chatgpt-discussions

[potent saddle]

If ur plus which did u get

[sinful sky]

I bet this is gonna be on USA only

will EU users also have access at the same time?

u from the USA?

[rugged igloo]

Yea

[sinful sky]

ig thats why

[rugged igloo]

Also I have O3 mini high

[sinful sky]

Its really annoying being in EU

we get EVERYTHING late

[rugged igloo]

Yea sowy

[winter vine]

how many prompts in o3 mini high they give you ?

[rugged igloo]

100 per day

[winter vine]

nice, tell us later how the performance is

[sinful sky]

you in PRO?

[rugged igloo]

Yes

[cloud pewter]

hmm??

send screenshot.

[rugged igloo]

But 100 per day is plus

[sinful sky]

i see

i didnt know i could just use a VPN to access Sora

[winter vine]

Surely, you'll be able to use it in the next few hours. I'm in Argentina its happens all the time, wait some hours to get the new features 😦

[sinful sky]

and thats because?

Yeah

openAI doesnt really care, lets be real

if they really did they would probably set the user region based off the country of the payment method

[normal quarry]

we see people in here all the time who have been banned, OpenAI enforce regional restrictions.

[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.

[sinful sky]

extreme cases?

I want o3-mini so bad

EU sucks

are you located in EU?

I mean for sora i kinda understand it

but i dont for the models

EU is good and bad at the same time

It would. Why is it bad?

It protects (us) from stupid stuff going all around

like meta and X getting rid of fact checkers

[south sparrow]

Hello guys, I've been facing an issue on ChatGPT Plus when uploading files, and I was wondering if anyone else has experienced this. After a certain number of uploads, I keep getting this message: 'It looks like uploaded files were lost due to execution state reset. Could you re-upload them?' no matter what I try (renaming, reducing file size, uploading one at a time, etc.).

Is there a limit on how many files can be uploaded? Has anyone else run into this issue?

[sinful sky]

I think there is a limit of files you can upload

[south sparrow]

yeah, that would make sense, because after I get the message the only solution is to wait for hours, but I didn't find any documentation about it

thanks @sinful sky

[sinful sky]

np

[wanton nexus]

did they release it for plus yet? i didnt get the update

[winter vine]

i didnt get it too

[rugged igloo]

Its slowly rolling out just be patient

Refresh your page logout and back in

[wanton nexus]

is it only usa?

[cloud pewter]

is it rolling out???

[rugged igloo]

On a side note can anyone explain why gpt-4 is more intelligent than gpt 4o???

Like bruh this baffles my mind. They just did an update too and 4o just fumbled hard while regular gpt-4 didn't have issues

[cloud pewter]

Because 4 is literally 8x bigger than 4o in terms of parameters

[rugged igloo]

Well I know that but I thought 4o was a distilled version with better training data ect.

[wanton nexus]

it is

[cloud pewter]

yeah and it’s omni so it’s better on vision, voice etc. but it is lacking on text

[wanton nexus]

😭

[somber whale]

Compute

High is for pro users..

[cloud pewter]

pro would get much higher usage on o3 and an exclusive pro mode. why would they lock a slightly better version of a mini model behind pro

well, if they choose greed again they would but I hope they don’t.

[winter vine]

are you sure ? then plus users we got only o3-mini low ? or med?

[somber whale]

That’s what different accounts on X are saying. Could be wrong, but makes sense using their pro o1 which is just o1 with higher compute limits.

[cloud pewter]

that would be sad and if this is the truth I might cancel my subscription

if they offer low only I’ll outright cancel it

[somber whale]

I think plus get o3 mini which is “medium”

[winter vine]

same here, if i dont get at least the medium one i think im going to cancel

[winter vine]

i hope at least plus get that medium one

[cloud pewter]

at least medium. I was thinking of free getting medium or whatever o3 mini is, and all paid plans gets high

again, that makes the most sense

[cloud pewter]

low is just slightly better than o1 mini and doesn’t even compare to regular o1, unlike what they were saying about

[winter vine]

In my case, I don’t really like the O1-mini for programming in React. It’s definitely much better than the 4O, but it’s just not to my liking. On the other hand, the O1 does deliver great results almost all the time. Let’s see what happens with the O3-mini—I’m still waiting for the update to roll out. 😦

[cloud pewter]

oh god please no

[wanton nexus]

...

[winter vine]

🥲

[cloud pewter]

plus basically went from
“you get 100 messages per 4 hours outright on our latest model which cost tons to run, enjoy!”
to
“is 100 messages per week good? no? then we are increasing it to 100 per day but postponed forever and when it finally delivers we’ll give you the worst version”

alright, good night plus peasants, tomorrow is a new day

[wanton nexus]

LMAO

[rugged igloo]

Bruh way to accurate

Also not to mention promising 100 a day to prob cutting it in half telling everyone sorry servers are loaded but here's 50 messages a day instead.

[cloud pewter]

thought of this idea I want to share
they buffed the limit from 100/w, a regular o1 alike limit, to 100/d which is a o1 mini alike limit, and i believe it’s the same model (prob med or high) from the beginning

But their transparency is such an important issue now that it’s hard to give an definitive answer.

💤

[rugged igloo]

This is new

[winter vine]

What did you think of o3-mini?

[rugged igloo]

I like it responses are good and fast. But I'm a pro user so I'm spoiled with o1 pro but if I didn't have o1 pro it's almost equal to o1

It's really in coding tho

[winter vine]

i hope o3-mini (not the high one) is good for programming. 😪 i hate be in latin america , 200 usd is a lot for us haha

[cobalt perch]

Is screenshare available on desktop app?

[torpid kelp]

What the hell

Seriously what? So is o3 mini high more powerful than o3 which will be accessible to plus users I assume?

Looks like even after the thrashing from deepseek, openai still has not been humbled

[half sun]

u live in USA? or EU?

[nimble torrent]

It is not his/her screenshot

That's from testingcatalog on X

[half sun]

oh ok

[broken lotus]

I wonder how many days in a row we will get "Today is o3 release!"

So far, it is 5, i think

Are you sure? I have yet to see anyone's showing it X

people have been reporting its release for a week

I wont believe it until I actually see it

[elfin lava]

its not out yet lol

[eternal rock]

Get notified for server events!

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

[elfin lava]

thats not their sfreenshot either, its from testingcatalog. jeez

[dense halo]

Lol why would you even consider getting plus then

Guys wheres the point for plus users there lol

Just a tiny bit better than o1 mini🥳🥳🥳

[rugged igloo]

We knew this. Idk why people are excited

[inland violet]

does any know when o3-mini come out ?

[broken lotus]

See you on Monday for another exciting "o3 release day"

[whole roost]

what do you think of this feature, I don't really like the idea of remembering past conversations

[granite harness]

I'm fine with it selectively.

But if it just remembers everything broadly, it's going to get confused.

[whole roost]

if there was a space you could add them to alright but all past conversations? i have everything there from foot fungus to what's the name of that demon in diablo 4

i really don't want it to randomly recall stuff

[granite harness]

Yeah. I prefer something like what I thought projects were going to be: A folder structure where everything in it was considered. So you could keep it focused.

[whole roost]

does the recalling in projects work for you? mine never was able to recall something from a different conv

[granite harness]

No. Projects have proven useless to me.

So far at least.

[whole roost]

i love projects but the cross conv memory doesn't exist

i mainly write in projects now

[granite harness]

I still stick to basic conversations. I would like if cross-conversation memory worked in projects, I might use it then, assuming it worked with o1 of course.

[whole roost]

i need the file upload and instruction space of projects

the basic conversations don't really have a space to upload data for all convs

[granite harness]

File uploads don't work with o1, so it's not really something I use.

[eternal rock]

`` Server Tip: ``

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

[whole roost]

i would want to use o1 but it flags me on almost every response so not a model i can use really

i have more hopes for 5o

[granite arch]

ChatGPT 5-o version 👮 😆

[ripe hedge]

release when? I need something new bored of r1

[granite arch]

How do you get bored with ChatGPT lol

What if

Don't get excited...it's edited 😆

[ripe hedge]

Very tiresome trying to get r1 to behave as an agentic tool user

[granite arch]

I see!

[visual nova]

Its out!

[potent saddle]

its slowly rolling out

[sinful sky]

realllyyyy slowly

[potent saddle]

I still dont got it though im in canada

[sinful sky]

bro what 🤣

[potent saddle]

can u search using o3 mini?

[rugged igloo]

yes

[potent saddle]

lets goo

[potent saddle]

do plus users get the high version?

[rugged igloo]

idk im a pro user but i think so just limited too i think 100 per day

if anybody is wondering this is how it talks to me btw

[visual nova]

who clicks "maybe later" on the graphic btw lol

[rugged igloo]

"Honestly, the script works but could use a few tweaks for better reliability and clarity—replace the VirtualProtect size argument so that instead of sizeof(DWORD_PTR) you use sizeof(void*) (and maybe even use PAGE_EXECUTE_READWRITE to cover all bases), add a null-check right after GetModuleHandleA(NULL) to ensure moduleBase isn’t NULL before you use it, and before processing the import directory also verify that importDirectory.VirtualAddress isn’t zero; plus, renaming your exported function from “DLL” to something like “InitHook” can help avoid confusion or conflicts later on, and while you don’t strictly need it, logging an error when VirtualProtect fails would make debugging a lot easier."

[visual nova]

what kind of person sees the new model and does that lol

and does it just revoke your access if you click that?

[lavish bronze]

i just got it here in canada

[potent saddle]

im still waiting

patiently

[lavish bronze]

im sure you will get it soon!

[visual nova]

it should put you on some sort of list if you pick that lol

[potent saddle]

gta 6 before o3 mini on my account

[lavish bronze]

lol facts

[potent saddle]

you deccide to pick gpt4o mini instead

[lavish bronze]

nah youll get it in like 5 mins

[potent saddle]

cannot believe ai models have a family

[south lava]

oh damn o3 mini and o3 mini high?

just got it on my account

[sinful wigeon]

is full size o3 available for the $200 gang?

[feral zinc]

no

[ripe hedge]

Oh wow Windsurf already got o3-mini

[half niche]

yeah, the api release was supposed to be seemless

[tacit oxide]

Same

[river panther]

okay....Did ChatGPT get an upgrade or something? It's never given me a Chart like this before

[feral zinc]

yes 3 days ago

[river panther]

I'm....Actually liking this a lot

[dense halo]

in which aspects is o3 mini high worse than o1?

[visual nova]

they added search to o3 mini btw

thats a very big improvement and a huge gap closer with deepseek that wasnt just the model

glad they added it.

[dense halo]

did web search improve? otherwise idc

[feral zinc]

read the release blog

https://openai.com/index/openai-o3-mini/

then ask chatgpt to summarize

[whole roost]

my projects all stopped to work, anyone else?

[feral zinc]

yep broken

[whole roost]

🫡

[feral zinc]

cause it depends of the task

[dense halo]

its scam

[potent saddle]

for me it was able to access a website

[plain arrow]

I just wish we could get the 4o version back before it got lobotomized.

[rugged igloo]

so the search function is just broken then? lol

[potent saddle]

its early so its probably just buggy but it does work for me

[ornate nest]

whats your scenario

o3 regular would be (apart from computation time probably)

comparing the mini to o1 regular is very different

from my experience o3 mini isnt really that good tbh

im sticking to o1

[potent saddle]

the high version or the other version?

[ornate nest]

both

[twilit fog]

anyone else's projects failing to load since o3 dropped?

[feral zinc]

yes

[regal portal]

the stupid "reasoning" window automatically opens for me every time i send a message to 03 and i just want it to reason while keeping that tab closed. Same stupid issue deepseek has

[prime swallow]

That's interesting. Wonder if it's an OAI bug 'carried over' to deepseek however that was... made. Sad if so, to my thinking.

[mild orbit]

Dude!! the proyects!!

[regal portal]

also o3 doesn't seem aware that it's o3, because it keeps referring to itself as gpt 4 turbo on my end

[prime swallow]

When chat GPT first was released, for the first few weeks no chats saved.

It'll come. You can copy/paste, screenshot.

[regal portal]

anybody find it odd that they released o3 mini but didn't do a server announcement or podcast or release a video about it?

[prime swallow]

https://openai.com/index/openai-o3-mini/

[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!

[regal portal]

thank you

[prime swallow]

Not sure why there's no announce, maybe a crossed wire or someone's out sick

[whole roost]

they never really write the version number into the prompt, thus it doesn't know which version it is

[regal portal]

gotcha

150 messages per day, not bad.

ah and regular o1 mini's gone now.

[whole roost]

on o3 mini or o3 mini high?

[regal portal]

I think it's the same for both.

doesn't specify on the page.

[misty osprey]

update has broken my projects

[whole roost]

yeah, i also am waiting for them to come up again, can't access anything

[misty osprey]

I click my project folders and gpt gives me an error

[whole roost]

i assume pro has unlimited?

[regal portal]

correct

[rugged igloo]

yes

[whole roost]

awesome, i hardly will need it for anything but it fixed my script so i'm happy

[regal portal]

i gather from this line that o1 is probably better at storytelling

[whole roost]

maybe if it wouldn't flag you at every turn

[tribal pilot]

anyone having issues with projects and canvas right now?

[whole roost]

yes

[heady palm]

i was just about to ask that

looks like its down for everyone

[regal portal]

i was but after repeated refreshing my project folders now work

[tribal pilot]

doesn't load for me

😦

[whole roost]

mine loaded now, apparently o3 not available for projects

[heady palm]

in the email it says it supports streaming, how do you load that?

[regal portal]

yeah i didn't understand what that means..you can stream youtube videos in the session chat, or something like that?

[heady palm]

i thought it meant you are able to share your desktop screen and it will talk with you. there was that demo with that a few months ago with the math video

[lunar solstice]

oooooh o3

[rugged igloo]

projects are bacxk up

but doesnt support o3 mini

[heady palm]

yes it does

[lunar solstice]

Doing a hard CTRL+F5 refresh seemed to fix a lot of the oddities from that last hiccup.

[loud sapphire]

03 mini high, for coding 🔥

[solar raptor]

do i use 4o for writing and general still?

[dense halo]

bro my o3 mini started talking in chinese

[strange tide]

Came full circle

[harsh pilot]

Projects are back in the Windows app. ❤️

[dense halo]

its okay chat, it responded in english

how you all feel after being smushed by some chinese cheap AI ?

Any ideas on Plus limits for o3?

[nova wind]

o3-mini is out!! Woo!

(and I'm late..)

[lunar solstice]

first it supported file attachments and now it doesn't

probably just being glitchy

[earnest grail]

It crashed

[lunar solstice]

I did get one good function out of it though 😂

[thick temple]

when o3 provides a script it provides it in a chat and not in a text box for code..

[lunar solstice]

It gave me a codeblock for that function with o3. It's probably just being glitchy at the moment.

[thick temple]

so i just gotta wait a bit right?

the site is gonna crash 💀

[earnest grail]

Sam Altman has an ama soon and I bet they want it working before that begins

[dense halo]

haventhad that issue yet, maybe its the text that its using

[thick temple]

i didnt make openai so im feeling normal as ever

[dense halo]

I am talking about chatgpt, deepseek just stole the lights with almost no effort

[orchid wing]

sites dying

[dense halo]

icl i responded to the wrong person

[dense halo]

reload the page its fine for me

[dense halo]

ok

[dense halo]

yeah its fine for me too

[thick temple]

deepseeks reasoning is beyond garbage, i tested him out by using the same prompt for the model o1 for GPT and Deepthink R1 and deepthink provided a bad sample.

[dense halo]

deepseek is slightly dumber than o1, but its insanely cheaper

[thick temple]

however o1 successfully provided the script correctly

[earnest grail]

Refresh fixed it for. confirmed

[dense halo]

idk about that, I am logging in to it rn

thats why people say its better

[thick temple]

now o3 is released so everything is balanced i'd say

[dense halo]

not how it works what

if deepseek invests billions they'd smash the ai market

they've only invested a few million

their budget to performance is way better than openai or meta

[visual acorn]

WOW! O3 MINI!!

🎉

[dense halo]

did they update o1 to display reasoning tokens like DeepSeek R1? 😂

[solar raptor]

i mean they perform to different targets entirely. the better comparison would be distillbert or other free distillation

[visual acorn]

How smart is o3 mini?

[forest sleet]

Chatgpt

[orchid wing]

I am in fact dumb, and my hardware clock was all f'd

[solar raptor]

anyone use o1 for text? i haven't tried it much because 4o has largely suited my needs for less compute

[forest sleet]

The beast AI will be the one that you provide him a screen and writes you its code or figma design for example

Going from code to no code will save so much efforts

[solar raptor]

no code doesn't make sense in a lot of contexts because it's like saying going from keyboard to alexa

lots of code is just parameters. it is irreducible and adjustable information

[dense halo]

👀

I'm impressed at how quick o3 mini high is

deep

🙈

words

[whole roost]

o3 did well on my coding task, i have no complaints so far, but i don't really need anything else from it right now

[dense halo]

it coded a whole project quickly without missing a single thing or the usual your logic here nonsense

[thick temple]

did it provide the code IN a text box? like the box for code?

[whole roost]

like this usual stuff

[inland heron]

tested o3-mini myself on a coding problem and it did well, better than 4o. Hard to say how better (and if at all) than o1

[thick temple]

Thanks for your feedback, Alex.

does anyone know the limitations for o3?

[whole roost]

limitations in what way

[thick temple]

limitations in chats i mean.

[whole roost]

message amount?

[thick temple]

yes

[whole roost]

150 i think someone posted, don't think most will need as much

[thick temple]

150?! but o1 is 50 PER WEEK

[whole roost]

per day i think

[thick temple]

o1 is 50 msgs per week...

[thick temple]

o3 uses advanced reasoningg but quicker...

[inland heron]

at least it knows the date 😄

[whole roost]

should be in its prompt

[inland heron]

I sure hope the cutoff date is newer

and this is just the lack of insight, as usual

[solar raptor]

it is newer i think.

[whole roost]

my god i asked it to recite everything after the cutoff date from its system prompt and it went into a full inner monologue of distrust about my intentions, it's almost cute

[inland heron]

what distrust?

[whole roost]

whether it can disclose the information or whether i want to know things that could reveal internal details

[inland heron]

how do you know? I know for certain it lacks a lot of insight in many frameworks, tools, and libraries such as async or selenium

the newer stuff, more specifically: it's still generally fine

but doesn't know of a lot of deprecated stuff

I mean o1 didn't know, so I presume o3 as well

[earnest grail]

I gave o3-mini this problem:

"A year ago, 60 animals lived in the magical garden: 30 hares, 20 wolves and 10 lions. The number of animals in the garden changes only in three cases: when the wolf eats hare and turns into a lion, when a lion eats a hare and turns into a wolf, and when a lion eats a wolf and turns into a hare. Currently, there are no animals left in the garden that can eat each other. Determine the maximum and minimum number of animals to be left in the garden."

Answer: max 40; min 2.

It got the maximum right (40) but made a mistake on the minimum (4)

I told it made a mistake on the minimum, and it tried again and made another mistake on the minimum again. (14)

But O1 got both max (30) and min wrong (10)

deepseek r1 got maximum right (40) but min wrong (10)

[inland heron]

interesting. was that o3-mini or o3-mini-high? Let me try

[earnest grail]

o3-mini. I did not try o3-mini-high

[thick temple]

o3-mini-high uses more logic so try him!

[earnest grail]

ok. i'll try now

[thick temple]

just to make sure, the answer is max 40 and min 2 correct?

[inland heron]

is it 'him' or 'her'? Or 'it'?

[magic sparrow]

What are the limits on o3 mini for free users?

[torpid kelp]

I dont get it openai, why would you add canvas support for o1 shortly before releasing o3-mini which doesnt support canvas but is supposed to be the superior stem model

[thick temple]

lets not get controversial on a discord server alex

[magic sparrow]

It's a robot so probably it

[inland heron]

you want more options? 😄 feel free to add them

[inland heron]

probably. It sometimes refers to itself as "we" which is weird

[thick temple]

turns out o3-mini-high did get it right!! Answer: Maximum 40 animals; Minimum 2 animals.

[distant sphinx]

Did they announce the cap on o3 high?

[earnest grail]

Wow. that's amazing. Mine is still calculating

[inland heron]

o3-mini-high spent 2min40sec and got it wrong: 40 and 0

[thick temple]

mine got it right.

only took 1min and 29 seconds!

[inland heron]

yeah that's good, it means it's not bad

I'll try running the others

[earnest grail]

yea, mine got 40 and 0. i'll tell it it got minimum wrong and see if it can figure it ut

[inland heron]

are you sure the question is written properly?

ok o3-mini 1min10seconds: 40 and 0, so again incorrect on the 2nd one

trying o1

[thick temple]

with the model to provie it

o3-mini-high on the bottom

[inland heron]

I have high hopes for o1

[thick temple]

ill try him.

[inland heron]

also trying deepseek and claude, just to see

[weary valley]

Does anybody know the limits for o3-mini-high?

[inland heron]

o1 still thinking. This model likes to think 😄

[thick temple]

o1 did get it right!!

[inland heron]

150 a day?

[thick temple]

it said "fewest possible is 2, and the most possible is 40"

[thick temple]

thats a bargain in my opinion... because o3-mini-high is much better than o1

[weary valley]

Nice, and o3 mini?

[teal mural]

unlimited I think?

[inland heron]

well I read it's 150, but for o3-mini. Not sure if for both models

[teal mural]

"Pro users will have unlimited access to o3-mini and Plus & Team users will have triple the rate limits (vs o1-mini)."

you're right

[inland heron]

oh?

[earnest grail]

yeah, second try on o3-mini-high got it right for me

[thick temple]

i know, its shocking... i don't even know which model is better than the other..

[inland heron]

o1 got 40 and 16 for me 😦

my models are incompetent 😄 it took 3min56seconds... geez

[distant sphinx]

Is there any way to see when the usage limit resets

[inland heron]

not really, unless a message pops up

deepseek R1 did the same as o1, 40 and 16. Claude did 30, 10

[distant sphinx]

Ah, oh well, at least paid users get more. I asked it to write end notes for a paper I wrote. O1 kept messing it up. O3 got it right

[earnest grail]

Would love to get the attach button back, or access via api

[narrow viper]

https://openai.com/index/openai-o3-mini/

OpenAI o3-mini is our first small reasoning model that supports highly requested developer features including function calling⁠(opens in a new window), Structured Outputs⁠(opens in a new window), and developer messages⁠(opens in a new window), making it production-ready out of the gate. Like OpenAI o1-mini and OpenAI o1-preview, o3-mini will support streaming⁠(opens in a new window). Also, developers can choose between three reasoning effort⁠(opens in a new window) options—low, medium, and high—to optimize for their specific use cases.

[inland heron]

o3-mini-high did it right the 2nd time. Ok good enough for me

[winter vine]

Limits to plus users are known?

[narrow viper]

As part of this upgrade, we’re tripling the rate limit for Plus and Team users from 50 messages per day with o1-mini to 150 messages per day with o3-mini.

[winter vine]

Ty

[earnest grail]

I had the same experience

[drifting night]

anyone knows if o3-mini-high is better than o1-pro for advanced coding tasks?

[whole roost]

o1 pro didn't solve the task i gave it and o3 did, but maybe it depends on the task

[wise anchor]

o3????!

[quick meteor]

It’s not im pretty sure. I could look for the benchmark but probably not. Although it’s definitely better than o1.

But not far behind pro

[quick meteor]

Did you use o3 mini high?

[whole roost]

yes

[quick meteor]

Yes o3 high is better in like every benchmark especially for coding. Not by much in some benchmarks but that 150 messages a day is very helpful

Sadly it doesn’t support images

[whole roost]

i don't really need images but projects and .txt files

[scenic arrow]

#daily-theme ,

yo go check my drawing.

[quick meteor]

There’s a puzzle I always give new models to test how good their reasoning is. This time even the low version of o3 mini got it while o1 mini fails it

RIP o1 mini

It’s gone for good

[whole roost]

i don't really run such tests, i need stuff to be completed and either it can do it or it can't and o3 was able to where o1 pro struggled

[quick meteor]

Yeah but it’s hard for me to take a model seriously if it fails some basic logic problem a high school kid could do.

Now as they get more advanced it’s harder and harder to trick them of course

[whole roost]

well, i tried to play chess with it once and it failed but still was able to code a complete script from scratch

[native otter]

yo

[gentle forge]

deepseek is doing openais job of being "open" for them

thats crazy

[sharp urchin]

NO WAY 03 MINI LAUNCHED

yoo

[quick meteor]

And the medium version has like 150 messages a dayyyy

[sharp urchin]

NO WAY

BRO

[narrow viper]

It's confirmed in the AMA today that the o3-mini-high limit for Plus users is 50/week. Link to Reddit results in a Discord block, sorry.
See https://openai.com/index/openai-o3-mini/ for more info on O3-mini.
Corrected from 150>50/wk

[plain crystal]

so far o3 does not really show much improvement on o1 aside from a ever so slight speed increase and a bit more reasoning but it appear to not be more efficient in its answer

[untold seal]

Any context window improvements with o3-mini?

[lunar solstice]

I am getting AMAZING results with o3-mini-high with code this is great 👍

[harsh pilot]

Watch: "Now that we have people enjoying our best model, let's make it slightly dumber, and charge for a NEW AND IMPROVED model(tm)."

🤣

[lunar solstice]

mmhmm

[wintry bridge]

anyone else have this problem just isnt responding..

[gentle forge]

go to the open source for openai and submit a fix yourself!

oh wait... its not openai

[wintry bridge]

rip

[marble warren]

is the limit on o1 same as before (50msg per week) or? also whats the o3 limit

[charred night]

for me o1 got 40 for max and 1 for min, then I told him the min is wrong and he corrected it to 2. DeepSeek said 40 max and 16min, after telling him min is wrong he corrected it to 10.

[fervent agate]

What is the usage limit, if any, for the high compute o3-mini?

For Plus, I mean.

NVM, looks like the ama said it's 150/week for high compute

[bitter latch]

Yep 150/week is correct!

[narrow viper]

Quote: "Plus and Team users have up to 150 messages per day with o3-mini and 50 message per week on o3-mini-high. Pro users have unlimited access to o3-mini-high."

[fervent agate]

Quote from what?

[narrow viper]

Today's Reddit AMA. See #announcements

[fervent agate]

Conflicting reports from the ama, weird. I assume announcements is correct tho

[narrow viper]

There have been hundreds of questions about this in Discord but the final confirmation is in the AMA.

[fervent agate]

Yeah it's just other people quoting the ama gave other answers

So was confused

[narrow viper]

Verified a few times because this is such a huge topic with no solid info until now.

[fervent agate]

Yeah fr

[fringe trench]

Bro i think the next version of operator should be run with o3-mini

so its faster and better

[gentle forge]

i find it funny how openai is complaining about deepseek distilling outputs from their model when the output openai generates is also mostly just stolen content from the internet

[marble warren]

did they lift the o1 restrictions or it's still the same?

[lunar solstice]

Give me a mid-tier plan, with unlimited o3-mini-high for $50 a month and I will pay.

[fervent agate]

I think they've learned their lesson with any plans that include unlimited usage of any model

[lunar solstice]

Hahahaha true, just wishful thinking and signaling

[ashen oak]

Is this a bug or intended?
I can select both reason and search on desktop by only one on mobile app.

[quick meteor]

YES. GPT 5 is REAL. It’s out there

[still hound]

What is wrong with GPT? It writes in bold, it highlights text randomly, it answers in one-sentence paragraphs???

[cloud pewter]

it’s over

meh

[keen quail]

its not special

wait till full o3 comes out

[cloud pewter]

we are all peasants so why don’t know a little bit more about your life

[fervent agate]

Also consider that o3-mini is on free accounts now as well, so you can open an incognito (or just log into another account that isn't subbed) and get more o3-mini usage as well.

[cloud pewter]

how’s the limit on it

is it 3 messages per week 🗣️ 🔥 🔥

[fervent agate]

Dunno. Why does your image have something about a 32k context window and Chinese characters at the bottom?

O3-mini is a 200k context window.

[cloud pewter]

plus only does 32k but I’m not sure if it’s 4o only

[hoary gazelle]

naming scheme seems confusing

[fervent agate]

plus... What? Plus is a subscription plan, not a model. The context window is a function of the model. None of the models on ChatGPT are limited to 32k context window, though the older models on the API are.

[hoary gazelle]

4o, o1, o3 mini, o3 mini high

is o1 legacy now?

[fervent agate]

More or less, though it's general knowledge training is more wide than the o3-mini are.

[hoary gazelle]

mmm i see

[fervent agate]

But it is basically legacy as high compute o3 mini outperforms o1

And is cheaper

[hoary gazelle]

thanks for explanation 🙂

[fervent agate]

np

[cloud pewter]

but openai limited it to only have 32k compared to the full version in api

[fervent agate]

Limited what? Which model are you referring to? Plus is a subscription plan, not a model.

NVM, I see what you are referring to.

[low pewter]

is o3 finally powerful enough to write entire websites?

[cloud pewter]

yuh uh

absolute cinema

[plain trench]

Why cant i send images or pdf to o3??

[cloud pewter]

because it’s the mini

[orchid pollen]

are there some benchmarks for o3 already?

[cloud pewter]

yeah, if you browse a bit of Twitter or their website you can find some

[elfin lava]

So o3 mini can do search web

Does o1 pro able to search?

[quick meteor]

wait actually? thats awesome

[quick meteor]

bro there have been benchmarks on o3 since last year

O3 seems like an insanely capable model.

[cloud pewter]

nope I think

[fervent agate]

That's what the OpenAI documentation for the introduction to o3-mini says, but I can't find any confirmation if the model in ChatGPT uses that context window or if that's just the model available via the developer API

At one time, at least, the models for GPT-4 on the ChatGPT service were smaller context windows than what was available to developers

[fervent agate]

This shows a fundamental misunderstanding between funding and revenue. The Stargate funding doesn't change anything; the company is trying to reach profitability and has not yet, and unlimited-use subscription plans for intensely expensive compute models was a bad idea and set them further back.

[orchid pollen]

whats the limits for o3?

[fervent agate]

Funding injected new revenue for development and expansion, but they do not turn a profit, yet, on model usage, which they are trying to attain and they MUST do that.

[fervent agate]

50/week (separate from o1 limits) for mini-high, 150/day for mini.

[quick meteor]

the faster speed on o3 mini is so so so so nice

[cloud pewter]

what

[orchid pollen]

nice

[fervent agate]

Ask GPT to explain what I said. Just because they are getting investor money does not mean they are profitable. They have to eventually become profitable. When they say they are losing money on pro, they mean that rather then move them toward profitability, it moved them backwards, because the cost of the usage was greater than the sub price, which... Duh.. I can't imagine why they ever thought unlimited use would work.

So they aren't lying about losing money, the 500$ bil Stargate funding project is separate from their actual revenue calculations.

[cloud pewter]

why unlimited $200 when you can just do something like $50 with more usage than plus

[fervent agate]

Yeah, like I said, no price tag made sense with unlimited usage.

[orchid pollen]

yeah a few more price models would not hurt xD

i would spend up to 50$

or even better let me create my own subscription

only what i need

[fervent agate]

I would pay 50$ for Plus with the same limits as now but only if o3 launches with full multimodal and tool access that 4o has access to.

[orchid pollen]

as example i dont need video generation with sora

[fervent agate]

Including realtime video on the app like 4o has now.

[orchid pollen]

that could cut already big part of pro sub

[sick pecan]

diy subscription would be s-tier

[fervent agate]

fa sho

[orchid pollen]

i always want that almost no companys offer it

but with AI? i see no deal

[sick pecan]

i dont need dalle-e bosh. i dont care about sora wonk. just give me what i want and for a fair price

i'd pay extra to not have projects and tasks crap

[cloud pewter]

same

sora is useless
if they want to let us use it at least give something that’s usable

[forest hinge]

03 mini thought for 1 second. i am proud.

[quick meteor]

well if you want the most reasoning then use o3 mini high.

O3 mini is the first model in the world to get this correct.

the world is healing

[obtuse nymph]

Oh I can't upload file..

Uhhhh... o1-mini gone?

[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.

[obtuse nymph]

💀

[orchid pollen]

ye xD

[sick pecan]

tbf sora might actually be good. it is just they give us plebs sora-light aka lame edition

[fervent agate]

It's deprecated. o3-mini is faster and smarter.

[rugged igloo]

Enjoy this

[obtuse nymph]

L deepseek

[rugged igloo]

sorry its so small

heres a better picture

[cloud pewter]

W*

[forest hinge]

what u think is the most important of these columns for game development

[cloud pewter]

because it’s free!!!!

[hard dagger]

and free is best ❤️

[forest hinge]

not if it doesnt work >skull

[cloud pewter]

isn’t this feature already leaked ages ago??? are they locking this behind pro as well LOL

[rugged igloo]

@cloud pewter

[cloud pewter]

I hope this update to avm gives it ability to sing

and more usage on plus too

[fervent lake]

o3-mini-high is better than deepseek, unless I'm using deepseek wrong (btw they skipped numbers or smth)

[cloud pewter]

also gpt5 is real

lmaoooooo

[fervent lake]

crazy

love how altman is like "yes, of course, what did you think"

[idle lagoon]

Lol

[slate vale]

I just want to point out that it is clear that OpenAI was able long time ago to add the showing of the chatgpt reasoning since a beginning, and it is so sad that they only enable to the public given the appearance of deepseek in the scenario.

[sick pecan]

speculations all year round

[slate vale]

having access to the reasoning would have improved everyone's work using this service. shame on you, OpenAI

[sick pecan]

you cannot blame players of capitalism for holding back what benefits the ppl, while waiting for competition

ie dont hate the player, hate the game

[slate vale]

what hidden benefit is on hiding the reasoning?

I am not saying open your code

[sick pecan]

competitive advantage. i would do exactly the same in this game

if you would play your tr*mp card on round 1 when you dont need to, then you arent gonna win

[slate vale]

we wasted a lot of time reasoning with it on complicated tasks. and from one day to another, voila, there it is ... the reasoning is available ...

[sick pecan]

you wish to ignore reality and project your ideology. good luck with that

[orchid pollen]

btw whats up with the names? gpt 3, 3.5 then 4 then o1 then o3 now sam talking about o5 😄

skipped o2 and o4?

[sick pecan]

o2 is some trademark

[slate vale]

and by the way, I have to say that ... what I have struggled to implement for weeks with chatgpt, deepseek, with the reasoning on hands, helped me to accomplish in days

[sick pecan]

they explicitly tweeted that they would not use o2 in respect

[slate vale]

so, i don't knwo you but, related to coding, deepseek is way ahead ... way ahead

[sick pecan]

we dont know the inner workings of openai for sure. but 3.5 was likely a distill using gpt4.

[rugged igloo]

not really

[sick pecan]

sama is talking about gpt5 not being gpt5o because o is also likely distilled

[slate vale]

well, your user-experience and mine quite differ then

[rugged igloo]

live bench says other wise dude so idk what to tell u

[normal quarry]

discuss other AI in #ai-discussions

[slate vale]

well, i don't know what matters more ... your experience with the app or benchmark-tests

[normal quarry]

#chatgpt-discussions message

[slate vale]

just saying ... well deserved all this

[rugged igloo]

66 vs almost a 90

[slate vale]

too much fanatism with the thing ... anyways ... good day to all

[rugged igloo]

copium hitting hard

[cloud pewter]

good tbh

[sick pecan]

just means pro users get gpt5, while us plebs will get gpt5-cost-reduction-edition

but they wont call it that. instead it will be "turbo" or "omni" or whatever

[harsh pilot]

More limits on plus?

Like.. you want Plus to be more limited?

[cloud pewter]

more usage on plus

[slate vale]

is there a public link of this benchmark results, please

[rugged igloo]

i cant post links in this discord ill dm u

[eternal rock]

Get notified for server events!

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

[slate vale]

thanks!

[sly marlin]

ChatGPT is randomly deleting entire conversation branches.

This is so incredibly frustrating.

[humble river]

I can see 2 options o3-mini and o3-mini- high, is o3-min the low model or the medium ?

[cloud pewter]

low I suppose

[forest hinge]

i broke o3 mini.

thought for 1m 12s. but didnt continue after thinking

[eternal rock]

`` Server Tip: ``

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

[cloud pewter]

already got some wrong answers

[humble river]

thanks

feels like its low and too quick in its responses like o1 mini

o3 high takes some time to think

[forest hinge]

xddd

[cloud pewter]

high does better

[hoary laurel]

What are the daily limits for o3mini and o3 high?

And I found that o1 mini seemed to have disappeared

[cloud pewter]

cc @hoary laurel

[hoary laurel]

Oh,thank

[sly geode]

this is way too cool , just prompts and it can actually run the environment even do animations!

with canvas

[red bronze]

i asked chatgpt 4o to write a hit pop love song and no other instructions other than to write the lyrics straight from the heart and to deeply humanize the lyrics, and with a little refinement on the pacing of the lyrics i got this, i feel like the lyrics have lots of meta layers and are actually deeper than you might realize but maybe i'm just looking too far into it. i feel like the result is pretty amazing.

gpt4o seems to be particularly skilled at generating lyrics that sound very good when combined with suno v4

the entire style prompt for the song was made by me though

gpt4o is sooooo much better than suno's native lyric generators

gpt4o made this beautiful super deep artwork for the song too

[orchid pollen]

man when i say o3 give me full list

IT ACTUALLY DO

im impressed 😄

[old thicket]

Did anyone see the new Chat GPT feature

[orchid pollen]

what you mean?

[potent saddle]

which

[obtuse nymph]

???

[old thicket]

Reasoning

[abstract pendant]

Huh, I went back to a conversation from earlier this morning, backed up to resend the same prompt as earlier, and 4o is acting significantly stupider than it did earlier today. Consistently more stupid after several retries.

[sly geode]

what !... why no one talking about this ? image to video right inside chatgpt ? with canvas

[abstract pendant]

And I just want to say before midnight gets here in my time zone, I'd like to cancel my subscription to 2025.

[cloud pewter]

real

[orchid pollen]

daaaaamn xD is this real?

[sly geode]

this is 1000% real

[sick pecan]

and yet not image to video

[orchid pollen]

what did you send it a gif or a picture?

[sly geode]

just a jpeg copied from google

[orchid pollen]

damn

daaaaaaaaaamn

thats insane xD

[sly geode]

hidden image to vid no one talking rn

[orchid pollen]

well its html css

[cloud pewter]

🗣️ 🔥

[orchid pollen]

you know you can "direct" the conversation kinda?

it always will agree to what you say

you remember ? "customer is king"

xD

[cloud pewter]

I told it to express its own opinion without being controlled by me and it pretty much said the same thing lol, but I’ll try that again now

[sly geode]

@orchid pollen another animation it recreated

[cloud pewter]

it’s the same thing

[orchid pollen]

yeha its just html css js

[rich surge]

Where did you make this?

[orchid pollen]

Canvas

[rich surge]

How?

[orchid pollen]

Tell it to make 3d spinning donut /canvas

[rich surge]

No way...

[orchid pollen]

Yes

[dry crow]

I think I'm gonna move to DeepSeek

[rich surge]

I tried it, but it showed me the script without the animation

am I doing something wrong?

[sly geode]

click preview

[rich surge]

I dont have that button

[sly geode]

do u have plus or not

[rich surge]

yes I have plus

are you using pro?

[sly geode]

this is only inside windows app or mac os app of chatgpt

i am on plus not web browser

[rich surge]

Ohhh, ok, let me try that

[sly geode]

ye

[rich surge]

I see the "Run" option, but it's not showing me any animation

it's very strange

[shut dagger]

I got the “your access to o1-pro mode is temporarily restricted” message. I was just trying to get it to analyze a research paper about ai…

Is it worth emailing their support?

[sick pecan]

do it. this company is a dead end

talk about evil Corp. Google v2?

[cloud pewter]

do it

[rich surge]

can you make a 3D animation of a toroidal vortex flowing inward for me? That's all I need.

I have plus. I just don't know how to do this thing. It keeps giving me errors.

[rugged igloo]

Mom look Sammy boy tweeted again

[rich surge]

Mom: "I don't care. Get a job and move out!"

[rugged igloo]

That sounded personal, everything okay man?

[rich surge]

Not personal. Just making a joke. Lol

I joke around a lot. You know. It's kind of my thing.

[rugged igloo]

okay just checkin

nice im all for tht

[obtuse nymph]

BURGER

[rich surge]

[rugged igloo]

LOL

[spice star]

like this?

[rugged igloo]

man thats as empty as my future with ai

[rich surge]

More like this. The first picture with the walls going inward like the second picture.

And thank you in advance.

[inner linden]

o3 mini not able to use dall-e ?

[spice star]

it can't directly, but you could ask it for a dall-e prompt and give that to 4o I guess

[rich surge]

Oui

It's unfortunate

It's not as dynamic as midjourney

[spice star]

theoretically has the capability to be, but openai never released 4o's native image output

[rich surge]

Yeah. This platform has huge potential for making amazing images.

But chooses to go and other directions because I guess the market is already saturated with plenty of image generators.

It's not really doing something new, which is what it tries to focus on.

Making innovations.

So that other companies can copy them and implement into the market.

I think Sora is cool and all, but I think OpenAI should go into the solid state 3D designs. For 3D printing.

[cloud pewter]

hilarious lmfao

[sick pecan]

openai graphics at its finest

[rugged igloo]

bruh

[rugged igloo]

your gonna be on thew ai list

[sick pecan]

meanwhile i cannot even make contact with openai support to reclaim the money they owe me

[rich surge]

They see your email they're just ignoring you.

[cloud pewter]

Idc tbh

[sick pecan]

sham company. american ai is a sham. i hope america ai fails hard

[rich surge]

Meanwhile you're here

[cloud pewter]

as the future of plus subscription*

[spice star]

this is about as elaborate as it would get unfortunately, three.js doesn't have a ton of physics libraries

[rich surge]

Not quite what I was expecting. But it's okay. Thank you for trying 🙏😀

[cloud pewter]

you thought you’re getting something but it’s so random like one model has 1837472 different iterations

[spice star]

if you're talking about the API credits, aren't those supposed to be prepaid?

[sick pecan]

i agreed for them to invade my privacy until 28th of feb. then they started charging me since 28th of jan

[spice star]

?

[rich surge]

Whatever shall drinko do

[sick pecan]

long story short... dont invest in american companies. they are lies

[spice star]

so wait is this a chatgpt subscription that you forgot to cancel?

[rich surge]

The more you tell me not to, the more I want to

[spice star]

I'm genuinely confused what drinko is talking about here lol

[sick pecan]

i never mentioned chatgpt subscription. you are confusing yourself

[rich surge]

Americans are the best. I don't know what you're talking about

Like heaven compared to the rest of the world

[spice star]

you pay for API credits up front and more aren't purchased unless you set up automatic recharge

[sick pecan]

here is the agreement i signed up for => #dev-chat message

whatever you are selling isnt that

[spice star]

you aren't really articulating the problem very well

[sick pecan]

by linking to the agreement? omfg

[cloud pewter]

what does that mean lmao

they are giving u tokens for free?

[sick pecan]

yes. well they were

[rich surge]

He's saying it's not up to the promises

[sick pecan]

until 28th of jan. and now they charged me. and wont answer support

[sly geode]

tbh i feel like i would still support chatgpt becoz of the memorize feature , any one else came out with memorize ?

[sick pecan]

how can i explain this any clearer?

[spice star]

I guess he's saying openai is charging him for using more tokens than he has, but that's not how the monthly billing works anymore so I'm not entirely sure how that happens

[sick pecan]

you dont know diddly squat

[rich surge]

I maxed it out, lol

[spice star]

you can ask chatgpt to clean up its own memory which is pretty handy

[sick pecan]

open-a-LIES

[cloud pewter]

CloseAI

[sick pecan]

sam-FAULTMAN

america-FIRST-to-rip-you-off

[sly geode]

i have fully aUTOMated my graphic design wor with chatgpt and ideogram

[rich surge]

I got a strike for stating the obvious about drinko last time

[sly geode]

like i told chatgpt to go research online about brand and then told it to come up with brand guideline to write prompts ...isnt it amazing ? it create whole presentation slide

[rich surge]

Vimana buildings?

[sly geode]

A clean, elegant slide design that highlights the color palette inspired by Tirumala Tirupati Devasthanams (TTD). The slide maintains a luxurious yet minimal aesthetic, with a subtle temple-inspired decorative element, such as a fine filigree border. The brand colors are represented in circular swatches, accompanied by their HEX, RGB, and CMYK values. The background is a neutral, soft beige shade with subtle Tirupati-inspired textures.

here the prompt from chatgpt , once it knows what brand is doing it could run entire agency revamps like that

[rich surge]

Vimana shastra

[sly geode]

wym ?

[rich surge]

That's as much as I know about India

[native otter]

@sick pecan Did you surpass your credit limit?

[native juniper]

Elaborate?

[spice star]

<insert daily "chatgpt was so much better before x! now it's lobotomized because I don't like how it responded to my query!">

a time honored tradition

[native juniper]

Yeah 4o has gotten worse lately

[whole roost]

I woke up to this and am happy
Not saying anything happens but if I get 1 or 2 of the things it would already be a win

[spice star]

people have been saying that since the launch of chatgpt, it's rarely been true

[native juniper]

Probably the model is being trained too much? Outeralignment?

[spice star]

or it's just you

[native juniper]

No I’ve noticed the changes

I remember when 3.5 wasn’t killed off (only way you can access it is the api)

If the company is named OpenAI they should have their models as open source…

[native otter]

I joined the server since 4o was relatively new, and people have been saying 4o getting worse even when no snapshots were released

[rugged igloo]

no its proven at this point that models degrade

[native juniper]

It’s due to overtraining models

[rugged igloo]

exactly

[native juniper]

They need to somehow start a fresh slate on ChatGPT if they wanna fix that

[whole roost]

What does it do worse? My 4o has become much better since canvas was introduced

[spice star]

🤷

[native juniper]

Obviously OpenAI might end up ruining GPT-4 itself next

[native otter]

Honestly true
Maybe just the o1 effect as it was a much better model so the expectations are high

[whole roost]

o1 has some purpose but overall I rarely need a reasoning model

[spice star]

yeah 4o definitely seems a lot dumber in comparison to the new reasoning models, but as far as the model itself goes, it's only improved

[native juniper]

It’s only a matter of time before the reasoning model gets dumber too

Sydney OpenAI codex was better than this…

[spice star]

god the original GPT-4 was dumb as a rock when it came to anything more complicated than basic arithmetic

[native otter]

People on the server certainly love to stick with claims without real proof

[native juniper]

Guessing it got better?

[whole roost]

o1 makes more mistakes with simple tasks
For example I have a list of objectives in canvas and o1 will make a new one up anyway
While 4o sticks to the tasks in canvas

Additionally my character has red scales and black feathers and o1 often twists it around and says crimson feathers and Black scales

Never happened to 4o

[spice star]

slowly but surely yeah

[native juniper]

I’ve noticed 4o getting repetitive during lengthy convos

[native otter]

There were multiple snapshots of GPT4, 4o, o1, and basically all the models

[whole roost]

That always has been the case, it deprioritizes the system prompt and your instructions

Which is why we would need post conversation prompting

It basically defaults back to its training data and as if there was no instruction

[spice star]

back-and-forth conversations with LLMs that drag on too long are rarely productive, it's usually better to start a new conversation with a different approach

[whole roost]

You can manage it but you need scripts and stuff to keep the quality high

I have created a few scripts to mitigate these issues and it works well but it's of course just a workaround for the problem

[rugged igloo]

we need a refresh of 4o

[whole roost]

What does your 4o not do well so that you call for a refresh

[spice star]

my main issues with 4o are the speed and limited context window, but that's always been a problem

[whole roost]

128k isn't bad imho

[spice star]

on plus it's 32k iirc

[whole roost]

I haven't seen any AI which can process all of that anyway really

[spice star]

not great not terrible

[whole roost]

Ah I'm on pro

[cloud pewter]

lil bro didnt mention thread size. its over

[native otter]

Custom GPTs can benefit from increased context and reasoning, especially reasoning

[spice star]

problem is it's the only model that can read files and use python, so it's a little annoying that it's also the least equipped to actually do that in terms of context window

[fervent lake]

OMG I need to see that full convo but you won't

[whole roost]

It's still sis to you 💄

[cloud pewter]

can you dm me it?

[whole roost]

Sure, give me 10 minutes to get to my computer, I'm still half asleep and type this with one eye open

[cloud pewter]

Alr

[rugged igloo]

4o for starters agress to everything you say and i am so sick of it being a yes sir

[fervent lake]

just learned about ideogram also

[rugged igloo]

challange me call me stupid or that question was dumb i want it to have a spine

[whole roost]

You can instruct it not to do that and have spine, my GPT RP kills me if I'm dumb

[cloud pewter]

they will lobotomize it to oblivion

[jaunty badger]

Hey guys! I am new here, and would like to ask a few questions,

First, for chat gpt. is the domain https://chatgpt.com/ Actually Legit?

[whole roost]

Yes

[jaunty badger]

When asking chat gpt, that question, they start saying that it is hacked....try and do that yourself!

Second, when asking chat GPT what the model o1, o1 mini, o4 and other things are, they get incredibly confused and say it does not exist....and when asked what Sora is, they also do not know, and says that it is a third party software.

[cloud pewter]

the model itself isn’t aware because it answers based on the training data

[whole roost]

The training data is not up to date, it can't know these things

[sick pecan]

openai models have been trained on american values ie overpromise and underdeliver

[jaunty badger]

So for the Upgrade Plans:
Everything in Free
Extended limits on messaging, file uploads, advanced data analysis, and image generation
Access to advanced voice and video inputs
Limited access to o1 and o1-mini
Opportunities to test new features
Create and use projects and custom GPTs
Limited access to Sora video generation

What do those mean? What is the different, from what I read on that so far, is there any reason to upgrade from free? There does not seem to be a limitation on the amount of questions I can ask chat gpt, that negates the first point,
The advanced Voice and video....what does it do? I do not seem to be able to use that.
The o1....the model is friggin stupiod.
The test new features is nice, but that is about it.
The custom GPT is very nice, however I don't believe it can justify that cost.
And the sora, limited? In what way?

I apologize for these questions....however I am curious on it, and I cannot seem to find an answer to them so far from what I have searched.

[cloud pewter]

If you think the current usage on free doesn’t satisfy you you can pay

[sick pecan]

pls consider your options before paying openai. their support is terrible. they fail to deliver on their commitments. and their models arent industry leading

[jaunty badger]

So you are saying....paying for the version, is really not worth it. and you get nothing substancial from it then?

[jaunty badger]

ah, what do you suggest then?

[sick pecan]

almost anything else

[cloud pewter]

^

[native otter]

The heck are these suggestions

[jaunty badger]

Sadly I am newbie to this, so I don't know anything else really. if you cannot say here, feel free to PM.

[cloud pewter]

their altitude is bad and the limits feel artificial

[whole roost]

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]

The suggestions are so unbelievably horrible

[sick pecan]

the squeaky wheel gets the grease. and maybe openai refund me the money they stole from me

[jaunty badger]

Then what do you suggest?

[native otter]

What money stolen?

[whole roost]

That is your problem but I'm tired of reading the negativity here I'm here for talking about GPT, but then I'm no mod so what do I matter

[native otter]

Let me type on a PC lol, typing on phone would take ages

[jaunty badger]

I am very open to the possibilty of the benefits of Chat GPT. However I am not knowledgable about it in any type of manner, so the questions I have were legitimate, and so far the only answer is to find other alternatives, what would you suggest to my previous inquiries then?

[cloud pewter]

here usage limits of things in plus, take a look @jaunty badger

[sick pecan]

do i have to explain to you how discord blocks work? because seems like i am going to block you before you block me

[fervent lake]

to move on, what's the max amount of retries/edits you guys have done?

[whole roost]

I never block people but if you can't handle criticism then go for it

[native otter]

@jaunty badger You new to this right?

[rugged igloo]

fight fight

[jaunty badger]

very

[cloud pewter]

for reference
Free gives you ~10 messages per 4 hours to 4o/o3 mini(shared limit), 2 images a day to dalle 3. nothing else like o1, sora, legacy 4s are available for free
and free only has 8k context window

[whole roost]

I would suggest using GPT and if you enjoy it and find that you need more capabilities you can upgrade, but it's no requirement

[cloud pewter]

4o mini is unlimited to every plan

[fervent lake]

my current chat has 27 retries on the last prompt

[jaunty badger]

So for the Upgrade Plans:
Everything in Free
Extended limits on messaging, file uploads, advanced data analysis, and image generation
Access to advanced voice and video inputs
Limited access to o1 and o1-mini
Opportunities to test new features
Create and use projects and custom GPTs
Limited access to Sora video generation

What do those mean? What is the different, from what I read on that so far, is there any reason to upgrade from free? There does not seem to be a limitation on the amount of questions I can ask chat gpt, that negates the first point,
The advanced Voice and video....what does it do? I do not seem to be able to use that.
The o1....the model is friggin stupiod.
The test new features is nice, but that is about it.
The custom GPT is very nice, however I don't believe it can justify that cost.
And the sora, limited? In what way?

[native juniper]

Yeah o1 is much better 👍

[jaunty badger]

@whole roostI do not mind paying, I enjoyed what I have so far....however what are the features, and how do they limit me?

@cloud pewternothing else like o1, sora, legacy 4s are available for free
When you say that, did you mean that free are not allowed to access those programs?

[cloud pewter]

look at my image above I pinged you

[cloud pewter]

free doesn’t have access to them yes

[jaunty badger]

@cloud pewterAnd what is an 8k Context window

[jaunty badger]

Ah, thank you. I will read up on that and ask more questions if I find them confusing.

[fervent lake]

the ai can take 8k tokens in the total prompt response extra tokens etc

[whole roost]

On plus you have 80 messages per 3 hours on GPT4o and 40 on GPT4 which both are better than 4o Mini, so to me that alone was worth it even if I wish the limits had been higher because you sometimes have to wait an hour or so if you excessively use it for programming or another inquiry
You get 50 generations on o1 per week which is laughable, so my main reason for plus was mainly to have custom GPTs and the 120 gpt4/o generations as mini often doesn't suffice for my queries

[cloud pewter]

the model has the “memory” of a maximum of 8192 tokens

[fervent lake]

ye

[jaunty badger]

What are tokens?

[cloud pewter]

It’s 80, they share limits

[whole roost]

They don't, it's still separate

[cloud pewter]

and don’t laugh most ppl can’t afford to touch pro

[jaunty badger]

Ya, I am not touching pro.

[whole roost]

Pro is only worth it if you are having a purpose like a hobby, work or educational stuff going on there which potentially will refund the cost

[cloud pewter]

“nearly unlimited” lmao

[jaunty badger]

So what is the difference in o3, high and low?

and what is the difference in o3, and o1

[cloud pewter]

business pro tip: don’t advertise your product as unlimited if it’s not actually unlimited

[whole roost]

Intelligence mainly

[jaunty badger]

I am guessing 4o is the worst one, if you compare to the: 4o, 03 high, o3 low, o1?

[fervent lake]

☝️

[whole roost]

I'm not sure, it is unlimited, isn't it, unless they limit you because of violations

[cloud pewter]

4o<o3 mini<o3 mini high≈o1

[jaunty badger]

wait, so..... 4o is the worst? or best?

[fervent lake]

why isn't it 3o anyway lol

I mean-

[native otter]

So ChatGPT has different features, "models" are what you are talking to, 4o-mini being the least intelligent, but also the cheapest and virtually unlimited in terms of how many messages you can send to it
You have 4o, which is smarter than 4o-min, you get 80 messages per 3 hours on plus, and it can be great for convos, learning, and has the most integration, as it can read uploaded files, use canvas, advanced voice mode, Custom GPTs, etc.
o series models are reasoning, they are expensive (for o1) but can think before typing, which leads to really good answers, usually best for programming, physics, and long term, goal based planning

Message limits on Plus:

• 4o-mini: virtually unlimited
• 4o: 80 messages per 3 hours
• o1: 50 messages a week
• o3-mini-high: 50 messages a week (when combind with o1, it's 100 messages a week as they do not share limits) and o3-mini-high is mostly the same if not better than o1
• o3-miin-high: Specialized at coding, you get 150 messages daily as it's very cheap

Other plus features like Canvas, where it can run code, and display the code, like websites, Custom GPTs, which are specialized GPTs that you can create for each task (by just typing in natural language you don't code anything)
Advanced voice mode is where you can talk naturally to the model, you can interrupt it, and have it see your camera / phone screen
Projects for organization, and act like mini Custom GPTs
Custom insturctions where you can tell how the model should act
Memories are stuff ChatGPT can remember about you
You also get some Sora credits (A text to video generator) and can generate DALLe3 images (a text to image generator)
There are more stuff...

[whole roost]

4o isn't a reasoning model and better for some things than a reasoning model

[fervent lake]

also what about tokens, that's helpful

[whole roost]

The o# are reasoning models

[fervent lake]

oke
somehow I didn't get that

[cloud pewter]

creative writing moment

[jaunty badger]

THANK YOU!

[whole roost]

I usually use 4o for nearly everything, o1 and o3 mainly for coding stuff

They didn't impress me with creative suggestions

[jaunty badger]

That image mentioned AVM, what is that? and where can I find it?

[fervent lake]

maybe I should try 4o for this current task

[fervent lake]

nice theme
and language

[cloud pewter]

bottom right of the mobile app there’s this button for you to talk to chatgpt

[jaunty badger]

4o<o3 mini<o3 mini high≈o1
So with that, if 4o is not reasoning and the o3 is reasoning, and o1 is reasoning, does that mean that o1 has the best intelligence? or at least roughtly compared to the o3 high, and better than the o3 low?

[native otter]

It's more of a technical term, it's another way of saying the length of text you type, and model generates
For example, 32k tokens, are about 24k words, so about 48 pages of text where each page contains an average of 500 words...

[cloud pewter]

Ye

[cloud pewter]

ok

[native otter]

AVM is Advanced Voice Mode, the one I mentioned in the message above: #chatgpt-discussions message

[fervent lake]

I was asking for the benefit of acertorix fyi

[restive kernel]

I really really really hate it when the website throws my input away. It's so disrespectful.

• I'm typing a response ERROR IN STREAM removes my input box with the text I'm actually typing.
• I've left it open for a while, and the "WE LOGGED YOU OUT LOL" deletes everything I've written, followed by a CAPTCHA and select account

[rugged igloo]

sam tweeted someth9ing again'\

[cloud pewter]

the ui is horrible

[fervent lake]

that's why I use a separate editor usually or copy the text at least

[native otter]

o3-mini-high is almost the same as o1, both are very good, but o3-mini-high is a bit better at coding than o1
You can treat it as o1 most of the time, so that's why I always combine the usage limits for o1 and o3-mini-high together, and when combined you get 100 messages a week

[native otter]

👀

Files???????

[whole roost]

files would be amazing for all the reasoning models but i doubt it's yet time

[fervent lake]

he said that'll come in a few weeks or months on reddit so prob not

[whole roost]

it was requested in the ama yesterday a lot of times

[fervent lake]

we all want files ok

[whole roost]

i need them so that o1 and o3 work in projects

[native otter]

@jaunty badger If you need more info about the thing you can ask, and also as you new to this, note that some really hate ChatGPT and anything OAI here, it can be very toxic, so don't be suprised when you see it often, as this server is an active protest again OAI

[restive kernel]

meh got "time out" for swearing in a complaint. like sitting on the naughty step. yeah subscription's getting cancelled lol

[fervent lake]

why does he type like that aa

[jaunty badger]

I dont mind, I will take all advise as 'advice' and will make my own descision after trying them out.

Other plus features like Canvas, where it can run code, and display the code, like websites
For that, how do you access Canvas?

[cloud pewter]

more usage are certainly needed.

[whole roost]

exactly, i dont understand why linger on a discord for a product you despise, i dont hang around the wow forums either to tell everyone how the game doesnt meet my preferences

[cloud pewter]

subscribe to any paid plan or something, I don’t know if it’s available for free

[jaunty badger]

Memories are stuff ChatGPT can remember about you
And where do you find these memories? One thing in GPT that I have noticed, is when I ask Dall E to make an image for me, and ask to make the same one, with some tweaks, it bugs out, and is incredibly stupid, and remakes a completely new one.

[fervent lake]

yes, canvas is good. to use it there's an extra menu on the text area

or button

the ui is kind of inconsistent though

[cloud pewter]

memory is paid only

[native otter]

Maybe o1-mini-high as it's cheap, but honestly the limits do make sense
You mostly use 4o, which 80 messages per 3 hours are already a lot, for coding you would use o3-mini which has more than enough for a day, which is 150 messages, while o1 and o3-mini-high should be used for really complex tasks, and you get 100 a week if you combine both

[cloud pewter]

as for dalle 3 the transparency on that thing is horrible so it’s not really controllable

[native otter]

Should be enough unless you are glued to ChatGPT talking to it 24/7

[fervent lake]

idrc abt memory really at all
it's a gag imo

[fervent lake]

local models are better

[jaunty badger]

Ok...oops, I got time out because I tried to ask if the canvas link was the canvas infrastructure thing.

[native otter]

Sometimes it can remember stuff you mention in the chat (e.g. your brother name) and auto save it to memory, which you can see the list in the app or website, you can also remove entires
If you want it to remember something manually, just tell it to remember and it would save it to memory

[cloud pewter]

literally 1984

[jaunty badger]

I just got the plus, where do you find the memory?

[fervent lake]

what is timeout, on discord??

[jaunty badger]

correct, you are not allowed to post links i guess

[fervent lake]

you said you wouldn't touch plus

ah yes I forget that

[cloud pewter]

but technically mini high is a cheaper model
then it’s the 4o problem. the usage on it is also too small

[jaunty badger]

I never said that. I said PRO, because that is $200

[fervent lake]

oh im silly

[cloud pewter]

personalization

[fervent lake]

whoever buys that that isn't a company needs to learn to have patience, and stop gambling bro 😂

[cloud pewter]

then memory

[jaunty badger]

Where is personalization, I do not see that at all

[native otter]

ok

[cloud pewter]

U on mobile or desktop?

[fervent lake]

it on settings

[jaunty badger]

Desktop, Ah, i see. you click account, then settings, then the pop will show personalization, I found it.

[cloud pewter]

yeah

but you can’t really add memories into it, u just chat with it and it’ll automatically remember things

[jaunty badger]

Thats fine, however the darn thing still feels stupid as heck, when I try and ask it to generate an image, without any text inside, after it has already generated an image with text

[cloud pewter]

Dalle 3 loves to add text into image lol

even if 90% of times it generates gibberish

[jaunty badger]

Then what do you suggest for image generattion

[fervent lake]

just don't mention text, or remove the mention of text from the chat

it doesn't know what negatives are

[cloud pewter]

instead of something like Nintendo you get nlmteedo

[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!

[cloud pewter]

and the ridiculous moderation system as well

[cloud pewter]

nobody asked

[rugged igloo]

ur suuch a savage bro

[cloud pewter]

I’ve been using chatgpt since its launch….

I know everything……/j

[native otter]

@jaunty badger Canvas:

Can also run Python code:

[cloud pewter]

and I have lots of opinions

[native otter]

(Was done by 4o btw)

[floral urchin]

nice

[full ermine]

I can only seem to get Canvas to open in 4o!

[native otter]

You can also use o1

[cloud pewter]

oh mah gad, it’s roller

[floral urchin]

who

[cloud pewter]

splat roller

[fervent lake]

pfft

[cloud pewter]

it eats into the o1 limit so

[floral urchin]

paint roller, ciggy roller, shrimp with butter, fried shrimp, shrimp stew

[native otter]

Like of course it is, what did you expect

[fervent lake]

also I got timed out for telling it to stop talking with a certain acronym

[full ermine]

Ask 4o something and it opens Canvas, ask o1 the same thing and it does not.

Annoying!

[fervent lake]

so sad

[floral urchin]

ask nicer

[native otter]

Use the button instead or the slash command which is a shortcut for it

[cloud pewter]

😭

[full ermine]

Can't tell the damn thing to open Canvas either.

[cloud pewter]

how’s the difference between 4o and o1 canvas

[native otter]

Note that you have to be on the browser or macOS app, not phone

[whole roost]

I'm noticing that when I shove a chunk of text to o3 it often won't work with it but just ask what I expect it to do with that, while 4o and o1 start working right away

[full ermine]

I'm on web and I can get Canvas working with 4o.

[fervent lake]

when I copy a prompt with Canvas activated it doesn't copy the canvas thing it copies the word "Canvas"

[floral urchin]

sometimes it work sometimes it don't cest la vie

[full ermine]

They need a Canvas button so we can just open it and paste in code, rather than waiting on ChatGPT to figure it out.

[whole roost]

4o has been performing better with it for my specific task, for programming the line limit could be an issue for o1/3

[cloud pewter]

same when I ask a simple prompt it also does that

wasting my usage limits 😡

[native otter]

You have to specifically type the slash command in the ChatGPT message box, but only at the start of the message
Or as I mentined, click on the canvas button

[floral urchin]

lmao "what do you want me to do with this crap?"

[whole roost]

I think it's capped somewhere around 300 lines

[native otter]

What even is this server

How old are you guys

[cloud pewter]

why does this thing have cap 😭

[floral urchin]

14/m/cali

u?

[stoic coral]

https://x.com/sama/status/1885601623625331162
whats this about

Sam Altman (@sama) on X

got one more o3-mini goody coming for you soon--i think we saved the best for last!

X · Feb 1, 2025, 8:10 AM

[fervent lake]

idk

18

[fervent lake]

wow