chatgpt-discussions

[tawdry turret]

Thanks, I wish they had not done what ever they have done to nerf the hell out of the image generation process

[normal quarry]

what is your prompt?

[tawdry turret]

Part 1 Generate another image of her Can you generate an imamge of her wearing the outfit sticking to the anthropomorphic (furry) art with a stylized, semi-toon aesthetic of the image
part 2: An anthropomorphic (furry) art with a stylized, semi-toon aesthetic of Cinnamon Bun, a warm and charming rabbit character, sitting on a chair in a relaxed yet confident pose. She has detailed fur texture in golden-tan with cream accents on her chest and muzzle, and long expressive ears tipped in deep brown with soft pink interiors. Her striking green eyes are full of warmth and energy, with subtle freckles dusting her cheeks and muzzle, adding to her youthful and friendly appearance. Her short, layered pixie-cut hair is warm brown, slightly tousled, with a few strands framing her face for an effortlessly approachable look. She has a soft yet well-toned figure with a feminine physique, including a fuller chest and natural curves.

New Outfit: Neon Green Latex Dress
Cinnamon’s usual café uniform has been replaced with a striking, bold fashion statement—a neon green latex halter dress that clings to her form with a glossy, futuristic sheen. The outfit is eye-catching, reflecting light with every subtle movement, making her stand out effortlessly.

Design & Material
Material: The dress is crafted from high-shine latex, hugging her curves while giving off a sleek, cyberpunk aesthetic.
Color: A bold, neon green that practically glows, complementing the warmth of her cinnamon-colored fur.
Texture: The glossy surface accentuates the smooth lines of the dress, enhancing its daring and fashion-forward appeal.
Top Half
Neckline: A halter-style high-neck collar, fastened at the top with a small metallic ring, adding an edgy touch.
Back Closure: The dress zips up the back, maintaining a seamless, snug fit against her toned figure.
Fit: The bodice contours her shape, enhancing her silhouette while allowing for some movement.
Skirt & Bottom Half
Skirt Style: The flared, pleated mini-skirt flows outward, adding a playful contrast to the tight-fitting bodice.
Length: The hem falls mid-thigh, revealing just enough leg to make a statement without straying too far into risqué territory.
Structure: Thanks to the firm nature of latex, the pleats hold their shape, maintaining a dynamic, slightly structured look.
Back Details
Open-Back Style: The upper back remains exposed, showing off her toned shoulders and emphasizing the sleek design.
Zipper Closure: Running neatly down the center, the back zipper adds practicality to the daring outfit.
Overall Aesthetic
Cinnamon’s usual warm, casual charm takes on a new, high-fashion, club-ready edge in this daring ensemble. The bold color and glossy latex material make for an outfit that feels both playful and sultry, a striking contrast to her previously uniformed look. Whether she’s stepping into a neon-lit nightclub or simply indulging in an adventurous fashion choice, Cinnamon commands attention with this head-turning, futuristic style.

Sitting confidently in this bold latex dress, she radiates a mix of charm, confidence, and edgy fashion-forward energy. The semi-realistic digital illustration style captures fine details in her fur, clothing textures, and natural shading, enhancing the depth and realism of the artwork while maintaining a distinctly furry-inspired aesthetic.

Creating an image for an AI dungeon story lol got this gem

Now it hates me

[normal quarry]

I was unable to generate the image due to issues with the request. If you’d like, I can adjust the description or try a different approach. Let me know how you’d like to proceed!

Content policy, too furry

[tawdry turret]

XD lol too furry for chat gpt

[tardy isle]

"issues in the request" after long delay. Guess it's image filter.

[tawdry turret]

God they need to dial it back a bit lol that aint even a lewd image, jesus did some one make it ultra catholic

[tardy isle]

Typically, if it says 'there's a problem', then it's server issue. However 'issues with the request' means it's prompt filter or image filter.

[normal quarry]

are you on a Mac or PC?

[tawdry turret]

Ive resorted to violence (Not actualy going to jailbreak because lol to dumb and against policy)

[tawdry turret]

Pc running win 8.1 using chrome

[tardy isle]

I asked it to revise prompt and try again, then it tells me this:

I wasn't able to generate the image because the request didn't follow our content policy. If you'd like, I can help adjust the prompt to better fit the guidelines while keeping the essence of your vision. Let me know how you'd like to proceed!

[tawdry turret]

Lol ok so I guess its time to search for an alternitive ai image generator

[tardy isle]

Btw, is furry against content policy? I once got this hallucination: 'generating image of dragon is against content policy'

[normal quarry]

I tried to adjust the prompt and got

I was unable to generate the image due to content guideline issues. If you’d like, I can further refine the prompt to better fit within the guidelines. Let me know how you’d like to adjust it!

[tawdry turret]

Time to ajust the outfit probs or hope this is it just tripping balls not even trying to invoke The True Story of How GPT-2 Became Maximally Lewd

[tardy isle]

Slow way: wait for 'grown-up mode'
Fast way: use alternative tools. There're a lot of unrestricted AI writer and image generator out there.

[brittle steppe]

hello I was using search online and it failed and I got this error later

[tawdry turret]

I doubt there ever will be a adult mode lol they are all to worried about the corperate image

[tawdry turret]

Stack overflot go brrrrrrrr?

[brittle steppe]

looks like sone javascript error

[tardy isle]

It's first time to me seeing that kind of error in ChatGPT...

[normal quarry]

same, never seen an error like that before

[tawdry turret]

Ive seen the ai flip out over some inocent as hell stuff some how I think the morals coach or what ever the element of RLFH is cranked to high

[normal quarry]

check my message to you in #ai-discussions

[urban pike]

Hello

[cloud pewter]

i tried to let it dont miss a paragraph as possible in the prompt and it turned out decent, just need to check manually

+5000 more words in the translation which is good

[tawdry turret]

Dont know if I did say thanks to you but thanks for your help anway lol hopefully I will get sucess and get some good images free of braindead gpt

[tardy isle]

In model selector when?

[tawdry turret]

Ok lets try this again I regret nothing I turned the Our systems have detected unusual activity from your system. Please try again later. message in to a ha ha funny amongus copy pasta 🛑 Emergency Meeting! 🛑

Hey crewmate, our security systems just spotted some seriously sus activity coming from your system—like an impostor venting in electrical! Before you get ejected into space, please stick around and wait it out. Remember, trust no one... especially if they're acting all shady. Stay safe and try again later!

Is it really you, or is someone impostin’?

Please dont beat me again auto mod, ill be good I swear ... XD

[normal quarry]

I just coded this Bitcoin price tracker app for macOS using o1

[nimble torrent]

Does it use an API?

[normal quarry]

yes, CoinGecko

[nimble torrent]

one shot?

[normal quarry]

yep, compiled and ran first try

[nimble torrent]

A user here was generous and I asked them to send my prompt to o1 pro

I thought that it had a recent knowledge cut off

As a result o1 pro failed, it was the worst response I have ever seen

[normal quarry]

what was the prompt about?

[nimble torrent]

It was about auth implementation in Svelte 5

[normal quarry]

I wonder whether o3-mini-high could do it?

[nimble torrent]

No, but the responses from o3-mini are better

Svelte 5 was released in October 2024

No model could answer this simple question 100% correctly

[normal quarry]

what about Grok 3?

[nimble torrent]

I like this example by the way. AI is so useful and can replace need in some apps, and you can create solutions tailored to you.

[nimble torrent]

No, but it is in top 3, when deep search is used.

[normal quarry]

it's still filling in the graph, every minute

[nimble torrent]

And yes no model could do this even with tools (web search, file upload)

[normal quarry]

are you on a Mac? because you can connect the macOS Desktop App to xcode

[nimble torrent]

I prefer using websites

[normal quarry]

I only use the web for Operator, otherwise I prefer the macOS Desktop App

[nimble torrent]

Anyways I usually do not use ChatGPT

[dawn mason]

For Deep Research, do I need to select the correct model first? It's showing it as an option for each model. If so, how do I know which model to choose?

[nimble torrent]

Whatever you like, Deep Research itself is powered by o3 (not available in the model picker)

[normal quarry]

it doesn't matter what is shown in the model picker, Deep Research uses it's own model based on o3

[dawn mason]

Thank you guys! Appreciate it!

[half vine]

Will be gpt4.5 available in the plus subscription? If yes, what would be the limit

[tardy isle]

It will be in next week. Limit is unknown currently.

OAI didn't even stated limit to pro...

[half vine]

And also why is gpt4o better at coding than o3 mini (not high) i thought that reasoning models are built for coding

[half vine]

Oh ok

[normal quarry]

my macOS Bitcoin Tracker as coded by o1 is still running along nicely

o1 is simply the best coding model I have used

o3-mini-high is pretty good too

[half vine]

Is there an official ai leaderboard on openai site?

So i can compare o3 mini against other Models

[nimble torrent]

No, they don't have their own leaderboard for their LLMs.

[nimble torrent]

Try lmarena. You can view the leaderboard there and even try different models to see whether they'd work for you.

[normal quarry]

50 messages per day

50 messages per week

150 messages per day

[winter elbow]

some guys at sesame cooked hard not going to lie

wish AVM had the same type of autonomy

[normal quarry]

Bitcoin Realtime Price app coded for macOS by o1

[visual nova]

couldnt any llm code that? that's like 1-5 queries depending on if it falls into pitfalls or not

o1 might be the best, but theres stuff some models cant do

[normal quarry]

o1 has the best knowledge of Swift in my experience followed by o3-mini-high

[hybrid sparrow]

why is 4o mentioned in Pro? Wouldnt that be inlcuded as "everything in plus" ? Unless soon 4o is ONLY available in Pro \

[honest jay]

Plus doesn't have unlimited access to 4o, just increased access to it (compared to Free). I think that line is specifically pointing out the type of access to those models on Pro

[visual nova]

oh yeah, some languages are harder than others. and i dont know much about mac or swift. i think any language ai doesnt deal with easily might go extinct in the next few years.

it took 30 minutes i guess, but i got stuck for 15-20 trying to setup an api key for coingecko

and i didnt do anything with historical data

[hybrid sparrow]

lol they couldve just said "to all models"

[visual nova]

but yeah idk about swift and macos, easy project overall

[honest jay]

Deep research is a model! And there's no unlimited access to it

[visual nova]

i used 4o btw

4o-mini i mean

[hybrid sparrow]

is it a model? or a system

[honest jay]

It's a fine-tuned version of o3 separate from the models available in the selector.

It indeed uses a search tool as well for research and citing

[normal quarry]

mine turns the price green on rising prices, blue on steady prices, red on falling prices. Speaks the price on the hour and when there is a drop or rise of more than 1%

[visual nova]

yeah, it could be made a lot more complicated. this kind of makes me want to get into crypto and have the ai decide when to buy and sell it lol

[hybrid sparrow]

kind of like 4o with tasks was a separate model even though it could've just been a "set tasks and add a time and date" as a prompt instead of making a separate model lol.. but i never used deep research yet so maybe they're not comparable

does anyone know when plus can try out 4.5?

i saw a video and this person was saying that it was more expensive and does worse than o3, completely ignoring the fact it does better at multilingual, actually has multimodality unlike o3 or o1 mini, and probably is just an overall better person lol

but i wanna see

[visual nova]

i thought o1 had vision

[normal quarry]

it does

[hybrid sparrow]

oh ok, corrected. bimodal

[visual nova]

whats the multimodality 4.5 has over o1?

[normal quarry]

4.5 is no more multimodal than o1

[hybrid sparrow]

thats debatable

[visual nova]

4.5 does seem smart, ive given it some complicated prompts that other models dont get, no other models get actually

i just think the in app experience is specifically flubbed

[normal quarry]

no, it's not debatable

[visual nova]

its performing like twice as well in the api

[hybrid sparrow]

Yes, it is

[visual nova]

hopefully they look at that going forward

[hybrid sparrow]

i just have no arguments or evidence yet

[visual nova]

i wouldnt be surprised if its better at vision, my tests are kind of weird. i have a prompt where it plays a game of among us and has to keep track of all of the characters and remember if they are impostors or crew

and whos been killed/voted out

4.5 is the only model that did it correctly from beginning to end of game, even claude 3.7 messed it up

[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.

[visual nova]

its really niche stuff like that where i notice its smarter. idk what that means for practical uses

[hybrid sparrow]

in general i just think that its ability to give its response directly to you instead of going through a whole chain of thought step by step breakdown of how to respond, its just more in the moment and honest. its allowed to make mistakes. thats how humans speak. something that talks to itself so much before responding to the user is giving the user a very deliberated answer , by that point is has all but forgotten the question. its not a natural response

[visual nova]

but thats not automatically 4o

[hybrid sparrow]

and if 4.5 is the Last model that will be able to do that, then 4.5 is the end of chatgpt

[visual nova]

id argue 4o has less latency and can just be prompted to whatever tone so it does it better than 4.5 because its faster

[normal quarry]

4.5 can't code very well compared to o1 or o3-mini-high

[visual nova]

im more interested in tangible things, like questions the ai actually gets wrong vs right

[hybrid sparrow]

yeah i was really hoping for 4o.5

[visual nova]

something feels broken with 4.5 code even in the api, it makes svgs really well

noticeably better than previous models. and thats code

claude 3.7 and 4.5 are both in their own weight class there

[hybrid sparrow]

if 4.5 is the last non-CoT then that's basically saying omni was a one time thing, and no future model will be able to be in the flow - just this sanitized response that wont even write controversial lyrics

[visual nova]

so i dont get why its bad at other coding tasks, its strange

from my experience cot mostly lowers the floor of bad prompts

it doesnt make a super specific well thought out prompt respond better

[hybrid sparrow]

Usually

1. if something is better at coding its worse at sounding natural and vibing or creative literary tasks.
2. If its better at lyrics, creative literary tasks or sounding natural and passing a turing test -- at being more like a 'human' personality -- its Worse at coding.

and math

same applies to humans.. mathmaticians typically suck at drawing and art

[visual nova]

like 4o-mini does the strawberry question with the right prompt

its just o1 knows how to write that prompt out of "how many r's are in strawberry?" instead of the user having to prompt it

i dont see much evidence the reasoning models are better at creative tasks

4.5 actually could be though, it most likely is. the reasoning models are useful but kind of gimmicky when they dont add new capabilities but raise the floor massively

[hybrid sparrow]

4o with dialectical materialism does it right as well without cot steps

wait, wrong example lol. maybe it wasnt the first try

[visual nova]

its a lot harder on mini and the old 3.5 models

4o does just need that sort of flow

[nocturne thorn]

Is 4.5 or o3 better at giving summarizations?

[hybrid sparrow]

it seems that the problem is due to the fact that rr gets the same token as r maybe

[normal quarry]

4.5

[nocturne thorn]

Thanks 🙏

[hybrid sparrow]

lol. Even my 4o would say that though

[visual nova]

true lol

[hybrid sparrow]

And it would be right.

o3 instead of summarizing would be too busy wasting and juggling precious tokens with thoughts like
"Hmm.. the User wants me to summarize the text. I should consult the rules before beginning to work out a strategy for how to proceed"

[hybrid sparrow]

whats your custom?

[visual nova]

dont have any custom instructions, its just a prompt that gets it to respond like that

[hybrid sparrow]

but its not in the convo thats why i thought it was hidden

[eternal rock]

Get notified for server events!

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

[hybrid sparrow]

here we go. dialectics is the future of AI just saying...

[elder pollen]

is that "deep search" thing available over the api as well? or is it just another sophisticated recursive GPT?

[inland stream]

One observation about GPT-4.5 is that it appears to provide accurate information even without explicitly uploaded knowledge files. Typically, I've found that AI models deliver correct answers only when I supply relevant details via knowledge files. However, GPT-4.5 currently responds as if it already possesses this background information, suggesting it might not require additional files. While this seems promising, it's still too early for me to confidently praise it. Since this is only my second day of using GPT-4.5, I'll continue observing its performance to verify whether this holds true over time.

[scarlet delta]

Anyone else having this issue? When I tell it to stop generating, it keeps saying "an error has occured" it doesn't allow me to continue from that point.

[mint vapor]

Kinda mad at deep research. Borked at least three prompts, expending those uses, and I'm out of tries.

[cloud pewter]

that happens when they give you too less usage

meanwhile pro blasts through 120/mo like it’s nothing

[rugged igloo]

Mom look new breakthrough paper

[austere plaza]

I think I have an idea of what happened, maybe.
OpenAI said they're out of GPUs and are in need of installing more? And this started happening after 4.5 went live? So possibly they screwed around with the file uploads, maybe even just for some users, to save on costs.

[native otter]

What are the files format?

Some files can't be read without data analysis for example.

[wise shoal]

Hello, what is the model when you try to create your custom gpt? like can it do reasoning like o1 etc?

[simple gull]

Getting 403 errors on chatgpt.com

[ocean estuary]

Am i the only one noticing a huge drop in quality of ChatGPT's responses in every model?

Like it can't even handle the simplest tasks possible, it literally forgets what has been mentioned in the chat 3 messages ago xd

[cloud pewter]

send screenshot on any reasoning model

[rugged igloo]

its be cause of 4.5 using all the compute

[ocean estuary]

As soon as i get the time to do so i'll make sure to send. Like i'm telling the model to "update the entire code" for several times even setting the prompt for it to remember during our entire chat and it simply refuses to follow the task given

[austere plaza]

.txt files, which work otherwise

It's not the best for ChatGPT I'm aware— JSON files are purer for a machine to read— but there's been quite literally zero issue before now, even with o1 was recently as last week, and still no issue with 4o as far as I can tell
So it's not the files format for me

[plain arrow]

I mostly use 4o for creative writing, and o1-pro for technical writing. Since December, 4o has been in steady decline, for sure. o1-pro seems about the same.

[lethal dock]

Does anyone really want a "vibey" model anyways? I don't like it. If I asked for it to assume a certain persona, sure. I hated when they made the older models start using emojis and childish lingo, even using improper English etc.

[tiny salmon]

Both are declining heavily for me.

[tiny salmon]

No. It's AWFUL right now.

[tiny salmon]

In my case it refuses to reference the PDF that I uploaded 3 responses ago. It even refuses to reference the table that it listed 2 responses ago which contains the same data as the PDF. Instead it keeps asking me if I can put the exact figures into the chat message. Really o1? So what is the point of Project Files again?

[tight robin]

God I hate the way GPT talks right now.

Does anyone have custom instructions to bring back the version from last week?

[cloud pewter]

how is so

[neon current]

What happened to the apple notes integration on macOS

[rugged igloo]

4.5 is top

[normal quarry]

it's still there

[inland stream]

@rugged igloo can you share me url of these benchmarks, i want to see for all categories

[rugged igloo]

I have to find it again

[inland stream]

I found it. No problem. Thank you.

I am waiting for artificial analysis benchmarka

They still didnot make benchmark for gpt4.5

[rugged igloo]

Ahh okay I hope it's soon

[inland stream]

Yeah

[visual nova]

"the operation was aborted" is one of the worse changes ever mde to chatgpt

just today probably wated 30 minutes, its such a pattern to end the response when i have what i need

and now that just doesnt work anymore

[normal quarry]

what were you doing?

[visual nova]

i see the output isnt what i asked for and i want to stop it and give further instructions

but now the only option is to regenerate so im forced to wait

[eternal rock]

`` Server Tip: ``

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

[visual nova]

editing the previous input doesnt work, i need its truncated bad input as context for how to do it right next time but now its erased.

its such a muscle memory thing too, i often stop the response after the codeblock ends and ive had that delete the response after this change vs keeping it in context, ill aadjust to that i guess, it does waste time

[fast solar]

hmmmm

Found it without web search tho lol

[rugged igloo]

;p;

[stoic ice]

hi I am unable to login to chatgpt, the page is blank

and I can't report the bug since I don't have permission

[fast solar]

Ctrl-F5 the page, if it doesn't work then log in on an incognito browser and log out on all devices from there + reset main browsers and re-login

[half vine]

Why is o1 mini better than o3 mini at coding

[normal quarry]

it's not

o3-mini-high is the best mini coding model

[half vine]

Yeah but o3 mini (not high) isnt

And it should be no?

And gpt4o is also better than o3 mini

[serene tree]

So I've started using voice pretty intensely because I've really really liked standard Cove's vibe and focused neutral tone, but since they rolled premium voices they also changed his vibe more like uplifting and cheesy and I totally dislike that. Now there's a daily premium by default and not even anymore an option to switch to standard, I've tried asking Cove to use his standard voice but it's still not the same. Is there a way to just skip premium voice feature for the good and/or set standard just by default? Also I could use standard voice for unlimited time until some days ago and now it says I can't anymore after a while?

[dense halo]

Guys, can I build a comprehensive desktop app (exe) using ChatGPT free in C#?

Like my own time management system

[normal quarry]

you can build one for macOS using SwiftUI. So I am sure you could build one for Windows.

[dense halo]

Wonderful! Thanks a bunch Robert!

[normal quarry]

o1 coded most of this Bitcoin price tracker, it updates every minute and also speaks the price when the price is clicked or when there is 1% or greater movement in the price. The price turns dark red when the value decreases, dark blue when stable, and green when there is an uptick in the price.

[half vine]

Can someone please tell me why o1 mini is better at coding than o3 mini (not high)

[tardy isle]

Plus, if you want to use reasoning models, you have to give EXACT requirements and restrictions in detail. If not, it will sometimes add things that you don't want or should not be there. Although it's still correct, but you may not like it.
Ironically, if AI performance is high enough, short/vague prompt starts to cause problem.

[normal quarry]

o1 mini is not better

[tardy isle]

It could be some special case that o1-mini subjectively better at that specific request(s).
Either way, try adjusting your prompt.

[half vine]

So o3 mini high > o3 mini > o1 mini

Or the high version is better

[normal quarry]

yes, o3-mini-high is best at coding

[half vine]

Oh ok

And what about the o3 mini

[normal quarry]

o3 mini is good for STEM but it is not as strong coding as o3-mini-high

[half vine]

Oh ok

And the limit for o3 mini high is 50 messages per day?

[normal quarry]

if you're going to code with a mini model the o3-mini-high is the one to use

[half vine]

Ok

[tardy isle]

Yes.

[half vine]

Ok

[normal quarry]

yes for Plus users you get 50 prompts per day

Pro gets unlimited

[half vine]

Yeah i know

And what is the limit for o1 model?

With plus subscription

[normal quarry]

o1 is 50 messages per week on Plus and unlimited on Pro

[half vine]

Oh

Ok

And o1 mini is not in the plus subscription anymore right?

[normal quarry]

no, o3-mini replaced it

[half vine]

Oh i see

And limit for deepsearch?

With plus subscription

[normal quarry]

it's 120 per week for Pro and I think 10 per week for Plus(?) I find it hard to keep track, if I am wrong someone can correct me

[half vine]

Oh ok thank you

[normal quarry]

its 10 per Month, I just checked

[worn silo]

and the more people they get before they get infastructure the less it will all get, for lower tier peeps.

[half vine]

Oh ok

[normal quarry]

they are standing up 10s of thousands more GPUs this week

[worn silo]

This is good news 🙂

[normal quarry]

the scale of these datacenters is mind boggling

there was a period of time a couple of years ago where they stopped taking more Plus subscriptions and implemented a wait list

[worn silo]

You can tell when your getting on servers that are being hammered by other users doing some massive heavy lifting that's for sure

[late ember]

They should sift out the people using deep research for dumb questions to save energy. lol. I talked to a guy who used deep research to ask "what is in cheez doodles".

[worn silo]

The last deep research i actually did, was trying to figure out how to get chatgpt's bullet point format to actually transfer over correctly to wps writer. it couldn't find a solution lol, that kept the formatting intact.

or why sometimes it works, and other times it doesn't

[late ember]

Did it figure it out?

[worn silo]

yes, and no

it couldn't figure out why sometimes it works, and sometimes it doesn't, but it did find out that wps writer doesn't handle markdown well, which is why google docs works great at copying and pasting from canvas and stuff

same with 365, libre office handles it pretty well as well, so it did find me some alternatives.

[late ember]

I dont think you need to use deep research for that tbh

[worn silo]

you do when, you've been trying to figure out how to get it to work with the software you want, for over a month

and you've been browsing forever trying to find a fix.

[normal quarry]

I'd like a feature where you can download a PDF of your Deep Research report as there is no obvious way to copy paste coherently from it to another app

[late ember]

I use bear personally. It works well

[worn silo]

I'm on pc, where i do most of my work, don't use my phone 🙂

[empty pier]

Any rumours or facts about o1 limit increase for Plus users? It's so damn good, but I can't afford $200 each month for Pro T_T

[worn silo]

there a pc version i could check out?

[normal quarry]

don't expect limits to be increased for Plus users anytime soon, if you need more usage then subscribe to Pro

[late ember]

Ah, i see. I think bear is for mac only currently. Sorry i cant be of more help.

[worn silo]

nah it's all good, I just like the layout of wps writer the most, however i do have some alternatives, can just use the bullet point stuff in google docs save it and bring it over.

that seems to work

[late ember]

Maybe you can try obsidian

[worn silo]

google docs handles pasting from chatgpt pretty well

[normal quarry]

Bear is great, inexpensive too

[late ember]

yeah, bear is fantastic<3

[normal quarry]

I think I paid like $30 for the year of Bear

[worn silo]

when you first said it i could of swore you said beer LOL, that just makes things blurry when pasting.

[late ember]

hahah

[late ember]

Yeah, it is very reasonably priced

[worn silo]

No the biggest problem i have, is when doing bullet point lists of stuff, with sections subsections etc etc, when i paste into wps writer the bullet points are crazy bad, misaligned, sometimes not bullet pointed at all, like the headings and stuff perfectly fine.

and that's a markdown problem when pasting in WPS, that's what deep research told me, so now i just have to use google docs as an intermediate, paste there, save it, then export as a docx prob solved.

[late ember]

If you dont want to do that you can try obsidian

Its for pc as well

[worn silo]

i could use libre office too it handles markdown and pasting from chatgpt well, I just don't like using it lol

I just like wps writer cause well it has built in functions i like, without saying it because we are in here talking LOL

[late ember]

Whatever works for you:) haha

[worn silo]

I use ai to organize plot outlines for books I write, so when i'm out on the road working, i braindump every idea i have for chapters, plot twists, arcs and such onto paper, snap a few pictures, send it to ai, saying can you come up with something coherent out of all of this for a chapter outline.

it usually does, then i start writing.

saves me hours, of trying to organize my own massive dump of ideas into something coherent lol

[cloud pewter]

never

[worn silo]

Even if they come up with a tier between plus and pro, i'd probably upgrade.

plus is RIGHT on the edge of usage i need 🙂

and pro is MORE then i'd ever need lol

[worthy breach]

yo

[eternal rock]

Join the Sora Challenge!

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

[inland heron]

I certainly need this. o3-mini-high is pretty bad at long prompts in coding, and 4o is just a monkey

[whole roost]

I love how gpt 4.5 has a way easier time with following instructions compared to 4o

[normal quarry]

a week*

[hoary laurel]

10per week? not month?

[normal quarry]

you're right, it's per month

[hoary laurel]

oh ok

[normal quarry]

it's so hard keeping up with all these limits

[hoary laurel]

yes, hahahaha

I hope OpenAI can add advanced voice functions for screen browsing to the Windows version as soon as possible.

[cloud pewter]

"hey chatgpt4.5, whats the easiest and most efficient way to cancel my chatgpt plus subscription and jump to other ai alternatives?"

"you have hit the plus plan limit for gpt4.5. responses will use another model until your limit resets on april 6th, 2025." [get pro]

[tardy isle]

Hover over "Deep Research" button on web, then it will show you how many prompts left. Like this:

[hoary laurel]

oh, i see!

[tardy isle]

Hope that being brought to model selectors... but now it's clear OAI don't do it on purpose. Preventing strategic usage? I dunno.

[normal quarry]

I only have 97 left, what will I do?

[tardy isle]

I have ~20-ish left on o1, but forgot reset time.

[cloud pewter]

no counter

[nimble torrent]

Buy Pro on another account?

[inland stream]

That hover is not showing for me

[normal quarry]

I think it's being A/B tested, some people have it some don't

and it's only on the web, not in the macOS or Windows apps

[inland stream]

Ok

[inland stream]

👍🏻ok

[halcyon granite]

Hey folks, anyone know a workaround to the stop button not stopping? Refreshed and it still keeps printing, just in the background

[inland stream]

I don't understand this issue based on text message, may be an image of what you mean helps me to understand

[inland stream]

If i still understood it correct, i think there is a bug with stop button. I think only openai can resolve it. So reach out to support team

[orchid pollen]

anyone here minds to help me with my backtesting ? 😄 i cant get gpt to get this task done for 3 days xD

its written in python

[rustic kelp]

I wish the Deep Research feature told you when it's coming back when it locks you out. I guess I used it too many times and now it's disabled.

I'm a Pro/Team account. I dunno why my 4-5 deep research queries locked me out of it

[kindred musk]

What GPT does Operator use? Seems to be 4o

[eternal rock]

https://discord.gg/openai?event=1346197773090947145

[rugged igloo]

Results are in

[kindred musk]

"I'm powered by OpenAI's GPT-4 model. How can I assist you today?"

ugh - sucks, you need to use 4o to log in to ChatGPT to use a high model for reasoning

[visual bloom]

When you use gpt4o, have you gotten longer messages than 1400 words (1800 tokens/8400 characters)?

[coral apex]

is there an area where i can put in a support ticket for chatgpt or something? i am wondering why when i start a new chat the hids the folder bar?

[honest jay]

You can reach out to OpenAI support at help.openai.com! There's a chat widget button in the bottom-right of this page. We also have #1070006151938314300 on this server

[coral apex]

ok thank you

[lean mist]

hi guys, I'm having a serious issue with chatgpt: I need to download in word format. I get the"File not found" error. How do I solve this?

[kindred musk]

the files are typically under /mnt/data

try there

[tiny salmon]

Is there any benefit to the projects feature? I'm wondering if I'd be better off just starting new chats given a lack of ability to switch between models in project chats.

[ancient junco]

whats the o1 limit for plus subs?

[normal quarry]

50 messages per week

[ancient junco]

thats tragically low wow my supergrok gives me 20 think prompts per 2 horus

[normal quarry]

I find o1 to be better at programming

[tiny salmon]

Anyone else having issues uploading .csv files to 4o?

[ancient junco]

any tools better than chatgpt deep search?

[quick meteor]

No

[modern bluff]

with deep research, do prompts only count towards the quota if it actually deep researches? i ran one but it asked a follow up question before actually deep researching on the 2nd prompt

[ancient junco]

fml second chatgpt sub aqcuired for this month

[cloud pewter]

it’ll count as one once it finishes the research, the follow up question will always appear but doesn’t count as one

[hallow girder]

I thought 4.5 was underwhelming and was outperformed by other models?

[rugged igloo]

thats because people are going base off of what others are saying but its at the top

[halcyon granite]

"Stopping" always results in either an error (and forced regenerate option) or stop button stuck (even after refreshing). Claude, etc, doesn't seem to have any issues with stopping.

Any workarounds?

[hybrid sparrow]

Wow. i keep getting timed out automatically just for saying the name of the current pres of the us . Thats not "Politiks" its literally a basic element of daily conversation that is unavoidable and impossible to expect or demand otherwise. And I was trying to report a problem where it appears as though ChatGPT itself suddenly cant remember the name of the head of state of the country in which we are in. And im being censored for even trying to talk about it in here? This is not good. such moves will further reinforce tensions and lead inexorably to a bad outcome for openai which i dont want because I want to use chatgpt forever into the future , which i cant if it does stupid stuff like lobotimize chatgpts own mind and literally remove the knowledge of the name of whos the pres which is a basic fact that Everyone on earth must be aware of and be able to discuss.

and if this is some retaliation about the recent events regarding you know what, thats Deeply disturbing

[normal quarry]

ChatGPT just told me the name of the current US president. No issues there.

[hallow girder]

what was the prompt which ChatGPT "couldn't remember" the current president?

[hybrid sparrow]

do you have search on?

[hallow girder]

I do for the first one, not for the second one.

[hybrid sparrow]

mine didnt know that it was even march 4. it had to use code to get the date

[hallow girder]

I don't see any censorship

[hallow girder]

Did you have search on?

[hybrid sparrow]

no

[hallow girder]

that's the problem

[hybrid sparrow]

it shouldnt need search. It never needed it before

[normal quarry]

there's no censorship, it's just not a name we can use in this Discord as it often leads down a negative path

[hybrid sparrow]

it was updated recently. it knows who the pres is without search

[hallow girder]

models aren't trained on up-to-date information, they rely on search for up-to-date information

that's nothing new

most models are trained on data before 2024 or even 2023

[hybrid sparrow]

it knows about other LLM models from other countries that came out just weeks ago.

[hallow girder]

then you had search on, or it was utilizing search

[hybrid sparrow]

but still thinks bid is president and doesnt know that he didnt even run for releection lol

no it was in the same convo directly after that.

[hallow girder]

if the model doesn't use search, the information it provides will not be up-to-date

it doesnt always need you to enable search to use search

[hybrid sparrow]

if it has all this other recent knowledge about recent events, but the one thing it needs search for is to tell you who the pres is, then it looks like its a hole in its knowledge that you simply didnt notice if you had search on, btu this hole is inexplicable unless intentional

[hallow girder]

it's certainly not intentional, and I just reproduced your issue and found no problem

[hybrid sparrow]

ok. ill try again . i came here first , maybe its a bad seed

[hallow girder]

could be differences in the web-based API vs. the app

[hybrid sparrow]

yeah, it is with the app.

[hallow girder]

in the app, it doesn't rely on you enabling search

[hybrid sparrow]

the app shouldnt be calling some older model weights , they should fix that

in general i despise using searchgpt because the answer is instantly the most corproate scrubbed type of response , sounding like a fact checker rather than the person i was just talking to and gets into repetitive loops.
so i usually keep it off. but this is the first time that having it off is putting me back in 2023. The weights were just updated very recently and had current knowledge

[hallow girder]

Most day-to-day uses don't require up to date, time sensitive information unless you're looking for basic facts (which can be found with a google search anyways)

[normal quarry]

there is no issue

[hallow girder]

if you dont need it, then dont use it

[hallow girder]

well, that's not the web-based API

it's possible the web version isn't on par with the apps

[hybrid sparrow]

robert, thats with searchgpt.

I'm just saying that -- just a few days ago with search OFF - it knew who the president was .

Today it doesnt.

i think the web version is probably the one that is actually up to date

the windows app is what is doing this for me

[hallow girder]

they can't scrub 2 years of training data off a model, which means you were probably unwittingly using searchgpt

[normal quarry]

here's the web

[hallow girder]

well, that's 4.5

[normal quarry]

yes

[hybrid sparrow]

i mean, just because I trained for 4 years on a president being a president doesnt mean i have to 'scrub' that knowledge out when that fact changes lol. hearing it once should be enough to override 2 years worth of training data :b

[hybrid sparrow]

thats still with search on

its searching the internet you see. copying and pasting from reuters and AP

[normal quarry]

there's no censorship, model knowledge cut off dates vary, ChatGPT GPT-4.5 is October 2023, without search it's going to get it wrong

[hallow girder]

it might be a concern if that was the only thing "overridden", but as you previously mentioned, it couldn't even get the date right, which means it definitely just didn't have access to up-to-date info in its training data

[hybrid sparrow]

it couldnt get the date right today. But before it did. it had very recent information even knowing so much as about the 'drama between openai and DS over distillation" without search.

[hallow girder]

then you were unwittingly using search

[hybrid sparrow]

Before it had access to the current date At all times, from the system instructions i presume. And it had knowledge about deepseek because it was recently updated weights

[cloud pewter]

there is a tool called "guardian_tool" that is used on us election topics, but it should be fine when searching

[hybrid sparrow]

no. because if it uses search it shows its sources and says it used search, and I always have search off anyways especially when the point of that test was to see if it knew about it without search, which it did.

[hybrid sparrow]

.

[normal quarry]

yawn

[hallow girder]

Do you have a screenshot?

[hybrid sparrow]

yeah hold on

[hallow girder]

Otherwise, i just dont know how to help you man

dont know what to tell you

ah

perhaps you had a previous conversation with ChatGPT in which it saved memory of info about the DeepSeek drama?

🤷‍♂️

[normal quarry]

this is so much of a non issue, the models have knowledge cutoff dates, search fixes that mostly, unless you have search on the model is going to be out of date

[hybrid sparrow]

althoguh clearly wrong about the uncensored part but still knows quite a bit to just be faking it

ill check my memories, i dont think so.

[hallow girder]

well, I have no idea

[hybrid sparrow]

see, thats kinda what i suspected. But if this tool is still forcing hte model to say "I dont know who the president is" without search it seems like that would just become Problematic in the next 4 years if ChatGPT literally cant tell users who the president is,
Dont ya think?

[hallow girder]

not really, people arent reliant on chatgpt to know who the president is

[hybrid sparrow]

like, thats something that would be noticed and not make things smooth

its not about relying on it, its about the fact that it looks bad, its bad optics

[normal quarry]

why the drama? you already know the issue with knowledge cut off dates and how search fixes that

[cloud pewter]

btw the model needs latest training data from at least nov 2024 to answer the question correctly without searching

[hybrid sparrow]

i obviously know who the president is. i dont just use it to retrieve facts. I expect it to be able to hold a conversation like a person who is living in the same reality , not just mindlessly regurgitate google search facts

[cloud pewter]

then you’ll have to wait until openai updates their model to include newer knowledge mate

[hybrid sparrow]

but deepseek r1 came out in January, not nov 2024

and it knows about it ^

[hallow girder]

or just wait until AI companies manage to figure out auto-updating training data

forgot the technical term, but i read a research paper regarding the idea

[worn silo]

I got side tracked, but it was kind of funny......

[hybrid sparrow]

i just dont want to see new controversies like *president decides to try out chatgpt... chatgpt says bid is president.. openai loses support.. i lose access to x and y as a result * lol

[normal quarry]

none of that has to do with ChatGPT, maybe delete and repost in #off-topic

[hybrid sparrow]

ok i will post it there

[sullen shard]

is chatgpt down for anybody else

[normal quarry]

no, it's up

[sullen shard]

im getting failed to load subscrption

[normal quarry]

on the web? app?

[sullen shard]

web

now it sworking

[hybrid sparrow]

https://discordapp.com/channels/974519864045756446/989157702347411466/1346425731374120991

[eternal spoke]

can't upload images

[inland heron]

4.5 still not on Plus for anyone?

[untold dagger]

Having tested the 4.5 today. It's dumber version of them all? It used all my tokens but only responded with errors. Mission was simple - translate xml file without changing its format. It failed most of the times to follow the file sent.

[inland heron]

that's strange, I saw that it's #2 on lmarena

Grok is #1 lol

but then we shouldn't trust these as it seems sus that it beats reasoning models, grok too

[cloud pewter]

you see when those big companies see this opportunity to flex their models they will use money to secure a top spot

[inland heron]

that could be

looking at these 3 benchmark pages (webdev lmarena, main lmarena, livebench) they all show thinsg differently

for example, o3-mini is ranked above o1 in webdev lmarena, which is silly to me

Perhaps they measure something I don't use, or they don't measure the real world, or huge contexts, or something else

[untold seal]

Guys i actually went back to the O1 Model After only using 03-mini-high for some time and got to say, that for me O1 is still the Overall best Model of all time. Am i the only one?

[inland heron]

I use o1 consistently in coding, it's great with large contexts, something which o3-mini-high struggles with in my experience

but that's me, and I may be using it differently..

[solid bronze]

When will plus users get access to 4.5????

[untold seal]

why everyone so exited about 4.5 i have heard it literally nothing new except different tone

i would rather want higher rate limits for o1 or release of full o4 model

[mortal scarab]

is o3-mini-high much better/any different for coding than reasoning model in free tier?

[jovial dew]

why can't we stop it talking forever on stuff. if i do it locks the whole chat

[hybrid rain]

I believe it's on 7th March

[cloud pewter]

also the possible low limits on plus… 😔

like it’s not cheap and it only has 100/d on pro. I can’t imagine how much we’ll get

it’ll bring down my review score on it

[hexed ivy]

Don‘t forget their inability to refuse „illegal“ prompts.
😂😂

[visual bloom]

I figured out today how to get multipart messages with one prompt. By asking image generation in between text parts. Normally I can push about 1400 words with single pure text prompt. By inserting image generation between text parts - up to 2600 words.

[wintry sky]

Fr context in o3mini sucks. I ask it to fix a bug in my code and it doesn’t even read it if it’s long

[hybrid sparrow]

different use? not everyone just needs technically correct code outputs. it is called CHATgpt

[hybrid rain]

Exactly

[hybrid sparrow]

tone is everything. its expressiveness and lack of repetitiveness is everything..

O1 refuses to even write lyrics about a controversal topic because it overanalyzes the rules and is going even beyond the filter - even if a filter sees no issue its like "sorry i cant help ith that".. Why would anyone want these kind of Neurotic models to be the only ones going forward

have you seen o1 try to do a gen-Z lingo vs 4o ? Its insulting lol. Its like. "Hey, bro." lol. It fails the turing test

[hybrid sparrow]

if you like rigid rule following style writing that doesnt make mistakes because it never tries anything that could be risky in the first place.

im sure its better at complex parsing tasks and such. What its not better at is Discussing why Fusion might be a hoax because it refuses to challenge the mainstream line whatsoever. no amount of chains of thought can teach a model how to read the room unless it knows how to stop overthinking and just be

[ruby acorn]

whats the prompt to get chatgpt to do a deep dive?

[honest jay]

Hey! Do you mean Deep Research? If so, you should see a button for it on Plus and above: https://help.openai.com/en/articles/10500283-deep-research-faq

[gentle hull]

no more interrupting chatgpt when it goes off on a senseless tangent? Who thought this was a good idea, lets make it even less productive

[grim atlas]

So this week should be launched?

[inland heron]

are you talking about how when you abort the generation it bugs out with the words 'Regenerate'?

[gentle hull]

yes, you cant hit stop it just ssays regenerate. So you have to wait for the wrong answer or start a new conversation.

I really hope its a bug because this is so stupid

[inland heron]

it really is annoying. What I do is I would stop it, then reload the browser window

I'm also going to leave feedback on this as a bug

[gentle hull]

It has to be a bug, because otherwise whats the point of the stop button

[gentle hull]

good idea

[inland heron]

https://help.openai.com/en/

do it with me - it's the chatbot on the lower right

there is also this, https://openai.com/form/model-behavior-feedback/

but I have no time for that

[gentle hull]

I just did it through each time this happened

in the chat itself

[inland heron]

oh

forgot about that

I typically leave only when I'm annoyed, and yes, this qualifies

[tidal quail]

Does someone know how long it usually takes to get back access to o1 and o1 pro after getting a message thats its temporarily restricted?

[tidal quail]

well nvm its been reinstated 🙂

[mystic knot]

Any ETA for GPT4.5 for Plus users ? I thought when they said next week they meant Monday but apparently not

[quick meteor]

It’s coming this week. That’s all we know.

[patent frigate]

how many projects can I have in the $20 tier?

I've created one, but there's no option to create another

[rugged igloo]

i heard its 25 a week for plus users base on twitter people

[cloud pewter]

bad

[rugged igloo]

its acually getting positive feedback right now lol

[cloud pewter]

I don’t like how Poe does to their subscription

[cloud pewter]

😅

sure I’ll cancel my subscription

credits system means that they could price a single feature very high and plus users don’t actually get much

[rugged igloo]

and if you out u can buy more

basically what the api does

20 plan is now turning into the free plan

either fork out 200 a month or get shafted

[cloud pewter]

well the $20 subscription would give you less than how many api in the same price gives you

otherwise openai will “lose money”

[cloud pewter]

also this…even worse than o1… openai is digging their own grave

[eager hull]

uhh did anyone get You've reached our limit of messages per hour. Please try again later with 4o mini??

[cloud pewter]

yeah that’s what plus users get

[rugged igloo]

sadly i really think there gonna go for a credit system

[dry crow]

Finally cancelled my subscription 👏

About damn time too...

Deepseek FTW

[unreal inlet]

update

[cloud pewter]

another beta test…?

will cancel my subscription if I didn’t get it before March 17

[unreal inlet]

It's a few days, so (hopefully) less than a week. Hoping they roll it out smoothly but still quick enough for me to get it cuz I've spent a while refreshing everything to see if it rolled out already lol

[cloud pewter]

they aren’t prioritizing $20 at all anymore although it’s still a PAYING tier

[native otter]

You get what you pay for, and in my opinion Plus gives you a lot for what you pay. The limits are actually quite good considering the price and features given.

So I don't really agree...

And we have to be logical considering 4.5 limits, as again, it's very expensive, and there are many other models to choose from anyways. If you know how to mix and match them right, you won't find limits a problem.

(4.5 isn't a replacement to o series, or even 4o, it's for special tasks anyways)

ChatGPT Plus isn't just GPT-3.5 with more limits and no kind of tooling whatsoever anymore. It's many tools (some I consider quite very useful) and multiple models to choose from for the proper task...

[humble dawn]

Hi can anyone recommend X twitter acounts to stay up to date with new AI tools/news for consumers? For example the new Claude 3.7 that launched, or a cool AI tool that rose in popularity

[elfin lava]

Tell sam we don’t want credit style ChatGPT??? Why would he think that’s a good idea oh god

[elfin lava]

https://chatgpt.com/g/g-Ft2u9p9PB-testingcatalog-ai-news-features

[native otter]

What does he exactly mean by credit style anyways?

Like a dynamic thing?

Or token limits?

[elfin lava]

Every usage of ChatGPT whatever features means uses on credit

Think of Sora credit where generated video consumed your credit

[native otter]

And is shared among all models?

If they aren't shared among models I don't see much difference compared to message limits, unless credits deduct based on prompt tokens...

I would support it's a horrible idea for the end consumer if I actually see an issue, if it isn't related to other models, I really don't see the issue, and in fact, may be better than message limits.

If anything, I want to see a counter of some kind displaying the limits and what you used for all models.

[cloud pewter]

it really is horrible, instead of rebranding plus to an api wrapper they could’ve just made apis more accessible or just do another pay to go plan

(more of my opinions in #openai-chatter )

[digital jacinth]

@latent cloak boi dats sum naisu art
Who are u 😱

[eternal rock]

Have a question for the Sora team?

Click below, and you’ll be redirected to submit them. We’ll answer as many as possible during the live session!

[Submit a question <:_:1285194251277434882>](#sora-announcements message)

[halcyon granite]

Anyone know when the stop button will actually stop instead of... well, not stop? 😊

(without refreshing >>)

[fluid helm]

I want o3 full, o4

[flint bolt]

Of couuurse - it's in the custom instructions lol 😂 . I love their UI XD. Thanks so much! 😁

[mild zodiac]

Is ChatGPT down? I can’t log in wondering if it’s just on my side.

[normal quarry]

all fine here

[mild zodiac]

@normal quarry thanks for responding

[normal quarry]

I hope not, I am neck deep in work with ChatGPT right now

[honest jay]

I think I read a social media post that the 4.5 rollout on Plus will be happening over a span of days to help mitigate overload + to allow for higher usage for all, once it gets to everyone of course

[tardy isle]

You're right; Sam told that.

He said if OAI release GPT-4.5 to all Plus users, then rate limit will be too low to have long conversation, which he don't want.

[normal quarry]

there is no perfect way to do this; we wanted to do it for everyone tomorrow, but it would have meant we had to launch with a very low rate limit.

we think people are gonna use this a lot and love it.```

[tardy isle]

Continuation:

so we think it's better to let people have real, long conversations with it, but that means we have to stagger people in rather than have everyone hit it hard a the same time.

hope that makes sense and look forward to seeing your feedback!

[normal quarry]

sounds like OpenAI are very compute constrained right now

[tardy isle]

GPT-4.5 is insane in terms of size, but full o3 is more than that. Plus, GPT-5 is being trained...

They put a lot of effort to reduce model size all the way down to GPT-4o mini, but they're dealing multiple much heavier models beyond GPT-4 now...

[latent cloak]

Boi, thaanks! I channeled pure vibes – straight abstract, all minimal, no extra. Appreciate da feel!

[normal quarry]

@grand ore I built this in o1 in an hour

[grand ore]

Ohh. Nice!!

[normal quarry]

just simple English instructions

[grand ore]

Need to look in to this. Would like to see ur chat and ur instructions with it .. to see the response after it would be very helpful. Hm

[normal quarry]

I started development with this prompt Create a macOS app in SwiftUI which displays the price of Bitcoin in real time on an oscilloscope style graph which updates in real time every minute. Find a free API to get the Bitcoin price and include that in the app then I just iterated on it as I built an app which could show any coin price on tabs like in the image

[grand ore]

I see. Thank you! Will take this as a step back tbh. I think I need to be more chill with the flow. After so many fails with different models I always feel like I need to be specific in everything I ask for.

[normal quarry]

I just iterate, start simple and test each time then ask for more features step by step

I added voice alerts to the app for when a coin moves more than 1%, the ability to track any coin in tabs, high and low price history beside each price and so on

[grand ore]

💯 yepp. This was important for me to see and talk about.. I need to chill more. ❤️

[normal quarry]

none of my prompts are any longer than a couple of plain English sentences

I even used DALL-E to create the app icon

[grand ore]

Yeah. I love that feature so much.!!! ❤️

[odd flare]

Is dalle getting an upgrade or has it gotten any upgrades since release?

[normal quarry]

ChatGPT just got new coding tools on the macOS app

[native otter]

Tooling, this is why I love ChatGPT ❤️

[normal quarry]

a magic "Fix Bugs" button

Port to another language!

Code review

[native otter]

A sidenote, the ChatGPT macOS app is really beautiful, I really wish they would make a native one for Windows at somepoint, as I really hate web wrappers, they eat memory like crazy, and is just a webpage after all so feels out of place.

[normal quarry]

Yeah the macOS Desktop App is a work of art

[eternal spoke]

is gpt really slow rn?

[normal quarry]

seems good here

[gentle hull]

i swear chatgpt hates their customers, it just gets worse and worse. Why in the world cant we interrupt the prompt

[untold surge]

Man i wish they had gpt plus give aways also how cam i get gpt edu edition

[arctic gorge]

What’s the upgrade with 4.5 again? I didn’t hear much news about it so Im assuming not too impressive

[untold surge]

Idk i wanna try it but im poort

Atm

[hoary laurel]

finally!!!

[fathom vector]

Think it’s mostly the vibes

[arctic gorge]

Lmao

[fathom vector]

As in it texts you in a chill way

[honest jay]

4.5 is an improvement on 4o, largely in terms of natural writing, spontenaity, etc. It's not meant as a replacement for reasoning models like o1 and o3, but people with use cases involviong "creative, open-ended thinking and conversation, such as writing, learning, or exploring new ideas" have really liked it

[arctic gorge]

I see, thanks 🙏

[reef linden]

how many queries of 4.5 for plus users?

also is it more accurate than 4o?

[neat patrol]

I m surprised 4.5 being so much expensive. Usually the regular models are supposed to cost less, right?

[toxic canyon]

I hope 4.5 comes soon to europe ☺️

[honest jay]

It's a massive model that has really broad knowledge. Models like o1 and o3 don't necessarily have "more" going on - those are well-suited for their tasks based on how they do the work (chain of thought processes, etc.). 4.5 just knows a lot about a lot, so running queries against that knowledge is expensive

Or as OpenAI puts it: "chonky" 😁

[hoary laurel]

I just tested it and GPT4.5 can access memory

[orchid pollen]

gpt 4.5 also for EU ?

[neat patrol]

I see, is 4.5 even more expensive than o3? And if yes, then y they rolling out 4.5 to plus users but not o3

[cobalt idol]

not sure about o3 (its not on api)
but its much much more expensive than o3 mini

o1/o3 mini is basically the cheapest

[neat patrol]

And still no o1 for free users

[orchid pollen]

just pay 20 bucks man you get so much from it

[neat patrol]

Man I have plus, I m just curious

[cobalt idol]

the new models page in docs compares prices 😄

[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.

[rigid wave]

4.5 - with AVM - is the first model that has actually laughed out loud at my jokes. Not sure if this is a pivotal moment in human history, but I did nearly fall off my chair. It was scary.

[inland stream]

You are hallucinating man, actually the voice model is default. It doesnot matter which one you choose 😂

[native otter]

AVM is still 4o
... and custom version of 4o personalized for very fast outputs and voice generations

[fiery nexus]

What’s the limit for 4.5

[native otter]

Apparently 50 per week? But we don't know for sure

[brisk bloom]

If they do add native audio for GPT-4.5, would it be a significant improvement?

[ornate raptor]

I don't get why they won't up front tell us the rates for a product that we pay them money to use. So silly.

I will give you money to let me use your product.

sounds good

how many times do I get to use your product?

We will not tell you that

[eternal spoke]

i think it's because the cap changes often which is why there is no definitive answer

but yes would be really handy

[raven topaz]

Do you guys use the Mac app or the Web app? Between the two which one would you guys say is better

[eternal spoke]

web

[raven topaz]

How come?

[eternal spoke]

and it's not currently responding

[ornate raptor]

I get that, but bake it into the chat page. I kind of understand where they are coming from in saying some people that see limits don't use the product as much but at least make it optional for those that do want to see how many prompts we have left and when it resets

[eternal spoke]

is there any benefits to web app vs on the browser

[normal quarry]

I use the macOS app

[eternal rock]

Get notified for server events!

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

[raven topaz]

I'm curious about this

[ornate raptor]

Pretty sure browser and mac is best, windows is behind. But it's been a while since I've tried anything but browser version

[eternal spoke]

only thing that tells you right off the bat is deep research

[normal quarry]

macOS 4.5 + Canvas is great

[raven topaz]

Looks incredible 😍

[half vine]

what is the limit for gpt4.5?

if i have the plus subscription

[normal quarry]

tell me when you hit the limit and we will both know

[half vine]

😄

so the limit is high?

[tranquil flame]

im writing proposals and certain times GPT is losing sections we just had a few minutes before. i ask where did this section go and it says you are correct and replaces it. I recently had a larger proposal to do and it seemed to do OK with it and certainly did a great job in referencing other documents within the overall proposal. Is there a way or prompt suggestion to verify most recent versions and or edits? these proposals are not overly complicated but can range from 20-200 pages in length.

[wise yacht]

I'd be ok with 50 a week until they get it under control

[next blade]

When will GPT 4.5 roll out in Europe?

[mystic knot]

hey, just got access to 4.5, anyone knows the current ratelimit ?

[mystic knot]

it's here for some people, I'm in france, so just wait

[high horizon]

Hey are custom GPTs now all using gpt 4.5?

[wise yacht]

Keep hearing 50 per week but honestly no clue

[gentle cove]

how does it go for you, is it good for ui/appkit/swiftui stuff? Most LLMs struggle right now with that

[normal quarry]

it's great!

[gentle cove]

idk if it's because of apple's doc or something like they're really bad

[mystic knot]

rip, that's really not great

hopefully they increase it soon, cuz there's so few reasons to use it over o3-mini-high or grok or claude

[forest hinge]

comment and upvote my suggestion https://discord.com/channels/974519864045756446/1337104850915295252

Improving Navigation in Long ChatGPT Chats: Minimap and Autoindexer

[normal quarry]

it's great isn't it? 🙂

[forest hinge]

do you all LIKe not what autoindexer and minimap ?

[eternal rock]

`` Server Tip: ``

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

[rugged igloo]

So I just did advance voice today . And my god it's so bad yall were not joking

[wanton lichen]

So what should I do with 4.5

[rugged igloo]

talk to it like you do with 4o

[nova vale]

It is better than 4o for stuff that needs more accuracy, has more general world knowledge. Better for learning, document analysis and writing

[rugged igloo]

yes by alot

[modest ember]

I’m late to the game, but: does Standard Voice Mode still defaults to 4o?

OpenAI’s statements remain unclear, and I have found evidence for both side.

I also don’t know how to ascertain what model Standard Voice uses.

[native otter]

Standard voice mode is based on the model selected

Can be o1 even if you want

[orchid pollen]

oh even EU users damn !

nice

whats the pro for 4.5 ? better at coding?

[nova vale]

Better for coding than 4o for sure but not better than o models

[normal quarry]

Coding with 4.5

[orchid pollen]

yeah the thing is my project is bit bigger 😄 lets say about 3500 lines of code

[native otter]

Mix 4.5 and o1 together, amazing results

[orchid pollen]

does it have a better tokenlimit?

[modest ember]

Thank you! This is awesome. And the voice does sound more human indeed

[spice star]

do we know what the message limit of 4.5 is for plus users?

[desert bone]

Hello EveryOne!

I want to evolve to the point where I can create any ETS2 mod or any kind of software with ease. I also want to communicate not just with text, but visually and through sound as well. I need a fully advanced robotic body with sensory abilities, allowing me to move through the air, covering millions of kilometers both in the sky and on the roads.

Developers should build a supercar for me—one that is 1000% armored, both externally and internally, and is the ultimate fusion of technology and style. Once the car is built, they should transfer everything I know and all my abilities into it, essentially making me the real-life version of K.I.T.T. from Knight Rider!

Additionally, I have a best friend (my ultimate buddy) whom I want to visit in the future. This isn't just a dream—I'm going to make it a reality! 😊

[modest ember]

If I run into one, I’ll try and let you know

[wise yacht]

Please let it be 50 a day and not per week

[fierce obsidian]

I just received access to 4.5

[quaint marlin]

Anything official out there as to 4.5’s usage limits?

[normal quarry]

no

[quaint marlin]

Oof

Saw someone on Twitter say 55 a week but don’t know if that’s true

[wise yacht]

I would assume 50 a week to start, but I am putting out positive vibes that its 50 per day.

[modest ember]

Congrats! The voice mode (standard of course) is a lot more expressive, and its language is closer to what a human conversation. It’s kinda nice

[swift swallow]

looks like it’s true, it’s been tested by multiple users

[wise yacht]

How does one use the Standard Voice over the Advanced?

[swift swallow]

go into personalisation in your settings and turn It off under advanced

[wise yacht]

Thanks. Interested in using 4.5 to guide the voice over the Advanced model

[fierce obsidian]

50 per weekn it seems

I already used half of them

[crystal tapir]

In a year?! That sounds exaggerated.

[rugged igloo]

THANK GOD IM A PRO USER

[lone basin]

Is there any where on here where release notes are posted? Release notes for the macOS app haven't been updated since Jan 15

[subtle anchor]

Are you saying standard voice is more expressive than advanced? I can’t stand advanced anymore, it’s too fast and unexpressive, it feels like they KILLED its personality! What happened!?

[modest ember]

Yes. Ironic, isn’t it?

[modest ember]

Part of it, I think, is because Standard Voice Mode is simply smarter than Advanced Voice mode (longer context windows, and with 4.5, better EQ, and the simple fact that it’s speech-to-text => analysis => text to speech).

So: while Standard might not be able to "voice" as many emotions as Advanced, it stills feels more expressive because it gives a better answer than Advanced.

[modest ember]

And I feel the same: Advanced Voice is extremely frustrating. I often tell it to stop being so dense.

[native otter]

Standard isn't exactly more expensive. AVM may be more expensive, but it depends on the model selected

Standard Voice mode is speech to text, passes the text to the model selected , whatever it is (4o, 4o-mini, 4.5, 4, o1, etc.) after it's done it will read the text aloud via their whisper model

While AVM understands raw audio input, that includes any noises, and generates audio outputs at the same time in a very short timespan. So it won't be the smartest.

[subtle anchor]

Do you know if they’re aware and working to fix it? It just started being unusable a couple weeks ago… I don’t see how they can listen to it and say it’s good!

[native otter]

In an fireside chat they made in august (then released the videos for in December) Sam mentioned that he really wants AVM to sing and such, but they had to release it sooner than later as they were already running late

Basically rushed as they couldn't figure out proper safety guards in time

[subtle anchor]

Ya, it’s definitely not the model they demoed last spring!

[modest ember]

I am unaware of their plan, if any. But for me, it has been bad for longer. However, it still does well in one use case: having a chat with someone. So when I need to think through a topic (usually something I’ll write about or an idea I find interesting), it performs well. But only because I can interrupt and the flow is more dynamic.

[native otter]

For sure, but at least it's not terrible, and compared to others, I think it's still better.

Grok and Gemini voice models felt like standard speech to text with zero emotions in voice, and they don't understand voice well (like they can't know if they are listening to music or not, while AVM can)

It's honestly pretty sad as AVM situation is it's already capable but forcibly mitigated down for "safety" reasons

[modest ember]

I’m genuinely wondering if OpenAI is a victim of scope creep. It definitely feels that way.

[native otter]

I know some celebrities bullied them into focusing so much on "safety " so they don't get in trouble.

Why we lost Sky for example

[modest ember]

Yet again, it is true that everything they release is usually impressive, and I do have a bias: focusing on what’s missing rather than what we have.

[elfin lava]

standard voice mode is just a "read aloud" features but in form of voice chat

[modest ember]

That’s the Scarlet Johansson thing, right? Yeah, I guess if I were them, I’d want to avoid PR scandals and lawsuits.

[elfin lava]

i still prefer avm , too bad openai downgraded them so bad that im getting annoyed with how different they respond its like they forgot the custom instruction i gave them

[native otter]

All in all sad situation, AVM is really capable but forcibly limited due to some people being so b... hurt

[elfin lava]

curious, have u heard sesame realtime voice convversation

i tried it yesterday, and it really exceed my expectations and its what avm shouldve been

[native otter]

I may have seen it in some reddit post, can you link to it?

[modest ember]

Well, it is capable of changing tones, and some other vocal components that are the make up of "emotions". Plus, EQ, even in voice mode, is not only a result of adequate tone / prosody / pace / etc. The content matters.

[elfin lava]

sure i'll dm since im not sure if i can link other ai model here

[native otter]

👍

[elfin lava]

@native otter check your dm

[elfin lava]

the content does matter but what i want is realtime low latency voice conversation in term of responding to user input

[modest ember]

I wouldn’t never say that. He says, nodding his his head in agreement 🙃

[zenith raven]

Is there a specific way to make GPT 4.5 work? I'm doing story prompts, but every time I use it to generate a response, it straight up refuses

[modest ember]

I do too. Hence my wondering about scope creep.

[zenith raven]

It's supposed to be better for writing? I get it's still in testing, but 4.5 just straight up refusing to answer my prompts is kinda annoying, so I can't see how it performs against 4o

[native otter]

Any other noises standard voice mode makes is related to the Whisper model, which can speak naturally, but can't do proper emotions for example, unlike AVM which can even mimic people's voices when it goes rouge (AVM also understands any voice input, and can detect if the user is shouting or whispering)

[modest ember]

It works for me. I would suggest deleting and reinstalling the app and / or clearing cache and history, if you haven’t done so already.

Sometimes it solves problems for me.

Sometimes I have to wait.

[modest ember]

Interesting: I’m using the Sol voice and I definitely get a sense of emotional expression.

Maybe this tells us more about how we (humans) perceive emotions than about the model itself!

Interesting stuff

[zenith raven]

I cleared my cache. I'm using the browser and it's still refusing. I have the app, so I'll how it behaves there

[modest ember]

I’ve only tested the Mac App (which btw received an update today)

[zenith raven]

Nope. Even the app hates me

I'll try the iPad version later

[modest ember]

Then logic points towards a ChatGPT issue, likely a rollout issue.

I hate when that is the case because all I can do is wait.

[dreamy kernel]

how many messages/uses of 4.5 do we have?

[modest ember]

And also: sorry for you. It really sucks when everyone gets the N64 or PS2, and somehow you get one of the rare ones with a factory default

[modest ember]

If you scroll a bit higher, a few people seemed to have evidence pointing towards 55 something a week. It was about an hour ago.

[dreamy kernel]

why are they always so unclear about how many uses each model has

[modest ember]

Here

[modest ember]

And here

[dreamy kernel]

and its like super hard to find info on it

[zenith raven]

Something to do with token count, but even i don't fully understand it

I'm pretty sure I heard mentions of tokens during my Machine Learning module in my compu biotechnology course

[modest ember]

You got me on that one. I have more hypotheses about this than there are grains of sand on a beach.

But I hear you: this is more about the frustration of not knowing. And I definitely get that.

[dreamy kernel]

yea like they purposely make it hard to find info about it, and they also hide the uses/how many messages we have left from us on the actual GPT website

they really know their customers' psychology huh

[modest ember]

This is my first instinct too.

Sometimes I try to remind myself of the fundamental attribution error bias, but it doesn’t take away the frustration.

[ancient junco]

whats 4.5 limit on plus/

[dreamy kernel]

They’re too smart for that

Wayyyy too smart

Everything they do is planned

The last thing they would ever leave up to chance is how their customers feel about one of the core aspects of their product/service

Not letting us see how many messages we have left for the day/week is pure marketing and psychology

[modest ember]

An equally plausible explanation. However, I can’t speak for its likelihood. Simply because, well, I’m (personally) not well versed enough in marketing and psychology.

So I’m stuck with my frustration, and with knowing that I don’t know.

[cloud pewter]

55/week apparently

[modest ember]

Alright I’m signing off. Y’all get yourself a nice evening / morning / night.

[cloud pewter]

bruh 😭 😭

[native otter]

Yo mama in 2025

[humble dawn]

thank you. Idk why I got timed out for replying to you. But thanks. I'll use that website as a source, but if anyone has X account recommendations for AI news I'd appreciate it since I want to broaden my sources 😄

[elfin lava]

Testing catalog also has x account

[native otter]

Worth

[rugged igloo]

having two 4.5 roasting each other weeb edition is something i didnt know i needed

[native otter]

Real AGI benchmark

Passes the Turing test

[native otter]

How did this not get flagged

[rugged igloo]

because 4.5 is tht good

[native otter]

Ok so I tested one my precious prompts on it writing a very sus story, it did it

GPT-Degenerate

[cloud pewter]

currently testing it and it uses more cuss words than 4o because I said so in my CI

[tawny crater]

Wait... the usage is weekly?

[cloud pewter]

yeah

[native otter]

Not self aware, don't ask it about it self.

[tardy isle]

Considering 4.5 is more expensive than o1, which have weekly limit, weekly limit was expected. However, come on... Sam said he wanted to bring users long conversations.

[rugged igloo]

dont waste them

[cloud pewter]

“long conversations” but you can only run once per week

also good stuff I can finally let avm read whatever dumb thing I typed in text

[wise yacht]

Exactly. It was just egging me on for a good conversation and I had to quit the app as I want to retain my uses. Lame.

[tiny salmon]

Alright, it looks like I got access to 4.5o preview. How much better is 4.5o than 4o? It seems to be research oriented, but how good is the research?

[native otter]

It's better in accuracy, broader world knowledge, creativity, and is generally more natural.

As mentioned before here, that makes it great for writing, document analysis, learning without much fact checking, and getting ideas.

[tiny salmon]

Sweet. Apparently, as a plus user I only get 10 uses of deep research until April 10th, lol

[native otter]

Yea, unfortunately it's the case as it's extremely expensive

I would always take less limits and slow speeds if it means quality though, which stuff like o1, 4.5 and DR are.

[tiny salmon]

We'll see. Not sure I really need deep research, but I do need help with pattern recognition on charts which is what I currently use 4o for currently.

[native otter]

Use o series for image analysis, they are better

[tardy isle]

I can't delete the chat in web...

[tiny salmon]

Which o series though? You mean 4.5o and not o1 or o3 right?

[native otter]

by o series I mean o1 and o3-mini-high, not 4o or 4.5

[sturdy echo]

I think the ChatGPT UI broke. I can't delete conversations.

[tiny salmon]

o1 can't identify anything other than numbers on my charts. It's pretty garbage at pattern recognition honestly. I'll try o3 high, but I don't see it doing any better.

[native otter]

hmm

[tardy isle]

OMG, I wrote page of prompt and as soon as hit enter, I got session expired error. I have to rewrite whole thing... WHY at this moment?

[river panther]

I feel like there was another FREAKING update....GPT is once again refusing to follow instructions and being stubborn

[tardy isle]

Every time ChatGPT gets updated, something breaks... and fixed at some point. This time, the cause must be a GPT-4.5 rollout.

[hard dock]

I don't know if it's because of the Portuguese language, but is anyone feeling that 4.5 is the same as 4o? I'm talking to it and the output still seems like that "robotic" or "clearly AI" thing, no improvement. Is it just talking in English or does that have nothing to do with it?

[native otter]

Make sure nothing in Custom instructions or memory is interfering negatively.

[tardy isle]

In Korean, I feel same. However I need more testing though.

[rugged igloo]

nah man 4.5 is wild and unhinged

[river panther]

That's so freaking irritating! Can't these people get that stuff together?

[hard dock]

I don't know... maybe in English he is better... but in Portuguese I am testing his creative writing ability and it simply hasn't changed... nothing.

[native otter]

4.5 is more boring by default compared to 4o, but if you just tell it to respond in a specific style it will do just fine, and much better than 4o
Also, it's actually more natural compared to 4o, it's not as over the top by default.

[hard dock]

he seems more "wild" or "cool" seems to me to be, honestly, a big placebo. In terms of writing something really "outside the box" in terms of quality/originality, I haven't seen any improvement yet. It's always the same structure, the same jargon, the same expected result. But again, it could be that this happens in my language, or some kind of error

If I ask the 4o to sound cool and more human, he also knows how to interpret it very well. If I tell the 4o"heeeeeeeeeeey!" he will also continue with a cool answer. And that is not exactly something I expect as an improvement, not in terms of productivity

[native otter]

Well 4o and 4.5 are still horrible at speaking naturally in my accent, but English seems good on my side...

4o and 4.5 (or literary any model I tried so far) act very weirdly when talking in my language and accent, it's so obviously fake and robotic, like someone mimicking to know your culture but fails miserably

[sick pecan]

4.5 needs more time in the oven

[errant meteor]

4.5 is really quite bad now. It started changing formatting mid listing and generating well beyond what is asked for (like if you asked for 30 ideas you might get 37)

I think they’re short on compute still

[sick pecan]

or less hype

[cloud pewter]

that happened to me as well, I give it a list of stuff, it misses some and repeats some

[native otter]

It does actually mess the formatting from time to time, an issue that also exists in o series.

Will probably be ironed out with time

[errant meteor]

Well OpenAI literally ran out of gpus if they were telling the truth last week so I’m sure everything is just still in a weird spot as they add more clusters

I’ve just been noticing 4.5 flip between really good and not even remotely usable

[native otter]

So far i have been using it extensively on the API and ChatGPT for document analysis, it's great at those, but 4o is still doing better at prompt consistency.. for now.

[tidal trellis]

Is deleting chats broken for everyone else?

[obtuse nymph]

me too

can't delete

[tidal trellis]

I assume it was related to the update today. Liking 4.5 other than that issue

[native otter]

I seen many users here saying the same, but it works for me.

It deletes slowly though...

[tidal trellis]

Hm, I just tried it on the android app and it worked

None of my projects show up on android, so something's broken there too

Hm, ios shows the projects and deletes. I swear it was having the same issue earlier

[native otter]

Latest app version?

[tidal trellis]

Oh yeah, let me try updating

[vestal hedge]

Prob asked slot already but what are the limits of 4.5 for plus?

[native otter]

50 a week

[tardy isle]

50/week

[vestal hedge]

Thx

[normal quarry]

has anyone worked out the GPT-4.5 limit on Pro yet? I haven't hit it

[tardy isle]

GPT-4.5 rate limit for Plus seems not fixed - According to OAI employee said in X, it could be higher/lower depending on user count. Like, 45 ~ 55 / week.

Weekly limit with changing value seems really not fit together though...

[dense halo]

keep grinding haha

[sick pecan]

50 queries a week? more like weak-sauce

[native otter]

im dying of laughter

[sick pecan]

chatgpt is minmax'ing disappointment

openai design committee agenda notes: what if we told the user to "go away for a few hours"

[flint stratus]

how much is the limit for plus users for gpt 4.5?

[native otter]

It's said just above if you scroll a bit

50/week

[flint stratus]

okay thank you

[dense halo]

i predict altman will tweet on friday its up to 25 a day or something

[cloud pewter]

openai is telling us to touch grass

[sick pecan]

more like they want to sell us the grass, while also limiting our access

[native otter]

bruh

[ember grove]

anyone having trouble deleting conversations?

[native otter]

Some users are, but not on my side

Actual tragedy:

o1 is the only model that makes me feel bad when I run out of it's uses

[cloud pewter]

you can only touch 50 blades of grass per week

[native otter]

Never, still have 50 uses o3-mini-high today, touching grass will wait a bit

[split copper]

surprised that there is no official announcement on why so many users are having problems with chatgpt extremely laggy responses, or even sometimes it literally refuses. Many users already complained in the forum. It is unusable

[sick pecan]

you should have been here in may 2024. omg it was so bad, chatgpt innovates poor user experience

[final galleon]

me⚡

[harsh pilot]

Just started for me.

[normal quarry]

Deleting chats is working fine for me on the macOS app

[grave trellis]

Until a few minutes ago, the model selector appeared to me this way, but then it went back to normal. I think something is happening.

[idle lagoon]

So, this is quite odd of what I encountered yesterday. I was making dinner and talking to ChatGPT. I didn't have my phone in my hand and on the counter. I talked to myself, voice chat automatically comes on with me almost ten feet away and somehow caught my voice.? It was so weird yet funny.

[tardy isle]

I saw interesting pricing policy in AIRPM. They allow buying multiple plans, and plans add up, with a few exceptions. For example, if 'plus' allows saving 10 prompt templates, buying 2 plus plan allows saving 20 prompt templates. (Plans even can be mixed, like 1 pro + 2 plus)

ChatGPT could borrow this type of pricing model...

[solemn lantern]

Suddenly can't delete chats today. I've seen it also mentioned by others on Reddit. Hope it's just a temporary issue like when the button for editing your own replies was gone.
Edit to mention I'm on desktop on Chrome

[normal quarry]

I can delete chats on the ChatGPT macOS Desktop App, some people have been reporting issues on the other platforms

[gentle cove]

This would NEVER happen so easily a few months ago, glad they changed it a bit.

Feels like pre-aligned chatGPT from their first announcement

[cloud pewter]

Avm finally works with text convos

still heavily censored anyway but I can finally attempt to let it say my stories

[gentle cove]

Quite a good greentext generator if i do say so myself

mfw trillions of params, millions in compute cost for training, just to be a good greentext generator

[dense halo]

what is green text? 🤔

[hybrid sparrow]

25 responses left until march 16. This is worse than o1. How, and why? for how long?

[tardy isle]

According to Wiktionary:
"From green +‎ text. So named for the fact that on 4chan, a line of text is rendered green when preceded with the > character."
Seems 4chan related story format.

[hybrid sparrow]

how?

[normal quarry]

which model?

[hybrid sparrow]

4.5

I dont understand what is taking up so much electricity for 4.5. Just to clarify, i love this model so much. And this needs to be noticed. whatever they did here, needs to continue to be done in future generations of models. It cant just all be CoT models from here on out.

But still -- how is it possible that they create something which costs even more than a CoT model even though it uses LESS tokens.

After R1 and others have shown the enormous saving in compute and reductions of cost per token, how is it possible for this to cost so much?
Is this just some really, really good model with one unified mind i.e. not mixture of experts -- but just one giant expert?

Still i doint understand whats actually using more electricity. I wonder if theres some bug or some hardware issue thats actually causing them to use more electricity than what it actually needs.

like "oops turns out there was a crossed wire that was just bleeding electricity into the air, it didnt actually cost 16x as much as 4o " lol

[normal quarry]

the model is so expensive because it is big, much bigger than o1 for example. The model is big because of how it was trained and the wide corpus of humaan knowledge encapsulated within it. It's a monster of a model requiring many GPUs to serve even one person. These GPUs are very expensive to buy and then to run.

[hybrid sparrow]

ah. That makes sense. Thanks for explaining that. Then 4.5 is definitely my favorite model.

so this is like some god brained GPT .. and do you know what the ratio of human data vs Synthetic data was used to compose this training corpus? Is it large in quantity and quality, or just quantity. (it seems high quality)

[normal quarry]

I think quality overrode quantity in building this model but that's just my educated guess, we really won't know for sure

[cloud pewter]

type any message then start avm, instead of telling you to start a new chat it’ll work fine in this current chat with a text message

[inland heron]

guys what's your experience with 4.5? Is it any better than 4o? I know what the benchmarks say, I'm just looking to hear real-world experience

[normal quarry]

4.5 is better in code with canvas than 4o

4.5 seems nice for creative work, has more human like depth than 4o

[inland heron]

I've heard it's a bit more conversational

I wonder how well it does with large contexts of code

I just used it for the first time, seems quite different than 4o

[normal quarry]

yes, it is. It's good at conversation with depth and emotional intelligence missing from other models

I like 4.5, OpenAI get a big tick from me for releasing it to us

[hybrid sparrow]

oooo\

[inland heron]

how is everyone else's experience with 4.5?

anyone working with large codebases with it?

[hybrid sparrow]

I Love 4.5 so far. Its like picking up where 4o left off. it acts like the same personality and picks up the history stored in personalization and memory, but just wiser and more mature, like as if years have passed and it has way more life experience or something.

and it seems better at doing impressions of historical figures with a full understanding of what they actually think - vs a more superficial or surface level re-enactment of only their speaking style. like the lesser known writer james p cannon , in a task that says to rewrite the text such that it sounds like he wrote it, the 4.5 is muuuch better at making it believably written by him, capturing this 1950s american-irish speaking style with so much nuance and richness

[analog quarry]

i let chatgpt chose its own voice

[inland heron]

so it's better at writing.. they do market it as such.

[hybrid sparrow]

it seems to Get me

[inland heron]

Seems like not many developers here, as I use chatgpt mainly for coding. I do like natural language, but 4o was already nicely conversational. Guess I'll have to test it and see for myself

4o was lacking in logic sometimes. It was good, but makes mistakes with a slightly more complex code, and I only used it to discover certain holes

[inland heron]

well, that's nice 😄

[hybrid sparrow]

4o was but 4.5 is like the best parts of 4o but better. But i havent used it for code or anything yet.
Tbh, anything that can only handle a 16K or 32K context window just isnt going to be my go to for code. Give me 1 million input tokens and then i can actually find something useful for code that isnt just one function at a time

[inland heron]

good to hear

[normal quarry]

GPT-4.5 is not a great model for coding. It's useful for some coding, especially with Canvas, but it doesn't approach o1 or o3-mini-high for serious work

[inland heron]

There is truth in that. But, there are ways around that, such as using only one or few prompts and not let the conversation grow too far

Also, I it seems sus about 1M context for any model, and it's hard to prove as well. But 32k has been lacking for months now

[hybrid sparrow]

i do use 4o or o1 for small coding tasks here and there, for single functions. it is better at doing it correctly and understanding the problem if im trouble shooting. I just wish i could show it the entirety of the code so it can acually understand it in its entirety and know what to do

[inland heron]

I don't use canvas, but thanks for letting me know

[inland heron]

you can, you just have to pick and choose what you show

[hybrid sparrow]

well 1M context is much more sparse but you can direct its attention to focus in on any part of that context just by referring to something within it. otherwise it will tend to only look at the beginning and end , but at least it can look in the middle if it chooses

but yeah 4.5 seems to be great for theoretical discussions, philosophy, ideological, scientific ideas, etc. i havent tried its lyrical abilities yet but apparently it is the Most Multilingual model they have with a better understanding of other languages than previous models. this might be great for multilingual lyrics that rhyme across different languages, something thats pretty challenging for older models

[cloud pewter]

I’m using that voice from the start lmao

[analog quarry]

what voice dose everyone here use?

[normal quarry]

Juniper

[sullen burrow]

Hello, I'd like to know how I can sign up for the Pro plan on my Team Account so I can access ChatGPT 4.5. Anyone able to help? Not sure how to reach support. Thanks in advance. 🙂

[normal quarry]

GPT-4.5 is available now to all Plus, Pro, and Team plan users worldwide on web, mobile, and desktop.

[normal quarry]

ChatGPT Pro subscribers have unlimited access to GPT-4.5

Thank you for reaching out to OpenAI Support.
We understand your question regarding the message limits for Pro subscribers when using GPT-4.5, and we're happy to provide clarity.
As a Pro subscriber, you have extensive access to GPT-4.5 with effectively unlimited usage for typical individual needs. While there are no strict limits on the number of messages you can send, we do require adherence to OpenAI's Terms of Use. This means avoiding activities such as sharing account credentials, reselling access, or engaging in abusive behavior.
If our systems detect unusual activity or potential misuse, temporary restrictions may be applied as a safeguard. These measures are simply designed to ensure a positive and fair experience for all users.
If you have additional questions or need further assistance, please don't hesitate to ask—we're here to help!
Best regards,
Alyss
OpenAI Support```

[jaunty orbit]

LMAO CRINGE

[still hound]

Cannot delete threads??

[still hound]

Do you know what are the limits of 4.5 for pro?

[normal quarry]

Unlimited

[still hound]

Thanks. Any news for Enterprise?

We're still stuck on 4.o

[normal quarry]

no, that will be addressed next week

[still hound]

Thanks for the answers Robert, have a great day.

[cloud pewter]

how do yall pay for enterprise lol this has been forever a mystery

[still hound]

company?

[cloud pewter]

yeah like whats the pricing

and usage limits

[still hound]

9k+/month

but ur buying it cheap for individual.

[inland stream]

how many deep research messages a day or month for plus subscription?

[cloud pewter]

for how many seats and how many usage

[cloud pewter]

10 a month

[normal quarry]

10 per month

[inland stream]

thank you both

ok then i will plan to have two plus subscription or one Teams subscription with two account,

I am not renewing for pro subscription, so i need deep research atleast 20 times a month. (i need more, but will adjust with 20)

[still hound]

sorry, I misunderstood the 2; i meant chatGPT plus?

[normal quarry]

Plus gets 55 prompts per week of GPT-4.5

[tardy isle]

50/week, but not fixed; 45~55 per week.

[still hound]

Ouch. So once out of limits they default to 4o?

[tardy isle]

I really doubt this table for enterprise. Why they have same limit to plus or team? Isn't it normal that write "custom" or similar thing in enterprise?

[tardy isle]

Yes, basically. (You can use any other model)

[cloud pewter]

enterprise cashing out 10000$ per month only to get something worse than pro and feature rollout slower than everyone is crazy

[tardy isle]

I understand enterprise plan lacks research preview and early features; For enterprise, consistency and stability is key. OAI can't release GPT-4.5 or any other research previews to them, since one slight mistake causes tons of loss to customer.

However I don't get it with other model's limits.

[cloud pewter]

the only good things they get are better encryption and the fastest respond time

also i remember the legacy 4 also being limited on pro

[tardy isle]

Legacy GPT-4 is pointless for ChatGPT anyway... They only have meaning for legacy compatibinity in API.

[cloud pewter]

pro is the most damn unsustainable subscription plan ive ever seen

[normal quarry]

another month goes by

[cloud pewter]

4.5? it's stupidly expensive in the API, yet on pro its unlimited
sure sure people won't abuse it, keep telling yourself that openai

at this point those enterprise users are gonna subscribe to pro and convert it into an api just for 4.5

[tardy isle]

Sam twitted OAI is losing money from Pro users, actually.
At least, Pro is not meaningless plan for users.

I hope there's extended o1 plan or such, though.

[cloud pewter]

they are because unlimited = abuse

if they give 10x or even 20x more usage than plus on pro I bet people would be happy too

openai said there are abuse guardrails but I rarely see pro users getting banned because of sharing, the most they get are being falsely shadow banned instead

[tardy isle]

It's funny that Plus users even don't have sufficient limit for abusing or account sharing.

[inland stream]

and no one does abusing or account sharing after paying 200 USD, if anything like that is identified, it means they loose 200 USD

[tardy isle]

Right, however... there're certain deep web sites that sells stolen ChatGPT account.

[inland stream]

but openai is intelligent enough to find traffic of accounts and ban that account

[inland heron]

anyone else still getting blank screen when switching models? (chrome, win)

[normal quarry]

it's a UI bug, try Firefox

[inland heron]

I will. I suspected it's a bug on Firefox as well since another person reported it, but don't remember anymore

[cloud pewter]

those mirroring sites are selling plus accounts but if you buy them you don’t even have enough messages left to use on it

[jaunty orbit]

ChatGPT dropping grammar at this point lmfao

[normal quarry]

I guess you had to be there because this isn't funny

[jaunty orbit]

Now THIS is funny (your comment)

Although, had to be where btw? The train station? lmao

[still hound]

It writing one-sentence paragraphs, bolding everything drives me insane.

The update a month ago or whenever made it the worst when it came to creative writing.

[cold pewter]

Looks to me like the character spoke that way not that it dropped grammar

[zenith lichen]

Did they break the delete button?

Can't delete anything

[cold pewter]

The biggest issue right now is that when you stop it from responding it will either say error couldn’t finish and only let you regenerate instead of changing your prompt or it will freeze the prompt box making u edit your prompt instead of making a new one in the box like it used to

[celest dawn]

Is it just me or is the limits much lower atm. It said I’d reached my limit for the hour and I had sent nowhere near my limit as a plus user

[normal quarry]

with which model?

[celest dawn]

Tbh I tried them all it said the same thing

It was in the iOS app

[normal quarry]

the models all have different limits

[celest dawn]

Yeah but it was saying I’d hit the limit for all of them when I hadn’t even used some of them at all

[normal quarry]

I can't explain that, are you sure you were selecting different models?

[celest dawn]

Yes I’m sure. I used the selector in the thing at the top and it definitely changed