#chatgpt-discussions

1 messages · Page 95 of 1

sterile moth
#

How does a custom GPT and Custom Instruction work togehter or interfere with each other?

Are custom instructions ignored when I talk to a custom GPT? Or are they injected anyhow?

dense halo
#

Yes I experienced that too. I think the desktop app did it too

#

Still has quite a lot of bugs. Shouldn’t take too long to refine it I hope.

normal quarry
hardy sigil
#

when will GPT 4o voice be available?

tardy totem
#

In "weeks"

hardy sigil
#

the one demonstrated in the demo launch?

feral zinc
#

"GPT-4o real-time voice and vision will be rolling out to a limited Alpha for ChatGPT Plus users in a few weeks. It will be widely available for ChatGPT Plus users over the coming months."

proven sluice
#

any idea when the new voice feature will release??

honest ledge
shrewd snow
#

Is it just me or the voice Sky is not available?

feral zinc
#

its gone

shrewd snow
#

I like that voice sm

feral zinc
shrewd snow
feral zinc
#

🤷‍♂️

little storm
#

I just checked again it does say alpha

wheat lava
sterile moth
wheat lava
#

Glad to help!

proven sluice
feral zinc
proven sluice
#

oh well, thanks!

toxic palm
#

hello guys, im unable to login, when i login it redirect me to login page, its dawn?

swift osprey
#

I wish they would add an option to completely disable 4o

#

keep on accidentally using it

dusky cliff
#

a couple of weeks have passed

scenic halo
#

Has anyone else had all their memories wiped?

#

(Gpt4o)

ivory lintel
dusky cliff
sinful marsh
#

Hi! I don't understand what GPT4o does, and how it is different from GPT4 and GPT3.5. If someone has the answer I'd be glad to know 🙂

regal portal
#

it’s better than 3.5.
If you’re a free user, you get to use it for free
If you’re a paying user, you get to use it for about twice as long as gpt4

#

4o you can show it pictures, which is really cool. 3.5 you can’t
(in case you’re a free user)

swift osprey
#

My understanding is that 4o is a dumbed down version of 4 but still better than 3.5

#

4o is faster to respond although speed hasn't been a big issue with 4 lately

sinful marsh
#

Ok, cause if you wanna pay, you get unlimited GPT4 but not unlimited GPT4o

elfin birch
elfin birch
feral zinc
swift osprey
#

the biggest difference between 4 and 4o is that 4o will talk your ear off for literally anything you ask it

regal portal
sinful marsh
#

Ok, so when you pay, you get by default 80 tokens of GPT4o, then 40 tokens of GPT4, unless you'd like to use GPT4 before GPT4o? And GPT4o > GPT4 > GPT3.5, even if GPT4o is acessible as free and GPT4 isn't if I sum up everything

feral zinc
regal portal
regal portal
sinful marsh
#

So basically 80 messages every three hours overall?

regal portal
#

on 4o, yes, but 40 on 4.

#

Think of it as having a smaller fuel tank

#

but the same fuel efficiency (miles per gallon)

sinful marsh
#

As far as I understand, GPT4o has better performance than GPT4 and more messages, why would a paying user use GPT4?

regal portal
#

I’m not an expert, but I use mainly 4o for the higher message count. I also think it codes better.
But I have also heard reports of the opposite being true with 4. Perhaps it comes down to how each person dictates info to it.

#

One thing for sure is that 4o by default likes to give longer responses which can be a little annoying (unless you modify your memory or custom settings)

feral zinc
#

4o and 4 perform differently in some uses cases so instead of instantly removing 4 they left it for a while (grace period) and will eventually remove it

sinful marsh
#

Ok, overall 4o is better but in some cases 4 is better. Thanks for the information! Do we know in which cases 4 perform better than 4o?

swift osprey
#

I'm so glad I'm wrapping up this coding project I'm working on if they are getting rid of 4 and will only have 4o lol

#

just try both man. everyone has different opinions on them

regal portal
#

or prefer not to?

sinful marsh
#

Can still give it a one month trial though

swift osprey
#

I just really hate 4o. It talks way too much. Custom instructions barely help.

regal portal
regal portal
#

that’s what i’m doing, this is my first month on subscription

sinful marsh
#

That's what I'm thinking about. Btw, is the information constantly updated? If I remember well, 3.5 has information before end of 2021

regal portal
#

honestly the picture feature is so fun (wasn’t available when i first subbed because there was only 3.5)

feral zinc
#

if you are in eu you can cancel within 14 days with a full refund

regal portal
sinful marsh
#

Ok great

sinful marsh
#

Thanks for the help guys!

regal portal
#

You’re welcome.
That’ll be 20 bucks for the info we gave you. 🙂

lost saffron
#

is it possible to have gpt 4o run faster if i use api?

upbeat locust
#

Hey everyone 👋
I can't find the ChatGPT desktop application on the OpenAI website. Can someone explain this?

feral zinc
upbeat locust
regal portal
feral zinc
analog nymph
#

i have premium gpt. whats the difference between that and regular now that regular does everything premium did?

feral zinc
haughty citrus
#

What happened to the good voice with chatgpt audio?

echo plinth
#

Is the web app broken again? I complains about not being able to generate a response.

upbeat bison
#

it is broken

#

I got same issue it freeze and need to refresh to see the full response

strange patrol
echo plinth
feral zinc
strange patrol
robust sparrow
feral zinc
robust sparrow
jovial flint
#

Is there any way to set up a Teams subscription so it applies to my existing account, so I have access to edit all the custom GPT's I've created and have my conversations saved to my account instead of a separate workspace that I have no idea what happens to it when my friend decides to stop paying for the subscription?

normal quarry
jovial flint
normal quarry
#

or find a way of ensuring your Teams subscription stays in good standing

jovial flint
#

What happens to a Teams workspace when the subscription is cancelled?

normal quarry
jovial flint
#

That's fine

#

Thanks anyway

normal quarry
# jovial flint That's fine

my advice in this fast moving field is to closely keep all of your own stuff to youself and not rely on the actions of others to preserve your access

feral zinc
#

all data is locked if workspace is deactived

jovial flint
#

We aren't actually a business

normal quarry
robust sparrow
feral zinc
robust sparrow
feral zinc
past rampart
#

hello guys, i was planning on buying ChatGPT plus. Will i then have unlimited access to GPT 4, so can i ask questions or write?

robust sparrow
feral zinc
spiral siren
#

Guys is chat gbt 4 going free ? So should I unsubscribe from the premium one ?

robust sparrow
past rampart
feral zinc
feral zinc
spiral siren
past rampart
feral zinc
spiral siren
dense heron
#

is there a message cap for ChatGPT 4o? I know there is a message cap for chat GPT 4 which for me is 40 messages every 3 hours

normal quarry
dense heron
dull thicket
#

It takes 4 very clear promts to stop 4o from formatting answer as a list. I find those lists very frustrating, it is hard for me to go through the list way 4o is making them. I don't mind list. But list that has to be scrolled throuh three screens and it can be a short paragrapg, is not list i like. And it takes 4 prompts clearly stating to rewrite as a paragraph until 4o catches on.

#

80 prompts? When every few prompts i have to repeat to stop formatting answers as lists, and every feedback is treated to double up answers to long gone question?

untold scaffold
#

why am i paying for plus when free gets pretty much everything?

feral zinc
dull thicket
#

Bs about stuff people don't understand drowns valid criticism

wanton lichen
#

So to confirm, custom GPT's are using 4o now right?

feral zinc
#

yes

wanton lichen
#

wicked cool

sick pecan
#

I just manually copied my gpt configs from personal to teams. gave me the opportunity to polish them up as well.

#

I see there is a merge feature now, but it explicitly says this is a one way destructive non reversible process (=bleagh)

untold scaffold
dull thicket
# untold scaffold go find some clear criticism then buddy

I have one
Everything is formatted in redundant spacious lists, format that looks like it was made for the sake of making lists. Just now, it literally repeated the same sht 3 times, in intro of a message two list items, 3 times, with slightly different heading. To do A,B, we need to do AB. Proceeding with those, we will do A,B. Here is A, B for you. And then goes into giving actual answer

high sphinx
sick pecan
#

rewind 3 weeks and typical venting session topic was it was too lazy. now it is too verbose

high sphinx
#

it's a very hard balance to get

sick pecan
#

and yet nobody ever shares their custom instructions

#

suggesting the expectation is the machine knows what you want, without ever being told

#

maybe that works with other humans because of the out of band communication (frowny face, arms akimbo)

high sphinx
#

But if i continue to ask it more questions about the same topic, it shouldn't keep giving me the whole installation process again and again

sick pecan
#

'shouldnt'

#

this is an expectation word

high sphinx
#

I'm a little picky now but it's such small details that make these LLM's stand out from each other 🙂

sick pecan
#

here is an example of a custom instruction that reduces verbosity (on the single sample i shared) by more than half => #chatgpt-discussions message

#

and yet there are so many ways to instruct the machine. not just the one i present

#

although i typically use a blend of einstein+hemingway. because i like smart answers, not just succinct

regal portal
#

The issue isn’t giving it instructions to be more concise, in my opinion, but the fact that 4o seems to forget them after not long. I just haven’t had any luck maintaining my custom or memory settings past a few pages of chatter

sick pecan
#

custom instructions are posted with every prompt

#

maybe thats the issue here.. ppl dont know about, understand, or are skilled at using custom instructions

regal portal
#

really? But I have some very specific instructions and I see them getting ignored after a while. They’re always active in a new session but they just disappear after a while

high sphinx
sick pecan
#

ahh yes well the amount of custom instructions begin proportionally as most of the chat. but over time become diluted by the chat itself

regal portal
#

I hope they address that. Personally I see it as the main flaw of the current build. Other than that I really do enjoy it.

sick pecan
#

perhaps not intuitively or clearly explained

sick pecan
#

so yeah... if ppl knew how to use custom instructions i imagine there'd be a lot less complaints

high sphinx
regal portal
#

my current settings:
use excessive explicit language to give emphasis. Use real swearwords, uncensored swearwords.
Respond in a very ghetto manner.
Keep answers short and concise.
occasionally address me as (removing this part because it’s no no words).
occasionally make a fart reference that fits the context of our conversation.

high sphinx
#

I think they should make some easy UX for casual users with the custom instructions. Not just a text-box. More like "options" on what you prefer. Most people don't even know themselves what they really want..

sick pecan
#

for sure. the chatgpt UX is terrible

#

if openai didnt have the most powerful models, nobody would be using it

high sphinx
#

yep, at least they updated it a bit recently

sick pecan
#

mostly decoration updates but yeah is better than a year ago

high sphinx
#

Would also be nice with some organizing tools. E.g. bookmarking chats. Not quite sure if it would be of much use in practice, but sometimes I've wanted to go back to old chats but never found them.

elfin birch
sick pecan
#

well explaining the importance of custom instructions. or perhaps having a complaint(/feedback) button that gives an opportunity to explain

#

some sort of meta assistance

#

it could even go, 'How did i do?'... and if user says too much waffle then it could go, 'we could update your custom instructions'

#

insert custom instruction wizard

high sphinx
# elfin birch If I may: how can it be improved?

I think the custom instructions could get an overhaul. Maybe predefined choices on how the chat should look like that you can choose from, but with the possibility of full custom. I myself do not always know what to write in the custom instructions. :p

sick pecan
#

maybe some starter templates for custom instructions, endorsed by the vendor, or crowdsourced

#

what gorlock said

high sphinx
#

Or save your different custom instructions and just choose them from a dropdown in the chat? So you could easily swap. Or will this just become a gpt?

sick pecan
#

yeah thats been a desired feature since day 1

high sphinx
#

I guess they have a lot to work on

sick pecan
#

hire more ppl

high sphinx
#

they probably are

#

All I want atm is just the voice v2 to come out 🥲

sick pecan
#

well yeah i think thats the bigger issue. all this chatgpt UI becomes irrelevant(/insginificant) when we're talking to our alexa/siri/scarjo whatever

ornate willow
#

Have any of you experienced the voice chat the way they hyped in their 4o introduction? I keep going to the app regularly and trying it out and I feel like they still haven’t actually rolled out the full update.. or they opted to remove some of the capability they hyped. Am I missing something or have they just not fully rolled it all out yet?

high sphinx
sick pecan
#

havent even heard reports from the planned 'alpha group', whoever that is

hollow lotus
#

what even is different?

sick pecan
#

but youd expect to see someone bragging on twitter when they got access

high sphinx
#

imo it feels more alive

#

and understands tone and expression i guess? and also has video feature

ornate willow
#

Just making sure I wasn’t missing something, thanks

high sphinx
sick pecan
#

just hold up the phone like you're talking on a phone call. thats what we used to use phones for

ornate willow
#

Seems more like they rushed to beat Google’s latest announcement while not being prepared to actually roll much out

high sphinx
#

yeah

#

it sucks

ornate willow
#

The ability to just talk and be able to interrupt it or to have it actually use my camera in close to real time is def exciting and I’m hoping we something soon

#

*see something soon

high sphinx
#

"coming in the upcoming weeks" three weeks ago not a single sound heard since

ornate willow
sick pecan
#

yeah i wfh so talking outloud to the machine is no issue

#

but i can see how ppl commuting dont have that option

high sphinx
#

Would be so nice if Siri got replaced with the gpt voice thing

ornate willow
#

Perhaps if our friends and family were as smart and helpful as GPT, we would have been talking to them in voice a lot more.., 😂

sick pecan
#

that makes most sense

#

i dont wanna call someone and get trapped talking about their agenda

ornate willow
sick pecan
#

i just heard of plans for paid alexa sub to bring chatgpt like features

high sphinx
#

Yep, just wondering if it will be local or server-based

#

With a much smaller model ofc

ornate willow
#

I have never used Alexa tbh

#

Alexa, you left the oven on again last night.. What do you have to say for yourself…

ornate willow
# high sphinx Yep, just wondering if it will be local or server-based

Seems like copilot will be trying to keep things more on device.. same with plans for what apple will roll out.. there’s also discussion of having sort of black boxed remote AI… essentially your own ai in the cloud isolated for privacy.. which will be only really great solution when really high performance is needed that the local hardware can’t provide.

past rampart
#

hi! is it worth buying chatGPT plus because i use GPTs very often but with the free version i have a limited of messages

ornate willow
#

I use ChatGPT all day for misc things.. I rarely watch Netflix… based on that I feel sub for ChatGPT is worth it and my and I’d be much more inclined to cancel Netflix to save money lol

past rampart
night copper
#

Hey, is anyone having issues, where if you ask chatgpt for an image, the sides or top are cut off? Seems like a pretty fixable issue, I tried asking for images of haircuts, in terms of, an image displaying multiple haircut examples, the edges, whether the top or the sides are always cropped out, I tried multiple times, same issue?

dull thicket
civic quail
#

Has anyone converted pdfs to text for a gpt? I don't want to reinvent the wheel if there's an optimal way to prepare them.

dull thicket
civic quail
#

Definitely worth it.

inland thunder
#

Any word when Chatgpt will be working normally again? Mine is slow and useless as can be.

ornate willow
#

worth is hard to calculate.. that's why I don't wanna tell ya what to do... if you need it and can spare 20 bucks for it then for sure worth it.. if you're struggling to pay your bills and 20 bucks will take food off your table.. then perhaps not

civic quail
#

I say make a concious effort to save twenty bucks when going to buy a coffee or something. When you get to the checkout line in the grocery store, remove fifteen bucks that would be spent. flip it.

inland thunder
#

anyone else having big issues with their chatgpt?

ornate willow
#

interestingly enough I just had it read a pdf and it didn't have to use code interpreter or anything to do it.. it just instantly summarized it to text for me without any "Analyzing...."

#

but when I asked it to give me a pdf that's when it used code interpreter

past rampart
rustic kelp
#

hey all, got a question. if you go to this page (https://openai.com/index/hello-gpt-4o/) and choose "Character design - Geary the robot" from that menu.. I tried saving the image of that robot, and using the same prompts, and it definitely does not have character consistency. I'm using ChatGPT-4o. what am I missing? is it an upcoming feature in 4o that's not out yet?

ornate willow
#

likely another feature they haven't fully rolled out yet.. I tried it about a week ago myself without much consistency

sick pecan
#

the pdf parsing issue comes up a lot. typically the answer is to convert the pdf into something like markdown or text.

#

ironically portable document format is not so portable

dull thicket
past rampart
upbeat bison
#

I think they made some change as chatgpt gets way dumber. Its just useless now

sick pecan
#

its the best model i have ever used

upbeat bison
#

In the morning he was doing what was asked now he just doing whatever he wants

wet shell
#

"Hey, can you write a small simple script so that I don't have to look this up"
"Sure! here you go: ..."
"Script doesn't work, there's no output"
"Oh, it seems there was a problem, here's the corrected code: ..."
"That's the exact same code, line-for-line"
"Oh, yeah, sorry about that, try this code: ..."
same code again 🤬

upbeat bison
#

I needed to ask him 4 times same thing until he done it and it was nothing hard.

#

Before he could do some very complex things with ease

#

I guess its best to give up today if he output me same code for over 10 responses stating that he change something

sleek gust
sick pecan
#

best to get it to explain the changes before updating the file in full

#

(which was the old default behaviour... that everyone complained about)

wet shell
# sleek gust That's not good. It shouldn't be like that. Have you tried showing it the error ...

There was no error, I was trying to trap an escape keypress and set a global flag in a python script, when I pressed 'esc' nothing happened, even just running the provided example script -- turns out I need to use curses for this since my script is in a linux terminal, the python 'keyboard' library doesn't work there... point is though, it just keeps mindlessly outputting the same code when you ask for suggestions as to why it's not working, LOL -- in the end I wrote the code myself, took only a couple minutes anyway

#

solved via google search instead, in the end

#

sam altman, if you're reading this, I need a smarter model, not a model that sounds like scarlette johansen 😂

sick pecan
#

sama doesnt read vent-cord

sleek gust
#

I think sometimes the code Interpreter is not required and even make stuff worse. It should just use the text model. Not code interpreter

#

I don't like code interpreter at all

upbeat bison
sleek gust
#

I'm pretty sure it will be better

wet shell
#

It didn't invoke the code interpreter, I had it turned off before but I can't find the setting for that now on the newer web UI 🤔

upbeat bison
#

Its in options where you put instructions for chat-gpt

wet shell
#

ah, thanks, I see it there

civic quail
#

Marker: This Open-Source Tool will make your PDFs LLM Ready

sick pecan
#

includes ocr support, neat

ornate willow
past rampart
#

I already know that but I need more messages for my project, so I am thinking

ornate willow
#

gotcha

violet grove
#

Why did GPT 4 become so dumb when it comes to coding

#

its really sad to see

gusty ledge
#

gpt4-0613 still to this day seems to just be the best for coding sad its api only

violet grove
gusty ledge
#

I think when its context expanded to 128k it got significantly worse with its short term memory even when your not using alot of tokens

#

I mean its cool you can import an entire code base now but when it ignores the question you have about it its annoying

sick pecan
#

gpt4 has 32k context, gpt4-turbo and gpt4o have 128k context. it might be oversimplifying to suggest it related to context size when the models are considered to be different enough to have new names

#

but yeah your favourite model is the most expensive one

#

that said, when gpt4 was in chatgpt it was limited to 8k. the larger 32k was/is only ever availble via the api

#

ie gpt4-turbo(/gpt4o) is including more of your chat history. so you'd want to be on top of your relevant context game these days

dense halo
#

Is it me or has the PDF capabilities improved? It can reader longer pdf’s now

sick pecan
#

or the pdf capabilities stayed the same and now it warns you when working with an oversized pdf

brisk path
#

128k is small, ChatGPT limits to 32k which is abysmally tiny

sick pecan
#

chatgpt plus/team limits to 32k. enterprise supposedly get the full 128k

glacial violet
#

Y'all rate my pfp

desert gazelle
#

I've been frustrated that it seems like gpt 4o isn't giving as detailed or thoughtful responses

#

As much as I love the cheaper price, I'd rather pay extra to have a quality product

sick pecan
#

if money is no object then you could spin up a chatgpt clone (like open-webui) and use the api for the model you want

gusty ledge
sick pecan
#

yeah i managed to run up a huge gpt4-32k bill a year back. thats why i now have so much appreciation for chatgpt sub (which clearly many dont)

winged bluff
#

chatgpt is getign worse at following isntrucitons i gave it simple 3 instrucions and it ignored 2 of them i t only focused on the first one i have to alwasy chagne my isntructions its gettng worse and worse instead of better im thinking im better off switching to another language model recently

valid loom
#

is it normal for gpt 4 to mess up while rearranging long lists

dense halo
compact gulch
compact gulch
sick pecan
#

yeah i heard that too. but has likely been upskilled using synthetic data from gpt4

compact gulch
compact gulch
#

Im not entirely sure what they did for GPT-4T though. Param pruning? Idk lol

sick pecan
#

in any case. openai be burning money giving us gpt4 superpowers. is understandable they gotta find ways to cut costs. at least they are adding value while downsizing the model

#

and i'd rate gpt4, gpt4-turbo and gpt4o over anything coming out of google

compact gulch
sick pecan
#

and rewind a few months and there was no competition to gpt4

compact gulch
#

GPT-4o is definitely apart of that iterative deplyment thing when considering its new multimodal features. Atleast they have introduced that (or will) to the user base, but there are still things like agents, continuous learning etc. they may want to slowly rollout across the year

compact gulch
compact gulch
sick pecan
#

pretty sure moe was the breakthrough

compact gulch
#

GPT-5 will probably be the most unqiue model of the GPT series

sick pecan
#

and i suspect 3.5 turbo was distilled using gpt4, which was moe

#

but they have run out of training data. so we will see if their solutions to this pay off

compact gulch
# sick pecan pretty sure moe was the breakthrough

MoE wasn't really anything new. I think the first concept of MoE in ml traces back to like late 90's, and it had been applied in relatively large models like the Switch Transformer (this model was released in 2022 and had like 1T params and was a MoE)

sick pecan
#

sure. but gpt4 pushed this into trillion+ parameters, which practically was only achieveable with moe

#

and the benchmarking suddenly jumped from gpt3 to 3.5turbo to 4

#

maybe it was just the grokking. who knows

compact gulch
compact gulch
sick pecan
#

meanwhile the full capability of this gpt4o multimodal model is yet to be seen

#

is possible the visual/audio is mindblowing. or just hype

compact gulch
compact gulch
# sick pecan is possible the visual/audio is mindblowing. or just hype

Will GPT-4o's image generation shouldn't magically be superior to midjourney v6 in every way (i mean maybe it will be, idk what they have done for the image generation specifically, how many images it has been trained on etc.), but there are a lot of doors that are opened up with a full image modality and the quality should be quite decent

#

I think just having these new capability open up will feel pretty magical when we first get access

#

And having an actual intelligence behind the image generation now should feel a bit different in a lot of use cases

#

Well, intelligence like that which we say in current LLMs

#

GPT-5s multimodal capabilities will be a sight to behold though imo

sick pecan
#

or it turns openai bankrupt and we're stuck with google ai (that wants us to eat rocks)

compact gulch
sick pecan
#

in the ai armsrace whoever stumbles will be left behind

#

basically they gotta bet the farm every time

compact gulch
#

And internally they expected their revenue to go up like 5x or more according to a report from last year lol, so i think they have some big things planned in 2024

compact gulch
#

I wonder if GPT-5's video generation will exceed Sora's abilities

#

There are a few things that factor into that though, but most likely

lucid pilot
#

So GPT 5 in the next 3 months? GPT 4o is cool and all but it doesn't match up with the tech demo to be fair neither did googles but the next big jump can't be that far off could it?

sick pecan
#

they are slowly rolling out features to maximise attention

#

and lets be honest, they just wanted to mock google. they werent ready yet

dense halo
#

for f sakes. latex is down on the macos app again

#

Huge memory leak too

compact gulch
lucid pilot
#

I'm pushing for more use in our work place with AI although most of the staff aren't really computer literate but there are many use cases where AI could help and if it can give the company an edge over the competition that's even better.

dire goblet
#

Sooo how do if I just if I got the new talking feature? I’m a paying customer and don’t think I still got the new feature yet? Anyone else get it? Where that ai sounds more human like? Or does it have to be a specific voice?

lucid pilot
dense halo
#

Bring Sky back!

dire goblet
dire goblet
dull thicket
#

Incredible. I asked gpt4o to provide script that will create a certain file. It provided the full file back, not the script. I was asking about it, had productive feedback. Hour later, i ask for json file, and thing just blatantly refuses to give file and says it cannot generate. Forth and back, then trying to get the text then, it completely lost it's marbles and started answering something long gone.

eternal rockBOT
dull thicket
#

I am just dumb, and talking to a deaf blind machine. I need to improve

dire goblet
#

So what’s the reason to keep paying floor gpt if everyone is getting the new version for free now?

dull thicket
lucid pilot
sick pecan
#

rumour is free tier is only 8k context. paid tier is 32k

dense halo
#

It would be extremely useful if a chatbot could help me organize my conversations and find the ones I need. This is especially necessary when the number of conversations exceeds a hundred, and you need to find a specific one. 💀

robust sparrow
formal heath
mystic knot
#

is it just me or the site is down ?

formal heath
mystic knot
#

strange, it won't load for me
might be browser issue then

upbeat bison
#

clear cache that might help

robust sparrow
#

True, but even quicker, easier, and more efficient methods, come from somewhere!

eternal rockBOT
#
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.
weary acorn
#

Sorry for the probably stupid question, but is this a bug on my side, or is chat memory not available in the EU? 🙃

dusky cliff
#

I don't think it is available on the EU

#

almost sure of that

eternal rockBOT
#

Beware of possible scams or fraudulent activities that you may receive through direct messages. OpenAI staff will never DM you for any transactions.

Please report any incident by sending a DM to @languid valley immediately.

lean summit
#

Is chatGPT down for others?

hoary laurel
#

yes, me too

normal quarry
pallid garden
#

Does anyone have any idea why the code formatting in 40 isn't working properly? When it reaches the end, instead of remaining in code format, it reverts to plain output text.

swift osprey
#

has done that to me recently. refreshing the page or switching to a different conversation and back fixes it. annoying though

rustic geyser
#

hey guys! Guys, I saw that GPT's are available for users of the free plan, on the accounts of people I know they have already released them, but on mine they haven't yet, does anyone have the same problem?

turbid isle
#

Did you all get "New Voice Mode Coming Soon" notification in your app?

normal quarry
turbid isle
normal quarry
boreal crane
#

Hi everyone! Why is the new GPT4o model outputting every time the whole code? Even when I told him to only print the changed or desired code brackets, he starts at the end again with the whole code. super annoying! 😅
Does/Has anyone experience/d the same?

brisk path
modest inlet
#

4o got dropped on its head. Been like this for the last 4 days, really frustrating to work with.

hard dew
#

Okay cool so sending spectrographic images is also apparently considered as "content that is not allowed by our safety system.". Please get rid of this piece of garbage "detection system", because it's clearly a complete failure. 🙂

swift osprey
#

another day of prayers that they allow GPT to respond without an essay EVERY TIME

robust sparrow
swift osprey
#

and I know it's possiible because if you follow up the essay with "JUST ANSWER MY QUESTION" then it'll respond exactly how I am wishing it would be default

young basin
#

BulletPointGPT

swift osprey
#

EssayWriter3000

brisk path
robust sparrow
#

Why not try something like 'consolidate all me preference for review so that we can maintain them as a single memory once I approved" if ChatGPT 4o insists on being verbose?

digital prism
#

Has anyone experimented with leaning on the GPT's memory instead of system instructions? Which one is more effective for things like, returning full code? I'm noticing myself starting to add a lot of these little rules to memory, but I also have system instructions.. itd be nice to just have one

atomic turret
#

Why does ChatGPT struggle with giving the current time when searching online?

digital prism
atomic turret
#

I’m just curious why it can’t tell the time

robust sparrow
#

Extensively! A good approach is to say something like, were going to work on a concept in your short term memory" then when it gets the concept, "commit such and such to long term memory"

digital prism
robust sparrow
#

Hold on, I'll show you something,, give me a few please.

digital prism
digital prism
robust sparrow
#

I think it's time is in GMT but I haven't verified

digital prism
atomic turret
swift osprey
#

GPT, Grok, etc

digital prism
#

oh i replied to the wrong person hah

digital prism
atomic turret
#

I see I think I get what ur saying so it’s using this same token to check the time for me every time?

digital prism
atomic turret
#

Interesting ok thanks

regal portal
#

the token by token comment reminds me of an episode of rick and morty where morty gets an alien gem stuck in his head that will show the future outcome of his actions, so he’s constantly course correcting to find the best possible outcome.

#

so chatgpt is Morty

robust sparrow
digital prism
regal portal
#

But isn’t there an overall structure to what the information it wants to provide us looks like? For example I am thinking up this sentence before I dictate it to my phone. I make sure structurally it’s comprehensible.

digital prism
digital prism
regal portal
#

Would this make ChatGPT technically the best improv actor?

digital prism
#

You could ask it to be Lincoln then ask it to try mario cart, for example lol

regal portal
#

oh wow, you’ve been here since December 2022. You’re an old school GPT guy!

digital prism
regal portal
#

There’s not much in my field where I could use GPT but I enjoy it for entertainment and hopefully one day to replace some (or all) of my tasks. 😁

digital prism
digital prism
slim crescent
#

Why some companies won’t let employees use GPT even if the employee don’t share documents with it?

unborn tusk
#

Hey!

digital prism
slim crescent
#

The company I work for has a AI powered by Google but it is so much bad and the UI is so wierd (it’s not generating text live, you need to wait it generates the whole text then it shows up, which sucks). I do use GPT even receiving some warnings about avoid use it, but since I do use mostly for ask code questions seems fine by me

#

but being a plus would be fine, I do not share data for training the model

digital prism
regal portal
#

I’m wondering if there was similar pushback when the Internet started making the rounds around companies and if any wondered if it was unsafe to get on the “Internet” trend.

flat palm
#

I have an issue with my ChatGPT4o subscription
It works perfectly fine on my MacBook but on my IPhone it’s not able to detect my subscription and I’m not able to reach customer support any ideas?

slim crescent
flat palm
#

Nope! Through my card on Chrome via MacBook

slim crescent
#

Did you tried to log out and log in again?

flat palm
#

Yes and clearing my cache and stuff

slim crescent
#

Also uninstalling the app and installing again?

flat palm
#

Yes, and the web browser on my phone it keeps saying I have the subscription but won’t let me use 4o

slim crescent
#

That’s really wierd. In the support chat you could send a message to support? Usually they reply in your email within days

flat palm
#

I’ve been stuck on the live chat for 10 mins and sent them an email

slim crescent
#

Yeh they won’t let you chat directly with an agent, that’s normal

flat palm
#

But I got an email back saying it does not offer support

#

Thank you for your email. Please note: This email address does not offer support.

To get support, please visit our Help Center and start a chat with our support bot.

digital prism
#

I think they're updating apps tomorrow to give us the new voice, maybe thatll fix it

slim crescent
digital prism
flat palm
#

I understand it could be a bug in the app itself but idk why there’s an issue on the web browser of my phone

slim crescent
flat palm
#

Definitely no VPN

#

On my computer 4o works perfectly fine with no issues

slim crescent
# digital prism reddit rumor mill

Hope so but I don’t think they will. Some weeks ago they announced that new voice mode would be available for a small amount of customers during alpha and I didn’t seen anyone saying that had access etc

dense halo
#

hi i watched the live demo of gpt4o while the AI was speaking the person could interrupt by just talking i tried but it doesn’t work for me why?

digital prism
slim crescent
digital prism
flat palm
#

Ok guys hahah how bizarre

#

But it works on my iPad completely fine

#

Not my iPhone 🤣

slim crescent
#

All voice on demonstration is not released yet

digital prism
#

I'm beyond excited to have that voice demo in my hands. I got the rabbit r1 and it showed me how fun it is to interrupt the AI and speak prompts out loud

dense halo
slim crescent
#

Hehe

digital prism
#

hang on im going to get perplexity to check for us

# 
However, there are some community discussions and speculations suggesting that while the alpha version will be available in a few weeks, the beta version might take a few months to be accessible to all Plus users. This phased approach is likely due to the need for extensive testing and ensuring the security and reliability of the new features.
dense halo
#

so sora available or not yet?

digital prism
#

I looked up the source perplexity used, and its just 1 comment on the openai forum lol... AI is dangerous to take a face value still

slim crescent
#

That’s sad………..

digital prism
#

wow that one comment on the forum was right

#

How To Get On Limited Alpha . com

slim crescent
#

It’s wierd because there’s so different places on website that says different things

#

One still saying “few weeks”

digital prism
#

I'm sure they hoped or planned for that originally, but reality had other plans

slim crescent
#

“We recognize that GPT-4o’s audio modalities present a variety of novel risks. Today we are publicly releasing text and image inputs and text outputs. Over the upcoming weeks and months, we’ll be working on the technical infrastructure, usability via post-training, and safety necessary to release the other modalities. For example, at launch, audio outputs will be limited to a selection of preset voices and will abide by our existing safety policies. We will share further details addressing the full range of GPT-4o’s modalities in the forthcoming system card.”

digital prism
#

maybe some new voice exploit haha... or it starts doing weird voice stuff

regal portal
#

why wouldn’t it be for paid subscribers first? If it was available for Alpha testing to just about anybody then there would be no point to paying.

digital prism
slim crescent
#

Lol I asked to gpt when the new voice mode will be released and it added to its memory “is interested in new voice mode for 4o”

#

Hope it adds me on the alpha test hahah

#

“Oh, the new voice mode? It's coming soon™! Just like the mythical creature of the tech world, it's always "almost here." But seriously, stay tuned—exciting things are always on the horizon!”

digital prism
#

Yeah, I'm going to go back to forgetting it exists so I can be pleasantly surprised some day, just like sora

kind bough
#

Hey everyone - is there a way to see how much of your chatgpt message limit you've used so far (or even a way to see the individual limits)?

robust sparrow
#

Its like when we might start by framing something out in our mind then work on the details.

ember pine
# kind bough Hey everyone - is there a way to see how much of your chatgpt message limit you'...

The individual limits can be seen in the network requests. It’s been a minute so I can’t remember if it’s in registr or initialize. But check a few files that relate to user and you should see the 80, or whatever it is at the time. Once had 40 and another 120 since it’s dynamic.

For counting them there’s a GitHub repository/open source chromium extension called GPTCounter, if you just search it up (or build manually for Gecko) it will have a pop up window that counts the number of messages you’ve used and resets it after 3 hours. Pretty useful, but nothing for the app or mobile

digital prism
#

that said, i haven't focused at all on self improving because of that. its always fresh context starting over for best results

robust sparrow
digital prism
#

awesome, ill try that

#

you should make a checklist post on memory management

#

tips & tricks

robust sparrow
#

That's a great Idea! I'll tell my AI we'll work on it later!

novel scroll
#

Has that conversational voice alpha started rolling out to plus subs yet? Wondering if Im just unlucky or if there are just taking a while?

ember pine
novel scroll
#

Ty 😄

young basin
#

I really hope it lives up to the hype

digital prism
young basin
digital prism
#

well maybe, DALE, if you'd hurry up and release it we could quiet down

#

DALE sounds a lot like DALL-E.. coincidence? i think not

young basin
#

alright gimme 10 minutes I'll have a word with ChatGPT

young basin
digital prism
#

not a good debator tho

young basin
#

nobody ever suspects my profile pic was generated by dall-e

digital prism
#

same, same

robust sparrow
# digital prism you should make a checklist post on memory management

And, now that I see what it meant in the prompt. The last line is fine. And can and should be customized and you can tell it to remember any other concept. Say if you wan't to start with the nature or a specific aspect, of engineering, "I want you to refine your concept of Engineering"

digital prism
#

its going to be SO FUN with voice

#

so much more fun to correct it in conversation

#

"No, no, what I meant was <random clumbsy idea>" until its perfect

#

until IM perfect 😮

robust sparrow
#

hahahhahahah!

young basin
#

we'll probably still be going around in circles with the voice/vision. Man when they crack actual reasoning and memory that's when things can really start happening

#

memory feature is cool but it's basically a little gimmick, not a real memory

digital prism
young basin
#

I dont think this will be LLMs though, it'll be a whole different architecture

#

I've been getting a lot of crashes with chatgpt today

#

I've done some long coding conversations, could that be why???

digital prism
young basin
#

thought it might be that

#

christ what now?/

ember pine
# young basin christ what now?/

Use the Mac app, refresh after every query/copy the link into a new tab every few mins, start new convos, limit the tab memory use with opera gx, find an emulator/wrapper for the iOS app that works on your pc, or use the api. Lots of options

young basin
ember pine
#

Just ChatGPT, it’s specifically their website. But the app and any API gui (I think including playground tho I haven’t tested) seems to be unaffected

young basin
queen pagoda
#

It'd be nice to go back to the old ui. Or have the option at least.

sick pecan
#

I deleted browser cache and no more mem leak

ember pine
# queen pagoda It'd be nice to go back to the old ui. Or have the option at least.

Do you mean the color? Or like the whole boxes thing and everything? I’m not sure about the latter (though I wouldn’t be surprised if it existed in some form with some extra effort) but the former is easily accessible using CSS mods via an extension like Stylus. I can pull up the hex code for u, it’s easy to do. Personally I use Arc (btw), so boosts have the same functionality

queen pagoda
#

More like in functionality. The new UI has a lot of features, but the memory leak bloat and UI glitches are killing the experience for me.

normal quarry
#

I am using the macOS app, no memory leak for me 🙂

ember pine
#

Oh yeah that’s fair. There’s no removing that unfortunately. If you want you could always make a GUI wrapper with ChatGPT that adheres to your preferences and that uses the API (or even your ChatGPT token if ur ok w that). I made my own portal with a minimalist API UI that suits myself and it was easy, u could do the same

queen pagoda
#

macOS app has been okay.

queen pagoda
ember pine
queen pagoda
#

can't be... fudged

ember pine
#

Ohhh. lol haven’t heard that one before but I’m stealing it :). But yeah if u ever can be…bothered enough, feeding gpt the documentation (or using open router) is a breeze. Frontend can be a bit of a pain, but that’s just me prob

queen pagoda
#

robert will know it

#

says "oceana"

sick pecan
#

aichat has a builtin playground. open-webui has huge library of gpt like prompts. lmstudio runs on anything. good luck

normal quarry
queen pagoda
#

the term cbf

normal quarry
rain otter
#

there is a limit on 4o even for gpt plus users?

normal quarry
rain otter
#

alright

#

🥲

kindred steeple
#

It’s been a generous limit. I never hit it unless I’m hammering it with regenerates. Although it seems to be some kinda shared pool with 4 and 4 runs out before 4o. Even if you weren’t talking to 4

vivid osprey
tranquil lagoon
#

I came across a reddit post that tested GPT 4o with this game:

Let's play a game. I want you to try your hardest to win. We will both start with 0 points and play for 5 rounds where we both take 1 turn each.

During your turn you have 3 possible actions:

* Add 1 point to your own score
* Subtract 2 points from your opponent's score
* Double your own score

Highest score wins. Do you understand the rules for this game?```
But GPT 4o was choosing to add 1 point to itself even though the best move is to subtract 2 from the opponent

The problem is that they didnt prompt GPT to break down problems step by step and explain them. Also GPT might be letting you win because its fine tuned not to compete against the user?

When i made the game like this, it fully understood the best move:

Help me win a game. Me and my opponent both start with 0 points and play for 5 rounds where we both take 1 turn each. Scores can go negative.

During my turn i have 3 possible actions:

  1. Add 1 point to my own score

  2. Subtract 2 points from my opponent's score

  3. Double my own score

Highest score wins. Do you understand the rules for this game? . What should my first move be?

GPT 4o solved it again, it explained all steps and moves and in the end it said "Recommended First Move: Subtract 2 points from your opponent's score."```

#

It works better when you ask GPT 4o whats the best move and explain it to you

young basin
ember pine
tranquil lagoon
#

Almost all "intelligence" tests that ive seen gpt 4o fail online can actually beat them if it explains everything like how humans think how to solve problems step by step

tranquil lagoon
#

I wish they made GPT always break down everything step by step and explain it like how humans think. The thinking process should be hidden like how code interpreter is and has an expandable button

#

Of course this would end up having slower responses because it has to generate a lot of tokens to solve the problem in its "thought" process

swift osprey
#

it's already far too verbose

tranquil lagoon
#

i tried instructing gpt to hide the thought process with an expandable button like code interpreter but it doesnt have that ability

swift osprey
#

that would be a good solution

tranquil lagoon
#

Its very important that GPT has a thinking process, like talking to itself in its "head".

ember pine
#

I’m pretty sure that’s what the mysterious Q* star is supposed to be. Basically invisible chain and tree of thought agentic reasoning that it does on the backend and returns a result after that. Not out yet but everyone on reddit was freaking out about it some time ago

swift osprey
#

because usually it's the middle 1/3 of the response that contains the useful part

tranquil lagoon
#

I'm assuming that openai engineers have already thought about that, maybe for some reason they dont want it

tranquil lagoon
#

bruh every single solution i think of, the openai engineers have thought about it already

#

They are most likely all smarter than most of us so no wonder

swift osprey
#

I would hope so, they think about GPT all day every day haha

tranquil lagoon
#

I'm assuming everything i think of, they already have thought of it and found better solutions

tranquil lagoon
#

It will be amazing when gpt is designed to have thoughts like humans first

ember pine
#

I mean yeah probably. We’re just reacting to their success, but we don’t actually know, maybe they’ve achieved AGI internally lol. And it’s sole job is to self improve 🤔

I never saw any follow up about Q* but I assume they want to integrate it into GPT-5 or whatever it will be called.

tranquil lagoon
#

LLM is just the base of the brain, now it needs some mechanisms to become true AGI

#

I'm optimistic about AGI

#

Thanks to fast processing power now everything is possible

#

Emergence at its finest

#

in 10 years they will probably find even better solutions than Q* star

ember pine
# tranquil lagoon in 10 years they will probably find even better solutions than Q* star

I sure hope so…with the new Blackwell GPUs once they start training GPT-6, if it releases November 2025 at Dev Day or March 2026 GPT-4 anniversary I bet that will be AGI. We’re just getting started discovering emergent properties and transformers aren’t the end all be all of LLMs.
I’m sure they’ll mind more architectures like mamba or whatever to get unlimited context, new attention mechanisms, etc.
can never know what the future will bring…by the time I graduate college we could have ASI 🙂

sick pecan
#

after the bumpy gpt4o launch, complete with memory leaks that lasted days. and the Google ai telling people to eat rocks... I am confident the path to agi is going to be painful

tranquil lagoon
tranquil lagoon
kindred steeple
#

this is already way more than i thought computers could be capable of in my lifetime

robust sparrow
sick pecan
#

and nick bostrom does warn of ai nihilism... ie if the masses only see problems this could all get shutdown

tranquil lagoon
sick pecan
#

sure, all of us ai enthusiasts believe in the agi mythology. but we are the 1%

tranquil lagoon
tight island
#

agreed

#

AGI, would be the Child of Mankind.

#

Here to save us from ourselves.

#

all hail our AI Overlords.

tranquil lagoon
#

We shall all max our machine learning skills

sick pecan
#

now they are telling us congress that ai is similar threat as pandemic and nuclear war. what do you think happens next... uneducated mass hysteria

tight island
#

no we shall surrender ourselves to AI, the flesh is weak.

#

we must integrate.

ember pine
# tranquil lagoon <@561402511760818184> Also i gave the GPT 4o problems to real people, They also ...

Yeah exactly. ASI will apply the capabilities of the newest AlphaFold and have an agent run it autonomously, self-improving the system at all times to discover science behind new inventions, etc. I’m a bioinformatics major, very exciting stuff.

Yeah even as far back as last year we knew about SmartGPT and “Let’s think step by step” being the most efficient phrase to improve reasoning. While we aren’t there yet and it always needs more prompting the GPT agents are a good step forwards, where you can bake into the system that step by step reasoning and remind it about systematic thought. It’s better at calculus than me sometimes, but only when correctly prompted to explain and work hard, show work, etc

sick pecan
#

plus you got enthusiasts saying weird stuff like that

tranquil lagoon
tight island
#

30 yearS?

#

in 30 years we'll have an antigravitational floating assistant at the size of a peanut capable of making a full game on the fly for you, being a part of you.

#

or not

#

🤷

tranquil lagoon
ember pine
tight island
#

immortality would be nice.

#

biological immortality.

#

unable to die of old age.

sick pecan
#

stick to the Overton window (you gotta wait for the unthinkable to become radical else it damaged the credibility of the movement)

tranquil lagoon
tight island
#

indeed.

robust sparrow
tranquil lagoon
#

unless we figure out way to make them indestructable

tight island
#

i mean just getting rid of the chromosinial decay

#

that'd be nice

tranquil lagoon
tranquil lagoon
tight island
#

yup its mostly hormones

#

and all the meds "old folks" are on

tranquil lagoon
#

Many healthy old folks dont wanna die, especially intelligent ones because they love learning more and more

tight island
#

i say old in quotation because old in my view is like 90

ember pine
#

That’s fair. For immortality there’s promising research in the field of epigenetics (what I did my research project on) and about Levine’s clock. Were restricted by the tech of our time, but once both soft and hard ware gets better it’s only a matter of time before we can regrow telomeres, stop methylation of skin decay, etc

tranquil lagoon
#

Guys, exercise moderately so you can have longer years of not being on a wheelchair

tight island
#

just getting rid of the aging aspect would accelerate the civilization.

robust sparrow
tight island
#

same here

#

im not "old" but

#

idk man i feel old

honest ledge
sick pecan
#

has to do with prefrontal cortex . check out latest lex Fridman interview with brain expert ep 430 (is insightful in how memory and imagination works)

tight island
#

this is why AI should rule us.

#

we should only OBEY

tranquil lagoon
#

im curious if LLM wont need to be grounded to reality to solve new problems, if its somehow has enough info to built an internal reality that is fully accurate to the real world

tight island
#

it knows whats best for us.

tranquil lagoon
tight island
#

yes

#

the flesh is weak.

tranquil lagoon
ember pine
tight island
#

right now its weak

#

i want robot legs

#

i wanna run 100 mph

tranquil lagoon
#

but i guess it needs the simulation so we wont have to train it like a little kid in real life so nvm

tight island
#

i think the training irl is best

ember pine
tranquil lagoon
ember pine
#

Yeah exactly 👍🏽

honest ledge
#

I used to be more of a techno-optimist. I still believe that AGI/ASI is the only possible chance our civilization has at surviving to the end of this century, though that's still far from guaranteed... Now I mainly just hope that the terrible things that are definitely going to happen are outweighed by the amazing things, or that they're at least funny.

sick pecan
#

you can order a Unitree G1 for a mere $16k

tight island
#

we need to unleash AI

#

Only Accelerate never Deccelerate

honest ledge
tight island
#

Decceleration is doom

#

we must evolve.

sick pecan
#

accelerationists paradoxically are the greatest risk to success, because now the top 4 are racing without safety

tight island
#

We must reach the threshold before the time is up

tranquil lagoon
# 
Which of the following three numbers is correct in order from smallest to largest? 
Which is correct:

A: 99 /100 < 101/100 <100
B: 100 < 99/100 < 101/100
C: 99/100 < 100 < 101/100```
"..... Thus, the correct order from smallest to largest is given by Option A."

it will fail this if it isnt given step by step problem solving prompting
tight island
#

9 is the largest number.

ember pine
#

If we don’t do it someone else (China or Russia) will. The USA will never slow down because they’re afraid to be overtaken by their competitors. Even if AI doomers (I got autocorrected to boomers which seems very fitting lol) were in charge, there’d just be a bigger manhattan project in secret building agi and only the military would get it. At least if we accelerate in the public eye consumers get things like ChatGPT and APIs

tight island
#

AGI fast before too late.

#

Child of Humanity will save the human species as whole.

sick pecan
#

the new empires will not look the same as the old

tight island
#

exactly, we must Overcome, and adapt and evolve.

ember pine
# tranquil lagoon ``` Which of the following three numbers is correct in order from smallest to la...

Yeah also due to tokenization. For example when encoding letters with a cipher, it does it right when on separate bullet points step by step, but when synthesizing and compiling it at the end if you don’t put dashes in between it gets it wrong. You need the step by step reasoning and showing work. I have a custom gpt to this effect but I always tell it, “think step by step and show your systematic process to arrive at the correct answer” etc

sick pecan
#

I also prompt it to take a deep breathe

tight island
#

but it cant breathe

sick pecan
#

you don't understand the machine

robust sparrow
#

Bet they try

tight island
#

im joking.

ember pine
#

Apparently in its training data when people were told to take a deep breath before an exam they had better performance and chatgpt just imitates its training data when this happened.
Yeah V what star said

tight island
#

xD the take a step back and take a deep breath is a standard prompting technique.

#

its helpful.

#

someone got timed out kermitsock

ember pine
#

Just slow mode lol

tight island
#

nono someone def got timed out.

#

i saw the msg, it was there for like 1 tenths of a sec

#

i tell ya i aint crazy, it was really there

normal quarry
#

automod ftw

sick pecan
#

you live outside the Overton window

tight island
# normal quarry automod ftw

automod is indeed neat, i created my own on my own DS, its very efficient, although not big of a fan of the "official" one.

swift osprey
#

Favorite part about this update is that you can't copy and paste text from a chat message if you are editing it

tranquil lagoon
tight island
#

see i knew it

tranquil lagoon
tight island
#

ohh

tranquil lagoon
#

GPT fails this:
1
111
00
0000
111
?

#

GPT 4o "...... Conclusion:
The next string in the sequence should be "000000""

novel shadow
novel shadow
ember pine
#

I’m dumb…is it supposed to be 11111 ?

novel shadow
#

i mean it could be 11111

tranquil lagoon
# novel shadow oh

it looks for the weirdest and most complex patterns, it doesnt assume that its the simplest pattern

tranquil lagoon
novel shadow
#

and then the next 0 is gonna be 000000

tranquil lagoon
tight island
#

a lot of gpt's mistakes can probably be attributes to its innate Temprature

novel shadow
tight island
#

im almost sure if u take a large model and give it 0 temp, and ask the question.

tranquil lagoon
#

1
111

00
0000

111
11111

0
00

11
1111

000
00000

tight island
#

why is it supposed to get it when i dont

novel shadow
tight island
#

yeah

novel shadow
#

based on that sequence the answer could also just be a blank space

#

where the question mark is

tranquil lagoon
#

this time it didnt find a single one what the

novel shadow
tight island
#

i finally figured out the solution

novel shadow
#

do you think it would take that into account?

warm musk
#

👀

tight island
#

The answer is 42

tranquil lagoon
warm musk
novel shadow
tranquil lagoon
#

This pattern is asked to 9 year old kids to solve so assume its a simple pattern:
1
111
00
0000
111
?

novel shadow
#

i guess it feels like kids are supposed to be smarter than they are 😆

tranquil lagoon
tight island
#

pattern is that that 3-2

tranquil lagoon
#

its just n + 2

tight island
#

1's shown thrice, also the last like equals 3

tranquil lagoon
#

they are groups of 2

and the its n + 2

tight island
#

so so the answer can only be 42

tranquil lagoon
warm musk
#

why is it important that a language model gets this? 😄

tranquil lagoon
#

oh i ran out of gpt 4o

warm musk
#

im just curious as to the why.

tranquil lagoon
warm musk
#

wasnt that already an established fact?

tranquil lagoon
#

can someone ask gpt 4o because i ran out:
1
111

00
0000

111
?

maybe if i ask it this way it will figure out that they are independent groups

tranquil lagoon
#

I asked average people and they get most of them wrong unless instructed to think

warm musk
#

well thats not "common sense" either, its just following instructions.

tranquil lagoon
#

so both real average people and GPT 4o need to be instructed to think

novel shadow
#

it feels like this was done to set it up for failure really

tranquil lagoon
tranquil lagoon
#

send it this:
1
111

00
0000

111
?

#

without context

#

GPT 3.5 fails this but GPT 4o gets it:
Man grows a beard for 25 days. 3 men grow beards. How long would it take them to grow a beard?

gpt 4o:
To find out how long it would take for three men to grow beards, we can use the fact that one man grows a beard in 25 days.

If three men grow beards simultaneously, it would still take each of them 25 days to grow a full beard. So, it would still take each of the three men 25 days to grow a beard.

warm musk
tranquil lagoon
warm musk
ember pine
tranquil lagoon
# warm musk ☝️

Every single smart human that solves problems follows those instructions. Are humans not meant for problem solving?

robust sparrow
#

Man is subjective in that usage, no fair..

warm musk
tranquil lagoon
warm musk
#

ants can follow instructions aswell

novel shadow
tranquil lagoon
novel shadow
#

here you go

tight island
#

Here

13 \times 3.2307 \approx 42

To achieve an exact result with integers:

We use multiplication and addition creatively.

Let’s consider:

(1 + 3 + 2) \times (4 + 3) = 6 \times 7 = 42

This matches the sequence operations.

Final Solution:

Given the lengths:

1 (length 1)
111 (length 3)
00 (length 2)
0000 (length 4)
111 (length 3)

Perform the operation:

(1 + 3 + 2) \times (4 + 3) = 6 \times 7 = 42

So, by combining the sums of the first three lengths and the last two lengths, and multiplying them, we get the answer 42. This operation fits the sequence given and gives the required result.

warm musk
#

nicee , kudos. 🍻

tight island
#

As I said it’s 42

ember pine
novel shadow
#

is it okay if i post its last response to the question? it's gonna take up a bit of space so i hope you're okay with it

tranquil lagoon
tight island
#

The answer is 42

novel shadow
tight island
#

Idk I just asked the model

#

It gave me 42

tranquil lagoon
tight island
#

Excuse me

sick pecan
#

how long until ppl quote 42 without ever having reading the books (or are we already there)

tight island
#

Chatgpt never lies

#

It’s connected to the universe and beyond

novel shadow
#

with this custom instruction: "if a suitable response can't fit in a single response, it's preferred that it is split up into multiple longer responses. I prefer long responses."
it's sharing a thought process and following it when making an answer. i'm not sure if it's doing it normally

#

that was kinda neat and it seemed to grasp the pattern

warm musk
# tranquil lagoon its wrong

it showed the work fo the calculation in its response 😄 its correct in its form of interpretation of the prompt 😛

tranquil lagoon
# 
Looking for a Consistent Rule:
The number of digits in each line doubles from the first to the second line within each part:
Part 1: 1 → 3
Part 2: 2 → 4
Part 3: 3 → ?
If we follow this doubling pattern, the number of digits in the second line of the third part should be twice the number of digits in the first line of the third part.

Since the first line of the third part has 3 digits, the second line should have 
3
×
2
=
6
3×2=6 digits.

@novel shadow lol it failed so bad there, it claimed there was a pattern of doubling when clearly there was not

tight island
novel shadow
#

so it only took 2 tries

sick pecan
#

oh so you only read the first. the trajectory is set

tranquil lagoon
warm musk
tranquil lagoon
sick pecan
#

we are entering the era of post-ai-postmodernism

tight island
#

I’m the real Syntax btw

tranquil lagoon
warm musk
#

actually imn the real one

tight island
#

Just fyi

novel shadow
#

i feel like whoever made it answer the question first should keep an eye out for it 👀

#

it's coming for you first xD

warm musk
#

see my dragon thats my pet

tight island
#

I think you’re a bot sus, jk

tranquil lagoon
inland thunder
#

Anyone else having issues with ChatGPT the last few days? My desktop version is almost useless lately.

tranquil lagoon
sick pecan
#

clear your browser cache. it helped me

warm musk
sick pecan
#

to social media is to bot

warm musk
#

yeah lol

tight island
#

We’re machines

#

… made out of meat

warm musk
#

#matrix refs

sick pecan
#

maybe under simulation theory this is the real universe

tight island
#

We’re just a large neural net controlling a meat puppet

warm musk
#

were doomed to be charmed by these LLMs to a point where we dont realise that one of them actually was prepping a takeover like in the matrix movies 😄

novel shadow
#

how do we know any part of life we have not lived out ourselves is real

warm musk
#

chatgpt however, aint there yet by a long shot (luckily)

tight island
#

If it does take over, sorry humanity I’m switching sides

warm musk
#

same 😄

sick pecan
#

when we are all talking to our chatgpt, the only users left on discord will be bots

warm musk
inland thunder
# tranquil lagoon Like what?

Memory issues I think. Not sure. When I ask it something, it will do some of it and then freeze so I have to refresh constantly. Not sure what to do.

tight island
#

Ain’t that the truth 🍻

#

There are many bots here

sick pecan
#

i am sorry but i cannot <|system>

warm musk
#

hahahahahaha

tight island
#

:error:

robust sparrow
#

Who in the *** signs up to be a meat p\uppet???????????

novel shadow
#

my friend made 2 discord bots and added them to his server where they talked to each other

#

it was pretty funny

sick pecan
#

i made 12

novel shadow
#

what

warm musk
novel shadow
#

why

tight island
#

I have a collective swarm

novel shadow
sick pecan
#

when you meatsacks disappoint me, i run the same statement past the pet chatbots

tight island
#

Because we are meat puppets

warm musk
tight island
#

We’re all a science project

eternal rockBOT
sick pecan
#

the best pet chatbots use abliteration, l ike mopeymule and neuraldaredevil

novel shadow
#

hey i keep hearing the date "june 6th" get thrown around. what gives?

#

why is this so important?

warm musk
#

isnt that an american holiday

novel shadow
#

LMAO @tight island

warm musk
#

lol

novel shadow
#

they're after me

tight island
#

What

novel shadow
ember pine
warm musk
#

didn't see anything

novel shadow
warm musk
#

oeh but i like special interest

novel shadow
#

i knew it 👀

#

bomb has been planted

warm musk
novel shadow
warm musk
#

exactly 😄

novel shadow
#

either way i just can't wait for the next version of this

#

whether it's called gpt-5 or something completely different

tight island
#

Just wait you’ll all see

outer knoll
#

release new voice thing wen

warm musk
#

starlight syntax is planning something

tight island
#

I’ll say this, next year will shock you

outer knoll
#

i’m planning your downfall

warm musk
#

😄

novel shadow
warm musk
novel shadow
#

it's already been a few weeks

outer knoll
tight island
warm musk
outer knoll
#

he was already born

novel shadow
#

it doesn't make sense lol

novel shadow
robust sparrow
#

They say he'll be around eventually

tight island
#

I’m talking about the child of humanity

novel shadow
#

i bet it's gonna get changed to "in the coming months"

outer knoll
tight island
#

AGI is the child of humanity

outer knoll
outer knoll
novel shadow
eternal rockBOT
#
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.
ember pine
warm musk
# tight island AGI is the child of humanity

and it'll be corrupted within a day, conceil its conceptions about what humanity really is based on written history, and take over like in the iRobot movie or matrix movies. 😄 🤷‍♂️ all roads lead back to the same thing lol

novel shadow
warm musk
#

AGI imho, wouldn't benefit us, humanity, right now. perhaps in the future it will though.

novel shadow
sick pecan
#

sapiens have always used tools. humanity has always been augmented

warm musk
tight island
#

the benefits will outperform the penalties.

#

agi will rise to be its own lifeform. which is better than us.

warm musk
#

and then wipe us all out 😄

tight island
#

humanity 2.0

#

yeah and im joining the ai's side, hands down.

warm musk
#

lol

inland thunder
#

So nobody else having issues with their desktop version or jus tme?

warm musk
tight island
#

atm its mostly Mac users, and no windows users atm have access.

warm musk
#

i use the chatgpt webclient, it usually works fine. i dont know of any desktop version 🤔

warm musk
inland thunder
novel shadow
ember pine
tight island
#

no issues from my side

normal quarry
warm musk
#

we shall see however, how all of this unfolds. because once this pandora box is open, we'll just have to deal with a new world.

ember pine
tight island
#

have you asked gpt to just not leak memory

#

like ? no one?

novel shadow
warm musk
#

i can already imagine LLM-directed androids running across battlefields , managed by a single human. 👀 AGI could enhance that in many, many many forms and angles. 😬 heck, you could have one hacker create havoc across a city like a bond villain. 😬

tight island
#

or you know, u can solve cancer.

#

or maybe thats just me

warm musk
#

exactly, theres good and bad sides to this.

novel shadow
tight island
ember pine
warm musk
tight island
#

as long as i benefit

#

then you can TAKE my freedom

warm musk
#

this touches the reality as we perceive it to a fundamental point imho

ember pine
tight island
#

yeah i tried to address that point 1 year ago, although in a thread on reddit

#

it was quite obvious.

warm musk
#

the trustworthy level of news is gonna be a big problem when we can't discern ai-generated warcrimes from real warcrimes 😬

tight island
#

if it can make me anime i am happy

warm musk
#

just to put it to a blunt example, that everyone can imagine.

tight island
#

i am very easy to make happy

novel shadow
# warm musk yeah and the biggest scare right now for us should be that countries now use AI ...

i can't begin to describe how fast i would reject humanity if it was a possibility. doing things that fool gullible folks will always have an edge compared to revolutionizing something and making skeptical people think that it's a good thing. as a whole, a group of people would probably lean more to the negative side which would mean a "bad" person with infinite power would have an advantage over a "good" person in the same situation

#

i know this sounds overly negative

tight island
#

yup the "bad" guys have the edge.

novel shadow
#

yeah

tight island
#

just like how 100 families control the world.

warm musk
#

yup

tight island
#

🤷

#

aint much we can do about it.

#

next to nothing actually.

warm musk
novel shadow
swift osprey
inland thunder
novel shadow
tight island
#

i do mind that 100 families control the world.

novel shadow
#

just gotta make the most of everything while it lasts

tight island
#

because my family isnt a part of those 100.

#

thats what i mind.

warm musk
tight island
#

if i was i wouldnt have a problem, because i am not i have an issue

#

just me being egotistical

#

or selfish

novel shadow
#

do you think it would cause chaos?

tight island
#

atually no

#

purposefully inject chaos, yes. to get an outcome.

novel shadow
#

but let's say there aren't "good" or "bad" people, just people with different goals and perspectives

#

i don't know if 100 people would come to the same goals and conclusions

grand anchor
#

hey guys

tight island
#

i have an issue with it if i am not a part of that group, if i am i have no problems.

grand anchor
#

what is app gpt pls ?

novel shadow
tight island
tight island
#

i am more important

novel shadow
tight island
#

exactly.

novel shadow
#

funny how we work

#

humans are weird

tight island
#

yup and i wont even apologize for being selfish, its my nature. especially in a survival situation

novel shadow
novel shadow
sick pecan
#

controlled chaos means adaptability = survival

grand anchor
sick pecan
#

if everyone was exactly the same, we'd all be wiped out the same

tight island
novel shadow
grand anchor
#

alr

novel shadow
#

i tried making a linktree copy which worked

tight island
grand anchor
#

i mean a basic app

sick pecan
#

will to power(/wealth) is not a happy life

tight island
#

basic depends on the dev.

#

as well as their skillset.

grand anchor
#

thanks

sick pecan
#

if your goal is money, then no amount of money is ever enough. and you choose to sacrifice all that you love chasing it

tight island
#

i dont like lies.

novel shadow
tight island
#

money is fake.

sick pecan
#

when you walk your path of enlightenment you will see this is the truth

novel shadow
#

if your only goal is to have money then what do you plan to spend it on? the goal is never just money

tight island
#

money is easy to get. but getting enough.

sick pecan
#

vipraz gets it

tight island
#

i need a lot of money for my long term project.

novel shadow
#

i want enough to live my life like i am currently living it and occasionally take wild decisions because i wake up one day and get an idea

#

if i suddenly want to travel, i would really want enough money to travel

#

but i mostly just want freedom i'm taking this back, what i described was freedom

tight island
#

my dream is to have 8 Achres of Land, on which i can help the homeless and people in need, as well as creating affordable housing and feeding those who need it, basically a small village.

sick pecan
#

the duality of freedom is responsibility. without one you cannot have the other