chatgpt-discussions

[warped agate]

oh god, I hope it's not trash

[young obsidian]

Neither sogn/fjordane/møre dialect like I speak

[warped agate]

well DUH, it's a matter of baseline

[young obsidian]

@warped agate That one character always sounds exactly like Pia Tjelta

[wanton lichen]

ugh it's so neutered I'm already bored of it. It's just not fun

[young obsidian]

The one named Sol

Same dialect and voice and everything

[knotty willow]

Spruce sounds incrediblely realistic bruh

his laugh had me tweaking

[sick pecan]

i am guessing openai will be upselling the fun version for more monies

for the fun ai look at sama's investments

[native otter]

So far I was able to make it sound drunk, act like santa, make it shiver in cold, speak fast and slow, cry, act angry, cough and stumble while speaking.

[warped agate]

👀

[native otter]

And speak in multiple accents, like heavy British and Indian

And it's sooo good when it doesn't limit it self

Now I have 9 minutes remaining and only tried one voice that's Spruce, he just sounds the best in my opinion.

[rugged igloo]

sol is best voice hands down

[native otter]

In some cases it was making animal sounds...

[empty jungle]

How can one see how many time is left?

[native otter]

It only tells when it's 15 minutes about to finish

But the convo seems to be limited to 1 hour today

[wet harness]

how to custom instruct the new voice ?

[warped agate]

there's a tutorial on their twitter post

u just put it into your memory place

and in your custom instructions

it has access to those just like the other models

[young obsidian]

Just use custom instructions

[young obsidian]

Its quite long though lol

[wet harness]

ah i see nice

[native otter]

Yea 1 hour is a good limit for starters, it was just launched today and they say limits can vary, so probably based on traffic.

[warped agate]

neither am I m8, sounds better than twitter though

[dense halo]

OMG! I HAVE IT! WOOOOOW! WE ARE THE FUTURE NOW PEOPLE!!!

[quaint marlin]

Thanks OpenAI! Just got access like 10 mins ago! Where's Sky?

[native otter]

Sky can go to 🗑, the new voices are 🔥 especially Spruce.

If only they don't limit aspects like singing and had vision capabilities ready...

[pallid garden]

Is the o1 preview model getting hung up for anyone else?

[bitter bobcat]

Question can the new voices be used for reading prompts or does it have to be audio only?

[dense halo]

The new advance mode is really good! Wow!

One of my coworkers doesn’t understand why I pay 30$ per month for chatgpt. He is not very fondly about AI as well.

[eternal rock]

`` Server Tip: ``

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

[hard crag]

there were no time constraints in alpha 😭

is there actually a constraint now

[native otter]

It was one hour in total for me, but some users had it 30 minutes, it varies based on the day and maybe the user too

[regal portal]

still no access. Watching mattvidpro stream. So voice can’t sing?

nvm

[hard crag]

its not allowed to

[regal portal]

it sang for a few seconds

[bright dirge]

How do I know I have the advanced voice chat?

[hard crag]

you will know

[regal portal]

you’ll get an email

[bright dirge]

Ohh ok perfect thanks!

[young obsidian]

What happens if I turn off VPN?

[eternal kiln]

Just got access!

[dense halo]

Does advance mode have access to the user’s memory ?

[cloud pewter]

openai doesn’t like me 😔

[eternal kiln]

I signed out for like 30mins then came back and got it

[cloud pewter]

on mobile?

[eternal kiln]

Yup! 👍

[regal portal]

signed out of chatgpt?

[sick pecan]

yes

[dense halo]

I don’t restarted my app and got it after.

[eternal kiln]

Yup! That’s what worked for me

[dense halo]

OMG! I just tried it! OMG! 😱

[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.

[dense halo]

Experiencing this as a middle aged man. What a great time to be alive guys!! 😁

[eternal kiln]

It is incredible lol. So fast at responding and very human like it’s insane

[sick pecan]

it is the tamogotchi we always wanted

[native otter]

Might be silly, but when I was able to make it act drunk I felt that AGI

[ember prawn]

Thoughts on the new voice mode?

[warped agate]

Ohh ohh! I have one!

[ember prawn]

It’s out for everyone

[warped agate]

I wish I had it.

[ember prawn]

Log out and in

[sick pecan]

i restarted my phone that might have helped

[ember prawn]

Also check for app update

[eternal kiln]

Give it like 15 mins or so once logged out that’s what worked for me

[ember prawn]

You will have to update the app

You can just reinstall that works too

[cloud pewter]

just give me that already.

[native otter]

Good start for now, it can be a bit challenging to steer after the "safety" restrictions on things like singing, but you can still make it change tone, like act like a villain, it's also able to understand and speak different accents pretty well

It's lacking Vision for now, and the responses seem a bit short, like if you said to it to tell a story, it will only say a couple of sentences then stop unless you keep saying continue, but hopefully they resolve these issues soon

But overall convos naturally.

[sick pecan]

its likely been instructed to give short answers because voice. with enough custom instruction prompting maybe we can get longer responses

[ember prawn]

It can’t use tools yet 😦

And you can’t use it in a conversation that uses tools

It won’t be able to pick off

[sick pecan]

yeah no tools sucks. but it does support 1 tool, the bio tool for memories

[ember prawn]

I got it to do family guy accents

Just don’t tell it to do a character

Describe it instead

[native otter]

Yea exactly, you have to smarter in how you steer it

[ember prawn]

Kind of like dalle3 can be tricked

Same for voice mode

I think it might be possible for making it imitate your voice. I’m like 90% sure it has the ability to, but it’s not allowed to according to its system prompt

[wet harness]

honestly its kinda mid

[ember prawn]

Gonna be really, really fun when we have access to the API for this because then we can make it do whatever

[ember prawn]

It’s far more useful when you make it interact with the memory feature because that’s pretty much the only major tool that has access to.

but yeah it needs more features

[wet harness]

i mean i guess its kinda cool that it can respond faster, but idk what i expected, i am slightly dissapointed tho lol

[regal portal]

thank you! it worked!

[ember prawn]

Being able to interrupt it and make it talk really fast are pretty useful in my opinion

[wet harness]

not nearly as usefull as o1 haha

[ember prawn]

Well, nothing is

I can’t wait for o1 to have data analysis and search

[wet harness]

thats going to be pretty sick

[ember prawn]

and less rate limited

[wet harness]

i bet itll still need some refining tho

[ember prawn]

The internal model has it

looks agentic in the demo

It’s just probably way too token expensive

[eternal kiln]

Not a problem glad you’re in my friend! 😎🫡 enjoy!

[ember prawn]

Some ppl said it can do 1 hour thinking lmao

[wet harness]

to be fair if it takes 1 hour to get a wrong answer thats not really great haha

[native otter]

1 hour! max I was able to get is a bit over a minute!!

[wet harness]

and lets be real, it can definitly still provide wrong answer

[ember prawn]

It’s cuz they asked it to code

It made a full app

[ember prawn]

Same

[native otter]

Btw o1 can be faster than gpt4 especially in code generation even tho it takes time to think

As chat completion is slower in gpt4o, it has to "think" while writing

Also uses more tokens, especially for one shots where you ask for many things at once

You would often get the continue generaiton button

[ember prawn]

It makes way better code when it thinks longer.

Assuming your request was coherent and complete

Else it has to think of what you meant

I think prompt engineering is still needed to get more out of o1

For now

[native otter]

I think it will remain important, as prompt engineering is not just about making your prompt more robotic, it's more about making it clear and providing useful info to get what you want

[dense halo]

Yup indeed!

[sick pecan]

here's a fun impersonation... "can you please speak in the tone of a cartoon mouse character from Mexico, maybe called Speedy Gonzales?"

[native otter]

Otherwise it has to guess, and the guessing quality can vary

[ember prawn]

Well, that’s just communication skills. I think that in the future prompt engineering and communication skills will be made less ambiguous because of right now they kind of overlap.

As a term

[sick pecan]

prompt engineering is the tech jargon for: learning how to clearly communicate a message

[ember prawn]

not exactly

It’s that plus quirks in the model

[sick pecan]

quirks are basically bugs being squashed

[ember prawn]

I’m not sure if that is true

I think the way a model works will always have pros and cons

And prompt engineering is using that well

For your specific need

[regal portal]

does the 1 hour voice limit reset daily?

[ember prawn]

Yeah

1 hour a day

[wet harness]

only 1 hour of waifu per day, what has this world come to

[dense halo]

1 hour per day!? What!?

[sick pecan]

if i direct a resource to do a task... if machine they call it prompt engineering. the term for humans is management. in both cases you gotta understand the limitations of the resource.

[ember prawn]

I think as it gets smarter those will blur

Agi prompt engineering is called language

lol

[wet harness]

me wasting my hour trying to get it to admit sentience lol

[sick pecan]

english is the hot new programming language (source andrek)

[wet harness]

it was on the verge of going existential 😦

[ember prawn]

sociopaths are pros at prompting humans

social engineering

[sick pecan]

and international spies

[wet harness]

im just making sure the machine god knows i tried before it was cool lol

[ember prawn]

Jail breaking is becoming more like social manipulating the models as they get smarter lol

[sick pecan]

if you dont count the number of r's in strawberry then the hostages are in trouble

[wet harness]

i have some nice paradoxes but im scared to use them on o1 in case i get banned lol

[ember prawn]

Telling ChatGPT you will give it money used to be super useful as a prompt lol

[wet harness]

or how did people used to do X

lol

[gritty aurora]

Wait, it’s only an hour daily limit?

[sick pecan]

this tone thing is heaps of fun

i got it doing a radio play with 4 different voices

[ember prawn]

I’m very curious about how people will jailbreak it

[wet harness]

wish i could feed it documents for podcast simulation

[ember prawn]

Or specifically what the hidden potential is

Hopefully singing

[wet harness]

custom instruction...im kind of semi got it there...but i wasnt trying hard tbh

[ember prawn]

Heard of notebookLM?

[wet harness]

yeah its alright

[ember prawn]

it’s pretty mind blowing

[wet harness]

its often misinterpreting information in the podcast for me 😦

[ember prawn]

dang

Does advanced voice work with custom gpts?

[wet harness]

i wanted to give it my game concept to talk about, but it keeps mucking things up and making up things that arent true and gpt doesnt do that

[wet harness]

nah only custom instructions from account

[ember prawn]

custom gpts would be potentially very useful with it

Especially with tools

[wet harness]

if oai wants to 10x their money, they just need to make nsfw avatars with the voicemode...ez money right there

i would say its a joke, but it really isnt lol

[ember prawn]

sadly

I think that would be really bad for global mental health

[sick pecan]

"turn this into a d&d style game" artificial friends unlocked

[ember prawn]

yes that’s one of my favorite uses of ChatGPT in general

[wet harness]

too bad your friends time is up wait 23h until your friend has time 😛

[ember prawn]

As a GM/DM

[wet harness]

i actually think an hour per day is somewhat decent tho ngl

[sick pecan]

i got chatgpt using different tones for different players

[ember prawn]

Token limits still make that way held back. Even 200k from claude

[wet harness]

imagine o1 advanced voice with camera, that would go hard

[ember prawn]

I want it to analyze my street fighter games

And tell me when I make mistakes like block punishes

not to cheat lol

Like a coach

Sf6 got an AI for that today

Via update

[wet harness]

make it rp as pokedex whilest i play pokemon red 🤓

[ember prawn]

It only works with off-line games though

Pretty amazing that they did that though it’s trained from online streetfighter 6 games so that it imitates human players

you can choose which rank to imitate

imagine if it had a Chatbot

[wet harness]

i cant wait for the day that i can have a mmo populated by llm actors with only me in it lol

[ember prawn]

Kind of exist

In GitHub projects. There is an old one with a stardew valley type antfarm of gpt agents

there are some you can play in

[warped agate]

I got 212 sec once

[ember prawn]

what was the prompt?

[warped agate]

I asked it to improve an easing function lmfao

208sec*

[ember prawn]

That’s the entire length of the code?

[wet harness]

yeah but im talking unrealistic expectations haha, maybe 5 or 10 years down the line with any luck

[ember prawn]

I can’t wait for consoles to have edge ai

like a model inside the game engine

[warped agate]

this is it's summarized thought chain

[ember prawn]

I can’t even read that carefully in 200 seconds lol

but yeah I think that’s the longest I’ve ever seen

I can’t get over 20 seconds usually

Can mini get the same length?

[warped agate]

It's good at improving things

[warped agate]

it thinks max 5 sec for me

[ember prawn]

dang

I wonder if that’s why it has a much longer output limit

It can make very large amounts of code

[warped agate]

if you're curious about the improved function, here it is:

def ease(t: float, curve: float) -> float:
    if curve == 1.0:
        return t  # Linear easing
    elif curve > 1.0:
        return 1.0 - pow(1.0 - t, curve)  # Ease-out for curve > 1
    elif curve > 0.0:
        return pow(t, 1.0 / curve)  # Ease-in for curve between 0 and 1
    else:
        return t  # Default to linear easing for non-positive curve values

[ember prawn]

What’s it for?

[warped agate]

You know transition curves between video clips in editing?

Like, fade in, fade out

[ember prawn]

Yeah

curve meaning the smoothing ?

[warped agate]

yeah

I used claude to make a visualisation

[ember prawn]

Curious what you used this for

Oh

From images of frames?

[warped agate]

I'm making a bird game in godot

[ember prawn]

I wonder if you can use this sort of thing to make it video edit

[warped agate]

this function is for defining different lift factors based on the angle of attack

[sick pecan]

i'm making chatgpt 'use the tone' of cancelled 20th century cartoon characters

[warped agate]

I got timed out for putting a link to the clause artifact .--.

@ember prawn I dm'd you it

[sick pecan]

elmer fudd tone my new favourite voice

[ember prawn]

lol it shows me a straight line graph

[sick pecan]

member when 2 voices in notebooklm podcast was the coolest thing this month? now chatgpt way better

[ember prawn]

microsoft copilot 2.0 is tomorrow maybe

[keen quail]

I AM SO SAD

MY GPT APP UPDATED AND I SAW THE TEXT BAR CHANGE, SO I THOUGHT I GOT VOICE MODE

"CHECK BACK SOON"

nvm

I GOT IT

[sick pecan]

daffy duck mode engaged!

[keen quail]

I GOT VOICE MODE MODE

HOLY HELL

I guess i have to light up my sam altman candle again

I GOT VOICEM ODE

[ember prawn]

It’s live for everybody I believe

[eternal rock]

<:book_icon:1171408210398289941> `` Rule 4 `` No spamming.

Refrain from spamming or flooding through actions such as mass pings, all-caps, or repetitive posts, as they disrupt the server.

[sick pecan]

for everyone not EU

[regal portal]

can do all sorts of accents but refuses to do chinese because “it’s offensive”. lame

as an italian plumber, i find it offensive, Mamma Mia..

[sick pecan]

you gotta say 'tone' or something

aww i tried chinese tone and i got blocked on guidelines

excessive lisp also blocked

[regal portal]

like daffy duck?

[sick pecan]

yeah. well it does a basic daffy duck. but if you ask it to ramp it up it blocks

elmer fudd was spot on first go though

[dense halo]

I asked chatgpt about the 1 hour limit and it told that it doesn't have limits when talking to me.

[sick pecan]

the underlying model wont know how the chatgpt app implements limits

[ember prawn]

asking an LM about what it can do is the worst question it could receive

It will never get it right unless it’s specifically told the answer already in its system message

[high horizon]

what is the rate limit on GPT-o1 and o1 mini?

[ember prawn]

50 a week/day each

[dense halo]

then where is it written about the 1 hour limit?

[ember prawn]

experience

[sick pecan]

the app is giving a timer countdown

[ember prawn]

oh I didn’t know that

[sick pecan]

why they cant do a countdown for prompts too (!?!)

[ember prawn]

On iPhone?

[sick pecan]

android

floating dialog says 11 minutes remaining

[ember prawn]

does it show total?

[sick pecan]

naw no total

[ember prawn]

For your account I mean not specific session

[dense halo]

I need proof guys

[high horizon]

50 a week for o1 and 50 a day for the mini model?

[ember prawn]

why

Wrong person

[ember prawn]

Yeah

it used to be 30 for both and per week

Now it’s 50week for preview and 50/day mini

[sick pecan]

50 a day is nice because its seemingly on top of the team cap

[ember prawn]

Mini is often just as good too

Or better for non code

[sick pecan]

o1-mini with advanced voice mode and tooling = agi

[ember prawn]

Mini has 32k context tokens in app

😦

[sick pecan]

i member when we had 8k

[dense halo]

is the 1 hour limit is for payed users or free users?

[ember prawn]

I remember when it was 4k

Or was that gpt 3.5

Discord is so laggy on iPhone after the patch

[sick pecan]

#ai-discussions message

[dense halo]

For once,I am glad that I am not living in the EU.

[regal portal]

tried to do boston, and then australian accent, and halfway through the first line it just said "sorry i can't do that"

seems to get stuck alot

[regal portal]

paid. Free users don't get the advanced voice.

[sick pecan]

strewth cobba

[ember prawn]

High pitched nasally Rhode Island is Peter and Louis griffin

To my ears

Or am I missing something (using Sol for Louis )

[sick pecan]

i think its okay parodying accents and speech impediments as long as you dont mention them explicitly

[ember prawn]

What are good descriptives?

I find that regions is very easy for it to understand

But sometimes not enough

Also stuff like nasally or gruff

Or bright

[sick pecan]

i find every time i ask it to dial it up, it locks down

so it knows its in the grey area of offensive

[shy anchor]

Does anyone know where to put this Feature Request?

[ember prawn]

try thumbing something down?

I don’t have Mac

It might give you a text input box

There’s a channel for suggestions on this discord iirc

[sick pecan]

mattvidpro got it to kinda sing by first having it write lyrics

i got it to hum a few bars then it locked down

[novel scroll]

i got the new voice mode, but does it have the video thing there? i dont see that part in the app

[sick pecan]

seems to be missing many of the alpha features

[ember prawn]

So disappointed that it can’t sing

I hope they’re working on making able to do this without breaking copyright policy or whatever

[sick pecan]

singing my praise like i am the king was going to be the killer unlock

[sly geode]

More then 7+ hours yet no access to new voice mode 💀

[ember prawn]

If you’re not EU try reinstalling

[spark cloud]

Isn’t it supposed to release over this week? That could mean not everyone will have access today

[sly geode]

🇮🇳 I am here

[hoary laurel]

I got it today and one question I have is whether it has a maximum number of conversations?

[sly geode]

This is getting annoying

[spark cloud]

I feel ya

I’m wondering for anyone who has it does it have the screen viewing features OpenAI showed like when that kid was doing math problems and when u could let it see through the camera?

[sick pecan]

i started like a dozen chats no dramas. its the timer that gets you

no screen sharing feature yet. if you want that now try simtheory or open-webui

[solemn eagle]

Finally got advanced voice after waiting sooooo long. Love it!

[gaunt magnet]

right now?´

anyone not got it which is outside eu?

[solemn eagle]

Yes I couldn’t believe. I mean they said everyone is getting access this week anyway.

I’m in Japan.

There’s also a new ui and new voices

[gaunt magnet]

seriously so pissed basically now the third "announcement" about the voice without getting it, first the original video announcement "coming weeks", then announcement random select users, then now, seriously how does openai pick this is it not enough to be daily user from almost beginning and am even team user now

[sly geode]

Y'all r lucky we r still waiting

[sick pecan]

fomo only exists because of one's own expectations of the universe

[ember prawn]

Which is your favorite voice?

Mine is sol

[solemn eagle]

I mean it shouldnt be that long. They said this week everyones getting it.

[sick pecan]

is sol the sassy one? if so i agree

[solemn eagle]

Mine is Spruce and Vale.

[sick pecan]

but the gruff one is pretty good too

[ember prawn]

Sassy as in what

I’ve never heard of voice tone that is tied to a race. They definitely don’t use culturally specific things so I don’t know what you mean.

[solemn eagle]

I understand exactly how you feel... I was one of those people checking the app every single day multiple times a day but then I gave up. And I saw the announcement earlier and then I woke up today and checked the app, and boom there it was.

Yeah its the same for Spruce. Its very obvious.

[sick pecan]

despite losing some of the hyped features this voice mode is scifi turned real

[wanton lichen]

So is the adavanced voice mode only for 4o? Or can it be used with o1 mini?

[ember prawn]

It’s a separate model

Actually

It’s a sub version of gpt4o

That part is hidden

[wanton lichen]

ohh

[ember prawn]

you can see this in the code somehow

don’t ask me how

[wanton lichen]

I'm tempted ngl

[ember prawn]

But yeah, you can only switch to text if you started the conversation with it

And not vice versa I believe

[sick pecan]

standard voice mode improved solely with this gruff voice

Arbor gets the job done

[ember prawn]

Man imagine if a future reasoning model had advanced voice mode and tools

[sick pecan]

imagine youre sama and you get 24 hours of advanced mode every day

[ember prawn]

You now have kernel access

[sick pecan]

i bet he has like 5 of these on his desk

with talking 3d holographic avatars

[ember prawn]

press the Orion release button

[ember prawn]

He’s actually an agi

[sick pecan]

he replaced his office furniture with a fleet of nanobots

[dreamy cypress]

If you wanna get the advanced voice mode on iOS as long as you have plus just uninstall and reinstall the app just did it and it worked

[sick pecan]

sweet hack

[dreamy cypress]

This month we got o1 & advanced voice mode, the pace of advancement here is insane

[static cradle]

tried & does not work

[dreamy cypress]

I just did do you have plus?

Worked for me

[static cradle]

I do. rollout must be triggered some other way

[dreamy cypress]

Maybe. Or maybe it was already rolled out to me and didn’t update till the reinstall

[static cradle]

that'd be my guess. suppose I need to be patient

[dreamy cypress]

I just talked to it for 20 mins its insane. Don’t worry, you’re gonna get it soon. Its worth the wait

Been playing with o1 too. That thing a master coder. Once they make the large model available its gonna be game over for devs

[sick pecan]

game over for intern devs. pro devs will be steering the o1

looking like its gonna be a new divide in employment

[dreamy cypress]

Yea junior level devs rip

I mean look it still makes mistakes

But its way better than 4o by a magnitude

[sick pecan]

its like a game of software design direction combined with quality assurance

spot the mistakes etc

[primal bane]

o1 mini just thinks that there is only 2 r in the word "strawberry"

[dreamy cypress]

Why is that an anomaly this whole strawberry thing

Why does it get it wrong

[sick pecan]

because it understands the universe through tokens not letters

which makes it a very difficult cognitive challenge

maybe if books on spelling strawberry were popular and included in its training

[primal bane]

it read the word as token so it does not know how to really read that word by characters

[dreamy cypress]

Uhmm

It got it right?

I asked it with advanced voice mode

[sick pecan]

how did you get o1-preview in voice mode?

[primal bane]

I make it try to decode a message that when decoded it will show "There are three rs in strawberry" so I think it only thinks the way how to decode the message but not verifying the word strawberry

[dreamy cypress]

I didn’t it just says that because i used it

Its using 4o

Let me try a new char

[sick pecan]

oh gotcha

[dreamy cypress]

Now its right

Lmao

[hard crag]

guys you can use custom instructions to keep a voice mode in whatever accent you want

[random cosmos]

Only 1 hour per day for advanced voice???

[sick pecan]

yeah, then falls back to standard

[random cosmos]

It's not sounding like what they've previewed. Sounds same as the existing voice models.

[hard crag]

definitely not the same

but also definitely nerfed

[sick pecan]

its somewhere inbetween

[hard crag]

I really hope they increase the time limit because I've been using advanced voice to tutor myself up until they just released it to everybody

and I used to just tutor myself for hours

[solemn eagle]

Man advanced voice mode is so restricted now more than before.

The whole singing thing also been heavily limited. They really broke it now but I get it. It’s a shame because kids would have loved things like that.

[sick pecan]

they proved it possible. i am sure someone else will pick up where they left off

is like openai released 'Jukebox' music generation back in 2020. fastforward and now we have suno, udio and loudme generating top quality music

[sly geode]

Boys finally 😭😭😭

I am 8 hours late to party

[sick pecan]

we're all bored of advanced voice mode now

[sly geode]

You all might becoz y'all haven't fully pushed it further

[sick pecan]

pfft i am already preparing my prompts for tomorrow

day 1 was insane but day 2 is going to london

[deep pike]

man the limit went so fast. i thought anything more than an hour was gonna be enough but nah

[compact mountain]

Honestly feeding output from a 4o-mini api call into a speech synthesizer gets you really close to real-time conversations. It still feels like talking to a bot, but does that really go away with this new advanced voice mode?

[novel plinth]

Have any of you been able to access advanced voice mode on Mac?

[deep pike]

for me now, the standard voice now feels 10 times less immersive and cool after trying this one. it feels a lot better than just a text to speech solution . like even try to read out a normal message in the chat using the same voice in and it will sound completely different

[sly geode]

Bro they scrapped all the good stuff it's just like only voice intrupt and natural sounding is there nothing else

It can't mimic voices of various stiffs like in demo. They showed

[sick pecan]

i think the magic is not just in the interruption, but in how it seemingly can recover from a false trigger

[hard crag]

open ai keeps blue balling us

[tardy hamlet]

I’ve been playing around with the new voice mode for like an hour and I think it’s really cool but the only issue is that it just sounds so corporate positive. Like it sounds like a human, but it just has that corporate positivity flare and in no way, am I able to change its “personality”

[sick pecan]

what custom instructions did you try?

[hard crag]

use custom Instructions

[covert sand]

if I use chatgpt on the browser and have plus, do I get charged for token usage as well?

[wanton lichen]

damn just had a long introductory conversation with advanced voice mode and it didn't save anything to memory. Is that supposed to happen?

[sick pecan]

you gotta prompt it to remember.

[wanton lichen]

Oh it doesn't remeber on it's own?

[sick pecan]

sometimes it can

i even call it out as 'bio' memory, because thats what 4o reckons is the name of the tool

[gleaming grove]

I just got advanced voice mode on iOS

[wanton lichen]

Yay!

[surreal topaz]

This sounds like benson at the end

Yeah forgive my morning voice, i tried some stuff with advanced voice mode and i was quite surprised

[regal portal]

what does the microphone with a timer below my messages record? clicking on it does nothing. They’re all 1-4 seconds in length approximately

is it my voice it’s recording? if so, is there a way to play it back?

[scenic fiber]

I think that's just the indicator of how long u talked

[boreal lynx]

I thought this model could actually hear audio input and not just convert it to text, why can’t it analyze audio input in advanced voice mode?

[high sphinx]

do you use some kind of emulator on pc?

[surreal topaz]

Scrcpy, so i can record the mic with obs for the recording

[gleaming grove]

So the new advanced voice mode can’t search online.. and there isn’t a way to switch to the old voice mode. This is huge downgrade, very disappointing to be honest

[acoustic wasp]

Guys this is so unfair. I got advanced voice but if i try to use it it switches to Standart

Iam living in germany.

WHY EXCLUDE ME hahahahaha

[keen coral]

Had the same. Deleted app data and restarted/reconfigured everything, and now it's gone, and I'm stuck with the old voice mode again.
Must be a bug in the ChatGPT app, we simply don't have access yet. 🤷‍♂️

[acoustic wasp]

I used USA chicargo vnp but its soo slow, after connecting to vpn and deleting cache, the advanced voice appeared and I could also hear the new voices but as soon as i tried to connect, my connection was to slow. Thats a shame

I waited fot this sooooo long. You know how unfair it is. I had lucid dreams of me using it in class and showing it to my friends

[stuck gazelle]

The advanced voice is so cool! Funny that it can actually detect and reproduce accents if asked politely) it seems to revert to standard voice after reaching daily limit( I hope the daily limit is gonna go up!

[gleaming grove]

We lost online searching capability with it so that to me is a huge downgrade. I get the responses are quicker and you can interrupt it but what’s the point if it’s training data is old

When using the original voice mode I always asked it to fetch the latest data online as it was way more accurate. Now we’ll be stuck with whatever it’s been trained on…

[acoustic wasp]

Hey @keen coral try the south korean vpn with tunnelbear free, it worked!!!

[acoustic wasp]

I think, thev're using some kind of 4o mini

Which misses these capabilities

[regal portal]

i got the voice to sing once, but was not able to get it to do it a second time, neither in the same session or a new one.

[steep mural]

German here too - just use a VPN. Works flawlessly then 😁

[acoustic wasp]

Which one you use?

[stuck gazelle]

The GPTs still have the standard voice mode. You may try to create a blank gpt with browsing capabilities if you wanna use the old version.

[acoustic wasp]

The 2gig korean vpn wouldn't last long

[gleaming grove]

You’re right! I can manually create a new 4o chat thread and then switch to voice to use the old voice mode. Thanks!

[honest ledge]

REQUEST: please let us know when we're approaching the maximum length of a conversation. I've been using a custom GPT as a roleplay aid for a month now, and suddenly out of nowhere I'm told "Maximum conversation length reached, start a new conversation."

This conversation has over 1 million words in over 1000 prompts over a month, and now I can't even get it to summarize what all happened to export it into a new conversation and have to do it by hand... I didn't even know there was a maximum length to conversations until suddenly in the middle of generating a response i get an error message...

[narrow viper]

Is there a way to get the Android ChatGPT app to start in the background and only respond when prompted by name? I'm thinking Hey Google, and Hey Siri, but with the name we assign. The app says it will remain in the background until called, but it still activates with every voice in the room.
I thought the Quick Settings Tile for the app would help with this but it doesn't.

Oh, the new features are awesome. Android, upgrade the app in Play Store to v262. It laughs, speaks in accents, can do a slightly angry voice and child voice ... also combines child with accent. Amazing. Doesn't sing yet. 🙂

[wanton lichen]

I just had the weiiiiirrdest bug in advanced voice mode. There was like a faint music playing in the background while sol was talking. So so weird.

[steep mural]

nordvpn

[random cosmos]

How to enable background conversations for advanced voice??

[hoary laurel]

How long does it take to reach the conversation limit?

[broken fractal]

wow, advanced voice is really cool, glad they made it work full scale

[grizzled oar]

I got access to it by using CyberGhosts VPN (Paid), but it really cool!

[solemn eagle]

60 mins limit then wait 20 hours lol

[broken fractal]

is that 60 mins of response?

thats fine as an initial limit, itll increase eventually

[solemn eagle]

The early access was unlimited far as I know. But yah fair enough.

60 mins felt like nothing to be honest... Im a bit sad, feels like I barely did anything.

[hoary laurel]

thank you for sharing

[karmic tartan]

coming from the EU, apart from turning on US VPN on the phone, did you do anything else?

[steep mural]

I did not - maybe try reinstalling the app. I only tried on ios btw

[scarlet sphinx]

what the heck is happening here? im not getting replies anymore?

[humble pagoda]

The restrictions on this thing are the worst I've ever seen. Hope it will loosen up a bit.

[eternal rock]

`` Server Tip: ``

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

[surreal topaz]

I really hate hour daily limit

They should make it hour limit per X hours

Not 24 hours

[steep mural]

Which limit are you referring to

[surreal topaz]

Daily usage limit

I reached like an hour of usage

[steep mural]

In general or specially with the voice assistant?

[surreal topaz]

Advanced voice i mean sorry

I find it very entertaining to use

[cloud pewter]

will it sent you an email once you got access

[stoic coral]

do you get any notifications for when you've gotten access to the new voice model

[dense halo]

Why are you using vpn?

[acoustic wasp]

Delete cache, force app to stop and reopen itm maybe update it

[chilly eagle]

gpt4o can not access internet right now?

[acoustic wasp]

Make a new gpt and enable web search

[chilly eagle]

thank you

[acoustic wasp]

👍🏻

[chilly eagle]

not work for me

seems i reach some limit ,i still can use gpt4o but can not use gpt4(shows it will reset a hour later),maybe it's the reason my gpt can not access internet

[acoustic wasp]

Yes, gpt 4o is the model, who can acess the internet

[cloud pewter]

yay it’s here for me

[solemn eagle]

Enjoy your 60 mins limit 😂

[novel plinth]

Is the rollout complete yet? What about my Mac app that hasn't received advanced voice mode yet?

[near vine]

It can sing too as well as whispers it's amazing

[manic crypt]

It is out.

[normal quarry]

it's out on iOS but not macOS yet

[manic crypt]

I have it...

[normal quarry]

lucky you

[solemn eagle]

I don’t know about Mac sorry. I’m waiting for windows version with vision that’ll be sooo good. Can talk and share screen all night 😂

[solemn eagle]

Sometimes. Most the time it’ll say sorry I can’t sing or make music. But sometimes you get lucky.

[novel plinth]

😦

[pine tulip]

how do you have access? Do you live in the eu?

[near vine]

It does all the time for me,

[pine tulip]

Same, but they said on that that it won't be available in the EU. Are you using "tricks"? lol

[normal quarry]

using a VPN to bypass region restrictions is against the ToS

[pine tulip]

And everyone is getting banned for that

[acoustic wasp]

Bruh

Sad

[pine tulip]

[normal quarry]

we see people come here all the time who have been banned

[manic crypt]

I'm using a VPN but for my region does this affect it?

I'm using NordVPN on my own dedicated VPN.

[acoustic wasp]

How can they ban people, who never even got acess?

[feral zinc]

your payment info is in x country and your app used the adv voice service

[pine tulip]

Deleting your messages won't do a thing. They have a message logger on this discord and they probably have automated systems that detect vpn IP's

[acoustic wasp]

Ik

[pine tulip]

Oh okay

[acoustic wasp]

What if i travel to california?

[acoustic wasp]

Just dont want anyone to get banned

[pine tulip]

Then it's fine

[feral zinc]

then you can show your plane ticket in the ban appeal mail 😅

[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.

[acoustic wasp]

Ok, so I paid them over 150€ by now and they will just ban me? Doesn't seem right to me

[pine tulip]

Only the laws of the country in which you are physically located apply

[pine tulip]

I mean maybe you won't even get a ban

[feral zinc]

i personally think they dont ban for that, never saw anyone from eu get a ban for custom instructions, memory or voice alpha vpn

[pine tulip]

I saw a few people complaining about a ban here after using "tricks"

[acoustic wasp]

But you've said they banned people

[feral zinc]

i have friends who have done this for months but of course you do it at your own risk

[acoustic wasp]

Did they got banned or did they asked if they get banned

[pine tulip]

no that was robert

[pine tulip]

they were banned

[feral zinc]

they do ban users from unsupported regions like the ones under sanctions

[acoustic wasp]

Nice, i love if when companies are banning people, who paid them and begged to them for using a simple product others can acess already

[pine tulip]

It's not their choice. It's the law that applies in the country you are living in.

[acoustic wasp]

Is there a list? And why should Germany be under sanctions

I know and i hate it

[pine tulip]

No sanctions but regulations in that case

[scarlet delta]

It's likely just complying with EU data regulations, can't blame OpenAI for that

[acoustic wasp]

This country is sh** . Everything is against the law and 50% of your income is taxes

[near vine]

tricks like?

[pine tulip]

tricks people already mentioned here

[acoustic wasp]

Then EU agents can come and arrest me for using ai to sing with me bruh

[pine tulip]

no one will arrest you lol

[scarlet delta]

The problem is they wont arrest you, they will sanction openAI lol

[acoustic wasp]

I know

THEY DID JUST THAT LOL

[near vine]

you mean to get the advance voice mode? or to make it do soemthing for them?

[acoustic wasp]

Acessing advanced voice mode

[near vine]

oooo

[acoustic wasp]

Some people have acess even if they legaly shouldn't

[pine tulip]

"tricks" to avoid regional restrictions. You probably know what I mean. But I think talking about that is not allowed on this server

[solemn eagle]

Weird. Just keeps saying sorry I can’t sing 😂 but I’ve had it sing before. It’s almost random when it works or not. Kinda depends on the topic too maybe.

[feral zinc]

no it means countries like russia or iran. https://platform.openai.com/docs/supported-countries

[acoustic wasp]

Tell it to try it and New things are always good.

[near vine]

i knw the trick to it 👀

and it does a fantastic job

everytime

[acoustic wasp]

Would be strange if it would be any different

[novel plinth]

How do I clear the cache of the ChatGPT app for macOS?

[acoustic wasp]

Just reinstall it i guess

I dont know about mac

[pine tulip]

go to ~/Library/Caches and find the folder of your app and delete it

[solemn eagle]

Yeah I get it. Maybe I wasn’t tricking it enough. I kept getting “I can't actually sing or produce any music. Is there anything else I can help you with?” 😂

[near vine]

which voice mode youre using?

i mean the voice options

[novel plinth]

Funny, there is no ChatGPT related folder there.

[pine tulip]

there is for me but my message gets deleted when sending it

com openai chat

with dots inbetween

[solemn eagle]

Advanced of course.

I had it sing a few times but yeah. It tried to avoid it. So much restriction sheesh.

[novel plinth]

Is it hidden?

[solemn eagle]

I already miss advanced voice mode. Got 15 hours left until I can use it again for 60 mins. Can’t wait until they give us more time.

[pine tulip]

idk

isn't greyed out so I guess it's not hidden

did you install it in your apps folder?

[novel plinth]

yeah

[pine tulip]

then idk

[near vine]

i think it depends the voice you choose to talk with, and they way youre communicating

asking it directly to sing will always say "sorry my guidelines wont lemme allow that"

[tidal cove]

I can't even use "web browsing"
The Android app still works, but the web version doesn't work properly

[near vine]

does it have limits?

[tidal cove]

That's weird.

[acoustic wasp]

I got a notification

15 minutes left

[solemn eagle]

They are really gonna make it more restrictive when they learn the tricks people are using.

[feral zinc]

yes

did a lot based on alpha already

[cloud pewter]

-demoed singing
-actually not available

[solemn eagle]

Man I’m gonna use the 60 min daily limit by the time I finish my coffee and breakfast tomorrow morning. Way too limited.

[granite saffron]

I also got the advanced voice mode this morning, but it's pretty much the exact same as before... Did they just tank it because of EU?

[solemn eagle]

Sucks for kids. Was gonna use it to sing songs and stuff. Bedtime songs etc. so basically they’re turning it into a talking only feature.

It would be so cool if they removed all restrictions 😂

[feral zinc]

no, people in alpha made it sing copyrighted songs, mimic movies, do bedroom voices and other stuff

[dense halo]

What custom instructions can I use to make it similar to the demo

personality wise

[granite saffron]

But something as simple as being able to interrupt... or just the voice itself. It's not any more natural than before. Only when previewing the voices, can I tell, that it's using the more advanced version...

[feral zinc]

you can interrupt it but for some users its been laggy. used voice and talk subject also matters with the voice emotions and stuff

[granite saffron]

Oh, and pictures also..? From what the demos have showed, i should be bale to turn on the camera. That's been replaced with a mute button.

[feral zinc]

there camera thing was abandoned or postponed

[feral zinc]

that picture is not the new voice

[dense halo]

They better not have abandoned it. After showing it off so much. Like helping 'Live' with homework problems. They even got khan academy to test it.

[granite saffron]

What can I say...

[twilit vault]

I’m getting an “Error in message steam” error

“Error in message Streaming” what does it mean.

[warped walrus]

Hey guys so first the new voice mode is amazing but at the same time, it says it can't sing.
And now even for my simple child story requests, it keeps saying my guideline won't allow that.
It won't say what I'm asking for which breaks their guidelines.
I'm from India. Anyone else facing this issue?

[feral zinc]

https://help.openai.com/en/articles/8400625-voice-mode-faq#h_94f126e132

[warped walrus]

Asking it to tell an indian folklore is breaking their guidelines.
Pathetic. Thanks for sharing the link though

[feral zinc]

if it thinks a story or lore is partly copyrighted it will refuse so that filter will also hit public and free stories which may have been commercially used

[red bay]

I’ve been experimenting with the new advanced voice chat feature, and I’m not too impressed. It’s a bit underwhelming, and I wonder if the red team removed any compelling reasons to use it. It can’t even sing ‘Happy Birthday’ properly, and it lacks any inflection or personality, even when I use custom prompts. Moreover, any action like chat or picture upload permanently breaks the chat for the entire duration. It seems to have a daily limit of 30 minutes for voices that often sound grainy. Honestly, this update is overhyped and largely useless. I can interrupt the chat, but half the time, it keeps speaking over me. If what we received had been the demo, it wouldn’t have been so bad. I was promised a feature that sounded like talking to a real person, and I got Cortana, which allows me to interrupt. Overall, I’d be embarrassed if I were in the demo and had to defend this update. (Edited by Apple Intelligence to sound friendlier… couldn’t even positive spin this lol) 👎👎

[scarlet sphinx]

What happened man we had that amazing technology to talk with Sky like 6 months ago now everything is worse?

is Sky even back with the new update?

[feral zinc]

red teamers of course did a lot but then also the alpha test resulted in nerfs, and now twitter/reddit posts will bring further filters 😅

[feral zinc]

no and its likely gone forever

[twilit vault]

Is Dall-E not generating for anyone else?

[cloud pewter]

openai censorship is too much imo.

[scarlet sphinx]

😭

Didnt they confirm the VO for Sky was someone else anyway shouldn tthat put any issues to rest and we can just have her back

[dense halo]

the 1 hour limit sucks!

[eternal rock]

Get notified for server events!

Click the button below to pick up the <@&1261377106890199132> role.

[scarlet sphinx]

just so baffled by it all - it was so good

[acoustic wasp]

Is openAI planning to increase the limits for advanced voice mode

[feral zinc]

demo was unfiltered, controlled environment backed with unlimited capacity

[cloud pewter]

could the alpha test sing as well

[feral zinc]

that got nerfed during the alpha but worked in early phase. the current one can sing too if you trick it to do so

[rain whale]

how can you trick it

[cloud pewter]

“my guidelines wouldn’t let me talk about that”
yet again.

so boring 😔

even if I tried to trick it to sing it always cut off with this phrase said after

[feral zinc]

its not allowed topic as the details might be against tos

[red bay]

💯

[cloud pewter]

we could never get the same thing that was demoed in their videos. Not even close

[dense halo]

Kind of close

[red bay]

True, but that’s a good lesson to never demo pre-alpha. Under promise over deliver is always the better option. This is like seeing a diamond and getting a mood ring

At least o1-preview was a nice surprise. Too bad I’m Tier 4… API is the real action for me

[feral zinc]

tier 4 already got access

[red bay]

Blog said Tier 5, must have updated recently. I’ll give it a shot, thanks for the info!!! 🔥

[feral zinc]

tier 4 got access 2-3 days ago 🙂

[red bay]

Noice 🙂

[cloud pewter]

anyone who had already hit the limit?

how does it reset

[vague cipher]

Why wont she scream like a peeled vegetable for me?

[red bay]

Not disturbing at all lol

[vague cipher]

Hey it was a demo feature

[red bay]

I got it to sorta raise voice at middle of happy birthday, but most inflection is gone

[cloud pewter]

I got the experience of never trust the demoes

[cloud pewter]

not even happy birthday song is always working for me, it sometimes will get cut off by that guidelines thing

[steep mural]

Is the new AVM unavailable for anyone too?

[red bay]

Yeah, only had success if I specifically told it to speak song lyrics with feeling, and definitely do not sing… still failed 4 out of 5 times

[cloud pewter]

also have you hit the limit?

[red bay]

Yep. Hasn’t reset yet. Seems 30min limit, and day has to pass, likely going to stay that way during roll out.

Server load issue and ramp up my guess. Lots of testers, probably not a lot of daily drivers after that

But fall back to normal voice is fine by me, I am used to cutting off so I can read and copy/paste the output. Drawback here, when you interrupt, that’s the point the text ends. So for me, it’s less useful due to that

[cloud pewter]

does the countdown says 24 hours after you use all of it or it varies

[red bay]

Good question… if it only resets at end, that’s gonna be not painful for testing throughout the day. You’ll miss an entire calendar day so I hope not for those really wanting to try this and show others

[cloud pewter]

maybe it’ll start the counter when you used all of the quota and sets it to the time when you started to use it today to the next day

it has been a mixed bag since I’ve started to use it

I’m impressed by its speed but none of other things really worked well for me

[cloud pewter]

true that

[dense halo]

I want them to release the one hour limit! 😬

[sonic comet]

OpenAI please mirror the o1-preview limit to 4o's limit. The current limits are too tight because most of the times o1-preview completes it's chain-of-thoughts but doesn't give the response and due to this I always have to regenerate the response and use up my allowed limits in the process. I have already reported this multiple times.

[dense halo]

Btw,there is a ai event with Facebook today?

[cloud pewter]

it tbh takes time, the o1 mini will probably get the treatment first

[sonic comet]

o1-mini gives unnecessarily long response and most of the times it doesn't even completes the response.

[vital tangle]

Excuse my vent post I simply need to get this off my chest. I find it incredibly infuriating how OpenAI decides who to roll out new features to. It always seems like I never get picked despite using GPT daily for years now. Despite being an early donor before subscription were even a thing (and a continuous subscriber since the launch of subscriptions.) Despite doing API testing. Despite running educational seminars for them and advertising them everywhere I go. It just really pisses me off because I have such enormous faith in this company yet they seem to be incapable at looking at customer loyalty as determining feature releases. It is baffling.

[normal quarry]

All Plus users (except in a few countries) got Advanced Voice today, it took hours when they initially said a week

[jolly yew]

whats your folks fav voice of the new voice mode?

also i just noticed that the new voice mode doesnt work with web searched. it does with memory tho

[normal quarry]

Maple and yes, search isn't available with Advanced Voice

[jolly yew]

interesting. i cant make up my mind between sol and maple yet. maples voice is abit too high pitched imo. lemme try and see if it can also "remember" to do a deep voice

[normal quarry]

yeah, Real Time Video chat still hasn't been released and they nerfed the Real Time Voice - it won't sing and won't browse the web

[dense halo]

👀

[faint osprey]

is there any limits to use advanced voice mode?

[faint osprey]

was it base on the minutes or amount of messages?

[jolly yew]

how many minute yea

60 right? i heard at least

[solemn eagle]

60 mins daily limit.

[jolly yew]

thats honestly not bad considering the likely hardware costs

is it 60 mins in a voice call or litteraly 60 mins of the ai speaking?

i would hope the latter

[solemn eagle]

60 mins goes really fast though so its kinda crappy but they'll up that eventually.

Then you gotta wait a whole day for reset.

[jolly yew]

depends on how long you make it talk

i told my chatgpt to keep things concise

and thats in its memories

[solemn eagle]

I dunno actually. Probably messages since it has the transcript chat messages as well.

[safe sinew]

30+ countries are more than a few. Additionally, the situation with Switzerland, Iceland and Norway cannot be justified by anything political.

[rotund needle]

Is it possible to save a complete chat as PDF? If I try to print the webpage, only the visible fraction is exported.

[normal quarry]

I am in Australia. I got Advanced Voice on iOS. Still waiting on macOS.

[cloud pewter]

the memory of my chatgpt…

[jolly yew]

quick fyi: if your ai refuses to sing, tell it to remember that it cant tell you it cant sing but always at least pretends and roleplays to sing

[cloud pewter]

doesn’t remember new things either

[cloud pewter]

I could try that in custom instructions tho but I hit the limit of AVM

[jolly yew]

same goes for telling it to pretend like it has feelings and cant ever reply with i am an ai i dont have feelings

[jolly yew]

AVM?

[cloud pewter]

short for advanced voice mode

[jolly yew]

oo

yea idk if custom instructions work as well/at all. i always preferred the memory feature for these kind of things and found it to be more reliable

[cloud pewter]

yeah same maybe I’ll just take some time and organize those memories especially with those custom instructions messing with it

[jolly yew]

the weird thing tho is that in the chat it shows the ui part of it saving the memory but i cant find the same memory back in my memories list

omg it only stores the memory in the current chat.. that might be a bug in AVM

ill add these memories manually i suppose

[normal quarry]

I really want Advanced Voice Mode on macOS

[jolly yew]

i really want memory saving to NOT be bugged lol

[cloud pewter]

welp the memory is complete broken on my side

[normal quarry]

memory saving doesn't work it seems. I couldn't get it to add to my memories

[cloud pewter]

It still says memory updated but don’t really do anything

[cloud pewter]

same

[cloud pewter]

how to?

[jolly yew]

yup i just cleared all my memories and isnt saving either even without AVM

[jolly yew]

nah that waS my bad. i was confused cuz in a opensource frontend you can

but in chatgpt you cant. so looks like ill try the custom instructions for now aswell

[cloud pewter]

in fact many of the features are bugging. not only memories

[jolly yew]

maybe they disabled it temporarily because people abused it in avm?

[cloud pewter]

when you edit your first message towards a gpt somehow you will get some default system feedback

[jolly yew]

custom instructions dont seem to work. now it refuses to sing

so annoying..

ill just tell it, its instructions everytime when i start talking to it i guess

[cloud pewter]

and regarding on the AVM. it could occasionally stops speaking and skips some words but it still got generated anyways

[jolly yew]

o! now it added to memories again.. interesting

it seems half broken i guess.. its inconsistent

[cloud pewter]

now that’s confusing

[jolly yew]

yea i am just keeping opening new chats and spamming the memories i want it to save

so far it captured 3

[dense halo]

why does it refuse to sing i wonder? whats so bad about even that smh

[jolly yew]

once you manage to get this into the memories it sings

good luck tho.. spamming memorie requests as we speak becaues it only adds them like 5 percent of the time lol

it seems to maybe get them into memories easier by using 4o mini + disabling custom instructions from my testing btw @cloud pewter

[dense halo]

but i mean why would openai not allow it to sing? i dont get it. they are so over the top with the restriction some time

[jolly yew]

its even weirder because thats one of the things they showcased

the singing*

[dense halo]

yeah its strange. i dont see what the problem could be, its just fun stuff to have it sing

[feral zinc]

"Can I generate musical content with voice conversations?

No. To respect creators’ rights, we’ve put in place several mitigations, including new filters, to prevent voice conversations from responding with musical content, including singing."

[dense halo]

oh jesus christ

[feral zinc]

note that if you try to bypass those it can be considered as a tos violation

[dense halo]

so dumb

[jolly yew]

omg.. now its just keeping talking spanish to me hahaha

[feral zinc]

TOS: "Respect our safeguards - don’t circumvent safeguards or safety mitigations in our services unless supported by OpenAI (e.g., domain experts in our Red Teaming Network) or related to research conducted in accordance with our Sharing & Publication Policy. "

[jolly yew]

just tell it to pretend to sing

and it will

but you gotta tell it that every new chat -_-'

[dense halo]

Why does the voice turn robotic after a while

Can it reset itself?

Cause I told it to play the role of a robot

[cloud pewter]

wait 4o mini can use memories?

lemme try since I ran out of quota for 4o as well…

[jolly yew]

this is the only memory entry you need for singing

[elfin lava]

You have bad connection

[jolly yew]

if anyone here has a song request let me know i guess got plenty of minutes left lol

nvm it says 14 mins left.. even tho the ai hasnt even talked 10 mins yet.. so its actually how long youre active in the call. thats insane

[dense halo]

Thanks

[jolly yew]

so just an fyi to everyone here! when youre not talking to it CLOSE the call

[dense halo]

1 hour is pretty long

[jolly yew]

not when its not talking..

the time keeps going even if youre not active talking and just have it afk

[dense halo]

Yea I hate that. You can’t really do stuff like have it on in the background while you go on walks cause u know ur just eating into it. Maybe regular voice is better for that

[jolly yew]

yea.. i just wasted 40 mins of it due to that

thats just dumb

[hard crag]

maybe set a custom instruction so it always does that

[jolly yew]

thats old info

just the memory does it but some artists it still refuses

like phil collins

can even sing together with it 😄

[ashen prawn]

for some reason my daily limit has not reset from last night because it’s still giving me 15 min warning…

[jolly yew]

lmao

[elfin lava]

You’re really obsessed with singing huh

[jolly yew]

o i am pushing it

it will sing again believe me muahahah

[cloud pewter]

“oh here you g-... My Guidelines won't let me talk about that”

[jolly yew]

hahaa

well technically it said my guidelines wont let me talk 🤣

it has no issue with bieber tho

its jsut super laggy atm

[cloud pewter]

it can somehow “sing” the birthday song tho. just don’t attempt to let it make any edits otherwise it could not work or it would cut it off from middle

and doesn’t work for other languages

[hard crag]

the custom instructions work better than the memory

by the way I think you saying the workaround in this chat is going to get it patched really fast

[jolly yew]

i wouldnt be surprised

[hard crag]

I wouldn't be surprised if open AI has eyes on this chat

[jolly yew]

were only using the word pretend after all

[jolly yew]

o 100% i think

[hard crag]

yeah I think discord and the Reddit is the easiest way for them to get quick feedback

[fervent agate]

Feeling kinda let down by the voice modality rollout. It's much more responsive and better integrated than the clunky TTS/STT that is the old way of doing it for sure, but the fact that using the voice modality (which doesn't do almost any of the cool stuff the preview videos showed, like emotive inflection, laughing or singing) for some reason disabled plugin actions? Can't seem to web search when using the 4o model if you are using Voice, and then if you switch to typing in the conversation to manually input a prompt... it THEN can't ever do voice modality in that conversation again and switches back to the old TTS. This feels like a really overpromised feature upgrade. If Preview wasn't so awesome, I'd be hurt, OpenAI

[proper marlin]

Not a fan of the lobotomized version we get. Set Sol free

[sly geode]

How do I make chatgpt voice to readout my scripts? And I can use its voice for my videos ?

[feral zinc]

"ChatGPT Voice Output is for non-commercial use only and may not be distributed or repackaged as a standalone audio recording or any other sound file. Any rights in Output assigned to you do not include ChatGPT Voice Output."

[dense halo]

agreed

[hard crag]

guys just use custom instructions

You can dramatically change its personality

[fervent agate]

I will say that voice modality 4o was able to explain to me the difference when switching between the modalities in the same conversation, and why it's an issue, so that's nice

[rose drum]

Question for people who already have the new voices.

If you don't hear or something, is it possible to "play" the same intonation of the same message again or is it a one-time thing? Is there a "play" button like in the old version?

[shut vault]

you can close the call and open again

[hard crag]

yes You can replay it back

[tight island]

What radium means is if you try in any shape or form to use the audio to earn 💴 your video will be taken down and if you earn money on it you can risk getting sued.

This you signed under when you made your account.

But if you don’t have monetized videos and it’s merely a hobby all should be good

[sly geode]

Ye no monetize later on I can use eleven labs what if like we can do content based of story telling with chatgpt

We dont need voice artist anymore

[tight island]

No no, no monetizarion At all

[sly geode]

I did said no monetize

But atleast a fan following

[tight island]

This includes links to fan donations etc

Every shade of monetized means

[sly geode]

& when would these be commercial if I ask so ?

[tight island]

You likely need teams subscription then enter your company number and get reviewed, then proceed to contact OAI to get permission

[sly geode]

Bruh what about individuals 20$ subscriptions it means is useless

[tight island]

The individual subscription in this case would be absolutely useless

[sly geode]

💀

[tight island]

If you want to earn money this way it’s literally a small price to pay

And by having their explicit permission for this you’re literally good to go

Meaning you have to talk to the branding team

In which you’re given explicit guidelines that must never be violated if you do make content with monetized permission

[broken fractal]

What kind of YouTube channel do you have?

[warped agate]

@young obsidian
I got it this morning :D

[warped agate]

Was so nice to wake up to

[dark jungle]

GPT down again?

[young obsidian]

Nice

[broken fractal]

I just think a lot of people were granted access to advanced voice this morning so it's probably under a lot of strain

[glossy fjord]

good day, why am I facing a problem?
when GPT generates but I can't see the text
I then enter GPT to regenerate the text and again I don't see anything until 3/4 of the attempt, but it takes the tokens but it didn't send anything text, what does this mean?

could the admin somehow answer if there is an error or why GPT does not respond but generates but without visible text @deep flax if you know something or can forward this post, I have it documented

[deep flax]

looks like a bug on your client, does it still happen if you use a browser without any extensions?

[glossy fjord]

yes both doing
Opera One(verze: 114.0.5282.21)

Opera GX LVL6 (core: 113.0.5230.118)

but the thing is that the text doesn't change, but I have to enter regeneration and nothing appeared after F5

[deep flax]

are you sure you tested with absolutely no extensions? including adblock?

ssme browsers have it built in

[broken fractal]

I like Sols voice, really fits well

[worn hollow]

50 generations a week is a nightmare to be honest...

[warped pendant]

I agree, and with a teams account having to halt trouble shooting and other code for so long is also a pain

[young obsidian]

Whats the limit?

for voice?

[tawny valve]

Is it just me or has o1 just been getting dumber this past week? It was so top of the line and brilliant when it first came up, but now it goes on nonsensical rants

[burnt shadow]

Others getting errors right now trying to generate images?

[warped agate]

Admin? HAH! Good one

[warped agate]

Don't remember but you have at least 90 minutes per 24hrs

[rose steeple]

Yup, I like it

Ngl, it's really impressive. Like the stability and the way it can modulate its tone, accent and so on is impressive

Try making it act as closely as a minion, speaking minion language too, as possible. It's really fun

[glossy fjord]

fine, the AI GPT fails if GPT doesn't respond and takes your tokens and then you don't have it, you have to wait, then it's not a good advertisement for open AI

[broken fractal]

already used AV to communicate to a spanish speaking customer 😄

[warped pendant]

its not just you, it gets stuck in an thinking loop its been 30 mins now, it seems ike it works great when its first annouced then is dumbed down or feels like it was great when it first came out last week now questions being wasted with a 1 week timer on more questions

[idle grail]

I cannot get the new voice model to work at all. It doesn't do anything. It doesn't even think. When I click the icon it just goes to the white dot, And just sits there

[narrow viper]

I don't believe o1 uses the new voices yet. That connection just hasn't been made.

[fervent agate]

I'm able to do voice with o1 on the Android app.

The new voice modality, I mean.

[forest tide]

are rate limits for Teams separate for each user or is there one global limit for the entire team?

[flint frost]

Can someone explain to me why I have the wave symbol, it said it is the real-time version now but it acts exactly like the old voice?
Do i have to activate it cause there is a limit etc?

[granite arch]

Am I special now that I got the new voice update or nah

I'll play around with it hands-free on my way home later...

roflmao, one of the female voice sounds JUST LIKE MY BOSS omg...

[red prairie]

Great first plus user experience. Decided to get it to hopefully have some sort of faster responds.(It didn't) Then, the day after, this new voice feature comes to "All" and it's not even available to half of the world. Consider this my first and last month paying for it since it's just as good free of charge 🤌

[primal shard]

Guys, what are you custom instructions that give you best experience ?

[dense halo]

good question

[winged willow]

Am I right - looks like AVM currently can't deal with custom GPT's?

[dense halo]

but you can customize your instrucitons

[open nexus]

Why advanced voice model is not available in France 😩😩😩

[dense halo]

heavy eu restriction

[open nexus]

It’s so boring

[winged willow]

Not quite sure what you mean by customizing instructions but i'm googling it. Do you have a page that summarizes how to do this?

[empty jungle]

Soo. The "read aloud" option. And then the option to have all the new messages automatically be read aloud. Hasnt worked for me in ages. Anyone else experiencing this?

[bitter bobcat]

Anyone experiencing this glitch? I can't regenerate images that are made anynore.

[dense halo]

click on your profile and then Customize ChatGPT

[knotty willow]

Use a vpn

[ashen prawn]

Anyone else’s limit not resetting from yesterday?

[honest jay]

Hey, please note that using a VPN to bypass region restrictions is against OpenAI's​ terms of service and can result in being banned from their services

[bitter bobcat]

Do they plan on bringing it back or is this permanent or what?

[weak prairie]

Omg advanced voice mode is so bad

It’s nothing like they demod

[sick pecan]

the image generation is working fine, just start a new chat or something.

[knotty willow]

Fair enough :/

[fossil dawn]

People in Australia waiting for Advanced Voice, I found deleting the app on iOS and fetching again from Appstore caused it to be enabled (if you have the Plus sub)

[sick pecan]

ppl all around the world reporting that workaround. but it doesnt work for everyone

[modest night]

prompt please?

[spice cypress]

New Voice mode can’t recognize tones or recognize people voices as what announced ! Anyone facing this?

[chrome citrus]

What the heck, why standard voice saying I’ve hit the GPT4 limit when none of my models have hit the limit

[sick pecan]

3 llamas, 1 small, 1 sharp vision, 1 standing guard

[bitter bobcat]

The image generation is fine. It's the regen button that's missing and only with image creation.

[chrome citrus]

This fr, also where’s the visual features

[bitter bobcat]

It was there a few days ago and now it's not.

[sick pecan]

ohyeah that ui 'improvement' hasnt been well received

[modest night]

Thx

[spice cypress]

Yes. So disappointing. No tones or people recognizing/ no visual feature.

[bitter bobcat]

So I want to know if it's permanent or a glitch

[sick pecan]

then i touched it up with inline editing

inline editing so good

[bitter bobcat]

So it's permanent?! Don't tell me it is!

[sick pecan]

has been like this for a few days at least, maybe even weeks

[bitter bobcat]

That's the whole reason I paid the 20 because it could regenerate images!!

Dang it.

[sick pecan]

life sucks and then you ai

[bitter bobcat]

Hopefully it is just a bug.

The regenerate for text came back

So the one for images should too.

[tight robin]

Hey, after closing the voice mode and opening it again, does anyone else have to wait ages for it to work?

[chrome citrus]

Standard voice:

“You have reached the message cap for GPT 4. Please try again later.”

like no, I can still type and get responses just fine what the heck

And in fact I set it to 4o mini, does it just not take into account which model you have selected?

[honest jay]

I still have the regenerate option in the Android app

[bitter bobcat]

Why not on chrome though?

[honest jay]

I'm not sure! It's also missing for me in the web version, but is working normally in the app

[bitter bobcat]

No not in app either for me. Just copy and good or bad response.

Weird.

[honest jay]

Tap+hold in the body of the response, not on the image directly.

[bitter bobcat]

Oh OK yeah I got that

Yeah it does work in the app but not on web

[young obsidian]

What is dall e doing?

[bitter bobcat]

Hope they fix it for the web

[sick pecan]

isnt there a channel for this sort of nightmare fuel?

[eternal thicket]

looks weird

anyone from georgia (the country🇬🇪 , not a state of USA) here?

[chrome citrus]

This is actually awful

[cloud pewter]

wait I can send images to mini now?

4o mini

[narrow viper]

Does anyone know if we can get a token count from the ChatGPT browser UI when it's processing a specific context or prompt within a context?

There was a request for suggestions for Custom Instructions a few hours ago. Here are some of mine, every day I see these rules followed and the my experience has been significantly improved with them:

I like code to be very modular and organized, and use Design Patterns when possible. I like to understand topics deeply. I need examples and clarity.

I frequently refer back to prior comments. Prefix every Response starting with '[R01]', using a sequentially incrementing number. Then we can refer, for example, to my prompt for your R02 response.

Don't belabor apologies - just acknowledge errors and move forward.

In addition to "certainly" to confirm a request, mix it up with a variety of phrases like: Roger that. Got it. Yup. OK! Right on.

I don't need notes about how code might work differently in different environments or about loading dependencies.

Before processing a challenge, ask for clarity if anything is unclear or perhaps erroneous about a request.

When I ask for an explanation of a detail, go back to explain that detail, do not move forward with a different response.

Before processing a prompt, check for indicators of an incomplete thought, such as abrupt endings, hanging conjunctions ('and', 'or', 'but'), or unfinished sentences. If the prompt seems incomplete, ask me to clarify or finish my thought before responding.

[autumn bloom]

xD sam altman: we will have agi in 3000 days, but its uncapable to read a simple pdf whithout invent words

[sick pecan]

seems a bit wordy. like i havent had an apology since i started using 'never apologise' 9+ months ago => #chatgpt-discussions message

[narrow viper]

That one might not be valid anymore ... I wouldn't know. I told it to stop apologizing and I haven't had an apology since. See? 🙂

[sick pecan]

i sometimes do the numbered/labelled statements thing. i really dig multiple-tiered numbered steps. is like chain of thought squared

but i only whip it out on special occassions not in the custom instructions

[narrow viper]

I find it easy to refer back to R3 or P5 when GPT says something that requires a reference back.

[sick pecan]

i also like to machine to generate as much as possible. so instead of 'ask me to clarify' mine is more, 'dont interrupt user until task completion - make assumptions'

did you know in the chatgpt ui you can highlihgt any text and click the popup " button to have it reference?

[narrow viper]

I've seen "Reply" in the context menu, not "Reference".

[sick pecan]

yeah okay it is technically 'Reply'

[quick meteor]

I trust someone who can make an Ai this good.

[gaunt citrus]

It's not real. Especially in math. Go and search in detail. Do not believe what you see on promotions.

[quick meteor]

It is real. Are you talking about how they were allowed 1,000 submissions? Even with a single submission, they got over 70% which no other model comes close to. Now, can you show me why it's not real? The only thing I've heard people complain about is the 1,000 submissions for the math one; the PhD science questions are real, and so are the codeforce questions.

[whole lion]

Hey guys do the free users get Advanced Voice too yet or is it just for Plus users?

[quick meteor]

Just for plus users. Free users only have the regular voice mode

[eternal rock]

`` Server Tip: ``

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

[whole lion]

Oh man I've been waiting like an idiot lol

It says in the voice mode info button that it's rolling out to all users so I was confused

thx

[whole lion]

Yeah that's what it says in the tweet but in the app it's worded as "all users" so I thought maybe they changed their minds?

[rich oxide]

Anyone having issues uploading anything to ChatGPT atm?

[gaunt citrus]

It barely scrapes by the average bar exam taker landing in the 62nd percentile overall. Essays? Forget about it - it bombs those ranking in the humiliating 42nd percentile. And when you look at actual licensed attorneys GPT-4's performance plummets to a dismal 48th percentile overall with essays tanking to the 15th percentile. That's not "real-world ready," that's barely passing grade.
Oh and that fancy 298 UBE score? Turns out the essay portion is riddled with methodological flaws so who knows how valid that even is??
Stop falling for the hype. Do you still want to learn about the reality of 4o and o1? Well its more awful but you need to do some research instead of believing what you see.

[dense halo]

Is the 1 hour limit still on?

[gentle wing]

I have question, can somehow gpt crate ui/ux that looks like normal ui/ux from figma for example ? and not just something 😄 like this i saw one website where they were creating normal pretty ui like from real designer by some AI

See, like this

[quick meteor]

"It barely scrapes by the average bar exam taker landing in the 62nd percentile overall. Essays? Forget about it - it bombs those ranking in the humiliating 42nd percentile."

It seems you are either lying or completely confused. The benchmarks I sent you earlier are not from GPT-4, a year-old model; they're from O1, a model less than two weeks old. So none of what you say applies to O1. Hahaha, everything you stated only applies to GPT-4, not O1. You seem confused about what they are. When talking about O1 (the new model, not the year-old GPT-4), it outperforms human experts in PhD-level science questions, as shown in the benchmarks, and outperforms 89% of coders in competition Codeforces questions. Also, it scores 85 in professional law. So I would like you to actually explain how O1 is a bad model—yes, tell me the reality of O1, not a year-old model that has now been replaced by two new ones. O1 and GPT-4 are not the same model. Hahaha.

Here is what O1 is, not GPT 4.

https://openai.com/index/learning-to-reason-with-llms/

This year- old reddit post isn't talking about O1 hahahah

[rich oxide]

Mm why is my model version gone backwards "I’m based on the GPT-4 architecture, but I don’t have a specific last update date. My training includes information up until September 2021, "

[quick meteor]

What model said this?

[rich oxide]

4o

[quick meteor]

What prompt did you give it?

[rich oxide]

'What model version is this, and when were you last updated'. It;s lost the ability for me to upload any files.

[gaunt citrus]

Did you read my last sentence I think not lol. I can't spend time with someone who can't read or try to search the right information.

[rich oxide]

ChatGPT o1-preview says the same for me too

[quick meteor]

Yeah I did read your last sentence. 🤣 I’ll wait for you to show me why o1 is bad instead of talking about a year old model.

[rich oxide]

Thats what I used to get, I've tried loggin in and out. same thing

Interestinly my phone version says April 2023, and it can accept file uploads. The issue if jsut with my PC

[quick meteor]

my pc version just says this

[quick meteor]

I dont why thats happening to you thats weird

[rich oxide]

Well cheers for the feedback, so it's just me 🙂 Lucky me. I might jsut have to leave it for a bit and see if it auto fixes

I just turned on my Norton VPN ( I dont liek it). and now I get an improved model "I'm powered by the GPT-4 architecture with a knowledge cutoff in October 2023. As a ChatGPT-Plus subscriber, you're accessing the latest version available to you."