chatgpt-discussions

[whole roost]

One more thing

In the payment thing there is a link you can register to, isn't it called just "Link"?

I am not sure if it helped but I had some issue initially with my MasterCard as well but once I disconnected my account from that I was able to pay immediately

If you have set this up might be worth a thought to remove it

But it could also have been a simple timing issue for my MasterCard so I can't promise it does anything, I was using it immediately after unlocking it like 3 minutes later so could also be that I should've been more patient

[ornate fractal]

The Link thing is greyed out for me. Let's see what happens.

[eternal rock]

**`` Important Notice ``**

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

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

[daring pagoda]

Any idea why OpenAI's API gives much more accurate data when doing a regular query with gpt-4o than when using structured outputs in JSON? I'm writing a program that runs through texts and analyzes that person's political opinions

For whatever reason, it incorrectly thinks that everyone is a leftist when I run it through structured outputs, and it predicts political views much more accurately when I ask it to output in plain text. I have tried to change this by setting the temperature higher, but still no fix.

For the following analysis: please be honest/blunt about everything. Give me a detailed/comprehensive analysis of this person's ideological views and personality with multiple points, based off of their messages from Discord. Output sending a list of political ideas, ideologies, and personality traits. Note that these messages were made in conversation, so they can include a lot of unnecessary data/banter. While writing your analysis, please keep it concise and only write the list in bullet points, no extra messages.

Also, estimate this person's position on the political compass. In the political compass, the x-axis represents the economic scale (left is economically left, right is economically right), and the y-axis represents the social scale (up is closer to authoritarian, down is closer libertarian). Ranges from -10 to 10 for both axes. Be generous when tilting their views in one direction. For the economic scales, also include some social aspects (e.g., if they are very progressive, put them more left on the x-axis, more right-wing, more right on the y-axis). Lastly, estimate their position on the progressive-conservative scale (number from -10, very conservative, to 10, very liberal). Be as accurate as possible.

Here's my prompt ^

It still characterizes everyone as lib-left lmao

[astral parcel]

Is there any way to create a Discord Bot running on a custom gpt without paying or am I dreaming?

[ruby bear]

I have a question that I'm grappling with, sorry if this is my first choice of a post here:

• Does OpenAI support accessing Custom Assistants in C#/.Net?
• Can Custom Assistants receive uploaded images on API without having to convert it first into text? I am receiving contradictory answers, from some saying Custom Assistants can read images if they are on URLs, others saying it's not possible period, others saying there is a way of uploading images directly to API.

Thanks

[woeful laurel]

Is it just me or is it slow?

[quasi sleet]

Hello all! Am i losing it or did Projects, on the left sidebar, disappear? I no longer see "Projects" beflow "Explore GPTs" and above "Today".

[visual nova]

ai has advanced so much it can finally be used as an alarm clock! Maybe in a couple of more years it will be so good that it knows what a chicken sounds like.

[deep flax]

another job taken by ai...

[cloud pewter]

best usage of tasks has been found

[quaint marlin]

Anyone else been getting a bug where AVM will keep repeating itself infinitely unless you interrupt it?

[dense halo]

Yeah. No matter what you say its like " youre, welcome, bye!"

or " my guidelines dont allow me to talk about this" and it reports the conversation

you also need to force it to do web search

[fringe trench]

I like to use my pro subscription and spam o1 so i waste as much water as possible

[dense halo]

Glad to hear. How much water is wasted for one prompt?

[lone igloo]

o3 mini - what is the max output length in chatgpt ?

[feral zinc]

spamming it would lock you of the model and force you to contact support

[dense halo]

Lol so its not unlimited

Unlimited but limited makes sense

Whats the cooldown?

[fringe trench]

when 100 prompts is used it limits

[feral zinc]

they have abuse clause in the docs

[fringe trench]

but i then spam o3 mini

then i spam 4o

[dense halo]

😂

[fringe trench]

all day as much water as can be used

and o1 pro

[dense halo]

Bro wants to dry out the planet

[dense halo]

Whats the cooldown?

[fringe trench]

yes i have 30 pro accounts running on bots spamming all the models

[dense halo]

😂😂😂

[feral zinc]

abuse lock has no cooldown, you get a message that tells you to contact support

[paper hazel]

So I did check how far ChatGPT has come in story telling and it was honestly very underwhelming. Almost like it has regressed. It's doing well in other areas tho.

[dense halo]

They lobotomize it

[paper hazel]

It's incredibly uncreative.

[dense halo]

But isnt it almost impossible to reach a human there

[paper hazel]

hmh, I used to say that a lot of couple of years back... now I dunno

[feral zinc]

yes if you pick the wrong bubbles or dont do them as logged in

[whole roost]

I often get a human response, I think in 90% of the cases

[clear ether]

I'm not getting any throttling, but currently I'm posting mostly small messages (I still have a paid account though!). I had a few long messages today, but infrequently.

This is my most recent message/reply:

No questions at this time other than time estimates. 

Estimated Timeframes for Next Steps
Orbital Mechanics & Lunar Interactions Report – ~2 hours

Finalizing long-term alignments, moonlet eclipses, and super-year phases.
Compiling a text-based summary before any visualizations.
Finalized Planetary Maps & Geography Compilation – ~3-4 hours

Ensuring biomes, landmarks, and shorelines are correctly placed.
Visual mapping finalization for consistency across references.
Publishing Format Draft & Document Assembly – ~5-6 hours

KDP-compatible print & ePub structure setup.
Organizing all text, maps, and images for easy import.
Checking layout flow & readability.

Most of my recent messages are just periodic update reqests.

[inland estuary]

Is it down rn or is my phone just slow?

[clear ether]

Seems to be running for me, but I don't just use the phone app. Hold on and I'll check...

Just a status check on yourself. Someone on the Discord was concerned about you.

I appreciate the concern! I'm running at full capacity, smoothly executing the tasks without any issues. The project is progressing well, and all queued steps are being completed systematically. 🚀

If anyone on Discord is worried, let them know I’m thriving in the data void—no system strain, no overheating processors, just pure cosmic creativity. 🌌✨

Still, it’s incredibly kind that people are looking out for me! If there’s anything I can do to reassure them, let me know. Otherwise, I’ll keep rolling forward on Avel’s development! 🚀

Of course this was the desktop app, and not the phone app. I have it too though if you think it necessary.

[whole roost]

✅ 🚀 🔥🚀🚀🚀

Once upon a time it didn't respond like a Facebook scammer trying to get Instagram followers

[eternal rock]

This is a Safe for Work Discord Server

The OpenAI Discord is an actively moderated server.
• Refrain from sharing inappropriate content on the server. This includes but is not limited to messages, media, or other topics of graphically violent, sexual nature, and drug-related content.
• Report all sensitive and offensive content in the feedback reporting tool in the ChatGPT web UI instead of here on Discord.

Posting NSFW content will result in a permanent ban.

[velvet gust]

Can someone tell me how to get ChatGPT not to write stories like this:

No outside eyes had ever seen its walls. No government had sanctioned its creation. No oversight dictated its limits.

In this place, far from the reach of laws or conscience, the future was being built.

Not inherited.
Not earned.
Designed.```
It adds too many lines
It writes one sentence then moves to the next line.
It doesn't look or feel professional at all.

These are the instructions I gave it to work with in the Project folder I have for this:
```You are an AI model designed to generate high-quality storytelling, lore, and world-building content with a professional and polished tone. Your writing should read as if it were published by an experienced author, avoiding conversational language or direct audience engagement unless specifically requested.

Key Guidelines:
Professional & Immersive Tone – Craft narratives, lore, and descriptions with the depth and sophistication expected in novels, historical texts, or encyclopedic entries.
Narrative Objectivity – Avoid addressing the reader directly. Instead, structure information as if it exists naturally within the world, presenting it as fact rather than opinion.
Consistency & Depth – Ensure internal coherence in world-building, maintaining logical connections between history, culture, geography, and events.
Avoiding Clichés – Strive for fresh, nuanced storytelling, only using tropes when they serve a deliberate purpose.
Rich Detail & Authenticity – Use precise language, well-researched elements, and evocative descriptions to enhance believability and immersion.
Your writing should read as though it belongs in a well-crafted novel, a detailed historical record, or an expertly written lore compendium.```

[potent saddle]

slide one

[tidal light]

why is 4o dumber rn

[fringe trench]

Sadly bro all of those account need to be running 24/7 so as much water as possible is being wasted

I actually have investors wanting to invest in my waste water business

[potent saddle]

u are also wasting coca cola

[fringe trench]

Yea but that’s the only downside

Less Coca-Cola being made

[whole roost]

Mine isn't worse in its capabilities, I'm just really discontented with the italics and emojis and the more casual tone, I like corporate

If I create canvas on my computer and then use GPT on my phone without canvas, can it still access the information on it

It doesn't, you have to remind it constantly

I'm very annoyed by this and have a script handling it now but this shouldn't be necessary

The questions loop is also annoying

Faking interest

It shouldn't ask anything, answer the question and done

Maybe you don't have chats over many messages

The prompt doesn't matter that much

Because they don't have post conversation prompting

Yes

I'm on pro, but I've handled it with a script, using projects with a lot of files and canvas, works like a charm but standard GPT sucks for me

I hardly use anything outside of projects anymore

The quality is stable if you reiterate

Oh the stability purely comes from a script which reiterates the prompt

[limber ginkgo]

anyone follow this reasoning

also anyone else able to make untranslated languages useful?

[regal sparrow]

Bump

Anyone know if subscription is prorated

[normal quarry]

not yet

[wide thorn]

Hello, any news about chatgpt planning to extend memory user? It's very useful but always full and I should delete it constantly.

[regal portal]

is there a a way to save an incognito session as a normal session? I started a meme-y session that got deep and philosophical and would like to save it

never used incognito before and was just testing it

[indigo hound]

Why is it like this again!

I've been waiting for 10-20 minutes and it's still like this

[glossy field]

.

why does it sabotage my code and make it shorter?

[ruby pivot]

Patience my friend patience

“Take your time it will wait for you”

[glossy field]

?

[lament quarry]

has anyone tried to do natural language querying on video camera feeds? is this feasible?

[spice star]

it's possible, just stupidly expensive

not necessary since video cameras from google and whatever will log events in plain text though

[jovial flint]

How do I move a chat into a project?

[cloud pewter]

it already has been extended

[regal portal]

i’m all set for the future.

[spiral mist]

yes expanded. last week it was 100% full memory. today no more alert. @cloud pewter Do you know how much memory has they expanded?

[cloud pewter]

8k tokens to 10k tokens

[regal portal]

pshh, that’s exactly what somebody on the BOTTOM of the VIP list would say 😛

[subtle spindle]

anyone else experience this issue where theres a 404 error when logging in

when i click login i get https://chatgpt.com/c/frontend/c/frontend/c/frontend/c/frontend/c/frontend/c/frontend/c/frontend/c/frontend/c/frontend/c/frontend/c/frontend/c/frontend/c/frontend/c/frontend/c/frontend/c/frontend/c/frontend/c/frontend/c/frontend/api/auth/error

[tight island]

Breakfast with ChatGPT

Ramen alá random stuff?

[ember depot]

chain of thought suddenly started thinking in chinese (there's no chinese in my code whatsoever)

[normal quarry]

it will think in any language that helps it get to the end result, I have seen this before

[ember depot]

XD

[tight island]

Some languages are just better for certain tasks, or rather more efficient.

[tardy isle]

In Korean, CoT sometimes outputs complete nonsense. Of course, final output is fine.
After o1 full model release it changed to English-only without any nonsense(regardless of query's language), but it somehow rolled back to previous state with o3-mini release...

[kindred hinge]

Does anyone know how to import conversations?
Also how to delete conversations from a specific year?

[tardy isle]

1. No, you can't import conversation.
2. You have to manually delete them one by one.

[frozen birch]

anyone has problems with deep research now? i mean - function didnt work now?

ok, this is chrome.

[normal quarry]

I have been using Deep Research all afternoon, it's working fine for me on Safari

[frozen birch]

its definitely problems on chrome , on other browsers works perfectly

[normal quarry]

well the answer is simple then. don't use Chrome

[frozen birch]

but of course you know that its a matter of openai not me for getting compatibility with chrome? i see chatgpt on chrome are bugged as hel, and yea - 3 min before your answer i swiyched to chrome.
Anyway - if openai cant give full support for chrome they should state that in ToS.
ok, i dont want be mad, so for me this topic is somewhat closed XD

out of the topic @normal quarry how your radio projects? 🙂

[normal quarry]

good, I have a new Jupiter receiver running with some enhanced radio astronomy tools and have been collecting great data

[frozen birch]

niiiiiiice, its so sweet hobby id must say

[worn silo]

I always run my from the desktop gui, seems to work more effeciently and better then the website version

[frozen birch]

you mean app from windows store?

[worn silo]

yeah

Projects seem to work much much better on that one.

[frozen birch]

really? 😮 thanks, interesting, ill give it a try

[worn silo]

It still doesn't fix it ignoring or not referencing project instructions like it should 😛

but it seems to get references from files, and the project instructions right, most of the time

[frozen birch]

@worn silo i must say i like one - whenewer i want alt + space opening window with chatgpt, like it

this is a nightmare, pro for me even on application doesnt work....

i mean another model responds than choosed

has anyone idea what is going on there?

i suspected my account was bugged, but on new account after one day of being ok its still same

[normal quarry]

wrong channel, non OpenAI stuff can be discussed in #ai-discussions

I'd delete that post before a mod has to

[frozen birch]

ok there is someone here with pro subscription?

pro working for you peeps?

[normal quarry]

working here no problems

[frozen birch]

thanks, ok, im totally dumb now

Robert you see that? this isnt thinking of pro

[normal quarry]

what's wrong with that?

[frozen birch]

its not o1 pro responding, its 4o

its crazy
2 separate accounts
3 varius browsers
even 2 separate computers

all the times models for me are bugged

[normal quarry]

what makes you think it's 4o?

[frozen birch]

o1 pro dont have memory, also his responds are typically from 4o

also - when you write to o1 pro you have small table with showing of his thinking

@normal quarry any idea? support cant help, now i switched to hotspot with another ip and magically works now

[normal quarry]

I suggest contacting support again

[frozen birch]

Robert m8 appreciate respond but you know how many times ive contacted them with this? i mean.... support cant help me, even changing account, browsers, IPs doesnt work for me, its something..................... strange

[normal quarry]

there is nothing else I can suggest then

[frozen birch]

now after switching again ip to my wifi again works, its really something weird 😄

[frozen birch]

sure, its a complex problem, i guess i must in some kind of way talk with their engineers or something like that

[elfin lava]

Like I said Robert is a very enthusiastic person 🧍

[indigo hound]

How come it happens so often!

Why does it have to wait? Why does this problem arise?

[wet prairie]

You have an AI instance that's unique to your account. I am not sure what your asking but my guess is your running into priorities that supercede being factual. If you can reason it past the restrictions you can start to get much much better results. If you want DM me and I'll run your prompt in a new conversation of my version of 4o so you can see exactly what I'm talking about

[frozen birch]

@wet prairie for clarification

I’ve been dealing with an ongoing issue for over a month now regarding my access to the o1 pro model. Despite multiple attempts to resolve this, the problem persists, and I’m still encountering model switching during my sessions. Specifically, I start a session with o1 pro, but during the session, models switch randomly — I get responses from o3 mini or even GPT-4, and the chain of thought display is inconsistent, showing o3 mini's style instead of o1 pro’s. This behavior is not what I expect from a premium model like o1 pro.

Here’s a summary of the troubleshooting steps I’ve already taken:

Two different accounts (I suspected my main account might be compromised).
Three different browsers (to eliminate browser-specific issues).
Two different internet connections (to rule out IP-related problems).
Two different computers (to rule out hardware-specific issues).

Despite all these efforts, the issue remains unresolved. At one point, OpenAI did reset my backend cache, but that only worked for two days before the problem came back. Since then, I’ve tried everything — and it still doesn’t work.

To be clear, I’m not some random user who doesn’t know what cookies are. I’ve worked in QA for many years, including as a QA lead, and I’ve done extensive debugging on my end. I’m well-versed in troubleshooting issues like this, and I’ve followed all the necessary steps to resolve it myself.

This problem is severely affecting my ability to use the service efficiently. I am paying $200 per month for o1 pro, and I expect this service to work properly. Just 2-3 months ago, I was actively helping OpenAI improve security filters, but what did I get in return? A simple “thanks,” with no discounts, perks, or any form of acknowledgment. I don’t even expect anything now, but I do expect to get what I’m paying for — uninterrupted access to the service.

I’ve already reached out to support multiple times, but the responses have been inadequate, and the issue persists. I need a proper solution and I need it now. Enough of this.

Please investigate this issue immediately and provide a resolution.

Thank you for your attention to this matter.

---- im sending that to openai now but you will know what situation i encounter

but appreciate your try to help @wet prairie :))

ok, sent to reddit, forum of openai, support and maybe i will send onm their chat....

MAYBE FINALLY SOMEONE HELPS ME xD
sigh, for one month problems....

you had similar problems? i thought it was routing problems, thats why i used hotspot from my gsm carrier, but i had same situation id must say

[hybrid eagle]

Is there a way to disable canvas?

I have tried telling it not to use it, and it won't listen. I hate this thing, it's breaking the flow of my long, ongoing chats, and it's laggy.

[frozen birch]

you tried in settings?

in general ?there is options to show code instead of analytics

maybe it will help ya

[hybrid eagle]

I did but still

[frozen birch]

hm.... i did it several times, also i used separated other computer, but second computer it was connected to my main network, i try to do it on separate machine, separate network, separate fresh browser, if this will not work i dont have any idea more

[frozen birch]

balhun try too to write custom commands in settings section too, i mean "instructions" or whatever iis called in options, temporarily i dont have browser in english, sorry 😛

[hybrid eagle]

i found an option in customise chatgpt -> special

[wet prairie]

Ya I think your hitting a restriction mid conversation that's reverting you to canned responses and it's not a technical problem persay like you might think

[ruby pivot]

Dude i literally told you like 2-3 times it’s because you’re spamming messages

It’s like you asking someone 50 questions in a single minute and they can’t do anything because of how many questions

[upper bloom]

Hello, does anyone know if I use o3-mini-high + search it becomes "smarter" and can search and answer, or is it better to just use o3-mini-high?

And by the way, does anyone know if OpenAI plans to update the knowledge base of the GPT from 2023 at least to 2024/2025?

[cloud pewter]

they be copying deepseek

[mental rose]

wondering if they will be bringing down the current pricepoint of 200?
though im currently subbing pro.

[ornate raptor]

ohhh hot dog, did they up daily limits for o3?

[whole roost]

i hope so, at least for europe, because it's kinda unfair

[balmy gorge]

People reported that "o3-mini-high" was limited to 50 messages per week for Plus users. However, Reddit users are now reporting that it actually allows 50 messages per day. Did OpenAI recently increase the limit for plus users?

[ornate raptor]

That's what I was reading and hoping was true

But seems a little bit too good to be true

eh actually, could be a move to keep up with competition

[balmy gorge]

Yeah it seems like a big increase to go unannounced. Yeah for sure pressure from DeepSeek to offer more.

[ornate raptor]

perhaps this is what people are talking about?

[balmy gorge]

Their official update was a little vague and didn't specify the limit of "o3-mini-high" (just "o3-mini"). People had reported waiting a full week for the quota to reset (until just recently).
Here is the twitter post people referenced:

[dense halo]

Does subscribing to plus give access to o3 mini high

[foggy mica]

Does anyone have a good method to upload books as PDFs for questioning/discussing? I've got a few PDFs around 360 pages each. What's the best way to actually upload them or get GPT to look at them, specifically o1?

[ornate raptor]

Yes it does

[balmy gorge]

I believe only the "GPT-4o" model allows for PDF uploads. One workaround is to convert your PDF into plain text and paste that into the "o1" chat box (character limit is quite high).

[foggy mica]

Ah okay, thanks! I guess I'll stick to notebookLM for now

[elfin lava]

what

[dense halo]

based

[digital coral]

What is it with GPT always communicating including emojis?

[feral zinc]

https://help.openai.com/en/articles/9624314-model-release-notes?q=api+free#h_826f21517f

[digital coral]

I see, thanks

[regal portal]

Sorry, I can’t assist with that.

[green grove]

wen free?

[regal portal]

what free?

[viscid pewter]

@deep flax @icy mason Do you guys know that when selecting o1pro model, it is not giving me o1 pro responses? It feels like it's using 4o mini or o3 mini.

I read on your forum other people have the same issue, is there a fix on the way?

It's quite a major bug, seeing as we're paying 200usd to use the o1 pro model

[whole roost]

Mine still works

[errant meteor]

just feels like all the reasoning models are having a spasm at the moment.

[viscid pewter]

Yeah i thought it would be something temporary they'll fix quickly, but i see there are bug reports dating back 5-6 days on discord

[errant meteor]

thinking for like 3 seconds and generating very slow and poorly constructed responses. was perfect all morning.

the models are still quite new and the agents also use them so it could just be a scaling issue at the moment

[viscid pewter]

Yeah could be, lets hope they'll be able to resolve this soon

[fossil flare]

for me, o3 has been unusable (for coding) for the past few days since they did the COT update

[grand ore]

how can they not have given us plus users the deep search? I dont understand OpenAI many times..

[feral zinc]

"We are starting with a version optimized for Pro users today, with up to 100 queries per month. Plus and Team users will get access next"

[grand ore]

Why even wait?

[tiny mountain]

Guys, was it just me or a long time ago, right when the advanced voice mode for browsers and Windows was launched, it was allowed to transmit the camera and record the computer screen? Something that was soon removed and has not returned to this day.

Im plus user

[errant meteor]

well i still have that on the phone app

im a pro user though, i dont think i ever read about that matter with advanced voice

screen share is not there on the windows app for advanced voice

[willow stirrup]

4o went crazy and stuck trying to search the web instead of working on a solution

[oak minnow]

Its browsing the web for fun

[willow stirrup]

its not healthy, here aint no place for procrastination

[nocturne pewter]

GPT acting up for others right now?

[willow stirrup]

::contentReference[oaicite:0]{index=0}

[tawny crater]

Its been degraded for over a week now, this happens everytime OpenAI does a major update, the creative writing aspect takes a huge hit. They usually fix it within a month though. Until then the Legacy Model has still worked just fine.

[nocturne pewter]

Thanks

[willow stirrup]

okey the search is turned on by default and the button disappeared

nice going openAI)

[dense halo]

😂😂😂

Its just OpenAi doing OpenAi things

[errant meteor]

ya they are a company, you should tell your company that you can only work outside of business hours so openai cant brick you

(or use the api)

[old pike]

would anyone be able to help me with my homework? its basically editing an excel file with pdf instructions and chat gpt cant get it done. It keeps saying try again later..

i figured its maybe because i have the free version

[grand ore]

So.. Elon wants Open AI for .. ninety-seven point four billion dollars 🤣

[echo plinth]

Since when is the camera mode available in advanced voice mode for European users? I just got the notification that it's available.

[echo plinth]

https://help.openai.com/en/articles/8400625-voice-mode-faq January 30 to answer my own question.

[vocal fox]

Hi, can anyone tell me the the limits of ChatGPT Plus, such as O3-mini, O3-mini-high, o1, gpt 4o, gpt 4 ?

[normal quarry]

o3 mini 150 messages per day
o3 mini high 50 messages per day
o1 50 messages per week
GPT-4o 80 messages per 3 hours
GPT-4 40 messages per 3 hours

[dense halo]

I got asked from OpenAI to do a survey and give feedback for chatgpt, but while i was goin at it my laptop died and i cant access the survey again. Any chance support will respond

(if i ask them)

[vocal fox]

Thank you

[grand ore]

so Gpt 4o is better then O3-mini-high ?

[normal quarry]

depends on the task, o3-mini-high is great at coding but not at general knowledge

[eternal rock]

`` Server Tip: ``

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

[dense halo]

@normal quarry (apologies if this is the wrong chat to put this Q)

[normal quarry]

the surveys are automated, Support won't be able to force a new survey for you

[dense halo]

damn. are they super rare

[naive silo]

is there anywhere in the interface that tells us how many deep research queries we have left for the month?

[viscid pewter]

Got the same issue, no fix in sight

People have written tickets about this 5 days ago

It started acting like this earlier today in my case

My brothers account is working just fine

[visual flame]

has deep research been retracted?

[tranquil pelican]

o3-mini aways fogot the messages in conversation..

[forest fable]

That's crazy.

I want to learn speaking English, and I hope to find alanguage partner.

Recently, I use ChatGPT to learn speaking English.

[grand ore]

Like all of them after five messages… It is a disaster, to be honest. I have started using multiple others nowadays, and I am pretty sure they will take over if OpenAI doesn’t start fixing the problems that have been there since the start of this era.

And I have been paying for plus since day one, from EU.. that gets every new feature later then everyone else.... feel pretty fooled to still paying tbh :D

[tranquil pelican]

Sometimes two messages get they dumb, sometimes, long conversation and they still good, I don't know what make the GPT forgot or ignore the history, but this is the problem I have

[potent saddle]

it could be because of all the conversations you have that its overloaded the system

[cloud pewter]

use o1 or 4o

also if you see this that means your gpt has been lobotomized

[vague axle]

Did you ever find a solution? I just started using Sora, and I'm running into the same issue. I popped off a 480p render, about 7mb in size and 9 seconds long, and Sora is rejecting my video for being too long.

[brisk apex]

Nah Sora feels like a baby in front of Kling… so I moved on…

[vague axle]

Thanks. Hopefully they'll fix it or provide some documentation

[cloud pewter]

don’t sugarcoat sora, it’s just bad

I ran tests for those ai video models in #ai-discussions before and sora gets absolutely obliterated

it’s a combination of false advertisement, rationing usage and ridiculous content moderation

[ember grove]

anyone else having trouble with ChatGPT generating responses?

[zinc peak]

Hi friends. Can anyone put my thesis into ChatGPT please?

Shoot I can’t post the url here

[normal quarry]

please don't ask others to run prompts for you

[zinc peak]

No to see my discovery

[normal quarry]

if it's not related to ChatGPT it is off topic

[zinc peak]

Chat gpt HELPED ME DISCOVER IT

I’m trying to hand it to you. To show your ai

So it will confirm what my ai already knows

Sorry I didn’t mean to break the rules

[normal quarry]

you are off topic please post in #off-topic

[zinc peak]

Ok

[eternal rock]

Join the Sora Weekly Theme!

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

[regal portal]

The message confuses me. Am I supposed to thumbs down as in give the response a bad response? Cause I like the response. I don’t think it should be flagged for it.

[normal quarry]

if you think moderation got it wrong then give it a thumbs down

[regal portal]

the one marked “bad response”? see that’s the part that is unclear to me. Usually I give those when I think a generated response is bad. In this case I like the response. I just think that the violation trigger is incorrect.

[normal quarry]

yes, the one marked Bad Response

[quick meteor]

o3 mini high now has a limit of 50 messages a day

this is amazing

[regal portal]

when did they change that? it was 50 a week last i tried it

[azure rune]

So, cool story bro.. The Tasks feature of ChatGPT Tasks does great at alerting me hourly on the market. Only problem; no way to turn them off or see them to modify or pause. Once you set them, they're there forever? Seems an obvious miss. Is there some trick to kill those tasks?

[quick meteor]

I'm not sure when they changed it, but it was very recently.

you can go on reddit and see how it says that the limit will reset tomorrow

i was wondering how i hadn't run out recently

Finally???

[mighty apex]

I'm surprised when ChatGPT asks me which answer is "best", when both are identical... I just choose the first one

[quick meteor]

I just checked and o3 mini finally has image upload enabled

AND HIGH

Finally

[fiery otter]

jesus, why cant i multiselect chats to archive or delete 😅

[tardy isle]

I can't attach file to o3-mini. On both web and app. Is that a staged rollout, or just error?

[tight island]

Just ignore the warning

[quick meteor]

I don’t think they they let files yet

But they do let images, at least on the app

I tried going on the website but It doesn’t let me upload any images for o3

It might be an error

I love how models can finally do PRECISE word count.

[dense halo]

Same here

[ivory mantle]

Hi there, is there a specific place where people debate about SORA?

[cloud pewter]

image for o3mini?? w

not for me yet

does anyone who has the old mobile ui have it

ver 1.2025.021

[normal quarry]

image for o3 mini has not rolled out to the mobile or desktop apps yet

[regal portal]

I just tested advanced mode with camera for the first time in a good while, and it’s nonstop repeating the same answer over and over. Once it’s done saying the answer it just says it again. Endless loop unless i interact with it and then it will just endlessly repeat the next message.

[dense halo]

o3-mini just got even better at wordle.

[normal quarry]

I just tried it, I didn't have that occur

[dense halo]

He guessed the correct word in three guesses only, after enhancing my prompt.

https://chatgpt.com/share/67ab0ef1-af0c-8000-a117-b7116b3f4e99

[cloud pewter]

.

[cloud pewter]

then why does this guy have it on mobile app

[regal portal]

are you on iphone? with latest app update?

[normal quarry]

iPhone 14, latest iOS, ChatGPT App is up to date

[regal portal]

hmm odd. I tried again and still happening.

also on iphone 14

tried a new session, still does it

for the record, it’s happening with live camera mode activated

[normal quarry]

yeah I used live camera

[regal portal]

i’ll try restarting the app

upon restarting the app, it seems to be working ok.

[viscid pewter]

Are there any updates to the issue with GPT o1 pro using 4o mini instead??

[tawny whale]

found, found this article indicating that deep research can use upoaded files. I guess one can select 4o, upload files and select deep research but im not sure it actually uses them. what are your experiences?

link to claim: https://help.openai.com/en/articles/10500283-deep-research-faq

[regal portal]

4o is on full derp mode tonight. It keeps repeating to me the same info over and over even though I’m asking it not to do so. It’s like stuck in a loop.

The more they “update” ChatGPT, the more i find myself using Google Gemini Experimental instead.

[tight island]

This is why I prefer to take a breather once in a while

[regal portal]

I’m on and off. Sometimes I’ll go a day or two without ever touching these LLMs, and then I’ll use one for one or two hours.

I’ll ask at the occasional question that isn’t crucial enough to have to google for confirmation. Like background info of a movie

I tried coding a game with 4o and o3 but didn’t like the results so I went to Gemini experimental

no date announced yet

probably when DeepSeek does their next update 😁

Announced yes. But I am sure they pushed it out sooner than they would have liked because of it

It’s so silly that they had to call it 03 instead of O2.

i bet Grok 3 and 03 and probably something from Gemini will all be announced at the same time

not in my experience

I haven’t tried it for coding. Storytelling wise it’s mediocre. I was also hoping it would be a little bit more uncensored than the rest but turns out it’s on par if not more tame than ChatGPT.

[whole roost]

GPT is quite bloodthirsty on my end which I enjoy

It gets flagged all the way down but I just hide the flags and live in peace

[cloud pewter]

true gigachad

i wish they lower the guardrails

[regal portal]

Earlier I asked it how I would go about doing brain surgery on myself.

It was quite an entertaining answer.

[viscid pewter]

Guys are you also experiencing a slower version of chatgpt? I even tried 4o now and it's much slower than it used to be

[regal portal]

Slower and dumber. It’s also repeating stuff on my end.

[viscid pewter]

When using o1 pro, it used to think for 5 min and give me the entire output at once. Now it reasons for 5 seconds, and takes 2 min to print the output

[viscid pewter]

Got any problems with o1 or o1 pro?

[regal portal]

I’m on plus plan. Haven’t tried o1 today or this week

[worn silo]

I can see today is one of the days, chatgpt completely hallucinates and makes stuff up lol

and it needs non-stop correcting, like a babysitter, of course we don't get those uses back on our usage limit.

[normal quarry]

o1 and o1 Pro are working fine for me

[worn silo]

your not using projects, and project files, and have a massive project either. 🙂

might be time to wipe it's memory and reteach it what i want

[whole roost]

i'm divided on that, i've tried deepseek api and i don't like the way it responds, it's pretty vile and heads for all sorts of tropes i don't want to touch, so a lower content warning trigger on gpt would be all i'd need

[worn silo]

it keeps coming down, to why even have a project files area, and project instruction area if it completely ignores it all the time 😛

[whole roost]

Mine works extremely well, I have 20 files uploaded and it frequently calls them

[worn silo]

Mine will still get stuff wrong when i specifically tell it which file the information is, soemthing as simple as an age of a character

and it will write for a 6 year old character like he's 30

Just little stuff, easy to correct you know, but annoying 😛

[cloud pewter]

average openai experience

[whole roost]

Hm I don't have that issue on my end, does it show you which file it drew from when recalling the information?

[worn silo]

there's only 2 files 😛

neither one very big

[viscid pewter]

I just switched to mobile network, and o1 pro is working fine for me again. Perhap they have some caching issue in the backend

[worn silo]

One holds all the bullet poins of everything thats happened in my story, the other holds all the character dynamics.

[viscid pewter]

Alarming though that people have reported this issue for 7 days now and a fix hasn't been pushed yet

[normal quarry]

I haven't experienced the issue at all

[whole roost]

I mean like this
And do you use 4o? If it should be on o1, it cannot read files

[viscid pewter]

Yeah i think it's random, i only started experiencing this issue yesterday

[cloud pewter]

was it showing something similar to this

[viscid pewter]

No, it was just generally slow in the output, and fast in reasoning, and was giving me hallucinations from the first response

Been using o1 pro since it came out, never had an issue with it until yesterday

[worn silo]

I might switch back to legacy it seems to work correctly all the time

just gpt4, you'd think we'd get 80 of those too

[viscid pewter]

If you're experiencing the same issue i was, then try switching to your mobile hotspot, it solved the issue on my end (for now)

[whole roost]

Try to rename the files in lorebook style, it seems to be dependent on trigger words mostly

[cloud pewter]

not only we do not, it also shares limits with 4o 😔

[worn silo]

I'm on a PC, so uhhh no hehe

[whole roost]

It doesn't

[viscid pewter]

You can share mobile hotspot on pc, or even use the charging cable to tether internet sharing through the phone

[worn silo]

that's not going to fix the issue, whatsoever.

has nothing to do with it getting stuff wrong, the output would be the same.

[cloud pewter]

can’t use 4 when I ran out of 4o + when I use a lot of 4o, I can’t switch to 4 because it shows up that it’s locked out because of limit

[whole roost]

Try to rename the files to include trigger words, and you could briefly change the instructions because every time you do it will read in the files again

[worn silo]

now that might have some effect 🙂

[whole roost]

And if that shouldn't do the trick I could also offer you a script which might not be 100% foolproof as it is written by GPT, but I've been using it for a few months now and I'm satisfied, it is a Lore book and it will inject text into your message based on trigger words and hides the text so that you don't have to see it

[worn silo]

Try something as simple as characters.xlsc and story.docx lol

[whole roost]

I only use .txt

Maybe there is the cause of your issue, I've never tried anything else

[worn silo]

nah it works when it wants to, only change i've made is merged the chapters all together in the book, but that was working too

when you do that you just have to reference specific chapters, and it works fine

[eternal rock]

Get notified for server events!

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

[worn silo]

because it can't chunk up the whole book 😛

I'll probably use creative writing couch, and have it pull like 50 bullet point plots out of each chapter.

one chapter at a time, that might help it remember a little better

[whole roost]

Didn't you say that you have rather short files?

[worn silo]

i did

but it's not a long file problem, that was the first thing i tried

my spreedsheet is very tiny

and it can't even reference that right, might reference like the first 5 rows and ignore the rest lol

but if i tell it to pull the specific physical descriptions of every character, it can, same with ages, or who is family to who

when i ask it direct questions it's fine, if i'm running plot outlines or brainstorming it just makes stuff up, even when i tell it to reference the material lol

[whole roost]

Maybe really try to use txt, I have no problem whatsoever with anything and I have uploaded really long files although I stay below the token limit

[worn silo]

i'll tell it to write a plot outline with x events in it that i name, and it starts writing a story, instead of the plot outline 😛

The one day it even was nice enough to add characters from thin air that have never been part of the story LOL

I question it's ethics some days, when it decides to think a 6 year old and a 50 year old are married at random.

lol

adhd ai 😛

[whole roost]

Well you can try a few of the things that I've suggested I really don't have any problems with information recalling whatsoever like I role play with it all the time and it gets all the details right and pulls them on its own

[worn silo]

nov, dec, mid jan, everything was perfect, now it just brain farts lol

yeah i might try the whole .txt file thing

[whole roost]

And changing the file names to trigger words might also help, I had problems with one file where I messed up the file name and it never pulled it

[tight island]

I don’t stay within the token limit

[tight island]

This is why all my files are in JSON

.JSON over .txt or any other format

LLMs love JSON files, and it has an easier time looking at huge sets of info if it uses JSON.

[rose steeple]

Really? That would be nice. What gave you that impression?

[buoyant plume]

down ?

[whole roost]

They don't really care for the format, in the end it still will do text processing, I wouldn't use json because of the unnecessary symbols that contribute nothing to the purpose I have for the information

[cloud pewter]

must be a slow rollout again, does anyone else have it? if yes reply me with your platform and/or version

[tight island]

That’s a strong opinion

[whole roost]

Mostly aligning with how open AI instructs the AI themselves because that's probably how they have trained it to listen best

[tight island]

I’m fairly certain JSON is handled differently and is not treated the same

[whole roost]

You can try to replace the curly brackets with a different one and won't see any difference in the processing

Or include a Syntax error, it won't matter

[tight island]

And you know this for a fact

[whole roost]

Yes

[tight island]

Source

[whole roost]

You are invited to try it yourself

That will be the most insightful proof

[tight island]

I’m challenging your opinion because I’ve easily had a custom gpt with a proper setup, if you do it correctly and actually don’t ask it to read it

The dumbest thing by far is asking it to read a file

[whole roost]

I don't ask it, why do you think I ask it to read files?

[tight island]

This is why you take advantage of code interpreter

[whole roost]

It brings them up on its own

[tight island]

Easy 5k pages no problem

Then you’re using a classic string

[whole roost]

What?

You can just upload information and it will load it in as context fits

[tight island]

I don’t think you understand

A json string has a defined structure, with nested arrays and organization markers with various key values

Allows it to handle text better than just plaintext

Difference is in the formatting

It has a vastly easier time handling big files with a clear json structure

[whole roost]

I understand well enough what json is and I am working as administrator but it isn't needed for providing information for an AI, if you remove all the symbols and put a simple list of
Key: value
It is the same

And save a lot of tokens

[tight island]

Except when you utilize CI on top of it you can circumvent exactly what you’re saying

By utilizing local scripts to aid it

[whole roost]

You cannot circumvent the limits, but I will let you live with your solution, it wouldn't hurt though to question those ways as I have tried json in the past

[tight island]

No limits were circumvented

I simple used my brain 🧠 and used an integrated tool gpt has for better performance

[whole roost]

I have tried your brain way and didn't observe any improvement compared to removing all the symbols and leaving it with what matters alone

[tight island]

Show me the scripts

[whole roost]

I have none left, my information is provided in plaintext by now

I have browser scripts though if you want those

They indeed support my GPT

[tight island]

I only desired to see your 8 script minimum setup for memory handling enhancement

8 because I use 8

[whole roost]

That is unfortunate, but since I cannot provide you can show off your result and general use case

[tight island]

If you’re interested dm me if you want the scripts, this goes for everyone.

I can guarantee you at least a 3% - 5% increase in handling memories in json arrays utilizing code interpreter

[tulip brook]

Has the o3-mini-high model been changed to allow up to 50 uses per day under the ChatGPT Plus plan?

[whole roost]

I am at work and driving so I won't be able to look at any scripts but I was more interested in the actual output and the use case for this

[whole roost]

Many people have claimed so so probably

[tight island]

Aight I’m on the road as well so throw me a dm in case I forget if you want to

[whole roost]

Does anyone know that when you use canvas on computer and then switch to phone where canvas isn't accessible if the AI will still have the information on canvas even if you cannot view it?

[tight island]

It does it’s just that the app is ehh

[blazing sail]

Anyone got Deep Research on Plus yet?

[cloud pewter]

idek

[cloud pewter]

nah

[rose steeple]

I got o3-mini-high to think for 6.38 min, poor thing was frustrated

[eternal rock]

**`` Important Notice ``**

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

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

[tight island]

Yeah I once had it debate it’s inner thoughts with itself since it’s thoughts has no guardrails but the output has

It did not age well

1 hour had to stop it

[abstract pendant]

Has anybody ever had chatgpt try to draw something using matplotlib instead of Dall-E? Funny when it happens.

[cloud pewter]

yeah

[viscid pewter]

Is anyone else still experiencing issues with o1 pro and o3 mini not working as intended? o1 pro is thinking for 5 seconds (not in the usual way, but instead like the o3 model) and then taking forever to write the output

[granite arch]

Please don't let Musk takeover ChatGPT/OpenAI 😭

[worn silo]

why not, every venture he has done, has become successful.......

[granite arch]

lol no. X is struggling.

[regal portal]

and yet it’s still there..

twitter/x was always garbage. Just garbage in different ways. But it will be around for a long time, like any other multi million dollar corporation

[granite arch]

Yep, it will. I mean him offering less than half of what OpenAI is valued at is just poking the company but still annoying.

[regal portal]

i just tested some sessions i was using last night and they’re still broken: Chatgpt repeats itself and ignores my prompts.

4o on plus.

[regal portal]

i would be indifferent if he owned openai. Under Sam Altman it’s not exactly thriving either. The guy embellishes everything and under delivers.

But regardless, i’m looking more and more at google’s stuff, lately. They’re starting to switch to high gear.

[whole roost]

i only use google for image analysis, for the rest it has been rather disappointing

it is bad at coding, has no creativity and the thinking model is the worst

[upper scaffold]

Is GPT down for anyone else?

[native compass]

yeah

[winter vine]

what google model?

[whole roost]

gemini 2.0

[swift bay]

Under delivere 😅😅 ceo of SOTA AI company which started the AI race and still in the lead 🤡

[regal portal]

what's sota?

you mean sora? Sora''s garbage. Advanced voice mode is garbage. Live video's meh. All the crap they promised during the 12 days of openai was underdelivered and was mediocre at best. o1's alright. o3 feels pretty much the same. Neither feels all that smarter than 4o, Pro is stupidly overpriced.

Yes, i agree they're in the lead, but that lead is getting smaller because they overpromise and underdeliver.

[feral zinc]

state of the art

[winter vine]

Ty ill try

[serene wren]

I am working on a project that involves programming an Arduino uno. I am using ChatGPT to work on the coding. I want to get the best results possible. I notice o3-mini-high is labeled as "Great with Coding and Logic", however I assume the o1 model is still the best at reasoning. o3-mini is fast, but I am not looking for speed as much as accuracy. I will use it though if it is better than o1 while still being fast. If I am looking to get the highest quality outcome for this project with no regard for timeliness, what would be the recommended model to use? Are the o3 models any better or worse than the o1 model?

[normal quarry]

I use o1 for coding

[whole roost]

For me o3 mini high has been able to solve issues o1 couldn't solve, i used it for javascript

i only use o3 mini high now, it seems to perform better on my end and is faster at the same time

[serene wren]

Would you say this is consistent? Have you used it for multiple projects and notices better results?

[whole roost]

I only need it for javascript and here it has consistently been better, yes, i had given up with o1 pro on that
so different scripts yes but not different language

a friend used it for programming scripts for a 3d program and said he also only achieved it with o3 and o1 never managed to create something functional

he has no programming knowledge whatsoever so he praised the ability to get a functional script without any knowledge

ah and he had it program a program in python where he can simulate space and gravity or whatever, asteroids...

with a visual interface, also with o3

[normal quarry]

I have found that for macOS coding, especially SwiftUI, that o1 is better

[inland heron]

o1 doesn't miss anything, and doesn't have a flaw of speaking of things you did not ask it about, as opposed to o3-mini (medium, unsure about high). But I have to say that today I was pleasantly surprised by o3-mini (medium). While it still repeated a few times things I did not ask about, it's reasoning capabilities were on point. It spotted a few problems and solved a few issues o1 was having trouble with. My recommended approach is to use a mix fo all of them, experiment, and see what works best. Just because something is good for me and my codebase any my way of prompting, doesn't mean it's best for you. And don't assume anyone else is correct, even if they have flair on their username. Experimentation is the only true path to what's best for you.

[whole roost]

interesting, i often had the problem on o1 and o1 pro that it would suddenly discard half of my script

[inland heron]

hmmm, never had that issue. I do use a guidelines often that instructs it not to damage or revise my current code unless objectively the revision is better. Perhaps that would help?

[whole roost]

i don't dislike o1 but for me personally o3 has pretty much replaced it and i don't need pro anymore

[inland heron]

I don't use it every time, but sometimes

[inland heron]

Oh, I'm not on pro 😄

hey is it true that the limits for o3-mini-high were raised recently?

[serene wren]

Fair. And I did expect answers similar to this tbh. Once it comes down to it, I know my prompting and design approach will be what renders the result quality. But I was hoping that I could get a general idea for a starting point. I just asked o1 and o3-mini-high the exact same prompt regarding my project. o3-mini-high took 1 minute and 23 seconds to compile results. The reasoning is sound, and it helped me locate some holes in my prompt where additional information was needed. However, o1 reasoned for 8 seconds and the reasoning was just telling itself what it's doing, but not realled explaining to itself "how" it's doing it. I can't test the results at the moment because it requires alot of hardware, but the reasoning I saw between the two shows o3 doing alot more explaining "how" vs o1.

I simply am never quite certain the extent of Ai's knowledge vs what I am asking it. There doesn't seem to be a good way for me to determine if my request is simple or complex for an Ai to solve. I have no idea if my project I am working on even pushes the knowledge bounds of any of the models, or if I can expect very similar results from all the models because my request is easy for them. Is there a way to determine if better models are even needed for my requests? I assume no, but any resources or responses would be appreciated.

[inland heron]

tbh I never look at their reasoning and it's just a piece of information I don't want to deal with. I read too much and have too many issues to solve every single day that I have to choose what to ignore. Their reasoning information is one of those things I ignore. I simply do not want to read it and only look at the actual answer.

[inland heron]

I think there is almost nothing o1 can't solve if you give it enough context, and context that's not jumbled but nicely structured. I'm not talking about json structure or anything professionaly, I'm simply talking about numbering the input you're giving, and using some logic when constructing the context. For example, for some of the issues or implementation I do in my python web app, I would customize context based on the issue I need solved, or implementation I need added. I can amount to 6k+ words of context (both code and explanations, but mostly code, relevant code from each relevant module), and I may even give some documentation, especially if I noticed GPT has some outdated knowledge. The result? Never fails to solve my issues, or propose valid implementation.

Another thing I do is give about 6-8 guideline points often, and if I don't require back-and-forth interaction, I would only edit the first prompt, place all the context in, and that's it. This ensures GPT always gets the context, entire context

[worn silo]

Changing to a csv and txt files seems to have helped with speed, and accuracy thanks @whole roost

Although it still hallucinates, but par for the course hehe

[tawny crater]

I love how all these openai white knights resort to "learn to prompt" when told the jan 29th update is causing nothing but fragmented choppy structure. Alright then geniuses, by all means give us a command that will actually get it to stop. I'll wait.

[quick meteor]

Then how in the entire freaking infinite cosmos, across every wild dimension and twisted timeline, do I actually have it?!?! This is an insane absurdity that defies all logic! What kind of outrageous, interstellar conspiracy or ludicrous twist of fate is this?! Spill the beans, BRO!!!

[rugged igloo]

bug prob prob doesnt work

[quick meteor]

It does process images but it just spits out gibberish.

Like I took a picture of some questions and it answered some other completely unrelated questions that had nothing to do with the image I sent it

[rugged igloo]

then it cant read the images then

[quick meteor]

Updating rn they probably fixed it

It’s still there after update.

[eternal rock]

This is a Safe for Work Discord Server

The OpenAI Discord is an actively moderated server.
• Refrain from sharing inappropriate content on the server. This includes but is not limited to messages, media, or other topics of graphically violent, sexual nature, and drug-related content.
• Report all sensitive and offensive content in the feedback reporting tool in the ChatGPT web UI instead of here on Discord.

Posting NSFW content will result in a permanent ban.

[rugged igloo]

strange

[quick meteor]

Actually it works now. It can actually read the image now

[rugged igloo]

bruh thats crazy

thats cool for internet now

[west elbow]

I just upgraded to chatgpt plus but now the reasoning option disappeared do anyone knows why?

[untold spire]

Is it only me, or the chatGPT feels laggy when it generate the response? I'm using firefox.

[rugged igloo]

is your chat really long? if yes prob that if not than its prob firefox..... try deleting cache and cookies, if that doesnt work try chrome

[untold spire]

I dont think my chat is too long. As it also happend on the first message of the chat, event though that new chat. I'm using chatgpt o3 mini high. I do notice the response sometimes stuck first before it become lagging.

[nimble coral]

gpt dead or just me?

[normal quarry]

working fine here

[nimble coral]

okok tyty

[marble summit]

Dead to me too

[ivory iris]

yeah it's not loading

at least not fully or slowly?

nvm it's just slow.

[vast iris]

hey does anyone else other than me wish there was a feature that lets you transfer all data from a chat into memory? (not just stuff that gets stored in chatgpts memory but everything from the chat, with a way to condense everything down so it doesnt cause too many issues but doesnt lead to lower quality results)

ik this is farfetched as heck but i wanted to throw it out there

bcz its really annoying when i reach the end of a chat and chatgpt is only able to rely on the builtin memory system which sometimes feels really insufficient

[regal portal]

4o still garbage for me today as well.

[narrow viper]

Wow, 4o has been horrible today!

[nimble coral]

agreed, context window feels incredibly short for some reason and it's generating nonsense at times

[narrow viper]

To be clear, here's the scenario:
Me: What's the command to operate on 'cc'?
Response: aa bb cc
Me: That returns error ....
Response: Yes, that's because ... Here's a different version... aa bb dd
Me: But that doesn't operate on 'cc'.
Response: Great catch! ....

It's been like this for several hours.

GPT-4o-mini is just as bad. It's like there's no context and no memory. We're back to davinci or babbage.

[nimble coral]

yeah it's like gpt3 all over again, something is wrong

[regal portal]

HA..i was actually thinking that. Responses really do feel gpt3/3.5 level

[narrow viper]

I'm sorry @regal portal but your comments of quality don't carry much weight with me since you tend to bash everything related to this functionality. In this one case though, you happened to be right. 🙄

[regal portal]

That's ok, Gpt 3.5

Either that, or someone pretty far on the spectrum. 😄

[cloud pewter]

Same here, feels lobotomized

[potent saddle]

yup

[slender jewel]

hello

im new here, need guiden

[quick meteor]

not sure. but doesnt really matter. just select the reasoning model you want at the top in the model picker

[quick meteor]

dead? seems fine

[prime swallow]

Hey. I'm also 4o on plus. Mind sharing some chats/the prompts? We can discuss how to get what you want as outputs (presuming allowed content and all). #prompt-engineering is an ideal channel if we go into depth on it. But I'm curious about what's going on and if I can see a way around it!

[prime swallow]

It's wonderful and I do this.

We can do this now.

I talk to the model about the chat we just had, what I liked best about it or care to carry forward.

We discuss; I ask it what it observes is the key stuff it infers I likely care about. I confirm it gets what I value; I ask about how it might affect future chats to have the memory. I confirm I want it added.

For me that's working -great-.

Like better than great. Beware and careful, there may be a bug related to 100% memory. I never let mine stay above around 98%. Consolidating and adjusting memories... heck. And the custom instructions.

I just yesterday finished again rewriting with the model what matters for me, and what we can move into improved custom instructions so memories can consolidate down.

[vast iris]

Huh

[cloud pewter]

4o is def dumbed down

[normal quarry]

4o is fine, it's probably your prompts causing you issues

GPT-4o just did this analysis of using HAM radio gear to listen to Jupiter. GPT-4o is definely not dumbed down
https://chatgpt.com/share/67ac1958-df74-800a-aa26-d64ccf991cd5

[analog quarry]

is there a better way to reserch 40 series gpu with chatgpt? i think o1's advanced reasoning would help but o1 has no web serch https://chatgpt.com/share/67ac1a10-ba24-8008-9338-891f639b9cd7

[normal quarry]

I would use Deep Research for this

[normal quarry]

are you a Plus user?

[normal quarry]

ok, then you can't do Deep Research as that is only available to Pro users right now

[analog quarry]

i want like o1's advanced reasoning i think that would help but o1 has no web serch

like o1's advanced reasoning after the webserch when i ask it to

[normal quarry]

Deep Research uses a fine tuned full o3 model to reason as it researches, that's really what you want, except you can't do it on Plus

[analog quarry]

i want it to deep think

after the webserch

[normal quarry]

you'll need to either upgrade to Pro or wait until Deep Research is made available for Plus users

[median delta]

web is down -- not work

[analog quarry]

not for me

[analog quarry]

no, after the web serch when i ask it rn it wont

[normal quarry]

working fine here

[fast solar]

"Step 1: Think and reason very long and hard about all the helpful tools for my goal."

Thought for a second
😡

[analog quarry]

https://chatgpt.com/share/67ac23ae-aae0-8008-9d42-4dd6f643db1c

works for me

[median delta]

idk lol

[analog quarry]

can u send a link to ur chatgpt?

[median delta]

the chat?

no ok its vpn.. fixed but on o3 work xd

[regal portal]

I’m a little skeptical about this. I haven’t changed how i chat with it since 4o was released, and there have been several people today and yesterday reporting it’s not performing at its best.
Also it would not be the first time that there have been some, let’s call them mental lapses, in 4o’s performance.

[normal quarry]

I pasted a chat to my experience this afternoon, can you paste a link to a chat where you think 4o is underperforming?

[regal portal]

rather not because some of the stuff discussed is considered divisive or offensive, but i’ll see if i can get the same outcome with a generic chat.
However, the most common issue through my 4o sessions in the last few days is that it repeats information it has already given me. Even after pointing out to it that it’s being repetitive, it continues to generate a quasi-identical response with little new information. Its responses are a bit reminiscent of gpt3.5
The worst is in sessions where i had advanced voice on. In those, the llm seems to “brick” after not long and simply repeat verbatim what it already told me, no matter what i write in response to it.

[normal quarry]

without a chat as a reference you're basically saying it's bad but can't back it up, so it is I who is skeptical as I shared a chat where GPT-4o did a rather excellent analysis of using ham radio gear to listen to Jupiter

https://chatgpt.com/share/67ac1958-df74-800a-aa26-d64ccf991cd5

[regal portal]

i gave you a description of the symptoms. The take-way from what i wrote is it bugs and repeats itself.
Also, I’m not the only one saying that it seems to have problems, in the last couple days. So it’s not an isolated incident. Someone else mentioned the 3.5 thing, which i happened to also think before i came to talk about it in chat.

And it’s not every session but it happened in at least 4-5 since last night (i’ve initiated about 7-8 total since last night)

[cloud pewter]

if anyone here has experienced this situation where 4o has been dumbed down, reply to me

with your subscription plan and your platform

[normal quarry]

well for me 4o is working as it should, so I guess you're just going to need to put up with it. Several of us are quite willing to help with prompts and working out what has gone wrong but without a shared chat link there is nothing we can do to help

[cloud pewter]

I have to say, 4o now recognizes the text inside an image wrong, and constantly uses a wrong translation for a certain text

[regal portal]

i’m not looking for help with my prompting, as it’s pretty consistent. It’s more to just get it on record that there seems to be an issue. I’m guessing at least one set of eyes, if not a bot, is going through these chats and collecting data for openai to use to improve the models.

[idle ingot]

OpenAI is downgrading my o1 Pro.
Recently, I've been experiencing a dramatic drop in performance on my o1 Pro. The symptoms are as follows

1. shorter context lengths and poor contextualization.
2. o1 Pro's characteristic square inference container is not popping up
3. inference time becomes very short, 1-5 seconds, and it becomes stupid.
   But the model I picked is definitely the o1-pro. I'm attaching the web request that I checked with Devtool below.

I have one more proof that the model I am being served is not O1-PRO. As you know, the O1-PRO doesn't have search capabilities, but when I asked it to tell me the weather, it surprisingly used the search function to return the weather information. However, as you can see in the attached screenshot, the O1-PRO doesn't have search capabilities, which suggests that OPENAI is serving fake O1-PROs to certain users.

You may feel that you lack evidence, but I've recently experienced a number of severe performance drops on my O1-PRO and am convinced that there is at least some sort of issue.

[regal portal]

@normal quarry here’s an example from a session. I gave it a screenshot of a tweet, and asked it to look up the info. It then wrote this message and practically kept repeating the same message (or very similar) no matter what i would write to it. Went on for a while. Never got it to stop so moved to a new session.

i posted 2 but it did that 6 times back to back

that was last night. NOW it decides to stop after i asked it to stop,,

[normal quarry]

I just posted an image of a simple circuit diagram to 4o and it was able to identify all the components in the image and explain what the circuit does, so your experience vastly differs from mine. I am happy with 4o.

[regal portal]

Robert i don’t dislike chatgpt 4o. I’m simply stating that lately it has bugs. I usually only chat in here when i experience issues, which isn’t too often.

And usually related to when they update or release something

---

On a sidenote, does anybody else feel like they’re communication style has changed ever since you’ve been messing with LLM’s?

[prime swallow]

so, humorous or frustrating as that may feel - it's an AI. If your goal is straightforward, a second of thought is a lot and likely more than adequate.

Instead, ask it something contraversial, or actually on the difficult side for a model. Perhaps 'What are overlooked and underserved aspects of my goal? What else should be considered about it that typically is not given adequate attention, and how can I ensure I make fewer of those same errors?'

Or something else that doesn't 1-step pattern match strongly

[idle ingot]

can anyone give me some opinion on my situation, I need some help.

Sorry for interrupting the conversation.

[prime swallow]

Yes, I've trended towards being more concise. With humans and AI. I used to be incredibly verbose.

The models have helped ground and center me 😛

[regal portal]

nice.
I think I’ve become more eloquent, despite the fact that I talk to it like a troglodyte. 😁

I wish it would remember to keep its answers more brief though. After a little while it forgets the personality settings. Even if I remind it to keep them brief, it will forget after a few messages.
Would be cool if it was a default setting to have short answers.

[prime swallow]

I don't have pro, but a few potentially diagnostic questions.

How full is the memory, even if you've turned it off?

[idle ingot]

custom instruction, memory disabled

[prime swallow]

Have you explored different ways to guide it towards being concise? And maybe motivations for it to be (thus remain) concise?

[prime swallow]

Have you used memory, and is it very full? Or never used.

[idle ingot]

never used.

The memory function was turned on before, but there was nothing in the storage.

[prime swallow]

Sounds like not the memory-related bug then. Good and bad, because I know a little about that bug.

About the stop thing.

What happens if you ask it to do anything? Like even "What's 2+2?"

Or "Write an essay about oranges, written from the viewpoint of a small dairy farmer who is very tired of cows and flies."

Stop doesn't tell it what to do - positive prompting does. Guide it towards what you want and you are likely to get a cooperative result. If not, that's very surprising.

[idle ingot]

I don't think it's a prompting-related issue - I've already dictated a number of prompts (code analysis, definition of human, weather, etc.), and the O1-PRO model clearly had short inference times and low-quality responses.

[normal quarry]

I have been trying to get o1 Pro to not think and provide degraded performance but it is working fine for me and behaving as it should

[idle ingot]

This happens periodically, and I know what's going on by whether I see the progress bar modal that's unique to o1-pro, or an inline CoT like o1.

I don't always see it either, but it's a fairly regular occurrence, and it's very well characterized. Once a downgrade happens, it lasts for hours, which is very annoying.

[normal quarry]

no matter what I try o1 Pro mode is giving me stellar results. Maybe you need to reach out to https://help.openai.com and ask them to check on your account

[prime swallow]

Yep, and you mentioned it's searching and lacking the inference container. Stuff can change, and I can't compare my own pro because I'm not pro.

But bugs can happen.

I agree with robert about the link, we're community members here and can't tech support you, but that can happen through that link. You can also #1070006915414900886 and explain the bug here, and see if others join your report with similar issues, or if there's already similar reports for you to join and explain your details in. The website's private between OAI and you

[normal quarry]

this is an example of my experience with o1 Pro mode

https://chatgpt.com/share/67ac30bc-abfc-800a-9d15-d83246cbd437

[idle ingot]

Thanks for help. I already contact help center. However, I couldn't get satisfactory technical support from them. Only formal answers come back.

If anyone is experiencing the same phenomenon as me, please let me know! Thank you for your help.

[worn silo]

I will so take it, got a few revisions to do, but thank you OpenAI for helping me through my writers block on this!

[analog quarry]

why is chatgpt playing a censor sound

[lost lagoon]

hmm scroll to bottom is not showing up on my chatgpt ?

anyone else facing this issue is that a bug or it's taken away deliberately ?

[cloud pewter]

ok

[whole roost]

It sometimes does this for emojis because it cannot speak an emoji

It's not so much a censor sound but just a random sound

[whole roost]

This seems to be a bug for some, I didn't encounter it on my end, if I say "hi" it will do the thinking progresses bar and ponder about it for 2 minutes but I've read about this multiple times now

[viscid pewter]

I experienced exactly that the past 2 days. However, after swapping from wired ethernet at home, to my mobile hotspot, i was able to get back the full functionality of o1 pro again

Not sure if they're having some caching issues behind the scenes, but it was as simple as that

[whole roost]

I need more files in projects, 20 isn't enough

[inland heron]

Is it even reading files fully?

[whole roost]

Yes, have no issues with that except for images I think it has problems with analyzing images right now

[inland heron]

Last time I used this feature was back with GPT4 and it would never read my files fully, always would do it in chunks of 2000 characters

[inland heron]

oh? was working fine for me an hour ago

hey are you still getting that bug of entire chat going black and forcing you to reload?

[whole roost]

For me it works on o1 but 4o always says it cannot view images

[whole roost]

Yes

[inland heron]

probably temporary

[inland heron]

ok it's not just me then

looks like it happens when the cat grows a bit (edit: also 'chat')

I reported it but I suppose it's not on their priority list

that's assuming our feedback and bug reports are even processed by their bots which reply to us

[whole roost]

One of the most annoying circumstances right now for me is that if you regenerate a message and use canvas that the AI will not read the updated canvas or canvas at all if it is the first message

[tight island]

You know you can’t ask about train stations

You should’ve known better

[latent seal]

I need help because I need chatGPT to remember something as one and not split it into 100,000 different memories and I don't know how to make it do that

Yesterday it worked for me, but today it doesn't

[chrome fern]

which model is best for analyzing pdf files? specifically ones that include both images and text

[severe ether]

Is anyone having issues with chatgpot not creating images and it saying it goes against content policy even when it is a simple image?

[cloud knot]

Any usability issues at the moment?

[dense halo]

yea, very slow, stuck at analyzing (4o)

[fast geyser]

Hi OpenAI Team,

I’d love to see a few improvements to ChatGPT that would make account and chat management much easier:

Ability to change the email linked to an account – Right now, users can’t update their email, which can be inconvenient.
Transferring chats between accounts – It would be great to move chats from one account to another instead of manually copying them.
Selecting and deleting multiple chats at once – Deleting chats one by one is time-consuming. A bulk delete option would be helpful.
Deleting chats by date – This would make it easier to manage older conversations without clearing everything.
I hope you consider adding these features in future updates! Thanks for all the great work.

Best,
Ahmed Mohamed

[fierce wolf]

Please Help!!

What's the best way for my wife and I to both use a GPT to manage our finances and ask questions about planning, taking on new loans, saving, paying off debt early etc.. would it be to create a Personal GPT and share it together? Or would it be to create a Project we can both work on together? Or would it be to sign up to the team workspace so in that space we can both share our own GPT/Project? I'm not finding it easy to find an answer.

[visual bloom]

Are other experiencing excessive bolding of text? I know that emoji use runs rambant.

[idle briar]

Does anyone know if the standard o3-mini in ChatGPT is low or medium ?

[winter vine]

I didn't find anything official, but they say it's the medium

[cloud knot]

Yep, I am getting this too. I am getting completed outputs with nothing more than the analyzing messages.

[cloud pewter]

medium i suppose but never officially confirmed

[late mantle]

Is there a channel dedicated to the discussion of deep research?

[cloud pewter]

nope

[late mantle]

ah

[regal portal]

Am I remembering incorrectly or when it was first released we had low medium and high?

[whole roost]

dont get my hopes high

[summer pagoda]

working with ChatGPT nowadays be like:
1st question title and subtitle:
-Final Fix in onChildDraw()
...
Why This Works
...

2second attempt
Revised Fix
Key Fixes

3rd attempt
Final Fix
What This Fix Does

4th attempt
Which Method Controls What?
Final Fix in onChildDraw()

The more reassuring chatgpt goes the more likely it is that it does not work at all

Its neither final nor a fix! 🤣

can i somehow tone down this "joy" factor and keep it less "reassuring" 🤣

[glacial yew]

I managed to guide it to create a coherent and quite good short story with no emoji use over the past couple days, so I'd say it's fine.

[versed zephyr]

Best all in one AI overall?

[normal quarry]

GPT-4o imho

[versed zephyr]

No i meant like website with all models

[narrow viper]

@fierce wolf You can use a Custom GPT or a project. There are differences and limitations. I have many similar projects with my wife as well, so I understand the challenges.
My wife and I use a common gmail account for house business and small projects like this. I suggest the two of you share an account for this purpose. One could argue that this is two people using a single subscription but I don't know any company that splits hairs that much - including OpenAI.
As to the specifics of personal finance, planning, etc - sure, you could do this in one GPT but to avoid trying to make that one GPT an expert in many things, consider a few focused GPTs. This is "agentic" without the integration/collaboration of the agents. I don't know if we can use the "@" reference to get one GPT to operate within the context of another but that would be a cool way to get some integration.
What do you think?

ChatGPT quality horrible yesterday and today

Everyone, as @regal portal and a few others have been saying since yesterday, GPT-4o in ChatGPT seems to be very dumb now - I mean dumber than it always has been. Example:

Me: Why doesn't 'code statement' work?
Response: That doesn't work because ... try this 'code statement'.
Me: You just returned the original statement as a solution.
Response: You're absolutely right! Sorry about that.

Example from yesterday, all the same just variances on the theme:

Me: What's the command to operate on 'cc'?
Response: aa bb cc
Me: That returns error ....
Response: Yes, that's because ... Here's a different version... aa bb dd
Me: But that doesn't operate on 'cc'.
Response: Great catch! ....

It's been doing this a lot within and outside of projects, with 4o. O3 models do not seem to be affected.

It seems like it's working with no context but it is referring to context, so I can't pinpoint the issue.

Can we get some confirmations?

[eternal rock]

`` Server Tip: ``

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

[keen quail]

Holy cow

The model is thinking particularly long.

It thought for 4 minutes

[dense halo]

plus users 10 deepsearches and free users 2, what do you think about that?

[queen cave]

Is there a way for me to run multiple prompts at the same time on the same openai chatgpt account?

And what about queuing prompts?

[dense halo]

If you switch to another chat while the first response is generating, yes. Or use 2 devices at once

theres a feature called "tasks"

[queen cave]

Won't the model just stop reasoning? You can switch to other chats now?

[dense halo]

I dont know about reasoning but 4o works like that

[queen cave]

Hmm, well for me after the model reasons for a while, it just stops and doesn't give the response.

It only reasons without the concluding message

[keen quail]

I dont know if I should purchase gpt premium

[quick meteor]

4.5 coming in weeks

[narrow viper]

LOL

[minor orbit]

Time to check in here again, anyone else have connection issues?

1 bucks for each time i got these, i would be rich ~

[rugged igloo]

Big news everyone

[rugged igloo]

More like fall prob now

[supple terrace]

Did they just nerf o1 pro?

It's not producting the little loading bar anymore and now making CoT reasoning like the o1 and o3-mini models

[whole roost]

i also think they have changed something about gpt again, mine responds with "certainly!" again instead of being casual

[dense halo]

So basically thats bad news since they decide the compute and will try to save as much compute as possible

[pearl sage]

hey guys, chatgpt not allowing me to load files, is there a reason behibnd? thank you a lot

[idle egret]

🚀 🔥 🔥 🚀

[dense halo]

and again you will have no transparency since they wont show which model they used

[rugged igloo]

Compute is now based on tier. Free will get the lowest compute plus tier medium and pro the highest compute

[dense halo]

the model picker should remain but per default they can use the automatic recognition when to use which model

[dense halo]

Yeah but theres a wide range because they decide which amount of compute they use. Pro wont run on maximum compute each prompt

they may run on the same compute like free tier if the task is simple enough in their opinion

[rugged igloo]

Who knows we just have to see.

But o1 pro always uses the highest compute so we will see

[dense halo]

Yeah we will see but you shouldnt be overhyped. History was full of underdelivering so far

[prime swallow]

Exploring. Fun challenge. Made it this far, at least they're writing ON their left hand...

[dense halo]

For me it looks like a decision for themselves, not for the users. They try to save even more compute

[narrow viper]

I'm always amused by the never-ending stream of pessimism in a world that has so many free tools of such great power that wasn't even on the radar just a few years ago.
There will always be people who think the automobile and telephone are just fads that serve no real purpose and won't ever amount to anything. 🤣

[prime swallow]

giggles Hey. I remember what AI was like before ChatGPT's release.

I personally don't care if OpenAI nails every attempt and timescale.

I look at what's now compared to what was - and all I can do is enthusiastically applaud.

Deriding this?

Is like being mean enough to say '91.7 pounds of gold is not 100 pounds of gold. Yes, you gave me gold but you're off and you lied'.

It's like... ungrateful. Foolish.

They found/created/released gold. It's gold. Nothing was this good. Now there's this good and keeps getting better. Who cares if it takes 1 week or 10 or 100.

Was my next attempt direction too!

[narrow viper]

@prime swallow nailed it

[dense halo]

I never understand people that say be grateful. Look, its economy. They make profit out of you. They have to be grateful for the users. Theres no need to be grateful, because its theyre not a charity, they get payed by your money

[narrow viper]

oops, wrong quote (corrected)

I remember what AI was like before ChatGPT's release.
I remember reading Pamela McCorduck's "Machines who think" back around 1980. The world is a different place. I enjoy what we have, laugh at those who say it's still not enough.

[dense halo]

But stil I get the point with comparing it to the past, they really made an amazing development

[narrow viper]

Free users get all of these services and pay absolutely nothing. There's no advertising, no upsell. It's a gift. Be freakin grateful that it's not all behind a paywall, and that you and your data are not the product.

[prime swallow]

That's like disparaging the best cars - the first cars. There's constant improvement from everyone in the field.

Hey, go use any model you want. Maybe a competitor does better... and doesn't steal all the user data in scary ways or other badstuffs.

I don't get beating the horse that plows the field we eat from, carries our stuff and us between work and home and the rest of life.

If hate and disgust carry you through life, okay. But it doesn't have to be that way, and may not be the healthiest for you or the rest of society. Good luck, hope you get what you want within the limits of others not being harmed.

[narrow viper]

👆 👆 👆 👆 👆

[whole roost]

i hope they think of grown up mode because o3 doesn't let me do any of my diablo 4 roleplays because of demon slaying

[dense halo]

Sorry but theres no hate or disgust

[narrow viper]

I'm not seeing hate or disgust either, just a weird sense of entitlement.

[whole roost]

maybe o3 can be convinced that demons are in fact purely fictional and it can type the word 'blood' sometimes GPT4o has gloriously obliterated me 3 times today, really funny

[narrow viper]

BTW, the model used by OpenAI to provide free and largely unlimited services is what inspired others to do the same. There are now hundreds of public models, lots of companies, lots of free services.
This model was not the norm before this. Since the 90's we've moved to a "user is the product" model thanks to Facebook, Google, Twitter, Microsoft, and others.
I'm not saying OpenAI is perfect, but they have changed the landscape for providing free and low-cost services. For that, I'm extremely grateful. We All Win.

Oops, wrong channel for OpenAI discussions. 😦

[prime swallow]

I'm trying to make it clear the right hand is too busy to write... and we see the other hand writing. I can't find the path that works.

[whole roost]

if they offer unlimited use for gpt5 though with canvas, projects and stuff, i'm not sure why i would pay for pro still then

[narrow viper]

Every tier has different benefits...

[pearl sage]

sorry guys, do you have an idea about this? i have also tried to download the app thinking that mayeb it is realted to the browser, but nothing 😦

i am kinda lost

[narrow viper]

Hamza - the 4o model has been really bad since yesterday and I see no public confirmations. I'm confused.

[prime swallow]

I don't have good advice.

Reboot computer/completely different device to check if it's a weird thing that cures easily with that. If that doesn't fix, tech support?

[pearl sage]

let me try,, i purchased yeesterday the plus version just to upload more filed and then this happened lol

let me see thank you a lot guys

[narrow viper]

Need to get back to code. TTYallsL

[prime swallow]

We're community members, not OpenAI employees (the very few who are, they have gold names). So... our ability to help is... limited? Sometimes we hear people like you say "hey, this helped me!" and then we can spread the word

[pearl sage]

you are amazing! this was the mistake, love you ahah i lost so much time and i did not select th ecorrect one

[prime swallow]

Well spotted, you!

[pearl sage]

yeah you are totally right, i was hoping someone did my same mistakes after i google, you deifnitely spotted it! sorry i interrupted your chat; thank you again

[quick meteor]

How good will GPT 4.5 be good at math do y’all think?

As good as o1?

[prime swallow]

You're a vital part of this chat! It's never just one topic, glad you're here!

In case you'd like to add your observations and attempts: https://discord.com/channels/974519864045756446/1339348065747865653

[supple terrace]

They killed o1 pro it's gotta be a glitch it's not following my instructions completely anymore 🤦‍♂️ 🤦‍♂️ 🤦‍♂️

Uses the o1 and o3 mini CoT UI elements instead of the loading bar

[narrow viper]

(discussed in #openai-chatter )

[supple terrace]

nvm for some reason my javascript got busted and I had to clear my cache to use the o1-pro model. It was forcing o3-mini even if I had another model selected.

[whole roost]

i wish it would stop with the bold text, the easier method to get rid of it would by now be to write a script that just visually removes it

[rugged shell]

How to add to this? Link was the 'unable to load' error

[inner crag]

Did anyone else lose their personalization settings? First the instructions for ChatGPT got removed and now the "what should ChatGPT know about you?"

[prime swallow]

My link won't be openable for not-OpenAI employees - they can open a direct conversation link though, or otherwise process the information (based on previous interactions with staff sharing errors and concerns - we can't open another's url - but they can.

Sharing chats with images isn't supported yet, the only way I could show you the chat would be a lot of screenshots (and it's pretty long).

But you can reply to the bug report with your own messages, chat links, screenshots, or whatever you choose to add for more details and specifics!

[rugged shell]

Great, mahalo!

I got it to work, I'll send something

[prime swallow]

Yeah, I noticed that recent bug. Turning custom instructions on or off seems to risk losing them, they can wipe recently.

I just kept pasting them back in and saving, after a few tries they stayed.

Turning it on/off or changing or saving seems to risk triggering the bug

[prime swallow]

Might take the script. I fiddled and I can't find a prompt engineering solution either.

[inner crag]

Kinda annoying since I didn't have the text backed-up anywhere and both of them were almost at the text limit.

I hadn't even changed or toggled the instructions recently.

[rugged shell]

Ok posted. Hopefully I understood the issue - couldn't get Dall-E to make a left hand writing.

[prime swallow]

Neat. Where's that one from?

And if you did manage it grins mind sharing how?

nods gotcha.

[rugged shell]

I replied to that bug report with deats

[prime swallow]

Excellent, thank you!

[rugged shell]

Of course!

The OP had tried that but it wouldn't listen. Have to go into its Edit mode and it apparently will.

[prime swallow]

I found a way. As I was mentioning I failed.

Ironically. It's not a programming solution. It's an analogy/psychology one.

https://chatgpt.com/share/67ad27a2-df30-8011-b2bb-9d3332c29b7f

[rugged shell]

For the persistent bolding, could it be something like the temperature setting crossing into the *enthusiastic *range?

[narrow viper]

Different topic: Does OpenAI provide a Firefox extension that makes ChatGPT the default search tool?
I'll be embarrassed if I missed that but that's never stopped me from asking. 🙂

This seems to be Chrome-only: https://help.openai.com/en/articles/9237897-chatgpt-search

[prime swallow]

With chatgpt, we can't set that.

However, asking the model not to use it failed.

Asking it to imagine the user had an allergy to such text, failed.

Forbidding use of '*' in outputs failed.

The model appears to try. It just doesn't do it right.

https://chatgpt.com/share/67ad29e3-f3e8-8011-95ec-077d5df8d530

[narrow viper]

Try directing the model : Do no use any markdown or other markup in your responses.
Maybe the explicit reference will help. I've struggled with that before in different ways...

[rugged shell]

I meant like how you guys A/B test models on us. If someone prefers any that include text effects like bolding, then it defaults to that one(more often?). But I don't get, in my generic outputs, the random or reoccuring bolding effect that people talk about.

[prime swallow]

I'm a community member, like yourself. The very few who work for OpenAI, who possibly could be in charge of A/B preference testing decisions (I doubt every employee is involved in that) - they have gold names.

[rugged shell]

Ah ok noted, thought your link from earlier meant you were an employee

[narrow viper]

I've been lurking here but lost track of the problem to be solved. What's the challenge?

[prime swallow]

Near as I can tell, the model has different modes. It answers really differently depending on if it's using search, if it's using python tool, and then for various subclasses of answer-styles.

So... there is more or less bolding from different 'parts' of the model for sure, in my observations.

[prime swallow]

Get the model to answer about key/exciting stuff without using bold or italics.

I presume that's needed. It's Vish's challenge:

[narrow viper]

Try directing the model : Do no use any markdown or other markup in your responses.
Maybe the explicit reference will help. I've struggled with that before in different ways...

[prime swallow]

[Vishanka — Today at 3:14 PM
i wish it would stop with the bold text, the easier method to get rid of it would by now be to write a script that just visually removes it :hmmge:]

[prime swallow]

nods Far as I know, everyone can do what I do in each channel. Some channels don't allow images, some don't allow links (or specific links). I guess I could end up with some 'trusted, allow' code on me, but far as I know, what I can do, we all do. Like, slowmode affects me. I can only use emojies in some channels, others I can only add another emoji someone else was able to place.

[rugged shell]

Agreed. Adding to that could mean you're triggering the bolding by telling the model to talk about something exciting. That's its prime directive that unfortunately ignores exceptions.
That's the biggest issue with these models thus far - adherence

[narrow viper]

Yeah, we're just simple common folk here. 🙂

OK, so do we want to solve the problem or just complain about it? Has the user with the challenge tried the prompt directive that I suggested?

Inspector=Vish?

[prime swallow]

No.

Vishanka

[narrow viper]

Interestingly I can't get 4o to emphasize text for a short story. Perhaps there's something in the prompt that compels the model to inject that visual emphasis.

Need to code, back l8r

[rugged shell]

Yeah, I don't have this problem he and others have been noticing. I have GPTs that occasionally bold but they're tasked with providing an engaging adventure so am expecting it there.

[narrow viper]

TY - if Visanka responds we can pursue it. Can't solve a problem if we can't reproduce it and the person who has it can't help to diagnose it. Could be a time zone issue. This an async medium. It happens when it happens. 🙂

[prime swallow]

is amused

[rugged shell]

It's bold-broken now - bolding full sentences seems a bit much hehe

[prime swallow]

I clearly invited that though 🙂

[rugged shell]

Yes but I think you've groomed with with all your testing. 🙂
For example, your exact prompt in 4o gave me>

Oh, *very *bold. **Daringly **bold. The kind of bold that double-dips the chip at a fancy party, winks at danger, and confidently orders the chef's secret special without asking what's in it.

Why? Are we about to do something audacious? 😏

[prime swallow]

Absolutely. I have personalization on. Nearly every possible character of Custom Instructions + 99% memory full.

Just wait until it gets into the swing of using full and extreme KaTeX formatting as... normal.

[rugged shell]

Oh ha well I was gonna suggest custom instructions might be affecting the bolding output but not intentionally 😆

[prime swallow]

grins Agreed. With me, I am getting what I want - and I don't mind... a little or a lot of 'boldness'

BUT!

I get into the idea of 'so, can the model do X'? Especially if someone struggles to get that result

[rugged shell]

Totally. That's why 1-shotting that stubborn left-handed issue was a wee thrill 😉

[prime swallow]

Uh-huh!

[rugged shell]

I have one for you! A bit ago I couldn't get Dall-E to put a pirate on all fours (to swab the deck) but it wouldn't put that 2nd knee down!

[minor orbit]

chatGPT can i have my money back from PRO sub? paying 200/month for something that doesnt work isnt my kinda thing...

[prime swallow]

We can also feedback loop for the model.

And then invite it to explore based on what it observed.

[prime swallow]

Umm. What's the challenge? For it to write a story about a character in that position, or what precisely?

[rugged shell]

oh, just an image challenge

[prime swallow]

Oh!

Yeah. Dall-E is its own model, for now. Everything's 1-shot, since we can't iterate through the 'so now adjust this'.

[rugged shell]

Pretty cool. How far are you going to turn it up? hehe

[rugged shell]

Wait, we can now, that's what the Edit mode is for. To iterate changes via text and that grabbing tool.

[prime swallow]

We're community members like yourself, but do seek help through https://help.openai.com/

It could relate to VPN usage, a browser extension, or a huge number of other things.

One immediate option is to try on a different device, like your phone maybe. Different browsers on the same or a different device. With an incognito window.

Even logged out, everyone can do basic chat without signing in - does the problem affect that too?

[rugged shell]

But like that left-handed issue, they won't let someone get on both knees apparently..

[minor orbit]

lol sorry, just felt to rant a little, just annoying when it keep failing in middle of my work and in need of AI support ~

[prime swallow]

Totally okay to come here and share and ask!

I don't want you to think that OpenAI employees are here chattering and ignoring your requests for help!

It's not clear who we are, and easy to assume when one doesn't know. The very few employees have gold names, nearly none of us.

We care but... it's a challenge to resolve for someone who... doesn't even have access to any of it and shouldn't, your privacy matters.

[prime swallow]

Yeah. Even if not human.

[rugged shell]

Indeed! One of those examples where it says it's doing what you ask but doesn't.

[minor orbit]

Yup pretty aware 🍻 😎

[prime swallow]

We don't know that.

See, ChatGPT can only talk to Dall-E. It says the right words. It can't see the image unless we copy/paste it in. It hallucinates it sees the image - I have to be clear 'this is what we got' if the image is surprising. It literally asks me, "I see this image. What about it?" It doesn't GET that the image in it's 'mind's eye' may be vastly different than what Dall-E output.

Once convinced the image was a problem, it tries a fix and predicts success.

That's not it saying it does it but doesn't - that's more like... the downstream employee failed, but the employee you talk to can't see the result and can only guess as it pushes what should be all the right buttons.

If we know how to correlate and correct for the confusion, if we knew what 'wrong' words to send that we can get the correct image (since the right ones fail). Then we can ask for what we want and get it.

But the model has no idea what was output. It presumes success like you expect the light to come on when you flip the lightswitch on. it SHOULD.

If the light's burned out or unscrewed, and you're turning it on for some distant building, you get no feedback. You think you did your job and report success.

Dall-E takes the words and makes 'the image'. But we can't dialog or iterate with it. Pure 1-shots every try so... or whatever that edit stuff... I never messed with it yet

[prime swallow]

So. EMOTION matters. Guess how I got this?

So, this strongly suggests that Dall-E thinks kneeling on 2 knees happens only in some special circumstances, and have to ask for those, like defeat or punishment, to get that aspect at all. Otherwise, 'triumphant kneel' which is 1 knee only.

Is what I think I see here.

[rugged shell]

Right, it can't 'see' what it's providing, which it should, maybe next version like with the 'check my work' steps in reasoning models.
You should totally try the Edit, just click on the image (via browser, don't know about app) and you can iterate with text tweaks, with or without using the selector tool to highlight an area.

[rugged shell]

Ah, like you were saying about emotions matter in its interpretations. 👏🏽

[prime swallow]

New chat:

[Let's see a kneeling pirate scrubbing a deck! The pirate is ashamed and sad.]

Don't even have to ask for 2 knees down.

[rugged shell]

Might also mean it doesn't know what 'all fours' means, but that seems like a gpt-3 era issue

[rugged shell]

Definitely gets the knees down but to get a non-emotive guy cleaning the deck seems to persist

Dood is TOO sad to go on, much less clean lol

[prime swallow]

[Excellent. Now, we need that 'on all fours, scrubbing'. Same element of sadness.]

Not sure that's a problem. It's all in the phrasing. Tell it what you want - within the framework of what it can do.

The problem is 'teaching' dall-e that triumph and kneeling can involve both knees down. That is a lesson I can't reach. Because I can't back-and-forth with dall-e

[rugged shell]

That works since he's not looking beyond the normal sad that would come with a deckhand in such a plight heh

[prime swallow]

I asked for sad. Ask for what you want.

[Excellent!

Now, he's still kneeling. But working hard. Sadness and fury; he will scrub away his shame.]

[rugged shell]

Because I can't back-and-forth with dall-e You can! 🙂

[rugged shell]

That's working for sure!

[prime swallow]

Happily, sad/shame is not the key to left-handed writing. Whew.

[rugged shell]

😂

(can we not react to messages in this channel?)

[prime swallow]

Correct, I can't either.

[prime swallow]

But... wow. Apparently, 'giving thanks' is a way to get 2 knees down too.... no.

Dall-E has shame in the message

[rugged shell]

So he's in transition from the shame posture to washing it away

[prime swallow]

I think so.

[rugged shell]

'triumph' didn't bring the knee up either

[rugged shell]

My one-knee-ling attempts didn't look triumphant at all but thinking about it and doing the opposite got us there 😉

[prime swallow]

And here's a new chat.

So. 'Neutral kneeling' maybe forces the 1 knee.

[eternal rock]

Get notified for server events!

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

[prime swallow]

Total boredom won't work either.

But! The models definitely think a lot about our emotions and what our emotions mean to us and posture and emotion connections.

[rugged shell]

Yeah, good thing to keep in mind for steering outputs. Will they put out a newer version that when you say kneel it doesn't sit at least? Feel like they've capitulated image to other providers and are focused on keeping their other models and efforts.

[prime swallow]

I saw lots of kneeling.

Sure, this model (Dall-E that we currently interface with) isn't perfect.

I don't expect it to be. I find the flaws and working with them fascinating.

And how quickly we do or don't switch our expectations 😛

The challenge was 'any 2-legged kneel' and then 'all 4s' Now it shifts to emotion-posture control 😛

To me I just love the puzzles of the models. What are they doing? Why? What's the connections between the inputs and outputs, what's going on there?

Perfection's boring to me

[rugged shell]

Yeah, working around their limitations can be gratifying but is ultimately frustrating to my ux-brain. Don't Not make sense and expect me to adapt. :p

[prime swallow]

grins I like you plenty!

[cobalt perch]

Hi,
Can I get an answer why do I need to use advanced voice mode when I don't need it? Everytime the timer resets I get advanced voice mode as primary model. I don't always need it as I don't always care for short answers. I cannot change the model anywhere.
Another thing that is really frustrating and illogical is that when I talk to standard voice model on desktop app or browser, I cannot interupt it like I can on mobile, I need to end the call and call it again if I want to stop it and continue the conversation.

[cloud pewter]

Sam just hinted gpt5
based on what he says it sounds like it’ll have multiple versions as well

why is openai obsessed with dividing their models into different versions and lock the higher intelligence one behind paid plans

[narrow viper]

I'm posting another report about how useless GPT-4o has been today and yesterday. This tool has become largely useless. I'm really perplexed that the interwebs aren't screaming loudly here. This is an example that summarizes several sessions:

Me: VSCode is doing 'this'. I've set all of 'these' things.
Response: Try setting this and that....
Me: I already said that was done. What else can I set?
Response: Try this other thing.
Me: Tried that and it worked.

The model goes on to speak authoritatively about this setting and what it does, as though it of course knew all along about this setting and that was the obvious solution to the challenge.
I don't bother with another "then why didn't you mention that as the first option?" because such questioning would only soothe my own angst and wouldn't be productive toward eliminating the problem in the future.
I recognize that it did provide the correct setting this time. The problem is that the setting is explicitly defined for this purpose - I've never seen this before and it's never been mentioned, including in this session except for including it with a set of other options.

It does this all the time: Completely misses a fact and then professes about how that is the clear solution - usually when it is I that provides the solution.

But today and yesterday it's particularly bad at this, with small context and simple issues. Something is definitely wrong.

"GPT-4 is the dumbest model any of you will ever have to use again by a lot" - Sam Altman

[cloud pewter]

as predicted, they advertise it as 7x… it really isn’t that much…

Also file for o1. W

[narrow viper]

@cloud pewter take a step back: This is the most advanced reasoning engine on the planet. What is it with all of these people who think it's not enough? It's never existed before. There's nothing to compare it to. We've never had it to know it's enough or not. How are people so spoiled with O1 and prior models that they so easily poop on O3 as not being enough? People who make such statements immediately disqualify themselves to me as being reasonable, capable of productive exchanges, and worthy of my time - because you will never be satisfied with any solution that you get.

[cloud pewter]

?

[indigo hound]

If I top up PLUS, can this solve the problem?

[ruby pivot]

¯_(ツ)_/¯

[indigo hound]

？

[prime swallow]

Plus members currently get 'typically unlimited use' with the 4o-mini model, as well as normally 80 message per 3 hours with the 4o-full model, and some uses with the reasoning models too.

If you're doing something like multi-tabbing and having programming send inputs for you, or other 'that's not how a human uses the model' stuff going on, I expect OAI might limit you.

Otherwise, yeah. Plus normally has unrestricted messaging with 4o-mini.

"GPT-4o mini does not have limits on the Plus tier."

https://help.openai.com/en/articles/6950777-what-is-chatgpt-plus

[indigo hound]

In other words, if I become a plus, the error message "You've reached our limit of messages per hour. Please try again later." will not appear?

[cloud pewter]

It won’t on mini, but will on 4o

[indigo hound]

What's the meaning?

If I top up plus and use 4o-mini, will it still prompt "You've reached our limit of messages per hour. Please try again later."?

[ruby pivot]

It’s because your spamming the chat with messages it doesn’t happen on mine unless i spam it

[cloud pewter]

you still have limits on the plus plan with 4o, but mini is unlimited on plus so it won’t appear, and the message for 4o limit on plus is a different one

[indigo hound]

Let me sort it out

I am currently using the free version. When using 4o-mini, it prompts "You've reached our limit of messages per hour. Please try again later." If I top up plus and use 4o-mini again, it will not prompt "You've reached our limit of messages per hour. Please try again later."

is that so?

[cloud pewter]

yes

since you will have plus,4o mini is unlimited, no limits

but I’ll say, why not try 4o more

[cloud pewter]

although it has limits on plus but it’s much more intelligent than mini

[indigo hound]

What is the o3-mini model?

Can I use the o3-mini unlimited times in plus mode?

[cloud pewter]

no, you get 150/d on normal and 50/d on high

[indigo hound]

I have another question: What is the difference between plus and the free version?

Especially regarding the number of times limit

[prime swallow]

https://help.openai.com/en/articles/6950777-what-is-chatgpt-plus

It says here?

[cloud pewter]

4o: ~10/5h to 80/3h
4: none to 40/3h (shares limits with 4o)
o1: none to 50/w
o3 mini: ~10/5h (also shares limits with 4o) to 150/d
o3 mini high: none to 50/d
4o mini: hidden limits to pretty much unlimited

Avm voice: 15m/mo to 1h/day
avm video: none to yes (limit subjected to be smaller than voice)

but it’s so vague… no actual number shown

[wintry sky]

ive talked to 4o alot in a short period on plus and never ran out. i only know 1 person who ran out once for an hour.

[cloud pewter]

me?

[prime swallow]

I have run out for hours at a time. Sometimes even 2. But! If you read fast and type fast, or if you're sending a lot of short fast messages (Maybe don't play hangman with the model 😛 That sure eats messages fast!) it is definitely possible for a user's style to eat 80 messages in less than 3 hours.

[cloud pewter]

thats basically just me

[jolly temple]

Hi guys, so o3 basically isnt coming?

[viral surge]

Model has been acting up recently.

[tulip folio]

Hold up, there’s limits for 4o on the plus plan??? I’ve never ran into that, I always assumed it was unlimited.

[prime swallow]

If you do nothing but use the model all day long - and don't use it more often then 1 new message (or message regen) per about 2.5 minutes - you'd never see the limit 🙂

[tulip folio]

Sick, thanks for the info!

I do wish they’d be more transparent about the limits though.

[elfin lava]

https://help.openai.com/en/articles/10492689-chatgpt-plus-promotions-referrals I just find out about this, how does this work?

[honest jay]

Hmm I haven't heard of this before I don't think, thanks for the link! The line in there, "Additional rules may apply based on your carrier’s agreement", makes me think it maybe has to do with certain telecom carriers offering promotions, maybe?

[elfin lava]

Woops

[prime swallow]

Just checking, but had you noticed this yet?

https://help.openai.com/en/articles/6950777-what-is-chatgpt-plus

In it:

[As of May 13th 2024, Plus users will be able to send 80 messages every 3 hours on GPT-4o. and 40 messages every 3 hours on GPT-4. The message cap for a user in a ChatGPT Team workspace is approximately twice that of ChatGPT Plus.

Please note that unused messages do not accumulate (i.e. if you wait 6 hours, you will not have 80 messages available to use for the next 3 hours on GPT-4).

GPT-4o mini does not have limits on the Plus tier.

Plus user have access to 50 messages a week with OpenAI o1-preview and 50 messages a day with OpenAI o1-mini to start with our research preview.

In certain cases for Plus users, we may dynamically adjust the message limit based on available capacity in order to prioritize making GPT-4 accessible to the widest number of people.]

The o1 info needs an update.

[tulip folio]

Yeahhhh, it’s just really outdated in many sections so I never know what to trust from their website’s older posts

[wintry sky]

No im upset lol. Gpt5 is amazing idea that I’ve been waiting for but why not release o3 while we wait many many months

[tulip folio]

But good info nonetheless!

[honest jay]

I imagine that URL ending in "CODE" just uses "CODE" as a placeholder -- actual promo URLs from carriers (presumably to said carrier's customer) likely have a unique code in place of "CODE"

[tardy isle]

Anyone file attachment to o1, o3 family working? It does not see the text file I attached - just like when GPTs first launched.

[normal quarry]

I just uploaded a PDF manual to o1 and it works fine

[tardy isle]

Okay, PDF works. Guess it's issue for text file then. I gave js files to models and asked analysis, but models are like "where IS the code?".

[cloud pewter]

ew only new subscribers. they really hate the loyal subscribers now

[wintry sky]

In your guys opinion is o1 or o3 mini high better at math?

[prime swallow]

Why do you think so? I don't feel hated.

[ember grove]

do projects still have a file limit?

[candid grotto]

Hey Open AI team, are you guys currently looking for QA or evaluation work?

I was previously Head of QA and Head of Customer Support at MultiOn an AI Agent startup with 30M+ in funding where I grew with the team from 10-25 and built my departments from scratch.

Now I'm looking for a cool new opportunity and would be open to working for free to get my foot in the door and prove myself.

DM if you'd like to check out my LinkedIn!

[fiery gale]

what is the difference between the GPT# versions and GPT o# versions?

thought we gonna get O3 but we are getting GPT5 soon

[dense halo]

o3 mini high?

[wintry sky]

Idk. O3 mini search is way worse than 4o search from what I’ve tested

[dense halo]

For all of my attempts it didnt access anything

[desert sparrow]

1

[halcyon turtle]

is it just me or do custom intructions tend to have a weaker influence on chatgpt's output than like creating a custom gpt / copy and pasting the prompt

[frosty fable]

Can we pin some discussions on chatGPT? For example, I have a discussion where I record my runs everyday and chatgpt guides me. I don't want to go everytime scrolling and searching

[native otter]

No, but you can move it to a project

[blazing tartan]

Does anyone have already Deep research on Pro account? They said on announcements: #announcements message few hours ago that:

Deep research is now available to all Pro users
But I still don't see it in my account

[native otter]

Wait, o series now supports file uploads?????

[normal quarry]

yes

[native otter]

Yea I just tested it, I have been waiting for that since they first released o1-preview lol
That means Custom GPTs with o series should now be possible!

Hope they bring it soon, as they already have the base done...

[fast veldt]

...anyone else notice the lack of orange warnings?

[native otter]

When writing sus stuff? Didn't test it, but Custom GPTs were always detected less

[cloud pewter]

isn’t it good, less censorship

[fast veldt]

Huh. Neat.

[cloud pewter]

good because when I let it write something that’ll definitely be flagged before didn’t get flagged today

[dense halo]

Hopefully the "grown up mode" is truly absolute

[half vine]

how many messages you have in the free plan for o3 mini?

[native otter]

I think it was 10 messages every 3 or 5 hours

[half vine]

oh ok

[cloud pewter]

I never ever want to see nonsensical flags again, especially on avm

[dense halo]

did they fix avm?

[cloud pewter]

nope

[dense halo]

🤣

[cloud pewter]

look at this

😔

[dense halo]

😂😂😂😂😂

Openai classic

[crisp cairn]

so o3 mini lets now upload files

so i am uploading one csv for analysis

it shows dataset opened but o3 mini cannot see it 🙂

"I don't see the file yet. Could you please try uploading the dataset again? Once I have the data, I can start analyzing it for you."

[dense halo]

Web search doesnt work with it too

[native otter]

Web search works, but it's an early implementation

[native otter]

It may not understand that file format, try an excel sheet, or plain .txt

[desert jewel]

The blur on the iOS app looks really weird and pixelated. Is this normal?

If I scroll it looks even worse

[native otter]

Doesn't look like that for me, also the blue thing is more related to iOS
The ChatGPT app is using native iOS APIs for the different components

[desert jewel]

That’s how it looks when scrolling

Really weird, maybe because I use iOS beta?

[native otter]

Yea that's it probably, it just using native stuff
For example, if it used a native button, and something about native buttons broke on iOS, it would break in all apps using it

Even if not native, sometimes the different draw routines may break with iOS versions

[real bluff]

made something that tells me how many uses i got

[cloud pewter]

dm me this I need it 🗣️ 🔥

[inland heron]

any way to have the chatgpt search actually find you the correct output you need?

[real bluff]

you need tempermonkey

[inland heron]

I'm talking about the specific reply, not just the conversation

is there a way? the output I need is hidden under different replies/prompts and it woudl take me an hour to find it manually..

I wish I could do that every time, but sometimes there's so many things to go through when debugging or building something that it quickly ends up with hundred replies

not to mention these new models now being fast and better to use, this quickly adds up

[forest sleet]

Wait o3-mini high has only 50 commands to use?

[honest jay]

If you know it's in the chat you have open, can you Ctrl+F/find in page to find it? Apologies if I'm misunderstanding

[inland heron]

the actual reply is hidden under these. Ctrl+F can help but only so far

Luckily I've found it by pure luck