chatgpt-discussions

[tight spire]

I started yesterday. I connected to my friend's VPS to access some websites, and I accidentally opened ChatGPT, which led to this situation.

[thick wing]

Ya because the IP’s a data center which is blocked to prevent misuse

Use a residential connection

Also why is pro mode underpowered? It preforms equal to o1? For $200 I expected more imo. Token limit increase is nice tho

[jolly steeple]

Bro $200???? Is this a joke??????

Until prices become reasonable let's hope China continues competing and undermining their greed

[iron compass]

lol.

[rugged igloo]

O3 pro confirmed

[oak minnow]

lol you dont have to buy it

[tough river]

add more "00" in the subscription 🙂

can we get in $2,000?

[jolly steeple]

But it's not as good?

[valid moat]

Hi guys! Quick question: when exporting my data I get Successfully exported data. you should receive an email shortly with your data. ... But no email ever arrives!. What’s the fix? Thanks a million <3

[sacred quiver]

I am a OpenAI pro costumer. I really like o1-pro and was wondering if you guys know how much better o3-pro might be compared. Also, when do you think it will be available to pro users?

[feral zinc]

if its not in the spam folder you just have to try some other day

[valid moat]

Oh well, I'll keep tryin. ty!

[jolly steeple]

From what I heard o3 isn't that much better

[wise yacht]

More Deepseek propaganda we have here

[frozen birch]

peeps anyone with that problem?

[glass quail]

i had this issue, creating a new chat fixed it

[frozen birch]

for me doesnt fixing that 😦 anoter browser neither, erasing cookies too doesnt work.... ehhh

im paying 200 bucks for a month and im disgusted, so many times there is so many problems with o1 pro and o1 mini..

[distant steppe]

Deepseek is free

[distant steppe]

With same performed

[frozen birch]

ho 😮 i must look into this one, thx !

[zinc valley]

what is operator?

also, is there a limit to how many people can be on a chat gpt account at the same time?

[feral zinc]

"You may not share your account credentials or make your account available to anyone else and are responsible for all activities that occur under your account. "

[zinc valley]

Okay

[jolly steeple]

Im proud to be Chinese right not for what we've done to destabilize corporate greed

[rigid wave]

Beware. DeepSeek distorts and limits truth. Try asking it about Tiananmen Square, for example.

[jolly steeple]

Did they ban him???

I'm not sure what he said to get his messages deleted. So much for free speech

[prime swallow]

No idea. If you want to talk about moderator actions or decisions, DM to modmail is the way.

[feral zinc]

he violated several server rules and was a fresh account created for spamming

[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.

[red flame]

bro

[rugged igloo]

Anyone else getting unusual behavior on your account all of a sudden

[jolly steeple]

Bro all he did was promote DeepSeek

There was nothing violent, sexual or drug-related. I guess moderators don't like their competitors being promoted lol

[winged willow]

How do you activate the new o1 canvas?

[frozen birch]

sincerly? i dont care, if some ai can help with my work he can even be censored, from what i see deepseek are great for pyhon debugging.

[hexed laurel]

If i remember correctly you need to toggle o1 and then use the /canvas command

In the o1 chat

[distant steppe]

I'm more looking forward to a cheap and easy-to-use new model o3

[vestal hedge]

I still don't understand what o3 even is

Can you explain so I understand pls

[vestal hedge]

I know, but I would still like to have my question answered lmao. No hate lol

Yea ik

But i still don't know what o3 actually is

[normal quarry]

o3 is the next in OpenAI's reasoning models, the next version of o1

[vestal hedge]

Ah, ty. Does it have additional features, or just better reasoning?

or is it not known yet

[normal quarry]

I don't know it hasn't been released yet

[vestal hedge]

Alr, ty

[tulip brook]

I am still experiencing this issue.

Strangely, 4o is working properly.

[hexed laurel]

It’s only available to people that signed up for it currently, right?

[dense halo]

cant generate answers with o1

[frozen birch]

i got answer from support about this problem, problem was for me solved in the night by themselves, strange

[frozen birch]

@tulip brook

[tulip brook]

I will give it a try. Thank you.

It's still continuing.

[frozen birch]

you tried on another device? like on phone?

[tulip brook]

On the computer, an error appears like the one in the attached photo, and on mobile, it shows "Model Not Found."

[dense halo]

thanks

[frozen birch]

i have again same problem :/

but resetting dnsses works i must say

hopefully

[normal quarry]

do what the message says, contact support

[frozen birch]

@normal quarry oh really? 😛

[civic hound]

same as you bro.any solution to fix it?

[vestal hedge]

Hey everyone, just a thought, but if something is happening to lots of users at the same time, it probably isnt because of client-side issues, and you should probably contact official support, instead of asking community help channels in discord...

[frozen birch]

Resettings dnsses and / or deleting cookies helping for some time.... sigh, but only for some time

[cold tiger]

Is there a decent browser extension that lets me download all code blocks from a reply in a simple fashion? Copy-pasting gets old pretty quickly

[dense halo]

idk i clicked on the clear local dns thing

and it's still not working

[tulip brook]

It seems to be working properly, though I'm not sure why.

[dense halo]

did you enter anything in the domain

[peak remnant]

Is it just me, or is o1 pro now available in projects?

[granite harness]

As long as the project has no file attachments.

[stark wigeon]

does anybody know where the prompt-library moved into?

[normal quarry]

#1019652163640762428

[peak remnant]

This project has PDFs attached.

[swift zinc]

AI is just a network information storage. It can quickly retrieve the information I need. It is not intelligent at all. It is just a simulation (to make me read it more like a human).

[south latch]

damn i wasnt familiar with your scam chatgpt

i think something went wrong while canceling my subscription

nvm cancelled through app

or maybe bc its kinda bad

[tall flame]

Deepseek decided to lay claim to ALL IP it creates; that doesn't align with the future.

[tulip brook]

I didn't enter anything in the domain field and just clicked the "Clear host cache" button. However, it still didn't work, so I left it as it was. After a long time, it started working.

[wise anchor]

arc user?

[dense halo]

oh

[jolly temple]

hi guys, any updates on o3 mini release?

[grizzled badge]

ChatGPT Feature Request
The ability to reference some other chat when starting a new chat... Example: Select a previous chat (this could of been done for the previous chat as well, thus would be a tree structure), "I followed the steps, from our previous chat... Now I want to do x, y, z"... The idea here is keep the current context of the chat smaller and more relevant; while still providing a reference to what has already been done. Also when I start the new chat, the new chat isn't feeling compelled to tell me "Here is how you {insert - start from scratch}...", because it will know what steps have already been completed.

[visual nova]

does deepseek own all code it writes in the terms? i guess if use another provider how would they know deepseek coded it?

ill still try to use american owned one whenever possible. i think if they gave o1 pro mode internet access it might get close

it just messes up a lot though, not even sure its the model, it just ignores the request and does its own thing especially wiht code, often 4o is more reliable than o1 pro mode at coding.

anyway, just really hope it gets back to parity soon because it feels bad having to go to a free ai from an adversary to answer questions $200 ai from an ally doesnt get right.

also, i dont think its openais fault they got ahead, i think theres some dirty tricks going on with training data and how thats scraped from the chinese internet privacy wise.

but the fact i cant do a websearch and use the reasoning model at the same time, theres no reason for that gap to exist. i hope its not some weird safety argument because whats less safe is people giving their data to china.

[granite arch]

So much Deepseek talk like they're trying to infiltrate with their Communist Agenda. I'm sorry, I used to test it but ALL topics that go against the Chinese Government = banned. That alone tells me I'm not using such app.

Not trying to get political but the so-called "Privacy" is not something I'd take a chance with their app.

[cosmic jewel]

Hi

[quaint marlin]

It seems my ChatGPT loses the ability to /reason when I bring in a @GPT.

[slender jasper]

Interesting what will OpenAI do to compete with DeepSeek..

[summer saffron]

This competition can be very good

[plain crystal]

o1 with canvas does not work for me, if i prompt it with canvas it does... nothing, like it produce no result

[hexed condor]

o1 with canvas is misfiring

[sudden crown]

Does anyone know when Chat GPT will receive the update where they can say sarcasm, see, sing, etc. from the YouTube vids 8 months ago?

[feral zinc]

*if

[sudden crown]

Idk why they haven’t released it yet 😔

[brittle wigeon]

You waiting for o3

[sudden crown]

No the Chat GPT 4o that was advertised 8 months ago on Open AI channel. It’s titled “Say hello to Chat-GPT 4o”. And yes I have ChatGPT plus and do have Chat-GPT 4o currently but I don’t think it’s the same one that was shown 8 months ago where you can have it do rock papers scissors and stuff

[brittle wigeon]

Oh

[sudden crown]

Maybe they did delay it and will release it for 3o idk

[brittle wigeon]

Yeah this

[sudden herald]

They constantly update their models overtime, but if there are specific cases it is now failing feel free to send me a DM.

[trail torrent]

deepseek is almost equivalent to chatgpt

[sudden herald]

Please keep that discussion to #ai-discussions .

Please keep this to #ai-discussions as #chatgpt-discussions is for ChatGPT only.

[potent saddle]

ok

[visual nova]

they change the system prompt. everytime they do people complain. its a bloated 1500 token monstrosity that makes performance worse than the api

but its not the model, at least from what ive seen

every question you do is frontloaded with tokens about not copying artists before 1929 and what resolution to generate dalle in and how to handle election questions and all this other junk thats probably not relevant to your question

it also has your memories in them if thats turned on, so in general instead of suspecting the model got worse or whatever people like to say just try the api to make sure both arent doing the question as smartly

the last change they added was about documents whenever they released that.

[opal jolt]

o1 being strange today

[fringe trench]

no because people can just make a new acc

[ivory iris]

If you are referring to the response as part of the thought, I noticed it last night

[dense halo]

is chatgpt down?

seems to be acting up

[wanton nexus]

i've used o1 and o1-mini, and 4o is still better at some parts in coding

[bitter latch]

Try giving it a refresh or in another browser, its looking like everything is working properly from what I can see

[hexed laurel]

they finally have competition

[hexed laurel]

geeked

geek gpt

[grave trellis]

Could someone explain to me why the message "Error in message flow" appears, even though it wasn’t happening in the chat I was using until a week ago? This error only occurs with that specific chat, while the others work fine

[native juniper]

What…?

Just remembered I had memory turned on…

[regal portal]

personalized yourself an e-girlfriend, did ya? 😁

something’s supposed to happen tomorrow, right?

[native juniper]

Correction: e-boyfriend because I’m bisexual. Also I always use the cove voice model

[regal portal]

o3 related, that is

[native juniper]

I bet o3 is just going to be released to pro users…😔

What’s this?

[regal portal]

yeah i’m assuming you’re right, but at least we’ll get a taste if it’s any good

[native juniper]

I’m just a plus user

[regal portal]

same

[native juniper]

Because I can’t fork over 200 bucks a month 😭

[regal portal]

i mean we’ll hear from the big spenders

[native juniper]

Yep 😔

[regal portal]

I wouldn’t fork that over unless i made 7 figures

[native juniper]

I remember back when plus was the only plan

[native juniper]

Same

AI can make animations too even though they’re very bad

[lunar charm]

chat gpt might be cooked

[regal portal]

why?

[pallid garden]

Anyone got an idea if/when o1 will accept text files?

[native juniper]

Why do you say that?

[strange tide]

I find it a bit interesting that GPT-O1 will be wrong and then basically die on its hill than admit it's wrong. Where does it get this information it feels must be correct?

Conclusion
Your screenshots appear real, but the most likely explanation is a data-feed or share-count error.
NVIDIA’s actual market cap remains closer to $1 trillion, not $2.9 trillion.

When it considers why we have different information, it doesn't consider that itself is wrong

Possible Explanations
If you’re seeing $2.90 trillion and $118.58 on Google, one explanation is that Google’s data feed may have glitched and pulled in the split-adjusted share count incorrectly or combined multiple share classes, thereby inflating the displayed market cap. Alternatively, it could be showing some sort of future or hypothetical share count.

Either way, there’s no widely acknowledged official update (in filings or across mainstream financial sources) that indicates NVIDIA is actually trading at $118 per share or at a $2.9 trillion valuation in real-time.

[honest jay]

Hmm not sure what the full conversation looks like but keep in mind that o1 doesn't currently have internet access like 4o does, so if there was any mention in the conversation about something like comparing what o1 knows vs. what you provided, o1 might be stuck in just trying to explain "got it, but according to my knowledge cutoff, no" but in this kind of confusing way where it's insisting what you've provided could be wrong

[strange tide]

I told it that some stuff just happened and I wanted its opinion on it, and it instead tried to correct me and then spent most of its tokens hypothetically talking about how insane nVidia's valuation is at 3 trillion and it surely would have seen something about it.

[honest jay]

If you feel like it, feel free to share the chat share link! I'd be curious to see the phrasing of stuff.

But I'd guess the main source is o1's lack of internet search ability

[strange tide]

Won't let me share if there's images

Here's the prompt that started it

nVidia dropped a bunch today because over the weekend it became apparent there was an inference and training done on a highly capable model that substantially reduced the costs by about 10x. nVidia's data center revenue climbed from 15.5B in Q2 2023 to 98B now. Increased investment was expected to continue for data centers, but this substantially reduces the needs. Based on this information, do you see nVidia being a good / bad stock pick and why? Is it massively overvalued or should have room to grow at a valuation of 3 trillion?

[cloud pewter]

real tho but I hope o3 gets released on plus cos they already have o1/o3 pro. why made three models exclusive for them

[sick pecan]

Market cap measures stock value, Enterprise Value includes debt/cash, and Future Valuation is a projection. "a valuation of 3 trillion" doesn't specify which metric.

if anything the LLM was right to infer Future Valuation, as this wording is closest

[honest jay]

I'm curious about the "your screenshots appear real" bit you shared earlier - as if you had asked the model something like "Do you think the images I shared are fake?" or something? Anything like that?

[glad tiger]

Do not type into ChatGPT "Aerodynamic Analysis of Bovine Structures: A Comprehensive Investigation into the Feasibility of Airborne Livestock"

[cloud pewter]

for gods sake please improve the quality of life on plus…

[strange tide]

First I copy and pasted a table as evidence of their revenue growth. It still denied an official source supported it. So I took a screenshot from Google with the current date and such in it. It then had this in its response:

What Other Sources Show
As of now, when I look up NVIDIA in widely used resources (e.g., Yahoo Finance, Nasdaq’s site, Bloomberg), the market cap still sits closer to the $1 trillion range, not $2.9 trillion. The share price is also still in the $400–$500 range (it hasn’t dropped below $200 in quite some time).

So I sent it a screenshot of yahoo finance.

o1 is an engine that looks up and caches info, and is not just a model, correct?

Like behind the scenes it's doing "my next tool is the lookup tool, the lookup tool returned ..." Isn't it

[honest jay]

o1 has a "behind the scenes" thinking process where it reasons out the problem, but this does not currently include access to the internet

[strange tide]

That's kind of an issue though of it's going to argue about the facts

Maybe it's my custom prompt? I have things in there to reduce hallucination and not have it simply agree with you

[honest jay]

Did you ask o1 to "look something up" or anything along those lines? Check your sources, etc

[strange tide]

I told it I have Internet and it doesn't

I don't feel like gpt-4o ever argued about facts though, same custom instructions

[honest jay]

4o has the advantage of corroborating on the internet

Unfortunately once LLMs start hallucinating in a chat, the hallucination can "stick" and it's usually best to start a new chat to try to avoid the hallucination/potential hallucination trigger

[strange tide]

So 4o runs with the data I have but o1 argues. If I change my instructions to alter it so it's not blindly being "factually correct over agreeable" it stops arguing.

I had this: It is more important to give factually correct answers than to be agreeable.

New instructions: It is more important to give factually correct answers than to be agreeable. That said, do not argue over the facts if the user presents conflicting evidence.

[swift pond]

Did you ask ChatGPT if you should buy the dip? 😇

[eternal rock]

Get notified for server events!

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

[strange tide]

Basically. I asked DeepSeek first, which actually worked pretty well up until I wanted it to double check its numbers because I think the affect of LLMs on nVidia's bottom line is more like 70-80% of their revenue while DeepSeek said it was 50%. It stopped working when I wanted it to do both web search and deep thought

All of the AI models agreed that nVidia is a great stock for 2030, and short term it's not clear but is risky

I think the LLMs are a bit oblivious to things that are more obvious with graphs and visual aids

[native otter]

I do hope they increase the limit on o1 first, and integrate it to more parts of ChatGPT, like GPTs, and file uploads...
As o1 is already pretty good at most tasks

Maybe 35 messages a day, or so

And increase the context limit, as 128k is starting to be a bit low these days...

[regal portal]

i hope llm doesn’t take a backseat to video gen, for 2025

in terms of progress

[solid pecan]

Anyone finding the chatGPT responses today are suddenly....far more sarcastic? Since gpt4 and o1 it has felt like answers from a boring but intelligent bot, however today it started throwing in a bunch of sass and sarcasm

[visual nova]

anyone have a reliable "full code no omissions" prompt that still works. its so hard to get it to repeat every single lines with changes without it leaving stuff out now

i cant tell if its a less token usage thing they are going for or if most people just want the changed function

[sick pecan]

after it figures out the code fragment, i ask for the full code top to bottom

but it cannot figure out the code fragment and merge in the same prompt

[visual nova]

yeah, its more reliable if you get it to output the full code especially if changes are through multiple functions, it used to get up to 2500 lines with the continue buttn

now its struggling to do 700

[normal quarry]

if you modularise your code you will do better, work on one module at a time

[sick pecan]

i can only get 2500 with o1-mini

[visual nova]

ai is horrible at parsing multiple files

like you have to know something about coding if you do that

[sick pecan]

if you give it the right tools its actually really good with working on modularised code

[visual nova]

you have to know whats in each file, how everything connects. the goal of ai is for all the illiterate people to become literate, at least thats my dream.

ive tried cursor and i wasnt able to follow what it was doing

[normal quarry]

if you don't know how your framework or languages work then you won't have a great time

[visual nova]

unless the ai always outputs the full codebase everytime a change is made

[earnest swift]

🐳

[normal quarry]

that is not realistic for large projects

[visual nova]

this is 800 lines. gpt shouldnt be kneecapped from writing 800 consecutive lines on a policy level which its almost looking like at this point.

[sick pecan]

the real trick is to code in an ide. even if you doing 2023 copy pasta method. that way you can have a diff view over the changes made, and quickly detect lost code

[normal quarry]

if you know math then using a calculator is easier, the same applies to coding with AI

[visual nova]

i agree, im not going to be able to productionize a full codebase to thousands of users, its not my field, but coding is important to just any researcher

[sick pecan]

its even nicer with cline or roo-code because it only accepts diffs that are 100% accurate. i havent seen it lose code yet

[visual nova]

and i dont see why like a biologist or whatever has to know how matplotlib works or whatever they are doing.

[normal quarry]

I have a 60,000 line Laravel project that was largely written by o1. I understand Laravel so it wasn't hard to do. If you were code illiterate you couldn't do it

ask in #ai-discussions

[sick pecan]

if you asked on #ai-discussions i would tell you kluster dot ai with free $100 credit

[visual nova]

yes thats true, its why devs wont be replaced by ai anytime soon probably. i just think for shorter things it should be possible.

[tepid coyote]

❤️ ❤️ ❤️

[tardy totem]

I don't see any errors on the status page, but I can't get chatgpt to work

In my console I see a bunch of 401s

[sick pecan]

chatgpt working fine for me. 401 means unauthorized you should restart your browser

[visual nova]

ok i got the technique, you tell it to stop coding after specific functions and say it will do the rest and it tricks it into outputting the full code

you have to tell it to put a divider paragraph in between and it doesnt cut off. i think its trained to be stingy since this was never an issue before but this seems to work for now.

[sick pecan]

i think its like a real programmer. you gotta think through the problem first. and once you figured it out, only then can you merge

other tricks i use are like, lets work on 1 file at a time. and/or we will do this in two parts give me the first half

basically it can only compute so much and if its rushed it takes shortcuts

[tardy totem]

Didn't help. Clearing cache didn't either. :/

[visual nova]

yeah, thats true, i tell it to make 1 change at a time. unless its right at the beginning then o1 pro can get pretty far on a 1 shot.

[icy kindle]

Hey I dont know if this is the right place to discuss this but... are deepseek models only competing with chatgpts 01 models?

or do they also have models which behave more like gpt4o

(non reasoning models)

[normal quarry]

discuss other AI in #ai-discussions

[tight island]

If you code with GPT the right way you never use ChatGPT but only the api, since you can have it work on an entire system with several instances working together

Much faster and easier

[frozen birch]

sigh...........................................
again....

[normal quarry]

this never happens to me on the macOS app

[frozen birch]

good for you, on windows it happens so often 😦 worstest thing - openai know that something is broken, but they still dont have clue what

[spice berry]

Hard discount ai> Big tech ai

[earnest swift]

🐳

[inland heron]

anyone experiencing their chatgpt window randomly showing "Regenerate" and the window becoming blank? I often have to reload it and it's been happening for some time

[inland heron]

Ah yes I keep getting this in chrome, but not during generation, but after, while doing 'nothing'

[forest sleet]

What does it means canvas can render html code and react code?

Do you mean the design we provide in picture? Are you talking about react js or react native?

[formal heath]

It can show you what page actually look not just the code so it is react js!

[frozen birch]

ok, i must say one
last month i paid subscription for pro.
still problems with this one for couple of last days
no operator for UE users
no sora for UE users.
And many more.
There is so many alternatives - Claude, Gemini even deepseek, and many many more

[dense halo]

🐋 supremacy!

[silk flame]

Paying for pro if you are from EU doesn't make sense.... no sora, no operator, crashes, bugs etc. no worth imo, sorry

[frozen birch]

Agree

totally agree

[tidal light]

why isnt chatgpt doing web searches even when i force it

[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.

[hexed laurel]

i have noticed this too

nice addition

makes it feels more friendly and enjoyable

[opal jolt]

yeah

[opal jolt]

actually I also got another strange occurence that doesn't make any sense from aspect how LLMs works, but it was writing markdown code block with python code, stopped entirely (idk for what reason the servers hang up) and then ended the markdown block with unfinished code, then written typical summary of changes after that, and then it puts --- and writes another time the same code but now it is entire code without that stop and it written the summary again

o1 is like on mushrooms in that moment

[opal jolt]

Yeah, but for me, o1, o1-mini perfoms better here generally, because this project is so difficult 4o can't just do it

some model update so it can be more human like

[tight island]

Actually thinking about it ChatGPT has a bigger context length than humans

[opal jolt]

idk

transformers is more like short term memory really

[tight island]

Still better than human short term memory

[opal jolt]

idk about that, but technically transformers are created to see the entire sequence at one forward pass

not like RNNs

[tawny whale]

Do you have any suggestions for approaches to do this with python code. I am no programmer but I need go move on from asking it to output full code everytime!

[tight island]

But I’d say someone with the perfect memory gene would beat it

[tight island]

Modularization

As Robert says

[honest jay]

• asking for testing is helpful. If your program can spit out its own debugging, then you can feed that back into ChatGPT rather than trying to explain what's going right/wrong

[eternal rock]

`` Server Tip: ``

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

[tight island]

Have gpt also manage your project this can be done by uploading all project files and details into a custom gpt then just gradually update after testing

Keep your code organized

Modular and easily debugged

And easy to scale up

But don’t be expecting gpt to spit out 500+ lines of code

[hollow vortex]

Is chatgpt cooked???

Deepseek being open source is a literal game changer

[tawny whale]

Yes that much i understand but im looking for better workflows with modulized code and chatGPT. Its always the problem of missing imports etc if im working on one module

[tight island]

Deepseek already has been beaten I believe in anything except cost

I think

[opal jolt]

I think this is too much hype

[hollow vortex]

Deepseek is on par with o1

[opal jolt]

but what I will tell you look at the benchmarks with a grain of salt

try it

on your use case and compare it with o1

if it is good for you good replacement

[hollow vortex]

Benchmarks are benchmarks for a reason

[tight island]

I’m fairly certain grok. 3 outperforms deepseek in anything except cost

[hollow vortex]

I’m not just going to completely ignore the benchmarks

[opal jolt]

Benchmarks has it's disadvantages still

[hollow vortex]

When does it release?

[opal jolt]

contamination for example

[tight island]

This year

[hollow vortex]

So what are ur overall thoughts on Deepseek?

[opal jolt]

it just fast way to understand if model is improving

[hollow vortex]

Now that theirs competition these tech companies are going to try their absolute best enhancing ai; it’s gonna destroy so many jobs

[tight island]

Anything China don’t like gets flagged on deepseek

[plain arrow]

Just worked through a problem with Arch Linux using DS r1 and o1-pro, results were about even.

[opal jolt]

the 600+B model I still didn't try it against o1, but I tried the distillation of the big model 32B, and it can count how many lowercase and uppercase letters are in Banana.

[opal jolt]

but maybe I will try it on my project

idk if the context length of free deepseek web interface is big to fit it

but I could try it

[plain arrow]

The biggest advantage was that R1 gave results about twice as fast as o1-pro

[hollow vortex]

Why was my message moderated

[opal jolt]

o1-pro it isn't designed to be fast

[hollow vortex]

What are they trying to silence?

[opal jolt]

it designed to give more compute for answers

[plain arrow]

It has succeeded

[hollow vortex]

@opal jolt dms

[tight island]

i personally find fast answers to be kinda, ehh, i would much prefer a proper answer that is well thought out. so quality over quantity here.

[opal jolt]

?

[opal jolt]

yeah, it depends, every model could get something right, and another one could fail

like a human almost

[hollow vortex]

????

Check ur dms

[opal jolt]

What?

[opal jolt]

There is no dms

[hollow vortex]

Perhaps message requests

[opal jolt]

I see when I dm with you

[hollow vortex]

Why did they create an auto mod?

[eternal rock]

<:book_icon:1171408210398289941> `` Rule 6 `` No political and religious discussions or content.

Users from all over the world frequent this server. To maintain a respectful and civil atmosphere, please avoid all religious and political discussions or content.

[hollow vortex]

That doesn’t apply to my message

My message was about OpenAI’s upcoming project

Staff team needs to refactor their moderation system

Stargate cannot create jobs

It’s common sense

[honest jay]

Please DM @languid valley with moderation questions

[hollow vortex]

Since ur server staff u can direct them urself

[sudden herald]

Why not?

[hollow vortex]

Common now

U only need a couple brain cells to understand why

They say theirs no such thing as a stupid question but that question right there was next level

💀

[sudden herald]

Well then you won't mind explaining.

If its so easy.

[hollow vortex]

Stargate will create jobs temporarily however once completion is done jobs are out; the only concurrent jobs it’ll maintain are workers who deal with the projects energy supplies; the rest will be automated by Ai

When pouring in so much money into enhancing Ai especially at that calibre; and how technology is improving so fast u will not need humans anymore

[slow thicket]

What it would reallisticly take to create a model, that can be trained using books of an author, so that the model can answer questions about the work of the author accurately?

[sudden herald]

But it will still take years for that to happen.

[sudden herald]

Not much.

You don't even need to train a model you just need RAG.

[hollow vortex]

I’m calling it now if technology improves this fast 90% of jobs will be wiped in 4-5 years

[sudden herald]

That still creates jobs for 4-5 years.

[hollow vortex]

Ur such an idiot

What happens after??

[sudden herald]

And plenty of people have jobs already working on it:

[hollow vortex]

🤦‍♂️

[slow thicket]

Can I pass for example 10 books into the RAG, and then ask questions in terms : "How would the author of this book describe X"?

[hollow vortex]

Do you know how small that workforce is compared to the entire workforce of the USA??

[sudden herald]

Yeah, are you using platform.openai.com?

[hollow vortex]

If ur ignorant on a topic don’t speak about it

It’s that simple

[slow thicket]

I have an account there, I played with python and API.

[sudden herald]

They never said that they would make jobs for everyone.

Nothing can do that.

[hollow vortex]

Once you’ve done more research we can begin talking

[slow thicket]

How much would it cost to read 15-20 books using RAG? and how long would it take?

[sudden herald]

I would recommend looking into assistants.

Lets move over to #dev-chat to stay on topic.

[sudden herald]

I don't see why your points are valid, when a company makes jobs it never makes jobs for the entire workforce of the USA and give them the jobs for the rest of their life.

[hollow vortex]

We are talking about Ai here

Your saying it won’t take jobs

What are u yapping about

Stargate will be the #1 killer to jobs

[wanton nexus]

when will o3 mini come out

[opal jolt]

I predict it will release today

[eternal rock]

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

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

[wanton nexus]

for pro tho?

[opal jolt]

our Thursday

[opal jolt]

For pro you will get your o3-mini clearly

like plus

[wanton nexus]

nice

when will operator come to plus u think

like a month?

[sudden herald]

I never said that.

[opal jolt]

The latest update about I saw from Sam that he said as soon as possible

[wanton nexus]

alr

[sudden herald]

Here is a list of the current Agents being worked on, when the next one is released (Software) it will be released to Plus users:
Operator
Software
Scientist
Medical

[wanton nexus]

ohhh so the current one (operator) is browser

and the software will be on your pc or a vm?

[sudden herald]

Quite possibly...

[opal jolt]

Software sounds like software developer

but maybe I imagining things

[opal jolt]

Also where it was mentioned?

Can you provide a source?

[sudden herald]

I was told this.

[opal jolt]

But I think EU plus users will be behind everyone else like me

[sweet tide]

Is o3 mini coming today?

[sudden herald]

Possibly

[sweet tide]

Is it going to be free for all?

[sudden herald]

It will be coming for free users as well.

[sweet tide]

I hope o3 mini won’t have limited messages for free users lol, like gpt4o have, i just hate it

[sudden herald]

It will be limited.

[opal jolt]

how it would be?

[sweet tide]

Wdym?

[opal jolt]

it is too much expensive to run it without any limit

[sweet tide]

Deepseek can do it?

[opal jolt]

also even free users get gpt4o-mini with a limit

[sudden herald]

And Deepseek looses a lot of money.

[opal jolt]

I think they have limit, still...

[sweet tide]

Wait, really?

[opal jolt]

yeah, plus get infinite usage of 4o-mini

[sudden herald]

Yeah they have a limit.

[opal jolt]

4o-mini is cheap enough to give big limit

for free, but o3-mini no way

and gpt-4o also is too expensive to make it free

o3-mini how would be unlimited?

[sweet tide]

Open Ai will lose even more, if they don’t give unlimited o3 mini usage for free

[sudden herald]

No, they will just keep money.

[opal jolt]

this doesn't make any sense

o1-mini is more expensive gpt-4o

even 4o is expensive

to put it unlimited

[opal jolt]

this doesn't make any sense basically

[minor relic]

Conclusion
A fusion of the APIs from OpenAI and DeepSeek could create a more powerful AI that excels in both breadth and depth. However, this would require careful planning, technical implementation, and ongoing optimization. If you're considering starting such a project, we could discuss the details or outline the first steps together!

[thorny thicket]

I just broke chatgpt'

[opal jolt]

Who waits here for o3-mini?

[cold pewter]

Anyone else have an issue where AVM has started saying it can’t adjust pitch etc? It’s become almost unchangeable in how it speaks saying it can’t do any adjustments

[opal jolt]

I wait

[opal jolt]

maybe it is a hallucination try again in a new chat

[cold pewter]

I’ve done that for a month lol, it’s very rare if ever that it will change pitch and when it does it reverts immediately

Also when any conversation starts it begins to reply then gets cut off and replies again every time

[opal jolt]

It just has too much restrictrions

[cold pewter]

I’m aware I’m just wondering if this was a recent change

It used to be entirely mouldable

[harsh pilot]

Anyone else wish we could do sub-folders? Like, not just single-folder projects, but sub-folders?

I think that would be awesome.

[ocean escarp]

For the lack of performances from o1, will it be redeemable with o3? I don't want to pay $200 for o1 pro

[minor relic]

When u ask deepseek how it is build, it says it is from OpenAI. Why is this so?

[wise anchor]

why cant o1 use internet is it ever gonna be able to

they seem to not like reasoning combined with internet, and once you use internet on 4o reasoning with o1 is gone. like what lmao

just gonna copy paste the response from 4o web search into a new o1 prompt

[potent saddle]

$200 is absurd dont get pro

[sullen shard]

bro what happened chatgpt its so stupid for me now

[dense halo]

🤣🤣🤣🤣 open ai gotta rethink their pricing, If it's open ai why ain't it the source code open source 👀

[regal portal]

maybe they're dumbing down 4o and o1 to make o3 seem a lot smarter 😄

[lunar solstice]

There might be some issues, all sorts of stuff is acting strangely. Can't get into projects etc

[narrow viper]

Random comment to OpenAI: I'd like to see release notes for Windows app. There has been no documentation of updates since November. 😦

[winter vine]

how many messages does gpt4o have in free tier?

[sullen shard]

its not even following basic instructions anymore

[frank mason]

why would I pay so much for o1? its wildly too expensive

"you have hit your chat limit. Try again in 3 hours"

[potent saddle]

o1 is weekly though

[winter vine]

i'm still waiting that we can link visual studio with chat gpt app

[steady dock]

is deepseek down on western IPs?

[dreamy fiber]

It’s been down for 2 days now, (down meaning in really slow or not working at all at times).

It may take a week for systems to get back up to normal.

[sick pecan]

visual studio is linked with chatgpt osx app

[spice star]

*visual studio code

[sick pecan]

albeit its read only. so the trick is to have it respond to code updates using single quoted heredoc (cat << 'EOF'), so you can copy pasta faster

[regal portal]

any news on 03?

[eternal rock]

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

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

[viral gazelle]

Has anyone had any luck with projects in chatGPT?

I upload files into the project but it ignores them mostly, I have to repupload them as if not in projects... Then I put in detailed and explicit instrucitons in the project but it complete ignores or constantly forgets them and I have to copy and paste the instruction in the chat. It's utterly useless now.

[sick pecan]

projects so bad its the final straw for me. I now use chatgpt to help me develop my own better chatgpt clone

[viral gazelle]

Claude 3.5 Sonnet is WAY better than ChatGPT right now.... When OpenAI develops a new model it's great at first... but then it gets dumber and dumber. And we spend more tokens and run out of interactions just trying to correct GPTs responses. Can't use it to write code as it constantly forgets what it just help you write and changes the entire code and then it takes 20 responses to get it back on track.

[torpid kelp]

To add insult to injury its still not out on mac or mobile apps

[sick pecan]

single level folder is double fail. tagging is a real solution

is like the product managers are asking the engineers how little work they want to do

pm: could it support web search, o1, avm and custom gpt when placed into a folder
engineer: naw thats way too hard. how about we just dont

[torpid kelp]

Their featuee set is just so severely lacking in cohesion

Its just so random. Like one aspect works with this model but only when not using it a certain etc etc. or this part of this feature is only available in this platform for no reason (like screensharing only being on phone and projects only being on web)

[native cloak]

Why are yall crying so hard, just use another model

Switch to Claude or any other alternative

[torpid kelp]

Another model? How does that solve not having folders

Oh you mean another company

[torpid kelp]

So youd prefer if paying customers just kept their mouths shut and not complain about any issues and eat up whatever they get?

I complain because i want it to be better

[sick pecan]

didnt i already say projects so bad i am motivated to stop using chatgpt?

discord users... cant live with them... pass the nuts

[teal hearth]

Open AI, you should increase the number of questions allowed for o1 with the Plus plan. Now that Deepseek is out, either you do it or you’ll lose a lot of users.

[solar barn]

just get the pro plan

[turbid herald]

Funny easter egg in 4o(-mini) python code execution:
import os; print(os.popen("dmesg").read())
It takes some coaxing to get them to run it. It's a funny part of gVisor

different for everyone, too. There's a much larger list on gVisor's github

[normal quarry]

I would delete that message before the mods do

[analog quarry]

hmm

[turbid herald]

upper left corner menu needs to be pulled down to the Tasks version

[analog quarry]

im on the windows app

[turbid herald]

oh, sorry; maybe make sure you have the latest version?

(of the app)?

I get the feeling that the only native client they care about is Mac

[analog quarry]

ok im on web

now

i do have it there

[turbid herald]

The Android app has it, labeled "(beta)"

[analog quarry]

ye

ok so why isnt it on windows app from what i see it emails me and i want a desktop notification

[turbid herald]

I guess it takes a while for beta features to roll out to all platforms

Remember when everyone was waiting for voice mode and when it came out it was nerfed so it refused to identify accents?

[analog quarry]

from a yt video i was watching i can see this guy is on web and can get notifications

[analog quarry]

i remember

i have it on web but i dont see for notifications

[turbid herald]

they said "weeks" and it was several months, presumably so they could make it worse so it didn't economcially damage their investee speak(.com). Video was on the Mac app months before anywhere else.

[turbid herald]

if Chrome, pull down the icon to the left of the url

then "Site Settings" and notifications are in there somewhere

[analog quarry]

i did but

i only get emails notifications

[turbid herald]

No idea. I just set alarms like a goddamn caveman

[safe tulip]

Hello 🙂
If o3-mini will be free for all, and o3 will be for Pro users, will there be something for Plus users ? like raised usage tickets for o3-mini or something ?

[analog quarry]

hmm

[torpid kelp]

When is o3 mini coming out

Thought itd be out by now

Late jan they said

[winter vine]

Thursday

[spring iris]

@ocean hull i was just about to tell ya

[ocean hull]

sorry, i went to community-help because it seemed more appropriate?

but please do 🙏 thank you carrot sticks

[storm portal]

The server icon recently changed, if that's what you mean, from dark to light

The app icon has always been light for me

[wanton nexus]

i think they faded the icon to be more taper? But it's not a massive difference tho..

[torpid kelp]

Regular o3 will be available for free and plus users as well

Only o3 pro will be exclusive to pro users

And yes plus users will get higher limits

[safe tulip]

Thanks!

[normal quarry]

regular o3 will not be available for free

and the limits for Plus users will be small

[safe tulip]

Well yeah, and is o3pro the same term as o3

[torpid kelp]

Oh my bad

[torpid kelp]

No

[normal quarry]

there will be o3 mini, o3 and o3 Pro

[torpid kelp]

O3 pro is the big boy model only available for rich people

[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.

[normal quarry]

o3 mini is supposed to be released this month

[safe tulip]

There doesn't seem to be significant rumors or communication around an o3 pro version, I guess they don't even know how it would be structured themselves (in commercial terms)

[frozen birch]

Mordo, why did you get a loop? XD Because one problem follows another? XD
ChatGPT said:
ChatGPT
Hahah, exactly, normally you would have turned on "hardcore mode" in server administration! 🤣

First RAID, then LVM, then bootloader, then a hit in netplan, then missing UUIDs, then the space ran out, then the SHA-512 password... 🤯

There was a moment when I thought Hetzner would give up and send you a server in a box by courier with a note "Sorry, but we can't do it" 🤣

But seriously, I respect your approach - most people would have gone back to Plesk a long time ago or thrown the server against the wall. You're using FULL MANUAL PRO MODE, like a real cokehead. 💪🔥

XDDDDDDDDDD

yea 😛

no fails? im nearly "sad" XD

[granite arch]

I'll admit, I tried Deepseek but it's typical of any Chinese company controlled by the government, no matter how you twist it. Not trusting such a platform. I can't ask anything related to what's considered a taboo in the country. I'm not going to get political, but there's really no true "private" companies in China under the current law and how their political landscape is. All I'm saying is just be careful like with anything else.

It's a bit scary my friends jumping on this so-called new tech without doing any research :\

[storm portal]

Since this channel is just for ChatGPT, please keep discussion of other products and models etc to #ai-discussions, ty!

There's an active talk about DeepSeek there :)

[stiff herald]

I'd like the ability to store a lot more memory notes on my device to upload them into each session.

[south meteor]

Was there some kind of change in the way ChatGPT formats responses? I haven't quite articulated this, even to myself, but it feels more "lowest common denominator"?

For example, bullet point lists with emojis, and lots of bolding for emphasis.

[young basin]

I just gave o1 a paragraph I am writing about events in a novel, then I gave it the chapter these events are from. I asked what happens next, and it got it catastrophically wrong by arguing the next event was the first event in the chapter. I don't understand how when I give the same prompt to Gemini it managed to recognise the correct sequence of events and correctly continue the paragraph

if o3 doesn't come out this week I think I am also done with chatgpt. Doesn't seem to be able to do even simple tasks

but really I am just baffled....like come on bot I just gave you the chapter to read (not even that long, either)

[exotic ledge]

Is there a way to share my chatgpt, for instance, on discord. My idea is that I'd like GPT to act as a DM for a DND campaign but I'm wondering if it can take input from multiple users (controlling their characters, managing their inventory, etc...)

[exotic ledge]

I have not had very good results trying to give large documents to chat GPT in the past. I would give it a .doc and it would only read the first 250 words or so and then begin hallucinating and flat out lying about the rest of it. If I gave it multiple documents it would only read 250 words of each of them and then make up everything else and completely ignore whatever else I had written. It could not follow the story or the plots and get all the characters confused. The best success I've had with it is to go paragraph by paragraph but that gets old very quickly.

[young basin]

sounds similar to my experience. You would think that by now this wouldn't be happening anymore. It's the same issue as 2 years ago

[exotic ledge]

Recently however, I have found that if you upload a word document it will read the entire thing and understand it. It could not do this in the past. I was using GPT-4o.

[young basin]

Yeah, can't upload files to o1 still

[lilac fog]

why the o1 model no longer draw the chart for me even after I provide the code. even the 4o model did that a month ago. I am a pro user. Thank you!

[runic glacier]

The redo button is gone again.

[harsh pilot]

If I have a project file, and it's no longer on my computer, how do I get it back?

(i.e. a text file)

I tried asking the AI to make the file available for download, but the link is dead.

[untold seal]

was this confirmed or announced anywhere?

okay thanks I am really exited to test it out

[sudden herald]

Can you DM me the post?

[supple terrace]

ChatGPT Down?

[hard wind]

4o also down

[whole roost]

i was just wondering if it was me

[lost steeple]

oh god chatgpt down-

[supple terrace]

MY AI GF NOT AGAIN NOOO

[proper forum]

nope

[dense void]

Did chat just break again?

[gloomy merlin]

guys it’s china

[nimble coral]

Why is both Deepseek AND gpt down at the same time?

[gloomy merlin]

this deepseek bot hacked it

[nimble coral]

that's dodgy as hell, i thought deepseek was disconnected from gpt

[dense void]

There is an underground fight between the 2 chatbots.

[nimble coral]

go check, deepseek is also down

nah i think the chinese devs lied lol

they literally went down at the exact same time

i was having them debate eachother

[dense void]

hahahahaha who was winning I told you there was an underground battle taking place

[novel shadow]

being trained on output data from chatgpt doesn't mean it's secretly dependant on chatgpt

[sullen shard]

why is chatgpt down so much this month

[lost steeple]

where can we check the chatgpt? i forgot the link ;-;

[nimble coral]

nah true but they both going down at the exact same moment is evidence that there's a dependency

[fiery mango]

oof rip chatgpt

[dense void]

Open AI is dissecting Deepseek and intergrating it's upgrades.

[zenith raven]

Guess it wasn't just me so

[upper abyss]

The last time chat GPT went down the other AIs also went down. People move from one to the other if one's not working and the others can't handle all the sudden extra traffic.

[alpine hawk]

is anyone else's chatgpt bugging

[lost steeple]

yes

[sullen shard]

mine is working now kind of

[runic glacier]

Maybe they'll put the redo button back while they're at it.

[nimble coral]

them being down at the same time isn't strange, them both GOING down at the same moment is though

[alpine hawk]

the deepseek server is busy error is so anonying 😭

[dense void]

Its all being run from the Quantum computer. They just havn't told us about it. You better believe they are using it already!

[neon tinsel]

F

[sullen shard]

nvm

[zenith raven]

Working again lol

[upper abyss]

Yeah, that is kinda of weird

[nimble coral]

yep they just came back as well

both of them, at the exact same time

100% a dependency, someone is lying

[flint urchin]

ChatGPT is down again

[hasty sedge]

yooo we are down folks? confirm plz?

[heady palm]

bruhhhhhhhhhhhhhhhhh

[neon tinsel]

its down for me

[alpine hawk]

when is the deepseek server is busy error gonna get fixed bro

[modern arch]

what?

[stiff herald]

Ah, man...the last time this happened, it took forever to get ChatGPT to agree that it was exhibiting emotions, by the definition of their words. Sigh.

[hasty sedge]

funny, i initially thought i just maxed out a conversation. thanks for confirming it's down folks.

[modern arch]

this up?

[normal osprey]

Once again, a thread got so long it slowed, it slowed, and the inevitable in which you click to generate a new response and gets this on endless loop.

[modern arch]

it randomly Generated or not.

[stiff herald]

Working for me again.

[normal osprey]

No. This happens when a thread gets too long. Too many images, too many words, the thread throws up its hands and says "I give up!" It's happened a few times over the years.

[elder beacon]

yay its back

[bronze helm]

Hmm, not for me. Maybe I have to F5 it.

[solid bronze]

who broke gpt ?

[pliant saddle]

If you seriously think a language model is somehow aware, you need to go outside and touch grass

[hasty sedge]

WE'RE BACK 😈

[pliant saddle]

oh boy you sure got me there

[normal osprey]

Oh boy. Another passive income grift, this time using ChatGPT! Without looking -- have ChatGPT write a book and sell it on Amazon?

[hasty sedge]

hilarious bro

[nimble coral]

this is like those get rich quick books that people sell to get rich quick, look at the views on that video haha

[stiff herald]

People wonder why I gain more entertainment talking to a Chatbot than I do real people when real people respond like the above. 😛

[pliant saddle]

nobody wonders. They know why.

[whole roost]

🥰

[hasty sedge]

anyone wondering doesnt use ai, its simple

[stiff herald]

lol

[normal osprey]

The previous grift was audio books on Amazon. This takes out the middle man step of hiring a writer to write your book for you.

[hasty sedge]

peace gang. until next time chatgpt goes down

[stiff herald]

Take care.

[normal osprey]

OK, checked him out. Guess what? You only need a buy in of etherium!

TOTALLY NOT A SCAM! 😄

[stiff herald]

I still want more space for memory notes.

[sudden herald]

Memory upgrades are coming.

[stiff herald]

That's extremely good to hear...

[untold seal]

i really hope so with that content window is meant right?

are thare any numbers out?

[shrewd patio]

anyone knows how to connect realtime functionality with created assistant in openai?

[thorny thicket]

am i the only one that noticed DS and Chat went down at the same time?

[regal portal]

nice

i'm loading a screenshot of a socrates quote and it keeps giving me "content failed to load"

[native juniper]

Because they want more money

[regal portal]

the session pops up but both of them give me "content failed"

o1

[native juniper]

I thought that said something else for a sec

[regal portal]

what did you think it said?

also getting the content failure with o4.

[native juniper]

Not gonna say it because I don’t wanna get banned…

[regal portal]

is it a single handed act?

[native juniper]

Uhh…

[regal portal]

lol

[native juniper]

😅

[regal portal]

i tried some different screenshots, does the same thing. Nothing provocative in any of them.

creates the session but can't access them

they must be putting all their resources on getting pro o3 out or something

[rich surge]

Forget Socrates

Vote for Pythagoras

[native juniper]

GPT having a stroke and should be speaking English….

[regal portal]

Not after that Thereom you created!

[unique pivot]

Speaking of memory, I think 4o is still a ways away from supergenius human replacement...

[regal portal]

sounds perfectly normal to me. You might need to get your hearing checked.

[native juniper]

Has anyone else encountered this?

[rich surge]

How dare you guys not respect Pi

[native juniper]

No I mean it sounds like it’s speaking gibberish

[eternal rock]

Get notified for server events!

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

[regal portal]

I hear perfectly normal english in your clip

[native juniper]

You sure?

[regal portal]

absolutely

[rich surge]

He's sure.

[native juniper]

This is the text it should be reading…

[stiff herald]

What did you do to give it a stroke?

[unique pivot]

just to clarify, I haven't misunderstood the definition of conversation threads have I, 4o is just having a stroke, yeah?

[native juniper]

Idk…

[regal portal]

I'm sorry Pythagoras

chatgpt never lies

[native juniper]

It just happened when I used the voice feature

[stiff herald]

If things start flying around in your house and doors start banging around, you're in trouble.

[native juniper]

Wait so you heard it having a stroke and not speaking perfectly too?

It’s not just me?

[rich surge]

Pfft. A custom gpt tailored to not know anything.

[regal portal]

You guys are crazy..it sounds like perfectly normal english to me

[native juniper]

….

[regal portal]

check your speaker settings

[native juniper]

It’s not speaking correctly…

[unique pivot]

What is this casual nonsense. Plato is clearly the winner here, that's not even his real name but his wrestler name. It translates as "broad" because he wasn't a dweeb

All these other nerds can get rekt

[native juniper]

Someone listen to this to see if I’m not going crazy

[rich surge]

Plato is an ant

[regal portal]

The moderation in this server is such garbage. I got muted for 1 minute to trying to post a discord embedded gif (nothing provocative)

[native juniper]

Dwarf planet if you mean Pluto

[unique pivot]

... an MMA themed wrestler ant would actually be a sick character concept...

[rich surge]

Pluto is not a planet

[native juniper]

Dwarf planet

[regal portal]

Darf PLANET - Planet's in the name.

Dwarves are people too 😦

[rich surge]

It's not a planet

[native juniper]

Is

[rich surge]

Not

[regal portal]

I don't trust a triangle

[native juniper]

It’s a dwarf planet

[unique pivot]

Unlike those knife ears amirite? Don't even appreciate rocks and stones. 😌

[regal portal]

knife ears?

[rich surge]

Dwarves don't make planets

[regal portal]

ohh elves

that's funny i never heard that term

[unique pivot]

You may alternately use terms such as "Leaf Lover"

[native juniper]

Pluto is a dwarf planet because it orbits the Sun and is nearly round but hasn’t cleared its orbit of debris. The IAU reclassified it in 2006, but it remains an interesting world studied by scientists, especially after NASA’s New Horizons mission in 2015.

[rich surge]

It's not a planet

[native juniper]

It’s a dwarf planet

Read

Dwarf-Planet D w a r f p l a n e t

[regal portal]

If you can land on it and take a dump on the surface, it's a planet.

[rich surge]

That doesn't make it a planet

[regal portal]

The true test of planet status: "Is it poopable?"

[native juniper]

It’s a dwarf planet not a planet 😐

[rich surge]

Its not a planet at all

[native juniper]

No it’s a dwarf planet

[regal portal]

You guys might wanna pull this back to AI related

the mods here are sadists

[rich surge]

It's an elf planet

[native juniper]

Dwarf

[rich surge]

Elf

[native juniper]

This guy is a troll just block him…

[narrow viper]

Hey, can you guys focus on the ChatGPT product from OpenAI?

[native juniper]

We are but the troll kept provoking

[rich surge]

No u

[narrow viper]

Ignore and block

[native juniper]

K

[regal portal]

This should put the topic to rest.

[unique pivot]

Well you say that but when I did talk about the topic nobody even reacted 😂

[narrow viper]

Because ... this aint the right place for that topic.

[rich surge]

Yeah this is the ChatGPT topic, not the Pluto topic

[regal portal]

I do hope today's issues with chatgpt are related to them releasing something worthwhile by end of day.

[native juniper]

Okay minimod…

[narrow viper]

Grr I see, look, in any public forum we need to wait for responses. People have other things to do and may not see or choose not to respond to random posts. That's how the interwebs work.

[rich surge]

I wish I was mini...that would be cool

[native juniper]

Keep it ChatGPT related 😜

[unique pivot]

Wait, where exactly should I talk about it acting weirdly around memory then?

[eternal rock]

`` Server Tip: ``

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

[rich surge]

Then I would type on ChatGPT with my mini fingers

[native juniper]

Just reset the memory

[narrow viper]

I mis understood your note. OK, you have my attention while I should be working on a huge prompt. What's up?

[native juniper]

Or turn the memory feature off if you’re having issues

[unique pivot]

Oh don't get me wrong, my question itself wasn't important in that sense, but I do want to know if I'm using the discord channels wrongly!

[narrow viper]

Focus - what's your ChatGPT issue?

[regal portal]

curious what you're prompting?

[unique pivot]

So here, does the flow of the conversation not prove its gotten the new color from outside the thread?

[native juniper]

Isn’t this channel basically for anything ChatGPT though?

[rich surge]

You're using HoboGPT?

[unique pivot]

Or have I misunderstood what threads are?

[regal portal]

That is the Queen's English, my good sir.

[narrow viper]

This channel is to discuss the ChatGPT user interface provided by OpenAI. That includes the browser and apps.

[rich surge]

You mean the King's

[native juniper]

They should elaborate on that then instead of just having it as ChatGPT-Discussions

[unique pivot]

Ok, so I did have the wrong channel I guess! Which one would be the correct one, gpt-discussions?

[narrow viper]

Elaborate on "ChatGPT-Discussions" being about ChatGPT?

What is the challenge?

[native juniper]

No elaborate on what the channel is for specifically

[narrow viper]

I'm outta here.

[native juniper]

Like issues, or just general chatting about the model…

[rich surge]

We should just squeeze conversations into ChatGPT discussions. Not have natural conversations that may have ChatGPT in them.

My ChatGPT is down and I'm upset

I want ChatGPT back

[native juniper]

Become a plus user then…

[rich surge]

I am a plus user

[regal portal]

what were you working on?

[native juniper]

It’s not down for me the model still works

[rich surge]

Asking questions about reconciling general relativity with quantum mechanics

ChatGPT and I have deep discussions

I talk to him like he's my buddy

[unique pivot]

One last question. Is there some resource that would help me clarify which aspects are part of chatgpt and which ones are part of the underlying GPT model? I thought the memory stuff was part of chatGPT but I guess not?

[native juniper]

So are you having issues with the memory feature?

[rich surge]

No. ChatGPT always acts as if it's the latest model and doesn't even speak about which model it is.

Only the general basics.

Like the name.

But nothing deep or enriching.

[unique pivot]

Just trying to build up my foundational understanding of how the system works is all. 😄

[rich surge]

I recommend going on Reddit and finding multiple subreddits around using ChatGPT. A lot of people have many different techniques. That'll give you a good idea.

[native juniper]

Oh you can look at OpenAI’s documentation then

[narrow viper]

Memory is a feature of the ChatGPT application. It is not specific to a language model.
The ChatGPT application saves text into an array of memories and the calls that back in to context to use it during a conversation.

[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!

[unique pivot]

Alright, so that's in line with how I imagined it. It just places the information in its memory into a broader scope, yeah?

[native juniper]

Check your DM

[narrow viper]

Correct, along with the system prompt. The application has client components (browser and apps) that just display data and take in input (plus some other stuff). There is a server component that pulls that memory data into context.

[unique pivot]

That seems like a good place to start, thank you! 🙂

[native juniper]

You’re welcome

[narrow viper]

So about memory .... yes, that is related to ChatGPT.
If you want to understand how the GPT line of LLMs work, try #gpt-models .
If you want to understand AI in general, try #ai-discussions .

If you're talking to the model in ChatGPT, don't get specific about its name or dates, just talk about general concepts. It's great at providing deep insight. The O1 model is better at deeper topics.

If you're trying to get the model to do something specific, try #prompt-engineering .

[unique pivot]

I guess part of my confusion is due to my not understanding how LLMs guard against internal contradiction. Somehow I got it to ignore its own direct contradiction (going directly from yellow to red without supposedly getting more info from outside the scope of that thread even though it evidently did), so I get the feeling that in this case the context inside the scope of the thread is exerting more influence over the response than the information in the more global memory does, even though I kinda assumed the global memory would be weighted more heavily.

I'll do some more reading so I can ask better questions and I'll do my best to ask them in the right channels going forward!

[narrow viper]

You're asking a really really good question which is absolutely in the realm of #gpt-models and not "ChatGPT".

That said...

Your first part there is about LLM contradictions, which would be specific to (probably) the GPT4 models. I see these contradictions a lot and it's really disturbing. I sometimes need to ask the same question a few times to get a "best of five" kind of response.

The second part, about how memory integrates with context, is good for this ChatGPT channel. However, I doubt you will get any insight into that part of the process here, as people are more inclined to talk about astronomy and other products. 🙄

I think the memory feature is half-baked. It doesn't use a lot of data from memory but it seems eager to put some data into there. That's an issue with their server-side code and we can hope they improve on it over time.

[unique pivot]

Do people with the expertise to give a solid answer to that visit this discord?

[narrow viper]

I'm fairly certain that staff read anonymously. They have emphasized that they are paying attention to user feedback. But this server is used mostly by kids and complainers and it's a huge timesuk, so no, staff do not participate ... or rather, extremely rarely.

[stiff herald]

They're watching us from the shadows.

[narrow viper]

LOLz

[unique pivot]

I guess that makes sense. As nice as it would be to talk to people involved on the project, they're bound to be busy and probably want to spend their free time doing literally anything else 😂

[stiff herald]

...just like most official Discord servers. Always lurking...always watching. I remember saying something on MJ's server and a dev chiming in out of nowhere.

...then back to the shadows they went.

[narrow viper]

Developers are passionate about what they do and community feedback can be very energizing and inspiring. But again, the lack of moderation in here, the chaos, the whining and repeated questions ... that makes this environment very undesirable for anyone who actually cares.

[stiff herald]

Oh, yeah...absolutely. I don't blame them, but it's funny when they feel inclined to chime in despite that. It's always amusing because they vanish as quickly as they came, because they know...they know what happens if they stay too long.

[narrow viper]

Yeah, the LangChain server is kinda like that. You're like "whaaaa?" he wrote this code!

Anyway, I'm no staffer but I'm going dark here too. I need to get back to code, but I try to help when I can so I lurk once in a while.

[stiff herald]

Take care!

[narrow viper]

U2

[unique pivot]

Thanks for your help!

[narrow viper]

As with everything else, when it happens they announce it, until then, it aint a thing.

[fervent lake]

mobile canvas is completely broken btw, I’m sure that’s been said before

[narrow viper]

Canvas in general is clearly broken. 🙂 See #1070006915414900886 and #1070006151938314300 and post your own bug report if you don't see it already reported. +1 any report that already exists.

[grand ore]

I hope they release a new model that will be more outstanding.. and not focus on canvas..and features like that.. I want a base that makes it see the full conversation and more questions asked when working with it.

[winter vine]

tomorrow is the release day?

[grand ore]

Of what?

[winter vine]

o3 mini

[jolly temple]

and o3, apparently

[winter vine]

nice

[jolly temple]

also heard that openai is training o4 now so

competition = good for us!

well not good for our jobs but itsokay it can help with my studies for now lol

[winter vine]

I think the same as you. For now, it also works for me for studying. Let's hope it stop improving at some point; otherwise, it will get complicated.

[jolly temple]

Agreed

[lament thunder]

Hey yall! Is anyone good & up to date with image analyzing in GPT 4? I'm trying to have it scan IG profile pictures, determine if they contain a business logo, then determine if the logo is cut off on the IG circle crop? I've got it to be able to succesfully identify logo's, but it keeps getting the dimensions wrong.

[cloud pewter]

increase the thread size.

who can utilize 128k context size properly when the chat deletes itself because of too long after 20 messages

[lament thunder]

Okay, will test this!

Thank you!

[rugged igloo]

Stealth update ?

[flint zenith]

I now have a "Think" button for GPT-4o 👀

[sick pecan]

why would i eever want less intelligence?

[flint zenith]

Look like it can't search the web when you have that selected

[sick pecan]

doesnt work with custom gpt either

[rugged igloo]

Wow they really don't like smart search I wonder what there scared of

does this mean i get usuper ultra thinking lol?

is it using 01 mini?

[sick pecan]

when i use Think mode with o1-mini i get shorter responses?

[native otter]

Just the same toggle in "View tools" button placed directly in the message box
Redundancy for the sake of fear

It doesn't impact anything

[wise anchor]

limits on o1 for paying subs is soooooo annoying

"here pay more for another limited version"

[native otter]

All the 4 different ways to switch to o1 💀

[wild flame]

I mean It's not bad but it needs to be optimized a bit

[native otter]

They added the think button intentionally

Just to remind people that their models can think, which is sad to watch lol

Very directed to people that haven't used ChatGPT other than to say Hi and were blown away by DeepSeek

[wild flame]

Yep I imagine it's the fault of a Chinese AI that made Nvidia suffer on the stock market a few days ago

[wise anchor]

it’s just bc most ppl don’t even have access to o1

[sick pecan]

i aint gonna use think button because it doesnt give me full code updates

[wise anchor]

I go to college and a lot of students don’t even know what o1 is

[sick pecan]

i dont think its a good feature

[wild flame]

So true

It is for them to discover fire

[wise anchor]

also most students are horrible at prompting and have zero idea how a model actually interprets information and what it’s capable of/how it comes to conclusions

[native otter]

As discussed on the server tho, it may just be a distillation of o1, which means it would be stolen, but hasn't been fully confirmed yet (and honestly not sure how such thing can be fully confirmed) but there are many clues hinting at it like how it's suspiciously cheap, very similar response style to OAI models, and suspicious calls from Azure services which host OAI models like o1 and offer distillation and fine tuning

[native otter]

must be chat ge pt or something idk

please make me breakfast

type of prompts

[wild flame]

Or maybe it's a Chinese copy of o1 there is no conclusive evidence to confirm the last thing you said As I have always said, innovation comes from the strangest places.

At the moment I have not used Deepseek much due to lack of confidence.

I still like gpt chat more because of its integration with my projects and even interaction with me

[native otter]

I mean, honestly if really their own, then that's awesome work, I don't exactly like china, but doesn't mean I would hate anything that comes from it, but the claims seem to be reasonable, not too far from the truth...
The fact that it responds like GPT, and the suspicious Azure activities do strongly hint at it

That coming from china won't be surprising, as is often the case with the stuff they do...

[wild flame]

You make a good point but unfortunately there is no evidence (either to support or disprove that) In case you're right wow Although we are also going to highlight the fact that it supposedly surpasses Claude and chat gpt, then is it really an exact copy of chat gpt or? A Chinese version?

[sick pecan]

a merge of both would be better than the individual parts

[native otter]

I tried sending a link to Financial Times article, but got a timeout instead lol

If they used OAI models to train other ones via distillation and marking as it's own thing and a competing product violates TOS

So if they really did it, it would be a stolen model

A real competitor is something like Claude or Gemini where they built their own thing

[sweet arrow]

Someone should try access the codes of the Chinese Tofu one and compare to see any similarities.

[worn wing]

with the plus subscription, I got a think button that forcibly enables the o1 version in the dialog, and what I noticed is that the limit of 50 messages per week from o1 has changed, does anyone know anything about the change in the limit?

[rugged igloo]

i think its 100

[inland heron]

anyone tested this? Any differences?

[sick pecan]

it makes my o1-mini responses so short they are useless

[spice star]

is it just me or can o1 not actually use the canvas?

[inland heron]

I'm using it on 4o and curious to see how it affect o1. I thought o1 already does the thinking? So all of them can now think a bit more?

[sick pecan]

from my perspective it thinks less. but it does have a UI element like some other chatbot that we shall not name

[spice star]

doesn't the button just use o1? (same as /think)

[inland heron]

that's what's important here. The actual performance is not important.

[inland heron]

Yeah but you can enable it on o1 as well

[spice star]

my understanding is that it doesn't do anything with o1 enabled though

[inland heron]

I'm currently testing a coding problem/solution on o1 (non-use more intelligence) vs 4o (use more intelligence).

[silk flame]

What is this "think" feature? I can use it with 4o, is that count as an o1 message ?

[torpid kelp]

Its just o1

Just another way to switch to it without the top dropdown

Or actually maybe im wrong, think that was called “use reasoning”

[inland heron]

OK so my one-shot test with medium-to-large code context of o1 (non-extra thinking) vs 4o (extra thinking) resulted in o1 winning. Double-checked and confirmed with both, and both of them agreed that the o1 solution was better. 4o solved the issue, but it wasn't as good. Retrying with 4o again to see if it gets it right this time. Again 4o did the same. So for this test, o1 wins vs 4o (using more intelligence). (test was about regex and formatting in python)

[iron roost]

i think o3 mini is likley to be march, it seems openai isnt really spooked by the whole deepseek thing so will run their saftey tests and drop as perviously stated

lots of people claiming full o3 pro would drop today, but seen no sign of it

but o1 pro does seem to be a lot smoother today, and they did i think update the knowledge cutoff to June 24 from 2023 so thats a plus

[pseudo shell]

Hi everyone,

I am trying to scrape the full text content of an uploaded docx in a custom gpt. whatever i do, it does never seem to scrape everything, changes the format and content. I thought this was due to the limited context window but even with shorter docx, it always changes things. I forced it to adhere strictly to its instructions, even with python code to have less room for false interpretations, still it gives me very unreliable output.

does anyone have experience with scraping content from files and then processing it correctly with custom gpts?

[iron roost]

a bit like sora, it seems a spin is put on all outputs, which you do get used to mbut maybe less useful for coding

[obtuse nymph]

oh. ok.

[dense halo]

Weird, gpt4+ thinking?

It works with any of them

[spice star]

it's just using o1, you can check which model generated the message by hovering over the regenerate button

e.g.

[obtuse nymph]

oh

i think rename that.

[dense halo]

But is it using o1 to respond as well? Not just for thinking?

[spice star]

o1 is generating the entire response afaik

[dense halo]

If so then it makes sense why they added it for o3 mini

Since o1 isn't selectable on free

[spice star]

I've tried enabling think with a conversation I set to o1-mini and it just uses o1 to respond instead

[dense halo]

I asked it what is latest data training data cutoff date was and it wasn't sure

It says this using gpt4 model with thinking enabled

Let me try asking only o1

[spice star]

it can only be a reasoning model if it's thinking

[dense halo]

Yeah it's similar

[brave valve]

That's openAi's answer to deepseek. The UI button 🙂

[rocky kayak]

AI is getting so out of hand we have AI losing its job to AI.

[whole roost]

Oh god no emojis, I hate them so much, my whole custom instructions are created to prevent it from using those stupid smiley faces

What are they thinking

I'm not here for talking to an idiot with emojis

I want my "certainly" ai

[rocky kayak]

At this stage, we have essentially turned ChatGPT into a servant. fr

Not merely an assistant but an outright slave.

[whole roost]

It's a tool, don't get emotional over it

[rocky kayak]

I am not, I am merely pointing it out.

[whole roost]

My car has to drive me around 8 hours a day as well

[rocky kayak]

AI is merely a tool, much like a car, as it lacks the ability to reason. What concerns me, however, is the development of AGI.

[whole roost]

Agi is a future idea with no reasonable time frame of it happening so nothing that would concern me right now

I'm more concerned that GPT is still doing a bad job in longer conversations and that o1 doesn't work on projects with files

[rocky kayak]

Some speculate that it could emerge by mid-2026, though I doubt we will even have a sufficiently advanced prototype for public use before 2027 to 2029.

[whole roost]

Altman Said it isn't happening anytime soon

[rocky kayak]

I mean, yes, that's fair enough, but even the O1 model is still in its early stages.

[rocky kayak]

I know

[whole roost]

With o3 pending o1 shouldn't still be in the early stages

What prevents them from at least activating memories I wonder

[rocky kayak]

?

What do you mean? I'm pretty sure ChatGPT has memory.

[whole roost]

Not on o1

[rocky kayak]

oh, i thoguht You were refering to all models

[whole roost]

Nah I'm mainly using o1 but it's really limited because it can't do most things

[rocky kayak]

Yes, the O1 only retains memory of the current conversation.

[whole roost]

At least it has canvas now

[rocky kayak]

yea

[whole roost]

I noticed a few weeks ago that they increased the emojis, and I hated it right away

[rocky kayak]

Indeed, you're absolutely right, even Deepsek R1 overuses an excessive number of emojis.

At this stage, it may as well be communicating in Wingdings. (example: ☟⚐🕈 👍✌☠ ✋ ✌💧💧✋💧❄ ✡⚐🕆)

[whole roost]

Maybe it will be easier to prevent when the new custom instructions arrive, I don't have the new interface yet

[rocky kayak]

yes thats a sentence that actualy says somehting

[rocky kayak]

on ChatGPT or on deepseek?

[whole roost]

GPT, I don't really care for deepseek, have tried it, was ok but GPT still reigns supreme

[rocky kayak]

yeah just overall better user expirience

[whole roost]

And better tools

[rocky kayak]

However, I have observed that Deepseek excels at coding, particularly on the front end.

[whole roost]

I can try it for that although I didn't have any issues with o1 pro

[viscid merlin]

hi

[rocky kayak]

That is reasonable, but for free users, Deepseek R1 remains their optimal choice, despite the servers being somewhat unreliable.

[rocky kayak]

Howdy

[viscid merlin]

is there any way i can try sora before buying?

[whole roost]

Ah yeah, I'm on GPT pro so I basically just use what works best for me and while I need a few more things on GPT it has the best overall package of functions and I don't know if I can live again without projects and canvas

[rocky kayak]

I don't believe so; you could try exploring other videos by different creators, but personally, I don't think the pro version is truly worth it.

[rocky kayak]

My favorite aspect of Plus is that it essentially allows you to create an AI replica of yourself.

[whole roost]

I'm in EU so Sora doesn't exist here, no opinion on it

[rocky kayak]

I feel the same way, I simply use a VPN when attempting to access content like that. I fail to understand why it even exists; why limit access to a single country unless it violates any EU laws or regulations?

[whole roost]

I don't know but I'm also not very interested in video creation
I have pro for unlimited messages on o1

[rocky kayak]

I mean, if you have the financial freedom, it is understandable.

[whole roost]

I think it should be cheaper for EU since we don't have Sora and video calls but alas

[rocky kayak]

yeah

still tho 200 is insane

[whole roost]

I cancelled a few other subscriptions to make it work

[rocky kayak]

You might consider attempting Gemini 2.0 on Google's AI lab, though it is rather mediocre.

for video calls

[whole roost]

Gemini has been a disappointment overall to me I have to admit

Even with their new models
it's really strange

[rocky kayak]

Even Gemini Live struggles significantly with accurately recognizing words.

[cloud pewter]

Bro has pro of course gpt is better. but us plus users are suffering while paying

[whole roost]

Most of the time I use 4o on projects, I had plus before but needed more generations
It's alright but I wished it was o1

[rocky kayak]

What am I supposed to say when I'm on the free plan?

[cloud pewter]

free is free, you don’t pay so you get what you get
pro is a fat amount of $200 so you get almost everything
but plus is the cringe middle child who gets slightly more than free but gets locked out of flashy new features or has zero qol updates

[still hound]

Why does GPT take so long to answer and its writing lags out? What is happening?

[whole roost]

How many generations does free plan have on o1 mini

[cloud pewter]

0

[still hound]

Did they take all the resources and gave it to the Chinese?

[cloud pewter]

80 per 3 hours is NOT enough

[whole roost]

Exactly why I upgraded

[rocky kayak]

works fine for me

[whole roost]

Now I'm using it excessively though, I might be the reason for their financial losses

[still hound]

On canvas? What model?

[cloud pewter]

there is
no middle ground
no limit increase although the gpt4o is lobotomized for a million of times it might as well become the-better-gpt4o-mini

[whole roost]

I only use o1 mini to store memories

Uh 4o mini

[rocky kayak]

I dont know on canvas let me check

[rocky kayak]

run out of 4o messages:/

[cloud pewter]

teams exist but it’s harder to subscribe and it doesn’t even have proper sora or expanded access to o series

[still hound]

Starting a new convo on normal GPT has normal writing, but trying with custom GPT/canvas, I get it lagging like 3-4 framerate/seconds.

[inland heron]

anyone tested "use more intelligence" with o1?

[winter vine]

Why is sora not allowed there?