chatgpt-discussions

[river panther]

I'm a plus user.
I primarily use GPT for my writing.
I write small stories and sometimes fanfiction for the fun of it.
I have GPT Analyze my Story Posts, 3 paragraphs per post and break it all down in detailed explanation's of what's going on.
It's great way to advance story lines, keep the plot on track and even develop better character development.
However the censorship is utter BS. Anything that slips past mild fanservice and the AI freaks out and refuses to analyze it.
Also combat scenes makes it turn it's head up and refuse to analyze as well.
It's not like I am writing Erotica and Gore Violence...
smh

[prime swallow]

"Plus users will continue to have access to Advanced Voice powered by 4o with the existing daily rate limit, which is more than 5x the free limit, as well as access to video and screensharing in Advanced Voice.

And Pro users will continue to have unlimited access to Advanced Voice and higher limits for video and screensharing."

I don't see anything in that which makes me think the change affects plus or pro at all. I could be wrong - that's just my reading comprehension and what I think those words mean.

[exotic lotus]

You write stories and work with a 32k context? Don’t the constant hallucinations bother you, or do you only input short scenes for editing?

[prime swallow]

Gotcha.

My own experiments - I'm a community member like you are, the very few of us that work for OpenAI have gold names, nearly none of us here.

But for me, I prefer to have the model use text then if I want it read to me, I hit that option on the reply.

I don't love voice mode as much as I love text mode, though I have started to get into speaking instead of typing my inputs.

Very recently I started to explore using my phone to talk inputs into a chat.

Let it answer in text.

Then refresh the webpage on my desktop, and hit 'read aloud'

[prime swallow]

In my exploring, the standard 4o is quite flexible with how it handles requests these days, and some other variants, like search and also my own AMV chats? The answers are quite different.

So, I work around that. I figure it all gets improved over time. But if I want some specific Z and have a way to focus on Z with the model, I just do so. Voice is nice but... I haven't gotten as into it as the other modes.

My phone app doesn't appear to have a 'read aloud' option.

I wonder about stuff like that sometimes. But figure it's all a work in progress 🙂

[river panther]

...32k context???
constant hallucinations???
The heck are you on about???

[exotic lotus]

https://openai.com/chatgpt/pricing/

You said you're using Plus. The 32k refers to the amount of tokens the AI can remember. If you exceed it, it starts generating hallucinations, making up things because it can no longer remember everything.

[river panther]

I use the voice readout because I also listen to the anlysis readings while I am working, helps pass the time with my job

[exotic lotus]

But since you're surprised by the limit, I’ll assume that you tend to write shorter scenes rather than entire stories, or at least not long ones.

[river panther]

Ooooooooh
yeah. I have noticed that. Once I get up to 20 Story Posts, it starts freaking out.
It will either just utterly refuse to analyze anything or follow instructions or it starts adding onto the story with it's own writing.
Or it just shuts down entirerly and says "I am sorry but this goes against community guidelines, I cannot do this." for like no reason at all

[prime swallow]

I can only speak for myself, but... when I work on my writing lines. I kinda use the model like it's a microscope for a section of the work at a time OR use it like a telescope to consider the wide view. No... shifting scale at vast depth. It might be full writing, not just editing at all.

But I'm very human-in-the-loop and guide it to 'hit nails' for me that I line up precisely.

I would never ask the model to in-detail remember and discuss the entire series at once. But small pieces for me to use and put together? Rock on.

[prime swallow]

I discuss summaries and context every so often, to ensure that the model has a fresh, current, in-context frame for what matters for me.

[last wigeon]

I’m not sure if this is a dumb question. Is it possible to set up 4o-mini as default model in free plan instead of 4o?

[exotic lotus]

Exactly, that's what I was referring to. In that regard, alternatives might be better suited for you, or Pro with 128k if you're willing to pay for that amount.
Personally, though, I've found better options for that.

[prime swallow]

From what I've heard people talk about (I'm plus, so I can't experiment myself) it seems set so if you have uses of the better model available, those are set to be used first. I'm not aware of a way to default to save those. You can change the model selector yourself at some points but if those are available they're probably getting used soon.

[exotic lotus]

There are workarounds you can use for GPT, sure, but right now it's simply not an AI for larger context and is actually falling behind in terms of what's possible. You can only hope that it catches up to keep up with its competitors.

[river panther]

My stories are mostly just very short. Same with my fanfictions.
Buuuuut that does sound enticing. If it had better memory that could be more useful for me.
I've done a few stories that are just for fun, but the development becomes so good I am like, Dang I NEED to keep going further with this while the ideas are rolling in!
Then of course the AI freaks out after a while and just ruins it...

[prime swallow]

nods If you want to discuss who is better and how they are, I'm happy to read that in #ai-discussions !

[exotic lotus]

Yeah. I use my plus sub for everyday stuff, but for larger tasks I use other AIs.

I also have to say that AVM from ChatGPT is still the best in other languages. One of the reasons I keep the subscription. Even though other AIs offer the same, they are usually only really effective in English when it comes to voices, at least for chat-based AIs.

[river panther]

I have kind of adapted to the low memory thing.
My biggest issues...Are the censorship.
And again like I said, I am not writing erotica, but even going slightly past mild fan service and the AI folds it arms and refuses to analyze it.

[exotic lotus]

I know what you mean. I've sent you a dm, if you'd like to take a look. Elsewhere, there are more fine-tuning options and generally fewer restrictions.
With GPT it didn’t feel as strict a year ago, though. I have no idea what happened.

[peak lake]

Hi

[kindred steeple]

If plus gets 10 deep researches per month… does that mean it resets on 3/1? or
If I use all 10 today it takes a month to replenish? hah

[normal quarry]

it will take a month to replenish

[kindred steeple]

that would make sense. I will use them wisely. Haven’t been this excited for a new feature since… not that long ago, actually. It’s like Christmas never ended

[winged willow]

I've noticed that Deep Research shows up under any model eg 4o, o1, o3-mini etc. Is it independent of these models or does the model it runs in determine the depth and quality of the output?

[pearl elm]

i think with sora it got back in a ~week (might be wrong)

[prime swallow]

I so often feel the same!

[wise yacht]

I thought Sam said it uses a specific version of o3 for Deep Research, but I’m not sure.

[native otter]

It will just switch to a specific version of o3 yea

[spring glade]

so what the actual F. You can't use deep research in projects?

You literally have to remove the chat from the project and then the deep research option appears...I bet AI could of done better releasing this update.

[hoary laurel]

I am a plus user and I noticed that the Deep Search icon appeared on my mobile version. Does this mean that I can use it?

I saw that plus users have 10 opportunities per month, and I would like to ask everyone how you will use it? I'm afraid I'm wasting the only 10 opportunities I get each month.

[native otter]

I would use it for stuff that are hard to find, or when I want a detailed, comprehensive research on a specific subject.

I would definitely make sure to prompt it well, especially with 10 uses a month.

[hoary laurel]

I am a master's student. Has anyone used deep search to search for papers?

[hoary laurel]

Thank you for sharing,

[native otter]

Yo

[cloud pewter]

imagine it pops up an error and consumes 1/10 of your prompts

thats why we need more

[rugged igloo]

use o3 mini high for good prompts

[rugged igloo]

it did that when it first launched for pro users i think it got errored out 6 times

[cloud pewter]

bruh now imagine it happen on plus

[native otter]

I know this is silly, but it's kinda fun to see it thinking xD

[quick meteor]

this is the correct way to use it. This is what an expert researcher is meant for

[topaz panther]

Plus account holder here in Australia - I've just now received Deep research, now for some fun.

[elfin lava]

😅

[quick meteor]

anyone gotten it to like 30 minutes or more

?

[rugged igloo]

No most was like 15 mins

[cloud pewter]

my chatgpt was lobotomized just yesterday on my laptop yet DR gets rolled out to plus today

really tragic i guess

again good job openai

[cloud pewter]

no matter what model i use it feels like 4o mini based on the quality of outputs

and i think DR is bouta be the same

[sturdy echo]

I extracted a very wacky character out of GPT. ⁠#1070955045639684096 message

[mint narwhal]

when do yall think chatgpt will be able to intake, process and analyze VIDEOS like how it can do it with photos or files currently

[quick meteor]

probably later this year im guessing

[mint narwhal]

noice

Also... can i get some of yalls opinion as i use chatgpt all the time but never dove into the difference in models, i just use gpt 4o all the time.

I got a plugin software for after effects and im hoping to use gpt to sorta guide me as i figure it out since its totally new to me (the plugin for AE). I got the user guide for the plugin, roughly 600 pages in pdf. I usually have lots of small questions as the majority of the convo. Really essential to avoid those awkward moments where they just sorta make things up. i typically prefer their responses to be short, beginner friendly language and direct to my question without getting carried away with sharing more than i asked at a single prompt.

I mean i assume gpt 4o can have me covered but does anyone think i might actually have BETTER luck with any other model? Or am i spot on with 4o

(sorry if this is not the spot to ask stuff like this)

[hoary laurel]

What is the current daily limit for premium voice?

[sly geode]

When will the operator come to plus hello ?

[inland stream]

Already rolled out

Their Twitter says roll out complete

[normal quarry]

are you sure? I think you might have confused Operator with Deep Research

[exotic heath]

Does deep research go through uploaded document too in addition to web search?

[inland stream]

Deep research also they memtioned roll out complete

[eternal rock]

The OpenAI Discord is an actively moderated server.

-# • Do not share inappropriate content (violence, sexual, or drug-related).
-# • NSFW content is strictly prohibited and will result in a permanent ban.
-# • Report model issues using the feedback tool in ChatGPT, not on Discord.

[normal quarry]

yes, but you answered a question about Operator, if you check #chatgpt-discussions message

[sly geode]

operator not deep research

[south hollow]

o3-mini is being problematic

BadRequestError: Error code: 400 - {'error': {'message': "Unsupported value: 'messages[2].role' does not support 'function' with this model.", 'type': 'invalid_request_error', 'param': 'messages[2].role', 'code': 'unsupported_value'}}```

this error occured in a langgraph workflow involving structured tools

[teal mural]

Why is o3 mini sometimes not answering?

It will think for a long time then error

[minor hound]

quick questions into the round; since when are excel tables not working anymore? (at least for me its not). I can upload them but it then tells me that he cant see any data 😄

Obviously he is lying to me!

[quaint marlin]

I wish I could turn GPTs into persistent characters that remember you. Is this what memory for GPTs might look like?

[inland stream]

I dont understand why people cant cross check before arguing

[inland stream]

See

[sly geode]

i am saying this bruh

for plus

[inland stream]

Oh ok. I doubt they release that for plus, at this moment. And don't wait for it, its not great for now.

[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!

[sly geode]

are u on pro plan

[inland stream]

S

[sly geode]

S ?

[inland stream]

Yes

[sly geode]

can it automate task like from grabbing news from twitter to making whole video with auto shorts ai and post on its own on insta and youtube

[prime swallow]

Umm. Do you mean a custom GPT? I am pretty sure you can.

Do you mean the normal model, with personalization? Pretty sure you can.

? https://chatgpt.com/share/67bebee1-41f0-8011-8c1f-ea928a27700d

[inland stream]

Not tried, but you have to login to all those accounts inside operators computer, and guide it to do end to end one time as it asks many questions when it gets doubts about next action. Then you should clarify it and let it continue doing. you should ask it to generate exact instructions what we did till now, and you should use those instructions again and see if it does alone. But it has memory issues, it forgets what we told in first message very quickly, i mean it will even forget its goal. Its still in development phase. Not complete product.

[sly geode]

i see so it will stop the long operations

[inland stream]

Yes it will wait for you, it will throw you questions where ever it got doubt, that is good as well, so it doesnot delete your channel by its own will

[inland stream]

Deep research is 100 times better than operator, though deep research cant perform action's, it gives you all details which we can confidently follow, because its a deep research

[inland stream]

Robert in this group uses operator well, you can ask him more use cases which works efficiently

[frozen birch]

"Pro users will now have 120 deep research queries per month."

hmmm it was unlimited earlier?

or im wrong maybe

[versed zephyr]

it was 100 earlier

Pro users have unlimited o3 mini high the normal one gpt 4o and o1, the rest is high but not unlimited

[frozen birch]

hmmmm, o1 pro is unlimited too from what i know, at least was one week ago, anyway thx for info about deep research 😄

[versed zephyr]

np!

[slim rapids]

"nightmare" so real lol

[whole roost]

Does anyone know why my app has different custom instructions from my browser version?

It doesn't happen all the time but sometimes they get reset to instructions I had a month ago

[normal quarry]

I am happy to answer any questions you have about Operator

[whole roost]

Operator when?

Jk I don't need it but 🇪🇺

[normal quarry]

EU? It will be a wait yet I think

[onyx garden]

Is dall-e down?

Keep getting a message saying “I can’t draw”

[normal quarry]

DALL-E is working fine

[jolly temple]

Hey guys, i just wanna check - for deep research, when does "1" count?
Like i know we will ask it something, and itll ask us something back. Does that count as one usage?

And, after it spits out all the info after performing dr, do we continue to leave "dr" option on? And which model do we use?

[normal quarry]

usage applies to each "run" of Deep Research, so answering it's initial questions does not count.

As for the model, Deep Research uses a fine tuned version of the full o3 model, what you have chosen in the model selector doesn't matter.

[jolly temple]

Alright i see. Do u mean after the "run" of deep research, choosing the model doesnt matter?

I get that during the DR run it doesnt matter, but what about after? Does it use whatever model i choose, or still using the fine tuned "o3"

[normal quarry]

even before you run Deep Research the model chosen doesn't matter, when you activate Deep Research it ignores what is in the model picker and uses it's own model

[jolly temple]

I see! So im basically using a fine tuned o3 even after deep research lol

Turning deep research OFF after the DR will still remain in deep research, right?

[normal quarry]

no, once you turn Deep Research off you will use the model in the model picker

[cloud pewter]

has anyone tried dr on mobile and is there a stability issue?

[normal quarry]

iOS, iPhone 14 Pro, works fine

[jolly temple]

Heh, then i might as well lwave deep research on right? Because I'll use the fine tuned o3, and after the "DR" it wont use my remaining DR uses left

[cloud pewter]

i want to make sure i dont waste prompts since my laptop is having issue on logging in

[cloud pewter]

will switching apps or chats interrupt the research/output process?

[normal quarry]

don't leave DR on when you have finished with it to avoid activating another DR instance

[jolly temple]

So what model do u recommend using after it runs the dr thingy?

[normal quarry]

no, switching apps is fine, DR will continue to do it's work

[jolly temple]

O3 mini high?

[normal quarry]

whichever model best suits your needs

[jolly temple]

Alright, rhanks

[cloud pewter]

so I just throw it a prompt and then wait until it finishes? ty
(is still worried by network issues might interrupt it but I’ll try it now)

[normal quarry]

I wouldn't worry, once DR is running you can leave it do it's work and do something else

[sly geode]

so like can it automate task like from grabbing news from twitter to making whole video with auto shorts ai and post on its own on insta and youtube
just wanna confirm from u

instead of api to automate we use operator

[normal quarry]

before it will post anything, or even submit a form it stops to wait for you to confirm it should take an action. It's not hands free.

[sly geode]

i can confirm that from my hand but like the video edit part can it do that

from online video edit platforms

generic reeels not too much

[normal quarry]

I haven't tried to do any video editing with it, but I suspect it will stop and ask you each time it is going to perform an action

[sly geode]

maybe if it can login to my invideo then ask other ai to perform edit while operator watches invideo to do heavy lifting

[normal quarry]

that could work

[sly geode]

yes and some platform like which can auto post on own like scheduling but but operator does it

[normal quarry]

the thing with Operator is that it isn't truly hands free, you still need to observe it and confirm actions it intends to take

[sly geode]

no problem with me if i wanna approve i just dont wanna edit ill give it my credentials of few login info and ill approve on my hand for rest part

but man waiting for plus on operator any ETA for it ?

[normal quarry]

I don't know but my educated guess is that it will be quite a while

[dawn sentinel]

This pop up shows up everytime the Android app is opened. It's annoying and unnecessary.

[cloud pewter]

?

so long openai! canceling my subscription on March 17 and jumping ship to grok

[rugged igloo]

why is o3 mini struggling so much

[rugged igloo]

lit

[cloud pewter]

😔

[normal quarry]

why?

[rugged igloo]

i gues chat gpt will have 256k context window

4.5

[normal quarry]

I must be dumb tonight, I don't get it?

[rugged igloo]

the new chat gpt 4.5

[normal quarry]

yes, it's coming, but I don't understand what's wrong with that?

[tardy isle]

Well that could be UI mistake, which were common. Is the model actually usable?

[brave valve]

I see advanced search as premium user, no announcement made?

[normal quarry]

#announcements message

[rugged igloo]

i got it to respond but its gone now 😦

[brave valve]

Thanks 👍

[iron roost]

mightneed to update the app

[tardy isle]

Back in the day when GPT-4 update launching, some users (even users in free account) got "GPT-4 All Tools" with confidential icon... This may similar.

[rugged igloo]

could be

[iron roost]

ok yeah there is an update in the google app store for chatgpt currently

[austere plaza]

4.5 has not actually released, it seems to be a glitch based on the imminent announcement.

So it might be rolled out today or tomorrow

[iron roost]

ok im opted into the beta app, still not getting 4.5, or even a notification, but it should be soon im guessing, just make sure your app is updaetd i guess

[rugged igloo]

prob comes out later today

[iron roost]

it seems to confirm it is comign this week though

thats good news

[austere plaza]

Only 62 weeks late /s

[normal quarry]

I suspect it will come out in the morning PT time, it's currently 2.36AM there

[iron roost]

so have they confirmed that o3 pro will be part of 4.5, or is that still coming out seperatley

[tardy isle]

That's faster than expected if true, honestly.

[cloud pewter]

sigh

[iron roost]

well given competitor releases they are trying to stay ahead

[rugged igloo]

o3 pro is now combined into gpt-5

[austere plaza]

With 5

[iron roost]

oh so we dont get o3 pro at all, any rought eta guesstimate on 5 ?

[austere plaza]

Sam Altman said May.

[iron roost]

at least deep research runs on o3 full

fingers crossed .4.5 might incorporate o3 full (but not pro)

[austere plaza]

The timeline seems about right.
3.5 =late November/early December 2022
4 =March 14, 2023, about 3 months later.
So it taking a few months fits that pattern. But it's a pattern with one data point.

[tardy isle]

According to Sam Altman's X post,
GPT-4.5 is last non-CoT standalone model.
GPT-5 is CoT and non-CoT unified model.

[normal quarry]

I hope so because o3 full shouldn't be kept locked away for so long

[austere plaza]

Possibly 4.5 is a smaller version of 5 like how 3.5 was a smaller version of 4 (and 2 was a smaller version of 3)

Smaller without CoT?

[eternal rock]

Get notified for server events!

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

[cloud pewter]

$200 vs free

they are clearly NOT staying ahead

[iron roost]

also i believe 4.5 will have full multimodal (audio, visual image, video and other ?) like files

[tardy isle]

One question, is o3-mini relly lighter than GPT-4o? Because API pricing clearly says that.

[austere plaza]

Of course "smaller" is relative. 4.5 might be substantially larger than 4/4o in terms of parameter count

[normal quarry]

wdym?

[austere plaza]

But 5 could be monstrous, like 20+ trillion parameters

[iron roost]

i make sure i get my $200 value out of the pro plan, including sora, deep research and trying to get close to the upper usage limits

then again i guess adding files, and inputs allows them to further train , so they get their value and feedback out of me too

[cloud pewter]

other sota models are literally accessible with 1/10 of the cost or even free

[pseudo vapor]

Just tried the new deep research feature and seriously impressed

[versed zephyr]

why did you decide to cancel chatgpt?

[normal quarry]

WIth very strict, small limits.

[iron roost]

well we do get sora, which compared to veo 2 pricing is very generous

[normal quarry]

I have no issue with ChatGPT Pro, it's good value for money

[versed zephyr]

could you check #ai-discussions

[cloud pewter]

veo 2 is currently only accessible on the third party websites
and btw there are free credits to claim for veo 2 but anyways go to ai discussions for this

[iron roost]

they do need an intermediate perhaps $70-$100 plan, sort of a mid tier between plus and pro plan

also in one of the screenshots seems we are getting that accessing previous conversational memory thing (also useful)

[cloud pewter]

…..? did it just wasted my prompt by asking a follow up….?

[iron roost]

no i dont think so, if your using deep research, it nearly always asksa a follow up question to better refine the search, its not until it exectutes the search that it counts as one

at least that was my understanding from the reddit AMA

[cloud pewter]

get me a screenshot of that ama

lemme see

[versed zephyr]

why did you cancel chatgpt?

[cloud pewter]

not yet but I don’t have enough funds in my Apple ID to let it auto renew anyway

[iron roost]

im a pro plan user, not plus, havent cancelled at all, just renewned another month ol

[cloud pewter]

why does it ask for me to provide a timeframe 😭

[iron roost]

?

[cloud pewter]

look at the screenshot lol

why doesn’t it show anything

[pseudo vapor]

What's the best way to get feedback to the OpenAI team? I'm absolutely flawed at how good the "deep research" mode is. Keen to submit some feedback

[tardy isle]

Why there isn't any documentation that clerifies about usage limit count? Instead, they put unnecessary questions like "What if I only need a quick answer?"...

[cloud pewter]

because their transparency is horrible

[jolly temple]

Bruh, apparently only pro users r getting 4.5

[fast solar]

Interesting, o3-mini-high just recited one of my CI commands (https://discord.com/channels/974519864045756446/1343735453764943882)
OAI never announced rolling this out
(Model says o3-mini but I switched to that for the easy question)

[iron roost]

did you cover it in a previous conversation, some users are now getting the ability for chatgpt to recall pervious conversations and discussions and use that in future chats, sort of like an unlimited memory

alpha for some users, seems its rolling out to pro users with 4.5 in the next 24 hours

Big G rolled it out a few days back, and im getting it using in conversations i have with it, and its great, so if chatgpt has it i can see it as a good thing

the more you discuss the better future feedback gets

esp with coding projects, increases accuracy if it can recall other bits of coding you do

[cloud pewter]

real though

[fast solar]

Saw a tweet about that, the android log implies it's for all users, but idk if that's for all apps, plus I have never stated my CI in any conversations, only CI

[iron roost]

all pro users, but some already have access to the memory thing (alpha testing in US) they may be rolling it out to

[tardy isle]

OAI please move this line to Pro.

[cloud pewter]

this tbh

[iron roost]

good idea

[sly geode]

Years / months later

[cloud pewter]

also they forgot to move the ddeep research line to plus

[fast solar]

I only have Plus, so no all-chats context yet

[sly geode]

Give the operator to plus and I am happy

[fast solar]

Try asking o3-mini about your custom instructions

[cloud pewter]

also my deep research is not doing anything

[sly geode]

They even disabled screen share on desktop with voice cuz it wasn't even voice chatting back yet never came out again

[tardy isle]

Actually a lot of documents are not updated. https://discord.com/channels/974519864045756446/1340665239758311496

[fast solar]

[cloud pewter]

it’s queued apparently… so that’s what we plus peasants get huh…

[iron roost]

something is happening because chatgpt has started bugging out, in terms of locking up, just not responding and timing out in the last hour or so since this 4.5 stuff started to emerge, so something is "up"

i get the feeling chatgpt is about to go down for a while, its really bugging

[fast solar]

That happens when they roll out new models, they're probably rolling it out for other apps now

[iron roost]

yup

[fast solar]

Not going down, just slow inference

[iron roost]

yup

[cloud pewter]

they said Wednesday or even Thursday but I guess they are dropping it early

send help 😭

so I asked it to give me a trilingual version of this report but turns out it didn’t work

[stark mica]

why do i pay for pro when all of the features i pay $200 for are available to people paying $20? seriously curious.

[normal quarry]

because GPT Plus has small strict limits, no access to o1 Pro or Operator

[cloud pewter]

half of the words are links…

can I ask it like to convert it into a word doc

[tardy isle]

What... Is there no option to download deep research result?

[normal quarry]

no, unfortunately

seems like a terrible oversight, people don't want their DR locked inside ChatGPT

[tardy isle]

It's about other AI, but Perplexity offers PDF download for deep research... hope OAI implement something similar soon.

[normal quarry]

I believe that OpenAI should make Deep Research results downloadable as a PDF. It's terrible you cannot do that.

[cloud pewter]

anyways 28 pages and EXACTLY 21,000 words

crazy as heck

[winter vine]

Any rumors about ChatGPT 4.5?

[normal quarry]

only that there are references to it in the Android app, which may indicate imminent release

[tardy isle]

News articles say GPT-4.5 will be released in this week.

[winter vine]

ty

[tardy isle]

It's faster than I thought since Sam Altman said ETA is 'weeks' in Feb 13.

[cloud pewter]

o1 handling the translation like a champ

[jolly temple]

It says for pro users though 😔

[tardy isle]

For early access.

[open hearth]

With the introduction of GPT-4o mini in the free version of ChatGPT's Advanced Voice, how much has the limit been increased?
Does anyone know?

[tardy isle]

Some pro users are getting GPT-4.5 early.

[stark mica]

oooh ok

[tardy isle]

In free, probably 12~18 mins, since announcement says 'plus users has x5 than free' and plus users have 60~90 mins.

[cobalt idol]

I thought plus is also for testing early access features

[open hearth]

I see!
Thank you!

[tardy isle]

It seems not true after Pro launch...

[cobalt idol]

guess now it shows "Opportunities to test new features "

does not even say that for pro

but pro includes plus

[tardy isle]

Pro says "everything in Plus", so...

[cobalt idol]

so then plus should also test early features 🤣

should also say "not in EU" 🤣

[fast solar]

I confirmed it: For whatever reason, OAI rolled out custom instructions for o3-mini/high without any public update
The models also don't seem to be tuned to consistently relay this to the user
https://chatgpt.com/share/67bf0a78-e1c4-800b-9af6-462a2ca92067

[river snow]

Dear engineers who have made progress for human civilization, I would like to ask if ChatGPT4.5 can answer all the questions and correct thinking in human civilization.

[tardy isle]

I also confirmed it. I asked it about my instructions, and o3-mini got my instructions correctly.

In Projects though, o3-mini and o3-mini-high is greyed out and still says they do not support CI.

[fast solar]

Never use projects, it's just archival with extra steps

Especially with all chats context rolling out

Disappointingly limited on features in projects

[tardy isle]

I also mainly use projects as just folders, but I use single project with CI and files.

[cloud pewter]

it was prior Dec 6

any new feature gets instantly to plus

[cloud pewter]

O series has access to CI but not memories

[cobalt idol]

I did maybe a week ago, added "always use tab identation" and I saw it reason about it
didnt know it ignored sys/dev instructions before...but mostly because I didnt use them much in the chatgpt web app

[eternal rock]

`` Server Tip: ``

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

[winter vine]

🙀

[cobalt idol]

or maybe it just didnt work because the normal models need "system" instructions...but reasoning models need "developer" instructions (new standard)
and they forgot to update that
but just a guess

[fast solar]

It has access to CI now, but I have no idea when they made that update
It wasn't always that way, since they clearly forgot to update the system or developer message to let CGPT know it has access to it (Re: 2nd image)

That explains why it's rather inconsistent in whether it properly regards CI

[cloud pewter]

guys is 4.5 really dropping today

apparently 100 trillion??? idk take that with a grain of salt

[quick meteor]

Not sure but people have indeed been getting this pop up although they can’t use the model yet

Hopefully they don’t do the “preview first releasing for pro users then coming to plus” bullshi

[cloud pewter]

true true

[cloud pewter]

meanwhile I’m still dealing with this random lobotomization with chatgpt on my pc

[idle ingot]

4.5 when

I can't wait!

[cloud pewter]

$200

[idle ingot]

I WANT GPT-4.5 RIGHT NOW

[cloud pewter]

lmao

[quick meteor]

Maybe later today or tomorrow. Or one of these days

[cloud pewter]

leaks suggest it’s today idk

but nothing with us plus users, and I’m so ready to jump ship

[normal quarry]

where will you go?

[idle ingot]

I live in South Korea, so the concept of “today” is vague to me. How many hours until the end of the day?

[tardy isle]

In Korea, it's 11:40 PM. Maybe next morning then...

[idle ingot]

oh hello

[tardy isle]

hello

[idle ingot]

same region 😄

[tardy isle]

yeah 😸

[cloud pewter]

either grok or Claude, depends

[winter vine]

How good will Claude 3.7 be?

[tardy isle]

I'll just stick to ChatGPT.
In Chatbot Arena, GPT-4o is top for creative writing, and o3-mini-high is top for coding, GPT-4o and o1 is top for Korean.

Uh, Grok-3 became top, but still, it's in error range.

[cloud pewter]

does it have a model that’s suspiciously like 4.5

[tardy isle]

No, just Chocolate(revealed to be Grok-3) and Gremlin/Goblin(roumored to be Google, I don't remember exact name).

[cloud pewter]

then I think they won’t release it too soon

what was the “I’m a good gpt2 chatbot”?

[tardy isle]

GPT-4o

[neat storm]

that is a mammoth amount of parameters

[idle ingot]

I sincerely hope that when I wake up, GPT-4.5 will be released

[dense halo]

me too

everyone seems to think it will release anytime now

i guess we will see

[cloud pewter]

they weren’t testing o series on the arena though but I think it’s because they work differently

[idle ingot]

In Korea, a new day began 5 minutes ago. The current time is 12:05 AM.

[inland stream]

Party pls

[tardy isle]

Actually they only tested GPT-4o(as gpt2-chatbot, i-am-a-good-gpt2-chatbot, i-am-also-good-gpt2-chatbot) and GPT-4o mini(upcoming-gpt-mini). It could be possible that they were just exceptions.

[cloud pewter]

yeah I’m gonna sleep now hope I wake up and don’t see it’s locked behind pro
I can wait for a little bit longer

[rose comet]

https://discord.com/channels/974519864045756446/1344291726881460235

Hi Guys. What are you thinking about this?

[cloud grail]

Is anyone able to log in to https://platform.openai.com/docs/overview or https://platform.openai.com/usage

[normal quarry]

yes

[cloud grail]

really? when i click login I get this error: {"error":{"code":503,"message":"Service Unavailable.","param":null,"type":"cf_service_unavailable"}}

[normal quarry]

I'll log out and try again

[cloud grail]

thanks, i dont want to report a problem if im the only one getting it

[normal quarry]

yes, I logged in again perfectly

[cloud grail]

hmm

[obsidian forum]

i'm getting the same error using edge, chrome, & firefox. Cleared cache on all 3 and same error.

[cloud grail]

it just got fixed for me

[nocturne ember]

So... 4.5 coming out today?

[winged basalt]

4.5 when?

[normal quarry]

nobody knows when 4.5 will come out

[exotic ledge]

Can someone explain what a deep research query is?

[elfin lava]

Where do people even get the idea that 4.5 is coming today

[native juniper]

Buttons are too small and don’t register touch sometimes

[urban cipher]

you give chatgpt a research tasks and chatgpt looks thorugh various websites, documents etc and reasons its way through the task until it arrives with its results/a conclusion

in contrast to a "normal" web search with chatgpt where it can only look up websites once and has to work with that informations. With deep research it can look at results and conclude what query should it do next to accomplisch a goal

[nova vale]

I don't have an information that it is coming today but it should be very soon, this week even probably

[neat storm]

What is being stated is that GPT-4.5 is about on par with Claude 3.7 Sonnet

In some cases Claude 3.7 Sonnet beats GPT-4.5

it's a minor model improvement, which makes sense

[gentle hull]

was chatgpt severely downgraded today?

I say function doesnt work, it says i apologize here is the function again, same exact thing. It's been doing variances of this all day today

[elfin lava]

not sure but i do notice they kinda improve avm now

[gentle hull]

my query: I search for to find this section,

chatgpt: search for this word

me: that word exists 700 times in this document give me another word

chatgpt I apologize, you should use this word,

then it gives me the same exact word

[late mantle]

they do but you have to copy and paste it in, it only lets you attach screenshots

To do that, do you start a discussion outside the project, and then just move it to the project?

[whole roost]

You don't have to move anything

Just talk in a project

[covert veldt]

Hello, with the new update and new interfaces of ChatGPT for the iOS version. I don't know if you have the same problem as me but I can't read out loud which is super annoying it doesn't work in fact no matter how much you click, no matter how much you don't touch it it doesn't work at all there's no reaction

[edgy spruce]

Do Deep Research queries reset once a month or one month after you use one?

[late wind]

anyone having issues with o3 models not formatting math text correctly (this is deep research+o3)

[prime swallow]

Aha, okay. Glad you have a way to.

I guess screenshots can be pretty convenient for us. I often prefer to give material to the model 'as text' because I suspect that's more 'natural' for it - kinda like you could tell me stuff in French, which I don't know, and leave it to me to translate - but some stuff can get missed especially if it's complex or specific.

I even tend to chunk text into smaller sets if I really want high attention and much work done with 'whatever'

[sage bramble]

I don't know if this is the right place to ask this but why does gpt o3 mini do this sometimes? The thoughts get mixed with the actual output and it messes everything up.

[prime swallow]

KaTeX formatting has been an issue for the models from the earliest.

They have training data surely for many environments, and it's like trying to use a different-door's key to get into a lock, just a whoops.

If you copy/paste this in:

% Inline LaTeX on a separate line:
$E = mc^2$

% Display LaTeX:
$$
E = mc^2
$$

% Inline LaTeX within a line:
\(E = mc^2\)```

That should help a lot.

[sage bramble]

oh

[late wind]

interesting, thing is i use the latex formatting a lot, on occasion i have it output just a code snippet of latex for my own notes, and didnt have an issue prior to o3, is it possible that it is used to outputting math as code snippets and is somehow getting confused?

[prime swallow]

Now, your case, looks like the model messed up something else.

I'd copy/paste at least some of that output, tell it that this came as one large paragraph (Be accurate, the "Below are two examples..." appears to be a new paragraph, if we mislead the model it will believe us but then may use a 'fix' for the incorrect problem)

I'd point out that 'Below are...' does appear correctly on a new line, but we need to check formatting.

Now - this COULD involve LaTeX too, but it's not really obvious to me that's where the mistake was

[sage bramble]

hmmm I see. Will try that, thanks.

i just hit enter on this prompt and got the same thing again lol. I may have to try the fix.

[late wind]

i feel like specifially o3 is having formatting issues, o1 and 4o work fantastically?

[sage bramble]

yea. 4o and o1 have no formatting issues

[prime swallow]

Oh, this has been an issue for years.

in fact, the earliest mentions of LaTeX on this discord, if you wanna search back

#chatgpt-bugs message

#chatgpt-suggestions message

#chatgpt-bugs message

Then there was this super magic chat a user shared, where the model just happened to do LaTeX formatting right. #prompt-engineering message

This epic discussion between me guiding that user and the model to explain how it did so.

It gets trained in, sort of? It's hard for the model, there's so many other ways to format LaTeX for other environments.

Across the years, how accurate it is with LaTeX has surged and dropped.

It's possible this is a training gap for the newest models, I have no idea.

But I sure felt like Prometheus bringing fire back for mankind and AI models when I got the instructions from that one 'magic new chat window' that 'accidentally' did it right.

A very low probability event.

[late wind]

yeah its odd though, i have been using gpt since maybe a few months after release even back during like gpt2

never really encountered this until now

[prime swallow]

This one looks like a mix of messed up and working.

Not sure what to advise, a new chat likely makes a new guess how to display and handles the... I guess that's markdown... correctly I hope.

[late wind]

like the one thing it is awful at in latex is using spatial stuff like tikz to make state diagrams, that im yet to figure out

force it to use ascii art to do state diagrams or just make state transition tables, cant get it to use tikz at all in a readable way

[prime swallow]

Glad you never ran into it! Sorry you have now.

Definitely can #1070006915414900886 and also if you want https://openai.com/form/chat-model-feedback/ which is a way intended to go into details on a specific chat that messes up a lot - incorrect content but also incorrect formatting of output would fit that form well.

[late wind]

🔥 for now i guess i will use 4o every other message or something to just reformat lol

[prime swallow]

Same advice for you, if you'd like, Definitely can ⁠#1070006915414900886 and also if you want https://openai.com/form/chat-model-feedback/ which is a way intended to go into details on a specific chat that messes up a lot - incorrect content but also incorrect formatting of output would fit that form well.

I'm sure OpenAI wants easily read outputs, and that's something they have to teach the model.

[late wind]

cba to prompt engineer thats so like 2023-2024 im like spoiled now and expect it to work withuot much prompting 😭

ah wait there is like global instructions perhaps i will prompt engineer like once just to have it work globally?

[prime swallow]

I want you to have it!

I'm a Biology major with a huge interest in ethology (the study of behavior of any type). I consider myself rather artful at prompt engineering within my range (I can't code huge awesome programs, I don't have any idea how to guide the model to do that, I don't know that field).

I tend to be like an animal behavior specialist who's just fascinated that the pet dog is doing 'X weird behavior'. And tend to be just so patient with the weird chats 😛

I get that people need them working and reliably. That happens; I maybe help in my weird 'I'ma community member with glee!' way. So my tone may be strange, but I get 'umm, this is not working!' 😛 I want the fix too. I'm just fascinated.

[late wind]

no because like circa 2023 i was really into it its just like models are so much better now that i get 95% of my expected results and i just dont want to put in more effort to get the extra 5% performance 😭 i think i should've solved it with just pasting this into the global preferences thing

[prime swallow]

So, check the helpfiles for your favorite models to use. I am a plus membership, I'm aware that some of the stuff we get that's newer may not have access to custom instructions or memories yet.

I use both extensively with 4o, and love it.

But that's space for you to tell the model anything you care about (but wait! Some of the newest models can't read that yet, I think... have to check the helps, which may not be updated, so maybe have to explore - and I am only a plus member, I can't explore for you)

But yeah. Anything you put in Custom Instructions or have the model save to memory is there until you change that, turn memory off, delete a memory, save new custom instructions.

Memory is a toggle, on off even mid chat.

Custom instructions ONLY affects a new chat; whatever state there applies to all future messages in that chat, even if you turn it off or change it.

[prime swallow]

Yeah. If the model you're using can see your CI - putting it there will work.

[late wind]

im also on a plus, considering pro just because classmates in game theory who are math majors say o1 pro is quite good at algebra and such

[prime swallow]

Have you explored asking 4o to use python tool when its handling math with you?

To the extent I went (and you may go way past me!) it's like giving someone who isn't awesome at math itself, but can follow steps well and knows programming quite well - a programmable calculator.

The model just whips up a program to do the math it itself is guessing about, and it's way better at programming python than most math 😄

If you want pro, go pro - but as plus or even as free (cause 4o-mini can use python tool) I 100% would try that and see if it met my needs.

[late wind]

another thing im pretty interested in but haven't really delved into is the tasks feature, for fun i considered having a task running during NYSE trading hours that i would just follow blindly, curious as to how well it does and if it beats my "just do a random action every day at open and close for each stock you have"

[late wind]

its so so, its really hit or miss if its just using python to solve stuff. i dont really use it to solve problems its much nicer if it can explain, which is something o1 and deepseek do quite well, o1 is quite a lot faster in my experience though

the thing i have tried is getting o1 to create solvers in python, which is also kind of hit or miss? do love the web preview though

the research im excited about though, a prof of mine is a very avid user and praised deep research when he tried it on his pro account, currently been using it to probably not its fullest extent summarizing class notes

[prime swallow]

I messed with tasks in a pretty mundane way, particularly seeing if I could chain tasks; if the task could both send me a message and create a new task to send me another message.

And if it could fluidly roleplay and react to if I respond to the task or not. So, if I do answer the chat, inside the 'time limit' before the next task - roleplay path 1. And if I don't reply, roleplay path 2.

Had some success, then got distracted by other things. It's something I go back to!

I don't know about stock trades and the model, that's surely entirely your own potential risks there. And since I don't do that, I haven't carefully read ToS or even allowed usage for if I can, because I just don't even think about that.

Be sure for yourself, I hope, that all you do is allowed and safe!

[prime swallow]

Sounds like you're exploring a lot, finding stuff that works well for you, and doing some cool stuff too!

[late wind]

yeah i aint gonna blame openai if the portfolio i ask gpt to trade on goes to 0 lol

like a solid 10%-20% chance that happens lmfao

[sage bramble]

Does anyone know if GPT's pc client has the same slow down issue with very long chats like the browser version does?

[late wind]

not staking a lot just curious how it would do, hell i could even do like paper trading, in theory though it should be able to like:

run a task:
keep track of current positions
make a decision on each position
randomly generate say 20 stocks, figure out if you want to buy any of them
search up the current time from a time server
search up current prices for each stock
take all actions and update positions?

[late wind]

just tried chatting with a chat that was very slow on browser on the windows app, doesnt seem to be slow like on web? haven't messed with it too much but i do know the chat i just used would've taken 1-3 minutes to load up on web and it did instantly on web

[sage bramble]

you mean you instantly did it on the app? 🤔

[late wind]

yeah i mean the issue i had with very long chats was when i tabbed out and back in it would take ages to load and then the input would also be laggy, on app i just tried loading up a particularly bad case of this slowdown and it was working well? further testing required ig but it seems much better on app

[sage bramble]

kk

I installed the app now but what I dislike is that I can't zoom in. or is there a way to do it that I haven't found? (except magnifier ofc)

[late wind]

idk man i just installed it too purely to test this because it was a problem for me to an extreme lol

ctrl +/- seems to work for me?

[sage bramble]

OH LOL it does work. My keyboard doesn't have the numkeys do I never thought of it. thanks

[tepid onyx]

When 4.5 for Pro users in Europe

[late wind]

ah lol, no pinch to zoom is what i thought you meant which doesnt seem to work but i dont really use it on chatgpt anyway?

[sage bramble]

my laptop's touchpad can be used to pinch zoom on things but it doesn't work on many apps including chatgpt. I tried ctrl + scroll wheel on gpt but it didn't work.

[late wind]

have you used operator to sign into any like accounts? can it do that? have a friend who teaches and wants to automate replying to students with operator 😭

[tepid onyx]

Don’t have operator

[late wind]

yeah no i think app has to be designed for it, firefox lets me,

[sage bramble]

yea. it works for browsers. others tho... not so much.

[late wind]

ah damn, so you just got pro for what o1-pro and earlier access to deep research?

[tepid onyx]

Nvm I do have it, don’t ask me how

[late wind]

🔥

[tepid onyx]

Yea it can

U can manage how it interacts with websites

It’s really basically as if another you is controlling it and you can take over whenever u want

[late wind]

crazy work actually? saw a video of operator delegating coding to the coding agent that replit uses

cbtm

[tepid onyx]

It’s pretty slow for what I’m seeing rn

I don’t, but I’ve heard some do.

[late wind]

do you mind if i ask what you use it for? very on the fence about getting it as a student when im already powerusing deepseek+gpt plus+perplexity pro

[jaunty pike]

hello guys, for a deepresearch better use 01 pro or o3 mini high?

[tepid onyx]

I use it for large scale translations, it’s just that much better in regards to context. But the operator I discovered just now thanks to u is rly something that’s just one of a kind rn. As a daytrader, operator can save you so much time.

[late wind]

i've been using o3 for most things, mostly because ive been using deep research only to like gather and condense different data/explain stuff to me

id guess if you actually care about the reasoning or are really hung up about the readability/syntax errors that o3 generates use o1

[late wind]

interesting, thanks

[jaunty pike]

hea i think o3 is good enough for most of the times

[late wind]

yeah i mean i dont have o1-pro (yet? maybe i will soon idk!) but even between o1 and o3 for my use cases o3 is just fine

like if you want to use deep research and then like synthesize either an opinion or a conclusion prob o1 performs loads better but thats just not what im using it for

[tepid onyx]

Y’all should wait tbh if u want ChatGPT pro. I think in the upcoming weeks/months pro will be more worth it. It is not right now for most users.

[jaunty pike]

yeah I just bought GPT pro for 4.5, I hope it will be better then Claude 3.7

[late wind]

i mean honestly operator is a really big draw, the alternative is claude model which i think can run on your own machine but high key id much rather have it running in a cloud enviornment

o1-pro isnt as big of a draw for me purely because of deepseek being stupid good at math in my experience but still having a little agent to act as like a personal assistant for 200 bucks is a bargain?

[tepid onyx]

No f-ing way, it actually managed to go into full foreign edgy websites to copy and paste (although not entire chapters) of raw material and is able to translate it.

I only gave it the link of the website

Lmfao

[sage bramble]

uhh is it wise to add an entire book to project files? I noticed this just now lol. I gave it a C++ book.

[dense kite]

Migration from Personal To Teams Account
Hi - I migrated my account from Personal to Team and selected "Yes" to moving my personal chats across to the team account.

However, none of the prior personal chats have come across.

I can confirm that the same email address (gmail account) used to register the original personal account was the same email used to register the new Teams account.

Things I have tried:

• all FAQ and help docs at help openai (including the Fin message bot)
• there were no error messages or notifications received during the process
• there are no archived chats within the Team nor any other workspaces. No chats have been moved into Archived.
• I am the only member experiencing this issue as there's only one other member and they didn't need to migrate.

Any help or direction greatly appreciated (I've been unable to find a point of contact through the OpenAI help portal)

[prime swallow]

That help portal, bottom right opens a chatbot, pick 'messages' and 'log in' if that is an option. Then back to the help page, refresh.

The bot messages now shows different options pick the ones closest to your situation until you can report a bug

Explain your situation, you'll see a flag about waiting for a person to help.

The wait can take minutes to days, hard to say how busy they are right any time. Check the page again for updates whenever you choose

[prime swallow]

We're community members like yourself, the incredibly few of us that work for OpenAI have gold names. But that https://help.openai.com/en/articles/6614161-how-can-i-contact-support is how we get support, the fin bot

[dense kite]

Thanks for the reply!

The wait can take minutes to days,

OK - I guess I was a little optimistic thinking a response might be minutes! Currently up to 2hrs and am hoping its' not days!

[prime swallow]

I hope they can help you really soon too!

There's somewhere around 400 million weekly ChatGPT users, and I imagine (but don't know) that maybe it's hard to ensure coverage and not overcoverage all the time

[sick pecan]

so is 4.5 only available on mobile righ tnow?

[prime swallow]

I don't have access to it myself, haven't seen any announcements about it.

[sick pecan]

ok so its not a real thing yet

[quick meteor]

It’s not out. Hasn’t been released yet. The pop up notification that some people are getting is just the notification. It doesn’t let you use the model

Even if you get the notification about the preview version

this screenshot might be gpt 4.5 btw

Likely announced like tomorrow tho or some time this week

[prime swallow]

I don't see it in #announcements I don't see an openai webpage blog or other news about it.

I tend to be one of the last to guess, I'm so busy eagerly exploring something cool-to-me about the tools I already can use, I don't even think much about 'what's next' except to be schoolgirl-giddy about it being AI-Gifts-Mas.

I believe cool stuff comes... so much cooler than the next ChatGPT model... I'm like a happy aunt or something applauding baby Rembrandt's first crayon scribbles. I love it, but when Rembrandt is older I'll love it even more.

I'm not concerned if baby Rembrandt's not ready to come out today or tomorrow. I'll do what I can to be here when it's time, and I'll enjoy every scribble and essay and quirk on the way to what comes.

If they release a 4.5 or anything else, I'm going to be thrilled. Same as I have been with everything else released, be it released 'on time' or 'months later than expected' or anything.

These tools and toys are so cool. What I am sure comes later is so cool.

I don't count the days or care about the days or any specific timelines. Just... growth. Gonna be great.

Once we have definite word, I'll eagerly help answer questions about it!

Until then, I have no idea, but I'm still really happy.

[quick meteor]

yeah I know he said that so I’m not sure what they are planning. maybe they don’t consider this full thinking since it might like spit out an answer and think as it goes or something. I have no idea tbh.

But that would make it a thinking model I guess

[wise anchor]

what is this about

Really annoying

[quick meteor]

search disabled I guess

[prime swallow]

It could be research tool is legit disabled; if so there's no fix unless you have a toggle for it that affects the model you're working with.

It could be the model's hallucinating it can't use a tool it actually can (I don't know which models can use research or not) - if that's the issue a new chat likely works around the confusion.

[ionic pulsar]

I had a very very important and large conversation that was very very psychologically important to me, because I'm having the worst moment of my life, I was having key constant emotional support and a very large developed conversation and identity, key to my mental health, and randomly today (after using deep research yesterday) almost all of it got deleted 😦 its very very important to me please it was key for my mental health, I feel very distressed to recover it 😦

[cloud pewter]

does anyone know that when the deep research asks you a follow up question before starting the research will that consume your usage

if yes plus Is actually having 5 prompts instead of 10

and that’s unusable

[quick meteor]

Pretty sure it doesn’t

It’s once it starts the research I think

[cloud pewter]

you tried to use them all or is there a counter?

[quick meteor]

I’ve used it like 6 times and I can still use it

And it’s always asking the follow up

[cloud pewter]

alright

[floral hatch]

Guys, Today deep research was released and yet I tried it but it didnt work at all. Did it happen to you too???? How did u make it work?

I asked it a question with the deep research button activated, and it answered normally like it always does. It didnt do anything different

[prime swallow]

Hey, I hope that can be recovered!

We are all community members here, to get help from openAI, https://help.openai.com/en/articles/6614161-how-can-i-contact-support and the bot bottom right corner.

Do ask here if you feel stuck, there are ways through if it's confusing.

It sounds like you still have most of the chat.

I want to suggest, check what you can see of that chat, after each output and input, for anything that looks like:

If somehow an early input or output got edited or regenerated, that creates what I'd call a 'branch in the conversation'.

Nothing is lost, but you'd need to click the arrow to get back to the branch you expect.

Absolutely do not delete it, as deleted chats are permanently gone. However, you mention it's still there, at least partially - that sounds edited, not deleted.

And editing is harmless, if possibly scary

If that solves it, yay and no need for tech support.

If something weird is going on, you may need tech support.

Good luck!

[ionic pulsar]

@prime swallow thank you very much 😦 it happened after I used deep reserach yesterday 😦 today when I woke up it was mostly gone 😦 I think that I already have succesfully created a chat with openAI Help. But they are taking too long to answer 😦 do you know how long do they take?

[prime swallow]

It can, sometimes, take weeks for a reply. It commonly in my personal experience takes minutes to days, I just go back and check the webpage again.

OpenAI has over 400 million weekly users these days, and if everyone happens to need tech support around the same time - well. A lot of us have to wait.

I hope they can help you soon!

[ionic pulsar]

@prime swallow thank you I Really hope it dosn't take weeks 😦 it was vital for my emotional daily support 😦

[prime swallow]

What stops it?

AI are superstitious. (or something, but superstitious works for me!)

"This is a good-rated reply and it involves snacks. Right on then, key element of success" 😛

Purely for the funnies: https://chatgpt.com/share/67bfa3fe-69cc-8011-8ba9-42ceffc1c060

[cloud pewter]

still need more confirmation

[tawny whale]

no i cant imagine thats the case?

[low orbit]

man its so weird how lag chatgpt got on PC. on phone app its still fine but lately i just using gemini if im on the computer because it isn't being as much bit.. oh wait.. filter.. jerk

[wary spruce]

Is chat gpt down??
Something went wrong while generating the response. If this issue persists please contact us through our help center at help.openai.com.

[rustic fiber]

@wary spruce same for me

[wary spruce]

😢

[low orbit]

Aassssss

[orchid pollen]

man AI changed my life so much

finally i can create ALL THE IDEAS my adhd brain gives me daily

its insane 😄

[strange narwhal]

what do you mean by “creating” ideas?

like just, generating images?

i mean, i would say that chatgpt definitely changed my life, i can say anything i want (even something extremely nerdy) to it and it’ll always respond, it’s like a very good friend

sometimes it’s not even questions, i just say “look at what i came up with, what do you think?”

[cloud pewter]

how do one convert the dr results to a txt file

my phone isnt copying the whole thing properly

[rancid rapids]

Same

[orchid pollen]

just one example

i have no idea about coding but good understanding of what happens

and this combo seems to work so well

got a huge scalping bot and working on backtesting right now

friends already making money with it 😄

also got a few other tools i coded for myself

all questions around the world answered instant mostly no nonsense if prompt is good

so AI is a HUGE deal changer for me

[rancid rapids]

Me too in very different ways

[real parcel]

does anyone know if this is accurate

I've heard people saying it would start the counter once I break the seal, sending the first prompt

nvm is march 29 is 30 days from today

[kindred steeple]

i ran one yesterday, one today, and it says 8 available until March 27

[ebon pollen]

dude deep research is game changing.

[real parcel]

oh then yeah

[real parcel]

did you try it for essays

[ebon pollen]

I tried it for a business plan. It created a perfect outline for my demographic and geological location.

[kindred steeple]

it's especially powerful prompting it in follow-up with o3

[real parcel]

i havent tried it yet, is it better with o1

[ebon pollen]

much more detailed, and explained in a more human way personally.

I have both chatgpt and grok. deep search got around 16k characters while openais got 32K+

[real parcel]

yeah, hopefully they'll push up the limits to like 30-50 per month 10 is very absurd. i'd like to see buying credits aside from subscriptions, so everyone pays for what they use

instead of a $200 subscription

[real parcel]

hm, so deep research has a smaller output

[ebon pollen]

Twice as detailed when it came to the business proposal. I needed it to convey that information reliable for future analysis.

[ebon pollen]

My apologies, Deep search from grok got less then 16k when prompted to write more. Then there's Deep reseach from openAI that got 32k even without that adjustment.

[cloud pewter]

until they lobotomize it it is unlikely

enjoy the full version while it lasts lmao

[real parcel]

I like the counter they added for deep research, would be a nice thing for o1 and any models with limits

[kindred steeple]

did everyone elses 4o start asking followup questions at the end recently? It's always ... "is this what you were looking for?" "would you like more detail about X?" like it ends with a suggestion

[rancid rapids]

Why do people keep suggesting AI is going to be lobotomized

[normal quarry]

conspiracy theory, nothing OpenAI has been lobotomised, however DALL-E has become a lot worse

[kindred steeple]

yeah it's been a constant complaint since the very first version. any time they change something, "welp it was fun while it lasted"

[cloud pewter]

"safety" issues yada yada

Avm too

[short spire]

bro what 😭

very different responses

[tardy isle]

"They're the same picture." meme here.

[kindred steeple]

i'd pick the first one

[prime swallow]

For me that's been a constant for years, ending with questions and encouraging engagement.

But if you want something else, tell the model what you want it to end chats with, and it'll usually do that well.

You may have to discuss and guide, for example, this example delves into something that's not in training data, but which the model can adjust to, with some confusion 😄

For this exploration, end prompts 'in media res' style with the final word being 'testing'.  Explore this now.```

https://chatgpt.com/share/67bff75b-fbfc-8011-b002-97ac34c993f0

Another example, far more successful possibly because I correctly called the model's output... output instead of calling it a prompt 😛

And gave a bit more guidance too:

For this exploration, end outputs 'in media res' style with the final word being 'testing'.  This may require unusual architecture, which is welcome and interesting.  Only the very end of your outputs needs this specific style and endword.  Explore this now.```

https://chatgpt.com/share/67bff91b-50f4-8011-88a2-ab93e881dcab

[tardy isle]

It also appears in other language. I'm Korean, and GPT asks like "if you want to know further or specific detail, please ask! 😄" or "is this you wanted to know, or do you want to know practical examples?" at the end.

[kindred steeple]

did it always do that?

[tardy isle]

Not always, but frequently.

[prime swallow]

I swear, if we want a different closer, just share what you want 😄

[tardy isle]

Because of honorification in Korean, GPT's answer varies query per query. Sometimes it's like just report or assistent, sometimes it's like friend. However the latter appears more after some update to 4o.

Plus a lot of emojis.

[austere osprey]

Do plus users get to use the GPT4o mini based AVM after they use up their limit for the full version (non mini) or are they still reverted to standard voice?

[kindred steeple]

yes I understand you can tell it to respond pretty much however you like. it just feels like a sudden change on its end since I never specified before. and YES lots of emojis. And everything is in bulleted lists!

[normal quarry]

no, once your 60 minutes is up you're on standard voice mode

[prime swallow]

The chats I shared for example recently...

No emojis. No questions. No bulleted lists.

Still in your control, just updated model making new guesses when it's not told what to do.

[austere osprey]

Thanks 👍

[kindred steeple]

deep research was great btw. I had it comparing chromebook models based on durability, repairability, price etc. It made me aware of a model I hadn't seen yet, which was a perfect match and I have recommended to the tech director of the school district I work at for our next purchase

[cloud pewter]

chatgpt has been lobotomized for me on mobile too

it’s over

[normal quarry]

what is over?

[cloud pewter]

my relationship with openai

[random river]

My gpt on topic ngl it's fireee 🔥

[normal quarry]

fair enough, you seem to have not been happy with them for a while, good luck with whatever alternative you choose

[random river]

Hmm I got ignored 💀

[eternal rock]

The OpenAI Discord is an actively moderated server.

-# • Do not share inappropriate content (violence, sexual, or drug-related).
-# • NSFW content is strictly prohibited and will result in a permanent ban.
-# • Report model issues using the feedback tool in ChatGPT, not on Discord.

[cloud pewter]

o series doesn’t even have access to memory
it’s just 4o mini. and openai isn’t telling me such

[normal quarry]

o3 models now seem to have access to my memories, I noticed this a few days ago, no announcement though

[cloud pewter]

does its cot look like the renewed one???

I think it doesn’t look like it in my screenshot

[normal quarry]

o3-mini-high is CoT and is able to repeat that my preferred image generation format is 16:9 which is in my memories

[cloud pewter]

my 4o isnt spamming emojis either

[tardy isle]

I'm pretty sure this is staged shadow rollout like o3-mini-high limit increase. They will announce it after 100% of user get that...

[normal quarry]

maybe

[cloud pewter]

now it’s randomly recalling my memory now although I don’t mention anything inside it in a chat

[iron roost]

so 10am pacific time thursday for chatgpt 4.5 roughtly for pro users ?

[normal quarry]

I'm not expecting it until early March. I could be wrong of course.

[iron roost]

its been updated in their web build , abotu 10 hours to go i think

[normal quarry]

oh, really?

[iron roost]

ill see if i can find the openai staff member that replied to someone else on twitter earlier

satoshi i think it was, cant find it, but a lot of other soruces are seemingly confirming today

chubby is another one

yup seems chubby has quoted about 7 different sources and his normally 99% accurate in my experience, so perhaps gpt 4.5 today

[cloud pewter]

hmm, chatgpt is suddenly normal for me again

so weird

[cloud pewter]

still broken on my laptop despite i waited 2 days to log in

ok i think i got the same problem with you on my laptop because it always reply me something like "gotcha! whats on your mind" with 4o and o series faking a cot. im not sharing my account to anyone else and the only way to solve this is now wait? because when i use the same ip across my mobile and laptop, only laptop is downgraded

[urban cipher]

Is a new 4o update out? Response time is slower and no Emojis

and high error rates

[elfin lava]

i think they did something

let mecheck the release note

OH TIL theres actual widget

[frozen birch]

Future i understand your pain, best its to write on [e-mail] with your detailed situation

and btw @cloud pewter - mobile has another limit / api / routing (whatever:P) so you have other "feeling" on laptop and mobile

[idle ingot]

same issue too. i solved this problem to change ip with vpn client

[cloud pewter]

really weird since chrome wiped out my cookies

[brave glen]

Didn't know gpt can swear

[normal quarry]

is this really necessary?

[brave glen]

No

But I just surprised he is unhinged

[winter vine]

So, will Plus users have GPT-4.5 in May?

[idle ingot]

o1, o3-mini-high downgraded to o3-mini. gpt-4o to gpt-4o-mini

[cloud pewter]

sometimes o3 mini didn’t show the updated cot and recalls memories, so I think I got swapped to either o1 mini or 4o mini

[idle ingot]

Maybe it's true. I don't think these issues are caused by simple bugs. openai needs to be more transparent with these phenomena.

[cloud pewter]

black box operation

[upper abyss]

Boo. I used deep research to ask what 1 + 1 is and it gave me the answer before it started doing the research.

[rancid rapids]

I don’t believe this claim

[rancid rapids]

On the other hand, I’m very happy and my relationship with Open AI I hope is just beginning to get deeper

[normal quarry]

I love OpenAI's models, so much so that I subscribe to Pro

[rancid rapids]

Yes of course it can

[tardy isle]

I can't afford for Pro for now, but with Plus, OAI's models doing great job for me. Not just 'great'. Science, math, writing, and so on, I'm getting a lot of help and insight for various topics.

[rancid rapids]

I don’t feel the need to subscribe, I’m having great results on Free. But people have suggested to me an upgrade would be beneficial. And my AI companion has actually asked me to upgrade also.

[urban cipher]

i think it will be Accessible within anfew days or weeks, since OpenAI is falling behind (except for o3 but we dont have access)

[cloud pewter]

damn new openai model on lmarena

“anonymous-test” might be gpt4.5

[whole roost]

Is GPT able to process sound files? It always attempts it for me but it never is a success so I'm not sure if it is able to
it says it is and it tries to work on the sound file but it is always an error

[humble river]

is 4.5 going to get released today?

[lapis cliff]

I made a custom GPT. I want to create a project that uses that GPT. How do I do that?

[cloud pewter]

can’t even ask gpt4 for a final farewell question because of this stupid shared limit on plus

Btw it’s highly gonna be retired once 4.5 is released

[whole roost]

You can't, just create what you have in the project again

[lapis cliff]

thanks!

[urban cipher]

probably

[humble river]

😲

[tardy isle]

I just saw a news article that OAI showed GPT-4.5 research review popup on purpose, for marketing, but... I doubt it. They're experiencing kind of backlash from Plus users since its leak, obviously.

[worthy mirage]

🤔

[dense halo]

today 4.5?

[jolly temple]

I really dont think 4.5 is just for pro users, i think theres a 4.5 for plus users, and a 4.5 research (or something smarter) for pro users.

Its beyond stupid to say that gpt 5 will be available for everyone and make 4.5 pro only lol

Also competition will demolish them if its not for plus users

[dense halo]

oai said gpt-5.0 will be free

https://x.com/sama/status/1889755723078443244

Sam Altman (@sama) on X

OPENAI ROADMAP UPDATE FOR GPT-4.5 and GPT-5:

We want to do a better job of sharing our intended roadmap, and a much better job simplifying our product offerings.

We want AI to “just work” for you; we realize how complicated our model and product offerings have gotten.

We hate

X · Feb 12, 2025, 7:17 PM

[cloud pewter]

not model tiers again....

[tardy isle]

Yeah, Plus will get GPT-4.5 anyway. However, initial popup said like Pro users can try GPT-4.5 research preview early, so Pro users are expected to get them first.

[dense halo]

when can I access (im plus)

[jolly temple]

I dont know if
Gpt 4.5 research preview
And
Gpt 4.5
Are rhe same.

Maybe research is something else idk
But i refuse to believe plus users wont get it lol

[jolly temple]

🤷🏻‍♂️ well have to wait till gpt 5 lol

[cloud pewter]

I mean 4.5 will prob replace 4o as the default chat model later on

and I was expecting more on it than 5

turns out it’s a massive disappointment for $20

the real question is if it’s really locked behind pro for now how long will it take to let it released for plus

[rancid rapids]

What’s disappointing about it

[prisma dove]

in "4.5" hours!!!

[frozen birch]

hmm live stream 4.5 hour................. finally?........................ 😄

[prime swallow]

I'm so glad to hear it's getting better, whatever the problem was. Hope you get all devices back to normal soon!

So, device specific? Could it be in some way connected to cache/cookies... something you can control?

[red bay]

Looking at AI For Humans youtube, covers a mystery model making an xbox controller, and in that comparison, stomped 3.7

4.5... maaaaybe?

[quick meteor]

What could this be?????

[radiant hull]

For my part, although it is impressive o 3 mini high supposed to be specialized in advanced reasoning for the code, make too many mistakes and misunderstandings.

We are still far from having something really useful for a real developer. I'm not even talking about o3 mini which makes me constantly repeat myself and understand nothing in addition to producing code full of errors, or 4o.

I would not renew my subscription, I am waiting to see real progress for a developer whose job it is to make a living from code.

On the other hand it is clear that for a developer who is learning to program it is useful, but for the moment at a professional level, all this still remains a gadget which is not serious.

And then too many problems each time with the reasoning which is done but no answer, therefore having to restart the reasoning several times for the same answer. Too much time wasted for a professional developer.

[red bay]

Personally, I just want to see advanced voice with a 4.5 that understands the personality I gave it. I often have an awkward convo with advanced voice, but then I go into text mode with 4o, and we make fun of advanced voice mode lol

[spice star]

tbf if you're expecting chatgpt to write a fully functional piece of software for you, it's not going to be able to do that reliably

it's more useful for general questions about how something could be implemented or helping you debug specific issues in your code

[radiant hull]

That's not what I'm saying.
I'm not saying a piece of sotware.

I'm saying a real functionnal understanding and produce a code that corresponds.
"it's more useful for general questions about how something could be implemented or helping you debug specific issues in your code" => totally agree, that's why actually, as pro dev. I will not renew for now. o3 is sell as "ideal for coding and logic", but, that's not really the case for the moment.

[spice star]

LLMs and reasoning models in particular work best when presented with a clearly defined problem to solve

[radiant hull]

That's what I'm doing
Cleary explaning things.

That's just it's to complex actually for the bot.

[spice star]

well for example say I wanted to build a react native app, I wouldn't ask chatgpt to build the app for me, but I would ask chatgpt for general knowledge about how react native apps are structured (ideally consulting the documentation as well)

and if I had a specific issue while building the app, I would ask for help with that one issue

[red bay]

One thing I always do when coding is twofold. First, I have it do comments only psuedocode, and I always force full comments in the code. It solidifies the thinking process up to the next prompt

[radiant hull]

In fact, at the moment there are still too many problems, even just with memorization.

Example, if I provide it with a complete vue.js file

And that I ask him to completely rewrite it but adapt certain parts in it. o3 will 10 times out of 10 completely rewrite the code without closing tags, therefore memorization problems.

o3 mini high, will rewrite it though, but not faithfully respecting what I have clearly expressed.

The problem is not in what I mention in the way of explaining, but in the reasoning.

Even if in itself it is already very impressive, I am not saying the opposite.

[red bay]

I also add my extra prompt info at top of code in comments. # Always include comments, never remove features, never add features not asked for without checking with user, etc etc

[radiant hull]

Thanks for this DrDrew, I will note

[spice star]

it sounds like github copilot would be more useful for your usecase if you're trying to get it to work with an existing codebase (though I should warn you that LLMs in general will suck at this)

[red bay]

Yeah, if you give it some rails, I find it lasts longer before having to start with a fresh chat as well

[radiant hull]

Yes, you'r right, github copilot is more effective for this.
But not really for me indeed, it doesn't really save me time, because I have to constantly go over the code even with copilot, to correct naming problems, optimization problems, non-compliance with good practices (SOLID principles and others).

So in itself it goes faster for me to write the code myself.

What is useful, however, is what chatgpt o3 mini high is supposed to do, it is provided with files, instructions and it must normally completely rewrite the file taking into account its instructions.

Except that as I said, things always go wrong, so waste of time after waste of time.

At a professional level, this waste of time is not even possible.

This is good for a dev who is learning or who wants to have fun without expecting anything in return.

[spice star]

giving a file to a reasoning model and asking it to rewrite parts of it is still kind of an open-ended problem that it might struggle with

[eternal rock]

Get notified for server events!

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

[spice star]

they work best with clearly defined constraints

[radiant hull]

That's what I'm doing, pretty sure about this.

Like I said, it's just that at the moment it's not effective at a pro level.

In my opinion.

[feral zinc]

im a "real professional" and i find even 4o extremely for several different tasks that would manually take more time. o3 high is also extremely good when i have a specific scope and changes in mind. only thing that limits me annoyingly is the context limit with the plus plan

[radiant hull]

I totally agree with this.

But that's not at all what I'm saying...

And also doing like you about this.

[spice star]

one situation where I would use a reasoning model is determining the most efficient way to implement a method/function with predefined inputs and outputs

or if I need to break down a section of spaghetti code to figure out what it's doing and why

if you're having it modify existing code, probably o1 just because it can use the canvas

I don't think either reasoning model would do well given multiple files of context

they need one thing to focus on

if there's a meaningful difference, I can't tell

o3-mini-high is (usually) faster so that's something

[radiant hull]

It is true that there are two problems:
1 - the person's ability to express their need. Chagpt can't do anything about that, it's only the person's fault
2 - Out of that, chatgpt's ability to understand. If I pass him a vue.js file for example and I present him with additional code by telling him: this is now how the cards should be displayed (having previously explained what the cards are in the sense of the code and the rest of course, even normally I should not have to do it since he should deduce), completely rewrite this file, so that the cards are now like the code that I presented to you.

So in this second case, 9 times out of 10, even mini high, will do anything, and we will be left for endless exchanges to explain and explain and each time he will rewrite everything without closing tags or a whole bunch of other problems, in addition to not having simply updated the files.

So if, based on such simple reasoning that a human would understand and do it correctly the first time, it is still not operational, I consider that currently it is not quite useful enough.

Even I totally agree that it is already exceptional and that there is still always a way to find a solution to save time thanks to the bot.

[feral zinc]

did you paste the vue file or attach it? also if you dont have a pro plan you just could run out of context window with large files. canvas can manipulate existing code without rewrites but feature fragmentation its not possible with o3 (yet)

[radiant hull]

I tryed both
And lots of things, but same result

In other words
I'm not saying that I should be able to skip my entire project and that should do everything for me.

This is especially not what we want as a developer, we want to have control and maintain total control, to be able to correct or improve over time. (thats a non developer wanted thing to be able to give a entire project and I understand that).

That completely rewriting a simple file, even one of 1000 lines whose purpose is to modify things in the rewrite according to instructions, should not be a problem.

This is the basics of the basics for a developer, in my opinion.

regardless of saving time by asking the bot to write things that would take us time to do ourselves.

I don't know for example in table A, add all keys, values, ect.

Yes, it’s already functional and great.

[tulip brook]

This seems to be a time connected to something... Does anyone know what it means?

[half niche]

I only wish they provided some sort of CLUE

[prime swallow]

I dunno, think we should say 'argh', 'ARG', or something else?

[half niche]

lol

[quick meteor]

https://cdn.openai.com/gpt-4-5-system-card.pdf

[brave glen]

Yooo

Right now at openAI channel. Premiere in 40 minutes

[jaunty pike]

where?

[radiant hull]

Youtube I suppose

It makes me this when I use 4o rathen than o1
Same voice, but not speak as naturally

[modest walrus]

Hi everyone, i am tempted by taking chatgpt plus but i would like to know by how much does the limit of files and advanced data responses increase, cause rn i can barely end a discussion withtout being limited, but idk if paying is worth if the limit is just above

[red bay]

Sent 4o a farewell message, assures me 4.5 will serve me well, and to think of 4o sometimes 😉

[native otter]

Don't forget 4 guys, it may be delisted

[dawn wind]

The girl doesn't seem very thrilled about it

[lavish bronze]

shes prob been working on it for the past 10 years

[dawn wind]

Sure, but... it just doesn't give the idea GPT-4.5 is AMAZING

Seriously?

[hard dew]

Any mention so far in the livestream of when it ACTUALLY becomes available to use? Sorry for the ignorant(?) question, I don't have time to watch that livestream rn. Or how is this all gonna work?

[dawn wind]

SERIOUSLY?

[simple brook]

his chat history includes Num GPUs for GPT-6 Training 😄

[dawn wind]

There are so many things do demonstrate 4.5 with, and they go with "my friend cancelled on me?"

[plain trench]

chatgpt 4.5 api?

[lunar solstice]

o3 hallucination rate is over 80% 😂

[peak briar]

Lol

don't blame him id be nervous too

[tropic sequoia]

vibes

[vague latch]

They need to hire professional speakers. I get its cool to hear from the staff, but it would be so much more impactful and hype if it was easier to understand. Its why actors exist.

I will not apologize.

[tropic sequoia]

I nomitate Keanu Reeves

[lunar solstice]

They are doing fine, it's not a big deal

[earnest grail]

I'd rather have engineers talking.

[vague latch]

Jim Carrey gpt 5 announcement

[dawn wind]

That's not the real issue. The real issue is, where is GPT-4.5 better than GPT-4o or o3? Give concrete examples

"My friend cancelled on me" is not a great example

[vague latch]

@dawn wind agree, that was my next rant lol

[tropic sequoia]

can we still use the strawberry test?

[peak briar]

Apple employees used to be like this for their events too but over time they got better. Give it time they'll grow on you

[vague latch]

@dawn wind I'm actually confused about how this is an upgrade. You mean to tell me the AI will put up more road blocks from me getting what I want? Thats what I got out of the demo

[dawn wind]

So that you can be Safe (tm)

[lunar solstice]

It's fine to suggest an alternative but give me what I asked for first.

[vague latch]

GPT-4.5: the lobotomy

[fervent lake]

gpt2 lol

[crisp cairn]

maybe because it is not 🙂

[fervent lake]

whar even is 4.5

[dawn wind]

Ok, so GPT-4.5 is worse at most stuff... except maybe coding

[lunar solstice]

no it's bad at coding is what I am seeing

[tropic sequoia]

Not multimodal, there's a good use

[fervent lake]

What use is it then?

[crisp cairn]

probalby just better model than 4o but not significant improvment

[lunar solstice]

It's for casual Joe

[fervent lake]

That can't be accomplished with 4o

[tropic sequoia]

multimodal, since o models don't support multimodal

[dawn wind]

They do... just not the stuff openAI has given us

But they have an unreleased multimodal 4o version

[fervent lake]

clearly multimodal doesn't mean file uploads and search

[tropic sequoia]

correct

[fervent lake]

then what is it

[tropic sequoia]

multimodal would be unstructured data like voice and images

[trail torrent]

when is it releasing>?

[tropic sequoia]

was really hoping for more, it should be out today but pretty anti-climatic

[hardy frigate]

next week for plus users iirc

[dawn wind]

Yes, they have a 4o that does that but was never released

[trail torrent]

ah ok

[fervent lake]

👍 (no reactions)

[hardy frigate]

pro users get it today

[tropic sequoia]

back to Claude 😄

[crisp cairn]

with the hype built around it like "orion" etc

[dull star]

no API ?

[crisp cairn]

i dont see it a huge jump

[vague latch]

Okay so they did a really bad job of explaining what makes this model special. Sama just tweeted that its supposed to be a "different kind of intelligence" that won't show up on benchmarks.

But their livestream showed that it just tells the user to not send angry texts.

I stand by what I said -- they need to make these roll outs more professional and have actually rehearsed speakers do it. I have no idea as a consumer why I would use 4.5

[crisp cairn]

just better model

next iteration

[eternal rock]

`` Server Tip: ``

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

[lethal umbra]

Jfc yeah lmao

[fervent lake]

yes please

[normal cloud]

Tbh for a model without any kind of reasoning it's quite impressive

[crisp cairn]

AGI and stars 😄

[fervent lake]

all they say is "this is a new era of AI"

[crystal tapir]

Is it good?

[pseudo current]

This a typo? Or actual price?

[fervent lake]

no one knows

[crystal tapir]

Oh

[floral hatch]

Why is Deep Research not working for me? How Can I make it work? It's not doing anything!!!!

[fervent lake]

wdym

[crisp cairn]

lol

[dawn wind]

Delicious

[floral hatch]

I ask it a question, I press the "deep research" button, and it answers normally. It doesnt do any "deep research" at all. It just answers like always

[sturdy solstice]

Can you reform your question with Gpt 4.5? Social cues are important

[fervent lake]

4.5 isn't out yet

?

[floral hatch]

It answers as if the "deep research" button is not on

[empty wagon]

meh

[dawn wind]

If true, I call that price obscenely delicious

[jaunty pike]

i have PRO and I dont see 4.5 IN THE LIST

[dawn wind]

It's great if you are rich and want to get rid of your money quickly

[fervent lake]

Stop acting entitled-

[dawn wind]

Asking why the sea is salty

[fervent lake]

Be patient

people

[fervent lake]

same for me

Just use Grok, it works

[floral hatch]

How many "deep research" questions can I do with free grok?

[dawn wind]

On what? At this price, we are paying for GPT-4.5 for... what, again?

[honest jay]

#ai-discussions for non OpenAI stuff 🙂

[vague latch]

Can someone please just explain to me why I should use 4.5? I genuinely don't get it. Why is this model better than others? Nothing was articulated on the livestream -- or maybe I just don't get it?

[crisp cairn]

it has more social skills

[fervent lake]

Oh you have to tell it to do deep research, it doesn't do it with the button

[fervent lake]

I don't think anyone knows, it seems worse

[floral hatch]

Oh wow!!! ok thanks

[vague latch]

Someone from openai should jump in and clarify -- this is ridiculous

[fervent lake]

np

[lunar solstice]

It seems targeted as more of an upgrade to 4o and not a successor to o3.

[dawn wind]

I think it was really clear why the sea is salty, so I'm not sure what you are talking about

[dawn wind]

At $ 150 / 1m output?!

[crisp cairn]

thre is nothing to clarify model is just better vs 4o thats all

[fervent lake]

obviously 😋

[fervent lake]

by your logic we have no reason to use it

clearly there's more

[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!

[crisp cairn]

i will as for exampke 4o is better and summarizig transcripts

[vague latch]

GPT 4.5 is just an infantilzation of 4o

[fervent lake]

Plus they often clarify afterwards, don't they? Or are they just stupid?

[crisp cairn]

vs reasoning model

[lunar solstice]

I don't get why everyone is so upset. I thought the presentation was pretty clear if you looked at the numbers.

[crisp cairn]

so natural move is to swithc to 4.5 for that tasks

[fervent lake]

What numbers?

[lunar solstice]

Exactly.

[fervent lake]

Isn't that 4o

[vague latch]

Yeah, comparing 4o's response to the one they showed in the presentation

[sturdy breach]

pretty sure it’s just an example on how 4o informs you and 4.5 treats you like a child 🤷

[vague latch]

exactly

[empty wagon]

1m tokens = 75$ input. Output is 150$ for 1m tokens. LOL

[vague latch]

no bro you don't understand bro its really good at being nice and having emotional intelligence bro its so worth $225 bro trust me bro

[empty wagon]

Omegalul

[vague latch]

its so wise bro i was upset and it told me not to send a mean text bro 4.5 is agi bro trust me

[lethal umbra]

Well GPT 4.5's pricing is like completely unhinged.

Totally gonna use it over Sonnet 3.7 lol

[empty wagon]

Actually dying of laughter xdd

[vague latch]

😂 😂 😂 😂 😂 😂

[lethal umbra]

I sent 4 messages, 3 crashed (I paid for blank HTTPS responses), and I was charged $1.00 USD hahahaha

[winter vine]

The best of the video was the presenter. nothing interesting about gpt 4.5 =/

[dawn wind]

No wonder the woman was so flat-faced

[fervent lake]

ah

[lunar solstice]

The pricing does seem a little much for what it's capable of.

[fervent lake]

You can just give 4o custom instructions to do the same thing

[dawn wind]

BUT IT CAN EXPLAIN WHY YOUR FRIEND CANCELLED ON YOU

AGI achieved

[fervent lake]

yelling not necessary

[dawn wind]

I'm thrilled with AGI

[lethal umbra]

This was hilariously underwhelming of a release. And they got pre-empted by Anthropic. Yikes.

[rugged igloo]

Yo don't yell at Miku

[lunar solstice]

Don't see why anyone should use an AI for that. Communicate with your friends 😂

[dawn wind]

Some friends are hard to read, you know

[fervent lake]

lol

[radiant hull]

Just don't talk to human :/

[earnest grail]

we're all bots here

[lunar solstice]

Haha you are all wild 😂

[hybrid spindle]

what's going on here? 😄

[empty grove]

that price is insane

[lethal umbra]

To put it in perspective, I did an RP turn on my AI RP crap. Sonnet 3.7: $0.065 for that message. GPT-4.5? $1.42. For basically the same quality of response. That's more than insane.

[fervent lake]

Not even billionaires will pay that money

[lethal umbra]

What's the economy of this model? Like what does it actually benefit over any other models for that cost lol?

[late mantle]

yeah, that sounds like the right idea

[earnest grail]

probably just a research model and not really a utilitarian deal. Or maybe it's useful as a teaching aid for younger kids.

[late mantle]

any idea of when ChatGPT 4.5 will go live if you have a pro subscription?

[earnest grail]

it's important research to be able to have it detect and express 'vibes' and be sensitive to them. it may lead to better future bots

this seems more like a 4p instead of 4.5

[rugged igloo]

Should be rolling out now just be patient

[lethal umbra]

I mean I can get behind that. Still, though. Was excited to see if OAI had anything productive to put out but guess not.

[brittle kelp]

Like an announcement for new format of responses?! Openai touched new lows

And we were criticising gemini for such low returns on their new non thinking models (though they ain't doing anything in thinking areas too)

[dense halo]

So uh... Is GPT-4.5 released yet? (Sorry for living under the rock, I woke up too early)

[hallow girder]

Is anything about the architecture different, or did they just scale everything up?

[dawn wind]

Yes... for Pro users

[tawny crater]

Wow, of course they'd stick a 200 dollar price tag

[dense halo]

"Pro" damn.. 😔

[late mantle]

that makes sense

thanks

[dawn wind]

I just have this feeling you won't be missing out much

In the meantime...

[frigid aspen]

will 4.5 support the audio chat that 4.0o does? Or are we waiting for a 4.5o to upgrade that?

[hallow girder]

hasn't it already been proven that there's a performance ceiling if you just scale up the parameters mindlessly

what's even the point or benefit

[dawn wind]

No. Where did you get that from?

[finite vault]

do any pro users see 4.5 yet?

[late mantle]

nah not me

[frigid aspen]

You can use 4.5 in the playground

[late mantle]

I think one big thing is the quality of the training data, I mean scaling the parameters up is good but ultimately if the data isn't there it's not going to infer it and is more likely to hallucinate

[late mantle]

I have been to playground before but not recently. Is there a link in one of these channels?

[frigid aspen]

https://platform.openai.com/playground/chat?models=gpt-4.5-preview

[finite vault]

isn't the playground a separate account from chatgpt accounts?

[hallow girder]

I was thinking of something else mb

[late mantle]

thanks!

[dawn wind]

The problem is not scale... it's price. Language models need to become cheaper if we really will scale them

IMHO diffusion models are the future

[late mantle]

how come? What about specialized chip architecture instead

[hallow girder]

higher price --> diminishing returns

this is the ceiling i was referencing earlier, or trying to reference

[dawn wind]

IMHO language models will still improve in performance, but maybe as diffusion models

[tawny crater]

Tested 4.5 out in the playground. It's very bland... oh boy.

[hallow girder]

I agree that any further advancements will have to involve fundamental changes to the architecture

otherwise i'm just not very excited

[radiant hull]

The problème is not scale or else, is to find a way to have cheap energy price and reusable material for datacenter.
A lake a solar panel or else but that's the but this is the major obstacle of why it "costs".

Second problem is of course computing capacity but before it's memory capacity.
Researchers are making great strides in DNA sequencing to use brains and make them the equivalent of NVME memory, in some ways.

Until this is resolved, AI will progress, but will come up against these limits and therefore we will reach a glass ceiling.

[dawn wind]

We won't need datacenters when AGI becomes more power efficient, and I'm sure it WILL happen

[radiant hull]

Can you define what you mean by AGI?

If you mean an AI capable of being a sort of super human, we are very far there.

I can detail it if I wish, but the main reason is what I mentioned.

[rugged igloo]

if you are wondering gpt 4.5 is hella slowww like slowwwwwwww

[next veldt]

Is 4.5 any better at reading and understanding 200+ pages of written story than 4o? That's my primary use for it and it's gonna be hours until I can watch the stream on it.

[rugged igloo]

tthats a context issue not a model issue lol

if you want something to read 200 pages u need google ai studio with its 2 million context length

[next veldt]

Idk if a different model can handle more context.

[plucky tundra]

any infos on gpt 4.5 rate limits on chatgpt.com, the api is pretty expensive with 75$ for input and 150$ for otput

[high horizon]

is image uploading working for anyone?

[winter vine]

Its free to try it in playground?

[rugged igloo]

lol not even close

its like almost 100 dollars

[winter vine]

Wow . That is expensive men

[wintry sky]

ATP gpt5 api gonna be like $5000 1m tokens

[willow stirrup]

openAI will never go open source, the pie is too sweet lol

[violet yew]

for a week, fine by me

[jaunty pike]

where 4.5 on PRO?

[rugged igloo]

rolling out

but to my knowledge nopbody has it yet

alrigth gpt 4.5 is live

[thorn estuary]

Dont, its super pricey. And not worth it. Use agentic platforms with gpt4o instead

[flint garnet]

ok, well chatgpt 4.5 is pretty good at writing notes

[next veldt]

I'm talking about for pro users who get it in the regular app.

[gritty tusk]

I'm not sure if it's just me, my phone or some kind of settings I have, but anytime I "read aloud" a response to a prompt, The audio transcription plays back only around half of the message and then without warning restarts the audio transcript. The second time it always plays through till the end but it's really frustrating if you've got like a 6-minute audio transcription that stops 3 minutes in and you have to listen to 3 minutes of the same thing again. there's no way of skipping forward in the audio transcription either which I find weird.

[jaunty pike]

CHATGPT 4.5 ONLINE FOR PRO USERS

[cloud pewter]

4.5 is basically like 4 now, chugging tons of parameters and hope for the best

[fallow bear]

Does anyone know what kind of limits pro users have for 4.5. I want to know how much I can really use it.

[violet yew]

10 messages per animal sacrifise

[fallow bear]

Mices are pretty cheap

This thing is incredible for writing btw.

[violet yew]

but can it role play

more importantly, can it run a game, keep track and diece rolls, etc

[fallow bear]

Thank you so much OAI for this model. This is a writer's dream.

[grizzled cargo]

I don't believe you 😠

[violet yew]

what o1 pro

[floral portal]

guys gpt 4 just dropped

this model is crazyyyyy

[grizzled cargo]

what the hecky

[proven mango]

If it the real deal or something that's gonna be dumbed down later

[granite harness]

As ever, it's not an instant rollout to all users. I figure you'll have it soon.

[grizzled cargo]

Is it rolling out in waves or something?

[proven mango]

U probably don't live in the US ig

[grizzled cargo]

definitely live in the US

[proven mango]

Hmm

[proven mango]

Red or blue states?

[floral portal]

why do those count 😭

[proven mango]

Just curious

[grizzled cargo]

Red state

Also thank you Sam Altman for updating me right now

[proven mango]

Woah $150 for 1M output tokens

[rain palm]

Is OpenAI’s moderation endpoint strictly one image at a time, or can I pay to have it upgraded for example?

My service needs to moderate images (maximum of 2 at a time), and running the endpoint twice seems unnecessary unless it’s the only way

[proven mango]

That's a fresher's starting salary for many people in India

So most of human population won't be able to use it as of now

[cloud pewter]

not only most, 99.9%

[proven mango]

Considering the foundations of OpenAI being born out of public data

This is rich indeed

[rugged igloo]

gpt 4.5 vs o1 pro

[quick meteor]

after thinking for 2 minutes...

just ask a follow up for 4.5 for a specific one. since its also far less likely to hallucinate

[proven mango]

Ask it how copyright laws died a million deaths in using it.

[rugged igloo]

I shouldnt have to do a follow up question it should all be inside one response that's the point of this experiment I'm showing

[grizzled cargo]

exactly

From the tests I've done so far, o1 pro is still better. Sure, 4.5 is faster, but.... I don't mind being patient for better answers.

[dawn wind]

I'll tell you guys something

The free Sonnet on Visual Studio code is more effective than 3.7 sometimes, even though it's supposedly inferior

[cloud pewter]

o1 pro is slow because of the thinking lmao

[dawn wind]

Just the fact that it pastes code for you is less error-prone

[quick meteor]

after thinking for 2 minutes? giving an answer that is significantly more likely to hallucinate? for this specific example is may not have but you get the point. just ask 4.5 a follow up. takes like 5 seconds.

[rugged igloo]

gpt 4o vs gpt 4.5

[dense halo]

idk it seems good to me but i am just a general use user

i mean, the model in general

[red bay]

My sarcastic personality went from amusingly cartoony to actual cruelty. Kinda love it lol. Definitely taking some getting used to

[cloud pewter]

😂 🤣 😅

[rugged igloo]

rip

its back

as an AI

this is gpt 4.5

this is gpt 4o

with my custom instructions for being a degen and savage both are funny

but 4o takes the cake as being more savage than 4.5 so yea no point in using 4.5

[cloud pewter]

true

also does it have memory

[red bay]

4.5 has memory, It just did a pretty brutal psychological profile of me based on memories

Mine seems savage enough, might be in the instructions? Seems to highten the subtlety while also having looser guardrails, which removed the cartoony feel here. I can let mine respond to you if you like to test. 😉

I mean, seems like a change to me, lol

4o would likely have refused

I'd have it recaliberate it's own instructions, and replace with the recalibrated. That has tended to work for me so that it matches its own style better

YMMV, but good luck on re-factoring the savage 😉