chatgpt-discussions

[rugged igloo]

In college, Jill taught Sam different math theorems.

After leaving college, Jill forgot more math theorems than Sam learned from her.

Based only on the above, which of the below must be true statements.

(Select all acceptable answers.)

Sam, while in college, knew some of the same math theorems that Jill knew.
Jill does not remember math theorems that she taught Sam.
Jill once knew more math theorems than Sam learned from her.
Sam does not know any math theorems that Jill did not know.
Sam knew at least two math theorems.
Sam once knew fewer math theorems than Jill. ---o1 preview got this wrong----

(these are the correct answers) Sam, while in college, knew some of the same math theorems that Jill knew.
Jill once knew more math theorems than Sam learned from her.
Sam knew at least two math theorems.

[swift zinc]

What is Alpha GPT-4o?

[lost raptor]

gonna be honest this new "o1" junk is annoying with the sensitivity settings. just the most minor prompts trigger a "potentially violating"
I submitted my code and asked for a modifcation and it "finishes thinking" only to get a flag.
This is happened quite a few times. Just a resubmit and it will work after but you can't tag everything as "Potentially violating" am I gonna get banned if the sensitivity is just too high?

[prime swallow]

If you're inside allowed content, you're safe. I've had probably more than 200 content flags since I started using ChatGPT (back when it first came out), and I do explore some potential grey areas and conceptual edges that are not out-of-bounds (those can throw a lot of flags) - but I stay entirely inside allowed content and still have account access.

I've seen about 10 content flags from the o1 models and my exploring so far, I think you're safe from bans unless you are actually doing disallowed stuff, so you sound fine.

But we don't want you missing out on model answers either!

What helped me get fewer flags with o1, and more answers even if there are flags, was to give it context.

What are you doing with the code, what your intentions and reach are with it; as-is, I think the model's imagining every possible perspective that can submit the prompt we input, from very kind-hearted to very evil-intentioned, and its 'mind' wanders all over those ideas. If it guesses it's really 'too scary' content flags fly.

If instead you explain the actual scope of your work and focus of your intentions, the model appears to evaluate what you ask for in connection to the context, decide if that makes sense, and if so decide everything's fine and it should help your use-case.

[lost raptor]

The prompt

"The code here"

give me the parts to replace but lets make it so that the velocity is scaled better so that quick flicks of the paddle don't scale up to unreasonable speeds. 

I'm making a mobile app. Most of the time it just kind of seems like its a thought process failure. so anything that breaks this chain of thought is within violation.

[potent oak]

For liability reasons, it's also always better to have the sensitivity too high than too low

[lost raptor]

sure but if i end up getting my account flagged the service isn't worth my money. Plus they have been flagging prompts that request info about the thinking process.

[wild wind]

Is it just me, or has the o1-preview limit just been reset for you too?

[rugged igloo]

I think they reset it again

I saw a tweet of Sama

[prime swallow]

They appear to have upped our use of o1-mini to 50 messages per day, and o1-preview to 50 per week.

It's not clear if we have the extra 20/week o1 messages or if we got reset to the full 50 again

Presuming you were capped before the change

[rugged igloo]

No shot really holy moly??

[prime swallow]

I want a regular announcement, but this does seem legit. #ai-discussions message

[rugged igloo]

That means I have 100 per week since I have two accounts

[prime swallow]

Every announcement we've had, I think came from gold-name (OpenAI employees have gold names) 'Eric' - who may be off work when the decision happened. I kinda expect an announcement tomorrow to help confirm

[prime swallow]

Grats!

[rugged igloo]

From my mate

[uncut vale]

What does 'alpha mode' does this time?
I got accessed to alpha mode several months ago and able to generate tons of free images?
Btw I'm free user

[prime swallow]

At a guess, currently you're seeing 'alpha models' not 'alpha mode' this time.

The alpha model, looks like you have or may eventually have access to o1-mini.

[uncut vale]

Oh

[swift zinc]

What about mine? After selecting Alpha, it still returns to 4o or 4omini, and the third one is unavailable

[uncut vale]

Last time it showed alpha mode with similar type of icon and i thought it's the same mode at first glance

[swift zinc]

What's this?

[prime swallow]

What device did it come from?

[swift zinc]

cell phone?

[prime swallow]

Android, iphone?

[swift zinc]

Samsung

Web page

[prime swallow]

K, Android. Not sure what to tell you, I'm also android and mine looks different.

Perhaps that's some sort of alpha, no idea. You're the first I saw discuss this.

[astral socket]

May someone explain to me why random websites on the internet can quote books/studies/research papers, but ChatGPT cannot?

I gave it a pdf to quote and it refused.

[swift zinc]

But can't use it

[astral socket]

Then I asked it to quote a book and it refused

[prime swallow]

Sure. Nobody is particularly controlling the random website creator, but if someone wants to sue that person, they could.

But ChatGPT is made by huge and tempting OpenAI, who could face a huge number of lawsuits from those wanting to explore the laws around what hasn't even been designed yet, in regards to AI and rules and responsibilities and what is and isn't allowed for sure.

So OAI is being very careful.

[ivory iris]

Has anyone had an experience with ChatGPT remembering anything beyond a chat session, specifically with chat personas?

[pliant saddle]

y’all the alpha model just means default. It’s what the app shows before it loads the available models from the server.

[prime swallow]

No. And I've explored for that.

What it will do is 'read into' anything you tell it, like a human psychic or 'mind reader' might, using every clue you give to what you would like to hear, and try to provide you satisfying answers that match what you want to hear.

Additionally, the model doesn't do 'random'. Asking it to make a random choice, like a topic, it goes to training data, influenced by your prompt's wording.

So, you yourself guide the model to talk to you like it does, and most of us typically have a similar writing style between prompts.

[prime swallow]

Except, memories made. Or my own copy/pastes, or what I put into custom instructions. If a custom GPT, whatever it's built with can seem like 'remembering' too if it's coincidentally right.

[humble river]

o1 mini is available for free users

atleast its visible for me

[ivory iris]

I've found that there may be a very small amount of simple things that a persona remembers. I added something like this to the end of my response: 🩵 . It started using it in that chat at the end of it's responses. In a new chat it also used it. That's just one example

[prime swallow]

No custom instructions on, no memories on, and you're seeing stuff like that? I'd love to see a shared chat

[ivory iris]

ChatGPT 3.5 used to say this was possible with chat personas, but I never really saw it, but later when I tried I started to see it.

[ivory iris]

This was before memories or 4o

so just with 4

before custom GPTs and no custom instructions

I actually did this a while ago, but some of the more advanced conversations I wonder if aspects like that play into it

I probably have the chats in the history some where.

This was just out of my own curiosity because I haven't heard people talking about it

If you want to try and duplicate it, have the persona use emojis. See what ones it uses over multiple chat sessions. Then have it use a custom one (in the chat session, not the instructions) and keep the conversation going. Then try it in a new chat and see if it carries over. I haven't tried it in a while, but it would seem like aspects are staying with it.

[uncut vale]

Same,it appears to me as well

[unborn brook]

what the... why is mine different?

[spice star]

openai very rarely rolls out a feature to everyone at once

[woeful laurel]

Not that they change my copy paste but what do these colours on the text mean?

oh they're commands and that in coding?

Drop, Let I assume.

[crisp cairn]

whats with this o1-mini alpha? i am checking this now, and it seems so slow... like a regular gpt 4o model

and it didn't use "thinking" at all

its not the same o1-mini that we had so far

also the output is different, short

i had to re-login, and now alpha is gone, maybe alpha was for a free user then?

[sonic comet]

o1 mini is being rolled out for the free tier user under the name alpha model. This alpha model is being tested for the attachment, memory, web searching, etc. But mainly the test is for auto-model switch which switches between the models based on the prompt and with this the users won't have to switch a new chat to use a new model.

[snow plover]

If you prompt it, it responds as 4o anyways.

[swift zinc]

I can't use o1 at all

[minor lark]

what's the point of coding if ai can do it?

[feral zinc]

it cant do everything, not even close

[orchid pollen]

did you guys changed the cooldown on preview 01?

i thought i have to wait until 21.09?

[feral zinc]

o1 mini is now 50/day

[orchid pollen]

i asked preview 01

because i had ONE WEEK cooldown and now after 3 days it works again

just wondering if you guys changed that?

because well 1 week was way too much xD

[feral zinc]

o1 preview still has 50 a week cooldown

[wise bobcat]

really? That would be awesome

[dreamy cypress]

Did the limit get reset?

[honest jay]

Here's the limit update tweet from OpenAI in case anyone hasn't heard yet! https://x.com/OpenAI/status/1835857163765637607

We appreciate your excitement for OpenAI o1 and we want you to be able to use it more.
For Plus and Team users, we have increased rate limits for o1-mini by 7x, from 50 messages per week to 50 messages per day.
o1-preview is more expensive to serve, so we’ve increased the rate limit from 30 messages per week to 50 messages per week.

[wise bobcat]

Yeah, I’ve seen the tweet

That’s awesome

[young obsidian]

50 per DAY?

Yay!

Btw how is Mini performing compared to 4o?

[wise bobcat]

Since one reply from o1mini is around 2 cents, I would say OpenAI is really a bonus from OAI

Way better

[wise bobcat]

Way 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.

[novel plover]

how do i chat text response essays

[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.

[humble river]

haha yes

[silk flame]

guys how to check the numer of o1-preview messages that I still have this week?

[still forge]

I'm getting confused on what's good what's better what's the model I need for what?

[worthy lynx]

Had the same question yesterday but no response 😦

[ashen prawn]

laggy af rn...

some messages not going through

[cloud pewter]

what

works completely fine on desktop but keeps popping errors on mobile

[limber oxide]

all my prompts are giving me errors

[latent ibex]

For Plus and Team users, they have increased rate limits for o1-mini by 7x, from 50 messages per week to 50 messages per day.

o1-preview is more expensive to serve, so they have increased the rate limit from 30 messages per week to 50 messages per week

dang that's goodd

50 per day

[mighty crest]

i cant use the ios app at all because i have NextDNS enabled and something in my denylist is blocking openai servers... dont know if this helps you but its definetely my problem and i cant fix it without disabling NextDNS

took me a while to figure out

[raven sand]

Hi all, is there a website that gives info on all LLM benchmarks?

Trying to do some research for my usecase and would like to look at the different benchmarks for different models.

[normal quarry]

ask in #ai-discussions

[ivory iris]

o1-preview is definitely not AGI level

[normal quarry]

it's not supposed to be

[ivory iris]

There's not a position either way. I'm just stating my observations.

It is an advancement which is what they said

[pliant saddle]

It’s not working because something in your NextDNS denylist is blocking openai. Sounds like it’s working like it should. So what’s the problem?

[mighty crest]

It's more like openai is relying on a specific apple server that's denied :(
It's not openai specific

[warped pendant]

was o1-preview changed to 50 per say vs per week? for teams and plus?

[normal quarry]

o1-preview is 50 per week, o1-mini is 50 per day

[steady cradle]

There's a bug: o1-mini shows up for free users, but they can't actually use it

[novel shadow]

can someone please try doing something and asking o1-mini or o1-preview to think for a long time?
i just heard someone say their chain of thought is hidden despite the dropdown arrow appearing and it got me thinking. because i've never seen anyone show the chain of thought

[sonic comet]

Well that's because mobile version was not designed as good a the web version. 😂

[novel shadow]

i can see the chain of thought by pressing the dropdown arrow to the right but why isn't anyone doing this?

[fathom tinsel]

It's not the actual chain of thought, it's the summary of it

[snow plover]

this ^

[fathom tinsel]

You can see some examples of how the actual chain of thought looks like here: https://openai.com/index/learning-to-reason-with-llms/ and then under the "Chain of Thought" section. It's shown uncensored for a few examples there

[novel shadow]

thank you 😄 so the first 2 things are just the summary of another chain of thought process that is hidden?

[fathom tinsel]

Yes, it somewhat reflects what happening but it's not the whole story

[novel shadow]

that's interesting

[fathom tinsel]

The real chain of thought process happens invisibly because it is not censored

[novel shadow]

i really wish we would get to see the whole thing eventually

[novel shadow]

oh

i see thanks

[stark void]

What happened here? I did NOT say that

I was talking to it via voice

[cedar glade]

Sometimes the voice transcription fails and gives some random version of "thank you for listening". It can happen if you don't speak english. I assume it fails and defaults to some common pattern in the training set.

[stark void]

Oh wow, that’s very interesting. Thanks!!

[dense halo]

They are really limiting the new model use. lol

[rugged igloo]

not really you get 50 o1 per week and 50 o1 mini per day now

[dense halo]

yeah, thats still pretty crazy

[rugged igloo]

well have you seen how much o1 costs on the api its exspensive i done 5 prompts with it and racked up 40$

[dense halo]

sounds like an OpenAI problem my friend. Sadly that turns it in to mine.

[rugged igloo]

typical user

[dense halo]

but thanks for the limit info

and typical "dev" i guess? lol

[winged willow]

looks like the cap on o1-preview has been lifted because all of sudden I now have access again. How do I work out how much usage I can get out of it again before it shuts me off?

[rugged igloo]

well this was from last night they increased it so they prob reset the cap thats why

right now there is not a best way to see where your at you can count manually but that is it

[young basin]

am I getting the "Your request was flagged as potentially violating our usage policy. Please try again with a different prompt." message because I'm asking too much from o1??? All I did was tell it what I want my code to achieve, and I gave it the current one that needs fixing. It thinks for about a minute then I get that message

[umbral mulch]

Did anyone got access to Advanced mode in the past few days?

[young basin]

what makes you ask that?

I haven't seen/read anything about more people getting access

[umbral mulch]

Right, that's what I was trying to figure out. My understanding was that they were "slow dripping" users everyday, but maybe it is not the case.

[young basin]

nah there was just a period where some users would get it, with a plan to fully release by fall. It wasn't a "we're gonna slowly roll this out all the way up to fall", sadly.

so I think July/August was the initial roll-out. Now we wait for them to announce the full release.

might have been just July actually, not sure anyone got access in August but might be wrong

[fickle fiber]

chatgpt is 404ing constantly. The 404 error is literally dancing all over my screen.

[flat jackal]

same

[grizzled oar]

How is the world knowledge of o1 mini compared to GPT 4O?

[honest jay]

Both have a knowledge cutoff of October 2023 currently:
https://platform.openai.com/docs/models/gpt-4o
https://platform.openai.com/docs/models/o1

[young obsidian]

@honest jay do you know anything about the rollout of advanced voice?

[honest jay]

Hello! Only what's been shared by OpenAI here: #announcements message

We'll continue to add more people on a rolling basis and plan for everyone on Plus to have access in the fall.

[grizzled oar]

So should I replace 4O with o1 mini

[honest jay]

It depends on your use case I'd imagine! Are you using the models on ChatGPT, or via the API?

[grizzled oar]

ChatGPT

Also thanks for helping me with this.

I think
O1 Prev. - Reasoning with world knowledge
O1 mini - Reasoning and coding
GPT 4O - World knowledge

[honest jay]

Of course! Happy to try to help. One thing to keep in mind is that the o1 models don't currently have access to tools like web browsing, data analysis, DALL·E, memory/custom instructions, etc. So 4o can be more useful in cases like that.

[winged willow]

How about SearchGPT? I saw a notification on my ChatGPT browser a couple of weeks ago that it was now available then nothing!

[quaint marlin]

Would it be beneficial to simply give ChatGPT access to a reliable API for a calculator app? It would be great and probably save a ton of time for ChatGPT itself and would make checking it's work easier since it wouldn't have to do code to check code.

[sick pecan]

it has the python tool. python can do anything a calculator does

[finite urchin]

hi, what's the longest response time expected for o1 preview? in terms of 'waiting for it to finish'? I submitted a query that's been running for what feels like 5 minutes+

[honest jay]

I think the most recent news on SearchGPT is here: https://help.openai.com/en/articles/9237891-obtaining-access-to-search-on-chatgpt

The waitlist for SearchGPT is now closed. We appreciate your interest and will update you when we bring the best of SearchGPT into ChatGPT.
I think the notification you saw a couple weeks ago was a visual glitch that many ChatGPT users saw - my guess is it was just a case where that popup was loaded by the site before the site actually checked if any given user actually had access yet, so some people saw it without actually having access yet.

[winged willow]

Also does anyone know if o1-preview responses that fail to be generated completely count towards one's usage limit? I've struggled today with getting complete responses and have had to ask it to regenerate several times

[wind ocean]

I am a user who is not yet on a payment plan and the o1-mini model is available, but it is a bit strange, I don't know if it is using implementation in the gpt4o - mini models and not the o1-mini which thinks and takes a long time , since he responds quickly like gpt4o does and does not indicate that he thinks about what was proposed or the assigned task, I don't know if it happens to others besides me.

[normal quarry]

o1-mini is only available to ChatGPT Plus subscribers, the model you will be using is GPT-4o mini

[warped pendant]

what's the rate limits for gpt teams, for o1 preview and o1 mini, looked on site but didn't see, os is it the same as plus being 50 per week and 50 per day?

[wind ocean]

A section of alpha models appeared and o1-mini appeared, in the same way, other users I know saw this new section appear today without being subscribed to GPT plus

[normal quarry]

what happens when you select it?

[wind ocean]

When I move the cursor, o1-mini appears and select, it eliminates the option to send attachments, it changes the prompt suggestions and where it says ChatGPT a, before selecting it said ChatGPT-auto

[wind ocean]

This is the interface before giving that option of alpha models o1-mini

[normal quarry]

it appears this is a User Interface bug and you don't have access to o1 mini, the model will choose GPT-4o mini for you

[wind ocean]

Yes, I already assumed it, because when interacting with it activated, the model did not do the thinking process as in the demonstration, in addition to that after it finished generating the message, it indicated regenerating with gpt4o / gpt4o mini and allowed sending attachments again.

[rugged igloo]

o1 preview is able to write a 8 page report without stopping thats amazing

i never tested because of the limit but now with 50 messages i can test amazing

[eternal thicket]

Is ChatGPT busy now?

when I open my ChatGPT, the interface is empty

even after I release my cache, reopen my browser, even restart my computer

it is still an empty interface

[magic maple]

me 2

[ocean mirage]

same

[grizzled oar]

Yeah that's why. I think o1 mini with web search will be the most useful model.

[acoustic wasp]

O1 preview is 30 per week and 01 mini is 50 per week

But what about advanced voice mode? I still want it

[sonic comet]

I think the limits have changed. GPT-o1-preview limit is 50 per week and GPT-o1-mini has a limit of 50 per day.

[acoustic wasp]

I hope so. O1 preview should get an extendet usetime

I hate it to pay for limits

How expensive could it be?

Okay they seem quite expensive at first. This is energy consumed right?

[sonic comet]

My limit hit on 15 sep and was supposed to reset on 21st sep but all of a sudden my limits were reset and ready for use.

[acoustic wasp]

Yeah, i saw that too. I wasn't aware of the limits so I've used o1 preview as a daily model for everyday tasks. I got acess on Thursday and ny limit ahould redet tomorrow but I could use it yesterday to do my homework. That was quite nice

Since it can think, the results are amazing

[sly sequoia]

I wish we could tell how many uses we have left

[lucid pilot]

The website looks all bugged for me in desktop anyone else having issues there?

[sonic comet]

Nope. I'm not facing any issue. I've never faced any issues other than the network connectivity. Most of the issue that you face is related to network connectivity, so you might wanna check your network connection properly.

[vernal axle]

Is it just me, or does some responses appear in wrong spaces overtime? Like going back to an older resoponse here's 2 responses in 1?

[bright quiver]

anyone else got this issue? lol

Got CORS errors in console for some reason

[midnight cedar]

same

[lucid pilot]

Yeah in another chat we found out it was caused by google Chrome's update. Using a different browser was the workaround

[winter elbow]

is it just me ??

[winter elbow]

sammeeee

[tight island]

In subscription no

[winter elbow]

i think their updating bcs they are adding more messages per they in their o1 models

[winter elbow]

lies

[eternal rock]

Get notified for server events!

Click the button below to pick up the <@&1261377106890199132> role.

[tight island]

Take 50 uses each day times that with 30, then factor in token cost input and output

It’s more than 25 usd

[winter elbow]

i am using chat gpt plus and it appears the same screen

[tight island]

OAI is being generous

[bright quiver]

are u in western EU perchance? Maybe its this azure region

[dark hawk]

I have the same issues and im in france

[mental rose]

seems to be a chrome issue, i launch it from edge, seems fine, but the moment i launch it from chrome it shows that

[eternal rock]

`` Server Tip: ``

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

[midnight cedar]

Thank you, I switched to Edge and it works

[tight island]

If you fully utilize full input and output of gpt o1 mini 32k input output for 31 days, let’s assume you knew somewhat what you talked about.

Now 25$ usd for subscription that gives 50 uses each 24 hours you’re looking at a price tag of 744 USD

SO in conclusion you guys should be very grateful you even got 50 uses of gpt o1 mini each day

The fact of the matter is you’re looking at a cost of 0.48 USD each msg roughly estimated

So complaining about lack of uses is absolutely UNWARRANTED

They’re bleeding money with this model for subscribers

[celest quartz]

truly a magnificent tool

suddenly deciding it can't draw bird pirates walking.

[worn finch]

i’m getting pretty bored and need chatgpt to browse the web and do automated tasks for me, i also have advanced voice mode and have had it for a month now, really bored and need automated tasks.

[winter elbow]

guyyyss

why i cant acess chat gpt

its still in the html page

[onyx horizon]

hmmmm

[feral zinc]

dont use chrome

[onyx horizon]

It's Microsoft Edge

[feral zinc]

then it has the same issue due using chromium

need to wait for openai fix

[onyx horizon]

It works on mobile tho

[cloud pewter]

I’m using the stylish extension to apply custom font to the interface but it doesn’t keep it after I refresh the page since like last week

[grizzled badge]

It's a cors issue...
chatgpt.com/:1 Access to script at 'https://cdn.oaistatic.com/assets/b6qqxvxchvce818i.js' from origin 'https://chatgpt.com' has been blocked by CORS policy: No 'Access-Control-Allow-Origin' header is present on the requested resource.

I wonder what claude is up to today? 😛

Please hurry fixing ChatGPT, I've lost the ability for independent thought. I need some brainstorming ideas.

[calm ember]

Hi

i can't login to chatgpt

what's the problem ? any suggestion please ?

[feral zinc]

use safari or firefox

[calm ember]

thank you so much bro

[cursive needle]

Seems like its a Chrome problem with version 129

I have the same

https://status.openai.com/

[bright quiver]

The fixed it!

[acoustic wasp]

I talked about the cost of running the model

[sonic comet]

It's better if you use brave browser.

[tranquil jackal]

Did OpenAI just make ChatGPT without subscription a free trial only?

Where did gpt 3.5 go..

[sonic comet]

GPT-3.5 is no longer accessible from the chatgpt website. Now it is only accessible through API.

[raven coyote]

Hello I am and advanced Engineer in Artificial Intelligence and produce rather distant data processing to find out the capabilities of Artificial Intelligence’s range of question answering, further talking. I am asking for questions that might startle the Ai in ways, and I will find a way to fix it.

[civic brook]

Guys, would you know to tell me if it's true that the usage limits of O1-mini are now 50 per day - or are they still 50 per week?

[sonic comet]

50 per day.

[civic brook]

Thank you for your reply, could you send me the official source so that I can check the information, please? alias, it resets midnight or when 24 hours of the first message sent?

[worldly mantle]

Any chance anyone else is getting issues with o1 finishing messages?

keeps getting stuck and then cutoff message

[worthy lynx]

Yep, I have. The message doesn't cut off, but after its output finishes, the message doesn't actually end. Recently, it DID start ending, but it took 2 minutes per message (again, 2 minutes to end AFTER it had clearly finished outputting its message).

[snow plover]

where do you get the model price/token at?

nvm, found the page

[tacit zinc]

Where can i create a support ticket

[swift zinc]

||I don't know||

[fallow bear]

Is anyone else having numerous issues with the standard voice mode. I'm getting tons of static, voice stopping, conversation suddenly not sending my replies making me having to restart the voice conversation and start over. It's been downright unusable for the past month...

[wild sinew]

iirc, around launch of o1-preview, I was able to use /search to make it search the web for things - but that seems gone now - the / command only gets "reason" as an option.

[ivory iris]

Maybe try #1070006915414900886

[restive swallow]

Anyone else notice the 4o and o models punting a task and saying they'll do it later? I am getting so many responses where it ends a response with something along the lines of "I'll provide you with the updated versions shortly" or " I estimate that this will take around 2-3 hours" or "I'll get back to you soon with detailed recommendations!".

And then it never actually gets back to me unless prompted again.

[sonic comet]

https://help.openai.com/en/articles/9824962-openai-o1-preview-and-o1-mini-usage-limits-on-chatgpt-and-the-api?hl=en-IN

[wild sinew]

Not sure if this helps - I saw this exact same phrase in Bing copilot just today.

[wet sorrel]

yo. any idea when o1 will fully release?

[winged willow]

Thanks for this only mine doesn't show when my usage limit will reset for some reason

[ivory iris]

Maybe in a few months? Seems like there are a lot of bugs to work out first.

[rugged igloo]

by his tweet december

[dense halo]

Dead app?

It's dead

down, buried, gone.

[undone saffron]

Has anyone else noticed that when creating a new chat it sometimes randomly switches to o1-preview/mini? I haven't had anything like this until now, most of the time I get 4o but then out of nowhere without even having touched mini/preview it switches by itself when making a new chat.

[restive swallow]

Interesting. Yeah at this point 4o is almost unusable because that seems to be the default answer I get to anything sufficiently complex. The older model seems to do a little bit better at actually following through.

[native otter]

I heavily disagree that gpt4o is unusable..

First it has the tooling like vision capabilities, data analysis, web search, and dalle3 for image generation, and if OpenAI actually keep their promises, we should also be getting advanced voice mode with live video feed, and the improved image gen capabilities

Gpt4o is fast and kinda cheap, also it can give pretty good answers if you know how to prompt it well...

[sick pecan]

it also has third party tooling, unlike the anthropics and googles of the world

ill be into o1 when it can access my systems/apps

[mint crypt]

Anybody else notice the the little info button in regular voice mode that said when advanced voice mode is coming is gone?

[ashen prawn]

still there for me?

[native otter]

UNLESS you were selected to try it out without realizing

[opaque herald]

I use chatgpt on a pinned tab in safari on my iPad. I used to stay logged in essentially forever though I've gone through periodic times when it would begin logging me out frequently, but for the past week or so it logs me out within minutes of last using it. I also notice when I log back in it often doesn't go to my most recent chat but loads a chat further down on the timeliness. Don't know if that means anything. Anyway, I cleared openai and chatgpt cookies but that doesn't solve the issue either. Any ideas? Thanks.

[onyx roost]

It's gone for me as well

[prime swallow]

So, the model follows patterns.

Once it says no or a conversation goes weird - start over, new conversation.

The odds of a 'went bad' conversation continuing to 'stay bad' is extremely high.

[glossy fjord]

GPT down ?

[late wind]

using right now, O1 and 4 are working for chat, unsure about api?

[abstract pendant]

Wow, 4o just got prissy. Earlier today, as part of debugging my custom instructions. I asked it to repeat them verbatim, and it did, just fine. Just now, it refused, due to privacy concerns. Twice in two different conversations.

[restive swallow]

Happened to me as well even though I wasn't working on anything remotely sensitive or private. I would simply reassure it that it was the same information we had been working on and urged it to continue, and that worked sometimes.

[abstract pendant]

One of the reasons I wanted it to repeat them was so they'd be in the context when I switched to one of the models that doesn't support the custom instructions.

yes, I can go fetch them ans paste them manually, but that defeats the purpose of having a smart AI. And it's kind of painstaking to have to do it twice. I'm visually impared and using a screen reader.

[dreamy cypress]

Is there a way to get advanced voice mode faster if im a plus subscriber?

Been waiting for ages for it lol

[limber oxide]

guys
o1-mini is more reliable than 4o? for study ( i study medicine)

[gritty thicket]

ofc with enough context

[limber oxide]

wdym enough context?

[abstract pendant]

I'm a plus subscriber and I still don't have it. I suspect free users won't get it until some time after every last plus subscriber gets it, if they get it at all.

[brave knot]

is there a problem with the openai across the board? The chat.openai.com is also not working with the 4o-mini model. https://community.openai.com/t/unknown-model-gpt-4o-mini/946600

I have tried chat.openai.com and exhausted all free messages to force using the 4o-mini but it still uses the 4-turbo model when I ask it which model is in use

[abstract pendant]

I never trust ChatGPT's answers about exactly which model it is when asked. They don't seem to train new sub-versions all that well on which specific model version they are.

[brave knot]

@Teggiie Good to know. However, API is confirmation that 4o-mini is not available for the last 12 hours. It is frustrating since it is not documented anywhere that there is an outage.

[naive wren]

Does anyone know how chatgpt inteprates pdf documents? does it just extract text via code or does it read the document somehow?

[harsh pilot]

I've been using 4o, and every so often I'll get a warning about (paraphrasing) running out of minutes with o1, or the other models. I'm not using o1, or the other models. Is 4o tied into the other models, so that whatever 'time' is used counts against them?

[vestal hedge]

https://chatgpt.com/share/66ebaf45-c768-8003-bbc3-3460e7870afc interesting that o1 gave korean in extended thought. (click da dropdown)

[meager belfry]

you guys chatGPT is working?

https://chatgpt.com/backend-api/conversation/init 403 (Forbidden)

[tight island]

Peak interface

[astral dagger]

Does anyone know why when tuning customgpt, it will forget previous instructions? How can I add instructions to it without the customgpt wiping itself off and only remember the current most instruction (when it was supposed to add to its understanding)

[inland heron]

how do I stop o1-mini from focusing on irrelevant issues from previous replies? It keeps going back and cannot focus

if anyone knows what specific thing to say in a prompt let me know. I've tried many things and it keeps failing

[brisk tapir]

use o1 instead?

[inland heron]

I need more prompts, and o1-mini is specifically suggested for coding

[brisk tapir]

how about starting a new conversation?

[celest quartz]

Ew

[inland heron]

that's certainly an option, but not a real solution, but a workaround. And I have to waste more time and energy to get all the context in

it's just a bunch of code from various modules in my app (context for the model), I ask one thing and we work through it and solve the issue after 5-10 outputs. Then there's a slightly different issue, but the model keeps focusing on the previous issue

I keep reiterating not to focus on previous issues, but it takes 2-3 prompts for the model to realize this

Similar issue with all other models, but wiht this one it's more noticeable and a waste of time and prompts

why would I say that?

I started saying to focus strictly on this particular issue, but doesn't seem to work

again, it takes 2-3, sometimes more, promtps to actually start working on the issue

but the issue starts before I say to focus

the focusing is what solves it

The issue is that it thinks we are still on the previous problems

[lucid pilot]

I use o1 for complex coding tasks I find mini goes back on previous issues from earlier in the conversation

[inland heron]

ok this person gets the same issue

how many prompts o1 gets?

I might just switch to it and go slowly and not waste prompts if possible

but on the openai website it's recommended specifically for code and math

[lucid pilot]

I haven't run out of prompts yet but have been super careful and try to solve as many tasks per prompt. Sometimes I'll use 4o to help write the most efficient prompts

[inland heron]

oh, smart

I'll try that

this is extra time and energy needed for all this, but might be worth it

wish there weren't so many things we need to do just to get the GPT to understand our context

always has been this issue

[lucid pilot]

O1 mini is great for functions and generating from scratch. If you need something more complex like refactoring an entire system over several scripts and a lot of functions that's where o1 comes In handy

[inland heron]

it's not bad, it's pretty good, just that it has issues, just like every other model

[fathom tinsel]

o1-mini / 50 per day and o1-preview / 50 per week

[inland heron]

thanks

[sweet void]

trying to find some info on the teams sub, I am finding things that say the plus is 40 per 3 hours, and the teams is 160 per 3 hours, is that the total team or each member of the team?

[tight island]

Each individual member so even if you buy more slots they’re only allocated to members or teammates

So you don’t benefit from more members

[rose gale]

does anyone have chat gpt with alpha version of advanced voice mode?

[sweet void]

thanks for info! will make the decision easier now to propose to the board heh

[tight island]

That is how I have understood it, I do recommend that you read the documentation

[sweet void]

I will go read it more, but it makes sense because technically if you get more than 4 team members you "loose" the amount you can do over individual subs and pay more

[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.

[acoustic wasp]

ChatGPT voice mode got a new ui and the I that days the advsnced vpice mode is in development is gone. What does that mean

Is advanced voice mode finally there? (My app still doesn't have avm)

[north coral]

Its been stuck like thos for few minutes now, cant continue the chat...

[jolly garnet]

My app isnt working either

[north coral]

Cant they use chatgpt themselves if they cant code... 😄

Yeah, no, it doesnt cancel....

[past fiber]

My question is
Is there a third model in the o serie?
So we have a o1 mini
o1 preview

But I also saw a o1 on openai site.
Does it mean that there is a much better model still waiting

[north coral]

I think they use the full full model for synthetic data generation and wont release it fully

[crisp cairn]

is there a way to see how many messages are left to o1-previev before the weekly limit is hit?

[crisp cairn]

they will release it, it's previev, full o1 is to be released (so current model no longer "preview")

[past fiber]

Don't think so
You just have to eye ball that

[crisp cairn]

i'm lacking this...

don't know how much is left so work with o1 conservatively

but maybe the limit is reset in 2 days and just could use it fully...

[past fiber]

You could work with o1 mini
this one gives you 50 a day

[crisp cairn]

i just used it with one project, o1 mini could not grasp the concept fully...

[past fiber]

Well then
You can only do 50 a week on the preview model
I would suggest not using it if you don't need to
Play it on Eco mode

[crisp cairn]

even with docs i shared, o1 mini did not understand the problem to be fully accurate...

only o1 is providing me accurate results

[past fiber]

There is a reason why it is so cheap to run

[crisp cairn]

yes, so i don't even say about gpt 4o

the problem is that the training data does not have those concepts, so the model needs to be fed with the docs and to learn from them, and there are some technical nuances that need to be taken into consideration to understand the concept fully

for now only o1 seems to actually grasp all the details from the docs

and can connect dots

[north coral]

Anyone has good prompts for structured output for o1-p? I mean so it uses structured output mode for 4o-mini in API code 🙂

[humble pagoda]

Gpt4o is bigger than o1 mini presumably but it needs to output less tokens so we have more messages.

[north coral]

I got a warning that I have 5 left

[crisp cairn]

okay good to know there is a warning, does it say when the cap will reset?

[north coral]

Nah... 😄

[crisp cairn]

okay nice 😄

[north coral]

But it does this, which bricks that chat forever... Gets stuck, stoping doesnt work 😄

[neon raven]

so wait does GPT resets o1 preview/mini at the same time for everybody?

[north coral]

And my last 3 o1-preview responses buged out and Im out of them... What a joke... 😄

[unborn trail]

Previous information seen says it resets counting from when you first used it. From your own first message. For o1-preview at least.

[neon raven]

Ok thanks

[fathom tinsel]

It's a rolling window, the first message falls out of the window again after 7 days, at least that's how it works for 4o and 4-turbo

And 24h for o1-mini / 7 days for o1-preview

[neon raven]

Thought it was 7 days for both?

Is there documentation on this?

[fathom tinsel]

They changed it, there was a tweet about it and also they changed the documentation https://help.openai.com/en/articles/6950777-what-is-chatgpt-plus#h_d78bb59065

[hard crag]

has anyone here had the issue of voice mode being unusable when connected to Wi-Fi

when I disconnect my Wi-Fi voice mode works perfectly fine. when I connect to Wi-Fi, I can only talk to voice mode for about a minute and then it starts buffering infinitely.

It has been like this for a month now. never had this problem before. I rebooted my Wi-Fi, uninstalled the app and reinstalled, turn my phone off and on, cleared the cache for the app and nothing works. I'm speaking in this chat because support couldn't even help me. kind of getting desperate. this is a really annoying problem to deal with especially as a paying customer

[raw hamlet]

Who is Mary?
part of ChatGPT o1's CoT.
I'm not Mary and have not used the name Mary anywhere in our chats. Has it decided to suddenly call itself Mary? 🤔
I guess it was an error... as Mary appears nowhere else in the CoT or response.

[rose otter]

site down?

[bleak sapphire]

Any news on the GPT voice mode? (The new type)

[shell goblet]

only this new ui for standard

@bleak sapphire

[shell goblet]

and this disappeared for me

[native otter]

I still have it 🤔

Android btw

[shell goblet]

Can you check for updates?

[proven prairie]

The o in o1 stands for observer, right? 1 observer.

Also anyone have the "observer" interject with a random phrase in another language while it's doing some thinking?

[honest jay]

Hmm good question, I can't find a specific answer for this myself, would be curious if anyone knows the answer for sure, but the o in 4o stands for omni, as in omnimodal. Unsure if that would be the same for o1 as a reasoning model

[sick pecan]

i thought it stood for orion

[proven prairie]

Oh makes sense! Thanks.
Don't ban me for probing o1-preview 🙏

[sick pecan]

we only see the summaries of the internal monologue. so it is always unclear if the summary hallucinated or the internal thoughts did

i'm sure there's a black mirror plot in there somewhere

[ruby cairn]

This is also like that meme of the couple in bed.. "partner assuming what he's thinking about" vs "what he's really thinking about"

[quiet mountain]

why do you guys think of o1 for generating prompts?

[sick pecan]

the smartest model farming out prompts to lesser models is the dream. but i've yet to see it work

(excluding distillation techniques)

[quiet mountain]

it's been working like a dream for me so far.

[sick pecan]

what kinda prompt you generating?

[quiet mountain]

i am creating a marketing ai bot that sells apple products.

he also gives tech advice around creative content generation

[gloomy brook]

ummm.... Asking o1 to help you understand why it did not follow prompt fully will now get you a warning. I was asking to help improve my prompt.

anyone experiencing this?

[quiet mountain]

what was the prompt

[prime swallow]

o1-mini is, in my opinion, 'hard to steer' especially after it 'thinks it knows what it should do'.

I recommend add to the new conversation's original prompt an overview of your intention, which must include that you mean to skip around and redirect its attention to specific parts of the code so it can do what you need, and it needs to watch for the phrase 'we're changing to this part of the code now' - or however you want to phrase it, ready to follow that direction.

I've explored this with story-game-lessons. The model intensely feedbacked the lesson part and ignored all my story and game cues; it wrote the story progression, period. However it nicely educated while doing so, lol, but flexibility with both was my intention.

I worked flexibility and watching for my cues into the initial prompt on a next try and o1-mini followed fairly well.

[native otter]

I check for updates every day
1.2024.254

[prime swallow]

When the model doesn't have enough context, the model can guess it is related to possibly disallowed content concerns.

I have had a few of those flags from o1 and o1-mini so far.

They're not dangerous for you unless you're actually doing stuff that's not allowed; a confused model makes a LOT of those flags. But it does waste a message for you, not ideal.

To work around it, I explain in my initial prompt a bit about the context of why I want the prompt followed that I ask. This seems to avoid concerns.

They do update the model and help it avoid throwing flags as they notice patterns of 'this shouldn't throw flags'; this is from a few days ago, versus today:

[tight island]

This much is obvious

It’s thinking is unchained, uncensored and free from any moderation

The thinking you actually see isn’t it’s real thoughts merely headlines and subheadings

It’s actual thoughts are hidden on the backend and are pages upon pages long

You should all know this it’s common sense, they give you all you need to know in their article

So many of these flags on simple prompts are merely false positives

[sonic comet]

This error states that your prompt is bigger than the allowed context length. I experienced this error because of the long prompt.

[neon cedar]

I'm a bit unclear on the query allowance for a paid subscription. It says 50 queries a week for o1-preview, and 50 a day for o1-mini. It also says "Thought for 5 seconds" with the most recent query I created in o1-preview. My understanding of how midjourney did pricing before using it was a bit off. Is the query 'cost' based on the time the query takes? Or is it a flat rate per inquiry?

Also, is this the right channel for a question like this?

[sick pecan]

flat rate. api is pay as you go

[neon cedar]

Is that more efficient, or roughly equivalent?

[sick pecan]

chatgpt is stupid good value, but limited on million dollar opportunities

(assuming you spread your queries working around the 3 hour cap/daily cap)

[neon cedar]

Thanks for the info. =). What is the 3 hour cap/daily cap? Is that for 4o/o1-mini respectively?

[sick pecan]

i speak as a poweruser. if yer only doing a couple o queries a day you'd really wanna go api

[neon cedar]

Roger roger. Tks!

[tight island]

Yes a couple of long queries with o1

With the api

Cheap

We should have a power user badge

I need two because one badge can’t hold my power

[chilly cove]

o1 and 01 mini just seem annoying...

[lone patrol]

starting today, chatgpt no longer works (I get just a page with a text input at the bottom) in my Firefox (Linux), tried the usual session, the usual session after deleting all the relevant cookies, different sessions without previous use of chatgpt, private mode... nothing works, other browsers seems to be working. Anybody else experienced the same problem ?

[rugged igloo]

yiur not worthy to use it

[lone patrol]

😛

[sick pecan]

have you tried disabling experimental QUIC and/or using a phone instead of desktop

[lone patrol]

it was working until yesterday, no new features were activated. I just disabled QUIC and restart the browser, still same. It does work on other browser, and the app in the phone works too, however I need it to work in Firefox for my convenience

[grizzled badge]

o1-preview is being broke... I was asking it for a code review for some code, it was helping me write. When the code we were working on suddenly was "violating terms" (or something like that) when it was not... Even ChatGPT 4o has no issues with it.

[celest igloo]

An animated scene of a cute cat resting comfortably in a cozy room, surrounded by soft pillows and warm sunlight streaming through the window. The cat has a peaceful expression, with its eyes half-closed, and the atmosphere is serene and inviting."

[grizzled badge]

It seems there was an issue generating the image. I'll attempt it again, making sure to capture the cozy and peaceful atmosphere of the cat in the room.

It's trying again

There it goes...

[cosmic hemlock]

For the GPT app, does it automatically have previously generated prompts/discussions when you first log in?

[plush oasis]

Another day, another GPT not working...

Stop letting free users use it

[sick pecan]

another productive day thanks to the power of chatgpt. hopefully the less fortunate gain even more access to this amazing tech

[latent ibex]

i have 0 issues

[stiff ruin]

Hello everyone! I’m looking for a bot or service that can take a document, optimize its content by removing unnecessary parts, summarizing key points, and then return the optimized file. I’m open to both free and paid options. Does anyone know of a solution that can do this? Thanks in advance for your help!

[normal quarry]

You can do that with ChatGPT

[haughty hollow]

omds the implementation of reading Custom GPTs is the worst thing ever created

it dosnt finish, it dosnt load, refreshing it causes half chat gone, broken send button

[tight island]

That’s just GPT in a nutshell

Now with an API setup you can handle big and complex documents but that’s another matter

[hexed condor]

is it down again?

[hexed condor]

every time I come here there's some self-entitled dumping on the poor

[stiff ruin]

do you have some example of this?

some code implementation with nodejs?

[pulsar carbon]

hi i any news?

[near summit]

Is it down?

[slate sphinx]

It's down.

[stiff ruin]

yep

[tropic pebble]

still down for me too

[slow hinge]

down

[young obsidian]

Anything happening?

I got an update on my phone

New UI in the app. Are they launching anything?

[lyric depot]

Yes ❤️ small bugs and fixes ❤️

[neon raven]

oh man

gpt let me down

[pulsar carbon]

no

fine for me

[slate sphinx]

Are you logged in on a teams or plus account?

[tropic pebble]

plus

[fickle fiber]

Nope, definitely not working.

[pulsar carbon]

plus

[slow hinge]

how long down?

[hard crag]

app is so broken these days

havent been able to use voice mode for a month

constant bugs

idk why i pay at this point

[red prairie]

There is a reason it is called a preview

[fickle fiber]

Based on downdetector it seems to largely be an issue in the eastern US

[fickle fiber]

It's not just the preview version.

4o is having the exact same issues, as is 4o mini

Haven't tested others

[hard crag]

i wish my voice mode would just work

[red prairie]

I am not too caught up in all the versions, but aren't they also under development?

[hard crag]

getting early access to advanced voice and never being able to use it drives me crazy

[fickle fiber]

No, only o1 and o1 mini are currently under development by my understanding

[red prairie]

You are kind of still answering your own questions. "Early access"

[hard crag]

its been 1 month. hasnt been usable the entire month unless i disconnect from wifi. which is rough bc at my home the service is terrible here

[fickle fiber]

4, 4o, and 4o mini are "complete". Besides, this is not a model error, but a server connection error.

[hard crag]

hes talking about advanced voice mode

[fickle fiber]

Specifically a server connection error that appears to mostly be affecting the eastern US.

[red prairie]

Just saw this edit. It's kind of odd that it doesn't work on wifi. Is this some common error from chatgpt? If others don't report it then I would look into your router settings perhaps or try to connect to a different wifi for a moment to see if that works

[hard crag]

its very odd because the voice mode works fine on other devices in my house

but for me it starts infinitely buffering after 30 seconds

i need to test it on different wifi i guess idk its weird

[red prairie]

Did you try the app and website on your phone? If you are using the app, try to clear cache of the app

[fickle fiber]

Then idk, as the current issues are connection errors, not unique to the voice mode

I never use the voice mode and am having the same issue regardless.

[hard crag]

i cleared cache yea. the website doesnt have voice mode i believe. uninstalled and reinstalled the app. checked DNS settings

[fickle fiber]

It does

Idk about the mobile website specifically but the desktop one does have voice mode.

I've never used it but the button is there

[hard crag]

im on mobile

but yea even chatgpt support couldnt help me

feels kinda helpless

[red prairie]

Someone actually said they had some similar issues. He actually narrowed it down to being his own wifi that didn't give proper signal. So his would work without wifi and also connected to other wifis, but not his own

[light river]

is chat down atm?

[hard crag]

idk why thatd be the case because advanced voice was working fine for weeks before the issue started

[plucky folio]

Yes, ChatGPT is down. I just get intermittent answers. --> Switch to Claude. Wait, just came back

[slate sphinx]

I'm able to login with Firefox, but not in edge or chome.

[light river]

thanks

was working fine about 30 mins ago, then started playing up

[red prairie]

Out of ideas then sorry. If it's still in early access, sadly things like this is to be expected. Maybe not a full month of not using it but. Bugs here and there at least

[hard crag]

i dont blame you

i just have to do some more tests to narrow the issue

[fickle fiber]

It could be an issue with your device's wifi antenna not picking up a solid enough signal? Significant packet loss between router and device can cause issues like that.

[hard crag]

could be that

[fickle fiber]

That could explain why other devices work fine on your network and yours works off the network, but not on the network.

[hard crag]

it has to be something with my phone specifically yea

it's just weird that I can watch a 4K video on YouTube no problem but for some reason I can't connect to voice mode lol

[pulsar carbon]

degraded performance on o1-preview and o1-mini aslong as gpt4o

[sharp stirrup]

hmm, chatgpt isnt working for me, and I have plus..

[bright nymph]

is openai down right now?

[heady moss]

it's all the goddamn time

[runic grotto]

just went down?

[bright nymph]

yea...

[runic grotto]

nooooooo

[heady moss]

can't use it here too

this crap has been happening a lot lately

[runic grotto]

i've been using it non stop for 4 days no issues

[heady moss]

I get this issue every 2 days or so

I use it very frequently too

fortunately there's claude, but I'm subscribed to chatgpt

oh, actually there's gemini too, forgot about that

claude is too limited on the free tier

[runic grotto]

I have both and cursor lol but I change between depending on the task

[heady moss]

gemini is basically free

and is just as good

[narrow pecan]

gg it's down , NA north carolina

[runic grotto]

yeah everywher eI think though It looks like the ios app is working..strange

[narrow pecan]

it's back again

[runic grotto]

woot

[heady moss]

not for me tho

[bright nymph]

yup its back for me too

[heady moss]

no no, it did come back now heheh

[bright nymph]

nice!

have a nice day yaal

[runic grotto]

you too! 🙂

[heady moss]

it was a nice chat with you guys, have a great day you all, see ya on the next outage 😁

[subtle spindle]

Is anyone experiencing limited or unresponsiveness on the chatgpt.com? I don't know if there is a chatgpt outage. I have changed browsers and even internet connections. The app is working ok though.

[upper flame]

Guys i have a question i bought plus but i dont have acces to it do iu just wait or what do i do exactly?

[pulsar carbon]

again shutdown

[latent karma]

dammit ):

was in the middle of something zzz

[pulsar carbon]

401 too many request thingy

[latent karma]

api still up ? lemme check rq

[pulsar carbon]

(pending) only

and 503

[latent karma]

rip

[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.

[pulsar carbon]

rip gpt4o

i guess they are updating o1-preview to support images and other stuff with chatgpt

[visual bloom]

The webpage does not load. The app takes a long time to think aka does not respond.

Typical Friday outtage.

[pulsar carbon]

when windows app???

[hybrid skiff]

i was in the middle of something 🥹

[visual bloom]

iOS app for iPad

[pulsar carbon]

🤷‍♂️ WINDOWS*

[strange hawk]

Other people are having issues right now, correct? I was editing a custom GPT and now I can’t login on browser but iPad app is still fine.

[narrow viper]

Yeah, completely down in all UI's.

[mint juniper]

yes

now it works

[narrow viper]

yup

I was hoping it was an unannounced "flip of the switch" to the new voice mode ... nah. 🥹

[astral parcel]

gpt was down

[pulsar carbon]

chatgpt is back

[astral parcel]

oh chatgpt my love

[indigo vine]

It's down

[astral parcel]

no

[pulsar carbon]

refresh

[indigo vine]

Do you think I tried that?

[pulsar carbon]

...

forgot it

[gloomy brook]

hmmm

[runic glacier]

It's down for me.

[winged heron]

its down lol, 502 bad gateway

[cloud pewter]

stop ruining my custom font already.

[runic glacier]

It back.

[astral parcel]

Nice try Diddy

[runic glacier]

👀

[astral parcel]

🫦

[eternal rock]

Get notified for server events!

Click the button below to pick up the <@&1261377106890199132> role.

[cinder linden]

is chatgpt down or something?

since it wont load.

idk its maybe ddosed or smthing, im not really sure icl.

[astral parcel]

chatgpt writes TOO MUCH

change my mind

[pulsar carbon]

no it's working

chatgpt icon got changed

now it's dark look for tab

[coarse swift]

Is the voice model still being rolled out?

[narrow viper]

better be

[lucid kindle]

Tell it to answer in a short summary.

[oblique swift]

Sure, also ahhh yeah that makes sense.. hmm I wonder what they could do differently because they can't just not store the chat... hmm...

[ivory iris]

Yes I think RAG could be used, but my knowledge is kind of limited with it.

If I understand it right, it vectorizes data so memory recall could be done based on proximity of what it's looking for.

[dense halo]

works now

[oblique swift]

Hmmm what about if the chat was stored locally on the users device, then a scaled down version that was in the app (or at least running on the local system somehow) that would exist purely for accessing the chat database, then the messages would get sent over to ChatGPT and stored for an hour or something and as the chat goes on, of course more and more messages would be included and sent over, but only the super common messages would be stored on OpenAI's side? That should reduce the need for as much storage... actually I just realized that that doesn't make sense since all of the chats need to be stored for the model to train on anyways 🤦

Actually I guess that might work since they'd only need to have all of that data to train the model which only happens like every year or two (seeing how they release a new model around that often -_-) so they'd only need to hold the actual data for a shorter period of time, and then maybe they can give the storage drives to like microsoft cloud storage or something during that year to make extra money, and transfer the files to other microsoft drives when they need to be used again for training?

ah whatever

i don't know

[ivory iris]

It's like a blank slate every time so every new message and response has to have everything from the conversation including memories

[oblique swift]

i just realized that wasn't even what you were talking about I think

it works for me somehow?

hopefully it should work for you soon-

[ivory iris]

all good

so it isn't really where you store the messages and data, it's what you give to ChatGPT for context.

[oblique swift]

??? what do you mean

[ivory iris]

Every new message you send you ChatGPT is like a new set of messages for it that it uses for it's reply.

[oblique swift]

a new "set" of messages? how so?

[ivory iris]

Every time you respond, both your messages and ChatGPTs responses are also sent from the current conversation. If you only sent your message it wouldn't know what was said before that.

It doesn't have an internal memory state like us, that's why you'd have to build something

to mimic something like it

[oblique swift]

Oh- I just assumed that was stored before it was sent to the user and then when ChatGPT receives a reply it would then save the users reply. or maybe you're saying that it does that and I'm confused-?

[ivory iris]

The chat sessions it saves are on a server separate from ChatGPT. When you are talking to ChatGPT, all the messages in the chat including your new response are sent and that's how ChatGPT understands what's going on. If you play around with the API and have it setup to send your message without all the prior messages, you will see what I mean.

[oblique swift]

Oh, hmmm that seems like it would use a lot of bandwidth though... I guess that's better than having to store everything, I suppose?

[ivory iris]

It's just text so bandwidth isn't bad. The processing requirement of LLMs is the biggest factor.

[oblique swift]

Wha? I can attach files though?

[ivory iris]

You can but it's just going to extract the text.

I think the file size limit for documents is 2mb

[oblique swift]

What about how it generates images?

Eh, I guess that's probably why they limit access to 4o and just assume people won't upload a document for every message

[ivory iris]

Yeah images are different. I believe it prompts a separate AI to generate it.

[oblique swift]

I bet there's a ton of optimizations to be made, I wonder what the theoretical limit is on how optimized it can get

[ivory iris]

There is a lot of room for improvement not just with optimization but how it operates. We still have a long way to go.

[hard dew]

Is the web app ever going to be fixed on mobile lol? It's been days since it's been like that for me now.

[visual bloom]

Try emptying the cache and cookies. This often does the trick.

[hard dew]

Unsurprisingly, no change lmao

besides turning it from dark mode back to white mode now, which I can't change either since literally nothing shows up

[oblique swift]

Sorry, I didn't see this until now- Yeah, I'm going to enter college as soon as I have the money for it to try to get a bachelors in AI and eventually a masters, I hope there's still stuff I can do by then..

[sick pecan]

i put my masters in ai on hold because the stuff being taught is outdated

[oblique swift]

It's less for the actual knowledge and more for the certification

[sick pecan]

if LLM isnt in your first year materials then maybe find a different school

[oblique swift]

lemme check

I'm not sure 🥲 If I send a link to the course schedule could you check?

[plain crystal]

well 4o1 is better but still far from perfect, as a simplistic example, give it the following prompt:
give me the results for the following problems:
4+1=
3+4=
2-1=

and it answers:
i'd be more than happy to help you out.
4+1=5
would you like me to solve the other problem

just there i lost one prompt, when the first prompt was clear to solve all the problemS:

when coding with it it will hit a point of no return at around 700 characters and it will start to either omit part of the code to correct in its answer or simply say it did the modification to the code when it did not, again making you prompt it again and loosing another prompt...

again, still better than 4 or 4o, but far from been really efficient

[oblique swift]

It's 1 page pdf so i might be able to just download it and send it here but I'm not sure

[oblique swift]

Hopefully that will get better with time, I'm going to assume that's just OpenAI's fault for limiting the context window on the second part, as for the first one... I don't really have anything to say to that that just seems like something that should have been fixed... i guess it hasn't been?

[normal quarry]

it's not 4o1 it's just OpenAI o1 it's seperate from the GPT models.

We have
GPT-4
GPT-4o
GPT-4o mini

then we have

OpenAI o1
OpenAI o1 mini

[plain crystal]

my bad. does not change the comment i made, but you are right

[oblique swift]

I'm so slow I can just send a pic of it

[eternal rock]

`` Server Tip: ``

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

[plain crystal]

i agree, i mean, i am not pissed or anything and i tottaly agree that they are doing progress and i am about certain that in the next year many surprise and capacity are to come that will surely surprise us

[oblique swift]

yeah, I'm excited for what the o1 release will do, I made a bet with a friend that unemployment in the US would reach 14% by the end of the year in January or Febuary, still not sure if I'll be right or wrong 🥲

[plain crystal]

definitely, i work in the event industry as a technician and i know for sure that ounce they figure out spacial awareness, plug that i a machine and machine will be able to setup stage and all, i hope they also understand that unemployment is also less people able to buy stuff

machine don't care about taylor swift concerts

[oblique swift]

🤣

[oblique swift]

That's another reason to enter the AI field, pretty sure that ai research and development will be automated last so i might as well do it.

[plain crystal]

but they wont need 8 billion people in the domain, so lets say 8 millions people work in AI across the globe , the world economy needs way more spending then 8 millions people spending to "work"

[oblique swift]

yeahhh but that's why things like UBI and ideas about a post labor economy exist

[plain crystal]

and it migh go faster than the first computer boom to reach human independancy

[plain crystal]

agree on that

[oblique swift]

damn. I hope. it would be nice to not need to work and simultaneously be able to witness the exponential advancement of technology and science

[plain crystal]

give it 20 years i think, i might not see it unless AI really improve the medical domain, but my kid will for sure

it could be faster, but i think human will slow it down at one point in order to mentaly catch up to the progress

[oblique swift]

maybe, but I kind of think we can't. even if we could, it's an arms race between nations, the one who produces the AI the fastest rules everything and uhhhh I don't see anybody slowing down on that

[plain crystal]

i agree, unless that nation can't keep up with it's AI, we still have mostly of a complex algorithm currently that sows word together without knowing what it's saying than a full fledge conscious machine "a la" ex-machina

[oblique swift]

???

[jaunty robin]

AI is already entering the medical domain, you just have to look at the areas where they don't have any choice (due to lack of doctors or lack of funding)

[plain crystal]

i know it is but it's not yet at it's max capacity

[oblique swift]

as in not as good as a human? i haven't looked into that part yet so i don't know

[plain crystal]

the current AI is more a ai than a truly conscious AI, it is more and more capable of appearing conscious and capable of reasoning, but it's still doing it based on how we tell it to do, for example when it say i cannot answer question X because it is not ethical, it is not doing it because it taught about it and went "that is wrong" it just does because it was told that it is not ethical.

[oblique swift]

regardless of if that's true or not, that doesn't change the economic benefits which is what we should be focusing on

imo at least

[plain crystal]

i agree, in my opinion it is the tool capable of leveling the playing field for all, mostly ounce it gets to physicaly interact with it's environement

[oblique swift]

leveling the playing field? what do you mean?

[plain crystal]

simplistically said, making it easier for everyone to have access to a better life... sorry english is not my native language

i am not talking about luxury here, just things like medical treatment, basic goods, education, etc..

[oblique swift]

it's fine, I don't mind 🙂

yeah, those would definitely be nice... ultimately I think it would be best if an ASI (that was friendly to humans and genuinely didn't harbor any negative feelings) were to reorganize society.

I know that there are some that think that humans should be in charge of humans... but we have clearly not been doing a good job...

[sick pecan]

if i am tasked with defining the ASI bias, then i am onboard with your plan. otherwise naw

[oblique swift]

fair enough- ah! I just realized- what do you think of the course schedule? You seemed to know something, at least more than me :p

but yeah, that would just be in an ideal world, of course in reality it could and actually probably should be more like an ASI would suggest things and humans would implement them, which makes perfect sense to me... assuming that the person who actually has the power is a good person... which defeats the entire purpose ;-;

buuuut it's got a lot of potential assuming it's not absolutely terrible imo! 😄

actually I guess the same could be said for anything- "It would be great if it wasn't so terrible" ;-;

[sick pecan]

looks like a pretty sweet schedule

obviously if you wanna go deep into ai then for electives you are choosing maths subjects

[oblique swift]

really? neat! :]

It says that for the electives, got any suggestions then? :p

[sick pecan]

but theres also gonna be so much cross-domain work that ai+anything will win

[oblique swift]

well this is just up until bachelors I'll probably know enough to pick the masters by myself by the time I get to that point- hopefully anyways

I mean I vaguely know what these terms mean but... not very well ;-;

[wet heath]

what is that schedule for

[oblique swift]

a bachelors degree for a course in AI I'm planning on taking

[wet heath]

oh ok

[oblique swift]

yeahhh- if you've got any advice I'm all ears! 😄

[wet heath]

that schedule looks realy "spread" i mean a lot of diffrent areas. and are they teaching you programming or just general use of stuff ?

[oblique swift]

it's supposed to be an AI focused course, but it's an introductory one so it's got some programming classes in it

it's just some university in the UK that I found that's cheap and doesn't include gen ed courses like colleges in America do-

I'm already familiar with programming concepts (at an intermediate level I'd say) so I don't need them

I can't find one that doesn't include them though that's still a degree

I'd take a certification exam and self study but that doesn't look as good on a resume 🥲

err- I think this belongs in #off-topic so I'm gonna go there :p

[dusky cliff]

wen scarlett joahnsen?

[wet heath]

best thing is to get a job and actually do something. papers means less to be honest. or do some projects you can show. if you do know programming.

[golden sequoia]

hi sorry to bother you im new here and im not fluent in english im french do you now ih chatgpt ave some trouble today ?

[dusky cliff]

uhlalah

<3

[wet heath]

yes there is ongoing issues https://status.openai.com/

[golden sequoia]

TY mathisen

[dusky cliff]

wen Scarlett Johansen?

[ivory iris]

There will always be more 🙂

[oblique swift]

fair enough-

[swift zinc]

discord=disco？

[native otter]

[ivory iris]

Is that with o1-mini?

[native otter]

Yes

[ivory iris]

That makes sense.

[native otter]

Honestly I don't measure intelligence at all based on the strawberry test :V

But thought it's funny how it's correct at first, then comes up with a weird excuse on why it's actually 2

[ivory iris]

That's why it's always good to ask the question if it's sure. This helps show if it actually understands or seems more like a guess.

[native otter]

Outside of o1, I heavily use 4o data analysis for more accurate results

[neon raven]

Does anyone know how to get chatgpt to read out lengthy text that I can easily paste but I need a workaround to paste lengthy texts

my iPhone speak to text Siri or whatever is crap

[tight island]

So you’re telling me you need it to read lengthy text in which IT DID NOT write to read out loud

Use an ordinary free to use tts app online for that

Or ask it to write it verbatim in the chat then use the read aloud feature

Of course this all depends on your definition of lengthy text

[neon raven]

That’s right i want to use that voice INSTEAD of the voice speak to text because it’s choppy with Siri

I’ve compared the same text with gpt and my phone and gpt is better

[tight island]

How much is lengthy

Give me a word count

[neon raven]

Novel length

[tight island]

Don’t bother

You’re going to need the api for that

Which costs money

[neon raven]

Sure I have the api hooked up

[neon raven]

Cost little no?

[tight island]

Novel length no

[neon raven]

They don’t make it easy 🤕

[next veldt]

Anyone else having the "Try again" button disappear? It makes it dramatically harder to work on making images.

[rose spire]

ChatGPT down?

[old skiff]

https://github.com/openai/tiktoken
is this updated to include token counts for o1?

[inland violet]

👍

[inland violet]

good morning

[old skiff]

can you show me where?

[lone ruin]

a quick question.

Does latest o1 model from openai know what it does not know?

Like lets say I give this prompt.

"Here is the title of the research paper. "RESEARCH PAPER TITLE HERE". If you already have context about this research paper or you have fully read this research paper, then give me the summary of this research paper. If you don't have context then ask me to give me context to you and I will provide you the content of this research paper."

Let's say if o1 does not have context about this particular reserach. Will it say I don't have context or generate random plausible response according to the reserach paper title?

[inland violet]

No sorry i dont tried it

[old skiff]

why did you reply to me then, blocking you

[strange condor]

is there a way to make GPT memorize only certain informations? I really like the memory feature but lately it seems to archive just about everything I say :/

[inland violet]

@old skiff It's your decision. I haven't done anything bad

[humble pagoda]

maybe some custom instructions will help?

[strange condor]

hmm.. i'll sure try. thanks!

[celest galleon]

Guys how do I manage my ChatGPT Plus subscription? It is telling me to click "My Plan" in the left sidebar but there is no such button

bruh... I had to click "Add Team workspace" and then there is a super tiny clickable text "Manage my subscription" in the box... That is pretty hostile design if you ask me

I'm not even trying to cancel, I just want to change the payment method

But ok, I found it now

[tight island]

As far as I can see it has not been updated.

[tight island]

Yeah there is

My question is do you wish to know?

If so I need you to choose between two colors..,

RED: 🟥 if you wish to understand the complexity of this universe
BLUE: 🟦 if you wish to remain ignorant

now choose…
You have to let it all go, fear, doubt, and disbelief, FREE YOURSELF make YOUR CHOICE **@strange condor **

The time is now, fulfill your destiny yeon the world is within your grasp make your choice which color will you choose

[strange condor]

wait what

[tight island]

Yes yeon it’s time to choose

[strange condor]

well i'd appreciate some help if that makes sense

[tight island]

The color yeon which will it be

[strange condor]

i didn't watch the movie but didn't he take the red pill?

i guess i'll do the same if you can help me with this memory options

[tight island]

Ahh I see you chose the red color, great choice

dramatic sci-fi noises

Are you a subscriber or a free user

[strange condor]

i've been subscribing for a while

[tight island]

Subscriber id assume so what you would do is this

Disable memory and custom instructions

Go to “Create myGPT” give it the knowledge it needs in form of actual files

[strange condor]

i just don't want it to memorize temporary events as if it's current

like if I tell it, for such, i hurt my finger, it would memorize that "the user is injured" or something.

[tight island]

And you do not want that

[strange condor]

i want it to memorize big big stuff like i guess my name

but not small stuff

i can't get it to memorize only certain details nor evaluate which to remember or not ;/

[tight island]

Ask if to clear all memory then proceed to tell it to add the following text to memory, “Whenever the user states a fact about themselves always ask if the user wants it added to memory” and this as well “Never forget to ask the user before writing to memory ALWAYS remember this regardless of user msg”

[strange condor]

ok, i'll try, thanks

[tight island]

Now I just did that in the fly but it goes something like that

I would ask a GPT to properly formulate the msgs

[deep pike]

why does it show this when i provide it a long code? it only happens on o1 models the mini does it more than the preview version. its really annoying

[tight island]

False positive

[normal quarry]

it's working fine, it's a problem on your end

this is a problem on your end

on the same network?

well, all I can tell you is ChatGPT is working, it's not down and hasn't been down in the last 24 hours, so it's a problem on your end

that was September 20, more than 24 hours ago

If you're still having problems then reach out to support at https://help.openai.com/

you can reach support using the chat on https://help.openai.com/

[sonic comet]

Aaaah Great kids quarral.

[tight island]

Robert…

You need to stop talking to yourself

[normal quarry]

🙃

[tight island]

I know I do it too

It’s hard to quit

[native haven]

hey

[pulsar carbon]

hi guys i cant believe this

o1 is based on gpt4????

not even gpt4o

[jolly osprey]

If you are asking ChatGPT itself, this is a fools errand.

It has no idea about itself. It'll give a different answer almost every time

[pulsar carbon]

fix it please openai

[jolly osprey]

fix what??

[quiet mountain]

LOL

[gilded ferry]

Do you think advanced voice mode is being released very soon? I say this because the current voice mode is taking a while to connect, and I am on good internet speeds. Maybe a sign of something? Anybody seeing the same thing where it takes a long time to connect via voice on the Android app?

[gilded ferry]

Happens to me too. I think maybe it's a limit error. But it defaults to the policy error by default due to a bug.

[rugged igloo]

24th

[rugged igloo]

but what i seen on how it performs its butchered and its not even half of what it was when they showed demons so i dont really care for it now

[gilded ferry]

How did you come to learn that?

[rugged igloo]

twitter and reddit where i get all my news and leaks and other employess talk about chat gpt

[onyx roost]

Advanced voice gets released to everyone on the 24?

[rugged igloo]

its more like another rollout than to everyone

[onyx roost]

Crap, been waiting forever.

[rugged igloo]

so you might get lucky and get or get it in the following weeks again

[onyx roost]

I swear only like 100 people get access in a rollout

[prime swallow]

Have you considered telling the model what you want from it? https://chatgpt.com/share/66eeec91-55ec-8011-b717-0009cbe887d5

[prime swallow]

Hehe. Typically, if people get access to a rollout and there's no problems with the stuff they try, they're using it, not here telling us about their access (at least until they run out of uses for a bit, even then most folks are more likely to discuss what they're unhappy with than what is going well)

[astral parcel]

x)

[bitter bobcat]

The manage my subscription button is missing.

[prime swallow]

Agree with Thursday there. 4o does not know it's 4o. If you have it search and compare itself to the description of 4o, it believes it's not even similar.

(amusingly, humans do this too. Tell a human about themself, even if it's a lie; many humans believe this even if it isn't true, especially if the lie started early enough in their life. But we all only know what we are told about ourselves or decide for ourselves, based on the available evidence.)

[bitter bobcat]

Nvm I found it

[pulsar carbon]

thanks for clairification! very appreciate

[wet heath]

not sure if they said it but it seem the 01 reset is always on mondays. as i hit my limit right now.

You’ve hit the Plus plan limit for o1-preview.
Responses will use another model until your limit resets September 23, 2024.

[bronze quail]

Anyone else affected by the “disappearing conversations” bug? (This: https://discord.com/channels/974519864045756446/1283020160768409640)

[hexed condor]

didn't they say we're getting 50 messages per day?

[wet heath]

for mini yes

[hexed condor]

✅ 🙏🏽

[winged moth]

Mine hold me the reset is on the 24th, so maybe it can vary for people

[honest jay]

https://help.openai.com/en/articles/9824962-openai-o1-preview-and-o1-mini-usage-limits-on-chatgpt-and-the-api

Your weekly usage limit resets every seven days after you send your first message. For example, if you start sending messages on September 12, your limit will reset on September 19 (00:00 UTC), regardless of when you reach the limit.

[stable quartz]

any eta on when the thinking models will be avaliable to the public api?

[wet heath]

good to know

[midnight nimbus]

Hey, I am new to working with GPT API, I tried everything including documentations and asking GPT itself, but my issue is not getting resolved, could anyone please help me? right now, its showing error of mentioning api key in client or through env variable, could anyone guide me how to do that?

client = OpenAI()
openai.api_key = 'my_api_key'
def get_product_features(obj):
    messages = [
        {"role": "system", "content": "You are a product feature analyst."},
        {"role": "user", "content": f"Analyze the following product title and its features: {json.dumps(obj)}. If any feature is wrong, return the corrected version."}
    ]
    try:
        # Making the API call (new interface)
        response = client.chat.completions.create(
            model="gpt-4o",  # Adjust this to your required model
            messages=messages,
            max_tokens=150,
            temperature=0.7
        )
        # Extract and return the response
        if 'choices' in response and len(response['choices']) > 0:
            return response['choices'][0]['message']['content'].strip()
        else:
            return None  # No valid response
    except Exception as e:
        print(f"Error: {e}")
        return None  # Return None if there's an error

The api_key client option must be set either by passing api_key to the client or by setting the OPENAI_API_KEY environment variable

[inland heron]

I'll help you when I get home. Everything can be solved, and this too

[midnight nimbus]

Thanks Alex! I know that, Trying to resolve it, I got through whole of GPT documentations and learned a lot of new things, but lol I failed to resolved this. 🙂

[sick pecan]

my_api_key is not a valid api key

[inland heron]

I have an app using gpt api, but I can't see currently what the issue is. I'll dm you

[midnight nimbus]

I know, I have just hid it, before making my code public 🙂

[sick pecan]

best to use process envvar to ensure code never contains secrets

[midnight nimbus]

will try from next time, but can u see my mistake right now?

[sick pecan]

this is common practice. save secrets into .env file etc. either source it from shell or in the code

you instantiate the client without the api token

the token is only useful when creating the client

[midnight nimbus]

so, what changes should I make now, can u please guide?

[sick pecan]

feed my advice into chatgpt and it will take you the rest of the way

[inland heron]

If there are beginner devs looking to create something great with ai, dm me. There are opportunities out there and I don't have all the time or skills (also jr beginner here so 😀)

[midnight nimbus]

I have tried GPT o1-preview about 15 times before again n again, it completely fails, but so you say, I will try again

[sick pecan]

i hear o1 mini is better at coding

because it's a lean reasoning engine not weighed down with knowing everything

[midnight nimbus]

ok, will try it on that now

client = OpenAI(api_key='your-api-key')

[sick pecan]

that's it

[midnight nimbus]

It responded correctly after feeding your advice into it, else it ruined my code completely 🥲

[sick pecan]

i find it helps to have it include professional comments. is giving it room to think more

[midnight nimbus]

Thanks, atleast i m moving now lol, was working on it since yesterday night

[sick pecan]

i will often prompt: give me updated version with professional comments, debug with values, guards.

those three things make for higher quality code(/ generation )

[midnight nimbus]

Sure! :0

[buoyant lantern]

What are the rate limits for gpt-4o-mini on the app, plus tier?

[sick pecan]

youtube was built by repurposing a failed dating app. that story has come and gone

[dense halo]

ensure you move your key to a .env file if you know how to do it. you shouldn't leave your API key in the code. use bearer or whatever idk how that works

[outer widget]

where can i check what chatgpt subscription do i have? there is no option to see it when i click on profile

[dense halo]

should be at the bottom of your conversations/chats if you're on PC

[limber oxide]

hey, do u think o1-mini is more reliable than gpt 4o to study medicine?
or just to give reliable and accurate info overall

[normal quarry]

you shouldn't use any GPT model for medical advice or medical information as it is not reliaable for this purpose

[limber oxide]

i mean
for study physiology and such

and its good

[normal quarry]

GPT-4o is better than GPT 4o-mini

[limber oxide]

yea, i know
but i asked gpt 4o vs o1-mini

which is more reliable and accurate overall?

[normal quarry]

o1 mini is good as it has reasoning but whether you need that for accuracy I do not know

[sick pecan]

they are better at different prompts/ problems. use both to minmax

[inland heron]

I've found the opposite to be true for coding EDIT: talking about o1 mini, my mistake

[normal quarry]

that's good

[inland heron]

sorry my bad, I was talking about o1 mini

not gpt4 mini

[normal quarry]

ok

[night egret]

hey everyone, is anyone else having a problem logging into chatgpt tonight?

[normal quarry]

no

[night egret]

I'm trying to get to my subscription settings

I'm on it on my phone but I can't find my subscription settings?\

nvm found it, they freakin' hide it!

[acoustic talon]

Are they gonna increase the limit for o1-preview

[normal quarry]

probably not

[sick pecan]

your first prompt should be, 'what should i consider regarding the use of openai keys found on github?'

[dense halo]

even if you found working keys you shouldn't use them its unethical

[keen quail]

oh my god

this is html btw

ignore the game i was just testing it\

[prime swallow]

We are excited that o1 made this for you, yes?

o1 is awesome... 4o can do it too. Though this did take 3 prompts; the first asked for a 'basic tetris game' (it worked and was exactly what I asked for, but so much less than your screenshot shows, my first screenshot shows the window).

The second prompt described the features your screenshot shows.

The third prompt shared a error message the updated code caused, and the fix allowed the second screenshot. It's Python, not HTML, but 4o can do it too 😉 https://chatgpt.com/share/66ef8b7b-bc88-8011-96d2-0f81324e9dfa

[keen quail]

made with gpt o1 mini

[tight island]

Yes

There are communities that actively look for keys inside repos only to spam requests

Then you end up with a huge bill 💵 if you haven’t set a hard limit

[left lily]

is there any model on ChatGPT that has a 100k+ context window

online it says 128k and i don't believe it for any of them

[sick pecan]

if you are enterprise supposedly you get 128k (in chatgpt)

for o1 i dont think it allows that much in the prompt

[muted trellis]

is this UI new ? it didn't show performing web search action. directly wrote the response and cited those links

[tight island]

But you’re not allowed to not believe it

The evidence is there

And you are not allowed to compare the context of the API compared to the web interface it’s two different things

They don’t work the same

Unless you’re enterprise the API always is better

[tight island]

Yes

[ancient trout]

this problem has been for days now anyone having same issue ?

[tight island]

How does its system prompt look like ?

[gleaming trail]

Hi, I'm trying to help a relative of mine who is a lawyer, I want to give chatgpt a pdf and I want him to reduce repeated concepts or in general help me reduce the document by transforming for example 10 pages into 7/8 pages but given that and still a legal document he must keep everything important at legal levels. I tried to give him a PDF to work on and he never did a real reduction of the text and replaced a few words but I can't find a way to get them to do what I need, if I wanted I would also have Openai credits and therefore I wouldn't have problems if there is a tool that can do this to use it with my credits. anyone have any ideas?

[ancient trout]

i can't check it just the error talking to it can't access it

[normal quarry]

you should not use ChatGPT for legal or medical matters

[wet heath]

it means that whatever API that gpt calls is having issues, so the endpoint is down or whatever access that gpt was using aka apikey/token has expired.

it is not an issue on the chatgpt side

[gleaming trail]

I know it obviously would be revised and because my father is getting old now and if a tool can help him understand when he repeats the same concept several times or uses too many words it could be very useful to him.

[normal quarry]

ChatGPT is not reliable to use for such use cases

[wet heath]

i agree with robert, legal text can be really wierd and not really "normal" text. i read in the past people trained models for this usecase just for legal documents. you should check huggingface if there is any public ones.

[gleaming trail]

if not the entire document could it just remove concept repetition and make the text parts smaller without neeeding any legal competition

[ancient trout]

yesterday it was on other bots too but today it's only on consensus that's why i think it's in chatgpt

[wet heath]

all chatgpt, server does is making a web post request to whatver endpoint and get back the data it requested. and your screenshot shows "error talking to" you get this error if the post request fails.

[ancient trout]

okay thanks for responding

[wet heath]

example here i made wrong URL by purpose after it said i am working and you see the error

[ancient trout]

okay got it thank you

[steady cradle]

Anyone else enjoys learning new concepts with ChatGPT? GPT-4o seems to be great in explaining concepts in programming for me

It's like a private teacher (tutor?) that you can ask again and again if you fail to understand the explanation, then it explains in a different way until you do get it

[hoary laurel]

Me too

[elfin lava]

Huh neat

[dense halo]

Jeez, so to get this you need to win the lottery

[elfin lava]

Either the rollout is purely random or they pick acc that doesn’t have any violation flag before cuz my other acc I made few weeks ago got advanced voice mode last week , my main acc that I created since last year didn’t get AVM 🤷

Fingers crossed for 24 September rumor

[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.

[inland violet]

great 👍

[gentle hull]

Failed to retrieve security token. Please try again.

already logged in 5 times

its not working, anyone have a solution?

is it down?

why is it so quiet here

[acoustic talon]

When are they gonna increase message limit for 01 preview

I understand no one might know but can we have some speculation? What happened to the days of speculation

[left lily]

if i feed it 75,000 tokens across three messages, with a code word at the start of message #1, it forgets the code word after message #2

i don't see any evidence whatsoever that any of the available models on the ChatGPT webfront are capable of more than 32k context window

[jade jetty]

so i have this issue any time i use the website, its this grey square that covers a good portion of the chat area, i can still copy and paste all of the stuff underneath the square but idk how to get rid of it and i would really like to

here's another example after i asked for a list of 100 things

[umbral bear]

and right now the page is not loading for me

[vocal cipher]

Hey, quick question. How many messages we have per week for the new model o1? And when they get reset?

[umbral bear]

I don't think they announced any limitations on it

[honest jay]

Hello, currently the limit is 50/week for o1-preview, and 50/day for o1-mini.

https://help.openai.com/en/articles/9824962-openai-o1-preview-and-o1-mini-usage-limits-on-chatgpt-and-the-api

[gentle hull]

last few days chatgpt has been hearing the opposite of what i say. Like if I mention I want to allow this it will think I am saying I dont want to allow this

[young obsidian]

@honest jay do you know the limits for 4o and mini?

[honest jay]

Hello! The default for 4o on Plus is 80/3hrs, but that can sometimes change based on server load: https://help.openai.com/en/articles/7102672-how-can-i-access-gpt-4-gpt-4-turbo-gpt-4o-and-gpt-4o-mini

As of May 13th 2024, Plus users will be able to send up to 80 messages every 3 hours on GPT-4o and up to 40 messages every 3 hours on GPT-4. We may reduce the limit during peak hours to keep GPT-4 and GPT-4o accessible to the widest number of people.
4o-mini is functionally unlimited - just a guess on my part but maybe in an extreme use case, you'd run into some kind of upper limit usage cap, but I'm not sure - it's described as "Unlimited" here: https://openai.com/chatgpt/pricing/

[vocal cipher]

Thanks, And this resets when?

[honest jay]

Sure thing. Reset info:

You can check the date that your usage limits restarts at any time by highlighting the model name in the model picker drop-down. Your weekly usage limit resets every seven days after you send your first message. For example, if you start sending messages on September 12, your limit will reset on September 19 (00:00 UTC), regardless of when you reach the limit.
Full details in that link in the original message if you're interested!

[feral zinc]

only enterprise has above 32k at the website and its all documented in the help articles and feature comparisons. api is uncapped

[elfin birch]

It is seen under Model Quality in https://openai.com/chatgpt/pricing/

[quaint violet]

I'm finding that o1-mini and o1-preview have stopped thinking before making a response, and I gave them a logic riddle it used to solve (which used to be after 6 seconds for o1-mini, 17 seconds for o1-preview) and now it doesn't solve it anymore. Anyone else have this issue?

[quaint violet]

Nvm, I restarted Firefox (and cleared cookies/cache) and it works now. (Might have been related to extensions I was using.)

[crisp cairn]

is it in a web view? doesn't seem to work for me

[queen kite]

Hello! I was just wondering if a lot of ppl already have access to new voice mode? Waiting time is really long omg

I'd much prefer if they announced features just before releasing them

[sick pecan]

the rumoured leak is by the end of this week

probably because google is giving their voice mode out for free to ppl with pixels (supposedly)

[elfin lava]

I do but not on my main acc surprisingly , they sent the invite on my alt acc I made few weeks ago 😐

My main acc was created last year 🤷‍♂️, oh well we can always hope for this