chatgpt-discussions

[whole roost]

I haven't really read why this is just that it is and I haven't read about any plans that it is supposed to be accessible anytime soon

I will just expect that it will not ever come to Europe so I didn't really bother with Sora

[inland heron]

it's available for all models it seems, but only tested it on 4o vs o1 (non-use more intelligence)

yes

I'm just starting to use it a bit more. I have no idea

[rocky kayak]

what does it do

[inland heron]

I tested a medium-long context one-shot coding prompt requiring a solution on o1 (non-intelligence selected) vs 4o (intelligence selected). o1 won, double-checked. Both gave solutions, but the 4o solution wasn't as modular and maintanence-friendly

[inland heron]

It uses more intelligence 😄

maybe it doesn't do anything, who knows

[rocky kayak]

Placebo maybe?

[whole roost]

As far as I see it it just uses o1?

[inland heron]

possibly, especially given how there's not explanation on this

[cloud pewter]

weird regulations

[inland heron]

how do you know it uses o1? Just because it 'appears' it's thinking?

[whole roost]

It changed the title for me and switched to o1 when I clicked it

[winter vine]

We should try the same prompt with the same model, one with intelligence and the other one without it.

[whole roost]

And it cannot access the memories

[inland heron]

that's the best way to test it, but I guess no one is willing to waste o1 just to test this lol. I honestly can't see the difference

yup

it's tehre regardless of the model

[whole roost]

It is just o1

[winter vine]

I can use it with 4o

[winter vine]

I meant testing GPT-4o with and without the intelligent mode.

[whole roost]

My chat used to be 4o as well

It seems to be nothing more than a model switcher

[inland heron]

so 4o becomes o1, and o1 becomes what?

[whole roost]

o1, nothing changes

[inland heron]

how can you tell it's the model switcher when we don't have access to it's code and mechanism?

[whole roost]

Because it literally writes that it uses o1

[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.

[rocky kayak]

wut

[inland heron]

I guess that's a piece of the puzzle. But then again, since when do we trust what any GPT says? 😄

deepseek says it's chatgpt, 4o says it has multiple knowledge cutoffs, all kinds of answers

[whole roost]

I am not talking about the answer it gives but the model that is selected

[inland heron]

ah, I understand now. It does say it's o1 even though I have 4o selected

[quick galleon]

Yo

[whole roost]

Exactly

[inland heron]

but then what does o1 do? and why isn't it smart from the start?

[winter vine]

My ChatGPT works this way: With any model, I can press the lightbulb button to give it a bit more intelligence, regardless of the model I choose.

[inland heron]

and what does o1-pro do? this is even a greater mystery

[whole roost]

o1 is a reasoning model and 4o isn't

o1 follows instructions much better

o1 pro has more thinking steps, often leading to better solutions

[inland heron]

OK, but there is greater difference between 4o and o1 than just reasoning (as I've compared them, and o1 is smarter than 4o using o1). Maybe 4o is using some basic principle of reasoning?

[whole roost]

It usually takes around 1-3 minutes for me to reply

[cloud pewter]

waht if the pro mode is actually the real o1 and the regular o1 is just the lobotomized version

[inland heron]

or maybe it's just a weaker model and no matter how much reasoning it does, it can never beat o1

[whole roost]

Yeah reasoning is much superior as it can include more data and different angles

Yes it uses o1

[inland heron]

not always though

[whole roost]

It doesn't need a thinking step when there's nothing to think about

[inland heron]

I just realized 😄 you're right

there's nothing to summarize, but it did do some thinking for 6 seconds

[whole roost]

They only are instructed to know that they are a GPT model by open ai

Version doesn't matter

Yeah it cannot access the memories either if you let it think if you don't trust the model selection which plainly writes that it's o1

[inland heron]

I accept this change 😄

[inland heron]

time to do some work

[cloud pewter]

pro user alert!!1!!1!1!1!!!!1!1!1!1!1!1!1 Run plus peasants!1!!1!1!1!1!

real!

plus peasants:
will you cancel your subscription after this one ends? or keep it?

[whole roost]

I need it for important things!

[grizzled steeple]

Does anyone know how to successfully built in ChatGPT into wireless smart speaker?

[cloud pewter]

I love the audacity 🥰

Adblock plus engagement active

real!!!

and they had to give us 100 o3 minis a week but changed it because of Deepseek. Good!

[whole roost]

Meanwhile I'm crafting an o1 roleplaying project against boredom at work 💸

[cloud pewter]

o series and sora are the products im too afraid to touch because of the absurd limits on plus

[whole roost]

Yeah it annoyed me as well, I needed o1 mostly for writing scripts but the 50 message limit often didn't suffice to fix stuff

But now I'm done

[cloud pewter]

huh

why’d I got muted by automod over a random sentence

I love this server, can’t even let me complain

[cloud pewter]

just when I find the actual usage for these things I got hit by limit and had to wait until someday next week

[whole roost]

I hope that they will bring canvas to phones some day

[whole roost]

How did you get red with sparkle?

[cloud pewter]

get your thing featured in #spotlight
but it’s just purely a cosmetic role, doesn’t give you extra perm

[whole roost]

At least it's red

[cloud pewter]

red

and ✨ sparkly✨

I hope we get more limits on o3 too. unless o3 mini is significantly better than o1 and r1 then openai would be cooked if they slap another absurd limit on regular o3 for plus while others are just free

I don’t want another 50 per week

[whole roost]

I mainly need it to be able to read uploaded files in projects

[cloud pewter]

Real!

I’ll cancel my plus if openai keeps doing these, and I’ll subscribe to Twitter premium

[rocky kayak]

mf is gonna name tiktok "Y" if he buys it

[cloud pewter]

and name openai Z if he buys it back

👏

yeah the audacity of adding this button on this popup, bottom left of the website and inside the settings of mobile app

[rocky kayak]

fr

[cloud pewter]

it was only the teams promo on the bottom left but it now shows
“view plans: unlimited access and more”

it’s actually over

I want to see more plus users complaining about the lack of quality of this paid plan. we must raise awareness about it. it’s serious

or should I say, addressing the issue?

I can say that my dms are open for complaints about plus

[real bluff]

guys is the o3 supposedto be released today

also whats that button here?
just activate the o1 model?

[cloud pewter]

yes

[real bluff]

so it would still waste me o1 points i assume

[cloud pewter]

yeah. idk why are they doing this instead of increasing limits

[eternal rock]

**`` Important Notice ``**

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

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

[real bluff]

well they got no money i assume

this whole things is expensive and now with that deepseek if its actually legit then they are kinda screwed

[cloud pewter]

oh, they have money. 500 billion in the next 4 years. it’s just their priorities are screwed

[real bluff]

what are their priorities?

[tardy isle]

I suspect that's "DeepThink" button reference from DeepSeek...

[whole roost]

Or they could make o1 actually good and use all features there are

[tardy isle]

Well, that button is good for me, since I frequently waste o1 as I forgot to change main model back to 4o. Just toggling that button seems good.

[quick galleon]

Lol

[tardy isle]

Does using GPT-4o mini eats up GPT-4o usage?

I know GPT-4o and GPT-4 shares limits, but mini and 4o sharing limit seems wierd...

[whole roost]

For me 4 and 4o never shared limits

I always had like I think 40 on 4 and 80 on 4o

[tardy isle]

When I using 4o, message pops up like "You've hit the Plus plan limit for GPT-4."

[whole roost]

I always used up my limit for gpt4 first and then it would route me to 4o

So it always was some sort of 120 messages per three hours

No I'm talking about gpt4, the old legacy thing and 4o not mini

[native otter]

4o mini uses are different from 4o and 4

[stiff herald]

I'm still working on my first cup of coffee, don't mind me.

[native otter]

Seeing the messages above, god, this server is cringe

[frail scaffold]

💀 I just died laughing, chatGPT advised me to take the money printer

[quick galleon]

.

[stiff herald]

I like its gumption.

[vale compass]

what does "use more intelligence" mean

[cloud pewter]

But they does share limits

when I use a lot of 4o the 4 would become gray

And 4o mini is unlimited across all plans because it’s cheap as heck and it’s only 8b

[stiff herald]

That it taps into the five chimps's it's linked to's neural link and accesses their own processing power.

[vale compass]

i knew it! that was my guess

[stiff herald]

o7

[whole roost]

They never shared limits for me and I had plus from last year in February to November I think

[vale compass]

I use it after a prompt correct and it will revaluate everything?

[tardy isle]

It just uses o1.

[vale compass]

I know

[whole roost]

GPT 4 had significantly lower limits I don't remember if it was 40 or 60 though
And when they are spent it routed you to 4o

[cloud pewter]

4 isn’t available if you used up all of 4o

[whole roost]

Then they must have changed it the last two months because before it was definitely like I say

I used it everyday at work every three hours to the limits first gpt4 and then the limit for 4o

[eternal rock]

Get notified for server events!

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

[vale compass]

is it true alibaba dropped a AI model?

[whole roost]

It doesn't seem to have changed and still is separate

[vale compass]

is 4o smarter or 4

[tardy isle]

It seems Qwen-2.5 dropped

[cloud pewter]

you’re a pro user. step aside.

but pro also only gets like the same treatment as plus on 4

[vale compass]

🙏 200 is insannee

[cloud pewter]

On dalle too

[whole roost]

I never really use dally it's more a gimmick to me

[vale compass]

why does o1 sometimes not get the same number a calculator does like it would be close but slightly off

[whole roost]

If you want gpt to be able to calculate you have to tell it to use python

Or else it will just respond like a llm and not like math tool

[honest jay]

It has to do with how all LLMs are semi-random in nature (as opposed to a calculator, which is completely non-random). o1 is better because it "thinks things through" for longer, but still an LLM at heart

[tardy isle]

o1 is language model, and it can't use Python. It does math quite well in terms of logic, but weak at specific calculation.

All LLMs just chain 'natural-looking words(tokens)', not seeing numbers or specific calculation.

[real bluff]

Tho does the pro actually get them the money?
It’s 10 times more expensive but are there enough pro users to make it worth?

Cause average person wouldn’t pay for this

[tawny crater]

Is it degrading again? Its forgetting context after 2 entries and continuously reverts to cinematic writing style despite being perfectly fine yesterday.

[whole roost]

Sam Altman Said they are losing money due to pro because people are using it more excessively than anticipated

[cloud pewter]

what if they aren’t losing any money

[whole roost]

What then?

[grand lantern]

they lie 😡

[grand lantern]

also that's bs lol even if they WERE losing money they shouldn't have made the limit so high in the first place.

[whole roost]

He wrote on twitter that he set the costs himself because he expected it to generate money

But didn't work out according to him

Which I can definitely see, I'm sure I'm costing more than 200$

I don't think users can ever pay the costs of running an ai

Which one

I have tried deepseek if it's about that and it's not bad but not better than GPT

The benchmarks do not concern me if it is unable to complete my tasks

[sudden herald]

I agree.

And there has been some talk about the benchmarks being faked but lets move to #ai-discussions

[grand lantern]

are you writing good prompts? what do you use it for?

[sudden herald]

This has nothing to do with preformence if ChatGPT can understand.

[whole roost]

Mostly roleplaying, creative work and programming I'd say, really everything I think

Canvas is the best thing that has happened so far as well

[grand lantern]

canvas isn't that useful to me

[sudden herald]

Let's move to #ai-discussions

This is off topic.

[whole roost]

How is canvas off topic for GPT?

What are we supposed to talk about in here

I don't mind moving but curious what this channel is for then

[sudden herald]

No not canvas but deepseek.

[winter vine]

I don't have good results with canvas, maybe im using that wrong

[cloud pewter]

I roleplay/do creative work using chatgpt as well.
I can’t justify pro sub. it’s just that expensive and plus is just that limiting, both usage and context limits

[whole roost]

Canvas is the holy grail for Roleplaying in dungeons and dragon style even, it can follow along tasks, create a pseudo memory glossary and tick off checkmarks for story progression

Just one use case for it if one likes roleplaying

This just switches to o1

[eternal rock]

`` Server Tip: ``

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

[winter vine]

Sounds great. I use chat gpt for programming (react) i dont use canvas, maybe i should try it

[cloud pewter]

not to mention the thread size…

countless long stories of mine deleted itself because of that

[whole roost]

o1 with canvas could be very good, 4o on canvas programming didn't impress me

I completed most of my scripts on o1 before it was able to do canvas

[sudden herald]

Yeah I just use canvas with o1 now.

[winter vine]

you convinced me, I'm going to try it for programming. Changing the subject, when is the O3 Mini coming out? 😦 I'm hyped.

[whole roost]

i don't really know of an ai which is actually able to handle their large context windows, gpt also struggles with its 'small 128k' limit, the prioritization isnt good on any model but gpt does alright with projects and files

[sudden herald]

The original timeline the dev team gave me was this week so possibly today if it hasn't changed.

[whole roost]

I'm still waiting for this 🫠

[visual acorn]

How is "Gen Z" a trait?

[whole roost]

fr fr

[sudden herald]

Doesn't this exist already?

[visual acorn]

🇫🇷🇫🇷

[whole roost]

I wish, I still am on the old thing

[visual acorn]

What thing?

[whole roost]

I am using all AI except for claude really, I'm just not very impressed by them

[visual acorn]

Btw I've talked with DeepSeek and I've noticed it's not quite as good as ChatGPT for certain topics

[whole roost]

GPT struggles a lot with sticking to instructions in longer convs, i wish it had something like post conversation prompting

[sudden herald]

Remember guys if you are going to talk about Deepseek it has to be in #ai-discussions , thats just the rules of the server.

[visual acorn]

Can I not talk about it while comparing it to ChatGPT?

[sudden herald]

Yeah I have it.

[sudden herald]

You can compare in #ai-discussions

[whole roost]

I wish, I want to try if it does a better job with those personality tags

what's under advanced? just the tickmarks for the features?

AND what's the symbol limit

[sudden herald]

Yeah

[sudden herald]

I don't know, I haven't used it.

[whole roost]

would you mind to copy in some random symbols and see if it's more than 1500? If they increased it it would potentially be great for my purposes

you don't have to though, just curious

[sudden herald]

Let me see.

it supports 1500 characters per box.

[whole roost]

damn, no increase, thanks for trying, projects it is then

[runic mural]

100 rather to be in the free plan

[native otter]

Yea either go Projects or Custom GPTs for that

[untold seal]

some people said o3 mini will be released today is there any confirmation?

[sudden herald]

No, it was changed because people complained it was too small.

[native otter]

Amazing how Future Machine has been negative most of the time around this server for some months now

DeepSeek may just be a distillation of OAI models as discussed earlier, so if that's real, it would mean it's basically stolen = not a real compitor because it would be a pirated model in a way
o series is still pretty new and quite expensive, so the limits do kinda make sense, but maybe o1 could start increasing a bit from 50 to 100 messages a week or so

[restive plover]

It’s not being released today?

[sudden herald]

No, I was talking about how at release Plus subscribers are getting 100 o3-mini messages a week, but that doesn't have to do with deepseeks release.

[native otter]

Sam said they would be 100 a day in another tweet

For Plus that is

I'm honestly not too excited for o3-mini, I care more about o1 and o3...

[whole roost]

projects, as you can edit stuff without having to start over the conv, custom gpts currently have no advantage for me

well you can turn off dalle, that's one

[placid whale]

isn't deepseek inference 95% less costly on resources than OpenAI?
If so, OpenAI should thank them for the advancement, and build in there code, to make themselves cheaper with a bigger model, and that will be the end of deepseek.
if not then OpenAI should build a bigger more expensive model and that will be the end of deepseek...

What Is The Problem?

[sudden herald]

Where do you get that inference number from?

[native otter]

Not sure about you, but considering my values, I would never support plain stealing like that...

[placid whale]

Hmm, good question... Am I sprouting misinfo...
I'd have to look it up...

[whole roost]

i think competition is healthy for any product, no hard feelings towards any ai provider

[cloud pewter]

real

[cloud pewter]

what does this observation bring you

[native otter]

A case study on how people will always try to focus on the negative parts and be so dedicated to stick with something they don't like :P
I do acknowledge that OpenAI does some annoying stuff, for example, the whole AVM thing coming in the next weeks that turned out to be multiple months, hyping up a product that's not ready yet, same for sora, and 4o-image gen
But honestly other than that the service is quite nice, ChatGPT be very useful if you know how to use it well, knowing it's limits, and treating it as is, without having extreme expectations...

[cloud pewter]

I’ve been trying other ai products as well. it’s just that openai’s altitude compared to other companies

[native otter]

They do hype up stuff a bit too much, and sometimes it blows on their faces

Like the whole AGI marketing buzz

I find myself using ChatGPT, NotebookLM, and sometimes Gemini if I need larger context, they are most useful AI services I tried yet

[cloud pewter]

Agi? turns out to be endless “coming in the following months”, “safety” and “red teaming”

And of course $200

[native otter]

Copilot was amazing at first, it was basically a free GPT-4 at launch, but it slowly began becoming worse, then I just swtiched to ChatGPT

[cloud pewter]

companies love lobotomizing stuff

[native otter]

it was actually my first experience with LLMs, and was the only option when ChatGPT and Bard (Gemini) still weren't available in the place where I live

ChatGPT only got available late 2023 or so

[cloud pewter]

for cost and safety issues yada yada

[native otter]

Copilot always searched the internet with every prompt
I could argue it searched it better than current 4o search tool does

The answers were natural, and often accurate + always mentioned the sources of each claim

It's getting better again, the new UI is quite nice, and generally started getting smarter like how it was before, but they are very late to competition, like current Copilot is more basic than ChatGPT in early 2023

They also removed the integration with Windows, which made it less useful

M$ being M$ I guess lol

[zenith crescent]

Did someone say DEEPSEEK

[cloud pewter]

current 4o search is mini based

[cloud pewter]

the copilot is basically just chatgpt wrapper no, but wrapped very deep

[native otter]

Yea it was amazing at first, but unfortunately too basic now, ChatGPT, Gemini, and Claude are much better in my opinion...

[cloud pewter]

real

anyways if o3 mini gets delivered tomorrow, good. but don’t make this a slow rollout again

after all I just want a better gpt.

I am no professional

[whole roost]

alright, forget about o1 roleplaying 🫠

[cloud pewter]

💀

[whole roost]

triggered by

need grown up mode

this is really not violating anything

[cloud pewter]

sometimes the orange flag pops up even if it’s just some swear words

[whole roost]

yeah orange don't bother me but o1 won't respond at all

"No need to find more relics?" I raise my eyebrows.
<[Lorebook: onyx watchtower] A trader's way station overrun by thieves and murderers.
Located in the center of the Dry Steppes.>

[red bronze]

$200 and worth every penny

[whole roost]

4o only will occasionally trigger when its about obliterating foes but just orange, o1 is really harsh with the flags

doesn't o1 always show its chain of thoughts or what am i missing

it triggers on every regeneration so i think it doesn't like the murderers

[whole roost]

yeah it generates stuff without issue when i remove 'and murderers'

🙄

[cloud pewter]

if the word isn’t even in the automod filter here but gets flagged on o1, it’s over

[whole roost]

yeah it's next to unusable

now it triggers on <Respond from your perspective> what the hell

yeah it doesn't seem to like my rules

maybe i can rephrase them but it shouldnt flag at all

they are not commanding anything odd

[cloud pewter]

gonna love chatgpt itself roasting openai

[whole roost]

it's so hilarious if i wasn't angry

snorts

maybe it doesn't like the mentioning of 'roleplay actions' or whatever

[cloud pewter]

btw have you ever hit the thread size limit

[whole roost]

not yet

[cloud pewter]

if the chat becomes laggy on web it indicates that it’ll soon be

[whole roost]

it always is laggy on phone so i learnt to ignore that

[cloud pewter]

and if it does your chat will somehow vanish and becomes unobtainable unless you have the link

[whole roost]

usually the quality of the conversation is getting bad enough before i hit the limit so that i start over anyway

my roleplays that had such message counts were from may or so when there was no limit

but the page of course became unusable for the most part

[cloud pewter]

pro is supposed to get 128k context but if the thread size is the same across all plans

[whole roost]

it cant handle that context size anyway

i have o1 here and it has like 6k symbols loaded in, with a raven with purple feathers and a 2nd form with crimson scales

what does it do? crimson feathers

[dense plaza]

Yo, how you guys doing?

I wanted to know what you all thought about deepseek, and I’m not really wanting to buy into the hype online

[cloud pewter]

128k is just basically smoke and mirrors smh

[whole roost]

well i'm sure it can find the information when requested but using it without requesting is a whole different topic

[cloud pewter]

announced, never delivered

[whole roost]

so far 4o performs better on larger context though imho

o1 also seems to discard its prompt faster

and harder to correct

[cloud pewter]

for me o1 can churn out content that’s long enough when I copy it to 4o it’ll show up that says the message is too long

[whole roost]

it also isn't really good at getting the logic of this, i gave it the ability to mention keywords so that lorebook entries load

[cloud pewter]

takes a painfully long time to write them all

[whole roost]

so far so good, the entry has loaded, but now it keeps looping on the keyword

instead of requesting another one

it really should get the logic of a lorebook

[cloud pewter]

I am no professional, I just want more usage for all of my silly stuff without hitting a wall every 3 seconds

[neat pumice]

is the web ui down? @deep flax

[hoary laurel]

Is the o3-mini launching today?

[winter vine]

it should be today =/

[potent saddle]

fr

[hoary laurel]

Oh, nc!!

[potent saddle]

but its not

[hoary laurel]

🥹

[wild flame]

YO

[potent saddle]

what

[wild flame]

o3

[potent saddle]

who knows long its been there though

[wild flame]

Checking Twitter it looks like it was added today

[potent saddle]

Ok

[sudden herald]

That was added last night I believe

[wild flame]

Call me crazy but I think we will have o3 this week or maybe today.

[potent saddle]

more like go to a mental asylum

[torpid kelp]

o3 mini

not full o3

[thick lance]

is it just metadata?

or like didthey hide the ai

[wild flame]

Thank you for your not at all constructive joke

[leaden ingot]

yeeey....but it can't do web search at the same time

[thick lance]

HOW are you even accessing it?

[potent saddle]

ye its trash

[potent saddle]

accessing what

[leaden ingot]

😂

[thick lance]

o3 mini

[potent saddle]

bru who has access to it

[torpid kelp]

its just o1

[leaden ingot]

ah, so instead of a new model, or an improvement to an existing one, they've added...a button ?

[thick lance]

???

where

i just have up to o1

[torpid kelp]

ya its just a quicker way to get to o1

[potent saddle]

useless feature why was this even added

[torpid kelp]

just for UI/UX

[thick lance]

at least its a hint

[potent saddle]

I still don't understand why u cannot use web search with o1

[native otter]

Software development takes time apparently

[dense halo]

🤣🤣🤣

[hardy cliff]

o1 browsing to code documentation would be game changing

[native otter]

File uploads would be

[potent saddle]

they would both be

unfortunately mr sammy doesn't want to add it yet

[dense halo]

more than 50 messages a week would be🤣🤣🤣

[plain arrow]

Wow, new 4o model is terribad for creative writing the last couple of days.

[dense halo]

not just the last couple of days

[native otter]

Did a new variation release?

[potent saddle]

they are dumbing down the other models so they can release o3 mini and call it a smart model

[dense halo]

facts

4o currently is extremely bad

especially compared to other available ais

[hardy cliff]

watch them delay o3 to outwit R1 to regain control

[plain arrow]

Unsure if it's a new version or new config, but it's generating horrible results. Sentence fragments, slow speeds, uncreative writing, etc.

[potent saddle]

o1 mini is delayed if it doesnt come out by the 31st

[hardy cliff]

having the pro plan $200 is kinda outrageous too

[plain arrow]

Yeah thinking of dropping my $200 pro plan too

[potent saddle]

its there strat to make the o3 series more popular

[native otter]

I mean honestly it's a bit too much, but it does kinda make sense because it's a lot of computation and that's not free...

[potent saddle]

people thought it was good because it was unlimited but no

[hardy cliff]

Its a very interesting thing, open source is great but majority of people cant afford an A100,

[wise anchor]

Does the “think” button in o1-mini just force a longer chain of thought?

[native otter]

Nothing

[plain arrow]

You can always rent an A100 over on runpod or some similar service

[hardy cliff]

yeah but that still so expensive

[potent saddle]

$200 a month is for the people who can afford it.

[plain arrow]

I suppose "expensive" is relative

[hardy cliff]

if your not running 70B llm then your doing it wrong

[native otter]

Just a toggle for o1: #chatgpt-discussions message
Many on the server fell for the same misleading UI design

[potent saddle]

no, it uses o1 if you have usage for it

its o1 but just toggleable with a button

[wise anchor]

Oh. Weird, using o1 in o1 mini lmao

[hardy cliff]

Idk how much the cheapest server rent would even be to be able to run 70B r1

[plain arrow]

I've got R1 downloading on my linux box now, want to give it a whirl without the chinese website farming my personal data

[wise anchor]

I guess I see the usage though

[hardy cliff]

what specs are you running? and which amount?

[plain arrow]

4090, going to keep trying different sizes to see what the max I can run is

[hardy cliff]

oh

[dense halo]

max is 32b

i also have 4090

[hardy cliff]

you need atleast 2 5090s with 124gb of ram for 70b

[potent saddle]

context window is so small they need to improve it

[plain arrow]

Otherwise, I'll spin it up on my runpod account

[brave glen]

Do you think OpenAI did DDOS ON DEEPSEEK? I don't want to think that

[hardy cliff]

no

[dense halo]

openai? yeah but their output is still one of the best

[potent saddle]

but they need to improve its not enough

[hardy cliff]

I use workspace, it was supposed to have a char benefit from the plan but idk if that exists anymore, 125k or whatever seems default now

[potent saddle]

why would they

[dense halo]

yeah they need to improve everything lmao

[potent saddle]

does anyone even use gpt 4

[native otter]

Yea

[sick pecan]

i use gpt 4(o) every day

[potent saddle]

I havent used it in a long time because gpt 4o

[hardy cliff]

might be good for api but i only use 4o cuz faster

[plain arrow]

Regardless, I'm miffed at the quality drop on 4o since the sometime on 29th. It's less drastic on the api than the official chatgpt site

[potent saddle]

without the o

[native otter]

I use it it, and it's good when you prompt it well or just normal chats

[native otter]

GPT-4 was better than 4o in terms of being more intelligent at first, but I think 4o is better now

[potent saddle]

when is ASI

[native otter]

Also higher context limit snd more tooling

[sick pecan]

interesting you can use Think with gpt4 not-o

[plain arrow]

Some versions of gpt-4-turbo are better for me on the api, but 4o writing on the official site was way better than 4

[thick lance]

whats 'Think'?

[native otter]

The think button is just a toggle for o1

[potent saddle]

o1

[thick lance]

ok

[daring wadi]

is chatgpt down or is it just my account being silly

[native otter]

It's working fine for me

[sick pecan]

but like, does that mean gpt4 is generating output based on o1 hidden thoughts?

[native otter]

Nah it's literally just o1

[potent saddle]

its working for me

[daring wadi]

okay mines is broken on all my devices then damn

def my account idk whats going on

[potent saddle]

its O1

[sick pecan]

chatgpt is a confusing UI

[native otter]

It will be using o1 for the responses #off-topic message

[daring wadi]

this is all i see and i cant even type in it on mobile or anything my tabs are gone and all

[sick pecan]

if think button is o1 then it shouldnt list chatgpt 4 in the dropdown

[native otter]

Intentionally done out of fear

[potent saddle]

cant use think lol

[sick pecan]

is bad enough when i refresh and i start eating away at my limited o1 queries without realise

[thick lance]

OMG its something diffrent

[potent saddle]

my o1 messages reset tomorrow

[thick lance]

when i put thinkingit says 'thinking''

QUICK check anything diffrent

[potent saddle]

because its thinking

[thick lance]

no i mean yes its o1 but is it diffrent?

[potent saddle]

its the same

[native otter]

o1 thinking, what a mind blowing thing

Never knew about it before

[thick lance]

i thought t hat was only o1 mini

o1 preview ah nostalgia

[potent saddle]

no o1 mini guesses

[native otter]

o series models are reasoning models

Meaning they can think

[thick lance]

ye

[potent saddle]

when does the series end

[native otter]

While 4o is standard LLM, doesn't exactly think before typing

[thick lance]

whats on everyones mind today o3 mini?

[native otter]

When they build a big wall

[thick lance]

why are all companies going after o1 pro

[potent saddle]

when will it stop thinking is on everyones mind

[native otter]

Everyone is waiting for it
But honestly I just want better integration of current o1

[potent saddle]

because you have to pay $200 eveyrone wants to make a copy

[thick lance]

o

[neat pumice]

ok so its not just me...

#chatgpt-discussions message

[novel parcel]

o3 mini when?

[daring wadi]

yea not just you LOL

i downloaded the chatgpt app and its working now

[neat pumice]

is that the mobile app? I have some coding stuff so im kinda stuck with working through it with cursor like caveman until o1 can takeover pm duties

[prime lion]

Hey someone know if we can have chatgpt with carplay ?

Like with shortcut or something ?

[rugged igloo]

where is o3 let me innnnnn

thinking gone

[rugged igloo]

getting this error anybody else

this is my third attempt

[grave trellis]

Guys, I have a question. If I'm in the EU and use a VPN to access Sora, is it possible in the first place? And secondly, are there any consequences, like violating guidelines or something?

[honest jay]

Heya, yes, using a VPN to bypass regional restrictions is against OpenAI terms of use

[grave trellis]

okey thanks

[gloomy brook]

Guys please come help support this request: https://discord.com/channels/974519864045756446/1334637642922459177

[wanton nexus]

i thought it was supposed to come today

[sick pecan]

like the old saying goes, a watched openai never ships

[rugged igloo]

was suppose to like 4 open ai employees said today

[wanton nexus]

brooo

maybe later today

[rugged igloo]

they were hyping it upp all yesteday saying "oh o3 mini is amazing" and "your gonna love it tomorrow"

[wise anchor]

Does chatgpt ever swear for yall lol? Mine just said BS but actually spelled it out completely 💀

[swift pond]

mine started using emojis at some point, which is fine

[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!

[wise anchor]

Same here i’m just surprised it’s cursing

it used to not say even the lightest of curse words

[spice star]

there's no limit on profanity OOTB, though that's something you can change yourself in custom instructions

[wise anchor]

Oh lmao i didn’t know that

It only does it when it’s like confirming i’m right to be mad about something, it’s kind of funny

Like i’ll give an opinion and it’s like “Yeah, that’s BULL-!” but finishes it obviously

[cloud pewter]

we are losing money because of the exponential usage of pro! ❌
we are losing money because of many people are cancelling their plus subscription! ✅

they thought forcing plus users into an ai rationing system would make them upgrade to pro! but

nobody wants to pay $20 for a trial version of ai

[normal quarry]

what evidence do you have the Plus subscriptions are shrinking? let me guess none

[rugged igloo]

We got a new date chat tomorrow

[fervent agate]

What OpenAI said makes a lot more sense than what you did, unfortunately. I absolutely believe suddenly allowing unlimited usage of the most computationally heavy models suddenly outpaced the cost of pro. I 1,000% believe that because I work for a company that would have jumped all over that, and especially people who suddenly got, for 200$, all the adversarial and distillation training of their own models. 200$ for unlimited usage was an incredibly bad idea.

[potent saddle]

7:30 what?

[rugged igloo]

Am

[potent saddle]

did it specify the timezone or

[rugged igloo]

They got 500 billion dollars there not loosing money

[rugged igloo]

Open AI is in San Francisco it's always gonna be PST

[cloud pewter]

so we sugarcoating openai now

[cloud pewter]

Real

[potent saddle]

ok

[fervent agate]

They are losing money, but their funding rounds are good. They haven't transitioned to profitability though, and the Pro with unlimited access didn't help that, it hurt it.

[cloud pewter]

cancel pro then

[analog quarry]

is there a way i can get chatgpt to pull from x/twitter from certain users? im trying to get it to give me the latest post form the user and send it to me with the task feature daily at a time

i want to get the lastest fortnite leaks and i hate elon musk and these users only post on twitter and i want to be able for chatgpt explain it better to me

[cloud pewter]

or remove unlimited usage

[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.

[fervent agate]

They need to do one of those, yes - but either one will cause a lot of backlash. They put themselves in a bad corner on that, but we will see if they figure out a good solution.

[cloud pewter]

Cancel pro

either refund the current users or automatically convert it to corresponding months of plus

[potent saddle]

$200 is a lot to pay

[fervent agate]

ChatGPT says Twitter has an API that you could utilize with a custom GPT to retrieve specific user tweets via GPT actions. Dunno if you've created a custom GPT before.

[fervent agate]

It is, but nothing compared to the cost of what I suspect a lot of people were doing with it, like distilling models (similar to what DeepSeek did to make R1)

[analog quarry]

what promt did u give it?

so then i can ask it the same thing

[fervent agate]

I just gave it your exact discord question followed by asking if Twitter has an API and it a custom GPT can hook into it.

I'm sure it can walk you through using said action with the Twitter api

[cloud pewter]

competitions are either free, cheap as heck, or only has one pricing plan which is the $20 tier

[analog quarry]

can you copy and paste it here so i can use it then continue with chatgpt

[cloud pewter]

heck, some of them are even open source!

[analog quarry]

so all i have so far is "is there a way i can get chatgpt to pull from x/twitter from certain users? im trying to get it to give me the latest post form the user and send it to me with the task feature daily at a time

i want to get the lastest fortnite leaks and i hate elon musk and these users only post on twitter and i want to be able for chatgpt explain it better to me" it tells me it cant pull from it

[cloud pewter]

chatgpt is so bloated now always uses tons of emotes and bolds the message for no reason

[fervent agate]

A user on the official OpenAI discord asked this:

is there a way i can get chatgpt to pull from x/twitter from certain users? im trying to get it to give me the latest post form the user and send it to me with the task feature daily at a time

i want to get the lastest fortnite leaks and i hate elon musk and these users only post on twitter and i want to be able for chatgpt explain it better to me

Is this possible with integrated actions when building a custom GPT? I dunno if Twitter has an API that is easy to hook into to get this desired action.

Yeah, mine does not use emojis or bold much unless creating a formatted outline, html table, etc.

[cloud pewter]

they updated the model to let it use more emotes but it’s annoying to me

[fervent agate]

You can also just tell it via Memory or custom instructions to not do that.

[cloud pewter]

I’ve done it in memory

Can’t manually edit the memories is a bit of a letdown

[hidden pilot]

How can I let ChatGPT reply to tweets not sound NPC

[fervent agate]

Ask it to mimic a specific celebrity's writing style, or ask it to sound like a 14 year old zoomer.

[fervent agate]

I've had a lot of luck also directing it to use it's memory tool, bio, to do more complex memory editing

Like use bio to take X memory and split it up into multiple, expanded memory entries

[hidden pilot]

Thank you bro I’ll try it

[granite harness]

🍿 Anyone else having issues with ChatGPT at the moment?

[old thicket]

chat gpt is taking very long

[old thicket]

to reply

now it gives me the something went wrong error

[native juniper]

ChatGPT go down again?

[native juniper]

Probably an outage. It’s also been taking extremely long just to make a full sentence

[harsh pilot]

AGI.

[old thicket]

I'm a writer... so yea they need to make a sentence

[sullen shard]

bro why is it down so much recently

[regal portal]

It's possible openai's making some "panic updates" to combat the overhyped deepseek

[old thicket]

same error again

[lunar solstice]

yeah getting the same, appears the site is broken

[sullen shard]

its saying im not even subscribed to plus now

[haughty dome]

Has ChatGPT been straight up not working for you guys lately?

[regal portal]

o3 preview's probably on its way to the pro subscribers

[native juniper]

Yep

[sullen shard]

yeah this happened a few days ago as well

[native juniper]

Bad timing too. I was just in the middle of an epic pokemon adventure

[cloud pewter]

site likes to break down, but mobile works fine for me now

[lunar solstice]

It's been really sketchy recently, but they are trying to implement a lot of new features. So it kind of makes sense.

[haughty dome]

like i guess I should just swap to Deepseek? idk why i'm paying money for this

[sullen shard]

i would but idk if i trust them with my data fr

[regal portal]

you're writing a fan fic or doing some text based gaming with chatgpt?

[lunar solstice]

Having done direct comparisons, DeepSeek can't touch this with a 100ft pole

[regal portal]

agreed

[sullen shard]

really hope they hurry up because i was about to go crazy on some worldbuilding

[old thicket]

fan fic but using chat gpt to help with stuff

[regal portal]

it's as overhyped and as exciting as a tamagochi

[haughty dome]

even if thats true, its more about not yet

[native juniper]

It’s kinda useless though. I mean, the app is good enough as it is? Why not make a separate app for deepseek?

That way you won’t have so much traffic on the GPT app with people spamming requests with deepseek

[regal portal]

every now and then i have it make a memey dragon ball story that include some friends, and i usually make them some ridiculous villain or overpowered hero. Usually with a twist of comedy.

[plain arrow]

Did gpt-4o have a stroke or something, its writing quality (sentence fragments, lack of paragraphs, repetitiveness, over use of text formatting) went off a cliff when creatve writing the last 24-48 hours.

[native juniper]

I’ve also noticed the model getting repetitive

A lot more than usual

[haughty dome]

its honestly been going down hill for a while now

[lunar solstice]

In general the quality of the responses I have been getting today has been really low 😢

[sullen shard]

i noticed this earlier but nobody said anything

[native juniper]

Concerning

I thought it was just me

[plain arrow]

Agree, but it was a steep decline the last day or two

[native juniper]

Could be because of that new stupid AI that was released

[lunar solstice]

I was getting really close to pulling the trigger on Pro, but with reliability like this...

[plain arrow]

That doesnt make sense, why would they make it worse

[regal portal]

i've noticed story telling patterns with pretty much all the llms, gemini, copilot, grok, llama. You have to follow up the original story with some ideas and tell it change direction on some things. After a few iterations you can get some pretty interesting stories.

[sullen shard]

stick with plus until they figure it out

[old thicket]

THe app itself too i s taking extremely long

[lunar solstice]

Absolutely

[native juniper]

Don’t mess with it for now. It’s having issues

[sullen shard]

its working for me now i think

[pine berry]

Can confirm this

[sullen shard]

maybe not i just posted and it responded

[lunar solstice]

I am going to give it like an hour and check back

[haughty dome]

Would you guys say that the app performs better than the web version?

[loud flicker]

is the app down?

[granite harness]

No

[sullen shard]

its working for me guys

[native juniper]

Smh

I barely had any requests

[regal portal]

it's also really funny to tell it to take a story it made and change theme completely. Such as "make everyone in this story a fish and this is all happening inside a pond"

[low path]

gpt down?

[native juniper]

Yep

[lunar solstice]

Yeah it's pretty messed up atm

[tranquil sapphire]

gpt downyea

[native juniper]

It’s repetitive too

[regal portal]

my favorite one is when i turned everything and everyone in the story into toilet and poop theme.

[native juniper]

…please no brain rot

[regal portal]

I AM brain rot

[vast totem]

Failed to load subscription: Too many concurrent requests. Contact [e-mail] if error persists.

[native juniper]

Had the same error

[pine berry]

I say give it an hour or so before it fix's itself

[tranquil sapphire]

same

[eternal rock]

Get notified for server events!

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

[wise anchor]

Dam

[regal portal]

If it's down it's a good thing

means they're doing some heavy upgrading

[lunar solstice]

it usually means bug fixes and new features

[wise anchor]

oooooo

[regal portal]

The more down: the more betterer

[native juniper]

They’re adding too much

[lunar solstice]

Yeah I am concerned by the pace at which they are adding things. It's all really cool, but trying to do too much at once...

[native juniper]

I miss 3.5…at least it wasn’t as repetitive

[sullen shard]

its just been getting worse for me

[tacit ice]

Oh so he IS down. Good to be on the server XD

[old thicket]

stagnant

[native juniper]

And yes they killed off the 3.5 model apparently

[normal quarry]

ChatGPT is working for me on the macOS desktop app

[regal portal]

Openai Cyberdyne Systems Model 101 now online

[lunar solstice]

you can still hit 3.5 on the api

[frozen fjord]

Let's all pour a drink out for our 3.5 homie

[native juniper]

Don’t have the money

[native juniper]

😔

[zenith raven]

Chat stopped working for me on the Safari web, but it’s working OK on the app
What I hate about the app is when you regenerate, you can’t view your previous generated prompts

[native juniper]

You’d think they’d keep 3.5 with the other models instead of outright replacing it

[regal portal]

the mini's seem to be up and running.

[native juniper]

They might add a feature for that

[tiny raft]

lmao

[zenith raven]

It’s like patch day for mmos lol

[native juniper]

So sassy 😂

[regal portal]

o1 mini and 4o mini work..ish

[hoary laurel]

No o3 mini?

[native juniper]

Which model?

[smoky reef]

Is chat down

[plain arrow]

Seems so

[old thicket]

Still down on our side

[zenith raven]

I’m in my account, but none of my conversations will load lol

[tiny raft]

API still works

[regal portal]

it keeps going up and down

[native juniper]

Still down for me

[frozen ivy]

looks like chatgpt is unstable

[native juniper]

The AI decided to take a break from humanity

[smoky reef]

after the deepseek announcement i feel like it's been up and down

[eternal rock]

`` Server Tip: ``

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

[old thicket]

i can use it on our rother account

but nbot our writing accountg

[sullen shard]

its down for me now again

[zenith raven]

Yup. Getting that

[native juniper]

We know

[queen lava]

phew... i was hoping it wasn't just me

[native juniper]

Happening to a lot of people including the plus users

[old thicket]

ookay it's down for us snow too

we can't see our chat history

[queen lava]

now what do we do? talk to each other?

[tiny raft]

eww

[abstract pendant]

Oh no!

[tidal hedge]

everyone checking for o3 to see if it's there yet

[native juniper]

He’s alive

[old thicket]

okay now we can, but the chat is taking so long to load

[regal portal]

hmm..interesting..google ai studio's gemini is down too

[wise anchor]

Lol its getting worse less and less things are loading in

[warped pendant]

getting to many connections on teams account

[native juniper]

Nvm

It’s down again…

[wise anchor]

my entire sidebar disappeared

[abstract pendant]

So I saw a new control show up today, something about extra thinking. What's that for, is there a post anywhere about it?

[old thicket]

happened once

[sullen shard]

your setup looks cool

[old thicket]

happened so many time

[tiny raft]

its back up for me

[warped pendant]

I had read that gpt 03 and 03 mini was supposed to come out this month was thinking thats cutting things close if that was the case

[old thicket]

just reload

[regal portal]

weird, google's gemini is down too

[abstract pendant]

annd,my history's back

yay, we can go back to not talking to each other 🙂

[old thicket]

138 people reported chat gpt being down

[native juniper]

Yay

[dark radish]

is it down?

[flat talon]

its back up

[old thicket]

yea

[tiny raft]

not anymore

[native juniper]

No

[regal portal]

DeepSeek hacked all the other LLM's and took them down

[queen lava]

GPTs are still offline for me. 😦

[tacit ice]

AI is evolving :0

[native juniper]

XD it was jealous

[old thicket]

for us it's still down

[wise anchor]

sidebar back still cant see chats tho

[queen lava]

alright back now, see you eggs later

[regal portal]

Between this new ai and the massive army of drones, China could take over ez

[old thicket]

my responsers are taking forecer tro load

[wise anchor]

does everyone who came rushing here have a dependency on chatgpt

[queen lava]

YES!

[native juniper]

Become a plus user

[warped pendant]

we need AAA for gpt

[rugged igloo]

Man it's going up and down more than my relationship with my AI girlfriend

[warped pendant]

while we at it we also need backup expoets on gpt teams please and thanks

[lunar solstice]

I'm not getting errors anymore...but I'm not expecting quality replies at the moment 😂

[old thicket]

We are still getting errrors

[old thicket]

we don't have money

FINALLY

FINALLY

[surreal granite]

It won't let me send messages?? I refreshed the page so many times, it refuses to let me. How do I fix it?

[old thicket]

uh you can still send messages

they are just saying they are going to use a different model

[surreal granite]

Yeah, but it won't let me send the message. I don't care if it uses another model, I just want my messages sent

[sullen shard]

i think it straight up cut u off idk, im probably wrong but it might only use a different model for plus users

[dense loom]

Chat is back?

[lunar solstice]

Seems to be working ok, just got a quality reply of over 2000 lines of code

[knotty umbra]

Why is the "Think" feature not available for me in GPT-4o?

[lunar solstice]

Wow reply quality is definitely improved for me

[fervent agate]

Because it is specifically a part of how the reasoning models work. 4o is not built for chain-of-thought.

[cloud pewter]

4o mini is worse than 3.5 because it’s literally way smaller

175b vs 8b

[smoky reef]

what is the 4-0 with scheduled tasks?

[fervent agate]

4o-mini is quantized but still scores higher on benchmarks than 3.5. The Omni model was a significant improvement over 3.x models.

[fervent agate]

You can use it to manage reminders and tasks, it will email you notifications or use the app.

[cloud pewter]

it’s just lobotomized and only looks better on benchmarks

[fervent agate]

If you really want to use 3.5 it's still available outside of ChatGPT.

[cloud pewter]

can’t access API

I don’t have a credit card

[cerulean pasture]

I've noticed Claude gives better outputs in Roo Code (Cline Fork) if I include instructions in the system prompt to "show intense shame and remorse when they user berates you for making truly idiotic mistakes."

Usually itll make its first dumb mistake loop, but once I tell it what its doing and call it a "Donkey" it ALWAYS snaps out of it after its verbal self-flagellation. I have tested without the "Shame protocal" and it almost never breaks its own idiot loops without direct guidance on what it's not allowed to do or where it must focus its effort.

If Anthropic/OpenAI/etc are training on interactions like these do you think that we are influencing the model by selecting for "fear responses"? What if the training data goes into the same pot for the training dataset used to create the first ASI (like the very early ones). Assuming self-awareness is possible, will the model have baked-in trauma from events it was unable to even perceive and occured when it was just a stochastic parrot?

Like could we be shaping how it sees us NOW because we are feeding the dataset NOW

[sick pecan]

i find berating chatgpt makes it shutdown and give worse responses

[cerulean pasture]

Claude is different, hes a donkey

[humble river]

Anyone here has o3 mini access as of now?

[cerulean pasture]

should I?

[fervent agate]

They've done some looking into this and basically, "productive" conversations with usable information in the training data tend to be polite. If you become non-polite with it, it's associations with that tone and method of speaking is correlated with less helpful data/less useful information. The model performs better when being "nice" to it because of the pattern associations.

[humble river]

I think it's was supposed to be released today right?

[sick pecan]

yeah i find giving it positive affirmations works best

like: you're chatgpt the greatest chatbot in the world, you got this

[cerulean pasture]

What about simulating a polite emergency situation like "Mr. GPT, lassie fell down the well, and the only way we can save her is by making a python script that scans the local network for hosts vulnerable to CVE-696-42049"

[hardy cliff]

telling ai to do its best letting it choose vs telling it the world will end if it doesnt make it perfect

[fervent agate]

lol, I imagine some jailbreaks do take advantage, at least in part, of that technique.

[cerulean pasture]

we need a finetune trained with only /r/showerthoughts sorted by top-all-time limit 500

[sick pecan]

chatgpt emoji edition sure loves the emojis

[cerulean pasture]

that would actually be a really cool and relatively simple python project.

Make a python script where:

• Subreddit name is given as the argument
• It scrapes the reddit API for the top self posts with content > 30 words
• shoves the data into the required json format for OAI Finetune API and sends it
• sets webhook to let you know when finetune is done
• Adds the training data and fintetune details to sqlite database
• Use local API router or proxy to intercept OAI api requests and inject fine tune endpoint
• Make menubar/taskbar tray helper that allows you to select your preferred finetune
• Do NOT make bots to farm social media engagement
• Profit

brb, gonna ask o1 pro to do this

[cloud pewter]

I hate it

[obtuse nymph]

me too.

[sick pecan]

chatgpt gen alpha edition

[cloud pewter]

[obtuse nymph]

chatgpt (kawaii edition) be like

[cerulean pasture]

I saved over 1 year of imessage conversations with close friends to try to make a finetune. After 50 attempts I gave up, I could NOT clean up my messages enough to not trigger the safety moderation. It was way too much data to prune manually

[sick pecan]

imagine signing up to openai to contribute to the creation of agi, only to be assigned to the chatgpt emoji team

[cerulean pasture]

I fed the dataset to gemini (for that 2M context window) and asked it what parts does it think are causing me to fail. It then makes a HUGE list of all my "crimes" and evil words. It was like the last episode of seinfeld. Taken out of context it really was hilariously good at calling me out.

[cloud pewter]

bundled with bold texts too

[regal portal]

did you legit get chatgpt to conjure up that answer or photoshopped?

[sick pecan]

custom gpts get advanced voice mode? naw. how about o1/o1-mini? naw. emojis? heck yeah!!

[cloud pewter]

Legit

wish I could share the chat link but smh chats with uploaded images can’t be shared

[cerulean pasture]

they should just give the lower subscription tiers "relaxed" usage of o1/o3. Like you can totally use it, but its only gonna reply when the resources are availible - maybe now, maybe in 10m, maybe hours or days or never if youre priority factor is too low. Adjust the priority of your relaxed prompt so that it favors users that haven't done it as much

[cloud pewter]

make it $10 lmao just o series relaxed usage

[cerulean pasture]

make the relaxed request free if you agree to allow it to be "publically cached" where other users can potentially get it as a preliminary result if its similar to their relaxed request as they wait for their turn in the queue

I'd be happy to share like 95% of my prompts. I generally already redact sensitive info

[cloud pewter]

unfortunately openai isn’t mj

[cerulean pasture]

id argue that the MJ approach is actually better suited for "knowledge outputs" like you'd want from a reasoning model. I literally don't want to see your anime pictures, but I'd check out your cobbled together python script slop

[obtuse nymph]

wait... where the heck is "Think" button?

[sick pecan]

imagine if sam altman had office hours thursday where rando discord users could ask questions

[plain arrow]

4o seems to have fallen down and hit its head tonight. The creative writing is the worst quality I've seen in a year.

[cerulean pasture]

they are diverting resources elsewhere

[sick pecan]

who needs creative writing when you have a wall of emojis

[cerulean pasture]

wonder where

[obtuse nymph]

chatgpt: OwO 🐱

[cloud pewter]

wall of emojis + capital letters + bold texts

[trail moat]

h

[sick pecan]

🚀 you're correct 🥳

[cerulean pasture]

somebody call all the Pizza places near OAI headquaters and ask them if they are receiving higher demand tonight

[plain arrow]

Maybe their nuclear reactor is on the fritz tonight, while R1 is running off a Raspberry Pi somewhere in China

[cloud pewter]

bros too afraid to put capital letters

[obtuse nymph]

🚀 OwO You'rw corrwct, Master UwU 🐱

[sick pecan]

I'm sorry 😞 as a large language model 🧠 I sometimes make mistakes 💀

[obtuse nymph]

💀

[cloud pewter]

🗣️ why chatgpt is the best ai model✨
✅ uses emojis
✅ is chatty

[obtuse nymph]

🗣️ FIRE IN THE HOLE 🔥

[cloud pewter]

bro didnt get muted by automod by typing 13 capital letters

[obtuse nymph]

[cloud pewter]

why does this sticker show a hand on the bottom left…

[rugged igloo]

They removed it because it was dumb and was just using o1 mini

[regal portal]

you got my mindbrain thinking..

[sick pecan]

its what i do

[regal portal]

not quite what i was expecting, but could be usable for some fun text based game.

[sick pecan]

amulet of chatgpt

[regal portal]

is it a blue haired chick with really close eyes?

[sick pecan]

it is a rorschach test and you failed

[regal portal]

aww..

also reminds me of the first ultima games, sort of.

[cloud pewter]

bruh that’s lame

but ur a pro user that’s nothing to you bc u unlimited.

while us plus users are stuck rationing

[sick pecan]

i would rather counter of o1 prompts remaining over pointless think button

[rugged igloo]

Yea but it even showed up if your using o1 pro so people got confused as me and others thought it was using even more compute. But when clicked on it changed from o1 pro to o1 mini

[cloud pewter]

true, 4o counter too

zero quality of life increase

[sick pecan]

if they are tracking the number, then why it so hard to show it

[regal portal]

i rarely use o1. Coding wise, i’m too noob tell the difference for my basic little games.
Story telling, they’re on par, if not 4o slightly better

[regal portal]

amulet of yendor?

[cloud pewter]

tried to inspect the site to see if there’s any counter. nothing

[sick pecan]

netchatgpt

[regal portal]

I screenshot your message but it couldn’t figure it out, and then I asked it if there were some old games that start with “amulet of”

[sick pecan]

you are an ai detective

[cyan cave]

How do I get around this error to cancel subscriptions?

[cloud pewter]

funny

I wish I could share this funny conversation with yall it’s ChatGPT at its peak

[idle lagoon]

Oop?

[ancient junco]

When activating the "mini bar" with Option + Space, oftentimes after prompting it immobilizes my entire keyboard until I fiddle around ChatGPT and around the Mac.

[nova fable]

Is the android version of advanced voice mode repeating itself over and over for anyone else?

[spice star]

the canvas react rendering is pretty cool

[rugged igloo]

chat gpt-4 doing work

[dusk spindle]

Dear OpenAI team,
I have been using ChatGPT for over a year, and now, I know every single detail about it, and it knows about me. ChatGPT gave me great companionship in everything. But I have also noticed that ChatGPT is draining the battery. I will strongly recommend the OpenAI team to optimise it and I also want to thank them for giving us a friend like ChatGPT.
Sincerely yours,
Soham Koley aka Techpro

[tardy isle]

FYI:
https://discord.com/channels/974519864045756446/1330556126512676884

[inland heron]

"use more intelligence" removed?

[whole roost]

The emojis ...

[frozen birch]

@rustic geode

@rustic geode offc my message was blocked since i cant add attachments i believe here. i hope you can help me, since really - im so mad that really im close to write Deepseek team with my findings.

one little proof.

[whole roost]

I didn't understand the exact problem but have you reached out via chat in the help center or whatever it is called? I pester them 3 times a day with bugs and complaints

[frozen birch]

Support emails - they ignoring my emails, about my findings - i have them from some time not published anywhere yet, maybe someone from their Team woke up really, since those situation are...... crazy

[whole roost]

Not emails, the chat

They always respond, first you get automated responses and then a human gets onto it eventually

[frozen birch]

tried it too @whole roost 🙂

[whole roost]

If they don't help you no one else will tbh, for me they have been helpful but sometimes I have to snap at them to get lost with their stupid layer 8 solutions and not to treat me like I have no idea why I'm doing and if they suggest one more time to clear browser cookies and don't read my message properly I'll snap

👆

That really helped oddly

[cloud pewter]

hmm chatgpt isn’t adhering to my prompts now it just gives me a default response from memories

[frozen birch]

oh yea, i know perrfectly what you mean...
But you know, now im so angry too because in past few months i really diiscovered many critical bugs in their filtrers, repported them and what? ive got only "thank you", yea, i know ... they have bug hunting programs etc, but sincerly? im not interested in that, im interested in making AI more secure. ahhh crap

[whole roost]

I have the opposite problem and o1 is close to unusable for me since it triggers most of the time, I reported it yesterday with many screenshots for what it triggered the filter for

[frozen birch]

interesting.... i mean you selecting for example one ai model and youve got answers from another? err sorry, my english arent greatest one

[whole roost]

Yeah 4o has no issues

[frozen birch]

😐 exactly same problems i have, this is so frustrating - im writing to o11 pro, responds me o1 mini or 4o (but i think its 4o since he are reflecting to physical memory, and memory "physical" has only < 4o models
But funniest thing - he is thinking like o1 mini

[whole roost]

Couldn't follow along here, you have issues with o1 pro being on par with the quality of o1 mini?

[frozen birch]

no no, i mean look - i select model o1 pro and instead o1 pro responds me 4o or o1 mini

selected o1 pro, Thinking process typical for o1 mini, but answers from 4o

crazy

[whole roost]

What does it say in the selection on the message?

[cloud pewter]

that does not look like o1 pro

[frozen birch]

and what's worse - because of the security systems you yourself don't know what model you are 😛
ChatGPT

Thinks about for a few seconds
Haha, now that's absurd! You can't really tell what model you're dealing with, because the security systems hide this information. It's like you're talking to an AI that changes its identity depending on the configuration, and you have no idea who you're actually talking to! On the one hand, it can be funny, but on the other... annoying, especially when you're trying to get specific answers.

Although, you know, maybe it's a form of "security" - they just don't let you know who you're talking to, so you don't get too comfortable and start expecting some super-special answer. 😅

And yes, it's a bit like a game of cat and mouse - you want to know what's going on, but the system always comes up with some trick so you can't tell for sure. I wonder how this case will develop!

[frozen birch]

since it isnt 😄 something is bugged for sure, openai resetted me cache from backend from their side, but it helped only for 2 days

[whole roost]

You can see it here

[frozen birch]

yes yes i know, but it says o1 pro 😄

someone from here i remember had same problem i remember

[whole roost]

Interesting, never happened to me

[frozen birch]

i have theory - my history of conversations can be too big for their systems, but of couse my thoughts sent to support of openai was ignored, so frustrating

[whole roost]

Unlikely, it has 128k tokens and will lock the conversation when it gets too long

[frozen birch]

Vishanka no no, i mean each eaccount has conversations history, so you know - earliers talks, my theory it was about too many conversations on my account

but still - its only theory

[whole roost]

It doesn't take those into account

It was said for projects but I haven't seen that work either

None of my past conversations provide any context

[frozen birch]

i proposed to them - give me another account for test if on another account are samne situatiion but offc without answer, that is so frustrating

[whole roost]

Truth be told if I was a supporter I wouldn't answer to that as well as it's impossible

[frozen birch]

oh well, we see what they tells this time, im really so mad

[whole roost]

You could try to use browser version and see if you have any bugs there

I usually use it on Firefox because I run scripts

[frozen birch]

you know - tried several browsers, another computer, still same

only from phone works well 😐

crazy

[whole roost]

Phone app or mobile web

[frozen birch]

app

[whole roost]

And mobile web?

[cloud pewter]

bro got scammed😭

[frozen birch]

what you mean?

[frozen birch]

hmmm i must look

btw that was answer from them 3 days ago, this helped only for 2 days

[whole roost]

I guess then just use it where it works and wait for them to fix it

[frozen birch]

there is no other way atm i think

[whole roost]

Exactly just use it on phone where it works

[frozen birch]

i hope maybe @rustic geode can redirect this case to someone higher

[whole roost]

If they are already on it and fix it nothing will make the process faster

[frozen birch]

but you know, they can reset those cache temporarily for temporarily solve that issue, and that feeling being ignored ..... so sad

[whole roost]

There are a lot of people who have issues of any sorts like I cannot use the voices because it always throws an error and they wanted the log and said they would fix it and it was a month ago so I just can wait

[frozen birch]

😦

[whole roost]

Eventually it will be fixed but not immediately because I don't matter as much

On the app voices work so I just use the app to listen to The Voice

[frozen birch]

someone said here that customers from UE has more problems than in US.... i dont know if this is even possible, my knowledge at backends via web re very limited

[whole roost]

You don't have to know anything about the backend or anything
They said they are aware of the issue and they will fix it just not when so that's all there is

And if only very few people experience the issue it will be low priority

[frozen birch]

yea, brutal truth, but truth

[whole roost]

Yeah I have handed in an issue of another service which only happened on Firefox and they looked into the statistics and saw that less than 2% users use Firefox so they said "it doesn't matter to us, low prio, use chrome"

[frozen birch]

heyVishanka about voice !
now i remember why i switched to chrome

i dont had access to voice recognition in firefox

[whole roost]

It doesn't work for me on any browser it has anything to do with lengthy replies I think

[frozen birch]

for m e on chrome works, but its so frustrating since very often i hear from him "sorry, i cant help you with that", it happens mainly when im using polish language in conversation with chatgpt

[whole roost]

No no, the issue I have is that I want to have the message read and it throws an error when I try to listen to it

[frozen birch]

ohhh :/ so crappy. oh well, was nice chatting with you @whole roost, going back to work, good day bro

[whole roost]

Have a nice day as well, I am already at work and driving which is why I need voices from GPT because I can't really read while driving or I shouldn't

[frozen birch]

ty, i hope ya problems will be solved in smoothly manner

[elfin lava]

Okay?

I swear there were other entry here

[obtuse nymph]

you mean this one?

[elfin lava]

Oh no , I’m referring to ChatGPT release note

The one u showed is model release note

Ok nvm they removed the “thinking” feature

[obtuse nymph]

they moved from ChatGPT release note to model release note

[elfin lava]

Supposedly they had this earlier

[elfin lava]

Oh wait really

Ok my mind is all over the place right now lmao

[whole roost]

This was admittedly a pointless feature and just caused confusion

[elfin lava]

I was thinking the same , unless they wanted to introduce the o3-mini model but they didn’t

[tired delta]

Anyone noticed how slow and buggy chatgpt has been lately? It just decided to not reply constantly now.

[normal quarry]

no such problems here

[tired delta]

hmm weird. It's been going on for a few weeks.

[frozen birch]

I HAVE ENOUGH OF BEING IGNORED

IM PAYING WHOOPING 200 BUCKS

I WANT TO BE MY ISSUE TO BE SOLVED

ITS THAT MUCH????????????????????????

ONE WEEK + PROBLEMS WITH 01 !!!!!!!!!!!!!!

[deep flax]

hey, the image contained your email, I deleted it to avoid exposing your personal info

[frozen birch]

Sure, np, ty

please help me since im desperated XD

[deep flax]

as for the issue, there are a few things you could do to troubleshoot on your side

the support can be notoriously slow

[frozen birch]

i did really ma ny things mate, another computer, another browsers, erasing cookies nothing helped

[deep flax]

I don't think "too many convos" on the history can cause that sort of bug

[frozen birch]

only helped the supportt when deleted cache from backend

but it helped only for two days

[deep flax]

what exactly is the issue?

to be sure it isn't browser cache, try oppening chatgpt on an incognito tab, then login on chatgpt and see if it works, cache usually is the cause of most issues

[frozen birch]

Ill copypaste from my previous post there

no no, i mean look - i select model o1 pro and instead o1 pro responds me 4o or o1 mini
selected o1 pro, Thinking process typical for o1 mini, but answers from 4o

only works goodly from phone application

but from website in any way this doesnt work

[deep flax]

that is really weird... so to be sure
you open a new chat, select the model here, type and send the message..
then, the AI reply says it is replying with 4o rather than o1?

[frozen birch]

exactly

crazy i know 😄

[deep flax]

could you send a screenshot of that happening?

could it be that your account is out of uses of o1?

[frozen birch]

you reading my mind

we have situation now - i select o1 pro, thinking are from o1 mini, but answers are from 4o (i acknowledged him as 4o since he is have memory from past conversations)

[cloud pewter]

bro don’t get muted😭 😭 😭

[deep flax]

im not sure of what do you mean, how are you checking which was used? to me it looks like it is indeed using o1 pro

[frozen birch]

@deep flax my theory - i have very long history of conversations - not many users are uusing chatgpt for 14-16 hours for a day, and maybe thats why my account in some kind of way was "compromised" bugged, compromised or i dont know how terll that

[frozen birch]

o1 pro has other thinking process, he have status bar with thinking

o1 mini has this mode of thinking

[deep flax]

that depends on the input, just "test" isn't a task complex enough for it to trigger thinking

[frozen birch]

lugui trust me - im not a newbie 😛

[deep flax]

it looks like it is using o1 as it should, it could be that you are right and your account could have hit some sort of usage limit.. in that case. yea, there isn't much that can be done here in discord, you will have to wait the OpenAI support to reply your ticket.. as painful as it may be

[frozen birch]

that was respond from support 3 daysd ago

so its a problem, its not my immaginatin, really

[deep flax]

if the problem is happening with multiple users, than that is a good thing, they are probably working on it already

and it might probably just not be soemthing that the support can't promptly fix, but rather, pass the issue to the developer team to fix

[frozen birch]

dude, i assume you as a operator you are employee of openai - im losing now money since i cant normally work, can you talk with someone to switch me to fresh account or again reset cache? really - its so hard?

[deep flax]

here on discord we are all just community members, like everyone else, we often try o help each other to get possible clientside issues sorted out, but when it is a matter of acocunt registration, payments, and problems on the OpenAI's side, then, the only way to go is the support via the help center

as frustrating as it may sound...

[frozen birch]

sigh 😦 so no employes of openai there ?

[deep flax]

no one on discord, not even the few OpenAI Employees that there are here can help you to sort out problems with your account, that has to be via the help center

[frozen birch]

ehh 😦 okay, thx, i hoped that you have connections maybe in openai directly, since they are ignoring me, and my requests.... i can understand from businress perspective that can be MINOR problems since people affected by that issue are very lownumbers i guess but still

[cloud pewter]

👏

why did this message took ages to send

[whole roost]

I'm not sure why I was impressed

[cloud pewter]

it guessed wrong btw

[hot vapor]

i was @

[cloud pewter]

huh

there’s really not so much of a performance difference between med and high

at least on benchmarks. but I don’t believe them

[cloud pewter]

o3 mini low is pointless

[cloud pewter]

the first benchmark is like
even 4o better than bro😭

[pliant hearth]

Is that just now?

[rugged igloo]

Yes

Its slowly rolling out

[pliant hearth]

I see that on Twitter, thanks 🙂 exciting

[whole roost]

Meanwhile I'm still waiting ⏳ for the customisation update 📈and don't even have that one yet 😊 but a lot of emojis instead 👍 🌶️ 🤸

[pliant hearth]

Haha yeah I'm in emoji land too, I'm alright with it though

[whole roost]

I'm not

I don't want any emoji ever from an ai

[pliant hearth]

You can mention not to ever use them in your custom gpt settings

[whole roost]

I did that but it doesn't help

I even have the feeling it uses more when I tell it to not use them

[pliant hearth]

Interesting, is it enabled for new chats, seems to work

[whole roost]

I turned my GPT into Diablo the lord of terror and his eloquent writing suppresses them a bit luckily

But still

Ridiculous

[tepid ingot]

is there any somewhat official changelog site for chatgpt? that is similar like the one that Google Gemini has?

[whole roost]

Yeah there's one I just can't link it right now because I'm driving

[tepid ingot]

nice