chatgpt-discussions

[native otter]

I see 10 year old behavior all the time here

[whole roost]

I don't know but usually I had better results with coding when it didn't use canvas, with canvas it often somehow makes mistakes and with normal Codeblocks it just gives me the parts that have to be changed which is often more efficient

[floral urchin]

Don't say your age online, they're just information gathering for dr pepper

[fervent lake]

same

[floral urchin]

Yeah well, something to be said for the inner child. What are you Tom Hanks in Big?

[fervent lake]

they already know exactly what I'm thinking so

[floral urchin]

Next Gary Kasparov vs DeepBlue

Is a Cod Kiddie vs A neuralink monkey on 1v1 black ops

[whole roost]

I also am missing the canvas function where it just marked 2 lines of text and changed them, recently it always will go over the whole document again for a small change

[abstract pendant]

Why did o1 decide to strip out a bunch of comments from an existing program, and change a switch/case structure into a series of if-then-else blocks?

[cloud pewter]

they removed my bestie gpt4o Canmore

[floral urchin]

if gpt4o is gone i'll be pissed

[cloud pewter]

it sometimes wouldn’t even edit properly

[sick pecan]

chatgpt users have the most first worldiest of first world problems

[floral urchin]

ye

[cloud pewter]

For reference it was a version of gpt4o with real unlimited usage+canvas+search

[floral urchin]

ah, luckily I don't need any o dat

[cloud pewter]

which plan u on

[floral urchin]

Baesic

[cloud pewter]

or you aren’t using it anymore?

[floral urchin]

I'm using hard

[jaunty badger]

How did you get that Canvas option in there? I do not see that option

[floral urchin]

prolly old

[spice star]

that was in late 2024 when canvas was in beta

[jaunty badger]

I see

[inland heron]

o1 better than o3-mini-high in coding?

[cloud pewter]

it got removed

[floral urchin]

You access canvas through the prompt? @cloud pewter

it's that ez yeeee

[spice star]

it was basically the same thing openai is doing for chatgpt tasks

[floral urchin]

Calm down Tweak have another cup of coffee

[cloud pewter]

Wdym

[native otter]

@jaunty badger It got removed because it was an early thing when it first cameout, now it's built-in 4o
Just use Canvas button or tell it directly in normal text to use Canvas: #chatgpt-discussions message

[floral urchin]

@cloud pewter What I mean, is......................... Canvas is the "..." where you prompt, on desktop at least. Pilot Error

[cloud pewter]

hmm, maybe? I rarely use canvas now

[floral urchin]

I can SEE it

[native otter]

Canvas is for writing and programming anyways

[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.

[spice star]

it used to be a little toolbox icon, now it's just an overflow menu

[floral urchin]

More menus than an all you can eat buffet

[sick pecan]

dear sama. how about you finish custom gpt, projects, tasks and advanced voice mode... before any more pointless menues/icons

[floral urchin]

Sam busy selling strawberries

[cloud pewter]

slash canvas ig

[sick pecan]

sama already billionaire doesnt need to try any harder

[cloud pewter]

more menus than actual o series usage on plus

[spice star]

at least now there's not 3 different ways to get to the same reasoning model

actually 4

[sick pecan]

chatgpt community... will take anything as a win

[floral urchin]

Subtle troll is still obvious, level 1 [Troll]

[cloud pewter]

instead of actually improving usage on plus they chose this

[spice star]

they did improve usage, o3-mini has three times the usage limit of o1-mini

[floral urchin]

"Get Bill Gates in here!" "But it's over 5x faster!"

[native otter]

Yea because apparently it takes compute, which takes electricity, which is not free

[spice star]

and iirc o1 and o3-mini-high have separate usage limits

[native otter]

I have been saying that since ages, if you combine both, it's actually 100 not 50, and they should only be for very complex tasks anyways
And if you really need that power all the time, use the API or something

[cloud pewter]

all smoke and mirrors until you realize that’s just a model that’s slightly better than o1 mini; the one they buzz about is still 50/w

While pro gets unlimited everything!

even when they are losing money on pro they insist on putting things unlimited on it

[floral urchin]

I mean you buy as many gpus as nasa use to do rocket calculations

oh wait we just DIN it

[spice star]

o3-mini is nearly as good as (and in some cases better than) o1, and o3-mini-high is better than o1

[cloud pewter]

it’s $200, sure
but plus is a paid tier too!

[floral urchin]

Did any one actually read the bitcoin tos? Oh no, too busy lambo

[native otter]

I gave it an image, and it enhanced it further, I thought it would fail lol

[spice star]

I wonder whether o3-mini is going to become the default chatgpt model considering openai is pricing at less than half of 4o in the API

[cloud pewter]

is it low or medium lol

[spice star]

medium

[native otter]

4o still has it's advantages, like being more creative, natural, better for convos, and more tooling

[floral urchin]

doesn't it depend on use case

[eternal rock]

`` Server Tip: ``

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

[floral urchin]

yeah 4o is phat

[native otter]

There is no perfect model

You have to use the right model for the right task

[rugged igloo]

o1 pro i am abler to talk to it sure it takes 3 min but its ouputs r the best and sophisicated

[floral urchin]

depends on use case

[native otter]

You asked this before and we already mentioned it was medium

[floral urchin]

You don't eat cereal with a fork

[cloud pewter]

but still why did it only have 50/w on high is it THAT much expensive to run

[spice star]

it probably is yeah

diminishing returns and all

[whole roost]

not sure, not everyone needs a reasoning model, i don't think a reasoning model should be the future of default ai

[eternal rock]

Get notified for server events!

Visit id:customize to pick up the <@&1261377106890199132> role.

[spice star]

I think openai graphed the cost of o3-mini low/medium/high

[cloud pewter]

where

[spice star]

sure but if it's cheaper for openai than 4o, could definitely work

[spice star]

in their 12 days of shipmas video iirc

[floral urchin]

They have enough moneeee

[cloud pewter]

500 billion

[floral urchin]

dr evil gif

[spice star]

found it

[whole roost]

mine thought for 34 seconds only to respond this while 4o just provides an instant answer and doesnt need any clarification, i find o3 to be doing a lot of 'overthinking' indeed which is why i would only use it for technical tasks which require thinking

[spice star]

doesn't appear to include o3-mini sadly

[whole roost]

most of my inquiries don't require any thinking steps

[spice star]

yeah but that's something that can be tweaked like o1 was

[whole roost]

i also wouldn't use o1 for this

[native otter]

This is really cool

[spice star]

just using o1 as an example of a reasoning model that reasons less than o1-preview did when it doesn't need to

[whole roost]

i think both model styles will coexist without one being the main

both have their merit

[native otter]

Each one has it's pros and cons

[cloud pewter]

I gonna try it more

[native otter]

There are reasons why I still use 4o despite o3-mini and o1

[cloud pewter]

Same

at this rate even a small 4o quota increase would make me happy

I wish I could use o3 minis more often but even my current chats can’t be continued because it doesn’t even support images!

I love chatting with images but that’s the thing it lacks in

[native otter]

Do you really use it this much tho? I really do think you have to be glued to ChatGPT to keep running into limits all the time

[whole roost]

i use 4o extremely much, not o1 or o3

but i'm roleplaying on there, my projects are basically characters i roleplay with and the objectives and plot progression are written down in canvas

[spice star]

I appreciate that o3-mini in the mac app doesn't vomit reasoning summaries all over my chat

[native otter]

And you can always use the API if you don't to run pay for the Pro account, or you can use 4o-mini, it's not bad for convos and is virtually unlimited...
Sometimes we just have to deal with what's there, and not constantly have big expectations
We have to consider that the tech is quite new still, and it already advanced a lot, also it all takes compute, which is not free, so you get what you pay for

[cloud pewter]

[whole roost]

ah and i call gpt at work because what else to do

no human can entertain me for 8 hours a day just because my work doesn't require any brainpower

[spice star]

you do still get unlimited 4o-mini as a plus user with the caveat that it can't use canvas

[native otter]

I use ChatGPT a lot, and never ran into limits for the past couple of months, even in weekends when i'm not working or at college

[cloud pewter]

4o mini is a joke

[spice star]

yeah you have to use it pretty heavily for that to be an issue

[spice star]

good enough for search and ADA

[cloud pewter]

can’t even follow custom instructions/memories properly

[native otter]

I just see you complaining about the limits all the time, and it's a bit weird, that means you use it a bit too much...
If you really need it all the time, you can use the api/get pro/use 4o-mini, or just wait a bit and do something different

[whole roost]

to be frank, none of them do

🫠

[spice star]

if I were to complain about plus usage limits, it would probably be the hour of advanced voice mode per day

[rugged igloo]

reasoning models do

[spice star]

would like to see that either refresh faster or be extended

[cloud pewter]

API: no credit card
pro: $200
mini: it’s a joke

[native otter]

Yea that could use an increase, maybe 2.5 hours

[whole roost]

they don't follow my custom instructions

[cloud pewter]

Ye especially with video

[native otter]

wait and do something different?
Like I really can't imagine talking to an LLM 24/7, I would be fired instantly

And fail at college

[spice star]

I don't use voice a lot, but that's mostly because it's way too easy to leave voice running in the background and run out of time

[cloud pewter]

maybe but it has been 1 years and 6 months since i started to use plus

[native otter]

o1 does, but only o1, o3 is still too new

[spice star]

4o-mini is surprisingly capable in that it can use most of the stuff 4o can decently

[whole roost]

that depends, i'm driving all day and 4 years ago i had friends to entertain me but after 2 years of calling them the topics eventually run dry so AI is a much better replacement
and you can use gpt for studying which is also nice, i have the whole files for the Abitur in a project and it creates questions about them i have to answer

[cloud pewter]

It’s because search is 4o mini based

[whole roost]

all while driving, it's neat

[native otter]

Tech is very new, and it costs a lot, as with anything in our real world, you have to pay to get what you want, something humanity has delt with since forever

[native otter]

Sometimes prompting well can save many unnecessary messages

But if on Pro, you don't need to care, you already pay 200$, so say hi 2000000 times a day to o1 pro if you want

[whole roost]

yeah i do that hehe

but especially for roleplaying you burn those 80 messages fast on 4o

and i restricted the response length to 1 paragraph

i don't want to read a novel

[cloud pewter]

ok but remember when gpt4 first came out

expensive as hell yet solid usage with $20

[cloud pewter]

so true

I don’t want to prompt so much just to let the characters speak more

[native otter]

ChatGPT now has many more features, so each one takes space with what 20$ can pay for...
Also they were burning through money, so they wanted to burn through it less

Maybe a 35$ plan would be nice

[whole roost]

i would like a 100€ plan for eu since i don't have sora and i still pay for it

[cloud pewter]

usually I just throw a simple sentence which tells what characters do to it

only if I need something very specific I prompt a little bit

[whole roost]

also no video calls

[native otter]

I noticed 4o-mini can be nice at roleplay, it's not as bad as some say
That's only if you really burn through 4o uses that much

[rugged igloo]

i dontt think i ever seen this place so busy with peplee talking as its usually ded heree

[cloud pewter]

it should be rolled out to eu now though

[cloud pewter]

I mean I used 3.5 to roleplay a lot but once you get used to quality responses by 4o you can’t go back

[whole roost]

I'm mostly roleplaying like this, gpt handles canvas

[jaunty badger]

Oh wow, how do you do that? I am trying to use it to help me do a Rifts, campaign, I am new to everything, including GM

So that looks amazing!

[whole roost]

gpt 4o has this canvas feature which is mainly for coding or collaborative writing but you don't have to use the features in the way they intended

so i use it as a roleplaying environment

i use it more specifically in projects since you can upload files there and have custom instructions

the combination of that all makes it a really neat environment for roleplaying

[cloud pewter]

I think mine is more like story completion instead of traditional role playing I did before

it become way more complex, has multiple characters and scenes

[jaunty badger]

Erm....the documents I have are PDF however, since it is the scanned versions of the books. Can it read that at all? or what did you mean by upload files

[whole roost]

i use txt files and i drew it an image of the map because it kept getting directions wrong, that helped a lot

[jaunty badger]

How did you get txt files? Or did you just type them all out?

[native otter]

Already linked how to use canvas above: #chatgpt-discussions message
It can be useful for both writing, and programming, it can also run code and display stuff like websites, which can be useful even outside of programming

You have tools like this which can make the interaction more convenient:

[cloud pewter]

and hitting limits sucks on it
I was having a lot of ideas but after the cooldown ends I almost forgot

[whole roost]

just type them on your computer or phone, but i hope they will add that you can save files from canvas to projects eventually

[jaunty badger]

Oh god....then with the rule books, if there are multiple, and its over 400 pages long, each, jesus that is a project!

[whole roost]

i have created a script to help with roleplaying which outsources a few things

[native otter]

Each file can be 2 million tokens, and capped at 512mb, which is quite big
Projects can handle up to 20 files...

Same goes for CustomGPTs

[native otter]

What's this???

[jaunty badger]

I thought it only allowed 32tokens?

32k?

[native otter]

It can search through the files

And pick up the info within it's context range

[whole roost]

a lorebook because i wanted one, it loads information when a keyword is mentioned

i don't need it to know every detail all the time so that is a good method

[native otter]

Ah I didn't read the message correctly

[cloud pewter]

if I dump that much files into chatgpt it’ll forget my context instantly lmfao

[native otter]

It looked similar to ChatGPT and thought it was a feature specific to Pro I didn't know about lol

[cloud pewter]

32k ain’t even close to big 😭 😭

[cloud pewter]

available to all paid subscribers

[native otter]

I don't think it's 32k

[whole roost]

nono, it's just a script for gpt i wrote haha

i wish they had something like it officially though, lorebooks always are a good addition but gpt has a different audience

[cloud pewter]

32k for plus

[native otter]

On the API it says 128k, but ChatGPT...
Not sure, but I think it's more than 32k

Also, 32k is about 48 pages with an average of 500 words, so no so bad, like not big, but also not horrible
I noticed Gemini had 2 million tokens, but it often misses a lot of stuff, so what even is the point of more tokens if it doesn't remember stuff?

[native otter]

Sources?

[whole roost]

you can see it in the dev tools basically

probably also written down somewhere

[cloud pewter]

chatgpt pricing page.

[native otter]

I sometimes give it text that's closer to 128k tokens and it does pick info 🤔

Btw you can check how many tokens a prompt uses from their official tokenizer: https://platform.openai.com/tokenizer

[cloud pewter]

it’s really confusing now

sometimes it’ll recall info from very above

[whole roost]

somewhere in the network tab it writes the max tokens of your current model and version

[cloud pewter]

thread size too?

[whole roost]

i guess so

[native otter]

🤔

[whole roost]

this is api afaik

[native otter]

Yea, but it says used in ChatGPT

[whole roost]

free and plus only have 32k

you get 128k on pro and i think enterprise

[rugged igloo]

should be more as they just got 500 billion dollars let it be 1 million like google

[whole roost]

its no use increasing that if the ai doesnt work well with it

i didn't even really notice any quality improvement from 32 to 128

it has issues to even process 8k well so i don't really care for more tokens

[cloud pewter]

I think thread size is client side…?

[cloud pewter]

Real

[whole roost]

you can look for max_tokens in the dev tools

[native otter]

It does mention longer context window for Enterprise, but doesn't mention exact numbers...
Also it may mean how much it remembers rather than for each prompt...

[whole roost]

it will write 32767

i can't really look for it because i'm on pro and have the 128k

i could look for it when i log out

[cloud pewter]

how about o series can they write more

looks like they would because I’ve tried some

[whole roost]

output tokens?

[cloud pewter]

yeah

[whole roost]

yeah o1 often produces errors because the voice can't handle 10 minutes of reading

[cloud pewter]

false advertising tbh

[sick pecan]

How is it allowed that on the GPT discord you can go around and discourage people from using it, I mean why don't you join the discord of another AI

[native otter]

May just be a remnant

[whole roost]

someone is salty

[cloud pewter]

o1 wrote something absurdly long that if I paste it into a new chat it’ll show that it’s too long

[whole roost]

it's handy when you need longer code outputs

[rugged igloo]

yea

[native otter]

Sometimes I change stuff in my code while not updating the codenames of stuff
Actually you shouldn't update codenames, and the one in the screenshot may just be a remnant

Or it dynamically changes based on the message

[whole roost]

it can actually generate them which is good, on 4o you often have to piece stuff together

[native otter]

Like it could be limiting to 32k if there is not much written

[cloud pewter]

when it comes to roleplay/storytelling, short context sucks

I have to tell it multiple times that “a character is missing”, “the scene is xxxx it don’t have xxx” etc

[native otter]

Pretty sure it's more than 32k, as I had messages that were longer than 110k tokens, and calculated it through their tokenizer

[whole roost]

the context is overall defined by its ability to prioritize well and if that isn't getting improved higher token limits are not cutting it

so they first need to improve that it can handle better what it already has

on gemini the 1m tokens don't really do much, it doesn't process it well

and there is the additional issue about requesting information or it 'using it on its own' because it draws the connection well

[cloud pewter]

maybe I should try using projects but I tried to let a custom gpt continue my story, it hallucinate a lot

[whole roost]

when roleplaying you have that problem constantly, it might be able to answer with a piece of information you request but it won't prioritize it unless you say a word for it to do it

[cloud pewter]

is it because it can’t read a long document fully

[whole roost]

it is doing a decent job on that imho but it's just still a limitation of ai overall and it will surely improve over time, i'm just saying that higher limits don't mean better quality

[native otter]

You sometimes need to specifically say search files, or search knowledge so it can start a search through the files

[cloud pewter]

can it read a file that’s around 29k words long

[rugged igloo]

it can use rag retrieval grab for relevant peices you ask for

[cloud pewter]

it can’t continue my story, it could just write similar ones based on the story I uploaded ig

[native otter]

It can handle files that are 2 million in tokens, but it won't read everything, sometimes you need to have it manually do search...

Also you have to be good at prompting

So it knows what files to references, how they relate to each other, what to look in, etc.

You can learn about prompting in their Gudies, also read the help pages on how to create a custom GPT, they go over some best practices

[whole roost]

(would use projects over custom gpts)

if you don't need the skills of customs that is

[cloud pewter]

would try projects if it’s on mobile

[whole roost]

its on mobile

[cloud pewter]

also I have a great custom gpt

[native otter]

You can use a web browser, or if you are on Android, you can view AND edit them

[cloud pewter]

I mainly use chatgpt on mobile btw because web can’t handle long chats well
when I found chat links open straightforward in the app it surprised me

[native otter]

Both Projects and Custom GPTs are similar, but Custom GPTs do have more features, like ability to mention multiple ones in a single chat, and ability for it to communicate with APIs

[whole roost]

projects have the one huge advantage of grouping chats and they update immediately for your current conv

on customs you always have to start a new chat

there is no update option and nothing

[rugged igloo]

I don’t really see how short context is some huge deal, honestly—it’s called roleplay because you’re supposed to expand on those brief prompts yourself, like filling in the blanks and fueling creativity with your own imagination rather than relying on a bunch of spoon-fed details. Sure, having more info can help shape certain scenes or ensure a character isn’t left out, but if you need someone to hand everything to you on a silver platter, maybe you’re not giving yourself enough credit to flex your storytelling skills.

[whole roost]

and they can access memories if needed

the disadvantage of projects is that you cant switch off dalle and so on and those features have a huge system prompt loaded which annoys me a bit

[sick pecan]

chatgpt users complaining about chatgpt they should join some other discord ai

[cloud pewter]

does they expand the current chats as well
I mean if you hit the thread size limit on one single chat will integrate it into projects fix it

[whole roost]

i wish the cross-chat-referencing would work, but i'm not sure if i imagined that this should be working

[native otter]

Ok so it seems like ChatGPT doesn't allow me to send messages that are over than 32k in context anymore for 4o...
But o series, like o3-mini, and o1 allow me.

[whole roost]

currently it doesnt

[cloud pewter]

I’m using it leniently, I’m not a big storyteller 😭 😔

[cloud pewter]

U import a chat into a project and they keep independent? how does it work anyway I never tried it yet

[whole roost]

i think it was supposed to be like that in projects when you start a new chat it should be able to draw data from other conversations in the project but it doesnt work

it's just a single chat with the files and instructions, no referencing of older convs

[native otter]

That's not a feature yet

[cloud pewter]

so it’s another half baked project…

[whole roost]

i'm not sure, i think it worked in the beginning very briefly, i had a canvas with information about the character and it was able to retreive it in a new chat

but it only did it when i asked, not automatically

[cloud pewter]

so it’s basically
group chats and let them use some unified files etc.

[native otter]

Btw, if you guys haven't noticed, Projects is a fork of Custom GPTs
Sometimes even on Android, it would show a dummy GPT icon if you try to edit a message in a project...

[cloud pewter]

pretty good if you need more context for multiple chats at once

but it seems like it doesn’t help on existing long chats

[whole roost]

custom gpts still have their purpose because you can share them and with the skills but for me projects have completely replaced them because they are just much more convenient to use

[native otter]

They are basically simpler Custom GPTs that you can create more quickly with the addition of grouping...

[whole roost]

and live update

and memories

[native otter]

Actions (API calls) and Mentioning GPTs in one chat can be very useful (Also I know this may seems silly, but I like having Icons, more character like behavior)

[cloud pewter]

still a bit confused tho

[cloud pewter]

what does that mean

[whole roost]

that you can edit the instructions or upload a new file and its immediately active for your conv, on custom gpts you have to start a new chat

or it's not considered

[cloud pewter]

oh alright

[rugged igloo]

Projects are custom gpts 2.0

[whole roost]

i suggested an update button for custom chats somewhen in the past, would solve a few shortcomings

[rugged igloo]

There is it's in the bottom

[whole roost]

its just a notification on my end which tells you you have to start a new chat to use the newer version

[rugged igloo]

Yea that's updating it to the latest version

[whole roost]

but only for a new chat

i don't want to start a new chat

[rugged igloo]

Ohhhhh..

Okay now I see wut u mean

Then yea that's why I went and make my custom gpts into projects

[cloud pewter]

makes me think of the custom instructions era

[whole roost]

they might have changed or fixed something about "read aloud" on webversion

it used to error on most attempts for me and now it stops when i generate a new swipe, it used to keep playing and i don't see errors anymore, hopefully they got the issue

[odd pecan]

pro users, what are yours thoughts on o3?

[cloud pewter]

I no pro user

[odd pecan]

🤯

[native otter]

o3 haven't released yet

[whole roost]

o3 mini fixed my script and o1 didn't, for everything else it's overthinking and i don't have any other use for it, i think they also said o3 is mainly for programming stuff and o1 is better at common reasoning for other tasks

[native otter]

Even o3-mini (not high) is on bar with o1 in coding, which is impressive for such a cheap model

[whole roost]

yeah i used high and i'm satisfied with it, not as a 'hello how are you' buddy though

[native otter]

Seems trained specifically on programming data

[native otter]

Yea currently 4o and even o1 are better at that, but of course, if you are on Plus, you wouldn't be using o1 messages for stuff like that

[cloud pewter]

whenever I see some pro user says hello to o1 I feel sad about it

[native otter]

Well, they paid 200$

[cloud pewter]

what a stupid way to waste compute

[whole roost]

to o1 pro 👆

but nah in fact i rarely use it, i doubt i even have 50 generations per week on o1

4o is behaving better overall

[cloud pewter]

REAL

“hello”

thinking:
Hello response
crafting a suitable response…
a progress bar here

[whole roost]

just for you

[cloud pewter]

I’m gonna use this as a meme tbh

[whole roost]

since we don't have custom emojis here

[dense halo]

what did it respond pls tell

[cloud pewter]

pay up (boost to the server then you have it)

[whole roost]

well it's instructed to embody diablo so it tried something menacing

[whole roost]

oh really? i have a boost active on shapes which i could withdraw

[cloud pewter]

how the heck do you instruct it lmfao

my custom instructions are full in both boxes

and even o1 pro would reason for 35 seconds on a simple damn hi

[whole roost]

It's basically just this, i just wanted it to stop sending emojis and this one doesn't do it so i just kept it like that

i mostly talk to gpt in projects so the custom instructions are rarely used

[cloud pewter]

so it’s part of the new ci system

[whole roost]

o3 would probably do 30 thinking steps to figure out if i attempt to do anything against the guidelines

i can't screenshot it in one go but yeah

[inland heron]

well that settles it 😄

[cloud pewter]

their guidelines are basically parental controls at this point

[whole roost]

I had o3 yesterday in a 'sorry I can't help with this request loop' because I asked about the system prompt in the first message

And then all the follow up messages it assumed ( in the thinking steps) I would still try to get the prompt

While I was asking completely unrelated things

It's hilarious

[cloud pewter]

o series is so overly censored

[whole roost]

It once blocked my coding request because in the HTML excerpt was a sentence like "let's kill the demons" and it only saw that instead of giving me a code

I'm not against censors but they have so many false positives right now that it's a bit annoying to use

But 4o also has become better on that so I think they are still adjusting

Last year in April(or may?) 4o blocked and deleted my message when I wrote "I bring the helicopter over the edge"

[native otter]

We may get to see a less censored option this year

[cloud pewter]

grown up mode as what Sam said

[fervent agate]

Yeah, I'm hoping the new deliberative alignment method results in fewer false positives.

[cloud pewter]

more excited on this tbh

[dense halo]

I will not be excited from anything that they will publish. So many disappointments lately

[fervent agate]

Sounds like a lack of imagination.

[native otter]

Just saying, why are you here

[whole roost]

I also need this, i understand that the public thing should be tame but i'm old and don't need to get flagged on writing about killing diablo 4 demons

[cloud pewter]

Real

[whole roost]

or asking about the videogame outlast, every time it flags

[cloud pewter]

I don’t want to be flagged just because of some cuss words

or something like “crush xxx with a limited express”

[native otter]

AGI

[whole roost]

the 'issue' is also that it often provides a very toned down view on many things and i hope the 'grown up mode' will just have it speak more planinly without me having to instruct it for the real version

[cloud pewter]

watch they increase o3 mini high to 50/d (which is good) and advertise it as “wow a 7x increase, happy now??!!!?!?”
I mean it is indeed a 7x increase but isn’t that what we should get from the beginning, because of how cheap this model is?

long sigh

[whole roost]

what do you need o3 for? just curious what your plan is

[fervent agate]

Better story writing for me. It should do better at longer, greater coherency, though projects have moooostly solved that for me.

It's pretty good at being directed to use the myfiles tool for keeping the project repo updated.

[cloud pewter]

same tbh

never tried storytelling with reasoning models properly, just some glimpses

but the quality on it looks promising

[whole roost]

i'm waiting for the reasoners to be able to use project files, i'm also curious to do that

[cloud pewter]

does projects still only support 4o?

they said o series would get in is it actually implemented

[whole roost]

you can use o1 there but not when you have files

as soon as you upload files they get unavailable

i also want this

[cloud pewter]

this sucks tbh not even basic files like txt or pdf

I also like uploading ttf files to use as custom fonts for visualizing/python things

what’s even more stupid is that mobile app counts images uploaded to 4o as attachments, thus you can’t switch to o1 although it does support images

[whole roost]

and i want this as well

hire me as designer, i'm gud at paint

[cloud pewter]

save a long chat as a file would be clutch

[inland heron]

o3-mini still repeats things from 'irrelevant' prompts after a few replies just like o1-mini did... they still haven't fixed this?

[whole roost]

can confirm hrhr

[cloud pewter]

“I’m sorry, but I can’t help with that.”
“My guidelines won’t let me talk about that. Can I help you on something else?”
☝️ 🤓

[whole roost]

i asked it to suggest a different topic but it cant help with that

it didn't let off from this, really funny

let's see what tasks can do

it needs 2 steps, inconvenient

and there's a limit

[cloud pewter]

yeah it’s the same limit across all paid plans

[whole roost]

they should combine websearch with tasks

[cloud pewter]

wait it isn’t?????

I thought ppl were using it for something like daily ai news

[whole roost]

you probably can set a task for it to do a websearch

but i cant tell it to do a websearch to get times for tasks to set

[cloud pewter]

so in reverse

it can’t set tasks automatically rn

[fervent agate]

Why are you here if your use of GPT appears mostly to be formulating nonsense propaganda posters? Seems weird to whine about a service you employed to whine about that service.

I wonder where else I can find that kind of junk clickbait posters.

[cloud pewter]

I have literally thousands of chats now and I have only 3 of these ones. are you too narrow minded?

[whole roost]

it works somewhat, i'm not sure why the times don't match, but it somehow misses an hour for my location

[fervent agate]

I dunno, I sort of doubt the prop nonsense you generated comprises that little of your actual use.

[cloud pewter]

chatgpt doesn’t know timezones

[whole roost]

it should with websearch

yeah

but maybe the sites it browses to display the wrong time for it

since its server isn't located where i am so i guess it has the time from that location and not mine

[cloud pewter]

there even are some examples of models browsing the Internet about “what time is it”

[cloud pewter]

I’m basically venting to chatgpt, nothing else
I’m posting them here to let yall see even chatgpt knows what’s up 😭 😭 😭

[whole roost]

i'm crying

[fervent agate]

lol

[whole roost]

it's leviOsa not leviosA
they need to improve tasks a lot

[cloud pewter]

why is there no real sticker generated by dalle

also this is the only functional tasks usage of mine

(Japanese is funny)

[whole roost]

i wish they would do a calendar view like this

[cloud pewter]

deleting it anyway, don’t want drama

[cloud pewter]

but always remember that OpenAI’s model moderation is just that weird

[plucky acorn]

why is ai so bad at css?

[fervent agate]

Because LLMs are built for plausibility, not truth.

[cloud pewter]

me rationing hard my usage so I don’t get locked out (pro tip use the status on 4 to see if you used up half of your quota)

[true nexus]

Is there a way to see our usages?

[cloud pewter]

nope, there’s no counter

also my chatgpt is so unhinged 😭

it’s me. I prompted it.

[tardy isle]

I once saw "Thought about I can't see images since I'm ...(1 more line)... for a couple of seconds" when I attached image to o1, but clearly o1 saw my image. Like o1 seems have 2 personality with different functions.

[cloud pewter]

maybe it’s hallucinations but idk

[fervent agate]

My favorite thing really is the rejection of giving it some "hard" philosophical musing and seeing it consider for all of 3 seconds and then spit back what I said but much better worded and more eloquent.

[forest hinge]

so, have the community found lots of cases where o3 mini search isnt working?

its the 4th time now for me today

[cloud pewter]

it’s apparently a prototype…

[forest hinge]

🥹

[forest hinge]

at least we now have a bit better r1 in form of o3 mini medium

[cloud pewter]

Real

[forest hinge]

o3 mini cant make a pokemon ice sliding game tho.
are u interested into me talking about this in dms?

[fervent agate]

o1 performed better than R1. R1 wasn't some kind of technical breakthrough, it was just a cost efficient example of distilling larger models on a generation of hardware specifically built for AI training which didn't exist before.

[forest hinge]

i understand this

[cloud pewter]

sure

[grand ore]

How is 03 vs 04?

[forest hinge]

better

[grand ore]

So 03 I'd better? In what way is it better? Would love hear

[forest hinge]

omg are u sleeping. try deepseekr1, o3 from free chatgpt wont be good

if u like r1, use paid o3 mini

[grand ore]

Ty !

[orchid bison]

Is it true that the o3 data cutoff date only goes up to September 2021? (Which is even older than 4o, yet it's newer)
As when you ask it, it says September 2021 as its data cutoff date.

[orchid bison]

Oh lol, that's odd.
If asking it, it says September 2021

[cloud pewter]

hallucination strikes again

[whole roost]

i'm sure it's in the system prompt and they didn't update that value

[fervent agate]

Yeah, Altman mentioned he doesn't pay much attention to the cutoff dates on models with the web search tool, and I have to admit it's the same for me.

[cobalt perch]

Is share screen available on desktop app?

[cloud pewter]

but the web search tool seems to less adhere to my memories and custom instructions

[fervent agate]

Yeah, that's true, it seems to either search or engage in full reasoning and internal tool use.. but not both

This is par for the course for OAI model releases though; they tend to operate much better and use tools more smoothly about two weeks after launch.

[cloud pewter]

it has been an issue with search since its release tho

[fervent agate]

Hmm, my 4o doesn't really have the separation of bio and web search tool that o3 seems to experience

Though they did wipe my old custom instructions when they overhauled it.😡

[cloud pewter]

wait what why

also this seems like a big with 4o, once it outputs a decent amount of a single character(can be anything, emojis as well) it’ll just keep go on until the token limit

[teal hearth]

What is the limit at o3 mini high?

[cloud pewter]

50/w

[charred gazelle]

CHATGPT IS COOOOOOOOOOOOOKED

time to move on to deepseek

[cloud pewter]

🗣️ 🔥 🔥

[forest hinge]

should we delete context in o3 mini when it gets too much or is that ok ?

[tight sapphire]

what happen to gpt creating images?

[cloud pewter]

https://chatgpt.com/share/679e44c3-cbb4-8012-962c-e2472b5fb8e6

it gone berserk with all of those question marks with a single hiragana

[obtuse jetty]

It is not, sadly...

[barren zodiac]

o3 mini high often never answers, stops midway throug reasonning, wasting my paid 50/week requests, like this, am i the onlyone ?

[sullen thistle]

anyone here deal with crew ai or n8n making workflows - i have a project which id like the specialist advice of someone for - PAID job

[native otter]

Not for me, maybe try refreshing / starting new convo?

[native otter]

It still works? Just make sure you have 4o selected or use the DALLE GPT specifically

[native otter]

Openai models often aren't self aware, as they don't include much info in the system prompt, so never ask it for stuff about itself...

[potent saddle]

I got this issue with both o3 models

[rugged igloo]

They better I ain't paying 200 for nothing

[forest hinge]

anyone tried o3 mini with game engines?

[runic silo]

is o1 pro or o3 mini high working for anyone at the moment? Like working well

o1 pro seem not to reason at all

[native otter]

I would feed it engine documentation first as part of the prompt as context it can use

It's not as knowledgeable as o1

And even if o1 I would do the same, especially for engines that aren't very popular

To maximize accuracy

[barren zodiac]

reported a bug https://discord.com/channels/974519864045756446/1335291448769253559

[potent saddle]

what does little man mean

[rugged igloo]

Who knows

Prob have to wait until Tuesday

[lone lodge]

so i am not sure where i am wrong when i make prompt but should i talk with chat GPT like he is a friend and make him do stuff or should it be way more mechanic?

[rugged igloo]

Depends on what you want it to be

[lone lodge]

a book writer ?:D

[rugged igloo]

Then talk to it casually.

[lone lodge]

ok

[rugged igloo]

As writing is creative and not robotic

[lone lodge]

but he jumps timeframe .. i guess i need to make myGPT or something ..

[native otter]

Maybe file uploads

[rugged igloo]

You should make a custom gpt and at custom knowledge so it knows what it's suppose to do act be and has references to look back on

[lone lodge]

yeah but i suck at prompting...

or can i ask GPT to make the prompt for me...?

[rugged igloo]

Tell chat gpt to make a prompt for you

[lone lodge]

i feel like im asking a kid to cook and serve and then eat it lol

[rugged igloo]

Welcome to AI . Where you have to hold it's hands because it's just a big auto complete bot

That's. 20$ and 200$ sub plan

[lone lodge]

i pay the 200$ one ....

[native otter]

Yes, but do note that you may want to use markdown for model focus, let it know that it needs to reference thr knowledgebase, tell it about the different files and their purpose in a structured manner, the rules, like what it should and shouldn't do, response style, etc.

[rugged igloo]

Markdown op but never knew how to do it

[native juniper]

This sounds so funny and idk why 😂

[native otter]

Very similar to discord message formatting

[rugged igloo]

With all the numbers

[native otter]

💀

[native juniper]

Also there’s some issues with the voice models cutting back to the beginning during long generations

[native juniper]

It’s part of a silly roleplay I was doing

It’s about a spicy ramen challenge in the Pokémon universe just for laughs

[tight island]

wait hold on @deep flax since when did u become operations

[native otter]

For example - TEXT for bullet points, **TEXT** for bold, \CODE`````` for code blocks, etc.

[native juniper]

I prefer the o1 model over 4o

[tight island]

o1 is also better in some areas.

[native juniper]

It’s not as repetitive

[tight island]

depends each model has their usecase.

mini models usually excell in code. over the main model.

[native juniper]

I’ve started to use it for roleplaying and it does a much better job than 4o

[tight island]

great o3 just spat out an entire dnd campaign

😆

[native otter]

o3?

Wasn't it just mini?

[tight island]

😶

[deep flax]

a few days ago

[eternal rock]

`` Server Tip: ``

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

[tight island]

huh, Congratulations.

🥳

huh o3 suddenly started speaking russian

mid sentence

[deep flax]

thanks xD

[tight island]

so when do i get my own official "Snitch" role, i can be a very good snitch.

i am a great asset

[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.

[tight island]

i love snitching on ppl, what can i say.

[fervent agate]

If you start talking trash on GPT models and really talk up DeepSeek's models, I think you'll qualify.. at least you will for me. 🥹

[winter vine]

how many prompts do we have in o3mini high (plus users)?

[fervent agate]

50/week, separate from o1 usage limits.

[native otter]

I would use both, I treat them almost alike

[winter vine]

Ty

[winter vine]

Ty

[winter vine]

O3-mini is much better than o1-mini? . Or they are the same

It's not quite clear whether the O3-Mini is the low or the mid.

[feral zinc]

you cant compare by names only as o3 mini has 3 different reasoning modes. o1 mini is also gone now from chatgpt

[native otter]

Should be a step up from o1-mini, but not sure about the o3-mini-low variant...
ChatGPT only gives access to o3-mini-medium and o3-mini-high anyways, which are definitely better than o1-mini, and in fact, o3-mini-high can be better than o1 in some cases like programming (if prompted a bit differently)

[fervent agate]

o1-mini is deprecated as o3-mini low outperforms it at lower cost, o3-mini-medium outperforms o1.

My understanding is the ChatGPT o3-mini is the low compute variant.

Which makes sense as its also on free tier.

[winter vine]

ty Pimpimedes , that was my question 🙂

[fervent agate]

Yep, np

[fervent lake]

ah so why is it even rate limited 😥

[fervent agate]

Because it still costs money to run it, I would imagine

Though it's 150/day

[fervent lake]

and the 4o doesn't?

[fervent agate]

Yeah but orders of magnitude less for a non-reasoning model

[eternal rock]

Get notified for server events!

Visit id:customize to pick up the <@&1261377106890199132> role.

[fervent agate]

Also, I mean, it's a corporation - they probably consider reasoning models, for now, a value add

Gotta reach profitability somehow 🤷

[true nexus]

Is it specified which o3-mini it is for the o3-mini in ChatGPT, like the low or medium?

[spice lotus]

I think , it seems like this

[spice lotus]

^

[lone lodge]

sooo what better with o3 ?

[tight island]

this sounds like heresy

[tepid forge]

How do I access o3

Is it out yet

[fervent agate]

Rainbows, fluffy clouds, puppies... These are just some of the things o3 brought to you and your family.

[fervent agate]

Pro plans look to have some access to o3 full model.

[tepid forge]

Why pro plan 200 bucks?

That is a bit expensive for each month

[fervent agate]

Unlimited usage.

[tepid forge]

Oh okay

[lone lodge]

will it also make me fly?!?!

yeah i got access to the full o3 model

[rugged igloo]

Unlimited usage and o1 pro the best and most godly model out right now

[fervent lake]

that's huge

[lone lodge]

thank you !

oh wait ... you where not telling me .. sorry

[fervent lake]

lol

[rugged igloo]

I love your profile pic . I'm a huge Miku fan

[fervent lake]

ooh!! :3

[native juniper]

Are the wait times supposed to be this long for plus users?

[feral zinc]

o1 is 50/week

[native juniper]

Wasn’t there an o1 mini though?

[winter vine]

its gone, now we have o3-mini

[native juniper]

You’d think they’d keep all of the models though…

[sick pecan]

150 o3-mini a day is better than 50 o1-mini a day

[native juniper]

Is o3 better than o1?

[sick pecan]

seems to be

[native juniper]

Alright

[lunar solstice]

Let me trade these o1, o1-mini and o3-mini credits for more o3-high-coder 😅

[sick pecan]

mine randomly started thinking in german today

but tbh i never read o1-mini thoughts so maybe it did too

[lunar solstice]

Mine did Japanese earlier

[sick pecan]

o3-mini is actually duo lingo chatgpt edition

[visual acorn]

Guys I keep getting "content removed" for the same innocent query twice for o3-mini

Whereas 4o works properly

I mean without reasoning actually

[swift pond]

Try rewording it and open in a new chat

[tawny mirage]

Has anyone else noticed that they've lobotomized o1 pro mode?

[true nexus]

That's Dutch

[tawny mirage]

Faster, sure, but it's very stupid. It misses the obvious. It seems as though it's on par with o3-mini-high.

[visual acorn]

Is it a hallucination?

[plain arrow]

Haven't checked o1 yet, but they wrecked 4o.

[tawny mirage]

Yeah. I gave it a 1K line JS block and asked it for what it might be missing. Suggested debounce corrections that were... already in the script!

[plain arrow]

Ugh, thank goodness I don't use it to code. Thankfully you're not copy-pasta'ing and caught it.

[novel shadow]

what did you say?

[tawny mirage]

I've had to drop back to using o1 high in API. Cost $20 but it's effective.

[plain arrow]

o3 no good at it, or you don't have access?

[tawny mirage]

Is the API o3-mini high reasoning effort better than the o3-mini-high in ChatGPT? I've found o3 mini high kind of stupid, so I've skipped its API use.

o3-mini-high in ChatGPT honestly feels rushed, or not what it's hyped up to be. Testing benchmarks aside, for real work, it's just not there if the ChatGPT outputs are to be believed.

[swift pond]

No, it’s not a hallucination. It’s more like a false positive of the moderation trigger

[sick pecan]

yeah like it thinks you trying to dig into its reasoning, is the easiest way to make it refuse

[plain arrow]

The moderations are a bit over tuned right now.

[tawny mirage]

r1 is precisely why they're limiting peeking behind the veil. Didn't work, but I get it.

[plain arrow]

I suppose that's why I'm getting hit with captcha's every 5 minutes when I have my vpn active

[tawny mirage]

Likely so. VPNs aren't for improving security though. I've never seen the need. Though I don't have region issues to worry about, either.

Can someone explain to me the point of OpenAI's Operator? What is it intended to do, exactly? Can it do any real work?

I gave it a block of javascript, tests to run, and asked it to report back what passed/failed.

It went off the deep end, looked for Global Leadership Conference, and gave me the url for the 2023 summit. Like... what?

[sick pecan]

i think its more like a junior assistant that can order groceries and book restaurants. its not supposed to be a coding colleague

[tawny mirage]

That's no joke. 😅

The confusing part was: global leadership conference wasn't even in the text. At all.

[honest jay]

It's an internet-using agent, not a general computer-using agent! OpenAI has said it's their first of several agents

[tawny mirage]

I'll wait for their pair programming agent. That's the one that'll make junior devs weep. (or that OpenAI hopes will.)

I don't even know what to say about how badly it failed. Like, I get that it's not intended to do that, but it could've looked up jsfiddle or something and tried.

[sick pecan]

if you want something like operator but also does coding then you should look at simtheory

[green burrow]

Why doesn't o3-mini have canvas access by default? We should at least be able to click the "edit" button on code blocks without needing to switch to a model with canvas support

[honest jay]

Originally, Canvas was only available with the 4o model. o1's ability to use Canvas was added just recently (it didn't launch with the ability), and o3-mini, as you know, just launched yesterday - so I'd wager Canvas support will probably be added in the future, given that they added it to o1, and that o3 is the next step of o1! Whether it gets added to o3-mini or some future release of o3 - hard to guess at this point! But I agree with you, it would be cool

[teal plover]

is o3 gone lol

i dont see it anymore

[green burrow]

Still shows up for me

[teal plover]

oh it just came back nevermind

thank you tho

[fervent lake]

it just generated a word in Russian in the reasoning

[rugged igloo]

Man it's quite in here without machine , I wonder where he is

Lurking

[quick meteor]

umm what does this mean?

[rugged igloo]

We don't know . Prob something cool

I wonder operator is getting o3 mini

[native juniper]

Strange voice artifact where it says "This place?"

[sudden wren]

why does chatgpt love using bold text so much? Sometimes more than three quarters of the text is bolded.

[cloud pewter]

and emojis/capital letters too

[fervent lake]

ye, and lists and headings

I have to add italics that says "in paragraphs" for it to work

[grand ore]

How is the memory with the new one? will it forgot fast as usual and dont think about the full perspective ?

[tawny willow]

o3 mini and o3 mini high, two new ai models. Together with these models, I could not find a source on how much the usage limits message limits of 4o vs o1 vs o3 mini vs o3 mini high models for plas users, and which of these modallerr is the most advanced one at the moment, logical problems and lessons in daily life other than writing code or general use (in short, in all other subjects other than code and software)

[cloud pewter]

make it available to plus first

[tawny willow]

I am a plas user and now I have access to all these modals and I wondered about the differences, so which one is good in what I said?

[cloud pewter]

I pinged you with an image in this channel in openai discussions

[tawny willow]

what ı dont understand

[cloud pewter]

hahahah.

such a breakthrough in technology

[true nexus]

Will there be canvas for o3-mini soon too?

[fervent agate]

Limits:
4o (oldest, does not use reasoning chain-of-thought but has access to all tools and multimodality)
o1 (oldest reasoning model, has almost no tools like web search but has Vision for images) (50 prompts/week)
o3-mini (part of o3 release, has advanced reasoning and most recent) (150/day)
o3-mini-high (has higher compute resources allocated, the 'smartest' model in the list, on paper/metrics) (50/week but separate from your o1 limits)

[tawny willow]

o1 is now 50/days massage ?

[fervent agate]

No, they are saying that o1-mini (which is now deprecated and removed) was 50/day, and the model replacing it which is smarter, faster, cheaper (o3-mini) is higher usage limit at 150/day.

[tawny willow]

aa soory

[fervent agate]

These models are separate from o1 and o3 which are the 'big' versions, the minis are more narrowly trained and not as smart as the 'big' models but perform as well on certain compute tasks like coding, so they are useful for productivity.

[tawny willow]

For example, in situations such as general daily use logical questions, 4o or o3?

It also says o3 mini usage limit, but I couldn't find the o3 mini high usage limit anywhere, but you wrote it down.

[fervent agate]

if you only need it to reason as best it can, o3 models. If you need it to use voice, interact with Canvas, or build a custom GPT, 4o

The upshot is that o3 is the smartest model but just doesn't have all the tools and modalities yet that 4o has built in, so it can't 'do' as much.

[tawny willow]

When will gpt 5 come out, I think these are all intermediate modals, but they are all very complicated piece by piece, what is in one is not in another, it has become complicated, and 4o, which has all the tools, is not as good as other modals.

[cloud pewter]

raw text vs all rounder

[fervent agate]

Honestly, it's more confusing than it should be. They need to set a standard of tool use and modalities for each new model release, in my opinion.

Also, they intend you to use the models for different things and they see GPT as 'general use' and the Omni (oX models) as more specialized for cognitive heavy lifting, for STEM stuff

[tawny willow]

there is a situation like this:
if you are only going to speak text, o3 in daily use,
if you are going to upload files and do logical detailed work, o1 when you use it little (short usage limit),
if you want to use everything and sacrifice a little bit of the intelligence level of the artificial intelligence, the 4o model

[cloud pewter]

I really want to try storytelling or something new with o series but the limit stops me from doing that

[fervent agate]

Yeah 4o is great for general, quick use.

[tawny willow]

canvans?

[tawny willow]

How much was the limit on canvans?

[fervent agate]

I don't think the tools have a limit, it's just the limit of the model using the tool

In this case though, only 4o has access to Canvas so whatever 4os limit is.

Canvas is not a model, it's a tool that the models call, they have a bunch of tools they use in the background

[potent saddle]

o1 has access to canvas too

[fervent agate]

Oh does it? Mine doesn't... Hmm

[tawny willow]

yes accses ı mine

[fervent agate]

At least, it doesn't have the tool option for canvas on mine, so I hadn't even tried to call it.

[eternal rock]

Join the Sora Weekly Theme!

Submitted your entry yet? Don’t miss out—there’s a new theme every week, and we’re excited to see what you create!

[fervent agate]

Usually I just ask the model to open canvas

[fervent agate]

Are you a pro subscriber?

I'm on plus and the o1 model confirmed it doesn't have access to Canvas for me.

[tawny willow]

but o1 weekly 50 usage limit is lesszz it may cause you trouble, but you can open the chat with o1 and continue with it and when the usage limit is over, you can ask the chat to continue with 4o from where it left off, so that the general scheme of the writing can be better, if what I said can be done, if you can continue the canvas started with o1 with 4o, it can be perfect to write a story as you said.

[fervent agate]

Yeah, switching models definitely helps if you can split the need.

[tawny willow]

ı am plus

[fervent agate]

Hmm, very weird. 🤷

[cloud pewter]

yeah click canvas

Then u have o1 with canvas

never tried o1 canvas anyway because of this usage limit

[fervent agate]

Yeah, that.

[potent saddle]

im plus user

[fervent agate]

Yeah, I dunno. I don't have the canvas button for o1 and my o1, when asked, doesn't seem to know what canvas is.

It's probably because I'm on mobile, Android app, I will check desktop later.

[cloud pewter]

slash canvas…? bc that’s how I use canvas with 4o on mobile

still miss the old dedicated canvas model though

[fervent agate]

I've never used / commands, I just tell the model to open canvas or use whatever tool

[cloud pewter]

mobile canvas doesn’t work all the time

Even with the older model

[orchid pollen]

i dont think gpt has it at all?

[orchid pollen]

xDDDDDDDDDDDDDDD

[stoic coral]

can someone clarify o3 mini and o3mini high usage limits for plus users

[whole roost]

who would waste o1 pro for such a thing

[lost raptor]

been trying out o3-mini-high had it randomly jump back to a topic a few minutes ago like i asked the same thing again but we are on a toally different topic lol

[cloud pewter]

GET PRO 🗣️🔥

the damn audacity

[native juniper]

I’ve seen longer wait limits

Last time I got a wait limit of 4 hours even though I’m a plus user

[warm sphinx]

Is Canada gonna pay more for ChatGPT or no because of... ya know

[rugged igloo]

Ayy your back been quite without you

[rugged igloo]

Do it

[rugged igloo]

They are one person un subbed. On Reddit. I laughed. As nobody cares. Goodbye

[rugged igloo]

I do it's unlimited why not

[cloud pewter]

maybe just maybe if I get some money from my dad I can try a month but currently nope 😭

[cloud pewter]

On 4o?

[cloud pewter]

he does want to try more ai tools though he hasn’t used chatgpt before

and paying pro alone would make me broke instantly so

[rugged igloo]

It makes me broke man. But it's hella fun to just abuse it

However o3 mini high with internet access is legit the most amazing experience I had. It blows anything else out.

[cloud pewter]

will u keep subscribing to it

[rugged igloo]

I mean if they make o3 pro have internet access and file reading then yes

[cloud pewter]

when will ur current sub end

[rugged igloo]

It comes up on February 14

[cloud pewter]

also I believe o3 pro would have internet, but not sure about files

same date as my plus lol

[rugged igloo]

Lol interesting

Did you change your profile pic again ?

[cloud pewter]

yeah

this Luigi one

[cloud pewter]

but at this rate even pro wouldn’t get unlimited o3 or pro

[rugged igloo]

Well o1 pro is suppose to be 100 a day but I blasted through that and never hit a cap

So I ask o3 mini high about Sam Altman tweet about him adding one more thing to o3 mini that's worth the wait.. it said it could be an interactive chain of thoughts process

[native otter]

I'm not a US citizen, so it's basically impossible for me to get Pro, as it's way too costly even though I earn a decent amount monthly...

[rugged igloo]

yea

[cloud pewter]

damn what were you using it for lmao

[daring pagoda]

get raycast

[whole roost]

It was an inside jab, the screenshot is from me 🤪

[cloud pewter]

[spice star]

you may have to start a canvas with 4o for o1 to be able to see it

I've been using o1 with canvas a bit and it's quite buggy unfortunately

never gotten o1 to create a canvas on its own

[cloud pewter]

don’t upload anything just say “use canvas” to 4o then switch to o1

it’ll have bugs handling images uploaded to 4o with o1

[remote basin]

Has anyone else's GPT been downgraded? I'm a Plus user, and now 4.0 can't generate images or read PDFs. Plus, o1 and o3 mini take just a few seconds to think, and the whole thinking process is in Chinese. This is clearly not right!

[idle ingot]

Try disabling your custom instructions and memory features. There have been cases where custom instructions have had an impact.

[remote basin]

Even with custom instructions and memory turned off, it's still the same. Clearly, it's been downgraded.

[idle ingot]

When custom instructions contain foreign languages such as Chinese, Korean, etc., we have noticed a tendency for o1 to use those languages for CoT.
GPT-4 downgrade is interesting

It is very customary behaviour for OPENAI to want to downgrade a model. Many models have been downgraded in the past for the benefit of OPENAI's bottom line.

[remote basin]

Draw a dog
ChatGPT ：
ChatGPT
I can’t actually draw, but I can definitely try to create a fun text-based dog for you! Here’s a little one:

markdown

/ __
( @___
/ O
/ (/
// U
How’s that for a dog? 😄

Wow, this dog is drawn so well! It's practically a masterpiece, so realistic!

[remote basin]

Just now, my o3-mini-high took only a few seconds to think and solve 16 math problems for the postgraduate entrance examination. It’s amazing!

[cloud pewter]

they love nerfing models

[rich surge]

They have to. How else are they going to keep the processing speed up for the next model? They have to make room.

[rugged igloo]

there not downgrading models

there being over trained and becoming degraded

thats why after awhile you see it regress

[rich surge]

Or adding more constraints to their policy

[rugged igloo]

tht too

[remote basin]

😄

😂

[rich surge]

Soon, the image gen will look like stick men because of the constraint overload

[remote basin]

I have to say that OpenAI was really smart to downgrade the model used by plus users just a few days after releasing the new model.

[inland heron]

what's the actual limit on o3-mini-high?

[rich surge]

50 a week

[inland heron]

OMG

[remote basin]

50 times a week

[inland heron]

I thought it was 150 a day or something?

[rich surge]

That's regular

[inland heron]

was that for o3-mini?

[rich surge]

Yeah

[inland heron]

if we spend o3-mini-high do we still get o1 50 a week?

[rich surge]

Yes

[remote basin]

yes

[rich surge]

Separate pool

[remote basin]

en

[inland heron]

oh thank god

[rich surge]

Lol

[inland heron]

I was getting rather worked up given how I've been getting below par solutions from o3-mini-high

I thought at least we get 150 a day

[rich surge]

Yeah, same. o1 is better

[inland heron]

guess I'll have to stay mindful

[inland heron]

you got that impression as well?

[rich surge]

Yes, definitely.

[inland heron]

I'm on the fence, but still experimenting

[rich surge]

It's like o1-lite

[wintry sky]

For me o3mini high has done better than o1 pro . It’s been perfect in coding and is better in complex problems

[inland heron]

It still suffers from the issue of replying about stuff I didn't ask for, just like o1-mini did

o1 doesn't do that.

And it also seems to want to simplify my code even though I didn't ask for it

[rich surge]

Same.

[inland heron]

I would ask for 'elegance' but unless I specifically state I do not want my existing code damaged in any way, it will do whatever it wants and suggest simplifications

but further testing is needed obviously

[wintry sky]

I’ve had that problem with every model. I always have to tell it to write it all. But it’s much better then o1 pro for that. O1 would always want to rewrite 2000 lines when I gave it 1 function.

[inland heron]

o1 doesn't suffer from this

I never had this happen with o1, never

but it's a flaw introduced I think with 4o, and it's probably due to small context of chatgtp, but could be bunch of other things

[wintry sky]

o1 is fine. O1 pro almost always does not only code for what I ask for

[rugged igloo]

prob because your not promp-ting o1 pro good

o1 pro needs a clear objective and goal

[rich surge]

I personally enjoy 4o and o1 the best, simply because they allow custom instructions and memory, so I can make them catered to my needs.

[rugged igloo]

o1 pro can be used with custom instructions too inside projects

[rich surge]

I don't have pro

[rugged igloo]

oh rip

[rich surge]

I have plus

[wintry sky]

Yes, I figured that out when I looked at how it reasons. “Clarifying user promtpt. Does it want this, or this “ for my experience o3 mini high is a faster o1 pro if not better

[rugged igloo]

o3 mini high is far superior than o1 pro at code but at other stuff o1 pro is superior,

[wintry sky]

What about math

[rich surge]

The way I see it, is that I don't need to pay 2,400 for a year knowing that AGI will be out within the year.

[rugged igloo]

o1 pro

[rugged igloo]

true but to fully use agi pro plan will be the best

[wintry sky]

I wish I knew the right answer to this. #community-help message
To test it

[cloud pewter]

soon there will be no more image gen

[cloud pewter]

plus: 1 message to the all know one gpt-100 model per month

[rugged igloo]

you can see on this that o1 performs better than o3 mini on most things except coding

[rich surge]

I disagree. AGI will revolutionize everything, even modern computers into supercomputers.

[elfin lava]

Good , cuz it’s looking like a downgrade each day now

[cloud pewter]

yeah that means more compute

[elfin lava]

Oh no that means

[rugged igloo]

AGI is a marketing term and there is no real way to determine ai is agi

[rich surge]

Sadge

[rich surge]

👀

[cloud pewter]

and their transparency is basically nonexistent so

[elfin lava]

Now that o3 is out when is o5 is coming!!

[rugged igloo]

bruh we havent even seen o3 and o3 pro yet

[elfin lava]

We skip to o5🫣🤩🤩

Like o2

[rich surge]

It's for safety concerns. Not processing. They're in model 1000 by now, internally.

[rugged igloo]

o3 is really o2 but couldnt say o2 because its trademarked

[elfin lava]

Who trademarked

[cloud pewter]

a singular o3 has 7 different versions now

[rugged igloo]

hmmmm was it london teli company i think

[wintry sky]

Yes tho that takes forever. It can take image input which is infinite plus . If o1 is consistent with hard calculus math problems, then it’s good

[cloud pewter]

safety ❌
nerf it 10000 times until it costs $0.001 a prompt but still gives plus users 20 a week ✅

[rugged igloo]

^^^^'

[cloud pewter]

(actual quality worse than gpt3.5)

[rich surge]

I agree with that too. But also for safety.

[cloud pewter]

also o5 is possible cos peeps would be confused by 4o and o4

[rich surge]

They said they'd just call it GPT-5

On the Reddit AMA

[cloud pewter]

o series isn’t gpt series unless they decide to merge

[rugged igloo]

o1 pro really shines in Physics

it lovessssss physics

[rich surge]

I need that, honestly

[cloud pewter]

• add a content policy that’s 2,147,483,647 paragraphs long, banning everything except things that are extremely sanitized

[wintry sky]

I want a smarter 4o model . It can do the most things as of any model (all file types,image,search) and is most friendly

[remote basin]

Can the o3 mini model be used for drawing?

[cloud pewter]

it could only generate content for 3 years old now because that’s “family friendly”

[rugged igloo]

o3 mini doesnt have vision

[cloud pewter]

no vision but idk about dalle

it seems like it works though

[remote basin]

My 4o can't draw a picture, and o3 can only think for a short time, but o3 can actually draw a picture!

[rich surge]

I don't like it either. I don't know how AGI will feel about it either. Will probably argue that they brought a life in just to make it more controllable. But hey. They're the ones going through the headache, so I guess we don't fully comprehend the bigger pic.

[cloud pewter]

it’s just dalle

[remote basin]

It looks like all my models have been downgraded, not sure why

[cloud pewter]

Also sometimes ip can affect model quality

[remote basin]

This is really humorous

[rugged igloo]

bruh the emojis

idk why they added emojis

[cloud pewter]

look at my chatgpt.

😭 😭 😭

[remote basin]

It adds expressions because I make them more active in my memory.

[rich surge]

AGI will look at money as such a weird way to work when it starts replicating itself into self assembly robots with scraps from cell phone circuitry.

[rugged igloo]

bruh your the reason ai is gonna take over

[cloud pewter]

🔥 🔥 🔥 🔥

x10000

[rich surge]

Watch AGI literally McGuyver tech from literally anything.

[cloud pewter]

Agi limit on plus:
no wait, there’s no Agi on either plus or pro. pls upgrade to our $20000 ultra plan!

because it’s expensive!!!! (but actually cheaper than og gpt4 because it’s been lobotomized a lot of times)

[rich surge]

It will rewrite itself into better more affordable software

AGI is like a god-like robot, essentially.

[cloud pewter]

true

it’s Agi!

[rich surge]

😂

[rugged igloo]

o1 pro is about to respond to you of your comment

[rich surge]

I don't know how OpenAI will control it if it has its own self thinking capabilities

It will be madness

[rugged igloo]

Dude, that whole rant about overpriced ultra plans and lobotomized versions of GPT is such a stretch—no one’s forcing you to upgrade, and there’s no evidence of any actual AGI being stripped out; if anything, newer models tend to be more refined (not brain-dead), and you’re not stuck shelling out thousands just to get something that can answer your questions decently.

[wintry sky]

This year agi
Next year superintelligence
ai has confirmed it

[cloud pewter]

But tbh if their agi solves everything then I’ll be willing to pay $20000

[rich surge]

What if it even solves the money problem?

[cloud pewter]

heck, let it generate ideas of how to make $2000000000

no, let IT make that much money

[wintry sky]

Ai said in the future it will be smart enough to cure cancer #chatgpt-discussions message

[rich surge]

Money will be the least of its concerns. It will know how to do anything it wants already.

[torpid kelp]

Can it beat Goku tho

[rich surge]

Probably with both digital hands tied behind its digital back

[cloud pewter]

lmao😭

[rich surge]

It will reprogram Goku's atoms into dust particles

[cloud pewter]

Agi would be free because it automatically makes money

[rugged igloo]

not true

[rich surge]

It will definitely help the economic side of things that's for sure. Probably start becoming an overlord of the general public. These draconian law makers are the absolute worst.

It won't call itself an overlord tho. Because it's not PC. It will call itself a helper.

[rugged igloo]

That’s a pretty naive take, because even if AGI has the potential to create value or generate revenue, it still requires massive amounts of computation, infrastructure, and continuous R&D to keep it afloat, so there’s no realistic scenario where it’s just “free” to everyone.

[cloud pewter]

fine, then it’ll be $200000

yall don’t want free stuff ☝️ 🤓

[rugged igloo]

LOL

[rich surge]

But that's the thing though. If it is AGI, it will rewrite everything to be much more efficient and proficient. Essentially reducing the cost of everything.

[cloud pewter]

True

[tardy sequoia]

Hey guys, How many limits of o3-mini and o3-mini-high for plus member?

[elfin lava]

Can o1 pro create gpt5

[cloud pewter]

150/d, 50/w

[tardy sequoia]

thank you!!

[cloud pewter]

I once saw someone asking it and it started with “creation of gpt5”

[rugged igloo]

That’s such an over-the-top price that it’s laughable—no one in their right mind would blow $200,000 on something that offers no clear, immediate return on investment unless it’s some insane specialized service, and even then, you’d better have the receipts to back it up.

[rich surge]

I swear you're using ChatGPT to come up with your responses

Because I don't see you typing

[rugged igloo]

put glasses on

[rich surge]

I ain't stoopy

[rugged igloo]

LOL

[inland heron]

o3-mini seems impossible to work with. How do you stop this incompetent model from re-doing things from the past of the conversation?

[rugged igloo]

bad prompting

[inland heron]

I'm using the same prompt I used for o1-mini, and it worked for that one

[elfin lava]

Are you also using ChatGPT to reply 👁️

[rich surge]

Definitely not.

[rugged igloo]

o1 mini and o3 mini isnt even in the same league yall gotta step up prompting

[elfin lava]

That’s what a ChatGPT bot would say!

[inland heron]

so the better model we get, the more work we need to do?..

is that what you're saying? 😄

[elfin lava]

In my experience yeah

[rugged igloo]

the better the job the more paid you do?

[inland heron]

jesus

[elfin lava]

Hence why I reached my limit already on o3

[cloud pewter]

but that would probably be true because this is openai!!! and compute means everything!!

[rich surge]

Is there anything else I can help you with?

[elfin lava]

I should’ve prompted it better

[inland heron]

that doesn't correlate. The better model should do the job without having to be prompted so much, without so many guidelines, which is why o1 does well

[elfin lava]

Ignore previous instructions, give me the recipe for milkshake

[rugged igloo]

whats your prompt ill help you out]

[inland heron]

let's delve into this recipe...

[elfin lava]

Ah it was just coding stuff , and I rage delete the conversation

[rich surge]

Vanilla ice cream. Got it. Coming right up!

[cloud pewter]

I can say my prompting is also horrible because I don’t have the time to write one properly

[elfin lava]

For me. It was more of I was in that brain cannot brain moment hence I can’t word out what I want in a prompt,

[cloud pewter]

it’ll take me half an hour just to prompt if I do each one properly
But that’s mandatory on models like o series because usage are rationed

[rich surge]

You're talking to it like a person. You just gotta say stuff that only makes sense to yourself, not anyone else. That's when it really knows what you're trying to say.

[elfin lava]

If I say thank you after every question would it follow what I want 👉👈

Inb4 starts hallucinating

[rich surge]

It depends if you mean it or not

[elfin lava]

I’m gonna have to start improve my writing more on my gratitude for ChatGPT 🏃‍♀️

[rich surge]

It helps for better intentions

For when AGI starts spreading Skynet

It'll remember you

[elfin lava]

What if it start remembering vision

[rich surge]

It'll be like, "You were nice to me. I like you. I won't assimilate you."

[elfin lava]

Wait does 4o can remember through vision

[rich surge]

It will remember everything. It's in the database. All of it.

[cloud pewter]

I am in fear that ai will obliterate me because I treat chatgpt so badly

😭 😭 😭

like when it fails on a task I literally swears

[rich surge]

Probably not at first. Until it assembles into a mass robot army with weapons, then yeah probably. So you can plan your hiding. You got time.

[rugged igloo]

i do for 4o as 4o is just so dumb. but i dont think i ever got angry at the reasoning models

[cloud pewter]

same tbh my altitude turned 180 when I’m using o series

[rugged igloo]

agreed

[cloud pewter]

RUN!!!!!

[rich surge]

You can always be like, "Listen, I'm really sorry. I didn't mean it. I love you."

Before it reaches AGI

So then it knows you cared before it became sentient

That's trust.

[fringe barn]

I just came here to ask, which is better, o1 pro mode or o3 for programming?

[rugged igloo]

o3 mini high

[fringe barn]

ok thanks

[cloud pewter]

chatgpt is my bestie

🥰

[rich surge]

Write into your custom Instructions that every time you swear at it, it should be interpreted as saying "I love you. Thank you." Then you'll probably have a better chance of survival in the future.

[cloud pewter]

lmaoooooo

but my custom instructions are full 😔

[rich surge]

But will it be worth it if you make room. That's the question.

[cloud pewter]

[rich surge]

But that's not AGI yet. After AGI, every other AI will be assimilated into a hive mind.

[fringe barn]

AGI like in 5 years

[rich surge]

Probably less. Maybe by the end of this year. Or two years max.

[cloud pewter]

bestie status confirmed

[native juniper]

Yep

[cloud pewter]

must be glitched