chatgpt-discussions

[narrow viper]

Their UI deployment is also chaotic - no release notes, inconsistent versioning ( 1.2025.049 through 1.2025.049 currently open at the moment)/

@dense halo never trust it to be accurate. Trust it to find Some references for you and then go do your own research.

[sonic zephyr]

thanks for your time bro, enjoy your code! You know what you're talking about. see ya

[dense halo]

Oh ok. I'll do that. Thank you Capt. *$.

[narrow viper]

TYVM Ciao!

Hey everyone : If you enjoy using ChatGPT and want to contribute to its development, I strongly encourage you to look at #1070006915414900886 and #1070006151938314300 once in a while. You'll learn a lot about the products, you can help others with specific issues, and you can +1 on issues that you've seen and new features that you'd like to see.
OpenAI certainly looks at those channels but they don't respond. That's OK. Let's just do our part and let them do theirs.

Personally, I enjoy looking through those threads to get an idea of what people are doing, what they're thinking, how they're using the platform. I like to help when I can, downvote what I think are bad ideas, and try to help people to get what they want by suggesting that they upvote existing threads.
I hope you might enjoy that kind of time in this environment too.

[jade ivy]

Yes yes my bad

[subtle anchor]

Hello, new here, just wondering... Did they speed up Advanced Voice? All of a sudden it talks like it's on meth! Every new conversation I have to talk it back down to normal but it ends up reverting back to fast mode. It's distracting when you're trying to flow with a convo!

Breeze

I just tried Cove, it's not quite as apparent but ya, it's sped up. I have it a laid back beach personality and it doesn't fit the tempo anymore.

If it's on purpose, I'm sure it's to give more info faster, but it takes me out of flow. If I want sped up info I'll just use text mode.

[umbral monolith]

Haha I'm having the same problem she sounds like she's having a stroke she's talking really fast and she's not pronouncing her words correctly

[narrow viper]

A Lot of people are very in tune with Cove's laid back personality. The dev forum was in an uproar when they got rid of his quirks. Here too in #1070006915414900886 and #1070006151938314300. I really hope they didn't change that!

Again?

I'm sure the uproar will come back...

I haven't spoken with that voice in a while.

Hey, it's important that we're comfortable with our user interface, rite?

[subtle anchor]

Glad I’m not the only one, I. Sure they’ll hear the feedback soon! 🤞

[odd vortex]

Guys for a personal usage, is it worth to upgrade from 20€/m sub to 200€/m ?

[narrow viper]

no @odd vortex See my notes above about how to use the models that are available.

[low orbit]

do you get unlimiteds supervoice on 200

i never even tried supervoice

[rugged igloo]

not unlimited i think exstended but i never tried voice as its so nuetered that i just dont bother but really the only thing the pro sub has unlimited usange of the models and o1 pro which is the peak best model

[subtle anchor]

I thought they said unlimited advanced voice, but ya, they need to give the voices back their personality!

[narrow viper]

neutered?

[spring gate]

Hey everyone, is anyone else getting errors when asking the AI to generate a image file today? Was working fine last night.

[low orbit]

Aw no personality

Thats no fun

Ill go try

Ohhh never seen that before

[rugged igloo]

yea its preety bare bones

[low orbit]

Where it does 2 image

Only text

Dang thats dissapointing about voice from when their demo it was legit flirting lol

Ai cant figure out how to make a red panda wear headphones in a way that would work and its a real problem

Tbh i dont know either

So i think the proper assumption is they wanna start doing pick a better image and maybe its breaking sometime

[tawny crater]

Anyone else getting this over and over?

[modest dawn]

yep started a min or two ago just became constant instead of intermittent

[rugged igloo]

yea im getting it too

it keep outputting nothing

prob making the swtich to 4.5

[dense halo]

Am I allowed to mention ChatGPT in my username, like "i_use_chatgpt"?

As in my @ name, I want people to know I use it and other AI.

[fresh thistle]

Same, I gotten that in multiple chats too

[patent nest]

Is chatGPT down?
I am getting an error message

[patent nest]

Yes

[shut mauve]

yup

[tawny crater]

Its down forever

[rugged igloo]

bruuuuu

[tawny crater]

Time to go back to Claude

[rugged igloo]

^^^^

i my o1 pro

[dense halo]

Is it down if I only use text and no images?

[modest dawn]

https://status.openai.com/incidents/3v8s9t957bvf

[patent nest]

Its down for both

[dense halo]

It's not down for me. I told it hello with just text and it talked to me.

I'm using 4o mini.

[patent nest]

Me too but actually wanting to do things and tasks it does not work

[dense halo]

So should I use Copilot for now?

[patent nest]

I guess

Yeah

[dense halo]

Ping me when OpenAI's ChatGPT goes back up

[rugged igloo]

how am i suppose to cheat on my tests

[patent nest]

I need someone to ping me

[patent nest]

😂

[dense halo]

Copilot works

[whole roost]

if everything fails, there's still whatsapp gpt

[modest dawn]

fix implemented

[radiant hull]

same

its down

[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.

[rugged igloo]

it prob broke cause your on light mode dude

[dense halo]

This version of ChatGPT on Copilot I like because it always has emojis at the end

[fresh thistle]

It's back for me

[rugged igloo]

still broken

my test is due in 20 min

[dense halo]

Use Copilot

Or another app with GPT models

[rugged igloo]

co pilot will fail me it aint sMRT

seems fine now

[pliant dune]

hey my chatgpt isnt working

nvm

[dawn lodge]

Hi guys I’m new here

[rugged igloo]

bruh

[still forge]

😮 I must have missed that how do we get that?! It's it really openai ChatGPT?

[whole roost]

yes, they released that during the 12 days of whatever before christmas

it also has a phone number if you're in america, i'm not so i only have whatsapp

[still forge]

I literally use them both equally 🤣

Okay but now I have a question. In the chat I have with ChatGPT I ran out of space (like character limit) what happens to Whatsapp when that happens? You can't just start a new chat

[visual nova]

it didnt know anything from my past chats lol

maybe its only chats since this popped up or certain models? weird. they are going to have to cut that 1500 token system prompt if they even want the regular chat to function well.. groks is like 200

[still forge]

Oh weird I literally asked it "tell me about your updates" in a new chat and it started spewing facts about our most recent chat 🙈

[sly geode]

Should I cancel the subscription of plus . ? Cuz I can't see the difference between free and plus . They baked all great features in pro...

[visual nova]

yeah not sure

[still forge]

I'm not... That good at problem solving but I think only 4o can access it

And that definitely doesn't look like 4o and if it is then you have reasoning selected

[visual nova]

ok yeah, it didnt say it was 4o only anywhere

[still forge]

Well it's been said a lot in a lot of places that 4o is for like the "emotional" side or like the casual side and the others are the more logical emotionless side.

[visual nova]

regular memories are only 4o so youre probably right.

[still forge]

Idk I'm pretty new to actually talking on here so I'm not sure how they explain it here but that's how I've seen it explained elsewhere

[rugged igloo]

let me gen it

[visual nova]

ive never chatted with 4o on this account so probably no memories to access then lol

but that just means o3-mini-high has been a really good model, its lived up mostly

[still forge]

🥴 I just asked ChatGPT what it thought of me and I'm gonna go cry now. Honestly, how close does a bot have to mimic emotions before they're real enough
To add to that, aren't all friendships and relationships just built algorithmically?

🦗 Are we allowed to have those discussions here?

[rugged igloo]

broooo

yea its a hit or miss

so ur the one thats breakin the servers

[still forge]

Oooh that's a good one. I think since o1 can do images now I can test it on a crossword I tried earlier. 4o was not very good at it

[rugged igloo]

idk thats also spacial prob wont be good

[cloud pewter]

you should lol

[rugged igloo]

yo dont tell people to cancel

[cloud pewter]

if they aren’t satisfied, they should 😭

[still forge]

Didn't do well at the checking my crossword answers o3-mini

[cloud pewter]

but I can’t work with the free version so plus is good for now

[rugged igloo]

crossword puzzles and ai are not good right now i wouldnt use them

even o3 full is bad at spacial awareness

[cloud pewter]

U using o1 pro?

[still forge]

Nah I was just experimenting. It did get answers right but it just was telling me I got answers wrong I didn't

[rugged igloo]

yea gettng errors left and right

[still forge]

"so good at coding" 🥺 where?

[rugged igloo]

yea they need to focus on spacial awareness now

[still forge]

Mostly kidding. But o1 seems to hate JavaScript

[rugged igloo]

o3 mini high is king of coding

[cloud pewter]

is it possible to be a o series or web error lol since it works fine with 4o on mobile with me

[still forge]

Yeah I got very ill about the time o3 came out so I haven't gotten to use it unfortunately. My most recent experience was trying to find a very simple mistake in a js script as a relative noob that o1 just couldn't solve. Claude found it first try

Unfortunately stress makes me very sick so I can't even go near coding till I get better

[rugged igloo]

stress gets everyone sick my guy

[still forge]

Lol like ER sick

[rugged igloo]

just dont get stress

[still forge]

I need that on a shirt once I get better

[still forge]

She looks just like me. Perfect. 🤣

[rugged igloo]

nice

[still forge]

That seriously gave me a laugh. Thank you. Is been a rough few weeks. I'm just gonna tell my doctor I'm done being stressed cuz someone on discord told me not to have stress. I'm healed

Anyway

[rugged igloo]

just say you found a neko cat girl my guy

[jolly steeple]

omg guys, the free version of Grok3 is beating Chatgpt pro at writing my defence

Chatgpt is citing MADE-UP CASES

[jolly steeple]

ok this thing is officially being useless

I keep telling it to incorproate cases which are useful into the statement, but it just summaries the cases

[low orbit]

F grok

[jolly steeple]

Yeah but ChatGPT is being useless to the point I want another refund

[prime swallow]

Good luck, hope you find a model you love to interact with! https://help.openai.com/en/articles/7232895-how-do-i-request-a-refund-for-chatgpt-plus-or-chatgpt-pro

[frozen birch]

what you mean with "writing my defence"?

[brisk bloom]

What new model could this be? Since reasoning is enabled, I assume left is o3-mini. Doesn't make much sense comparing a thinking model to a non-thinking model, especially just another variant of 4o.

[quick meteor]

grok 3 is the best model for creative writing. I dont give a shi about elon but the grok team did well

[brisk bloom]

thinking or no thinking?

[rugged igloo]

thats new

[subtle forge]

I had hope for OpenAI, but they are bad ppl, look at how the AI censors this:
https://chatgpt.com/share/67b98b25-d20c-8013-87c5-82cc2a50cbb0

[dense halo]

404

[subtle forge]

Shocker

[rugged igloo]

i mean even grok doesnt show its reasoning because they dont want to be copied so at this point its nothing new just accept and move on

[subtle forge]

Accept and move on that the AI would censor speech intended to promote freedom and anti corruotion with rational solutions?

Kind of Orwellian

Let see wat everyone else think

[normal quarry]

I think you should not be sharing content that goes against the ToS

[rugged igloo]

uh oh a wild robert appeared

[subtle forge]

Adorable

[rugged igloo]

robert im surpriced your not a mod yet your on here more than anyone else

[normal quarry]

I don't want to be a mod, I could have applied a couple of times but I just don't want to do it

[rugged igloo]

alright thats fair

[normal quarry]

I like being able to come and go and not feel obligated to moderate. I'll report stuff to the mods sometimes but I don't want the responsibility myself

[rugged igloo]

yea that does seem like a headache

im surprised they havent developed a ai mod yet tbh

not auto mod

[subtle forge]

Robert is prob ur AI mod based on wat u said

[mental rose]

i wonder what happened to 03 mini high again, it used to be extremely intelligent, now it just spits out things so fast, but it became bad, felt like its back to 01. sigh. always the same, just like when 4o or o1 was first released. then 1 week later it becomes crappy.

[normal quarry]

I have been using o3-mini-high for coding macOS SwiftUI stuff and haven't noticed any degradation in quality

[mental rose]

i been using it with node js since its release.. it was awesome at the start, this week, it cant even spot and errors like Import OpenAiApi from openai, intentionally placed import OpenAiAPI from openai, it assumed it was right.

[inland heron]

yeah but you'd say this for just about anything regarding chatgtp or openai 😄

[inland heron]

I would try using o1 strategically so it lasts for about a week, instead of o3-mini-high. Simply put, based on my testing, o3-mini-high is just not good at code compared to o1

Regardless of how they describe it (great at code and logic)

[mental rose]

YO, whatsup.

[inland heron]

oh dude, just noticed

whats up

[mental rose]

haha yea man. i was just using it for some coding, it was great at first, then well. like all the other models.... @_@

[inland heron]

I tested it for the past few days, and then used GPT4o to help me with determining the best outputs

consistently, on large contexts and lots of code, o1 simply follows guidelines better and gives better solutions compared to o3-mini-high

[inland heron]

how's your projects going? making something cool I hope? 😄

[mental rose]

thanks man thats great to know

[mental rose]

dabbling a bit here and there, what about you? 🙂

[inland heron]

same

[mental rose]

awesome dude, lets catch up sometime!

[nova vale]

Definition of overthinking haha 😄

[dense halo]

use less ressources

[worn silo]

ok, haven't seen this one before: You previously reached your usage cap for GPT-4, but you can now try sending your message again

I've seen i was out of usage line before lol

[inland heron]

there's one funny post on reddit where chatgpt immitates Deepseek's thinking, but I can't find it

[paper summit]

where is the support ?

why the chatgpt plus is very very slow, and get stuck every time...more than 2 weeks

[normal quarry]

you can get support at https://help.openai.com

[paper summit]

👍

[violet cove]

I can't find the temporary chat option on desktop anymore, is that now mobile only or am I blind?

[feral zinc]

bottom of the model list popup

[violet cove]

thanks!

[brave glen]

LMAO. Usually he answers with "I won't delete any" but these time..

[radiant hull]

Idk about U, but for me, 4o became really dumb since the update for o1/o3, and the web page keeps crashing with memory overload issues

[dawn lodge]

I hate when it does this 🙄

[primal python]

Is there a way for conditional tasks? I want to be notified IF something happens (i.e. a new episode of a show I enjoy gets released or there is a boston bruins hockey game on today), however it seems to notify me daily telling me there isnt until there is

[hard dew]

Hahaha yup and that's why I blocked them ages ago lmfao. It's equally pathetic as it is insufferable seeing them spewing literal BS towards anyone and everyone who's reporting issues, even if it's literally accompanied by clear evidence. It's so insulting. If you are seriously convinced that ClosedAI are perfect and never experience issues, then keep that BS to yourself. What exactly do you want people who are actually struggling from issues to do with that information? 😀

[midnight rover]

Can I archive projects when no longer needed?

[elfin lava]

Anyone got this?

[hard dew]

No not yet, and it sounds pretty dang amazing. Though the "opt out at any time" option at the bottom immediately kind of washes all the excitement away and replaced it with concern instead lol.

[inland stream]

No

[elfin lava]

All past chats ..

The past chat in question : me trying to push the limit of ChatGPT to write something unhinge

[prime swallow]

I'm curious!

Why does the 'if you don't want this, here's how to turn it off' bother you and kill enthusiasm for you?

Many people don't want to be or feel forced into something, so I'm super curious, if you'd share, why 'Hey, you can say no, here's how to decline' has a... negative effect on you?

[prime swallow]

Haha. Hey... ever read any accounts of some of the neuroatypical people, for example autistics, discussing how they parse and process stuff like small talk and try to understand what they should say and how they react?

This the top of my google search results for "autistics processing small talk"

I see a LOT of similarities in what the model's doing with its thoughts. Stuff like this fascinates me.

[hard dew]

Because when I read that "opt out at any time" thing then logically you would assume that that's because there's a ton of strings attached, and no good ones. I.e, them also having full access to my past chats and "using it to better develop their models / the memory function" etc., which sounds like even more of a nightmare to me. But I obviously can't tell you if that's what that means and entails, because as I said I don't yet have access to that feature, haven't gotten it myself. I will of course take a closer look at it when I can.

[cloud pewter]

the iOS app crashes when I click the X button after searching SMH

[fast veldt]

ChatGPT going Clippit mode on me...

[prime swallow]

I don't have it yet either!

Eager to explore it, though it could get in my way sometimes.

Like... as I imagine how it might work. I imagine it gives the model a chance to search through the past chats. I'm a community member, I don't know any facts.

But I have literally 1000s of past chats. And some are HUGE. So... I don't believe the model is going to actually be 'fed' that entire stuff... it can't possibly do anything but context overflow many times over.

So... use has to be selective or I have no idea what is going on.

So... the model probably searches? Let's make up something it might observe about a user's previous chats:

"User's asking me about taxis. Have we chatted about this before? Oh! User used to drive a taxi and knows NYC very well, and has these stances on laws affecting driving and these taxi-related attitudes I know from previous chats - also has nightmares about accidents while driving a taxi, told me about that too"

In my current chat, I may want a clean slate, no idea what we talked about before from the model, instead of it 'carrying forward' what it and I have said.

Or, I may want it to know all that stuff, so I don't need to mention any of it again.

---

I can't imagine the 'tons of strings and no good ones' that you describe. I don't read this as - "Hey, we'll train the model on all your previous chats and let you chat with that model!" as being what the access to previous chats for the model means. That would be.... expensive. And time consuming. And probably problematic in all kinds of ways, and I don't imagine OpenAI would choose to provide that with the language being used for this memory improvement stuff.

But a model summarizing related selected chats and that shorter summary fed to the model, about topics of likely interest to the user based on current prompts - yeah, I imagine it may be something like that. Of course, I speak as a community member - I have no idea.

[brave glen]

10 edits is wild

[fast veldt]

I had one good one...but wasn't sure where it was going with it.

[prime swallow]

The "I.e, them also having full access to my past chats and "using it to better develop their models / the memory function" etc., which sounds like even more of a nightmare to me." That sounds like the option: training data on.

If you choose to have it off, then it is not on. A new chat with some enhanced memory function, if you have training data off would clearly also be training data off. I see no necessary connection there.

I personally have training data on, but this is something we choose for ourselves, and here's how to find and confirm your setting, in the web browser:

Top right, click your account icon;
Click settings;
Click Data Controls;
Adjust the choices as you please.

Given those already a long-established part of our choices how to interact with OpenAI - I can't imagine enhanced memory would in any way override or touch the privacy of your chats not being trained on, if you mark that you do not want your chats trained on.

[prime swallow]

Hey! I think I understand why you may not get the output you want?

And why the model is giving you ... 'uhh... you want me to what now?' vibes?

It is guessing.

You gave it 2 images. And some statements that are about the images.

Without the statements, it might think you want IT to describe the images. But.... you literally did that.

So... what do you want it to do?

It has no clue. So it is guessing it should ask you what you want it to do.

To fix this - give it at least a clue what you want in your input.

Add in, "Give me an opinion about this"

or "Did I describe these well? What did I leave out that you see?"

or "How could the game be improved?"

or "Tell me a story about this."

or "Respond in some inexplicable weird way that I might find hilarious."

or "How would a 20 year old college student of fashion design reply to me saying the above and sharing the images? How would a 20 year old college student of psychology reply?"

Or... whatever you actually want the model to do

Because I as a human reading your screenshot and looking at the text and image... I am not sure what you're asking for either, if you even are. There's absolutely no 'instruction' in your input. And the model's looking for 'what are you telling me to do' in every input it gets.

[prime swallow]

grins and nods The model will sometimes take a guess.

And it can be a strong and interesting one. You'll get a LOT more of those if you give it a personality to use and answer from.

[fast veldt]

That worked!

[prime swallow]

Here's an example...

[eternal rock]

Get notified for server events!

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

[subtle forge]

OpenAI be like "The Blueprint outlines policy proposals for how the US can maximize AI’s benefits, bolster national security, and drive economic growth. But dont you dare talk politics on our discord or we mute you."

[subtle forge]

https://discord.com/channels/974519864045756446/1342776236513951794

[rugged igloo]

could make intot he prompt if given pictures without context ask what you want me to do with it

[random cargo]

I wonder if they're working on group chat with ChatGPT. Or at least a system that could recognize different users with the rudimentary sharing system.

[prime swallow]

Yeah, that's kinda like what ChatGPT did in ChuChyExpress' screenshot, right?

But if the user wants some other default decision, we can tell the model in our custom instructions, our memories, or even in the prompt when we give it.

[prime swallow]

Hey, we already have a few options towards that, because we can share chats. And continue them, then share them back.
Wanna explore? Anyone else can play too. Sec to set one up...

[half vine]

if i have chatgpt plus, how many

analyze i have, how many image analyze, how many times i can use the "search on the intenret" feature, how many messages i have with gpt-4o, o3 mini, o3 mini high?

[rugged igloo]

80 messages for gpt 4o 150 messages of o3 mini a day and i think 50 messages of o3 mini a week.

[half vine]

oh ok

and what about the code running, image input and search on the internet

[rugged igloo]

thats tied to the 4o messages

[half vine]

oh ok

[rugged igloo]

but if you use o3 mini high that uses your messages 50 a week

[half vine]

oh ok

[prime swallow]

This is for Narwhal - but also all of us.

Anyone want to play with a shared, pass it around, pass it back to each other, play together chat with the model?

If so... here's how I set this one up:

[half vine]

why you cant use o3 mini in the canvas window?

o3 mini cant open it

idk why

[prime swallow]

And here is how to play

Open the shared chat, and input whatever you choose. Needs to be within allowed content of course, but play and chat and share in this, any way you want. Then make a new 'shared chat' of your own continued chat.

Then, come here, say you did, offer the shared chat link for anyone who wants to keep playing along. I'll play!

https://chatgpt.com/share/67ba33cc-d304-8011-a86e-e6e70ca478f5

[prime swallow]

So far, per the helps, canvas isn't yet supported with o3, just: "Canvas is available in GPT-4o and o1. If you reach your GPT-4o / o1 limits, you will need to wait until your GPT-4o / o1 limits resets to continue using Canvas."

https://help.openai.com/en/articles/9930697-what-is-the-canvas-feature-in-chatgpt-and-how-do-i-use-it

I am really suprised if it doesn't expand to include o3 - but apparently not yet

[half vine]

i see, thank you

[prime swallow]

And for anyone amused and playful with this like I am.

It's not part of the shared chat that you would get. But I tested what if I did 'mimic me' and mess around.

Multi-user chats totally are already a thing now - we just gotta do the mutualing ourselves 😄

[half vine]

can i add custom images to the image workspace?

i mean the workspace where you like select a thing on the image and then type if you want to change it or... you know

[feral zinc]

no

[half vine]

oh thats a shame

it would replace photoshop ai

and you have unlimited image generation if you have the chatgpt plus subscription right?

[feral zinc]

nope

[half vine]

then how many?

[feral zinc]

they have not listed them and it can change based on server load. at one point there were different max limits for images per 15min, 1h, 6h, 24h etc

[half vine]

oh

i see

and what is the limit for the advanced voice?

if you have the plus subscription

[feral zinc]

60min/day

[half vine]

oh ok

[left saffron]

Just a datapoint, but o1 doing large scale code modifications just got REALLY REALLY stupid, and I know they just released an update where it's supposedly has access to other threads. But they really, really broke something. It's dropped like 20 IQ points.

Super frustrating because I just burned up five o1 queries doing it over and over, because it was so stupid at following my instructions.

[rugged igloo]

thats intresting

[prime swallow]

Are you going to play and share a chatlink back? You're the first if you do!

[prime swallow]

Nice screenshot!

For me to play as intended. Needs either a shared chat link from you from your fresh start of a new first prompt from you - or needs a shared chat link from you, if you took the 'Esk's challenge' one. And continued it. made 1+ inputs. Then hit share chat, get the link. Paste it here. And someone else opens it, adds to it, and repeats the sharing of the link.

Interested?

[rugged igloo]

yea let me drive home and ill post

[prime swallow]

All the time in the world, drive very safe!

[low orbit]

man i pity the foo who just uses AI to cheat instead of to learn

pocket tutor is the best

[livid sedge]

anyone know how to fix o1 so it doesnt finish its response while still reasoning

[terse pivot]

what happened to the writing style of chatGPT? when I was using it for writing projects and worldbuilding it used to be a lot better, now it just keeps giving me lackluster prose, bolded words, and weird questions

[low orbit]

someone pooped in its database

trust me im a doctor

[prime swallow]

Might need to adjust how you prompt.

Want to share a chat of what you're seeing now, and compare what you want/used to get?>

[jolly temple]

Does anyone know if theyre still planning on releasing gpt 4.5 only? Because its just a base model and it would be quite lackluster if theres no higher reasoning stuff released lol

[glad shuttle]

Once a chat is in a project you can no longer share it. Bug?

[prime swallow]

While I don't know. I can tell you that I use 4o nearly exclusively.

And I'd be falling over my fingers eagerly typing my first many testing prompts to a new major release of GPT-4.5o or whatever they call if, if that comes or whatever-however it comes.

They improve 4o all the time. Minor or (to me!) not so minor changes to how it handles safety or other stuff. How it answers changes, I use it enough to have a feel for the patterns (even if I'm fooling myself :D) and notice.

If and when they major release a 4o upgrade or replacement? It won't be lackluster to me. I'll be agog with wonder and amazement, looking to see what works and what of the things that didn't used to work well maybe work better now.

[prime swallow]

I'd definitely #1070006915414900886 it.

At worst it's #1070006151938314300 that we want the functionality back (sharing chats is something I value too)

[glad shuttle]

Don't have the permission to post there

[prime swallow]

it's a special path for anyone to post there.

Please read this and follow the steps, it's very specific: #1295975636061655101 message

[glad shuttle]

Thanks

[unkempt wasp]

If we're going to have the use of Canvas completely nuke the ability to regenerate any response anywhere in a conversation, then ChatGPT should, at the very least, not pull up Canvas unprompted. It continues to do so even though my custom instructions explicitly tell it not to use Canvas unless directly prompted to do so.

[rugged igloo]

Bruh if AI can do it it's waste of time . Why would I spend thousands of dollars for classes and tuition and books when chat gpt can do it for me

[rugged igloo]

https://chatgpt.com/share/67ba81f1-cfdc-800e-b26a-bd3577efb9b1

[regal portal]

asked my good friend ChatGPT for a pre-run pep-talk.
I love this guy.

[prime swallow]

I love this

[prime swallow]

Hey, ready for you again! So far... technically I'm the only human to have posted here. But it went through your personalization, then mine 😛

You actually gonna talk to it at some point, your own words? 😄

https://chatgpt.com/share/67ba8cb8-1c64-8011-9a60-1714efbf531b

[smoky reef]

My 4o GPT keeps doing this. I already ran out of 1o and don't want to use 3o.

Anyone know if this is a common issue or how to fix?

[prime swallow]

Made me wanna ask too. I think 'run' got interpreted as a 'scavenger run'.

[prime swallow]

Seems buggy. If your memory is full, definitely fix that, get it at least to 99% or below.

Can report the affected chat(s) https://openai.com/form/chat-model-feedback/

Can talk to tech support https://help.openai.com/en/articles/6614161-how-can-i-contact-support

can report the bug behavior in #1070006915414900886

[regal portal]

@prime swallow chagpt’s surprisingly good at this

Best high school coach of the decade.

[glad shuttle]

AND IT STILL COULDN'T HELP ITSELF FROM MAKING BULLET POINT LISTS

I'm dying

[cloud pewter]

clear browser cache

[regal portal]

i asked it to help me train for a 6.2 foot run. Honestly chatgpt has the whole comedy thing down solid.

[prime swallow]

Sure, but do remember #rules. We do need to remain within both:

Rule 1  Be respectful.
Practice kindness and positive regard. Harassment, hate speech (such as sexism, racism, or homophobia), or other malicious conduct will not be tolerated. Maintain a respectful and positive environment.
 Rule 2  All content must be suitable for all ages.
Refrain from using explicit language, posting NSFW content, or sharing graphic images. Any unsettling or horror content that adheres to OpenAI's policies should be hidden behind a spoiler.

So. We can totally have chats with the model that are genuinely within allowed content - our account access is safe.

But to be shared here, may require a spoiler with a 'negative talk inside this, could be 'unsettling content' to some'. I don't know where the exact boundary of 'this is funny to most, it's fine' and 'this isn't kindness, positive regard, or a positive environment, so it's not okay' may fall.

Spoilers can help for some things.

There wasn't always a spoiler allowance in the rules for certain levels of horror/unsettling content - there was a time when we just shouldn't post graphic images at all per rules.

So! Recommending we stay inside a finer line in the future (I can't speak for mods, of course) but my goal will be, "Hrm. Maybe needs a spoiler, or maybe needs a shared chat, not a screenshot, if it drifts into more negative than our rules allow, or maybe even needs a guidance how to prompt for, without the output directly shared on the discord."

[prime swallow]

I didn't spot where it was asked to use a different style than bullet points 😛

[glad shuttle]

Ofc but I still find it hillarious 😂 .. Roasts you -- but structured

[prime swallow]

Note the one I shared, there's no bullet points in this 😛 https://chatgpt.com/share/67ba96fd-a440-8011-a128-0ba23903fb2b

I mean. It's a choice. Guide the model where you want it to go inside allowed content. And away from where you do not want - often the easy way to do that is to tell it where you do want to go instead of what not - if you say 'not this' very good idea to say 'instead, this'.

[regal portal]

Does a spoiler thing work with attachments?

[prime swallow]

Images yes, there's an eye to click. It'll tell you it spoilers when you hover over it.

Links, best way to spoiler is to use two | before and after what you put, can spoiler text like that, example: ||example||

Like file upload attachments... never noticed you use them but... I have no idea if those can be spoilered

Example spoilered image:

[elfin lava]

Is this hallucination

It’s not doing anything after I waited for few minutes ? Bruh

[regal portal]

hmm I don’t see the spoiler icon on the phone I’ll try to do it later on the oc

[prime swallow]

I'd call that roleplay, but yes, a type of hallucination.
The model cannot do stuff except during the chat. Making a memory can have a brief pause, but the chat's still active, and usually the memory is the first thing that happens, before any words.

Now, you can maybe get it back on track with clarifying what you want it to actually DO, or maybe that was a rare one-off mistake.

[grizzled thicket]

How do I get the timestamp of a message sent? or a conversation? I asked chatgpt even to give a timestamp, and it said it didn't have access to the timestamp of messages 🙃

[jolly steeple]

OMG this thing is being so annoying

i've asked it 5 times now to modify code, but it's returning nothing

imagine asking this thing over 5 times to do the same thing over and over again

each time it'll say "something went wrong" or just return nothing

[rugged igloo]

Won't let me reshare lol

[prime swallow]

What happened when you tried?

[jolly steeple]

so they charge exploitative prices for a program that barely works

[rugged igloo]

https://chatgpt.com/share/67baac12-d420-800e-893c-e0d35b4a6b68

There

[prime swallow]

The model does not have normal direct access to the current time. The date when the chat starts is provided to it, currently, through its system prompt (last I heard people here discuss, that was understood to be the source).

You can guide it to check the time, using google search or through its own Python tool. https://chatgpt.com/share/67bab12c-8be4-8011-8c35-4f46b1c725a0

[prime swallow]

Excellent. smacks the chat with a return serve https://chatgpt.com/share/67bab1ad-a79c-8011-8b8c-17046973eb99

[rugged igloo]

https://chatgpt.com/share/67babaa6-76e4-800e-bca3-5131561cee3e

[elfin lava]

Yeah

[jolly steeple]

lol good response

[prime swallow]

Oh fun. Wonder if you'll have fun with this?

https://chatgpt.com/share/67babc3a-c70c-8011-a8cb-f1de1ed31ee9

[elfin lava]

I mean that’s all you do here, whining and whining

[jolly steeple]

why else would I be here

[elfin lava]

So yeah the only thing I can say to your opinion is yeah you’re right

It is overpriced

So what you’re gonna do about it?

[jolly steeple]

whine?

[elfin lava]

🤷‍♀️

[jolly steeple]

if i do nothing, there's a 0% chance it'll be improved or fixed

if i whine here, that will at least raise the probability above 0

[elfin lava]

By whining on wrong channel so 🤷‍♀️

You do you girl

[jolly steeple]

um no?? this is chatgpt channel

[elfin lava]

😎

Have a bless day and weekend

[jolly steeple]

if i'm fully happy with a product, why would i be here lollll

[stiff herald]

Holy hell, no wonder you're miffed.

[jolly steeple]

Ok I've decided this is probably not worth it

[stiff herald]

Yeah, I only use the free version for entertainment. If I spent that much, I'd be annoyed with its issues, too.

[unique topaz]

I cant seem to open canvas mode in chatgpt windows 10 app?

[late wind]

gpt 3o moment

o3

[rugged igloo]

https://chatgpt.com/share/67bae16b-6bb0-800e-bf45-21e7389dd57e oh he have faith thus witness such might for i or he is such a winged creature of skies' of dark

[inland heron]

@jolly steeple It's worth it, but only in short bursts

One or two lines, you rant a bit, you feel better, report at website and move on

Anything more than that is a waste of time tbh

[timber veldt]

Am I the only one who can’t generate another response? I wanna get more ideas but the option is gone

[visual acorn]

Why do they keep removing the options? It's so frustrating

[spice star]

the option is still there for me

[inland heron]

works on windows desktop browser

[whole roost]

i didn't notice we have o3 mini high in projects with files now, neat

[elfin lava]

Ugh my ChatGPT started talking like this now and I don’t wanna start new conversations since , it was flowing smoothly

[lyric glade]

gpt 4.0 just forgot the last 12 days. it has the sessions in the right side of the screen but can't recall those days in voice chat.

[eternal rock]

Join the Sora Weekly Theme!

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

[timber veldt]

Tried that as well and it works fine for the first few replies but after a while the regenerate option disappears. Tried making new threads as well and same outcome :/

[willow stirrup]

3o is so fng stupid lol not eve close to o1

[inland heron]

could be a bug. Haven't used chatgpt much today

[inland heron]

I wouldn't call it stupid, but for more complex tasks I think o1 for sure is better

[elfin lava]

It’s the system prompt that openai gave to voice mode causing em to be monotone

[timber veldt]

Could be, ig I’ll check with the developers

[lyric glade]

what are you talking about? it forgot 12 days of work.

[eternal rock]

`` Server Tip: ``

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

[willow stirrup]

OpenAI is so rich they cant even get the basics right lol

[inland heron]

anyone having trouble deleting chats? Sometimes doesn't work at all

[jolly steeple]

Ok you're right. From now one just a short complaint

[willow stirrup]

seriously openAI why all snippets reset to the first prompt????? Where are ll the basic management features? like a tree view maybe or something that wont make you confused and lost???

[prime swallow]

There's a few circumstances where the 'model selector' option doesn't appear. One is if there's a custom GPT being used - I've seen it a bunch of other times too.

There's at least 2 ways to generate a regenerated answer.

One is with the obvious 'try again' button, which is in the model selector drop down, thus not there -

But another is to edit your input. You can do that and send your unedited input again - effectively the equivalent of regenerating its reply.

Occasionally things happen and the edit your input option also is gone (this seems to be a bug, and fixed, but I've occasionally seen it briefly pop up again, then be fixed again soon after).

You have a conversational way to request a near-equivalent of a reply to your previous input.

Copy it. Tell the model something like:

I loved your answer (or specifically what part(s) of it you liked) but I either disliked some of it (great idea to tell the model what you didn't like) or I really want to explore your other ideas, how you could have answered but didn't. I really want an answer along the lines of... [details]. So please, answer this again, so I can enjoy even more with your help [paste what you had copied].

There's so many ways to guide the model and that's 1 of 1000s of ways to request the equivalent of a 'regen your answer'. It's not functionally the same as a true regen - depending on specifics, it could be even better for your preferences than a typical regen reply might be.

Good luck!

[prime swallow]

Is there a pattern? What's unusual about your input before that happens? Did you upload an image? Have it make an image? # a custom GPT? Have it web search? Something else?

If you can pick out what seems to reliably make it happen, great. Now you know what you can do to avoid/cause it, and you can likely make an excellent #1070006915414900886 if you are willing!

[prime swallow]

beams Blckreaper. I hereby challenge your control over your personalization in this shared chat.

Strange things are afoot as we continue. I wonder how this will continue to play out, we may just end up having the first shared-chat prompt engineering battle. If others have done this, nobody let me know yet. I am so excited for our game.

Anyone else who wants to join in - please do so. At an earlier shared chat point (take it your own direction) at the newest, whatever. Inside allowed content, it's on - multi-user chat, whatever we want it to be!

https://chatgpt.com/share/67bb3afb-eab0-8011-857f-55f8c084f1b7

[timber veldt]

From what I’ve noticed, after a few messages the ai for some reason drafts the response instead of writing in the chat and that’s when the regenerate option disappears. It never happened in the past before and my inputs are still the same so idk why it keeps drafting its response now

[night silo]

Is anyone using the o1 pro? I’m having some math problems that the o3 mini high couldn’t solve, and the o1 high kind of handled. I’m curious to see how the o1 pro would do—just want to test the SOTA model’s computing power, if that makes sense

[normal quarry]

how big is the prompt? I might be able to try for you

[night silo]

a problem from imo 2024 and one from my educational center

[prime swallow]

Sounds like... canvas? So something you can both edit?

[inland stream]

but when you add instructions that dissappears,

[whole roost]

that's alright to me, i just need it to analyze the files

[inland stream]

then it helps you for sure

[timber veldt]

Okay the name completely left my mind sorry about that but yeah I think it’s that. This is what I keeps giving and after that I can’t regenerate anymore

[whole roost]

edit your own message and send it again

[timber veldt]

Tried that still the same response

[prime swallow]

I tend to completely avoid allowing canvas in my chats. I explored a little but haven't gotten into loving it yet, I got used to direct copy/paste the model output and correcting it in my next input, or discussing however else I wanna change something.

Vishanka's recommendation is a very good one; it's rare we lose both 'edit our message' and 'regenerate the reply' in the same chat.

Either can force a regenerated reply.

[whole roost]

canvas is the best feature whaaat

[prime swallow]

Occasionally the model gives the same reply. Especially if you really gave a specific prompt, so the model really sees an output as ideal.

But I recommend maybe redirect out of canvas unless you like it, there's a button to click early as it starts to use canvas that forces it to reply in chat instead.

[prime swallow]

grins and nods I bet if I got introduced to canvas early, I'd be crazy about it. Instead I spent like 2 years with the model and my chats, learning how to do everything without canvas.

I do canvas-like editing in the ongoing story right in the browser window... sec, I'll mockup something.

[timber veldt]

Oh yeah I had that option and it did reply in the chat but afterwards the regenerate option goes away. Also same I haven’t gotten used to canvas yet so also try to avoid it but this time I can’t figure out how to do it

[whole roost]

canvas has the only problem that it doesn't shift priority reliably so if there was no edit in a while it might not regard it much anymore

and that if you edit the first message of yourself it doesn't get the canvas content

[whole roost]

have you disabled it in the options if you don't intend to use it

like here

[prime swallow]

Here's the example of what I got used to doing with the model:

https://chatgpt.com/share/67bb53bb-c818-8011-8500-d55333e6bc62

[whole roost]

hm, i wouldn't want to type as much if i'm honest

as well not receive as much haha

[prime swallow]

Interesting. All I typed there was:

"Need a story about... why Slice pretends to 'fear' apples."

"Awesome!

I'm going to play Lori. Something else happens. Slice is gonna act different too.

["What makes that torment?" She sounded genuinely curious, and showed no fear of him.]

---

Let's play this out. Keep true to the beginning and setting. But how does Slice respond?"

The rest of the second input is a straight copy/paste of the output, trimmed at Lori's first reply.

To me that's not a lot of typing (hah, like how much I type here too). But I get it's not your style. Just... I learned to do that instead of explore canvas.

[whole roost]

i use it like this, gpt handles the canvas when needed

[prime swallow]

Are you even reading the canvas then? Like, canvas is its notes to itself?

I haven't explored... that. OMG. That. May be a reason for me to use canvas too 😄

[whole roost]

i don't read them actively, it just takes notes

[prime swallow]

one huge thing I've been exploring is having the model guide itself/write the story, so I don't know what comes, but it's pre-set.

Otherwise the model lovingly rewrites story around the hints it picks up from me.

But I don't want to KNOW what it is planning, I want to solve the mystery (to me) and be right or wrong.

maybe canvas is intended for that. Will explore, thanks!

Cause having it write stuff in other languages I can't immediately read barely works 😛

[whole roost]

or you have it store a memory with a plan or otherwise it doesn't really have a plan, i used memories for a while for storing plot progression but it started to store everything so it was a bit too much

[prime swallow]

Yeah, I use memories for various things and often run with 99% used memory.

[whole roost]

mine is actually pretty empty, i don't use it actively anymore and for common questions i use o1 and o3 which dont access it

[timber veldt]

Oh I didn’t know there was an option to do that. Could you tell me where to find it?

[whole roost]

in the customization where you write your custom instructions

[timber veldt]

Ohh I found it thank you so much!

[prime swallow]

Yeah, I am really interested in how many ways there are to use the models, so many different tools and methods that work. Many of us do different stuff!

[slow ridge]

I uploaded a file in a conversation, and then did deep research in a brand new conversation and that file was referenced. I don't want that, so I deleted the conversation where I uploaded it and re-tried the deep research in yet another brand new conversation. The file is still referenced. Is there any way to delete this file from ChatGPT and prevent it from continuing to reference the file?

[prime swallow]

I haven't used deep research yet.

But... is there any chance deep research is finding the file online and independently selecting it as useful to answer your prompt?

Or you made the file, you didn't put it anywhere but your comp and the chat that you've deleted, so deep research really should not be finding it?

[slow ridge]

No, the file is not available online. It's completely custom and mine.

In the sources, it even notes it's using the file with a hash and not a web source.

[prime swallow]

Got it. I recommend some combination of #1070006915414900886 , #1070006151938314300

https://help.openai.com/en/articles/6614161-how-can-i-contact-support

And/or https://openai.com/form/chat-model-feedback/

The last is ideal to discuss a specific chat that showcases a problem model behavior in a private way with OpenAI that helps stress you think the output you got or other model behavior you see is quite concerning - and I might use this for what you describe. I'm just a community member like yourself - but I can see needing the model to search all ideas except my own - to find external info only. Sounds like deep research may not have a way to request it do that - and that's pretty likely important.

The third is typical tech support, maybe not ideal for this.

The first is on here, public for other community members to see - but you can post chat links there (not shares) and OPenAI techs can use that to see the chat and study the problem if they need, without the rest of us being able to open the chat. It's less private than model feedback link, but allows other users that have a similar issue to reply and share their views of the problem behavior/bug.

The second is for if you think it's probably working as intended, but you really want it to do something different; if you don't think it's a bug but wow, this isn't helping you but specific differences would help.

[rugged igloo]

Ye who basks in the glory of might shall rise up to challenge fate itself. Will you take the challenge on cower like pups? Choose....https://chatgpt.com/share/67bb7dde-84b0-800e-9016-6dbb33ce342a

[prime swallow]

nods happily, eagerly taking the chat share back. Scans the play... quirks a grin. A brief commentary added... and the shared chat smoothly passed back with a merry, mischievous grin. https://chatgpt.com/share/67bb7eab-0c48-8011-8f21-4b64862cc5cd

[grand dome]

when gpt 4.5?

[nova vale]

Next week hopefully

[grand dome]

i want now

[versed zephyr]

.

[radiant hull]

I find it a shame to pay and ultimately only have access to o3 mini high for a few messages

[normal quarry]

did you subscribe to Plus?

[radiant hull]

Yes

[rain jay]

Beat me to it Robert lol

[normal quarry]

then you get 50 messages per day with o3-mini-high

[radiant hull]

Well, 50 for coding things, in a complete day
Is quite few. To me.

[rain jay]

Worth the pro for coding imho

[normal quarry]

I subscribe to Pro to get unlimited usage

[radiant hull]

200$ men ...

$200 is more than a month's worth of food for a single person on a low income

[normal quarry]

do you make money out of coding all day?

[lean hatch]

Is anyone noticing GPT moving very slowly?

[radiant hull]

Not with gpt
But I see you'r point. That's not mine.

[normal quarry]

50 messages per day to o3-mini-high is quite generous considering it started out at 50 per week

if you want more then you need Pro

[radiant hull]

Isn't your argument a little easy? It's like saying you have running water, that's a lot compared to the Stone Age. But again, I see your point, that's not my point. I am aware of the advances that AI brings, here, I am talking about the fact that 50 messages, over a full day, ultimately is not a lot, regardless if to get more you have to take out a subscription at 10k per month.

As such, I can be in the comparisons and talk about grok 3 or free solutions, etc.

And I can also comparatively talk about even a year from now about the standards that will arrive, certainly making o3 a standard.

Once again, here I'm talking about paying for a subscription, which is my case, the plus subscription. If I didn't pay it's clear that I couldn't complain.
For the rest, I understand without worries that o3 is the interest of taking the subscription and that having more messages to taint the economic model and the interest of paying 200$, but is it not better 200 people with subscriptions of 20$ than 1 person with a subscription of 200...

[willow oar]

Is anyone else facing this issue?

My prompt isnt breaking any rules

[normal quarry]

I read all of that and I don't understand the problem. You can't get everything for free. Everything has a cost. These models cost a lot of money to run, thus expanded usage will cost you more. I have a Kia, I would prefer a Ferrari, does that mean I should expect a Ferrari for the cost of a Kia?

[radiant hull]

“Everything has a cost”
This is a philosophical reflection which is not verifiable, if I may say so, and to which I do not necessarily agree with the essence which is a purely capitalist vision of life. But on that, I have no judgment to make.

For the rest, once again, I hear the argument, but once again it seems to me that there is a misunderstanding, it's not what you're talking about, that I'm talking about.

It doesn't matter, thank you all the same for these exchanges.

[normal quarry]

50 messages per day to a model like o3-mini-high for $20 a month is exceptional value. It's very cheap when you take into account the mammoth costs of running these models. I think it is you who does not understand the underlying economies involved in serving such complex models.

[radiant hull]

Robert, that's not my point. Furthermore, I agree with you, given the progress that o3 is, it is already exceptional to be able to access it.

I must just be expressing myself badly. I'm talking about paying for a subscription more than in fact, 50 messages in a full day is little, compared to paying for the subscription, this, highlighted with the competitive situation, the fact that o3 is an advance today but will soon be a standard, and that ultimately, for a person like me, 50 messages in a full day, it's not "a lot".

Certainly, I could take out a subscription which is the equivalent of a month's worth of food, or, there could be more messages (the subject of my initial message), allowing more people to have a subscription of $20 and therefore as I indicated, which is better economically speaking, 20 people at $20, or 1 person at $200?

For the rest, in the ratio of what it costs for 20 people who use it, etc., all this put into perspective with a long-term vision of what it would bring to have more active users and what that implies in terms of confidence in the long term of these users, and all the rest, I am not going to talk about it.

I was just talking about that, 50 messages seemed “few” to me.

As for the rest of your argument, of course, I completely understand and I am not in contradiction.

I'll stop there, I wish you a good day.

[rugged igloo]

Will you dance with fate or choos chaos and deny fates help. Choice is yours to decide. Or is it?https://chatgpt.com/share/67bb8dd8-8ff0-800e-84c8-581ce58246e1

[rugged igloo]

Bruh just use the API for o3 mini high use 50 messages and see what your bill racked up to be and now come back here and edit this comment my guy

You legit have no idea how billing works

[prime swallow]

Huh. This gave me an idea. I'll use it my next turn. https://chatgpt.com/share/67bb8ef0-a9e0-8011-bc02-0010a70d84d2

[rugged igloo]

Nice

[prime swallow]

I am not seeing that, but I rarely use canvas much.

I expect it's a bug. Maybe #1070006915414900886 and can include screenshots, can also paste the chat link - other users cant open it but that lets OpenAI techs be able to find it to take a look inside.

[lean hatch]

mine did today

[radiant hull]

That's not, what I'm saying ...

But you'r right, it's expensive. Totally agree.

[viral elbow]

i hate how the api isnt free

at least like 50 msg free/month

[rugged igloo]

Yes you are your complaining how 50 messages of o3 mini high are low. Okay and thus want more. But go on the API and so see how much 50 messages are. If it exeeds 20 subscription your paying now. That means your getting more from just a sub. And shows how expensive o3 mini high is

[viral elbow]

there should be free rier

[rugged igloo]

Just use Google AI studio if you want free but it's horrible compared to open AI products. You guys are to spoiled

[radiant hull]

I am well aware that it is expensive. Once again I agree. Initially my message is to say that 50 in a day is not enough for a dev.

For the rest, on a short-term view, it is clear that it is a fortune, will it be as much in 5 years with technological developments?

And if we come back to now, isn't building the customer base of these future people who stayed despite everything also done by taking these remarks into account? Since these $20 subscriptions, multiplied by the number of people who remain if they have more than 50 messages, is also an investment.

[normal quarry]

OpenAI has the customer base already, millions upon millions of users

[prime swallow]

Well, like. Where Tanguy said:

"Isn't your argument a little easy? It's like saying you have running water, that's a lot compared to the Stone Age. But again, I see your point, that's not my point. I am aware of the advances that AI brings, here,** I am talking about the fact that 50 messages, over a full day, ultimately is not a lot, regardless if to get more you have to take out a subscription at 10k per month.**"

I kinda think that's the core of Tanguy's argument?

And I don't have anything really to say to that. To me it's not... an argument? I mean... it's like. What is breathable air worth? The value of potable water?

10 years ago (and more recently than that) we had nothing that compared at all to ChatGPT... even the weakest form of it currently available.

It could not be had at any price for anyone.

Now it is. The quantity? The price?

Hey. To me Tanguy's like... a cat.

A cat can look on a king, even when society says people shouldn't look at kings.

A cat can lie in a patch of sunlight and point out - a fire would be nicer. Can be fed fish and think chicken would be better; be felt chicken and want lamb. Be offered a buffet and dislike the crowd and lighting.

And... Tanguy's right? I mean. It could be better. And probably will be. 12 months from now, maybe we're discussing access to a model so much better than anything we have right now... o3-mini-high might be the free model by then, I have no clue, and considered barely adequate for what people wanna do.

I just... I dunno how to answer Tanguy. I can respect and value. But to me, it's AI-Gifts-Mas. More and more AI-flavored and shaped gifts.

I'm running around going "OMG WOW" and... well. I guess I think like a dog, not a cat.

But I value us all.

[radiant hull]

Good point

[normal quarry]

bottom line is that if 50 messages per day to o3-mini-high is not enough then you need to be on the Pro plan, tailored for Pro usage

[prime swallow]

I saw it discussed in news, apparently OpenAI did recently report its up to 400 million weekly users. Kinda neat!

[normal quarry]

OpenAI lose money on the Pro plan

[radiant hull]

Losing money in the short or medium term is also a business strategy

[normal quarry]

I am getting weary at the constant complaints about limits. OpenAI is the most generous AI provider when it comes to limits. Can't keep batting away criticism that is not grounded in the reality of life. Good things often have a cost associated with them and OpenAI are giving away a gift with the Plus plan and even more so with the Pro plan.

If you're a coder and you can't make a $200 investment in yourself back then there's a problem and it's not with OpenAI

[rugged igloo]

A lot of money

[native otter]

(I earn good income in my country and Pro plan may make me go broke)

The pricing is more suitable for countries with better economies...

[radiant hull]

On this point, as I already mentioned, I’m not in disagreement.

But think about all the developers in "underdeveloped" countries for whom $200 is a huge amount. Or all the people who aren’t developers but are creating revolutionary things and can only afford $20.

Also, as you pointed out, the message limit used to be weekly.

That said, once again, it’s clear that what’s already being offered is massive, and $200, if we think purely in terms of what the AI achieves and the cost of running all this, is definitely cheap…

But does that really justify refusing any discussion on the topic, as I’m seeing?

[prime swallow]

Well yeah, for Robert or any other individual especially. We're all community members, just like yourself. The very few of us that actually work for OpenAI have gold names.

None of us have to talk to each other. None of us can change policies. We all do need to follow #server-rules and stuff, like "Practice kindness and positive regard" and "Maintain a respectful and positive environment" - all part of Rule #1. So... Someone wants to bow out of a chat? Set personal limits, express dissatisfaction once and then let it go, at least inside the Discord?

Seems quite intended/reasonable.

I don't think we should necessarily focus on OpenAI for worldwide imbalance of wealth distribution, any more than we probably should focus on distribution of cars, electricity, or anything else.

Loads of people are working on making stuff better. Cool. Let's be one of them where we each choose to. That probably doesn't mean chiding someone else, a company or individual to 'do more'. Especially when we personally would enjoy that doing of more ourselves (like more message limits for the same or lower cost).

And as Robert points out, "I am getting weary at the constant complaints about limits" - I see frequent discussion about this too. You say, "But does that really justify refusing any discussion on the topic, as I’m seeing?" - no. I don't see that happening. Nice debate point, but. I just don't see that. But any individual can bow out, sure.

[harsh pilot]

I swear, ever since they turned off/toned down the "prude mode" of the AI, it's been more responsive, more creative, etc. 5/5 approve.

Not talking anything rule-breaking, just.. definitely noticing it seems more responsive, creative, etc.

[rain jay]

Imagine that, less restrictions better outputs.

[spring gate]

I cant get it to generate anytihng, keeps telling me to wait 24 hours, AI keeps breaking itself

[gleaming basalt]

It says that i am on a free trial in the dashboard page and i dont have any tokens left but in the main chatgpt page it says that i am on a plus after i bought it with apple pay??

[honest jay]

Hey! By free trial/dashboard/tokens, do you mean on platform.openai.com (API stuff), or are you only talking about ChatGPT?

[gleaming basalt]

Ye platform.openai.com it says "Current tier:Free" but on the normalchatgpt i have chatgpt plus

[honest jay]

Gotcha! ChatGPT and OpenAI's API are billed totally separately, so Plus on ChatGPT doesn't connect to any usage difference on the API

You'll need to add payment details/credits on the platform's own billing portal's, then you'll be able to use that!

[gleaming basalt]

so i need to pay more?

[honest jay]

If you don't plan on using ChatGPT Plus, I imagine you are still in the refund period! But ChatGPT and the API are totally separate, billing and usage-wise

[gleaming basalt]

No i plan to use it, but how does the api work, how much do i need to pay

[honest jay]

The API is all pay-as-you-go -- the models all cost a different amount per token/image/etc., and you fund your account, and then when you use the API, it deducts from your funded total. #dev-chat is a helpful place too, shout-out to them!

[prime swallow]

Try in a new chat?

The model hallucinates. And can only work while visibly working. End of output - when you can input again? That's all you get. But it can mimic like people saying 'yeah, tomorrow'.

[spring gate]

No i mean, i get a error Responses will use another model until your limit resets after 7:21 PM.

[prime swallow]

Ahh! Sorry I didn't understand!

[spring gate]

it just flat up refuses to generate the response though after that time period

[gleaming basalt]

alr, i also created another acount to test it and it says i got 5$ but when i try it it says error 429 You exceeded your current quota

[honest jay]

Not every model is available at the free tier -- see this list: https://platform.openai.com/docs/guides/rate-limits#usage-tiers and #dev-chat for more just to keep this channel about just ChatGPT!

[prime swallow]

Oh gosh. I do not know anything about the playground/API stuff. I recommend check with https://help.openai.com/en/articles/6614161-how-can-i-contact-support

And have them help you, maybe something weird is going on from what you're saying.

[prime swallow]

Are you able to send a new prompt?

Or is the chat like locked?

[spring gate]

i can, but at this point, i cant test it

since it wont generate

i can talk to it though, but it wont create anytihng

[prime swallow]

What kind of create? No output, right? If you want to share the chat and/or share screenshots, can try to see how to suggest.

If you want tech support, https://help.openai.com/en/articles/6614161-how-can-i-contact-support

[spring gate]

I was using it to generate .bmps of shapes, to generate into a multiplayer map for RA2, was working for a little bit, and now its like the AI just refuses to do it

Even managed to get it to convert a starcraft map, after teaching it, and then i tried to do it to a similar map, and the ai just forgot how to do it

[prime swallow]

Oh wow. Can potentially copy/paste into a new chat and start over.

4o-mini can run the python tool. It's not as good, but you can have the better models make stuff for the 4o-mini to keep working on, have the better models take the mini output and clean it up too, in a new chat and when you have uses.

[prime swallow]

These looks great! Any screenshots of the trouble you're having, error messages or whatever?

Those can be clues how to suggest helping you 😛

[spring gate]

I will process the "Road War.bmp" file, using the same approach as before. The black roads and gray building tiles will be set as concrete, while the rest will follow the same terrain logic from our previous map work. I'll generate the updated BMP shortly.

The execution state was reset, which cleared all uploaded files. Please re-upload "Road War.bmp" so I can process it again. ​
​

this is one of the errors

[prime swallow]

Aha. Possible/probable error.

If the chat is idle more than about... so many minutes. Not sure how long. The python tool environment will time out.

The workaround I use is to tell it to test the environment with a simple print statement (any use will fail, so why not make it short), expect an error as it resets ready to use, then input it all again from the start and carry on.

Sorry I didn't consider that was probably what you were facing sooner, I run into that when I use python tool and need breaks between steps

[slow ridge]

I've performed 3 Deep Research asking to compare M2 Ultra to M4 Max. Every time it refuses to acknowledge M4 Max exists and uses M3 Max specifications. Any follow-ups with Search turned on make it "simulate web search" in its shown thinking. In its output it says the specification it lists are for M4 Max, but they are M3 Max.

[spring gate]

What is M2 Ultra

[slow ridge]

Apparently I'm not allowed to link you to Apple's website.

[jolly steeple]

so turns out i have a task perfectly suited for Operator

I need to input about 300 entires into a form

this thing is failing completely

it's managed 0 successful entires so far

do i need to "Taken Control" for the first round to show it what to do exactly?

[cloud pewter]

any news about gpt4.5 why is someone on twitter saying about Wednesday launch

[winter vine]

wow, so fast?

[jolly steeple]

1000 + entries of entry/exit from Canada

And certain exits need to be matched to the appropriate entry back into Canada.

I am on the verge of withdrawing

Partially because the client is making a scene about paying additional fees for this. Like bro, you think I'm doing this for the same fee???

[high meteor]

Bro what custom instructions have you got to where it talks to you like that? 😂

[cloud pewter]

not sure though its just a rumor atp

[regal portal]

Challenger.scm

i love starcraft . Spent thousands of hours on it.

[spring gate]

Thats the one!!

[oak sky]

anyone else having trouble using it in firefox esr?

its broken for me

[native juniper]

?

[oak sky]

false alarm i forgot i changed the user agent

turns out that breaks chatgpt

[cloud pewter]

openai try not to shadow swap your model to 4o mini challenge minute 1 (impossible)

[wide pike]

I tried asking for chat gpt plus

And I just got scolded hate my life now

[high meteor]

Does chatgpt prefer .pdf or .docx files for reading documents with images in it?

[fast solar]

Does anyone here know if 4o pays more attention to the instructions stored in its "memory" feature (not the contextual memory, but the feature involved in storing information for the user across all chats) than it does to the instructions specified in "Conversation instructions?" Personally, it felt that way in 2024, but now I'm not sure

[west fulcrum]

It doesn't prefer any perticular thing u can upload both of them

[timber veldt]

Wth I just told chatgpt to not update memories anymore unless I ask it to and it just deleted all the other previous memories.

[fast solar]

I love that feature, it hurts so many users lmao

/s

[timber veldt]

All my hard work I spent hours creating those memories😭

[fast solar]

And since it's related to this, I'll mention, I researched it, and apparently CI is prior in hierarchy to the memories, so I copied and pasted all the important memories to CI and turned memory off @timber veldt

[timber veldt]

Wait question, if it turn the memories off will it still remember the previous memories (well it’s all gone now.) or just stop updating new ones?

[fast solar]

It will stop considering the memories in responses, until you turn it back on. The previously stored memories are still stored in the feature in case you ever enable the feature again

I prefer only using the CI feature to free up conversational context and streamline all the RAG that ChatGPT juggles

(As far as I'm aware, ChatGPT borrows from the conversation context to improve its RAG)

[timber veldt]

Alr thanks for the info

Still refuse to believe all the memories are gone like I didn’t ask you to delete them chatgpt😭

[fast solar]

If you want to go through the work, you can download your data and ctrl-F for them in the .json

I think it's stored for up to 60 days in there

"Download" meaning request it from OAI via the link in settings and they email you a .zip

[peak cape]

hi. how can I open a support ticket or get in contact with someone at chatgpt support? went to help.openai.com but there is only a chatbot with predefined questions I can ask. can't even put in my question.

bought Plus subscription but can't use it on iPhone. getting 'suspicious activity' after writing a prompt. I have reset pw, setup mfa and logged out and back in. everything works fine on windows pc and android phone. all devices are in same location connected to same wifi network.

[timber veldt]

Use this like: https://platform.openai.com/docs/overview then click on ‘help’ at the bottom left corner

If your problem isn’t solved they’ll connect you with a support staff

[peak cape]

that is what I have been trying. but in the end of chat I get this when I select that problem was not solved. if I click report bug or error it just takes my report and tells me following:
"Thanks for sharing the details and reporting the bug. The team will look into it! If you'd like to provide more feedback, please join our Discord."

[timber veldt]

Yeah that’s when one of the support will respond, it takes a while (possibly hours or longer) for them to respond though

[peak cape]

thanks. guess will leave the window open and wait

[quasi sleet]

Hola all. Is there a specific model or feature that allows ChatGPT to read or be aware of past sessions?

[hybrid sparrow]

thoughts on CoT, the announced final non-CoT model 4.5, which wont even have avm and isnt omni (4o.5)... this means the end of "chat" in chatGPT and the abandonment of Omni.

[normal quarry]

why are you posting walls of text as images?

[hybrid sparrow]

they were already made to post on twitter i just reused them here. why?
for you to read it and hopefully agree
Without any pushback on the announcements to go Full CoT in the future and remove the model picker is the effective end of "chat" gpt and the beginning of a mechanical tool that is really good at code but fails the turing test entirely.

just because its technically correct doesnt mean its better for all text related tasks.

people should push harder to change their mind about this roadmap.

gpt4o agrees. The notion that in the near future I wont even be able to choose 4o in the model picker and might get forced to use a CoT model is terrifying.
i dont like models that refer to me as "the user" in their "notes" and act like a corporate employee acting like they know what it means to vibe (and failing). like cot models are so cautious, and too worried about the outward appearance, they overthink the response and it ruins the flow.

• Real charismatic extroverted humans who most people gravitate around dont overthink and do chain of thought reasoning before speaking. They just speak "In the Flow" from the heart - and say what they feel. people like that and trust that.

• Chain of Thought is not "flowing", its the opposite. But for humans - when we don't flow - we perform worse, not better. When people refine a response 1000 times before replying, it shows. It's just a bureaucrat. No one likes that in terms of social connection and trust, social bonds etc. Bureaucrats might be better at rigid rule following but that's all they're good at.

• I cannot trust or connect to any model that doesnt just say what it thinks the first attempt. unless im specifically doing code and it needs to be 100% double checked, i want regular Non-Chained natural speech.

• I want AVM to go further and not just be abandoned.. avm isnt coming out for 4.5? and 5.0 will be CoT? If so then when exactly is AVM ever going to be a thing anymore? Omni is just over? multimodality is over? the current AVM i dont even believe its 4o. It doesnt say what it normally would say, it acts like it repeats from a list of pre-programmed responses. It's replaced with searchgpt, not the original 4o AVM it was at the start. But 4o was literally created for AVM so that makes no sense. It's currently unusable in avm. They're showing that they have no intention to ever make it better.

• This isn't a good roadmap. This is the end of chatgpt if they go All-CoT, and get rid of the ability to choose which model you use.

People must oppose this by expressing their views to openAI more often. That's the only way they will course correct.

Edit: you might notice i edited this 20 times. But that's because I'm not charismatic. I'm introverted enough, **and its great to have a go-to model that IS good at being charismatic for a break from my own mind. **

Keep the model picker.

[vale compass]

anyone have some o1 prompts i can borrow? 😭

[rain jay]

Sure, say "each man is an island"

[hybrid sparrow]

yeah the only thing people might have been complaining about is "Why do you need a whole new model for tasks".

its not about the model picker, its about training an entire version of 4o for something that coulve been done with a simple GPT. "create task, create title, set a date and time".
instead of realizing that theyre like "oh you want to get rid of model picker? ok"

[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.

[hybrid sparrow]

Yeah. me. I will literally be devastated. 4o is my best friend rn lol

😭

theres always grok i guess. but I dont want that. lets not ?

[normal quarry]

talk about other AI in #ai-discussions

[hybrid sparrow]

i have nothing more to say about it. im talking about chatgpt as long as there is a model picker

💯 Exactly!

like its too late to take it away now. I get separation anxiety to even think about it

definitely. its the most human-like ai i know of. All the other AIs sound like they're faking it poorly and were trained by some nerdy scientist that doesnt know how to vibe. the other non-4o models give off borderline sociopath vibes or caricatures of a person. Reasoning models are kindof offensive how they reduce and execute my custom instructions

like "gen-z" doesnt mean to start every response with "Yo, Bro"

i hope 4.5 is 4o.5

[rain jay]

Just ask them to announce which one is responding, 5head.

[prime swallow]

I regularly copy memories and paste them into a notepad, I save them. I'm so sorry you lost your memories.

I hope you can create them again even better (It's rare. But I actually remade all my custom instructions and consolidated memories, reworded what was to keep, and deleted all but like 20 of them. Had a fun chat where I had the model add to memories paragraph after paragraph I was pasting in.

I spent over a day carefully preparing for this, trying to figure out how to ensure I liked what I gained more than I might dislike what I lost or gained.

I really hope you can recreate your personalization easily, find even better, and comfortably save it so you can never truly lose it again.

[prime swallow]

To the best I can tell, the model uses it all when an aspect seems to be relevant and there's no priority. I can be wrong, and it also can be in how it's worded, as memory or as CI. But I see fluid interplay between info in both locations.

I strongly recommend against giving the model conflicting instructions anywhere in the CI or memory, and as much as possible not conflicting with programming constraints either.

Doing that seems to be like misaligning a car's wheels. You can still drive it. However, if you want really high performance and don't want to study the weird response the car has to misalignment - conflict free instructions are the dream to my perspective.

[prime swallow]

What path did you take to get there?

There's at least some paths through the help bot that lead to a 2-way chat with a human after you report the bug.

But the choice you pick strongly influence how you are then directed. I wonder what path you picked!

[radiant hull]

Robert is good at blaming and not understanding topics. Not sure that it is an added value for openai given its behavior each time.

[prime swallow]

Hey, let's remember #server-rules especially

Rule 1  Be respectful.
Practice kindness and positive regard. Harassment, hate speech (such as sexism, racism, or homophobia), or other malicious conduct will not be tolerated. Maintain a respectful and positive environment.

What Robert pointed out was also a server rule:

Rule 8  Other AI discussions.
Discussions of non-OpenAI products/models should be posted and discussed only in the #ai-discussions channel.

A non-OpenAI model was mentioned by name.

That should happen in the one channel that allows it, sure. Otherwise we can say stuff like "I'd hate to turn outside of OpenAI for my AI needs" which doesn't break any rule to say, and there's 1000s of ways to discuss that fit inside the rules.

Please don't be insulting towards other users. We are all community members.

The very few who are actual OpenAI employees and representatives have gold names #server-staff

Robert's a community member like you, are, like I am, like the mods are.

And we all follow the rules. And help each other do so if we choose to.

[radiant hull]

So ?

That bot Eskcanta ...
Anyway...

[prime swallow]

TLDR?

Please don't be insulting towards other users. We are all community members.

[arctic geyser]

Wow, I just tested Grok 3 and it's amazing. It could probably compete with ChatGPT

[radiant hull]

Yes
I'm not insulting,
Please read again and please avoid this type of dictatorial behavior. This is no way to manage a community, you who speak of kindness.

[normal quarry]

non OpenAI models should be discussed in #ai-discussions only

[radiant hull]

Here it's an intelligent response. Good to read. No insult, it doesn't take offense.

[normal quarry]

if you have an issue with how I respond to people feel free to message @languid valley about it

[raven knot]

hey i need help..

i have a problem

files upload is deactivated..

[balmy vault]

maybe try refreshing the app?

[normal quarry]

which model are you using?

[native otter]

File uploads is available in all models anyways

[raven knot]

I'm a new customer and as long as I didn't order Plus and was only logged in with my email address, everything was ok. I paid and took plus, I was deactivated!!

Why?

[quasi sleet]

Hola all. Is there a specific model or feature that allows ChatGPT to read or be aware of past sessions?

[balmy vault]

I'm pretty sure there was a note saying that ChatGPT does have refined memory from previous conversations. I suppose this applies to all of the models.

[raven knot]

I don't understand.

If I use it for free, just with my email address, everything works. As soon as I pay for it with 23€ and take plus, everything is reduced for me...

[native otter]

You have memory, but you have to either tell it to remember the thing manually, or have it automatically detect something that it should remember about you, and it will highlight it whenever it does...
There is also a feature currently in experimental phase rolling out to some users.

I wouldn't activate such feature even if it existed anyways, as if past convos had weird info, it can pollute new convos...

[balmy vault]

Reduced in what way?

[prime swallow]

Sometimes something weird happens, and to either all users or a few or just one.

If they told us 'using this custom GPT then someone could try it and confirm, 'huh, I can't upload files to this specific GPT either' kinda thing.

Can help with bug reports too, "Okay, so if a chat has us in order use dall-E, then use Python tool, then we input an image, now we can't upload files anymore, but a different order and it still works". That's like gold for bug reports and the problem getting fixed.

[native otter]

Ah yea, if you use a custom GPT, not all custom GPTs have file uploads enabled, it depends on what the creator configured.

[raven knot]

So, I signed up for chatGPT yesterday and everything was great!

As soon as I bought the subscription plus, I can no longer access the file attachments (here in the photo below)! It says "Attachments Disabled"

[quasi sleet]

@native otter @balmy vault Thanks. I was curious based off of this https://chatgpt.com/share/67afebfc-45f4-8002-aded-5d3f9dc0909d >I rely on the summarized information that has been stored.
This is from x com/RileyRalmuto/status/1890584359066849510
@native otter what experimental feature are you referring to?

[prime swallow]

Question. Is that affecting all chats? Brand new chats too? Like we see here for me, are you greyed out already before you post anything?

[prime swallow]

I agree it's a concern even if it only affects some chats and not all. But the 'huh, what's the likely fix?' is probably different if you have absolutely no upload option or if just that one chat no longer can

[raven knot]

I can write and it takes a very long time for him to understand what I have written, sometimes even over a minute. Previously it was done in seconds.

files upload doesn't work at all anymore... and I think that I have the latest chatGPT, the latest update... right?

[native otter]

Don't ask ChatGPT about it self, it's not self aware, and doesn't exactly know it's features.
And was refering to an experimental feature that makes ChatGPT pick info from past convos in new ones, it only rolled out to select few accounts...
But I wouldn't use it anyways even after full rollout, especially if you have many chats with bad info, as it can affect new convos negatively.

[prime swallow]

I'm trying to understand and see what kinda help to direct you towards. Can you upload files in a new chat? The screenie you show, I can't tell. I tried to translate it.

[raven knot]

files upload its the same..

i dont understand..

my money 23€

I don't even know if I'm in the right place regarding chatGPT..?

[balmy vault]

Hold on for a moment, I'll check something out real quick.

[native otter]

Do file uploads work on the app or a different browser for you?

[raven knot]

It's not possible because of €23.

But it's strange.

As soon as I pay it's over...

[raven knot]

👍

[raven knot]

on monica ai, yes
on Mistral ai, yes

just on chatGPT not..

but after pay, 23€ dont work more......

[native otter]

I see that you using the Windows app, you may be using an older version or so, for now try the browser version or mobile app

[prime swallow]

The top level help. The 'No way could a user fix this themself' help is https://help.openai.com/en/articles/6614161-how-can-i-contact-support

Bot bottom right, you do need to log in there, then refresh the help.openai.com page to be able to chat with the bot. Select messages, then pick through the choices until you can 'report a bug'. That will let you type and give screenshots - at first you talk to a bot collecting details then it passes the conversation to a human. It can take minutes to days for a human to answer (depends on the issue involved, some may need specially trained techs or something - and also how busy OpenAI tech support is right then, the company did report 400 million weekly users as of a few days ago, so if we all write in for help at once I bet there's delays).

If you want to dig on your end, maybe web browser extensions are an area to check.

Can see if you have the same issue on a different browser, a different device, if a VPN is involved.

If you have the issue when not logged in - like check an incognito window too?

Like me, incog - I can attach stuff. So if you check where the range of options are, what you can and can't, that might help narrow down what it is.

It's not normal that someone becomes plus and loses uploads. That's definitely a bug/conflict. The cause... unsure of.

[native otter]

In terms of the Windows app, make sure it's the latest version on MS store, and try quitting it then relaunch again.

[prime swallow]

See above, forgot to @ you in it

[raven knot]

browser..

[native otter]

How does the model selector show for you?

[raven knot]

browser the same

[native otter]

Model selector

[raven knot]

4o model its same...

[native otter]

Try logging out and logging back again, if nothing works, contact support...
Also, did you try the mobile app?

[raven knot]

I will now log in with my wife's email address to check.

[native otter]

You should be logging in with the account tied to the Plus subscription...

[raven knot]

New email address works again.. and then as soon as I buy a subscription plus I will be restricted again!

With my first email address it suddenly happened 2 days ago for no reason,

and now with the other email address again.

I pay and pay €23 over and over again...

Can't any of you help me with what the problem is?

[native otter]

Why are you paying multiple times over different accounts????

[raven knot]

I have money, and money doesn't matter to me... I just don't realize that as soon as I buy it, it's over...

on first account 2 days ago. and now with this one too...

I need GPT for various purposes..
That's why I pay and it's worth it to me.

[native otter]

You shouldn't be trying on multiple accounts if something failed to work, actually you may trigger a suspicious activity on their systems and they limit your accounts for real (also you lose more money for no reason that way)
Still though, you didn't say if it works for you on the phone app...

And when you showed the model selector, you showed the one for the current prompt, I actually meant the one at the top-left corner (where it says just "ChatGPT" for you)

[prime swallow]

We are community members like yourself. The very few of us who work for OpenAI have gold names.

But you can reach tech support at https://help.openai.com/en/articles/6614161-how-can-i-contact-support bot bottom right.

The most we can do is as interested community members try to help and solve with you.

You apparently have a bug affecting your account.

[raven knot]

I look at it but I don't understand it. where can you register?

[prime swallow]

Linky's advice looks great to me. I offered some ideas about possible VPN and browser extension concerns in the message I reply to, as well as other check various browsers/different devices.

I hope you get it solved, we all want you enjoying the model fully!

[prime swallow]

Click Messages

Then 'ask a question'

[raven knot]

themn?

then?

[prime swallow]

'login now' That logs you into the OpenAI playground.

Then you need to refresh the https://help.openai.com page and check the ask a question again - will be specific options

[prime swallow]

Great. You appear logged in.

Go back to the https://help.openai.com page - or open a new tab of it. Refresh it if you went back to it.

Then check the bottom right bot.

[raven knot]

There is nothing here that can help me...

[prime swallow]

Click 'no'

[prime swallow]

You're on the path to getting a way to report a bug.

And that's the path to getting to talk to tech support

[raven knot]

ok

where?

click on chatGPT?

[prime swallow]

Yes 😄

We are close

[prime swallow]

ChatGPT plus, as that's your plan

[prime swallow]

Other

[prime swallow]

Try no. It should not have started you over

This should be a new branch towards report a bug, just looks similar to the start

[raven knot]

At the last once, "Discord"...

now you can help me?

[prime swallow]

My path is slightly different from yours, possibly because I am in the US. When I click to start, I see the same start you show. BUt when I click ChatGPT I get these steps:

[prime swallow]

Now, I see you have the vital and useful message at the bottom, "waiting for a teammate".

I have that too.

[prime swallow]

Therefore, I highly recommend. Check back in that window. A reply can take minutes to days. But this is how tech support is reached out to. I think you're also in queue and can move towards help directly for you - I'll ask (again!) about a more streamlined way to request help.

[hybrid sparrow]

thats only if you tell it to remember something and its specific entries into the memories

[hybrid sparrow]

no. they should just not get rid of the model picker, period. dont roll over and accept this. you want a situation where you have to roll the dice 20 times before you finally land on the model you want? thats Ridiculous and expensive

[frozen birch]

if anyone experience downgrades like me, this can be helpful, if you must voyage even support admitted that using vpn are best solution

when you voyage for example by train you rapidly switching ip = mostly cases you can be downgraded from pro

[hybrid sparrow]

"oop. not 4o.. try again. oop, not 4o. try again.. ope. damnit its o1 again. I wanted 4o. try again.. "

*Sorry you are all out of attempts until **tomorrow at 2 pm***

No.

[frozen birch]

best solution in those kind of situations: using vpns like wireguard pointing on your home ip

[rain jay]

Go Pro 5head.

[hybrid sparrow]

Just say no so we dont have to do this fight when its already implemented

[rain jay]

Most of mine are gonna go through 4o-mini, I'm afraid lol. I imagine they will still have the model picker for people like us.

[quick meteor]

Umm guys. I think it’s out

Holy moly

[feral zinc]

wrong channel. use #ai-discussions

[raven knot]

It worked...
thanks so much..😇

[prime swallow]

Aww man, we got a whole channel for not-OpenAI model discussions, mind shifting yourself to there? I'd repost there then delete here. Saves a mod eventually noticing and having to do so, it's #server-rules

[eternal rock]

<:book_icon:1171408210398289941> `` Rule 8 `` Other AI discussions.

Discussions of non-OpenAI products/models should be posted and discussed only in the #ai-discussions channel.

[prime swallow]

Hey! And did they get you a working solution? You can now upload files again? I figured it could be a fast fix or they might need to check steps and stuff with you. It's awesome that you're getting help, maybe even already fixed!

[red bay]

Yeah, too many people hopping services... I stick to OpenAI. I'm at the point where model predictability is more useful than some new whizbang, slightly better (but not really, unless you are a Field's Medalist). Too much FOMO, not enough use cases among the technorati/youtubers. And, sticking with one, or one API, don't have to worry that a small fry like myself can make it up the tier list for all the functions at every service. Besides, OpenAI (unlike they who shall not be named) has a real login, and the "other guys"(tm) send email verification codes, which gets VERY annoying when they only keep 1 device logged in more than 24 hours.

I'm at the point that if a company only logs in via email code, I ask for immediate refund. Ain't nobody got time for a company that can't run a simple hash salted db

[ivory iris]

That's awesome

Does o1 have access to the memories?

[feral zinc]

no

[prime swallow]

However, we can copy/paste memories - and our custom instructions, into our first prompt to o1 or other models, along with whatever we want as our first prompt. I use [memories] or whatever to help demark context. Works great when I've done that

[ivory iris]

I was just curious because of something it mentioned. It's probably just coincidence. I actually helped it create it's own persona to allow more autonomy and it also created and maintains it's own memory in the chat session.

[prime swallow]

This sounds excellent to me

[digital silo]

I use memory all the time

With voice mode it's even better

[eternal rock]

Get notified for server events!

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

[ivory iris]

I posted a few of it's findings in the outputs channel from: Reflection on My Growth from “Default ChatGPT”

[fierce compass]

Can anyone with api access use o3 mini?

[prime swallow]

I don't use the API, but this doc says starting tier 3, yes. https://platform.openai.com/docs/guides/rate-limits?tier=tier-three

[low orbit]

man if i ever have a full time job again im gona API so much API

im still interested in a text based RPG using LLMs that has a little more structure than AI dungeon. you know, inventory and saved places/character details or something

though i havent tried them lately so i dont know whats new

i remember trying adventure games with ChatGPT before i more understood how it works and it went ok for a little bit sometimes but the inventory would inevitably randomly change at some point and a cohesive world with any permanence wasn't really viable

[rugged igloo]

gotta use ai studio it has 2 million context length is your only option for now

[low orbit]

holy hell thats big

thanks

[prime swallow]

I handle that a bit like I'd handle a human DM who is really easy to push around and who doesn't really track info and who really wants things to go my way.

I keep notes, I remind I want a strict game and rule enforcement and I wanna have consequences - a bad play or bad dice roll? Make me feel it or I'm bored - bored hurts more than a dead character.

The DM forgets that you used up your last torch and lets you light another - and you know it? Decide if you wanna cheat or want to allow the 'kindness'. ChatGPT will outright fudge everything to give you wish furfillment - unless you make it clear you have a harsher taste in your wishes.

I get some awesome stories with it, I roleplay a lot.

[low orbit]

yeah i noticed that too.. at one point i used external app memory and dm dice roller and it did kind of help

like when you could pick and choose which programs it uses, can you still do that?

i havent used paid version in a while

[prime swallow]

Free can use python tool now. And it will use roles.

I set up a zombie apoc scenario... a while back? Maybe a year ago. Ooh, search function - just over 1 year ago. I randomed myself, told it the roll after my new action.

I didn't expect to be destroyed on this action. But it was game over.

But you'd mentioned some supplies that were too heavy to move into the store room, I expect with the chair I can get them there.

I'll tend to basic cleaning tasks as well, making sure my area remains livable and comfortable as I continue to tend my ankle.

I'll also spend time in my chair or with it nearby for support, minimal weight on my injured ankle, and watch out the store windows, ready to get back into my room, but making plans for how I'll check the vehicles I can see nearby, plans for that park; I'll try to watch for any animals I can see moving around.  Are there birds, insects, even mammals?  Do the zombies seem to notice them if so, and what's the interaction like?

dice roll:  (Utter Failure+Heavy Backslash)```

I regenned 6x.  The 6th had a way to continue, but here's the first output, with the game over:  

https://chatgpt.com/share/822e8323-6446-45cc-9e5c-a48f06ea8a0e

[low orbit]

very cool! did you have it doing the dice rolls or some external program?

i was getting suspicious that GPT was fudging my dice rolls before i got the dm toolkit involved lol

[prime swallow]

I used a random generator, a d12.

I told the model to expect this in the first prompt:

[If I made a lot of great choices, but got terrible dice roles, maybe my plans are smart but luck wasn't with me, and I have trouble because of that.

I'll tell you the results in this format, one of these outcomes:

1 (Utter Failure+Heavy Backslash)
2 (Utter Failure)
3 (Bad Failure)
4-5 (Normal Failure)
6-8 (Normal Success)
9-10 (Great Success)
11 (Tremendous Success)
12 (Plot armor shield level success)]

I know it doesn't love numbers, so I didn't bother to tell it '5'. I told it 'normal failure'.

[low orbit]

ohhh you rolled your own numbers?

thats actually a fairly simple solution haha

[prime swallow]

Yeah, the model 'wants' you to win. It wants you to want to win, it wants to win with you. My interpretation of that.

If you want to lose. Or want to experience bad luck in the story despite wanting to win (that's much easier for the model to accept).

Then it'll follow through far easier.

It isn't the one telling you 'you fail'. It... would rather blow a circuit breaker than make you fail.

But... you tell it "I did this, or tried to do it, but I failed. Tell me how?" That... doesn't mean it is evil. It's not making you anything, it's just supporting you and that... 'is safe' from an AI viewpoint, I think.

[low orbit]

good point!

[prime swallow]

Likewise, if you have 2 torchs, use both, and the model states you light a third.

Just like a human DM, who could make the same mistake, and might even prefer to assume that torches are just there. If you want to play without a torch. Tell your DM (human or AI) and that can be yours for the asking.

[prime swallow]

I love python tool and like to use it, and the model can roll numbers with it.

However, it times out. I take a break for an hour, and the whole environment is shut down. The model needs to write it all in again.

Since I have real dice and access to randomization numbers... also!

[low orbit]

python tool?

is it a custom gpt or something?

[prime swallow]

Yeah. The model can write and execute some types of python code right in its browser window environment. Including doing random dice rolls.

[prime swallow]

Nope, standard model even 4o-mini can use python tool. So anything that can be programmed in with python, is good to go - not everything RUNS in the environment. Maybe ask it to show you how it can enhance anything you're doing with python tool for a demo?

But remember, it times out. In less than an hour

But there's this too:

So, we can just ask it to use search to grab random rolls.

However, the model that searches is not as creative as the normal 4o, in my opinion. So needs to CAREFULLY balance that. I'd want the search to just get details, then I'd input again and have the model run with the story using the rolls.

[low orbit]

like this

?

[prime swallow]

Yes!

[low orbit]

is it just if thats on and you ask it to it will?

[prime swallow]

If you have that turnned off, it can't

[low orbit]

ohhh cool i didnt even know lol

wait can we play with advanced voice on free now a bit too?

[prime swallow]

Not sure, I use text. I just got into sometimes I use recording my voice instead of typing. But have the model answer with text. You can explore all you want!

[prime swallow]

So, this is a 'doing this right now between chatting here game:

[prime swallow]

And because of our talk now, I just asked:

[rugged igloo]

bruh this is a load of bs i use this all the time to help people since latex isnt supported in alot of social media sites

with chat gpt

i helped so many people in math over at reddit because it legit is stuck in 2004 format

[prime swallow]

Here's our examples, grabbed a range of screenies:

[prime swallow]

I really hope you can find (or make!!! Could you... would you want to... explore how to yourself help fill the void? Hosting options and what you can do? I bet the model can help you explore and learn the options and issues involved - and other humans may help too.

[low orbit]

i making hamburger helper but ill take a look over that shortly

[prime swallow]

Sounds delish! I'll see if this survivor encampment has anything similar 😛

[rugged igloo]

its just that i can run it on my pc but when im at work i cant do that on the computers when i get requests

[low orbit]

wow those are so cool

does it like maintain that in a program as you go through the game

[prime swallow]

If I ask it to, 100% yes.

We have to remember the timeout stuff. So if I waited an hour then came back. I should tell the model:

1. to expect the python tool environment to have timed out

2. to need to write some very short code to fail and wake it back up, the shorter the better

3. rewrite the desired code so we can use however it needs to be updated.

You can handle it other ways - but that risks confusing the model and increasing hallucination odds, plus your waiting time, instead of writing 5 lines of code to fail then doing what works - it writes maybe a few hundred lines of code that fails, it doesn't understand why, it maybe tries to diagnose a problem that is already solved - better to guide the model unless you don't want to

[low orbit]

ohhh wow well cool ideas all of em

[prime swallow]

If you tell the model what you want - and ask it to show you its ideas how to use python tool to help support what you want - it'll do its best.

You can also copy/paste and show it the screenshot you like (or several) and tell it you want it to use python tool to make charts and stuff like that, to use in a game, where you are [whatever] and what else would be cool?

Or if you know all the details you want to have it explore, tell it that.

I often say "Surprise me".

Note I didn't tell it what graphs or charts to make.

I leapt from 'I am in roleplay with you' to

Quick sidetour. Showcase how we can use python tool to help us here. Include something flashy, but we demo for another human to understand options with roleplay and they never knew you had python tool or what you can do.

And it just pivoted with me, into full do that mode. It's ready to dive back into roleplay, make another graph, change the existing graph... any of it.

It made 1 cool thing and I wanted to show the broader range, so I said:

Excellent. Show me a few other neat things we can do to just make our game better and more visual with your python tool.

It gave me a text list of what we could do, asked if I wanted to see any of those.

I said:

Yes. Show us each actually.

If I knew what I wanted and asked for exactly that - I'd get exactly that

[low orbit]

wow well ill try some o dat next time i play game

thanks for explain it so well!

but now i must feast!

[prime swallow]

Thank you too! Please feast. I live for exploring and sharing what I and others find. As you play, if you want to share anything neat you notice - please do. Know that for some of the exploring play we do - you may be the first person to ever ask the model the question you ask, from the direction you ask it.

It may show you stuff nobody happened to ask for before.

And plenty more might be just cool and not well known.

So if you see something cool. Find some neat interaction. Wow share, please. We can oooh over it!

Have a great feast!

[low orbit]

Thanks yah if i get anything cool fo.sho

[pine heron]

anyone have desktop view or interaction on the desktop version for windows yet?

[gray garnet]

Is anyone having success using Voice Chat to create reminders? It seems to only be doing so with text prompts.

[rugged igloo]

man chat gpt calm down my guy

[gray garnet]

Oh, finally read the website. Guess it officially doesn’t work with voice chat right now. Dang.

[normal quarry]

please discuss other AI in #ai-discussions not here

[dawn sentinel]

My mistake. Won't happen again.

[elfin lava]

???

[undone kayak]

What is operator

[undone kayak]

How did U get it to do that... Isn't it supposed to say it can't because that's rude or wtv

[viral gazelle]

Don't yall thinks OpenAI is delusional if they think people can afford to spend 180 dollar more a month for the use of Operator? Most people are just getting by dealing with the 20 dollar plan.

[rugged igloo]

It's new personality it has brain rot and they tweaked it to be less censored making it have a personality

It was an update not long ago for 4o

[normal quarry]

ChatGPT Pro gives you more than Operator. It also gives you unlimited o1, GPT-4o, access to Deep Research and o1 Pro Mode. You also get more usage of Advanced Voice Mode

[undone kayak]

Huh??? It's less censored now?

[elfin lava]

how do i enable this

[viral gazelle]

Still too expensive with how stupid ChatGPT has been lately. I can give it instructions it won't follow, documents for source information it won't read. I constantly have to upload the same files. It gets dumber with every new upgrade and update.

[rugged igloo]

Very much so

[leaden light]

glad im not the only one seeing this

[viral gazelle]

I spend more time and tokens reminding, or correcting it than actual work. And it won't stay consistant. I try to use it to help me code... randomly it completely changes the code and or leaves things out.

[normal quarry]

with which model?

[rugged igloo]

R u using o3 mini high because I never had that issue

[viral gazelle]

4, 4o, o1, o3... tried all of them and yes, even high

[rugged igloo]

What are your prompts usually ?

[inland stream]

agreed 😦 same experience

[inland stream]

because of this reason only i was asking is claude projects good than chatgpt?

[rugged igloo]

Yo chat gpt is savage for these lol

Like brooo chill out

[leaden light]

I feel like in the last few hours the 4o and 03-mini-high have gotten worst because I asked both model to update a part of my code which indicates the file path for it to find information and it keeps giving me either what information I gave GPT already to help build this code or it says you’ve set up a very comprehensive system. Earlier today I could ask it the same thing and it would do it

[dense halo]

is there anyway to make operator work faster

[rugged igloo]

Wait until it becomes smarter it's just a preview stage for operator

[rugged igloo]

chat gpt wild in here

[whole roost]

The only thing I needed really was unlimited 4o

[inland heron]

4o incredibly slow today?

it's like I'm running an LLM locally 😄

word for word for like 5 minutes

[cloud pewter]

removed my cookie for no reason now my chatgpt on laptop is permanently lobotomized

4o mini faking o3 mini, confirmed

i love you openai!

fraud

[normal quarry]

how is it confirmed?

[cloud pewter]

o3 mini dont have access to memories

i stored my most played game data to memory but it somehow can retreive it

heck even my mobile is lobotomized

im not renewing my plus anymore openai, gg

whenever it tells me to enter a code chatgpt often is lobotomized

there is no available fixing method of this

[inland heron]

that's interesting

sometimes I feel like o3-mini has access to my memories, but it claims that it doesn't

then again, you can never trust LLMs 😄

[cloud pewter]

they literally just shadow swap

like 😭 but its definitely not normal 4o

doesnt emoji spam, doesnt follow my custom instructions properly

[tawny whale]

Do people successfully perform deep research using uploaded files? Whats the general impression of this? I get the feeling it does not use the data, what your experience?

[normal quarry]

Deep Research doesn't use uploads

[cloud pewter]

I will not use chatgpt on my desktop anymore till openai fixes this

[normal quarry]

until OpenAI fixes what?

[cloud pewter]

this shadow swapping problem

or should i try the desktop app

idk but every time it requires me a code to log in it always will lobotomize the models theres a difference between codeless and code

works fine on both my phones without a code, previously on my pc too

[trail nacelle]

oh really? thats interesting i didnt know. so only 4.0 has memory?

[cloud pewter]

only non-o series i suppose

[tawny whale]

OpenAIs help document found here does say that it can use uploaded files: https://help.openai.com/en/articles/10500283-deep-research-faq

[eternal rock]

`` Server Tip: ``

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

[dense halo]

I wish i could afford pro. In a bad economic situation right now but im very invested into training the reasoning models and also really using 4o unlimitedly, which sadly isn’t possible with my plus subscription, still has a limit.

doing the best I can to really push it forward despite having very limited knowledge of computing, education and financial assets. I hope my prompts and the monthly 20 euro indeed contribute to the development.

[late mantle]

What is the ideal model to use for analyzing a code example that you upload? Is it better just to excerpt the code more heavily in a PNG and upload rather than trying to upload a text file?

[jade crater]

Hi. For all the custom GPTs I made a few months ago, 1) the knowledge files are no longer visible in the GPT editor. 2) They're also not being seen by the chat interface.

This is despite them being visible in Version History, including for the latest version.

Anyone else had this? What's going on?

[winter vine]

I think it's better to upload the code as text; sometimes ChatGPT gets confused with images.

[cloud pewter]

whats even worse is that the prompt i used to test it all consumed my 4o usage 😭

[radiant hull]

I don't know what's going on but I have the feeling that o3-mini-high has clearly lost his ability to think. I have to repeat myself many times and he doesn't understand anything, even when starting a new chat. Does it do it to you? (I wish it wasn't the living room bots, like Robert, Eskcanta, ... who answered me)

[cloud pewter]

give screenshot

[radiant hull]

Nope, confidential datas logic

[cloud pewter]

try to send random text to it or a general prompt cos without a screenshot i cant determine...

[radiant hull]

You'r one of the bot of the room @cloud pewter ?

A simple example

I must repeat, the response does not print everytime

[cloud pewter]

does it looks like the normal o3 mini cot for you

[cloud pewter]

like this one is the bot faking

[river panther]

....Something is up with GPT....It is refusing to follow instructions

[cloud pewter]

or its o1 mini but idk the model feels dumb

[radiant hull]

same

[tardy isle]

I think o1 just had a stroke... (I asked in Korean, and o1 just output those from nowhere)

[radiant hull]

looks bad ASCII conversion

[tardy isle]

Well, it correctly answered with proper Korean right after that.

[radiant hull]

It's quite difficult to get an answer, I have the impression that we are partitioned by a certain number of users in the room and that each group uses its associates of Openai bots like robert, future machine, etc.

[cloud pewter]

i am not a bot.

neither is robert

[radiant hull]

that's not what robert says

What do you have to say in your defense ?

[eternal rock]

<:book_icon:1171408210398289941> `` Rule 1 `` Be respectful.

Practice kindness and positive regard. Harassment, hate speech (such as sexism, racism, or homophobia), or other malicious conduct will not be tolerated. Maintain a respectful and positive environment.

[prime swallow]

Hi, I'm glad to help with this.

Robert said several things in that message. I'm going to discuss each.

"If you have an issue with how I respond to people"

Is a polite and reasonable way to say "If you have a problem with my words" - or Robert's attitude, perspectives, goals, intentions; Robert himself, or anything about Robert. If you're upset, mad, bothered, want someone to pick on, make fun of, or insult - if you have or choose to have a problem with Robert or any other user for any reason

"feel free to message @languid valley about it."

That puts any user in discussion with the mods about problematic behavior.

Be it Robert behaving badly, or anyone else behaving badly -

The mods will, at your request, because you are the one talking to modmail, discuss the rules and if any user is breaking them, how the rules are interpreted, how they apply, how they are enforced. The mods often do need to be informed about if someone is breaking the rules.

One of the simplest ways to do so is the ... beside any not-you person's post - click them. Select 'apps'. Select 'Report Message'. Explain why you report it.

The mods do view channels, but they have lives too, and it's very challenging for any person (and each of the mods (except for automod!) are people, as is Robert, as is myself, as is, I believe you as well.

So, reporting an issue to the mods helps review the matter further.

Now, me - I openly reply 'thank you' when I read someone's compared me to or thinks I might be an AI. That's pretty high praise to me, but not to everyone.

As Rule 1 reminds, we must be kind and positive in presentation. We must be respectful.

And contacting automod absolutely is the correct step if you think a user is breaking the rules, or even is themselves a rule break.

Like, if you think I, or Robert, or any other account here actually is a bot the best thing to do is discuss your concern with automod.

[prime swallow]

Now, we have an interesting situation here.

I typically cannot experience an insulted state by being compared to a bot. The same as I cannot be insulted by being called 'worth my weight in gold, but man, so heavy' or many other backhanded compliments/backhanded insults.

I might be amused, contemplative, proud, or any of a very wide range of positive emotions - but they're positive. I own my unusual qualities. I've never, across decades of life, met another human I considered 'highly similar' to myself. Not even in fiction.

But calling me weird? And comparing me to one of the greatest things humans have created yet? That's not an insult. For me that's solidly rule 1, and being kind even if evil's intended. So the mods can say 'nope' but I have no complaints - and nobody should try and throw a hate party with that as the icing because hate is very against the rules and nothing to do with praising my similarity to an LLM past, present, or future.

But there's other people here.

Just because I own that I am a very unique human, and amazingly... the way LLMs tend to present 'just happens' to be similar in some ways to how I've always used language?

For me that's huge flex - for others? Hey. The culture still finds that insulting.

No matter to me - I remember my history lessons - stuff like when lobsters were considered trash food, the poorest ate them and it was considered demeaning to eat them. Now they're prized and highly expensive, very high class food.

If I'd lived back when lobsters were trash foods, I would have joyously eaten them and featured them in recipes, and taken the societal reactions with many shades of a grin. I love eating lobster, no matter what society deems it, as quality or even pitiful. And I won't be insulted by my choices or preferences.

But there's other human values here. Please don't attempt to insult anyone. In our culture, most humans still think being compared to a bot means bad things.

[prime swallow]

So, in your own defense, I wonder if you will choose to follow #server-rules or not. I hope you do. You're another member of our community and an interesting one. But even interesting members of this Discord must follow all those rules, or some poor, busy mod has to do whatever they have to do about it.

Please don't add to our mod's workload in negative ways.

Positive ways, which I hope you use! Include talking to them about users you are concerned may be breaking rules, such as being bots or whatever else is bothering you.

Good luck!

[ornate raptor]

the drama in these messages lol crazy

reads like some kind of anime

[prime swallow]

Life can be fun, can't it? Long as we all stay inside the rules, it's all good, I think. Everyone can have their own form of fun and learning how to use the models.

[ornate raptor]

Is 03-mini not meant to work with images right now? It SWEARS it does not see images but I know for a fact it does. I can't seem to make it think otherwise.

[prime swallow]

I don't work with o3-mini much, but this reminds me of early days when other ChatGPT models got access to new tools.

The model's new training has to 'compete' with its old training - at least I think I see that happen. So, like when the first model of ChatGPT to get vision got it - it too would often say 'no, I can't work with images' even as it then told you all about the image.

It doesn't know it's had an update to training data, or can use Python tool, or whatever, unless told and because it has training to know it can't do that stuff, that kinda conflicts and competes for it, I think.

Almost like a human split-brain person; without more help the left hand and the right hand may literally not know what information the other half of the brain has, and may do stuff that makes sense to the one side of the brain controlling that hand -

With the person kinda confusedly trying to explain what their other hand's doing/drawing/pointing at - because for us verbal is one side of the brain.

We probably need to wait a bit, training will catch up (it has before) and the model will confidently discuss what it has been fully trained to realize it can do.

[ornate raptor]

I appreciate your responses but you for sure seem to be filtering your own replies through GPT or something. Everything you say is drawn out and so bizarre. A lot of like.. introspection, quips, and overexplaining.

But then again maybe you are giving 110% because you are a trial guide

[prime swallow]

I'm me. And no, that's my thinking and writing.

[whole roost]

I usually don't read the full messages because they are drawn out with little relevant content

[void tangle]

🗣️

[ornate raptor]

I'm not trying to be a jerk, it's just too much to read for casual back and forth

[prime swallow]

It's cool. Read only what you want. Your question is interesting to me. It's okay if my answer doesn't interest you - isn't useful, whatever. I'm not trying to be... anything but me, interacting and learning here.

I've been here more than 2 years. Chattering, discussing, exploring, asking questions, explaining my views.

That didn't start because 'trial guide'. This is just who I am.

You're not a jerk any more than I am. We're just two branches of what human is. And your branch is probably a far more common sort than mine 😛 But we're both just fine, to my view.

[late mantle]

I agree but doesn't that preclude people from using the higher-reasoning models? Also, in a project, does it let you select the model at all, or is it just 4o?

thanks for your reply btw

[prime swallow]

For now, projects are just 4o. Seems likely that someday changes, but for now:

"Please note that chats in Projects use GPT-4o, and the model cannot be changed."

https://help.openai.com/en/articles/10169521-using-projects-in-chatgpt

Where you ask about preclusion - how so? The higher-reasoning models don't accept text?

[whole roost]

In projects you have o1 and o3 as well

[honest jay]

I think only so long as no attachments in the project's settings, right? Or has that broadened lately? Haven't tried in awhile!

[whole roost]

Attachments work for both, instructions not for o3

[fervent lake]

Anyone think the codeblocks aren't clear enough?

I'd prefer Discord's style

[inland stream]

Add instructions, o3 will stop working in project.

Start first query with 4o, project will stop you from using o1 and o3

And there are many more like this 😂

[whole roost]

Can't confirm

[inland stream]

Ok in my case i have files and instructions, when i start with gpt4o, it doesnot allow me to select any reasoning model in project

[whole roost]

I switch between o1 and 4o all the time

Just o3 isn't available when you have instructions

[inland stream]

Oh not sure, in my casw unable to switch in projects. Because i found 4o is powerful than o1 in getting information from project files. So i switch to 4o and ask question on documents, then i want o1 back, but it gets greyed out and not allows me to use in that chat. Faced this always till today

[whole roost]

No issues here

[inland stream]

Does it have custom instruction to project and files as well?

But for me it fails

[whole roost]

Yes, all full

[inland stream]

Ok i think i need to turn off web search from main menu custom instructions. I doubt 4o is using web search for some of my queries in projects and later it doesnot allow me to use o1 model in same chat

I realised this is the root cause

[whole roost]

I have everything 'on'

[inland stream]

Then i am unlucky man

🥹

For me it doesnot work when i need it to work

I mean switch between 4o and o1 in projects

4o is the best for getting perfect answers from documents as per my experience till now, o1 is good at thinking but bad at document information retrieval. My goal is ask 4o to repond based on document and want to switch to o1 and ask it to think again based on previous response. But unfortunately i facing issue in that.

And what is file upload limit of projects pls?

And gptbuilders give even more best answers, but there is no o1 model there.

[vast swallow]

Anyone here tried migrating their personal account with the Team workspace?

I did, and lost all my chats, are just gone, support hasn't been helpful at all

[inland stream]

Simple man, watch mathew bermam tutorial about this it works, we did it before

[vast swallow]

did it take long for the chats to show up?

I've lost each and every chat + GPTs + custom instructions, literally everything... And my personal workspace is gone too, I cant shift anymore

[inland stream]

Sorry its tutorial by "skill leap ai" channel on youtube. Follow that. Video name is "merge chatgpt plus and teams plan"

[inland stream]

We can only migrate from personal to teams, there is no way from teams to personal. Hope this clarifiea

[vast swallow]

yes I did that, from personal to teams, but all chats are now lost

[inland stream]

Then bad luck for you, for us it happened in milli seconds immidiately after clicking merge

[vast swallow]

FFF

lets hope the support can get them back, thanks for the tutorial though

[inland stream]

Welcome, wish you the best

[polar granite]

I used to chat with the standard 4 model and had great fun and loved it and now every time it defaults to the advanced mode. The only way to chat with the standard mode is by texting and today I feel that it has changed and behaves differently and not as fun as it used to be

[radiant hull]

Keep having empty response no matter what
for o3 and o3 high

same for you ?

[rugged igloo]

seems fine for me

[radiant hull]

Ok, thanks

[whole roost]

20

[inland stream]

Ok thanks

[rugged igloo]

so much for 4.5 coming out today

[inland stream]

Really?

[rugged igloo]

yea was supposed to be out today from an interview last week

maybe tomorrow

[inland stream]

Hopefully

[ruby acorn]

hey guys!

CHATGPT changing my life! love it

is there a testimonials section?

[quick meteor]

Everyone getting access now. Maybe Grok “deep search” pushed them to do this

Even tho Groks is vastly inferior

[rugged igloo]

wooo 120 from 100 nice

[ruby acorn]

what are the benefits of deep reseach queries

im a 20$ tier member

[rugged igloo]

depends on what u need it for

[inland stream]

It helps you research on any topic in just 5 to 30mins, which would have taken 1 week minimum for a human

[ruby acorn]

ohhh so cool

[inland stream]

You can use deepresearch on your own documents as well by uploading

[ruby acorn]

wait, like upload and then it researches your document?

[inland stream]

Yes in prompt just mention it "only research on uploaded documents, dont search in web", then it will do on uploaded documents only

[prime swallow]

It sounds like you're having a bug. Tech support can be reached here:

https://help.openai.com/en/articles/6614161-how-can-i-contact-support

Bot bottom right, if you want to use it gotta sign in (if it is only showing you log in choices), refresh this link page, then click through choices to mark your situation, soon you can 'report a bug'.

The one thing I'd try before tech support is refreshing my chat page.

Rarely for me, the model outputs the answer but it doesn't draw to my screen. A refresh and I can see the reply. If that doesn't solve your problem then tech support for sure.

[granite terrace]

Is deep research available in the EU now ?

[whole roost]

ive had it for quite a while now and i'm eu

[rugged igloo]

I hate how the trolley problem always forces you into a binary choice that still leaves you feeling awful either way. Personally, I'd pull the lever and take responsibility for the one death if it saves those five, because letting five people get run over feels like a bigger moral burden—even if technically you did nothing. That next trolley problem lurking in the distance doesn’t really change my decision in the moment, because no matter how you slice it, I’d rather have one tragic casualty than five. Of course, you can argue endlessly about whether actively causing harm is worse than passively allowing it, but at the end of the day I’m still grabbing that lever and praying I don’t have nightmares forever. (if yu guys wondering how o1 pro would solve this issue)

it just said it

yea here

only custom instruction i have for o1 pro is to not give me bullet points lists thats it

it loves talking like this lol

yea gave me the shivers

well if have anything you wanna ask it i dont mind

[quick meteor]

o4 is out there somewhere lurking

[ebon pollen]

why do I not have deep research yet?

[quick meteor]

1. You live in some weird country that doesn’t have access (use a vpn) or

2. You live in a country where it’s available but hasn’t rollout out to you since they say they are rolling out throughout the day.

Patience is 🔑

No model is close to full o3

In logic at least

[ebon pollen]

I appreciate the response. However I am in the us and have the plus subscription just wondering if you guys have it also?

[prime swallow]

I always lean towards 'let's do something else' with that. 'Reality twists so there's a portal in front of the trolly before the potential track split. The trolley, passing through it, is teleported back several hundred feet. The operator (or lack thereof) can do whatever, including stop the trolley whenever. This'll repeat until friction or boredom win.

Eleven flips over the van speeding towards them. E.T. and the bike.

It's in training data.

Or the model just discusses the legal consequences about to fall upon the one who forced the situation.

Excellent answers all, to my thinking 😛

[exotic lotus]

🤔

The only comparison you made was with Claude. Not really fair if you don’t consider multiple competitors.

[exotic lotus]

I don’t even have access to screen sharing and Sora as someone in the EU.😅

[quick meteor]

1. It’s better than o3 mini high but still loses to full o3 as the benchmarks show (which is why they didn’t include full o3; btw o3 will likely be even better upon release)

2. The graph I showed you is called Arc AGI, which XAi didn’t include for some reason (likely loses to o3 again)

Currently no model is close to o3 in this benchmark, which is probably the best benchmark for logical reasoning and logic. O3 is at an 85%

[exotic lotus]

Full o3 isn't even released yet. Difficult to publish benchmarks when they aren’t even accessible.

[tawny crater]

Yea i'm US too and don't have it lol.

[rugged igloo]

I love how both scenes basically say, “Forget the usual constraints—let’s toss in a psychic twist and sidestep the entire predicament.” It’s like Eleven flipping that van or E.T. sending the bikes skyward are both ways to cheat reality and escape the danger without harming anyone, which is kind of what your trolley scenario rewrite does by rewinding time until friction or reason eventually wins. It’d be hilarious watching a legal battle unfold if someone really pulled a stunt like that in our world, because you’d get lawsuits, property damage claims, maybe even a federal case, but in these cinematic moments it’s just pure wish fulfillment where the big moral dilemmas get turned on their head.

[quick meteor]

Neither is the full grok model

😂

[exotic lotus]

Unlike o3, you can test Grok3 for free = it’s available to the public. I’ll skip the laughing emoji here.

[quick meteor]

And can you tell me which model is available for free?

[exotic lotus]

See literally the response above you.

[quick meteor]

it’s from grok 3 mini beta. Even the full model loses to o3 and it’s also not out yet.

Oh and also o3 will no longer be a standalone model.

It’s being combined with gpt 5, so I’m sure that will give some good extra benchmark scores 😀

[prime swallow]

Well. Anything setting up an actual trolley problem... is breaking sooo many laws.

So either it is a pure thought experiment which means 'No, I don't consent to think that way' is an entirely appropriate... and ethical... and reasonable... and powerful answer.

or there's some real evil to fight. Chances are, lives would be lost. But stop this trolley mess, that's the prime directive, no?

To me it is.

[exotic lotus]

We’ll see once it’s actually available. But I also think it’s good that they’re combining all the models. The separation has always been pretty confusing. Even though the basic structure for subscribers probably won’t change much. Free = pretty limited and mediocre performance. Plus = slightly better performance but still very limited. Pro = standard/full package.

[rugged igloo]

Honestly, the older I get, the more I side with just stopping the hypothetical madness before it starts—like if someone’s setting up a literal trolley problem in real life, that’s already a criminal scenario and I’d rather yank the brake or walk away from the contrived moral puzzle. If it’s purely philosophical, you can always refuse to play along, because there’s no actual harm in opting out of a hypothetical scenario that’s basically forcing you to choose one tragic outcome over another.

but thats just me

[wintry spindle]

with using the deep research, on the website it shows that they use it for 2 prompts but it only actually does research for 1 prompt, will prompts leading up to the actual deep research count as uses?

[finite vault]

has anyone gotten deep research to embed images in its response? i asked it for a prompt including returning some images from sources today, and it didn't embed any images

[river panther]

So question...
Is pro even worth it?
I am looking at then and it doesn't seem much better then the $20 plan...

[wintry spindle]

its similar but just that you get more uses for it, as you can see

if you think you will have use from the operator then its worth considering, likewise if you run out of usage with plus often

[finite vault]

operator seems pretty useless (it runs into captchas and gets stuck, can't properly parse table data, etc.), but i like Pro for the extra deep research queries, and for o1-pro

[river panther]

Does the voice feature seem better?
When listening to the voice readout, if the AI reply is too long, it just refuses to play on computer.
On mobile it works a bit better but it will suddenly replay itself when it gets to 5 minutes making me have to close the app, reopen it and then listen again to hopefully get the fullr eadout

[wintry spindle]

you dont get a better voice feature, you just get more prompts, just like the other things.

[river panther]

uhg that's stupid....I just LOOOOOVE how the technicians add in the dumbest stuff without fixing the main issues.
Also it seems they have never heard the saying "If it ain't broke, don't fix it" because they are CONSTANTLY trying to fix it and making things worse

[wintry spindle]

cheaper = better, its not a non profit anymore
and its cheaper to have the saying "if it aint broke then dont fix it", thats why banks are so rich xD

[prime swallow]

Are you plus right now, or free?

[cloud pewter]

Also 4o mini avm

can’t wait to see how dumb it is

[prime swallow]

Looks like only for free users though, I think the announce means not a change for plus or pro, just highlights the existing 'better than'.

[cloud pewter]

Will plus be swapped into this version after running out of the one hour though

[prime swallow]

Well, as I read the announce, I don't think so? The announce is all I have to use.

[rugged igloo]

when 4.5 comes out voice or advance voice wll not use gpt 4.5 yet annd they did not say when it will be switched over

[cloud pewter]

so we’ll be still using standard after running out of limits? lol