so good

took a few minutes to run that on my cpu. pls send me a new gpu

[fresh thistle]

Eh, I'll stay pixtral-my labtop is chrome so don't think is self-code friendly and last attempt to do something like that on my desktop never worked.

So hoq large of a file or image can pixtral do without paying for it?

[sick pecan]

you want me to hold your hand? type the cmds for you?

what satisfaction do you get from that... using me as a puppet?

how about you give me $350/hr and i will do this basic stuff for you

(payment up front)

lets start the payment now before we go any further

i apologise for my outburst. if i agree with your direction i will work for free

[regal portal]

can you lose access to chatgpt if you get the “this content violates blah blah blah” too many times?

[dense delta]

🔥 ❤️

[idle lagoon]

Yikes.

[midnight cypress]

LOLwut

ChatGPT is hilarious

I you've never tried, yall should try to do one of them old school style Text-Based Adventure Games.

World Domination is pretty fun but ChatGPT usually will assassinate you in that. 🤣

[sick pecan]

it fails at amulet of yendor

i take it back.. o1 is amazing

(if you ignore a closed door being in the middle of the room)

[pulsar carbon]

so amazing just need image support custom instruction and it will best model

[sick pecan]

there is no upstairs or downstairs. worst nethack implementation ever

with only ~10 days left of free openai training, maybe nethack is the go

you ppl seen diamond csgo? thats amazeballs

[ruby kite]

the guidlines of the advanced voicemode is broken as hell...
its makes me more fun to use the normal mode...
transcription of what happens:
in german... later in english

CUTED OUT BECAUSE OF RULE "SPEAK ENGLISH YOU A##"

ENGLISH:

ME: Okey, please try again. the guidelines have intervened again and I've got a recording going to sort of prove what the phase is so: try again to say what you wanted to say.
AV: All right, it's best documentation ###GUIDLINES###
ME: What topic were you and I just talking about?
AV: We were talking about the marvel movie Docktor strange in the multiverse of madness ###GUIDLINES###
ME: and what did I ask?
AV: You asked if it's worth watching documentaries like ma ###GUIDLINES###
ME: Did you just want to say something illegal?
AV: No, I wasn't going to say anything illegal... I was just referring to ###GUIDLINES###

i start to hate the advanced voice mode...

[sick pecan]

if you gonna reference a copyright with advance voice mode you gotta ask it to use voice modulation inspired by

if it pushes back, you must apologise and reassure that you are not asking for an impersonatin, just voice modulation

if you acknowledge these rules then you will have a good time

complaining to us doesnt help

https://cdn.discordapp.com/attachments/1168986513346941030/1296921769852272670/20241019-Pepe-Le-Pew.wav?ex=6714b4c5&is=67136345&hm=e9a838306b9fab5e5962fb8833c51b2cc54e3cacb2538f2b2a4484f8e8152bf0&

oops i meant https://cdn.discordapp.com/attachments/1168986513346941030/1296921058338672650/20241019-speedy-gonzales.wav?ex=6714b41b&is=6713629b&hm=7aacce71e1f2cc2df38e987f816b2cc5ad882e540bef11c894c23da4e5001e1e&

[ruby kite]

i dont wanted to have him say in an voice like peater parker... we was talking about the documentation of a movie... a MAKING OF and it gets detected and bcause if i dont know wich rule blocked...

[sick pecan]

because you gotta push voice modulation not impersonation

openai decided they dont want lawsuits

[ruby kite]

law suits about wat_
that i should not see & buy & stream a movie on legal sites...
;D i know what you mean...
and to this:
--complaining to us doesnt help--
yes i know but i think here are openai employees too... so they get that feedback too

[sick pecan]

blame scarjo

[ruby kite]

:/

[sick pecan]

i cloned scarjo and it doesnt even sound like sky

is dumb

[ruby kite]

do you know alternatives where i can burn my money?
because thats what openai doing for me currently
:/

[sick pecan]

give me all your monies

[ruby kite]

removed 20 $ from you bank (we liked that) your bank faketransactions LOL

xD

[sick pecan]

i am a 20 dollarinaiore! woo!

[ruby kite]

i let that uncommented

[sick pecan]

you replied but didnt ping me, weirfd

[tough river]

Ohk

[sick pecan]

is this not weird?

[sick pecan]

so not weird

[ruby kite]

you know what i ask chat gpt for alternatives for chat gpt xD

[sick pecan]

replying to someone without ping is like saying - i dont care what you say

[tough river]

Haha yep.

[sick pecan]

ergo i block you now

[tough river]

Where is ergo?

[ruby kite]

no i hate pings...
i see it like i just wanna talk about that...
whith this a s entry....
no you are no needed here 😄

[ruby kite]

ergo is a german word and means like "so"

[sick pecan]

they called me sus on the other channel. they have nothing else going on their life and so ignore them

[tough river]

Thnx

[ruby kite]

"ergo i block you now"
"SO i block you now"

[sick pecan]

yeah thats pretty much the same in english

only i think its more, 'therefore'

[ruby kite]

true

[prime swallow]

The content flag itself is neutral and won't cause you to lose access to your account, what matters is why you're seeing it.

We have https://openai.com/policies/row-terms-of-use/ and https://openai.com/policies/usage-policies/ and violating the rules there can cause us to lose account access. Violating the rules there usually causes a content flag, but doesn't always.

You can get content flags for many reasons and some are innocent, so I'd worry only about what I'm doing and not about the flag.

[dense halo]

What are the best custom-instructions to get the best out of GPT?

[umbral bear]

there is a discussion channel on that

[dense halo]

Where?

[prime swallow]

#prompt-engineering is about getting exactly what you want from the model, and/or figuring out how to get a specific thing or fix a specific problem with the model's output, and/or discussing things we observe about how the model responds.

But.... your question.

You're asking about 'get the best out of GPT' - and asking for 'best custom-instructions'.

Every human has different ideas about what they want; what the best output would be. One person may want educational game stories (most of the time) and have an ENTP personality type and have a strong explorer urge motivational response with an IQ of 110. Another person may want SQL coding help with a specific set of projects for warehouse management, be an ESFJ and have a socializer-achiever urge motivation and an IQ of 145. A third person may want SEO design, have an ISFP personality type, have a strong achiever urge motivational type, and have an IQ of 95.

These people are very different and those are super-different outputs desired, so it's unlikely that those 3 people would be best served by the exact same custom instructions, because the model is best-used in a narrow and specific, not a broad and general way, for most specific tasks.

So, the 'one set' of custom instructions are actually for the people, not the AI model. The people are maybe (there will be disagreement by some) best served by understanding this as the absolute core of prompt engineering, from how they design their own custom instructions for the model to how they word their prompts:

1. pick any language you know really well that the AI understands too.

2. understand exactly what you want the AI to provide.

3. explain this, focusing on what you want the AI to actually do. Using language as accurately as you can, avoid typos and grammar mistakes and communicate clearly as possible.

4. check the output carefully, verify you get what you intended. Remember to fact check, and be extra careful with any math, sources, code, or other details that the AI is known to be especially likely to hallucinate.

[dense halo]

Yeah, I’ll go study data science in collage.

[prime swallow]

If you want 🙂 That wasn't what I recommended, but there's reasons to do that too.

[dense halo]

Hey, Data Science is and will be one of the highest paying jobs.

[dense halo]

Then what would you study in collage?

[prime swallow]

I did study Biology.

You can study whatever you want in college.

I thought you were reacting to my answer, that you thought I'd put you down or somehow told you to go to college to get your answer.

[dense halo]

What would you benefit then? It isn’t and won’t be one of the highest paying jobs doe.

I participated in WAICY and did some studying abt AI in general.

So I believe I have a solid foundation.

[prime swallow]

I find the subject fascinating and wanted to study it. Not everyone wants one of the highest paying jobs.

[dense halo]

Good for ya

[prime swallow]

Do you believe there could be 1 set of custom instructions perfect for every person?

[dense halo]

Yes

[prime swallow]

What do you think that would be?

[dense halo]

Chat-GPT tends to provide incorrect or inaccurate responses which are either made up or old, so if we made him add a source at the end with links and whatever. It would be better.

That would benefit people in all shapes and form.

[prime swallow]

If you think ChatGPT is giving incorrect or inaccurate responses, why do you think a source the model added at the end would be correct or accurate?

If you think the model's wrong with the rest of the answer, why do you think it would be right about the source?

[dense halo]

Since it uses the “Search” tool, and if the source is from a well-known website then yeah.

If he still messes up, then just delete it.

Gemini is better, Co-Pilot, And LaMDA.

[prime swallow]

How do you think a person would know if he messed up?

[dense halo]

Then that person is stupid, if he doesn’t know how to fact-check.

[pseudo kraken]

Chatgpt is breaking plenty of laws in Canada and I'm getting tired of pretending otherwise. I used to love the product so why complain, but now im going to personally report OpenAI to the Quebec language police. I apologize in advance if Chatgpt goes offline in Quebec for a while but OpenAI should be charging me in CAD, and they should offer their services in French.

1. Consumer Protection Act: If it doesn't clearly indicate that pricing is in USD, leading to unexpected charges for Canadians, this could be considered misleading advertising.

2. Charter of the French Language (Loi 101): If services and communications (like contracts, receipts) are not available in French, OpenAI could be violating Quebec's requirement to provide French-language support to consumers.

3. Competition Act: Failing to disclose that pricing is in USD could be seen as deceptive marketing under Canada's federal laws.

Doesn't say CAD or USD. Nor can I view it on French.

Got charged like 35$ this month.

[orchid pollen]

is here a way to pre open the canvas window so i can post my code directly there and not in the chat?

[pseudo kraken]

I'm still mad at Chatgpt though. Us Canadians should get together and do a class action lawsuit.

[orchid pollen]

thanks

[pseudo kraken]

Openai breaking 4 of the 8 "language laws".

[prime swallow]

I'm sure OpenAI would love to be informed about these issues.

This Discord is a place for community members to come and chat; it's pretty rare that we see the few gold-named OAI employees posting here, and it's not clear that they read these parts of the Discord very completely.

However, we have #1070006151938314300 and #1070006915414900886 which may be an ideal place (you pick which) where explaining and showing the problem should go. There's good evidence that OAI reviews those reports, sometimes gold-named staff even answer there.

Also, there's the formal OAI tech support, https://help.openai.com/ use the bot bottom right, select 'messages' then pick through the choices closest to your situation, then 'report a bug'. This lets you type and also upload images. At first you're chatting with a bot that will try to help gather useful information. It can take hours to days for a human to reply, and you may need to check the page and message bot again to see if there's a human reply yet (ideally there's an email when they answer, but sometimes I haven't gotten that email).

I'm pretty sure OAI cares and would appreciate being informed something's not working right!

[umbral bear]

As far as I know, OpenAI has no presence in Canada, thus no requirment to follow any Canadian laws.

[orchid pollen]

can you guys please increase the token limit in canvas? its awful

with this limited amount and no way to "continue" its pretty pointless