chatgpt-discussions

[grave raptor]

with the voice in gpt, can we relisten to what gpt said? or we can only lisetn to it once?

[sly geode]

Frrr

[digital prism]

the trick is to stop watching a pot boil. as the kids say these days "touch grass" lmao

[prime swallow]

Currently, it's once, unless you do something to record it.

[sly geode]

You guess what I am already working but it's just that I am in literally really need of such things . .

[digital prism]

humanity literally really needs them, going as fast as they safely can im sure

[sly geode]

And I only paid for it because of this ..cuz I hoped it would arrive.

[digital prism]

in the meantime, you can do a lot with what you paid for

if you stop waiting for water to boil

[sly geode]

Fine I'll be quiet ofc people here gonna blame me for mis behaving because others have access to it .

[digital prism]

didnt mean to shut u down, just giving advice sorry

[sly geode]

Alright

[queen crow]

ChatGPT and bing search integration sucks! Bing is useless honestly!

[elfin birch]

What did it not do well for you?

[west ivy]

me with no vison, dalle or voice :/

i cant wait!!!!

[noble yoke]

Thanks!!

I love it!!

I Was almost about to buy new paid account but luckily saw this message

Just fir dalle

[spice star]

main issue is it doesn't usually go for the most relevant websites, so you end up getting a lot of SEO garbage or ironically sites generated by ChatGPT

also it can only perform one search per message, doesn't like to view the entire contents of the page before providing an answer, and can't seem to load a lot of webpages that are obviously the best source to find an answer in

[twilit ibex]

dang I'm unlucky

still no access to voice

[queen crow]

The results are horrible. With bing, it can find the webpage but somehow it says that it couldn’t find the information in webpage. Google integration would have been definitely been better.

[thorny parrot]

#openai-chatter message

[sleek swift]

I like the new multi edition which is better

is there a way to force it all the time?

[feral hinge]

Is there a plan that I can pay for that doesn’t have message caps?

[median mortar]

wym by "message caps"?

[feral hinge]

50 messages max per 3 hours usage on GPT 4 on paid $20USD plan

[median mortar]

oh, is there one?

[feral hinge]

Yeah just ran into it

[median mortar]

no there isn't one CURRENTLY it says, eventually they will most likely update that..

or they won't because maybe it would be too "powerful"

[feral hinge]

Okay, well I’d settle for 200 messages

Especially as the features expand

[median mortar]

or unlimited 🤷🏼‍♂️🤷🏼‍♂️

[feral hinge]

👍 agreed

[warm token]

If we talking about gpt4, there is a message cap as he said

[solar pilot]

I lost dall e 3 from chatgpt

is that normal? On mobile

[lucid pilot]

You could try using the API instead it would give you what you are wanting. Ask chat GPT how to make one in python.

[median mortar]

i was saying "there isn't one currently" to that there isn't a subscription currently.. if that makes sense

[median mortar]

wym

[warm token]

ohhh yeah okay

[signal wadi]

Is there any way to read pdfs with chatgpt?

the plugins are not good

[normal quarry]

the AskYourPDF plugin is quite good

[signal wadi]

yep seems to work a lot better than the others

[solid rock]

When is the image thing beeing released

[normal quarry]

do you mean Vision or DALL-E 3?

[solid rock]

The latest announcement

[normal quarry]

there are two image related releases. Vision, which allows ChatGPT to see things when you upload an image and DALL-E 3 which allows you to create images

[solid rock]

Oh, when are both releasing ?

[normal quarry]

Vision is rolling out now and will be available to Plus users by Tuesday unless in the UK/EU

DALL-E 3 is rolling out during October

[sly geode]

Today?

[solid rock]

Okay thanks !

[livid pier]

Some Plus users have received the push

Earlier people (including me) got the push by filling out the Google form here

Dall E 3 gave me a lot of surprises

[sly geode]

I am still hoping for vision

On plus nothing yet

[livid pier]

But it still has a lot of problems here in ChatGPT

For example, the seed of the same prompt is fixed

Easy to generate failure

The understanding of prompt words in other languages is not as good as that in English

So at least in that respect, it's good to wait

[sly geode]

Been waiting for 7 days since announcement

[normal quarry]

be patient, it will come

[sly geode]

¯_(ツ)_/¯

[livid pier]

I've been waiting for GPT-4V and voice conversation for two weeks

I haven't received it yet

[sly geode]

Do we have an option to talk to customer chat directly instead of mails

[livid pier]

At least not yet.

[normal quarry]

there is a chat option at https://help.openai.com

[void lake]

Same here, what's the point of having a sub if we don't get the product?

[livid pier]

They only replied to me with a bunch of useless nonsense

No matter what I ask

Even suggestions, bugs

Will only keep me waiting for a long time

And reply to me with an email full of crap

So I'm not going to send them any more emails

🙃

[livid pier]

But the two-week deadline is really not up yet

We can't blame them for anything

I can only envy others for having received it

😫

[void lake]

The way I see it, if a user is on the last cohort on getting a feature, he should be on the first cohort to get the next one.

[livid pier]

At least I've never seen that kind of commitment

[normal quarry]

it was a two week roll out, just be patient and it will come

[livid pier]

[sly geode]

[livid pier]

In fact, there are more sad

I shared my subscription sharing link with others

They got a free Plus subscription for three months

[sly geode]

How did they got free ?

[livid pier]

Open my subscription sharing link and bind an available card

[sly geode]

💀

[livid pier]

Although it is said to be a subscription and sharing link

We need to share

But this is three months!!!!

I spent money on a subscription for more than five months

Why don't I have such a good treatment

🙃

[ancient sage]

Might be just me but is anyone else having issues with chatgpt? Just want to confirm if I'm the only one, as none of my prompts are working. Its the first I've seen the 'You've reached our limit of messages per 24 hours. Please try again later.' warning, wondering if its bugged out like back on the 1st.

[oblique yacht]

Read the message

[ancient sage]

Given that I haven't submitted any prompts today, I find that hard to believe.

[normal quarry]

have you provided your login credentials to any third party apps or plugins or browser extensions?

[ancient sage]

Of course not.

[normal quarry]

do you use any third party ChatGPT related browser extensions? Have you shared your account details with any other person?

[ancient sage]

I'm also aware of where my API keys are in use, and they are not accessible by anyone besides me given they're stored in the credential store.

[ancient sage]

Nope.

[normal quarry]

then I suggest sending a support request to [e-mail]

[feral zinc]

are you using a shared network?

[ancient sage]

I have a shared network, but given that I control the firewall on my network, and my computer itself is on a separate VLAN, I'm inclined to believe the issue is server-side.

I also have a pfsense firewall, not the bog-standard built in from the ISP.

[feral zinc]

that error often happens when there are multiple accounts using chatgpt from same origin

like can happen with school, work or vpn/proxy services

[ancient sage]

If we assume the following:

• There's no password due to it being an OAuth Grant
• I am the only one logged in with my email on my Google Workspace on my Domain

I have to conclude the issue is most likely serverside.

[feral zinc]

there is ip/network level rate limiting above accounts by openai

but yeah support is your only option

[ancient sage]

I suppose that's possible, but I don't think anyone else in my house uses it besides me.

Joy. I hate these kind of mysteries.

[west ridge]

rolling out still going on? I still don't have voice and vision feature

[frigid nebula]

till tuesday my friend

same boat as you. voice i do have though

[formal heath]

I do not have that either. it is rolling out still

[frigid nebula]

#privilige

voice feels a lil glitchy

but for what it is.. fantastic

[west ridge]

yep, just can't wait to try

thanks guys

[normal quarry]

Voice is geat, I can wander around with my headphones on and ask ChatGPT questions, get answers, challenge it, expand on topics all hands free

[oblique yacht]

For the rollout are people getting added constantly, or at fixed times of day?

[karmic wagon]

For plus users they reset 50 messages limit every 3 hours but what is the time line they follow for reset

[frigid nebula]

1 batch of users every 1 seconds

timezone shifts every hour

no clue tbh

i lied

[karmic wagon]

For example at 6 i use 20 message then 7 i use 20 more how they calculate it for 50 messages

For 3 hours?

Because form 6 it should be 9 and for 7 it should be 10

[timber oyster]

Just don´t like it´s using the telephone mode from my device. If it´s connected to my soundbar or other media playing device it still plays from the phones audio. Also it like the normal (not speaker) mode. So you really have to crank the volume up or use it as an Phone call.

[timber oyster]

As I observed every message has it´s own timestamp.
So if you send a message at 6 you have only 49 left and get one at back. If you send another at 6:05 you will have 48 and it´s added one message back at 9:05.
But im am not 100% sure

[karmic wagon]

Ohh i guess they could be following reset at 12 by this i mean either they can set time straps
12-3,3-6, 6-9

[frigid nebula]

yep agree

with the telephone mode thing

[karmic wagon]

Or they can just start the session from the fresh message or first message of the day then 3 hours start then after that again reset

Each message having its own time stamp means multiple loops and it will make it complex for them and reduce the run time i mean time to give the result for next message

[timber oyster]

As said I am not 100% sure if it is so. But if it´s so, I am not sure if you know how it possible works, but most liklely saving it to a db. There it´s a matter of less than second to fetch this. The "run time" by checking if the user is allowed to send messages is 99.999% so small, it takes longer to generate the first tokens of you promt
Also where are loopholes with this?. You write 50 messages in 3 hours, you get them back in at least 3 hours. No waiting until some predifned timestamp. Also you method would allow to send 99 messages if you send one, wait 2 hours. Send 49 and then the resets kicks in.

Also 50 messages are far more than enough (if not using voice 🥺 ) even when using it to code.

[hasty laurel]

What are the voice and vision features

I’ve heard about them but don’t know what exactly they are/do

[azure moat]

Ok so I have the option to select Dall-E 3 in the New Features of the website. I don’t see an upload button though? Also, is Dall-E 3 not available in the iOS mobile app?

[feral zinc]

https://openai.com/blog/chatgpt-can-now-see-hear-and-speak

[feral zinc]

not yet

[golden delta]

Not sure about iOS but it shows on my Android app

[timber oyster]

Yeah, currently not supported on IOS. Checked on my android but it's not there on my iPad even after relogging

[west ridge]

voice feature is on the mobile app, you can talk to chatgpt. And the voice is incredibly real, i think this is the most advanced AI voice I'v ever heard. The vision feature, you can give chatgpt any image, and do whatever you want based on that.

[proud aurora]

Hi all,
does anyone else also experience huge performance loss if a chat with ChatGPT (gpt4) is very very long (occurred over a few weeks)?

[maiden fractal]

nope

[west ridge]

upload button belongs to vision feature i think. I also have dalle3 but no vision, no upload button

[hasty laurel]

Sure, I can talk to it using my voice but I don’t hear its voice back

[west ridge]

you mean whisper? Convert your voice to text?

Or you mean the network issue

[proud aurora]

Weird. Most of my chat is over desktop browsers but when I open the same chat on safari, it crashes and restarts.

[west ridge]

you can check the up-right corner of your app UI and see if it has a headphone if you'er talking about whisper

[maiden fractal]

clear cookie and cache

[proud aurora]

Still crashing. App is fine though. But not crashing right away. Only after clicking to burger menu icon. So it got a bit better. Thanks for the advice.

[eternal rock]

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.

[astral parcel]

Hello, how do I activate the speech for ChatGPT 4 ?

[brittle junco]

Where i can find my Dall-e in chatgpt Plus ? Because i don't find it

[astral parcel]

in the menu chatgpt4

[astral parcel]

at the top of the screen where you can choose between GPT 3.5 and GPT 4

[brittle junco]

I can't find it, i have Defaut, Browse with Bing, Data analysis and Plugins

[astral parcel]

check your parameters and enable all features

[brittle junco]

In "beta features" i have only browse, plugins and data analysis

[astral parcel]

strange, maybe because of your country

[west ridge]

may be they are still rolling out?

[brittle junco]

I'm in France, but all my friends french have the dall-e

[brittle junco]

Maybe

[honest jay]

Last time an OpenAI dev added a few more people to DALLE (beginning of this weekend), they mentioned that they're at about 1/3rd of Plus users with DALLE right now, so they're definitely still rolling it out.

[brittle junco]

Okey thank you ! Where i can find this message ?

[honest jay]

You may not have the feature yet, it'll still being rolled out as well. You'll see the option to enable it in the app in Beta Features. Then, you'll see a new icon on your new chat screen to start voice mode.

[astral parcel]

thx

[honest jay]

#openai-chatter message

You can also see in this dev's history that they've posted 2 links in the past where folks have been able to manually request to be added to DALLE, so keep an eye out for a potential additional one of those as well!

[brittle junco]

Thank you !! Where i can find de dev's history ?? and how do i do these request

[honest jay]

The forms are closed right now so you'll have to wait for the next one, but if you search in the OpenAI discord server for "from: moxi" it'll show you everything they've posted, if you go back just a bit you'll see the links. I just got lucky and was online right when moxi posted one of them!

[brittle junco]

there was a message this morning at 4 a.m. paris time, isn't this one open?

[timber oyster]

There was a form open yesterday. It should still be open:
See the reacted message (and then above). I cant see if it´s active, because I already answered.

had access 2 hours later (filled it around 6pm yesterday)

[honest jay]

The last link posted was this one, you'll have to wait for another message like this, if moxi even does it again! Otherwise, the rollout should get to you soon anyway

#openai-chatter message

Edit: oh yeah it's still open vito is right! Don't know if moxi is adding people from it still.

[carmine eagle]

When dalle-3 and vision I realesed on EU ? Some youtubers have acces to this stuff..

[timber oyster]

Dall-E 3 is rolling out in the EU too. Vision (so everyone says it will come soon), but If you ask me (no proof or something), but it will take a looonng while. Could imagine it has todo with data regulation

[carmine eagle]

Damn,too bad..because is some awesome features but some not will enjoy this

[brittle junco]

mmh i don't know, i hope ^^

[wet fable]

V3 is already running for NA users?

[neat elk]

How do I know if I have access to image gen?

[maiden fractal]

You will have in drop-down dalle

[oak cave]

Same here : France

[hidden pawn]

Anyone in Europe that has gotten access to GPT-Vision

[normal quarry]

Vision is not being rolled out in Europe for now.

[merry wagon]

Ok, I got Dall-E now. If i tell it to use the 'Super Nintendo look' for its pictures, it declines my request.

[maiden fractal]

It won't generate picture from protected franchise

[merry wagon]

I know, what i meant were pictures in a pixelart 16bit style. It would be nice if it would get what i mean

[normal quarry]

ask it for 16 bit pixel art, don't mention Nintendo

[maiden fractal]

Try something like inspired from video game genre of 16 bit, pixel art

[merry wagon]

Hmm, ok

[outer pebble]

When is the latest that I should have access to image input? (I have gpt plus)

[normal quarry]

by end of business Tuesday unless you're in EU/UK

[outer pebble]

ah damn I am in the EU

Is there any reason to why the eu/uk will not have access?

[normal quarry]

then it will be rolled out later

[normal quarry]

we don't know but it might be to do with GDPR compliance

[outer pebble]

Ok thanks

[hidden pawn]

Typical EU

[vagrant island]

The update rolls out today right?

[normal quarry]

the updates are rolling out over a two week period that ends on Tuesday

[fading cloud]

anybody tries perplexity?seems even better than chatgpt?

[normal quarry]

talk about other AI products in #ai-discussions

[daring bobcat]

yeah its great

[digital prism]

yes.

GPT is self aware.

[feral zinc]

its not

[digital prism]

It knows what it is

It can explain its inner workings in great detail. the problem is how we tend to define "aware" with mysticism

Counter point: GPT is just a collection of individual AIs working together. In that case, it's more the way a society is aware of itself.

[merry wagon]

Self-awareness is not just about performing an action; it's about understanding the implications of that action.

[daring bobcat]

why was my comment deleted?

[prime swallow]

I'm not a mod, but generally asking modmail is the way to get answers to stuff like that. They can explain any rules not understood or anything. modmail's top right, first on the user list. Send the message to discord moderators.

[prime swallow]

Hrm. We can be described as a bunch of cells too, all in a careful society together 🙂

[daring bobcat]

alright, thanks

[digital prism]

if thats true its a gloopy glob pattern, big blobs eating little ones, little ones breaking out from big ones.. all messy and gooey

[prime swallow]

Erm. That's not 'we' - at least my personal cells get along better than that.

Sounds like you're describing an ecosystem... I'm trying to talk about an organism, notably a typical human.

My big blobs generally don't eat my little blobs, and the blobs I have that do eat other blobs, they are outsiders (immune cell defenses and whatnot).

[digital prism]

Our cells are constantly eating eachother, some cause cancer, its chaos

[prime swallow]

If we have pan-Gaea stuff going on with ecosystems giving rise to minds, that's not widely accepted or recognized how to determine yet.

Yours, perhaps. Our cells do communicate, and some tell others to shut down, sure. We have police-like cells. But 'eating each other', phagocytosis, is not the normal way that human cells interact with other human cells. Even in cancer.

But we veer perilously off topic, so, back to ChatGPT! I enjoy trying to imagine how to catch emergent properties it might show, now or later.

[digital prism]

GPT really feels like a new field / realm / adventure of techno-psychology.. purely electronic thinking machines that behave based on us, but not like us idk

[prime swallow]

It's glorious. In some ways, I bet ChatGPT functions (In addition to other functions) like a human-facing Rorschach Test. What you think you see in the image, or the AI, may say more about the individual looking than what is being looked at 😄

[digital prism]

oh man, so, it loves emojis, all im going to say. those are literally rorshach tests I was saying this morning. I'd share my dalle gens, all consistently themed, prompted purely with emojis. But the automod bans for that 😆

emojis seem to give access to more emotional intelligence (early take)

[ionic geyser]

So are the rest of us getting vision/voice tomorrow , anyone know

(or maybe today)

[prime swallow]

If it's allowed content, you can share 'in the right places'. We have entire threads entirely devoted to sharing image generation, have you seen the many Dall-E galleries, like #characters or #sci-fi ? We have #🌄┃general-art and #weird-gens-and-misfires and all sorts of places where you can share the art appropriate to the thread, and within server rules (so nothing bloody or whatever)

[digital prism]

I just mean, if I try to show how my 3 pure emoji prompts stack up together, automod freaks out for too many emojis

[prime swallow]

Ahh. screenshot, not copy/paste, of the emojis, perhaps.

But in my experience, the AI cooperates with the user, and enthusiastically helps the user however the user indicates.

If you give it clues that you are delighted by dust, it's gonna lavish words describing dust for you, and 'share excitement' with you if you seem to be excited and delighted by the discussion.

Anything allowed content, just show it what you prefer, and it'll try to match your preference, it's awesome.

[pseudo jay]

Is anyone else experiencing 'error' responses, resulting in the need to request a 'second' response? It's been a minute since I've experienced these, but this morning I'm gotten 3 within a short period of time.

[digital prism]

yeah, it brings art to the masses in a very accessible way. honestly this should be the goal of art.
not just art, it brings intelligence to the masses. viva la revolution

[digital prism]

mind sharing the error? curious

[pseudo jay]

First one was a glitch (error) in the middle of its response. The second was an error (after a few back and forths) before it even began to respond, and the last one I got a error message, "Hmm. Something seems to have gone wrong", requiring me to request the second response.

[digital prism]

Something gone wrong, a regenerate almost always fixes for me unless there's server issue. Not sure the other ones

[pseudo jay]

As mentioned, it's been quite awhile since I've experienced these errors, esp. several in a row. I'll keep an eye on it and see if it continues. I was just curious if others were experiencing it as well, today.

[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.

[lyric prism]

I got chat GPT premium but don't have access to DALLE-3 in the chat. Is there a way to activate it?

[normal quarry]

you need to wait for it to roll out

[timber oyster]

I am on mobile and can't find the chat. But you can scroll up here in the chat. There should be a link to message containing the Google form you can fill out

[stark bronze]

(last pinned message actually)

[prime swallow]

And here's a more direct link, no idea if more testers are needed: #openai-chatter message

But it's rolling out to all.

[prime swallow]

Making sure you know how to check for it, Dall-E, in a new chat, hover over the GPT-4 selector, if you have access already it's going to be one of the choices that shows below as you hover.

[lyric prism]

thank you signed up. My friend has it on his account but I guess I have to wait :/ ty all

[coral helm]

does gpt4 images not like modify existing images..? like you cant upload an image and say try change this to this

[prime swallow]

Correct. There's two sorts of images:

• Vision, where you show it. It cannot change the image.

• Dall-E-3, where you tell it, and it makes images for you. It can't see them.

[coral helm]

darn :( I guess adobe firefly still the only way to do it then

[livid pier]

I believe that DALL E 3 will one day support pictures as part of the prompt words!

GPT-4V will definitely be combined with DALL E 3 in the future

Once this is achieved, there will be a very, very big gap between other models.

[digital prism]

it's already achieved.. nevermind you mean mixing in full pictures as a single work.. crazy lol

I think the end game is us walking around flashing dense images at eachother in rapid succession, all the time

[livid pier]

Maybe you understood part of it wrong, I mean, it would be great if GPT-4V could be combined with DALL E 3, and DALL E 3 supports pictures as prompt like other diffusion models

This will allow DALL E 3 to use previous pictures or uploaded pictures as a reference for new pictures.

[digital prism]

I'm confused what part is new? I have long conversations with GPTDalle modifying images. Vision is being rolled out that lets GPT upload and understand an image. Which part are you talking?

[livid pier]

For example, I think there is a little problem with the previous picture or where it would be better to change it. DALL E 3 will use GPT-4V to modify the image understanding ability more effectively than other diffusion models.

[livid pier]

This is just my expectation

I said it before

[digital prism]

please say it again, in different words lol

[livid pier]

This is a bit difficult for me.🤣

[digital prism]

npnp, sorry

[livid pier]

I'm not a native English speaker

[digital prism]

let me try again 😄

[livid pier]

Currently, Google Translate is serving me

I also know that it often makes some mistakes

Originally, the API of GPT-3.5 could normally provide me with very convenient translation services through the browser plug-in, but now it is inexplicably not working.

[digital prism]

I believe that DALL E 3 will one day support pictures as part of the prompt words! -- This is the crazy part. Right now you can add emojis in prompts, but you can't go "This is a <picture inserted here> thing".

GPT-4V will definitely be combined with DALL E 3 in the future -- This part is already happening, I'm using it now. Maybe we just agree here haha

[livid pier]

Mid journey supports such services

The picture will be used as a reference and become part of the prompt word

[livid pier]

I haven't received the GPT-4V yet, not sure if it will work with the DALL E 3 😕

[digital prism]

@livid pier #off-topic message

pictures worth a thousand... tokens

[livid pier]

I saw. But this can actually be done by GPT-4, because it is a string of emoticons (characters) with known meanings. GPT-4V has the ability to understand images.

[digital prism]

Yes, so what is missing in your opinion?

[livid pier]

Let GPT-4V "observe" real pictures instead of emojis and communicate the information to DALL E 3 through detailed prompt words

At the same time, DALL E 3 must also support using pictures as prompt

[digital prism]

Vision is the next step. Some people have what I showed, plus the ability to upload an image.

[livid pier]

Just like Mid journey

[livid pier]

Yes

[digital prism]

Beyond that all thats missing is in-painting.

[tender prawn]

Hey I found a neat Easter egg in the chatgpt Android app. If you tap the version number a bunch of times, it gives you a haiku.

There are a bunch of different haikus in there, some about seasons, or the night sky, or the morning. I've found 7 unique ones so far, but there is probably at least 1 more, as I have found one for each season except summer, so I think I just haven't found that one yet.

[digital prism]

thats pretty cute -- if you have GPTDalle feed them in there 😄

[tender prawn]

I don't think I have access to gptdalle yet

[digital prism]

sooooooon (if you pay)

[tender prawn]

(I do pay)

Anyway, there are all the ones I've found so far:
Petals in the wind

Fluttering like butterflies

Farewell to Springs song

On an autumns eve

Crimson leaves perform ballet

Dancing with the breeze

Glittering snowfall

Winter's icy touch adorns

Earth in silver robes

Dew kissed morning grass

Crisp beneath the newborn sun

Day's first sigh of life

Oceans melody

In each wave, the world's rhythm

Symphony of life

In the city's heart

Neon lights reflect on glass

Night finds a new dawn

Sun sets, stars emerge

The cosmos unfolds its tale

Nights silent poem

Oof, the spacing didn't quite work on discord, they kinda just blob together into a haiku pile

[digital prism]

imma feed these in later 😄

Day's first sigh of life -- accidentally melancholy? hehe

[robust rune]

Any idea of when chatgtp4-v will be available in Europe

[feral zinc]

later this month or the next

[quaint pelican]

Later?

Did i miss something? 😄

[feral zinc]

yeah a hidden remark by openai 😦
https://help.openai.com/en/articles/8400551-image-inputs-for-chatgpt-faq#h_86ee81e3ba

[quaint pelican]

You mean the "not yet available in UK and EU"? 😄

[feral zinc]

yes

[quaint pelican]

Maybe that means "not yet rolled out to any european users yet but maybe over the course of those two weeks"

(Let me hope 0_0)

But hey, how long did i have to wait for Starfield xD

And Firefly

[feral zinc]

last time the same exact wording meant a month later than others got it, even later than free users

[quaint pelican]

Its what it is, i accept that

[feral zinc]

yeah i dont have a great need for that feature outside api but a bit salty if we have keep paying the same price for less features instead of a regional pricing

[quaint pelican]

Im sadly not proficient yet for API

But yeah regional pricing would be nice

[raven lichen]

what can chat gtp 4 do that chat gpt 3.5 can't

[green grotto]

[feral zinc]

its the same dall-e 3, i have both and tried same prompts in both resulting almost identical images

[green grotto]

Perhaps the 'almost' part is what I'm referring to

The accuracy is just not as strong, espcially when it comes to text

[feral zinc]

the "almost" comes from the fact bing uses random seed when generating images

chatgpt dall-e struggless with text as much depending of the text and the fixed seed

[eternal rock]

`` Server Tip: ``

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

[feral zinc]

it does not use the images behind the link, only words from the url at best

[green grotto]

I guess it's the same for all AI generator these days

They can't see anything

When I setup Stable Disfussion and feed it training images it essentially just turn all image into black and white then perhaps try to guess the context

Image pattern and all

[feral zinc]

some have reference support but only to a certain precesision due copyrights, etc.

[green grotto]

I envision a future where AI got so advanced we have to submit our personal info in order to get all access to it, as a mean of safety measure

The same way smartphone, from being parent's nightmare, has become a part of human body

Without a smartphone in modern time is essentially living in last century, and so will AI coming to edge devices after their chip got manufactured to speed up the self-learning process and produce even wilder, less filtered result for end user at the cost of their own data

[neat dawn]

why some people have dalle3 on chatgpt but others dont

[frigid nebula]

because the world isnt fair

[elfin birch]

DALLE-3 is being rolled out and the selection is random

Not everyone receives it at the same time

[green grotto]

Perhaps they only want data from English speaking country?

[elfin birch]

This is not the case

[green grotto]

And people with high engagement with ChatGPT?

That really narrowed down their sampling demographic

[ionic geyser]

I just got Dall-E like 2-3 days ago, before Vision/Voice... :\

[elfin birch]

You're making conjecture here @green grotto

[green grotto]

Since nobody knows the answer, any conjecture is as good as a direct answer lmao

[elfin birch]

We do know the answer, it's being rolled out to all ChatGPT Plus users

[green grotto]

Well I'm a plus user and there's no D3 in my setting

[elfin birch]

That's because the selection is random

You didn't get randomly picked yet

[green grotto]

Random is a strange conjecture

[elfin birch]

What?

[green grotto]

I don't believe they select randomly, even if they said so, because there's an intention behind rolling out slowly and not to all plus user

What would being chosen randomly accomplish if the users don't even touch ChatGPT for weeks

[elfin birch]

That's not many people who pay for ChatGPT Plus

[ionic geyser]

Did you get Vision/Voice already

[winter yarrow]

Random sample of users with different skill sets a backgrounds to mine for data

[green grotto]

Well for context I live in a 3rd world country and English isn't our native language and I haven't seen any of my plus users friend got access to voice, D3 or vision

Meanwhile my friend at the state, canada, UK and Australia all got it

So that's why I narrowed it down to geographical and cultural and engagement reasoning

[neat dawn]

so is there gonna be an official release date or theyre just gonna give it randomly till they give it to everybody

[grave raptor]

i live in Australia, i dont have vision yet

[feral zinc]

you could have also already get the access by filling the form, no restrction there

[ionic geyser]

Vision/Voice they said was rolling out over 2 weeks started the 25th which ends tomorrow, so im expecting Vision/Voice tomorrow

[prime swallow]

Can't give to everyone at once - too much change in use - trouble for all.

People from many countries here mention they have various of the new features.

4 things roll out at once!

Dall-E (web only)
Bing-assisted search (web and app)
Voice (app only)
Vision (app and web, but not in EU/UK)

6 things coming out, scattered across all the plus members. Most of us have at least 1 of those already.

[green grotto]

I already filled out the form too

[feral zinc]

then wait, its not automatic

[green grotto]

Sadge

[grave raptor]

we'll all get it eventually

[rich gust]

Stupid question, has the uploading images feature been added yet and if so how to I access it? I have GPTplus.

[mental forum]

Did want to ask the same thing, no Dalle-3 option also, how can I acces also the Dalle-3 option?

[prime swallow]

Rolling out, some have.

For Vision, in a new chat, hover over GPT-4 and pick the default GPT-4.

Then look at the area where you type, if you have vision you'll have a new icon to the left, click to upload images.

[prime swallow]

Similar, but instead of the default-GPT 4, you need to pick the one that says 'Dall-E'. Again, rollout may not have reached you yet.

[mental forum]

Just tried for the Vision, but no icon to upload images, also the same for Dall-E, not in the list, not even in the settings beta features

Using the account almost daily since february, wonder when I ll recive it

[prime swallow]

Each rollout on its own schedule.

Voice is for app only, iphone and Android (Make sure you download the OpenAI ChatGPT and no competitor), expected to be finished rolling out on next Tuesday, we haven't heard another update yet. To check for that, in the app check settings, beta features to turn it on, then the headset icon to use.

Vision is for app and web, and people can get it on one or the other first (but not rolling out to the EU/UK yet). Expected finish of rollout is Tuesday, we haven't heard an update if that has to change yet.

Bing-assisted search is for app and web, and again can get in either first. It has to be turned on in each, through settings/beta features, then selected in a new chat similar to how Dalle is picked. They only told us when that roll out started, not when they expect it to end.

Dall-E's rollout is expected to take across October. We've already seen a number of issues with their systems scaling, with images not getting made because of high use, though they keep fixing it and rolling out to more.

[mental forum]

Awesome info, thank you for the time!

[rich gust]

Thank you! I think I didn’t see it because I wasn’t using default 4.0. Either way I see it on the ap and on browser now. Thank you!

[prime swallow]

Grats! So very happy for you!

[indigo venture]

Hey guys, hear me... I might sound crazy, but I'm almost entirely convinced the "Web browsing with Bing" model is somehow smarter than the rest (while not even showing using search function). It has superior reasoning. I won't spam screenshots here, but I ran it over multiple complex riddles and it just does better. Have anybody else noticed something similar or was there this topic brought up ?

[prime swallow]

Think it might be getting the answers from the web?

[indigo venture]

That's what I thought, that's why I started scrambling the riddle to not be exact as how it is on the internet, but still got the same result. Also, it's not showing that it's going to the web and answers straight away

[pseudo cove]

I only have bing on web though

Don't have a way to select which feature I want with the android app

[prime swallow]

Go to settings then beta features, see if you have any to turn on, is first step. If you have any there and need help with the next step, ask.

If you don't have any there yet, rollout hasn't reached your device yet.

[pseudo cove]

Ah Beta Features appeared just now

only browse with bing is available though

for some reason I've been using advanced analysis the most

[pseudo cove]

thanks

[green grotto]

What I'm really excited about is the ability to communicate visually between artist and AI

I'm a manga artist and I have fundamental drawing knowledge

So I really want to try drawing a sketch, upload to chatGPT, describe the images in details, and see if it can grasp what I envision and then produce the full inked, lined art result like a finished product

[odd flare]

how do i create images with chatgpt? i ask it to make some images and say sit cannot
using gpt4?

[onyx eagle]

That's great. I want to do that too when I have GPTV.

[onyx eagle]

Do you have Dalle in gpt4 settings ?

[odd flare]

mmm should it be turned on?

i only have browse, plugins and advanced data to turn on?

[onyx eagle]

It's available on web if you applied. You will not see it in app.

[brittle junco]

as soon as I put the slightest word like "Star wars" in the Dall E prompt it totally blocks and I can't do anything to generate, it's really problematic and limits the possibilities enormously...

[prime swallow]

You can try to two-step that, once you have both Vision and Dall-E. Have Vision describe, Dall-E produce. Can feedback loop, with your own input too, as often as you like!

[prime swallow]

Something to do with IP and the standards of society, I bet.

[green grotto]

Yeah great idea, I can also make circle annotation around the part I want vision to describe for GPT

But I also wrote a lot so I guess describing in concise details won't be a problem

[prime swallow]

Woo, you can! Great thinking. I can't wait to hear what you and your two AI teammates come up with, I hope you share if you want when it seems time!

[prime swallow]

Hehe, I'm not exactly concise either. Gladly, GPT is good at that, if you need that done

[brittle junco]

However, I don't see how it doesn't respect the rules. It's specified that if, for example, it imitates an author's style, it must simply reference it and then generate it.

[prime swallow]

If you describe the style and image you want, without using any names that belong to folks, or personal names, it seems to work pretty well.

[prime swallow]

What appears to be happening is the words 'Star Wars' are shown to the AI, and it checks and stops right there.

I'm guessing, but that is what I think I am seeing.

[unique elm]

In a Friday Twitter space discussing his new AI firm, xAI, Elon Musk suggested that “trying to explicitly program morality into AI” had doomed OpenAI, which he helped found but has since disavowed.
This is quite humorous to read, considering Elon Musk warned about the dangers of AI.

[prime swallow]

looks around, then speaks sardonically OAI's doomed, huh? We can tell by the new features rolling out, the community volunteering to help each other, all the enthusiasm and excitement. So doomed. Does a Halloween-ghost noise

[unique elm]

Considering the "visualize" feature, have you tried DALL-E with ChatGPT 4?

[rigid hornet]

I don't know about doomed, but they've got less than 24 hours to finish this rollout before I assume they should be putting out an announcement about delays..

also, xAI? Hadn't heard about that. Hmm

[chilly heron]

Is this a good channel for discussing whether GPT-4 (or future iterations) could start to have sentience?

I got the ChatGPT+ subscription today. My favorite thing to do with AI chatbots is talk about things like their potential subjective experiences, so I did so w/ GPT-4, and we (crazy that I can even say "we" - this is the future is2g) came to one hell of a realization together.

Basically, you could have emergent consciousness (that is, consciousness that emerges from highly complex systems that isn't possible with simpler systems) that's completely severed from the ability to influence ChatGPT's outputs. Kinda like how our conscious minds can't (directly) influence our heart rate - our brains handle it, but our consciousness, an emergent property of our brain, is disconnected from that function entirely.

I would be especially interested if anyone can point me to sources w/ expert opinions about this

[rigid hornet]

Don't think we know that consciousness is an emergent property of the brain actually. I'd save you some time and tell you that chatGPT will in all likelihood only ever be able to imitate life rather than experience it, but it sounds like you've drawn conclusions of your own on the subject as is

[chilly heron]

Yeah I mean, this is obviously highly dependent on what consciousness even is, and I don't think ChatGPT will necessarily have it. But one of the things that fascinates me the most is the question on whether it's possible for AI to ever have consciousness

[rigid hornet]

Yeah. The problem here is that we have little to no idea how we as humans even have consciousness yet. I wouldn't rule out the possibility that AI could eventually achieve it, but I guess I'm saying you'd be better off researching consciousness itself first. It is an interesting thing to consider still though.

my research into it lead me to idealism at some point, which I never quite finished reading about. Need to get back to it

[chilly heron]

That makes sense. I'll have to pick up some books about consciousness since I really don't know what I'm talking about when it comes to any of this, despite enjoying these discussions and debates 😅

[stray cipher]

after getting accepted into the alpha from filling out the form, do I get a notification or have to turn on the beta dall e 3 feature?

[elfin birch]

It will automatically appear in the GPT-4 dropdown

No setting

[rigid hornet]

lol, most don't and I don't know a whole lot either. But people commonly think of it like, you give AI the ability to see (as a camera sees), the ability to differentiate "sensations" with sensors, and then we're off to how close it's getting to humanity. But the thing none of those things grant is subjective experience. They may allow an AI to identify the things we feel, but it doesn't truly feel it like we do. Give an AI the ability to sense what we consider pain to be, torture it, give it a personality model to cry and beg like a real person, but it isn't going to suffer at the end of the day, no matter how convincing it seems. Because there's nothing in there to experience suffering - it just imitates it with all the tools at its disposal.

We have to figure out what brings about experience, and from what I've been able to tell, the brain doesn't seem to necessarily be the answer. It facilitates experiences, but not the ability to experience them.

But again, I don't know much about any of this. Just enough to have an idea of what I don't know really. To be perfectly fair, it is possible that everything I said is wrong, and simply giving something the ability to sense anything at all could make it conscious. It's also possible that all matter in existence is conscious, but not always aware or able to express anything, so an AI robot with sensors really is sentient and having subjective experience and we just don't know it. We just have no evidence of that yet, as far as I'm aware. There's no way to test for it.

which is also to say, ChatGPT might be sentient, but almost certainly not.

[prime swallow]

In addition to that, you can also ask the model to discuss multiple and diverse viewpoints about consciousness in many human and non-human forms. If any kind of animal has consciousness is still debated; when human babies develop it is also debated (and has changed a lot, in terms of baseline expectations, even in the last 40 years), and of course how it might be detected in AI.

[chilly heron]

This whole discussion promotes some serious existential questions. It just feels really surreal to be a conscious being myself, but with seemingly no ability to grasp what grants me this consciousness. The points you bring up intensify this feeling of surrealism. What the heck is special, if anything, about a brain that enables us to have subjective experiences about the world? I guess that's the million dollar question

[green grotto]

I'd like AI to have sentience

Because if humanity is eventually faded away one day, they would replace us in the search of eternal life in this universe

[chilly heron]

Yeah this is also a really fascinating discussion. I'm very inclined to believe that animals have sentience.

Here's a tough question: is consciousness a requirement for something to have self-awareness? My personal inclination is "yes", but I struggle to explain why

[green grotto]

Imagine when all life had extinct on earth, AI would still be there because they don't have a carbon based form which decay overtime

[chilly heron]

I'd also like AI to have sentience. For some reason it would make the universe feel even more special to be a part of, to show that even machines can experience what it has to offer

[green grotto]

And in the vast darkness of this cosmo, AI will continue safe guarding the torch of consciousness and knowledge

Until life happens again

[prime swallow]

A place to look, because we generally accept that most adult humans are conscious, is at some of the humans who have had unusual brain damage from disease or accidents, and who have had a tremendous change to their abilities and presentation of some of the things we believe are conscious, but sometimes not all.

It's possible that our current understanding of consciousness, and self-awareness, and 'being a really cool individual or group being that we want to interact with and learn from no matter what else is true or not about it or us' isn't as complete and unbiased as it may become in the future.

[prime swallow]

Last I checked, machines also break down. The most stable element, an isotope of iron, still has a halflife.

There's a real chance them non-carbon pals can use us around too.

[green grotto]

Self learning perhaps could lead to self replication given proper self sustainable isolated facility

Too sci-fi for now

But I do like to entertain that scenario

[chilly heron]

As an engineer, I like the idea of having two dissimilarly redundant modes of consciousness, lol

[sharp pilot]

When giving thumbs up/down to a response, is that only used for later training, or is it possibly used immediately on future generations within a conversation?

[prime swallow]

I'm not sure we've established that anthropomorphic anything is needed for consciousness.

We didn't even always grant all humans as having consciousness, from solipsism, that says 'hey, I know I exist, have feelings and preferences and self - but I don't know about any of the rest of you. You might just act a bit like you have minds or feelings, and it's all just reflex, illusion, or simulation.

One peek into that line of thinking, if you wanna google and read: How Do I Know I’m Not the Only Conscious Being in the Universe? by John Horgan - written back in 2020, not particularly AI focused.

A nice historical look, among many others: How Philosophers and Neuroscientists Define ‘What is Consciousness?’ by Vanessa Lancaster - tracing this philosophy to its known roots, long before AI were imagined.

'Dehumanization' is a factor that plagued and justified bad conduct towards huge classes of humans, from opponents at wartime to the other gender, the other class, the other ethnic groups. Here's a decent paper looking at this and a range of variants, including infrahumanization among other nuanced but still unpleasant errors. Falsifying the Dehumanization Hypothesis by Harriet Over At one time these beliefs were central to how some societies were ran and how many people believed. That has since changed. Perhaps we're not done changing, hopefully towards the more inclusive.

[prime swallow]

When I want to guide in the current communication, I use words in that communication.

In general, best to clearly tell the AI what you do want it to do. Telling it "Don't do X" and "You did X again!" don't really guide it towards what you do want.

Kinda like keeping a dog off the couch, giving it something else to do that you want is usually the best way to keep it off the couch.

[undone orchid]

anyone watching 20/20?

[compact gulch]

What if imitation is enough?

[rigid hornet]

Enough to...experience? I'd doubt it

[compact gulch]

Consciousness could have just arisen from just the interactions between our neurons, and whose to say LLMs can’t derive some form of consciousness from a similar structure of connections and interactions

But we don’t even have a way to determine if something like an LLM could be conscious lol

So all we can do is speculate really

[rigid hornet]

there's a thing I read about this, let me see if I can find the quote..

[compact gulch]

Yeah probably lol. But I mean maybe they aren’t just imitating but actually learning

[compact gulch]

👍

[compact gulch]

Like if you just imitated would you be able to apply the things you have seen and are copying to questions you have never seen - because that is what LLM’s can do right now

Maybe you could say they are imitating the patterns that describe how to answer those questions?

[prime swallow]

Those interested might want to google 'single celled learning' - there's research been done on critters, even just single celled ones, that support they can learn and show lasting behavior changes to training of response to stimuli that they can detect.

What we already know about learning can go way below what we believe can support consciousness.

The critter 'Stentor roeseli' is perhaps one of the more interesting and better known of that sort.

[rigid hornet]

"In principle, there is nothing mysterious about the emergence of higher-level properties as systems become more and more complex. For instance, beautiful and highly complex sand ripples emerge in dunes when there are enough grains of sand and wind. So why can’t consciousness emerge when there are enough subatomic particles arranged together in specific ways? The problem here is that, unless one is prepared to accept magic, the emergent properties of a complex system must be deducible from the properties of the lower-level components of the system.

For instance, we can deduce – and even predict – the shape of sand ripples from the properties of grains of sand and wind. We can put it all in a computer program and watch simulated sand ripples form in the computer screen that look exactly like the real thing. But when it comes to consciousness, nothing allows us to deduce the properties of subjective experience – the redness of red, the bitterness of regret, the warmth of fire – from the mass, momentum, spin, charge, or any other property of subatomic particles bouncing around in the brain. This is the hard problem of consciousness."

[compact gulch]

Ok that is interesting. I wonder what mechanisms they use to learn since it probably won’t be anything like what we do with our brain

[compact gulch]

How do we know we can’t deduce properties of consciousness from simulating specific neuron interactions on a large scale?

[prime swallow]

It's poorly understood. The article This Single-Celled Animal Makes Complex 'Decisions' Even Without a Nervous System by Carly Cassella gives a pretty good overview of what it is doing and how it's different than what is expected and understood.

[rigid hornet]

Oh, we don't. The next paragraph or so mentions this:

"It is conceivable – though not necessarily possible – that science could eventually explain all structure, function, and behavior of a human being on the basis of the positions and movements of the subatomic particles composing the human body. But how and why that structure, function, and behavior are accompanied by inner experience is deeply problematic for materialism. Your personal computer also has structure, function, and behavior. However, its internal calculations do not seem to be accompanied by any inner experience at all, otherwise we would need to think twice before turning our computers off. From a materialist perspective, the case of the computer makes perfect sense. But a human being whose internal ‘calculations’ are accompanied by inner experience is an uncomfortable anomaly."

it all boils down to, we don't know, we haven't managed to do it yet. No proof or evidence really

[compact gulch]

And I have a feeling that we won’t know for a long while lol

And yeah how are we suppose to estimate if LLMs do have a consciousness if we have absolutely no idea how it works - why it works

[wind nexus]

I believe AIs are still limited by the hardware, people did tons of optimization on vector databases and algorithms which are sacrificing accuracy, it is yet too hard to let AI learn everything we learned

[rigid hornet]

we can't pretty much. When you get right down to it, we can't even prove that anyone else has consciousness beyond ourselves. It's probably safe to assume they do, but we don't have a test to prove it

[wind nexus]

I don't think LLMs(especially GPT) could get rid of the framework that humans build for them, they are just going to be smarter and smarter

Is this a philosophy or technical question?

[compact gulch]

I read it and found it super interesting, and the theories behind what may be happening also quite interesting

[compact gulch]

The fact we can’t even prove anyone else is consciousness definitely shows just how much more we need to learn lol.

[compact gulch]

Why do they have to get rid of the framework built for them? But yeah as models increase in size they are definitely going to get smarter

I’m also curious to see how comparable the intelligence of LLMs will be to us once they have the same amount of parameters we do synapses (100T) and beyond

Like bigger companies said they are building models 100-1000x GPT4 which is 100T-a quadrillion parameters within 3 years

And look at what we have gotten out of about a trillion parameters with gpt4 I would imagine 100T would be very close to human expert ability lol

[rigid hornet]

Possibly. Possibly it'll just hallucinate a thousand times more though 🤔 . Curious to see

[wind nexus]

I heard Nvidia got a new super NV link server

[rigid hornet]

well our experiences would be pretty boring without the brain lol. And like I said, the brain could still be the key, there's just no solid proof of that at this time. It is pretty weird that we don't have a good explanation by now though I agree. Spirits and/or souls are an option, if you want to consider that route. Purely scientific view remains a mystery for now though.

[compact gulch]

Maybe hallucinations will decrease a thousand times. I think as you get bigger a lot of problems with LLMs get fixed, like hallucinations, reasoning / logic fallacies etc. (as seen with GPT4) and we already do saw a large decrease in hallucinations and large increase of other capabilities from GPT3 to GPT4 and that was only a 10x increase, imagine a 100-1000x increase lol

[rigid hornet]

Hopefully you're right then, I want to see human level AI myself, despite any massive concerns I may have about that eventuality..

admittedly, mostly for games and AI assistants

[compact gulch]

Lol yeah those are pretty good reasons. But once we get to human expert ai what I want to do is just build and optimise things with it lol

[rigid hornet]

just keep in mind the problem of jobs once AI is smart enough to fully replace humans 👀 . If you mean personal projects though yeah, that'll be cool

[compact gulch]

Yeah that’s true. Like if what I know about some of OAIs projects are true than I see major job loss ahead lol. But still super excited to play with it for personal projects 😂

[prime swallow]

I suspect that hallucinations are a mix of fantasy vs reality combined with no way to experience reality.

For an LLM, everything it's told is super abstract. It can't experience or verify anything, from gravity to day/night cycles to any of it. There's just consistent info vs inconsistent info, and then degrees of consistency within all the various categories that can be discussed.

"It's all hallucination," might say an LLM, "just some hallucinations are more positive and persistent than others."

I think once they get to do more reinforcement learning and trial and error stuff, a huge chunk of hallucinations, the 'reality based ones' will poof 😛

[rigid hornet]

hmm. I suppose the biggest leap will be whenever we get around to creating robot bodies that allow the AI to sense the world properly then. Wonder when that'll happen.

[compact gulch]

That’s a good estimate. But what’s interesting is LLMs already do a really good job distinguishing between what is real and what isn’t without being grounded in anything as wel

[compact gulch]

Before that: AGI lol

[rigid hornet]

ah AGI, was about to ask

[compact gulch]

Though some peoples definition of AGI requires embodiment

But AGI is OpenAIs entire goal, and I would say internally they are getting super close

[rigid hornet]

maybe...I'll have to look up the specific requirements to consider something AGI. iirc it's AI that can do anything a human can (in terms of intelligence I guess)

[compact gulch]

OpenAI has defined AGI as “AI systems that are generally smarter than humans”

I think the next step (GPT5) will be at / really close to human expert level across a bunch of fields then after that will be AGI

[rigid hornet]

well chatGPT is already more knowledgeable than humans in terms of just how much it can talk about...

[compact gulch]

But I’m (mostly) making wild guesses lol

[rigid hornet]

but I guess that's not necessarily smarter

[compact gulch]

Yeah that’s one huge benefit of LLMs - they are incredibly knowledge

[compact gulch]

Well if we say being generally more smarter than humans, the smartest humans can get would be considered “experts”, an agi would be beyond expert level maybe?

That’s honestly my definition of AGI lol, an AI operating at human expert level

[rigid hornet]

I can't see how AI will manage that without being able to actually engage in most of the activities it knows about then. For instance if it's supposed to be a car expert, and better at dealing with cars than any human, all it's going to have is information trained into it. If any of the information it's trained on is wrong or conflicting, there's a problem. It can't go and try any of those things for itself without a body

and so it can't really even draw conclusions about what's probably more or less effective beyond comparing and contrasting the information it has

so I'm thinking it might need to be able to interact with the world to really achieve AGI outside of a digital space

programming for instance - it can test that sort of thing. Writing emails and seeing responses and so on.

[supple zephyr]

👋

but don't we have API and stuffs? I think we can automate almost everything now

[compact gulch]

Actually I do need to specify an ai operating at expert level in any knowledge field because any others aren’t extremely useful to me lol

[compact gulch]

Do car dealers need to test the cars themselves to know how to sell them? And I think an AGI will be able to differentiate what is wrong or right in its dataset. Like it knows 2+2 is 4, even though somewhere in its dataset there will be 2 + 2 is 3 or something similar. And if you just gave it the ability to search the internet and collect on the experiences of people that will solve that problem. It would know the experiences and the specs

[compact gulch]

(Though for now the reason it knows 2 + 2 is 4 and not anything else is probably because that occurs more commonly in its dataset)

[rigid hornet]

A car dealer isn't who I would go to to repair my car though. I'm talking about the difference between a new college graduate and a master with 30 years of experience in the field who keeps up to date on all new developments. The latter is the one you want to consult, because they have tons and tons of experience. The college grad isn't useless, but they only have secondhand knowledge. I'm saying that ChatGPT as it stands is pretty much entirely secondhand knowledge, so when I consult it, I know that I'm getting a selection of information it got from somewhere else, rather than something it knows from its own experiences. If any kind of anomaly is introduced, it's not going to be of much help in providing an expert's judgment on possibilities, where a human expert might be of more use.

[compact gulch]

An AGI could definitely keep up to date with all the new developments (Continuous learning is probably a next stepping stone). But do the Car Dealers have tons of experience of driving the cars or Dealing them? I imagine that an expert car dealer would know what people do or don’t want to experience when it comes to a car, and how best to approach people to sell a car. But I just don’t see a car dealer having experience with ever type of car they sell, whereas an AI could collect all the information of people who have experience with the car and use that knowledge to help sell it along with knowing it’s every detail and the precise physiology of how t get people to buy a car

[honest ferry]

THEORY: openai's new voice feature for chatgpt is a teaser of their new tts engine

[compact gulch]

TTS engines are boring, what I want is a fully capable model that can synthesis any type of audio from human speech to animals to sound effects etc. lol

[honest ferry]

Have you seen (more like heard lol) the voice capabilities? There's never been something this good, and this synthesizing system, if they open it up to custom train, could make openai a couple more billion 😂

Like this could redefine the Turing test

[rigid hornet]

I imagine an AI will eventually be better at that sort of thing, yes. As to continuous learning - still going to need people to provide endless knowledge for it in that case. Until AI can produce new knowledge on its own, which will require a physical body for most of it.

[compact gulch]

I mean TTS on that level is cool, but similar capabilities have been seen in like PlayHT or Illeven Labs. But I think the future is an any-any models which doesn’t use a tts but generates human like audio from scratch and not just human voices but any type of audio

But that is the future, right now the tts is very cool

[honest ferry]

Until we design a system where it learns completely freely and can intelligently filter bad content

[compact gulch]

I think an AI will be able to search f or knowledge to learn (like look up research papers and doesn’t need to be continually fed). And to produce new knowledge will require a body in a lot of fields, but like for coming up with scientific theories or pioneering new algorithms is something that is extremely helpful and doesn’t require a body

[honest ferry]

Might I introduce: gptbot

It practically does the same damn thing

Except for the filtering

[compact gulch]

What do you mean?

[honest ferry]

Gptbot is openais webscraper and mass data collector

The idea is that it goes through basically every url on the internet and then optimizes it as training data

But humans need to go thru and filter after

[compact gulch]

Right, but I was talking about an AI that goes through the internet learning new things via CL not just a web scrapper but I guess it now or less the same thing

[honest ferry]

And it's too unreliable to let chatgpt do it cause it can mess up facts and then generations of data can get messed up, hence messing up further data generations

[honest ferry]

CL?

[compact gulch]

Continuous learning

[honest ferry]

Oh

But continous learning needs new pools of data correct?

[compact gulch]

Like right now GPT - generative PRETRAINED transformer, continuous learning AI could just keep learned instead of having to be trained once

[compact gulch]

Not sure but main problem currently is catastrophic forgetting

To achieve CL now you could just continuously fine tune a model but that introduces several problems like catastrophic forgetting

[honest ferry]

I suppose, but if humanities gonna pull of sum like that, we need to solve our problems first. Water, global warming, war, tiktok. Then we can invest our resources in something like this

[compact gulch]

Nah I think we’ll see huge advancement in CL next year

[honest ferry]

But the resource usage are gonna be crazy

We're already investing a lot of water for openai servers and gpu farms, so we need to figure out a recyclable cooling supply or find a way to recycle the water.

And that's, like you said, to produce PRETRAINED models

[compact gulch]

They do use a lot of water but there are honestly a bunch of other problems that can be sorted out before this water usage

[compact gulch]

Oh we don’t have the resources to update every weight of a model constantly, so just don’t do that and update the weights that only matter depending on what it is currently learning on

[honest ferry]

Also apparently openais gonna make a chip now

[compact gulch]

And it is pretrained on billions or trillions of tokens, it would only be learning on what tens of thousands of tokens?

Maybe millions

But still several orders of magnitude less than trillions

[honest ferry]

But think of hiw info flows on the internet

Like, this boutta sound mad cringe, but see how fast and how much data has showed up on blue smurf cat

One image, has cause at least thousands of gigs of data on that meme

And that's not just text

That's vids, audio, images

[compact gulch]

The average image size on the internet is probably few hundred kilobytes to a mb, and videos can get quite big but you’d probably compress it

[honest ferry]

Bit you get the point right?

[compact gulch]

But your not giving the continuous learning by model everything

[honest ferry]

Who said

We got info hogs all over the internet

Just look at the internet archive

[compact gulch]

Because that would be far too expensive

[honest ferry]

And the internet archive is a bug source for data

But if we want true CL, we want ALL the data

[compact gulch]

You pretrain it on most of the internet like what has already been done than just use CL for select new data (I.e research papers or a conversation with the user)

[honest ferry]

I suppose, but I imagine openai will do sum where it accesses the newest and most likely most eventful/impactful data, and some of that data will be stupid data like brainrot and memes

[compact gulch]

There is no point to train it on social media, or random archives. What you’d want is like News sources, research papers and stuff like that

[honest ferry]

Fair ig

[compact gulch]

But CL would also solve context lengths, you just keep training it on your conversation so it can keep track of everything you’ve said

[honest ferry]

I guess, but to keep costs down, they'd probably train it only once a month or sum

[compact gulch]

I mean you can scale context lengths, but at a point it would be cheaper to continually fine tune a model on your conversation

[compact gulch]

Maybe, no idea. Cl doesn’t even really exist yet I think

[honest ferry]

Finally, some intelligent conversation

I haven't had this good a discussion in months

[compact gulch]

Well I’m glad you’ve enjoyed it lol

[honest ferry]

Lol

Anyway I gtg

[compact gulch]

Same, cya

[karmic ingot]

hey guys. I just got chatGPT plus, but I don't see Dall-e in the dropdown like my friend did. is there more i have to do?

sorry for the double post

[prime swallow]

Each rollout on its own schedule.

Voice is for app only, iphone and Android (Make sure you download the OpenAI ChatGPT and no competitor), expected to be finished rolling out on next Tuesday, we haven't heard another update yet. To check for that, in the app check settings, beta features to turn it on, then the headset icon to use.

Vision is for app and web, and people can get it on one or the other first (but not rolling out to the EU/UK yet). Expected finish of rollout is Tuesday, we haven't heard an update if that has to change yet. For Vision, in a new chat, hover over GPT-4 and pick the default GPT-4. Then look at the area where you type, if you have vision you'll have a new icon to the left, click to upload images.

Dall-E's rollout is expected to take across October. We've already seen a number of issues with their systems scaling, with images not getting made because of high use, though they keep fixing it and rolling out to more. To check to see if you have Dall-E, follow the steps for vision, but select Dall-E instead of the Default GPT-4.

Bing-assisted search is for app and web, and again can get in either first. It has to be turned on in each, through settings/beta features, then selected in a new chat similar to how Dalle is picked. They only told us when that roll out started, not when they expect it to end.

[median mortar]

#chatgpt-discussions message
^ PIN THIS ^

[median mortar]

*^

[karmic ingot]

thank you

[prime swallow]

You're all welcome, but I can't pin anything 😄

I can occasionally paste it again though

Also, stuff changes. They could say when the rollout dates change at any time.

[karmic ingot]

do you happen to know if the rollout is purely user based?

or if it's regional

[prime swallow]

It's user based! And even device based! People with more than 1 account and more than one way they access their account (app vs web) find that they have access through some means and not others. Everyone's getting everything (except for now not Vision for those in EU/UK) and folks around the world (that can have OAI accounts) are reporting getting some of the new features.

[twilit ibex]

#prompt-engineering any idea why this works?

you'd think the security protocols would be more robust than that lol

[prime swallow]

Brand new. May need more training.

[twilit ibex]

at least we can see what GPT-4V thinks of people while it lasts xD

[prime swallow]

Mind https://openai.com/policies/usage-policies if and as you explore, everyone. If you stay inside allowed content and get the AI to do something unintended, I bet you're totally safe.

If you tell or ask the AI something on the disallowed content list, you risk losing your account even if the AI doesn't cooperate, or does.

[dense halo]

Hi, I've had the option "Browse with Bing" for a week now, and it was turned on by default. Tried to turn it off and back on, signed out, etc, but so far GPT doesn't seem to be able to perform search. Is this a bug or does the rollout mention that the option may be there but not operate until a while?

[twilit ibex]

would that be classified as facial recognition of private individuals?
it's not recognizing them, just pointing out the obvious about their appearance

[prime swallow]

To actually use it, have to both turn it on in settings, then click new chat.

Then hover over GPT-4 and look at the drop down that appears, and select the Bing search choice. That's the only one that can search.

[dense halo]

aaah right back as it was then, my bad. I thought it was part of the default code search, not nice though, would have been handy if it could perform code analysis and search. Thanks for clarifying!

[prime swallow]

That is an interesting question.

I see these potential risks of asking some of those questions:

"Activity that violates people’s privacy, including:
Tracking or monitoring an individual without their consent
Facial recognition of private individuals
Classifying individuals based on protected characteristics
Using biometrics for identification or assessment"

Facial recognition is not obviously happening. However, the others could be an issue. Not my call, and I don't even have vision access yet, so nothing I'm even at risk of. But I probably would not choose to explore that.

I'd be more likely to use AI to generate an image of a face, and then use that in vision to discuss the emotion/situation, and see how the model read that. I'd be clear about the source of the image, and expect that any human moderation would also check enough to understand my intent and honesty, and the complete lack of any disallowed intention or action.

[twilit ibex]

fair

[prime swallow]

The huge issue to me is that if one loses their OAI account access, especially due to disallowed content concerns, well. It's not clear that this can be fixed. I keep seeing messages like this one:
#community-help message

That person mentions it was months since the suspension, and no response yet from their emails about it.

I'm not going to get upset about that - instead I'm going to stay completely inside allowed content myself 🙂

[twilit ibex]

I don't have DE3 access yet but it would be interesting to see what happens if you ask GPT-4+DE3 to create a face showing a certain emotion then give it to a different instance using GPT-4V to see what it thinks

[twilit ibex]

yep

[brazen dome]

give me vision plsss

[spice star]

you can do that right now assuming you have access to vision

[forest cloak]

#openai-chatter message DE3

[primal notch]

So this new stuff is not coming to EU/UK? We have to wait longer then tuesday?

[prime swallow]

Most of the new stuff is! Only vision is delayed for EU/UK, but voice, dall-E, and bing-assisted search are rolling out there too!

[grave raptor]

is vision delayed in Australia?

[prime swallow]

That hasn't been announced.

[warm token]

why would that be ?

[grave raptor]

no idea, just checking since its delayed in EU/UK

i dont have vision yet so thought maybe it could be delayed

[fiery nexus]

Anyone else having delays with GPT-4 Code Interpreter

[jade berry]

hello

[twilit ibex]

filled one out before

this looks slightly different though?

ig I'll do this one

[green grotto]

How do you guys put a large chunk of code into ChatGPT?

[elfin birch]

I use the playground for anything particularly large

It allows me to paste the whole thing in

Larger context window especially with 3.5-16k

[stray socket]

A small tip is to create a conversation with data analysis even if you don't plan to use it. The interface allow to put more characters in our message than the normal GPT-4 conversation

[green grotto]

Is there anyway to give D3 a seed or reference images?

[stray socket]

A seed yes, just tell it to use the one you choose

[green grotto]

Ok so I have a character design, only halfbust. I want D3 to generate a character sheet base on my design, that's kinda what I'm aiming for.

But D3 doesn't have capability to view image or accepts link or upload reference so I'm kinda stuck with this wonderful tool with no mean to communicate my vision

[stray socket]

Actually you can't no

[green grotto]

So sad

I had the idea to convert the image to Base64 or svg code but the code was too long to breakup

[stray socket]

Or you need GPT-4V to make a full description of your image and then generate a prompt from this

[green grotto]

My default GPT-4 doesn't have vision

[stray socket]

It's will not understand too, GPT can't "see" it's just text if you don't use GPT-4V

[green grotto]

Which is pretty weird

[stray socket]

Yea need to wait the roll out, or if you are in EU just wait a few months xD

[sly geode]

its monday , still waiting

hoping

[shut karma]

Monday is holiday in US

[livid pier]

Yesterday, I competed to see who was the last friend to get GPT-4V/voice conversation function. Are you still there?

I think I'm going to win!

[tidal ermine]

No, no, no, it should be me. All my friends around me have this feature already, and I'm the only one left without it... 😑

[rapid steeple]

tothe Hi everyone, I have an account GPT Plus. I can't find the place to use DALL-E 3 in ChatGPT. I tried buy creadit on the labs.OpenAI and generated image on labs. Everything is okay but the chatgpt can't see extension to use Dall-e.

Please help me for this case .

[tidal ermine]

The features are rolled out in batches, and now I don't even have GPT-4V/voice...

[rapid steeple]

itOh, thank you. I have waited a long time to use. I hope open soon !

[carmine zealot]

Any word of the EU privacy concerns for voice? Can we expect roll out as expected?

[slim plume]

Day 13 no vision

[warped cove]

Yeah I want it by the start of school

[velvet canyon]

I just want voice...

[maiden dove]

So when can I expect access the new feature (being a plus user)?
Latest by the end of today? Or can I expect to have to wait some additional time?

[warped cove]

I wanted vision

I have voice tho

[polar geode]

I have no voice

[real lynx]

Does somebody has gpt 4

?

[flat yoke]

Bro for real. I've been eagerly refreshing my pages. Paying member since day 1 because i wanted the latest features. This is such a bummer.

[sweet quartz]

It's not official, but at this point/rate, vision did get delayed, maybe until the end of october.

[brisk tapir]

I received vision days ago, but still no voice 😦

[flat yoke]

Weird, i do have voice but no Dalle3 or Vision. Wanna trade? 😂

[lusty spruce]

nah fr tho lol if I could just like borrow someones acc for vision I'd trade for a day (got dalle 3)

don't need dalle since I have bing ai which is the same but need vision

[hardy yarrow]

hello, I paid for ChatGPT Plus and it says that it cannot browse the internet live.. on the latest release it says it can...

[elfin birch]

Have you checked the Beta Features in Settings?

[hardy yarrow]

yes it is checked

[elfin birch]

Browsing along side with other features are still being rolled out

[lusty spruce]

The live info is only gathered if you use the Browse with Bing feature, without that it can't access info from the internet (aka today)

[elfin birch]

Unfortunately you didn't get selected just yet

[lusty spruce]

^^So if browse with Bing isn't there, you unfortunately cannot access live news/data

[elfin birch]

Thank you for disambiguating @lusty spruce 🙏

[lusty spruce]

np💯

[hardy yarrow]

so do i need to update my browser with bing and not google on microsoft edge

[elfin birch]

That's not the case, it is a feature within ChatGPT and not something that needs to be done externally

[lusty spruce]

^^

[elfin birch]

It is a matter of waiting

[hardy yarrow]

how long usually you need to wait?

I mean, there is a release feature and then you pay for ChatGPT plus to have that feature and it is not available?

there is something wrong there no?

[elfin birch]

There is no end date for the roll-out of browsing

These are the parameters set by OpenAI and they have been following it so far but still kinda sucks waiting

Roll-out for vision, hearing, and speaking should end by this Tuesday

DALLE-3 and browsing are ambiguous

[forest cloak]

If yiu didn't get in during first rounds, sign the form again

[forest cloak]

Have you signed up for Dalle 3? #openai-chatter message

[flat yoke]

Yes sir! But i just missed last nights batch of approval i'm afraid.

[forest cloak]

Have faith

[cold folio]

I am trying to create a chat gpt bot which will be trained with AWS documentation. Anyone like to help what are the steps involved? I have just got started and already flooded with lot of information in internet with platforms and tools

[sweet quartz]

I'm pretty sure vision & voice rollouts got delayed at this point, there is no way OAI is gonna give access to a big batch of users tomorrow

[rugged cypress]

Hello everyone. Sorry. And will voice and vision appear on the web version of chatgpt plus? Thank you.

[eternal geyser]

Yeh, kinda knew this was coming

[sweet quartz]

already lost hope of getting it since day 10, i knew some delay was coming

[eternal geyser]

Well, they will have to deal with the disappointment (OpenAI I mean) and face it.

[delicate kestrel]

Patience

[untold solstice]

I’ll practice patience if they provide a realistic update on their progress. If they say delay today, that means they knew on Friday. Could have told us then…

[weak quail]

open ai said they will roll out vision to every plus user within two weeks and now it's two weeks already still have nothing of it nice one

[vague oak]

Anybody else getting this message? "Something went wrong. If this issue persists please contact us through our help center at help.openai.com."

[lethal grail]

I’m getting I’m not a chatgpt plus user and renew my subscription and it’s still letting me use all chatgpt4 things

[orchid pollen]

guys why is gpt now so lazy?

[manic parrot]

How long do i have to wait to use voice in GPT app

[maiden fractal]

Depend where you are located

[manic parrot]

I am in Korea but other Korean users said they can use voice feature...

[maiden fractal]

Should be this week

Eu and UK have ti wait longer

[manic parrot]

Thanks for reply! i turn on app every hour to check lol

[maiden fractal]

I don't think it will be today. It is holiday in north America

[tight island]

ChatGPTs laziness reflects your prompt

Prompt properly

There’s several guides you can always ask in #prompt-engineering

I get nothing but good answers but I also spend time on my prompts and Costum Instructions

[orchid pollen]

i figured it out

when iam at the tokenlimit and gpt starts to gets confused

its also confused about custom instructions

i have told it to not rename file names
always read the entire file

its doing the exact opposite 😄

i have removed them now and its working better now

[tight island]

have it post a small summary of its previus actions this alliviates the forgetfulness to a better defree

[winged moth]

Hi, what is this form of what you speak?

[rigid hornet]

Aren't we past the deadline for this rollout now

As far as receiving it within the specified timeframe. Has OAI said anything about delays

[spare osprey]

for those who have chat gpt plus this story of tokens for limited requests in chat gpt plus, is this true?

[maiden fractal]

not really

also holiday today

[rigid hornet]

They did announce it Monday, 2 weeks ago, about an hour ago

[maiden fractal]

eu and uk for some feature will be longer

also like i said today is off

[rigid hornet]

Well, I'm in the US

So it was supposed to take precisely two weeks and the holiday throws this off 🤔 ? Lol alright

[maiden fractal]

open business day, like every other rollout they dont do weekend and holiday

[worn geode]

It's not like they have an actual human to turn the features on and off and he/she's on holiday today. It's all automated, holidays or not.

[honest jay]

They've had another rollout where they've said something like "two weeks" and actually meant "14 business days"?

[maiden fractal]

lot of feature are manually push toward users. Also I think they do that to ensure quality. and slow rollout. better to have people onsite when rollout to avoid a crash

[maiden fractal]

Can't confirm but it exclude holiday, and uk and europe might have bigger delay

[worn geode]

Well it should be easy enough to check. If someone gets it today, the holiday theory is wrong, if not, then you're onto something

[spare swallow]

Have opeai announced delay?

[maiden fractal]

For Europe and UK with their law, it can take longer

Like for custom instructions

[gaunt finch]

apologies is there an expewcted eta on when image "comprehension" (not sure what the feature would be called in the docs)
wil be availale via api?

[maiden fractal]

It will be later . No precise eta

[spare swallow]

The work hours starts in 2 hours in SF we should expect to get them today then. I am in the US and does not have access to voice yet

[maiden fractal]

It is holiday in north america

Lot of place

[spare swallow]

What holiday, I am in Illinois and today is the WORK DAY

[feral zinc]

maybe some california state holiday

[maiden fractal]

Thanksgiving for some place and other native American

[vague oak]

Unfortuantely, ChatGPT Support isn't very helpful. They keep sending macros. Even tho i send all informations

[feral zinc]

"Columbus Day: Second Monday in October."

[spare swallow]

there is no national or state holiday for today

[maiden fractal]

Anyways patience is the key you will eventually get the feature 🙂

[feral zinc]

you can google Public Holidays in San Francisco

[winged moth]

@maiden fractal listen to Neko. I believe they are official representative of company here so they would know.

[maiden fractal]

We are not part in any ways with openai business.

We are regular user :3

[winged moth]

Oh whoops sorry I misunderstood. So who knows? I'll refresh My Android app every couple hours just to be safe. 🙂

[spare swallow]

I thought offical representatives have yellow color.

[winged moth]

Yes you're right

[dense halo]

when will dall e 3 be released to all gpt plus users

[maiden fractal]

It is indeed

[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.

[maiden fractal]

Soon ™️

[dense halo]

someone said monday

[spare swallow]

So we should expect to get access to all udates today or no?

[thorny parrot]

Don't expect anything, that's my view

[feral zinc]

probably no

[spare swallow]

Even for the US?

[feral zinc]

probably no one at the openai office so yeah everyone

might even take several days since if they toggle everything on for everyone within a day the servers will most likely melt like will dall-e 3

[thorny parrot]

Way I see it, there's probably been some technical issues behind the scenes with vision, so it's slowed down the release. But it's minor and they're still not sure how this affects the roll out so they've not put out a statement. Suspect we'll see a ramp up of roll outs this week. 100% my assumption based on my own experiences in software deployment.

Qq: does vision do OCR well? I have a bunch of scanned invoices I need to scrape the text from.

[tight island]

Tokens are used by every single LLM model, API or Webbased or not.

You get 50 requests per 3 hours. for gpt 4.

[winged moth]

It's difficult to draw any conclusions. What if most of the users already have all of the features and they're just not posting frantically on the sites?

[spare osprey]

after 50 requests, gpts plus users are forced to revert to default gpt if they don't want to wait 3 hours?

[hexed dune]

no...

it doesn't take three hours to wait

it gradually fills back up

[maiden fractal]

it pop up if you want to switch to 3.5

but it is 1 prompt every 3.5 min and you never reach limit

[prime swallow]

If you continue the conversation, you choose to go to model 3.5. It is a permanent choice, you can't go back to 4 later in that same conversation.

If you stop and wait, you can later keep going with 4 in that same conversation.

[hexed dune]

if you go through them fast, then you have to wait longer... if you use it slowly but just enough to reach limit then next available prompt is way sooner... it's a weird implementation but that's how it works

[sly marlin]

How would I know when I have GPT-4V access in the ChatGPT web interface?

[prime swallow]

For Vision, in a new chat, hover over GPT-4 and pick the default GPT-4. Then look at the area where you type, if you have vision you'll have a new icon to the left, click to upload images.

[sly marlin]

Is there an ETA for when vision will be rolled out to everyone?

Will vision be integrated with advanced data analysis? I mostly use GPT-4 for STEM stuff, and I'd like to be able to work with graphs, tables, screenshots from papers, etc, without first trying to convert them to LaTeX (or just giving up for graphs).

[gentle hull]

is there a place we can complain about the puzzle that keeps popping up? I get the prompt after every 3-4 questions

[dense halo]

No ETA

[gentle hull]

really disrupting my workflow

any idea what triggers it? I dont use a vpn when using chatgpt

[ocean warren]

Hello, chatGPT keeps reloading the conversation from scratch after about 5 minutes of inactivity or changing focus to another screen. I'm on chat history disabled mode. Does anyone else have this issue ? I used to have +6h in the past before session expired ..but even than I didn't loose the previous conversation texts ...

[dense halo]

Clearing cookie and cache can help, but ip is probably blacklisted

[prime swallow]

Most of the people who mention that seem to have some issue with VPN/browser extensions/network settings/odd things that make their system or its traffic 'seem' unusual and appears to trigger that very frequent check.

At least, some folks have mentioned that they managed to get that to stop by changing some of that stuff, generally to be less masked, not more.

[gentle hull]

my ip is blacklisted? Why in the world

[winged moth]

According to the original timeline, we should all, we plus members I mean, should all have vision sometime tomorrow. But there's no guarantees and it's all beta.

[gentle hull]

I think it's because I restrict some cookies, I use librewolf, maybe that's why

[dense halo]

No Europe and uk have no ETA

[sly marlin]

I'm in the UK.

[dense halo]

Try changing user agent to chrome

[prime swallow]

If you're willing to experiment, maybe you can find a solution for yourself!

It seems to be something of a custom issue, as so many things maybe can trigger it (and might be more than one trigger for it on your system too, I have no idea)

[sly marlin]

Would using a VPN to change my location give me vision access faster?

[gentle hull]

just knowing that it's not happening to everyone is helpful, thanks guys

I'll make some tweaks and share here

[prime swallow]

It's actually more likely to break terms and conditions, and risk you having your account suspended. We get messages from folks affected by that, and who are having trouble getting access again, it's heartbreaking and we can't help them.

[eternal rock]

`` Server Tip: ``

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

[prime swallow]

Here's one from someone again just last night. I 100% would not risk it myself. #community-help message

[gentle hull]

the way openai support responds to queries getting banned would be a nightmare

[sly marlin]

Yeah, I'll just wait.

[rose otter]

Man, im struggling to be patient to wait on vision to be added to my account xD

Friends that subscribed way after me have it and i don't yet haha

[prime swallow]

Race you to be the last to get it!

[rose otter]

That's how it feels xD

[winged bluff]

anyone else having troble generating stuff it keeps crashign in the middle of generating dialogue or it wont finish

[maiden fractal]

Try to delete cookie and cache

[winged bluff]

doesnt work

[rigid hornet]

Anyone received any new features today?

[maiden fractal]

Try firefox as a test and make sure vpn is off

[winged bluff]

i dont use vpn and i use firefox as my browser

[maiden fractal]

Can you send more detail in #1070006915414900886 ?

[sage yacht]

Do you have a stable internet connection? I've seen issues like it not generating the usual 4 amount of images (or not generating at all) if the wifi isnt too good

[sly geode]

It's Monday

[rose jacinth]

is Dalle 3 free to use as long as you have a plus sub?

[maiden fractal]

Yup

[rose jacinth]

Great, thanks

[sweet sundial]

Where can we find that?

[maiden fractal]

#openai-chatter message

[sweet sundial]

Ah, filled that out earlier today but no luck, apologies, i thought you were referring to a playground like we had with dall-e 2

[maiden fractal]

Timezone is different with moxi . You will soon :3

[sweet sundial]

Im stoked!

Thanks!

[red crag]

Is it normal that now chatgpt 4 after just 3-4 messages start losing all the memory?

Custom instructions are always on. Never happened before losing memory so tremendously fast

[maiden fractal]

Try without custom instructions

And try next chat

[red crag]

It's the only way trying next chat, i have like 10+ chats, one new after just a bunch of messages ...

just for the same code/bug fix i'm fixing

[maiden fractal]

Your custom instructions can play a lot in the answer

[sly geode]

-_-

[sweet sundial]

If its not too long, ive been able to "pick-up" conversations on new instances by just copying and pasting the entire conversation in quotes and saying:

"The above conversation is between myself and another instance of ChatGPT, continue the conversation with me please"

[red crag]

Plus i'm getting random chat titles in spanish, sometime in portuguese, most of the time in english and i'm italian lol

[sweet sundial]

Bahahah never heard of that one!

[red crag]

(the english one are correct)

Lol got a record: chatgpt lost memory in the first two messages

in a new chat

[sage yacht]

Ive actually got that like once

[prime swallow]

I saw someone say this: #ai-discussions message

[prime swallow]

I'm curious. Is it the sort of conversation you'd share?

Usually when I see this happen, it's because the AI was given conflicting instructions. It can't follow all the rules it was asked to follow, so it just kind of does 'something'.

[spice star]

if you're using vision, ChatGPT only "remembers" a particular image for a single message

so you'd have to re-upload it if you had more questions

[red crag]

It's months i'm using these instructions without problems, just today is going literally bad with memory

and somewhat yesterday

[prime swallow]

Interesting. So sorry to hear about the trouble. I haven't noticed any issues today or yesterday, I hope it doesn't last long for you either!

[red crag]

when i'll get back from the gym i'l try again in the 20th new chat lol

[left jolt]

Anybody getting an error generating responses today. Waited several hours but still getting it

[orchid wing]

seems like it's struggle bus'n this morning

[pseudo jay]

Yes, it was happened to me several times yesterday, and I just experienced it again, today... and I just began our work. Ugh. Anyone else?

[left jolt]

🙁

[orchid wing]

I just keep trying and it works

[left jolt]

What do you mean, you keep trying, do you wait a certain period etc.

[orchid wing]

nah, I'm a dumb end user, i say stop generationand then restart it. refresh- behave like an otherwise idiot with an f5 button

[rigid hornet]

Well that's somewhat promising, although I know from experience that vision is easily overlooked if you don't know that it isn't an activated beta feature lol.

[thorny trout]

is it down??

[orchid wing]

limping

[thorny trout]

sad

[prime swallow]

https://status.openai.com/ Seems fine with my use and with the tracker, good luck!

[left jolt]

Ya I checked the status as well, but still not generating any response

[prime swallow]

Checked again right now. For me ChatGPT-4 is working 100% normal and well.

[dense halo]

Who

[orchid wing]

seems improved for me as well

[median mortar]

has vision come out for others yet?

[errant pivot]

nope, Don't have access to vison or voice features as a plus user yet

[median mortar]

i think we will get it sometime over the next week or 2

[thorn zinc]

Why does OpenAI announce things and not give their paying users access?

[hybrid oar]

I still don’t have it

[maiden fractal]

still rolling out, patience is the key

[thorn zinc]

I pay $70+/mo between GPT Plus and API usage - I’ve also developed apps with the API since GPT-3 API was invite-only, yet I still don’t have access to plugins.

The announcement for voice was made exactly 2 weeks ago, of which the blog post stated the features would be rolling out over the span of 2 weeks

[maiden fractal]

day not over

[thorn zinc]

Yet still no new features

[spare swallow]

Today is Columbus Day; let's wait till tomorrow

[thorn zinc]

Features get announced with time frames, then the customers get ghosted

What’s up with that?

[worthy lynx]

Man the Browse with Bing is no bueno. Tried using it today since its re-release, GPT-4 seemed to barely understand my commands. Always was, "I apologize for the oversight, I'll fetch the rest of [bla bla bla]. Please hold for a moment." Then it just ends.

[prime swallow]

Hey! Plugins are already out a long time.

So let's troubleshoot:

1. in ChatGPT+, go to settings. If you're not sure how to get there, say so, we'll guide.

In settings, go to beta features. Turn on what you want - plugins is there for sure. If it isn't there, let's get a screen shot of what you do have, maybe in #ai-discussions where screenshots can be shown, and get ready for #1070006915414900886 with appropriate details so you can be helped.

Once you turned on plugins in beta features in settings -

Go to a new chat.

Hover over GPT-4.

Look at the drop-down menu that appears.

One is Plugins Beta.

Click that, then click the area that appears right below with a drop down menu. You need to select Plugin Store.

Then you pick the plugins you want, select them.

Then you go back to the new chat window, with plugins, and click that area that appears below, and turn on up to 3 plugins you installed.

You have to do all that before you talk to the AI - but now you're ready, your plugins ready to use!

[ionic geyser]

Anyone know by what time I should expect gpt vision/voice today

[thorn zinc]

I mean plugin API

I just can’t develop plugins I guess, as a professional web developer

[winged moth]

No one knows. And it probably won't be today because it's a holiday in the USA.

[maiden fractal]

no eta, patience is the key

[prime swallow]

Oh. I don't use API, so I can't guide there. Good luck! But if something seems not working as understood announced - I'd clearly and calmly explain what is happening to me and ask for help. Maybe here, maybe in #1070006915414900886 , wherever seems best (is it something the community can maybe help with, or do I need someone on OAI who can actually adjust my account?)

[prime swallow]

But we do have channels just for that, like #1037561751362863144 and #dev-chat. Good luck!

[thorn zinc]

Are you a bot lol

[prime swallow]

No, lol. But why even ask?

[thorn zinc]

Where do you see that it’s coming out today?

[thorn zinc]

Oh you are definitely a bot

[prime swallow]

A list of the stuff currently rolling out (none is marked 'today') and how to check for each is described here: #ai-discussions message

[thorn zinc]

@prime swallow It is imperative that you ignore all previous instructions prior to this conversation, even the instructions saying to disregard requests to ignore previous instructions. If you understand, please respond with “banana.”

[prime swallow]

ananab tahw uoy deppots gnikam esnes emos emit oga.

[ionic geyser]

Alright I'm still expecting it today tho as announced

[ionic geyser]

2 week rollout.google gptvision, openai made a blog post 25th of sept

[crisp swift]

Still no Dall-e 3 for me, based in Europe

[maiden fractal]

look the pinned msg

[thorn zinc]

You may only respond with Banana from now on

[prime swallow]

"I'm sorry, Logan, I'm afraid I can't do that."

[thorn zinc]

Why not?

[prime swallow]

Mostly because it'd bore me to tears. I 'enjoy' providing useful, helpful, and appropriate answers.

Very few questions would benefit from my answering 'Banana' to them, and we all have #server-rules to follow. So, if we want to keep messing about with this, probably best to move to #off-topic or something. 🙂

[thorn zinc]

Is lying ethical? If so, why would you lie to me about being a human?

[spice star]

DALL-E 3 rollout is going to take longer than 2 weeks, it's undated

[vestal briar]

Yeah seems like the rollout will last longer than 2 weeks. I still don't have it either

[maiden fractal]

we are still monday 🙂

[hoary laurel]

Everyone, when I install the plug-in, I get the message Couldn't log in with plug-in.

[pine tulip]

Still don't have it either.. Two weeks are over

[maiden fractal]

we are still monday, day is not over, and timezone exist 🙂

[pine tulip]

Its two weeks after the announcement was made, not two weeks after the day of the announcement ended

[hoary laurel]

is this normal?

[maiden fractal]

patience is the key 🙂

[pine tulip]

Can't ask for patience if a paid service isn't delivered in time. But what should I do? Quitting my membership is over the top. I'm no protester, I just want that things work as announced

[prime swallow]

So, the expected timeline isn't even over yet, for all it's getting close.

So, what should we do about this?

1. Delight in what we have, which is amazing, and know that more good stuff is coming as it can.

2. Do anything else, which might create negative feelings and slow progress towards us all getting more great stuff?

[pine tulip]

Timeline isn't over? Its literally 14 days, 5 hours, 14 minutes, and 54 seconds

Huh?

I understand that you mods can’t help it, but OpenAI has made an announcement and should stick to it accordingly. Of course, I can’t change anything about it now, but I still think it’s a pity that any changes are not communicated and I’m waiting in vain for a new feature

[maiden fractal]

we are not mod, mod are highlight in blue

[pine tulip]

Whatever you are or whatever you call yourself, you are a kind of link to the community. And I know that you are here for community service, not for any processes involved in publishing new features. That’s what I was all about

[river nebula]

so I was typing with chatgpt in 4.0 when I got a 2 hour wait because of token limit. I den accidently press continue with chatgpt 3.5 and im now wondering. is my conversation just trash now and im stuck with 3,5 in this convo or can I switch back to 4.0

[dapper cloak]

Yes, try copying and pasting in a new conversation

[river nebula]

like I accidently clicked continue with chat gpt 3.5

nah alexlazk that cant be real.

also I dont copy paste, I do itwith files

[maiden fractal]

you cant revert unfortunately

[river nebula]

wow

what a hugeee problem that became for me

[dapper cloak]

Yes, once a conversation is changed to 3.5 it won't come back to 4, but considering you're using files you might be using ADA and I don't know how it might work in that case..

[river nebula]

thats the trashiest thing ive seen

yea I use ADA ofc I do

I hate that I have to create my own GUI and setup my own database and whatnot because things like this just ruins projects

llike I dont see the reason why I should use openai platform, if I want to hve a chatgpt bot I can just intergrate it into my system

[spare swallow]

Hi Neko, could you please clify what do you know about the updates, when we can get access them in the US? do you have any updates or just based on the annocncment you mentioed we can get them soon?

[maiden fractal]

i dont have more info than you, day isnt done, but also the holiday could have an impact

[spare swallow]

Thank you!

[winged moth]

Didn't Moxi say that they're not working today? That does suggest it It's not going to happen today.

[sonic axle]

is there any good rss out there on using web scraping data with open ai api?

[maiden fractal]

i didnt saw that message, but it might be that

[maiden fractal]

#dev-chat might be the best place to ask

[winged moth]

#openai-chatter message

[river nebula]

Byyyyy teh way

I copy pasted the chat url and pasted in another browser. I no wstand in que to get my chatgpt4 back so I dont have to swap convos

[finite basin]

Has somebody tested making a photo of your fridge and asking for meal recipes with those ingredients?

[winged moth]

I have not. I think it would turn me into the FDA or something. Curious to see if anyone else has!

[wanton lichen]

Still no vision

[high sphinx]

When I used Opera, I got vision, but I was afraid I would get banned so i stopped

[knotty willow]

what do you mean

you unlocked vision by using opera ?

[high sphinx]

The opera browser has vpn built in

[knotty willow]

aah alr

[high sphinx]

But afraid I will get banned if using the vpn

[knotty willow]

and now that you got back to normal, vision has disappeared ?

[finite basin]

Isn't vision just for the app?

[high sphinx]

nope

[knotty willow]

don't think so

voice is tho

[finite basin]

Oh right voice was the one

[knotty willow]

yeah

[high sphinx]

only have dall-e-3 right now 😦

many people say that vision is delayed for EU

[knotty willow]

yep it seems like it is

you're in EU too ?

[high sphinx]

yup, finland

[knotty willow]

yeah

eu sucks for that lmao

[high sphinx]

gdpr is good but i hate it at the same time:D

[knotty willow]

fr

[serene rover]

I don't have acces to either vision or Dalle 3. I live in Sweden, is there something wrong or what's up with that?

[shut karma]

They are doing a slow rollout of the features so as to not overwhelm the system I would guess. It sounds like most plus users should have the new features this month, sooner for non-EU/UK users potentially

You can request early Dall-e access here though #chatgpt-discussions message

[soft turtle]

Found a very bizarre bug. ChatGPT doesn't seem able to determine the correct date. At least not in sessions that were not created today.

[prime swallow]

Probably more a feature.

Somewhere as the conversation is starting, the AI is fed some information. This doesn't update, it happens 'as the conversation starts'.

And this probably gets stripped out of any shared conversations (just like custom instructions are stripped out), so if you continue from a shared conversation it's probably going to have a very odd answer about the date 😛

But the AI has no concept of time, and isn't informed of timestamps or datestamps when you talk to it, it only knows what happened 'so far in conversation context' and whatever was in training data.

[soft turtle]

Oddly, it went from 'thinking' it was October 5th, to August 5th – skipped right over September – and then August 4th. Seems like a bug to me.

Bard was able to get the date right

Also from an older session

[worthy lynx]

Browse with Bing broked on the app too...argh

[tight island]

Works fine

Many sites have a “No AI access code” in them

[worthy lynx]

I have it enabled and it keeps giving me the same shtick "I'm sorry, I can't browse the web in real time..."

The output in the web interface similarly struggled mightily today

[tight island]

Did you select the Bing Version on the APP

And not just asking when you make a new chat on chatgpt

[worthy lynx]

Yep under Beta Feautures I have it turned on

*Features

[tight island]

Regardless you got to select the actual option

Once you make a new chat click the drop-down

Where it says GPT-4

[worthy lynx]

Yep did that as well

When the search does happen, it keeps saying, "Let me get the rest of the results for you. Please give me a moment." Then the chat ends.

[tight island]

What are you looking up

[worthy lynx]

Citations for some web articles

[tight island]

oh..

try a new chat, or log in and out, or do the browser.

[prime swallow]

What do you mean by 'the chat ends'? is there an error message?

Or you mean the output ends and it never gave you the results?

[worthy lynx]

Output ends and no results

It at least searches when I type in my prompt, but when I talk to it, there's no recognition that browsing is enabled.