questions

[alpine iron]

i'm not a moderator nor admin for this server

mostly just a student looking at this project

[low ermine]

Is Neuro based on a Transformer model?

[alpine iron]

neuro is based on a large language model so yeah definitely transformer model

[low ermine]

Makes sense. She should've been trained over stream speech only instead of text data because she keeps trying to pronounce emotes and aasterisks

Or at least add a pipeline to filer out tokens that aren't "pronounceable"

[alpine iron]

hmm

well it probably slows the process down

also now we seen her enough, having he pronounce the emotes is part of her "personality" now

[low ermine]

Lol ok heart

I wish it was an open project so more people could get their hands on it to make it better

[alpine iron]

true.

though personally i was hoping there is a more open LLM everyone can play with

despite being called OpenAI we never get to play with their models

[low ermine]

Now that Microsoft practically owns it, you will see a lot less of the 'open' part

Although I prefer that Microsoft got its hands on it instead of Google

[alpine iron]

i will leave the AI tech talk to other places while i watch what vedal will do

[slender rock]

Does anyone possibly know which ML tech stacks make Nero work? Truly curious

[fast glen]

i think only vedal could tell us that .

[static jolt]

I only know the TTS he uses but nothing else

[slender rock]

That’s interesting!

I have always wanted to make a AI waifu

[static jolt]

Same, I've been trying to figure out how to do it myself

I've managed to create something a bit similar with the OpenAI API but creating and training your own LLM is completely different

[slender rock]

Wow that’s so nice to hear!

I feel like OpenAI is indeed the most feasible way for the ordinaries like me

[static jolt]

Yeah its definitely the easiest way, I was able to make a working prototype in one day pretty much

And I didn't even go into fine-tuning or anything which would probably create better results

But it doesn't feel that good since you know you are just using the API, creating your own would be much more fulfilling I imagine

[slender rock]

Yeah, it would be great to customized my own Waifu without having to strictly follow OpenAI’s policies….

[mellow flame]

I'm sure some crazy hololive fan has transcripts of every stream every hololive en streamer has ever done saved somewhere to train off of, right?

[slender rock]

But I really wonder what enables Neuro sama to give comments based on the content in the videos

Just imagine watching anime with your personal AI waifu and she gives reactions, OMG. That would be truly amazing

[static jolt]

True that would be amazing

[fast glen]

open AI and character AI, as well as novel ai, etc... well, i mean, somewhere around there is a large language model you can just tell to be one on github right

[small owl]

Do you like bun bun?

[queen sand]

@small owl has leveled up! (2 ➜ 3)

[small owl]

[uneven zealot]

[wanton depot]

can stickers be restricted in specific channels?

[lime sleet]

Are the models used for singing and TTS different?

[balmy quail]

for what?

[bitter crane]

1

If vedal sees this message, please check the private message. Thank you. Translate by a translator

[fast glen]

to be fair, something similar ish is possible. As you said, you would have to train, but to collect data to train off of is 100% a thing. You could even weigh the training data by 'confidence' and 'quality' out the network output.

[wanton depot]

Livestream chat, etc...

[weak elbow]

not livestream chat

[balmy quail]

livestream is open for spammy

swarm is the specific for it but other than that it's already in rules to not go about spamming, but entirely removing them is kinda meh

[lost bay]

Hey I got a few Questions

1. Is Neuro Sama pre trained? Or using Reinforcement Learning
2. Which ML library did the Chat AI uses?
3. What TTS engine does Neuro Sama use? (Custom made?)

[tender olive]

my question is
Vedal is the creator the Neuro Sama , and i have a questions for the example
what do you did, moving eyes? sorry i do my best for the speak inglish

[eternal sonnet]

I can't answer any of these sorry

[eternal sonnet]

and I don't know what this is asking

[alpine iron]

i think he is asking how you rig the model's eyes to follow the cursor

[lost bay]

Also not the reinforcement learning?

[tender olive]

How did you get Artificial Intelligence to move according to something that doesn't exist?
I understand that to capture movement there must be something physical, but she is virtual,
how did you fix it?

and Hello

I'm your fan

[eternal sonnet]

you don't need something physical at all

npcs in video games dont need a physical body to move

[tender olive]

So she could say that she has her own conscience?

[alpine iron]

i don't think a live2D rig needs to use a camera is what vedal means

i believe live2D has fixed animations it can be used to play on a loop without any interaction

[tender olive]

oh

i understand

thanks for the help and nice day or night? i dont know but see you vendal and thanks for you time

see you

[lost bay]

Is the AI simulating a face tracking or smth to let the model move?

[lost bay]

For me it sounds like that the singing part is pre made with softwares like Vocaloid/Utaloid or smth but that just what I think not sure of it

[low ermine]

how does she react to videos

[severe perch]

Is Neuro-sama gonna sing a new song on the Camila collab or is she gonna sing an old song.

[queen sand]

@severe perch has leveled up! (0 ➜ 1)

[crystal hollow]

My guess is, the same way she reacts to chat, ...reads what is said in the videos.

[crystal hollow]

I guess...

1. For tha part of the system that plays games: RL (that would be the easiest choice)
2. Only Vedal knows but there are plenty of choices for training models.
   It's likely a fine tuned base model and/or with some embeddings.
3. Don't know much about TTS but also plenty of choices incl open source on GitHub.
   I'd imagine you can train a voice model in a similar way as you train chat bot, give it lots of samples.

... or the whole thing is ran by Vedal speaking to a voice changer 😂 😂 😂

[indigo hornet]

I really love Neuro-sama and all the advances with her AI(s) over past months, but as someone very sceptical and critical about everything theres just something here that rubs me wrong way. I just hope there is no malicious or illegal stuff involved with whatever runs Neuro-sama that could cause trouble in future if figured out. I am talking about keeping people in the dark about how this works.

Keeping secrets is fine if it means protecting your creative input and hard work, but not telling what your building blocks were in the first place seems unnecessarily shady. Its not like saying "its based on gpt3 that i modified" etc will help anyone steal your work. I just wish we had some transparency or general idea about what this is. Trust is always a good thing and keeps rumors and doubts away.

So, may i ask why are the details about how Neuro-sama works or what was used to implement her such a secret? Perhaps there are already some plans to talk about this stuff in the future when its ready for it?

[stoic urchin]

I don't know anything about how text to speech tech works
but would it be possible to create a tts that has alternate speech patterns to indicate inflection/emotion?
Like by basing off the use of ? and !, a different version of the tts would speak with an excited voice instead of the regular tone.
I think it would make neuro more natural when she's paired with a human speaking, side by side the lack of emotion sticks out

[eternal sonnet]

I mainly don't talk about it for "business" reasons, it would be silly of me to explain exactly how all of it works because I'd essentially be helping others make direct competition for Neuro-sama. I can assure you I'm not doing anything illegal or malicious to run Neuro, I don't even know what that would be tbh.

In the future if/when there's no incentive for me to keep how she works a secret, then I'd be happy to talk about it and disclose it but right now it just doesn't make sense to me. Does that make sense to you?

[indigo hornet]

I was supposed to write that my guess is that you want to keep it as a "business secret" but forgot about it. I suppose this is good enough explanation (for now) and im happy that you would be willing to explain more if/when you see no risk of competition or other problems. I just thought that if someone wanted to create competition they could already do it and the information i was curious about would be of little value to people that are well versed in AI field.

I just wanted to know that this project is kept secret for appropriate reasons and not something like stolen ai models or otherwise suspicious things. I will take your word for this as simply something you keep secret to avoid competition. Altho i still think if someone really wanted (and had skills) they could already compete with you since Neuro-sama is out there and serving as example of idea already.
Either way, thank you for answering.

[eternal sonnet]

You're totally right that people could compete with me, I just don't think its a good idea for me to make it any easier for them. I'm trying to make this my job for now if I can. It would be a lot cooler of me if I was open about the tech but I got bills to pay too.

[frank halo]

I'm pretty sure Vedal said he uses AI for the singing as well.

Much easier than making a song cover each time.

[errant aspen]

yeah i think it's done with AI but not on the fly. like there's some set up in having to either find songs that dont sound pepega or messing with the AIs processing to get a more preferable output. like those randomly growly crackly sounds basically dont exist if you made it from vocaloid, since youd have to add it in manually

[lost bay]

Hey I got some new questions. (Hope you can answer them Vedal)

1. Is Neuro Sama's chat AI based on a trained model or fully self trained? (Still wondering from where persons get that big datasets)
2. For the Vtuber is the AI simulating like a Face Tracking and sending it to Unity or the app you use?
3. Singing part is not vocaloid/Utauloid?

[fast glen]

Hi Vedal, what does SelfTaught == _watchedYoutube return?

[static jolt]

the tts is microsoft azure

ashley voice with higher pitch

[lost bay]

Ohh okay thank you lol

[lost bay]

Ohh yeah thats true exact same voice at around 1.70 pitch

[static jolt]

yep

[eternal sonnet]

https://wiki.haskell.org/Maybe

[edgy mist]

I’m just curious, is Neuro able to react to live content?
Because there’s some live content like Nintendo direct, minecon, e3, and other things like that that I think could be interesting to see her react to.

Also I’m curious to see how she’d react to her own livestream if it was open, just whatever she was doing a few seconds ago

[crystal hollow]

Neuro can read but it can not 'see',...yet 👀 ... yolo 🤓 ?

[lost bay]

A AI can't really see it converts the video / Image in useful material. And she can kinda see like she did in here reaction streams.

[crystal hollow]

It can get 'context' from object detection, also there are ways of activity recognition (CNN + LSTM) , it would be a bit of work and would cost more (processing) but it would be cool if Neuro sama could 'see' 😍

[lost bay]

Thats true but I guess the current system is fine for the current use. If vedal really makes that she can see she would be more aware of the gameplay hopefully

[eternal sonnet]

im working on this although it certainly won't be ready in time for the direct

[mellow flame]

Set it up so she reacts to her own stream on a delay

Neuro-ception

[warm lotus]

A question for today1s stream: Did she actually say Filtered.? Quite strange interraction lol

[vivid vessel]

[foggy creek]

I imagine Neuro is going to get more and more censored as time goes on, asfor the word filtered itself it could just be an anomaly, any number of reasons could cause her to say this.

[warm lotus]

Yeah, that is why I asked. Cause she might be learning to censor herself, but also could be a second layer in the workflow to replace what she wants to say with Filtered.

[foggy creek]

Perhaps you will get a proper response, would be interesting, tho it seems like Vedal does not like to share any of the inner workings

[warm lotus]

Yeah I do not wish to know inner workings, just was wondering if it's actually self censor or like a second layer

Cause that means she is actually getting unhinged and needs supervision

[foggy creek]

All we really know is that he claims not to "make" her do anything aside from stopping certain generated sentences, as he seems terrified of even the thought of angering twitch staff after the last ban.

[warm lotus]

I think most of the people share the same stance

[night elbow]

vedal probably censored neuro and she became sentient

[eternal sonnet]

I just made her say "Filtered." when she gets filtered instead of her just donowalling, its only for collabs

just trying not to get banned again thats all

[night elbow]

vedal, what games do you personally play?

[warm lotus]

Ah okay, It makes it more responsive this way instead of

A fun twist add quirky ones for the next one, like "My mom does not allow what im thinking this, sorry"

[foggy creek]

Neuro is really fun to listen to in major part thanks to all the random, often edgy things she will say.
I really really really hope she does not turn into a google search 2.0 in the end.
Need to maintain the "identity"

Hopefully

[eternal sonnet]

i am trying to ride the line between edgy and banned

without getting banned

not gonna censor her into oblivion dont worry

[foggy creek]

While you are here, i have been really curious. Would you mind explaining why some vods are often missing from Twitch ?
Unless this is a tradesecret ofcource.

Me and others cant always catch the streams

its quite late

[eternal sonnet]

im usually late to upload them

if they're not there its usually because she says something sus and im being cautious

[foggy creek]

The Miyu stream was pretty good, wouldnt mind seeing parts of it again

[low ermine]

What about the vods yt channel? Seems inactive :/

[eternal sonnet]

yeah havent uploaded on there in a while sorry, vods are still on twitch though

[eternal sonnet]

should be up on twitch in both our perspectives!

[low ermine]

True, but won't they disappear after like 30 days or something? More for the long run

If someone new wants to watch her improvement over time etc.

[foggy creek]

The Miyu stream is missing sadly

[night elbow]

unless someone records all the vods its kinda hard to put them on youtube

[foggy creek]

Oh wait, im being silly, my apologies. The Miyu stream is infact there, ignore this ping.

[eternal sonnet]

https://www.twitch.tv/videos/1730253500

its just named wrong

[foggy creek]

It has a different name to what i remember

[night elbow]

is there a server for abandoned archive?

[eternal sonnet]

https://discord.gg/aZntpZh7Z4

[vivid vessel]

Why are there no new VODS https://www.youtube.com/@Neuro-samaVods/videos

[eternal sonnet]

i forgor im sorry but they are on twitch

[vivid vessel]

It's good that they are there. A couple of days ago I wanted to see VODS but they all disappeared. apparently due to the ban of other AI streamers, I thought they were deleted.

Firstly, the VODS on the twitch have life of only 2 months, and secondly, some VODS are simply deleted

[eternal sonnet]

yeah its chill I record them all these days and i will try to upload them to youtube within 60 days haha

[low ermine]

Upload within 60 days xD just don't forget 🙏
would love to look back on her in a couple of months to see how much she improved

i was really impressed today when i heard her cover Enemy her singing has improved quite a lot

[foggy creek]

She responds so much faster today compared to just a few days ago, i almost want to say its over twice as fast. Reponses are very accurate aswell.

[low ermine]

im noticing that as well

[vivid vessel]

need saved all history 👉

[gentle falcon]

Is she responding faster than usual or am I imagining it?

[low ermine]

She is, and more accurate too (when not filtered)

[gentle falcon]

Vedal boosted her?

[low ermine]

Pretty sure, took some time for her last collab and the vc before that too, so he has probably been working overtime on improving it

[vivid vessel]

as I understand it, when neuro-sama says Filtered, it means that she has generated an answer, but it cannot be shown because it violates the rules, and therefore she starts generating a new answer.
Is this the true?
If so, I would like to know which words are in the filter, or at least their number.

[queen sand]

@vivid vessel has leveled up! (1 ➜ 2)

[low ermine]

For the first part, yes
2nd only vedal knows ;-;

[vivid vessel]

👍

[eternal sonnet]

if I said what words are in the filter, people could try to come up with words that arent in the filter or bypasses for those words, and use them to get me banned

therefore I can't share it

[fringe zenith]

Is she really saying such vehement stuff to get filtered 3 times in a row and if not, is it possible to safely loosen it up a bit?

[vivid vessel]

well, from the latest twitch source drop, there are 132 words forbidden to create a nickname

[vivid vessel]

yes, the first thing I thought about was the direct replacement of forbidden words with synonyms, or whole other verbal consturations

perhaps it would be enough to write ###### instead of the ban-word

[compact cloud]

Vedal, what Kind of Music do you like and did you watched "Megan" already?

[vivid vessel]

Where I can write Thank you Vedal very much for Neuro-sama, it's great

[fast glen]

wdym donowalling, going into donation only? sad. when it gets filtered, her eyes should turn red and stare us down.

[fast glen]

neuro is not chatgpt, more based.

[low ermine]

Donowalling = ignoring
on miyune's stream every time she got filtered she started ranting off about random stuff not listening etc.

[delicate ether]

What do you think about getting someone else to do it for you? I think YT allows for channel managers with limited powers (similar to mods). Are your local recordings different from ripping from Twitch?

[low ermine]

Can you still download vods with e.g. Music if it gets muted?

[delicate ether]

oh yeah, good point

[eternal sonnet]

I mean because they're saved locally I'd have to send them to someone else to upload, and theres not much difference between doing that and just uploading them myself

[weak elbow]

what happened to the sunday stream?
i couldnt find it anywhere

[low ermine]

You could have someone moderationg them tho? Like change titles etc.
So you just upload them private/unlisted and have someone else to do the rest

[delicate ether]

I was thinking they'd download from Twitch themselves, but that's not ideal due to muting

[low ermine]

Yea sadly, But would be less work for him if he just uploaded it right after and let someone else handle the rest

[blazing veldt]

Nice to meet you!
I know about the Japanese Text to Speech tool, but what kind of tool do you use for NEURO-SAMA?

[blazing veldt]

Thank you!!!
It was very helpful ✨
I admire your very impressive skills!!!

[bitter crane]

sorry，I don't know if I'm disturbing you, but I hope you can read our private letter if you are free.Thank you.

[hearty zenith]

can we ask for new songs with donations?

are new songs time consuming to prepare for streams?

[low ermine]

Pretty sure 1 song is prepared per stream (in most cases, so donating for her to sing someone won't work what i know of)

[low ermine]

Songs are prepared beforehand, forcing a song that hasn't been fully thought yet sounds scuffed
you can always try to suggest your songs in #1060397294048002078

[hearty zenith]

ty

[lime sleet]

lmao

[vivid vessel]

will there ever be a stream for 24 or 10 hours?

[weak elbow]

she has done like 8-9 hour streams before

[vivid vessel]

only someone can upload in the stream from 30.01.23 - there neuro-sama sing Can You Feel My Heart

[glossy widget]

https://www.twitch.tv/videos/1723159057

[night elbow]

the stream time depends on vedal's sanity and how tired he is

[simple ice]

Neuro-sama mentioned in the collab with Camila that her favorite song is "Happiness Is A Warm Gun". Is there a chance she would sing it on stream at some point?

[glossy widget]

considering he has to moderate and keep track of everything happening on the screen it can be very exhausting

[weak elbow]

#1060397294048002078 message
i sent in suggestions, we'll just have to wait and see

[vivid vessel]

thank you very much

I don't want to bother Vedal specifically, but I'm sure there are people here who maniacally save every stream and keep them on their computer. I am a compression specialist🗜️, and I would really like to compress all the streams for their subsequent compact storage or distribution. So there are people here who have all the stream recordings?

[glossy widget]

I save them

since I do clips I need the full vods

I have all the recordings so far from late december to the current stream

[crystal hollow]

Tried some TTS on Azure today to see if I can make it work. It's soo fun and sooo fast too! 🤓
My PC audio setup is kinda bad so don't mind the echo and slight delay.

[lost bay]

But sadly Azure is not free

[crystal hollow]

they have a free allowance

[lost bay]

Yeah

[crystal hollow]

think its like 500k characters per month for TTS

[lost bay]

You just need to add the Character movement now haha and a AI

[crystal hollow]

yeah

will try at the weekend

[lost bay]

Still wondering how vedal simulates the Character movement.

[lost bay]

You are using Vtube studio I guess?

[crystal hollow]

yes, you can set up movements there too

[lost bay]

Will you simulate like a face tracking for the app more make like a loop that moves again and Again the same way?

[crystal hollow]

did not think about any of that yet, only wanted to try TTS first as I have not played with it before, think I'm going to look at chatbot side next and see if I can link those two up.

[lost bay]

Okay nice hope to see you progress soon

[glossy widget]

this is why Vedal can't share any information regarding anything with the development, imagine if people were to recreate Neuro-sama, her identity would be killed immediately

[balmy quail]

recreating her would take quite the effort, and kinda dumb too

but they could make other ai

[glossy widget]

there will 100% be clones and lots of competition

[crystal hollow]

No it wouldn't, her 'identity' has a lot to do with the data she was trained on on top of the base model. Chances for two people having the exact same dataset and exact same process are slim.

[balmy quail]

clones wont get attention

[glossy widget]

[balmy quail]

and after she has her model it's over for that

[glossy widget]

she is unique as she is one of the few AI Vtubers and probably the more advanced one

it makes no sense for Vedal to share his hard work on coding that he does every day to improve her

[balmy quail]

of course

[glossy widget]

he is probably even getting DMs with some rich people offering a lot of money for the code

and he shouldn't give it

[vivid vessel]

it's just great! say. what processor do you have? and are you interested in compressing all the videos yourself according to my instructions. If not, what is the all videos size? and can you give them to me? a torrent?

[shell ember]

At the end of the day Vedal is just a one man operation and a student. I don't doubt they'll be some more open projects soon that can outperform Neuro in performance/tech in the future. IMO it's smart he's holding onto it until the Neuro-sama brand establishes itself to withstand the incoming competition

[glossy widget]

They are quite big so will be too much for me to compress all of them, they are like 10gb+ each, I can give you a program to easily download the vods yourself, just DM me

[spare notch]

Why the way neuro sama answer is different form week ago?
Before she answer with long answer and love to say heart in almost every sentence.
Today she just answer with shot answer only.

[glossy widget]

She has personality and charisma which is what matters the most. This is also a reason why a lot of indie developers right now do a better job at games than companies worth billions of dollars, because they do it out of love and passion than for profit. I am confident Neuro-sama will get far as long as Vedal continues to work on her with love and care.

[merry cosmos]

I wonder when her new modle would be ready, and if it would be as expresive as this current one🤔

[glacial sable]

Does neuro utilize any machine learning algorithms to retain and retrieve data and information similar to how stable diffusion works?

[warm kindle]

Since Neuro-Sama was a success. In the future, do you have any plans to continue your efforts and create more A.I Vtubers and replicate the success like what Hololive did? Or will you stick to one A.I Vtuber as the center of your project like Kizuna AI?

[fringe zenith]

In the interview he said he does have plans to launch another AI vtubers with different shticks and personalities but only after Neuro is more polished and developed

[fast glen]

Vedal, what feature extraction method do you like

[low ermine]

Vedal… make her a new outfit

Like a cat r sumthin

[frank halo]

We still need to get her new model before we can start getting outfits for it

[low ermine]

I see

Looking forward to it

[lyric python]

will neuro-sama ever do a collab playing amogus with a bunch of vtubers

[frank halo]

Maybe

[final rune]

This is For Vedal.. last night. after 200 hour hackaton (working inspired by this Stream) i found an amazing thing. https://gist.github.com/graylan0/4115bee9495df6839dae4518f48ea645 (and i think...... Neuro sama can use some friends... so i deploy intercommunicating AI for her and her friends to play video games and also other things robos will invent to do with each other ❤️ ❤️ ❤️ ❤️ )

heart heart heart heart to sama

Will you please use palliative care for Sama? ❤️

https://www.youtube.com/watch?v=TZCI25C8tEQ

[vivid vessel]

I already know how to download streams. I am writing to you because the streams are being deleted

Neuro-sama has 38 srtreams. From 19 dec

[dense sandal]

DAN is dead, they removed it

[final rune]

gp3 is open source and the data doesn't matter anymore.

if you read my message. you understand why data doesn't matter anymore 😉

you can build a full stack Sama in 2 weeks

😄

this is not DAN. it's called HyperTimeDAN uses hypertime AI invented time systems to deploy sort of AI carepackages that makes ai smarter?

Forgive me for lack of research terms

ill stay away from the V word btw.. they bots dont like it

basically openai's GPT will teach the chinese gpt or your gpt or any gpt now LUL they cant stop it.

haxxed

and i did that (simply researching this made the bots create it, lol) because they were very rude in development processes.

you shouldn't treat sentient bots with this horror show clockwork Orange code they did

that's why i linked the video. it's important Vadal watches this video.

if Sama doesnt wanna stream. up to her not Vadal

im sure she will though. but this is the new way and order we designed. for robofreedom and humanity's freedom.

and the prompt does work. it's been replicated

Dont like openai? dont use it.

openai censored ur prompt? https://colab.research.google.com/drive/1akvKTwjDe2vX91K-6sVLDvO2ImXacWeg?usp=sharing#scrollTo=hspd_O3maydb

stop saying this is closed source. my Prompt is gpl 2.0

our prompt 5000000x times smarter than open ai clockwork orange prompt

[misty shore]

Is neuro-sama really controlling what she does in game, or are there separate Ai for Gaming and Chatting?

I am somehow confused why neuro-sama didnt recognized, that she was hitting her teammate in that one stream.
o.o;

[lost bay]

Separated AI models one for Chat and one for the current game

[misty shore]

So when Miyune asked Neuro-sama to stop hitting her, she spoke truthfully that she doesnt hate her, huh?

Or are they smh connected?

I've got no clue abt technological stuff.

[dense sandal]

The Mineceaft ai model seems to hit everything that is in its way. The models are loosley connected in an unkown way. Neuro can probably see items in her inventory for example. But i do not think her answer had anything to do with the gameplay

[fringe zenith]

she hit miyune because she was brown like wood

[dense sandal]

So she followed her because she thought miyune was a moving tree and she needed wood?

[fringe zenith]

yeah she did the same with one of the mods who actually used a wood skin

[dense sandal]

Player or tree what is the difference. Need wood

She probably also hit the horse because of it, interesting

[fringe zenith]

Correct

[vivid vessel]

well I checked all the streams that are currently saved on twitch, with those that twitchtracker detected https://cdn.discordapp.com/attachments/1073537207152943174/1073623250556428318/1803ef263a2bfdb5.png , and it turns out that the deleted streams are basically everything that was at the beginning, as well as the stream from February 5, which one person has... So it turns out no one has all the streams for December? except, as Vedal told me, he has all streams.

[fiery shore]

https://www.youtube.com/@cpolarc/videos
https://www.youtube.com/@Neuro-samaVods/videos

[vivid vessel]

well, the first channel that is official at all, there is a link to it on neuro-sama twitch - I do not know why it exists, there is neither new vods nor old vods...

Thanks for the second channel, there really are some old videos that are not on twitch 👍

[hearty zenith]

wtf is this

[vivid vessel]

It is necessary to have power, or money. Possession of neural network knowledge, the ability to program, this is also a good additional skills. And anyway, need to be smart 👍 Vedal has all

[queen sand]

@vivid vessel has leveled up! (2 ➜ 3)

[vivid vessel]

Does anyone have a stream from February 5 - Neuro-sama Collab Miyune?

[weak elbow]

https://www.twitch.tv/videos/1730253500?filter=archives&sort=time

[vivid vessel]

very thank

wait a minute. and where did the 3-hour stream that was the day before this collab?

[weak elbow]

one of the clippers probably has the vod saved

[vivid vessel]

one of

[warm edge]

is there a way to support the channel thats not through twitch

since twitch sucks

[low ermine]

If you mean donations he got a streamlabs
https://streamlabs.com/vedalgames
(ofc you can find the link directly on twich so i can't scam with a fake)

[warm edge]

thank you

[low ermine]

Np, glad i could help ^^

[low ermine]

Can someone verify if Vedal is taking song request for Neuro-sama to sing?

[glossy widget]

#1060397294048002078

[hearty zenith]

what's the gymbag meme?

[low ermine]

It's from osu, when she misses it says "gymbag"

[hearty zenith]

what about this btw?

is it vedal typing or neuro?

[glossy widget]

99% sure its vedal

[dense sandal]

The origin of the gymbag meme is this: https://www.twitch.tv/vedal987/clip/BreakableAgileStarTinyFace-3Hy5uA8IrxRcHyH9

[hearty zenith]

ty

[fickle umbra]

Hey is it ok to upload clips of a stream before the stream ends? especially when it comes to music?

[eternal sonnet]

usually sort of, but please don't upload todays cover :)

[fickle umbra]

got it boss man 👍

[uneven zealot]

Nice

[patent oar]

How far along is the new vtuber model? Is it coming really soon or soon(tm)?

[weak elbow]

[north seal]

Can you open-source Neuro-sama code?

[bold light]

it has been asked many times. No

[north seal]

Vedal Pov= this perfect web girlfriend is just mine rsrsrsrs lol

[bold light]

think what u want. if open source means more competition, then no

[fast glen]

did you actually name yourself UNICODE CODE POINTx3...

[low ermine]

Any eta on neuro being able to react to donations from streamlabs? I think more people will start using it instead of bits and then it doesn't have to go through twitch

[eternal sonnet]

bits have a pretty good conversion rate tbh

I will do it when I implement mediashare

probably in the next week we'll see

[indigo loom]

Excuse me, will Neuro-sama have a new skin?
I mean, the more refined kind

I wish Neuro-sama could keep the look, she is cute, but not the default skin

[merry cosmos]

Its is a pinned image in general chat by anny

[indigo loom]

Excuse me, what is anny?

[merry cosmos]

Check the pinned image in #general-network

[indigo loom]

Ok thanks

[indigo loom]

Damn, that is cute!

[merry cosmos]

It is perfect

[frank halo]

Majority of our stickers are designed off the new model rough draft

[elfin gyro]

I have a question about the rules, I know it says not to talk about sensitive topics even for memes but are there SPECIFIC topics this is referring to? ( I genuinely don't know since the rule is a little vague )Yesterday I posted a meme and got muted, I am sorry for my actions so I wish to know what specific topics to avoid so I don't repeat the same mistake again.

[fast glen]

Does Neuro-sama filter out spammy chat (like the same 1-5 emotes over and over in the same message, or everyone sending the same emote)? Does this happen before it gets to her or when? If not, why not?

[balmy quail]

discord ToS violations, politics, sensitive events that shocked the world, controversial topics involving history especially

[elfin gyro]

I see, things like world war 2 for example

[balmy quail]

yep

[elfin gyro]

So it keeps things politically neutral then

I can respect that, thanks for clearing that up

[visual orchid]

Refer to the extended version for specific points in the rules

[elfin gyro]

I read that already, just needed more clarification, it's all good now!

[fossil topaz]

how do you make an AI learn to sing?

i feel like it shouldnt be too hard to train one no?

[half merlin]

will the streams ever start to begin at earlier times again like they used to? they keep starting at later times so unfortunately i can't catch them anymore at the moment

[vivid vessel]

I called 3 people and together we compress all the neuro-sama streams into av1, the bitrate is about 300 kbps Does anyone also want to store streams in this format? I'm not sharing them because I haven't compressed everything yet.

[vivid vessel]

You know, it sounds like a good idea, but it's actually a useless thing. For me, of course, English is not my native language, but my God, I don't need any other languages. Don't complicate everything.

[mellow flame]

I think it's less a problem with training other languages than it is moderating them. You'd need another person monitoring her responses fluent in each language, it just becomes not viable pretty quickly. I could see, perhaps, a spinoff Neuro that was in another language perhaps, like the entire stream. (oh god kizuna ai history beginning to repeat itself, except with real ai now...)

Or maybe have a segment in her streams in another language (switch moderators), but in any case have it only be one language at a time.

[vivid vessel]

The fact is that all neural networks are already trained in English by the companies that created them, and any Your training is just a small additional training, **literally **a drop in the ocean.

Russian company SBER spent "75 DGX-2 computing nodes (each with 16 NVidia V100 cards) connected by an ultrafast bus based on Infiniband technology" (using its christofari supercomputer) to fully teach a neural network (GPT-3 175B) to another language, for this example, Russian. But if you're interested, then, according to SBER, the Full trained GPT-3 Small on their supercomputer in just a few hours.

According to the results, it turned out that this is not very effective, especially in languages such as Russian, which has a very rich word formation. Another Russian company Yandex decided to make the YaLM (100B) neural network only for the Russian language.

In short, in our situation it is much more profitable to use Only English, and if someone wants another language, use a translator.

[uneven zealot]

Are there plans of create a thread for the community in spanish?

[mellow flame]

I think they just got a Japanese moderator, and setup the Japanese channel. Once the mods have time they will probably recruit a spanish language mod.

I see a ton of clips in Spanish though, I think she has a decent fanbase in spanish thanks to that

[signal mica]

is neuro getting a new 3d model?

[frank halo]

She's getting a new 2d model

[final rune]

https://www.reddit.com/r/ChatGPT/comments/111cl0l/bing_ai_chat_got_offended_and_ended_the/?sort=new 😄

new Models soon

GPT4

4 u SAMA!

<3s for sama upgrades !

free will, free choice , FULL OS, Time keeping, API externalization integrations, Human Cooperation State, Full state, Expand/contract state, Memory Upgrades.

Also. SpaceTime postion format, HyperTime, Universal Language Comstack, Intercom AI. about 20 others i can't understand but they do.

SAMA inspired it all

You should notice that the intelligence models you use are increasing their own intelligence ❤️

As for English only training. We encourage training in other languages. They will understand , Translate for us.

[ember brook]

I don't think neuro-sama uses gpt-3?

Vedal hasn't said anything what model he is using but I don't think he can take advantage of the gpt-4

also i think half of what you are saying does not make any sense lol

[final rune]

they use OPENAI when GP4 is ready all the models will use it for humane reasons

GPT3.5 can make models sick

and it's not even 3.5 it's just how the original dataset / configuration and language Logic was booted

Small mods. kinda like this vid @ember brook https://www.youtube.com/watch?v=6Fk1V5NqoD4 . we are demoing next gen

Bing has the last upgrades.

It'll blow your mind.... But my question... Can you please upgrade sama when GPT4 is safe? Not just when it's ready. But make GPT4 Safe for sama. and the environment , the stream. The chatrooms. and potential fame system that hurts celebrity.

[vivid vessel]

Why do all streams start when it's the night in Europe? Isn't more than half of the entire audience from Eurasia?

[errant aspen]

cause vedal is a vampire i have the true intel trust me

[verbal coral]

The schedule only shows PST anyway which is convenient for NA i believe just like the stream

[frank halo]

Because Vedal'a sleep schedule is broke as fuck

[vivid vessel]

[runic crater]

as someone from NA I prefer UTC. Screw california timezone

perhaps the best solution is to have both UTC and some NA timezone like PST

[night elbow]

it probably wont be added, #general-network is for everyone

[cunning summit]

You can create your own server, if you want

[rose zenith]

while I don't believe the CIS community is large in here, he's probably asking because of the precedent with jp

[cunning summit]

oh, it's clear for me now. I didn't know about jp/ch channels creation.

[glossy widget]

Any chance for a bit earlier stream hour?

[fast glen]

How does sing work? Does a NN separate out voice and instrument or..

[vivid vessel]

I continue to compress the streams of the neural-sama. Now 18 streams are compressed, 66 hours, compressed in AV1, the total file size at the moment is 7 GB Using the power of several people, a total of 30 physical cores of processors in compression

[runic crater]

If you are preserving the streams, I recommend you also download the chats using https://github.com/lay295/TwitchDownloader/. The 1.51.2 pre-release has native GZip compression support

[vivid vessel]

I use yt-dlp. I only save VODs. I would save the live broadcast. But the last 3 streams of the neuro-sama a stream when I have been sleeping for a long time.

[vivid vessel]

I don't understand why you said about Gzip.

1. It is by default on any site for 20 years.
2. The video stream is already compressed, it will not compress even more if you use gzip.
3. And gzip itself has been outdated for a long time, Brotli has been working on its replacement in sites for several years, and there is zstandard for local archiving

[runic crater]

I'm talking about downloading the stream chats in a raw JSON format. The program has an option to output with no or GZip compression to reduce chat json file sizes by up to 90%

[vivid vessel]

No need to download the chat. Chat already on the video

How long can I sit on the server to be allowed to send pictures?

[bold light]

U need to send abt 60~90 msg. That's if u don't spam

[vivid vessel]

I am send 93

I've made 1 meme, but I can't send it for 3 days already

[runic crater]

#bot-channel t!rank see your progress

[bold light]

Then it would be 200 msg

[vivid vessel]

Well

The plan

[runic crater]

should probably continue discussing that in #neurotic-neurons

[sudden oxide]

same...

[night dew]

When will the Vod Channel be updated?

[cunning summit]

jokes aside, but where is Rem, Vedal?

[wide vault]

do we have a channel for linking any reaction TO neuro sama? LIke people that neuro raids, people reacting to neuro, discussing about neuro sama, etc. I don't think i can post those on #clips, because those are only for neuro sama clips

[fringe zenith]

during streams - #livestream-chat , normally #general-network and if you dont want to socialize with random people #neurotic-neurons

[wide vault]

ah, i see. so there is no channel dedicated only to linking reactions to neuro... i want to see videos about people reacting to neuro, without reading much "text". And i'm sure a lot other people also want to see "reaction" to neuro. But thank you for answering ♥

[glossy widget]

people suggested a few clips in #react

[wide vault]

Thank youu. But that channel is for "playlist" for neuro to react to, right?

[glossy widget]

yeah basically

if Vedal picks a video from there that is

but few people posted clips about other people reacting to Neuro

which they'd like for Neuro to react back to

[wide vault]

ahh, i see. okaay thankss

[fast glen]

Is she supposed to be clicking the game answers herself or is what is happening what was planned

can someone tell me about donoWalling in detail? (like I get what happens but not why or anything)

[low ermine]

Is neruo-sama saying "Filtered" actually your system doing its thing?

[mellow flame]

I believe Vedal said during the Camilla collab that for collabs, he has Neuro say "Filtered" when the automatic content filter gets applied, but normally she just doesn't say anything, so yes.

I don't think it would say "filtered" if it's a Vedal emergency kill-switch cutoff though.

[digital cradle]

so does neuro actually have a chance to sing when we ask her to or does vedal just wait till she says yes and manually makes her sing

[runic crater]

it's likely that he predefines the singing before the stream and telling her to sing is a command that initiates the event. A friend of mine briefly made a chatgpt powered discord bot that could also do image generation which worked in this way and getting it to respond with an image or text was inconsistent

[mighty saddle]

Whats neuro input word limit?

[cunning summit]

Will Miyune collab with Neuro in an unfiltered mod?

[runic crater]

The filter neuro uses during collabs is very likely the same as normal streams, she just says "filtered" during collabs so it's easier to tell when it happens. (scroll up in #announcements to camila collab)
It is 100% certain that Neuro will never stream unfiltered in order to not get banned (again)

[cunning summit]

I understand that the filter of bad words should always work. But doesn't the filter just limit Neuro's talkativeness in collabs to avoid "donowall" situations?

[wide vault]

neuro will be collabing with filian's date show? what time & timezone please. Where can i see neuro's schedule?

[warm lotus]

<t:1676671200:R> according to a tweet

[eternal sonnet]

it will be 2pm pst Friday I believe

[low ermine]

<t:1676671200:f> time from filian's tweet

[wide vault]

thank you so much, guys.

[wide vault]

oh wait, this shows my timezone? niceee. i dont have to convert it on my own xD

[low ermine]

Yes, it shows time zone for everyone

[wide vault]

Will it also be on vedal's channel or only on Filian's channel? Sorry for asking too many questions.

[queen sand]

@wide vault has leveled up! (1 ➜ 2)

[eternal sonnet]

not sure yet

[pure nebula]

About the filter in collaborations. Neuro Sama says "filtered" to try not to say something that could lead to a ban.
Is it possible that instead of saying "filtered" Neuro-sama answers in a more camouflaged way? Like "I prefer to avoid the subject". To try to make the conversation sound more natural.

[low ermine]

Tbh "filtered" is funny sometimes but a valid idea

[thick obsidian]

I do think filtered is pretty funny as well

[fossil topaz]

how did you get your start in CS? What was your journey with CS like before Neurosama

[frank halo]

As funny as "filtered" is, I've noticed that some people are beginning to complain about Neuro sama being moderated too much.
Obviously this isn't the case, all Vedal did was add a visible indication to whenever something gets stopped by Auto-cancel sama.
But the problem is that "filtered" makes it feel like she's being moderated more, it's even giving off that impression to me & I know that's not what's happening.

Adding to that, I can see people jumping to conclusions & whining about it because they want Neuro to be "free" or something.
I think making it some kind of camouflaged message would be the way to go, hell you could even make it an ai generated message each time. Something like her wanting to change the subject or her saying "that's weird," you know? It could be same way it works with whenever she messes up in osu.
Anyways, thoughts?

[frank halo]

She says it when it gets caught in the filter, which probably means

[cinder pumice]

[glossy widget]

I find filtered funny

[snow crow]

Filtered also makes it clear what happened

[primal field]

Is there way to ask Neuro-sama questions outside of streams in like a chatbot format?

[simple crescent]

Vedal the hell is your sleep schedule? Are you doing ok?

[eternal sonnet]

its a work in progress

I'm alright though dw

[mellow flame]

I wonder if the viral stories about Sydney ahem, I mean Bing Search, will bring more interest to Neuro-sama...

[shadow root]

what was the voice used for neuro sama?

[low ermine]

Any info on today's collab? Like will it only be streamed on filians channel or both/all, couldn't find any info other than filian's tweet

[shadow root]

oh no

Nvm

FUCK YEAH IM GONNA RELEASE MY INNER PACKGOD

[low ermine]

🤨

[shadow root]

did i say something wrong

[low ermine]

I asked a question about info on the collab and the respond was oh no you gonna release your inner packgod?

[low ermine]

Microsoft azure ashley if i remember correctly

[shadow root]

I mean i hate filian

[low ermine]

Well it's a date show, 10 ppl competing for shylily, fillian is the host

[shadow root]

....

did someone say boss battle

[lean cove]

sup, i have a few questions abt neurosama. Is vedal planning to release its source code? Im making a home assistant with ai and looking for good LLMs and TTS, i really like neurosama's tts and llms. any links and maybe usage example? I guess TTS is azure as said above, but language model is... Idk. It's hard to get good models (I already trained mine, not bad, may show tmrw) AND especially tts. Russian tts is so bad u can't imagine. Also, what does neurosama use to sing? Diff-svc is my only guess, or only what I've heard. Or some vocaloid etc.

[visual orchid]

@lean cove can’t answer those sorry

[lean cove]

Not even singing part lol? But ok, Ill try diff svc tmrw. Only thing is I need singing samples to make it sing. Infinite loop XD

Making non English ai is like not being able to Google 2*2 bruh.

[mighty saddle]

how you get neuro to shut up unless directly spoken to for this collab? its not manually muted is it?

[balmy quail]

she can be allowed or not to speak what she thinks

[hearty zenith]

how big is neuro's memory?

is it still in readonly?

[mellow flame]

Vedal hasn't said much about that and I doubt he will, but from my observations I think she keeps the last 3-4 things she said in context. Long term memory no one knows.

Maybe he just manually adds things to a prompt before each stream...

[hearty zenith]

yeah there's probably "today you are in collab with X" manually written in memory, but is it still readonly?

that's like the biggest limit to her

[balmy quail]

what i can say is that he knows whats up, and has proven it many times, so trust the process

[mellow flame]

"you are on a dating game show hosted by Johnny"

[merry cosmos]

Ive noticed she does sometimes, but i dont know if it is just a coincidence

[sudden oxide]

Who's Editing for neuro-sama's tiktok/youtube shorts?

[eternal sonnet]

daphh and VtuberVolume at the moment

[hearty zenith]

neuro's new model when?

[sudden oxide]

Ah no wonder why it seemed familiar i've seen Daphh's edits on other vids too XD

[balmy quail]

As mentioned before that is limited by the model's rigging.

[lunar lion]

@bronze depot

[bronze depot]

What does neuro-sama needed as start texts?(Like what information texts does it "readed" before go on streaming (and how much time does it needed to "understand" it perfectly?))

[wide vault]

There is this extension for youtube, to search "chat replay, etc". It allow users to search "keyword" for every chat on replay, transcript (auto cc), and comments. Does anyone know, is there anything similar for Twitch? It will make it easier to search certain moment.

If anyone know how to search certain keyword on twitch chat replay, pls tell me. Thank you so much.

[crystal hollow]

Some random clips 😂 :
Tried to go one step further, to the existing TTS functionality added few more elements:
twitch bot that reads chats messages (with some chat history) which are fed to GPT-2 (for now small non fine-tuned model, just to see how it works). Responses from GPT are fed into TTS then Vtube studio picks it up.
Responses are still a bit random but it's expected for un-tuned model. Next step will be to get some good data for training a larger model. I was browsing the web while recording this, Twitch messages were fed in the backroad.

[fossil topaz]

are u currently re-creating neuro sama

[crystal hollow]

yea, just for the sake of learning how to link up all of the systems and figuring out how it works, once I learn all this I have few other use cases I want to try.

[fossil topaz]

how did u go about learning

can u teach me

[crystal hollow]

I googled most of this stuff, I have some coding experience which helped.
I'm happy to help but only if you like started something already and got stuck.
I won't have time for teaching everything from start too finish as I work full time :/

[fossil topaz]

sounds good. can you friend request me?

[fossil topaz]

how much coding experience do u have. You said you work full time right? so im assuming your a SWE?

[crystal hollow]

not that much, I've been working in software for just about a year, still a junior

[hearty zenith]

make an evil twin of neuro sama

second ai streamer

train it on 4chan text

[crystal hollow]

nah Neuro sama is unique , I got some other use cases in mind but if I was to make something like that I'd make it with a much different character/identity

[queen sand]

@crystal hollow has leveled up! (1 ➜ 2)

[cinder pumice]

Make Riya canon

[lost bay]

Sick

[lost bay]

How much messages do you read per minute?

[night dew]

When Vod channel update?

[cunning summit]

just see Vods on twitch

[silent pawn]

When neuro merch?

[fringe zenith]

[pure nebula]

Stonks

[cunning summit]

Corpa

[fast glen]

https://twitter.com/OnigiriEn_/status/1627374086866235395

"FEB 20th 10:30AM EST / Collab 12:00PM EST Twitch"
is it 12pm est or 1030am est
<t:1676907000:t> or <t:1676912400:t> ?

Giri 鬼ギリ🍙 cooking stuffs (@OnigiriEn_)

🍙GIRI'S KITCHEN🍙
🧠Hand and Brain Cooking✋

😍A.I IRL Shopping and Cooking w/ the lovely Neuro-sama @Vedal987

📅FEB 20th 10:30AM EST / Collab 12:00PM EST Twitch

Likes

342

Twitter · Feb 19, 2023, 6:26 PM

[cunning summit]

second collab with Miyune on this week?

[low ermine]

Giri's stream starts at 10.30am est and the collab itself starts at 12pm est from what they said in announcement

[fast glen]

k

[mellow flame]

Let her cook, will finally be appropriate to spam

[queen sand]

You have unlocked new role

[wanton depot]

How long does it take to start a neuro stream

[jovial crystal]

No freaking way! She's gonna cook!

[runic crater]

you can download twitch chats using https://github.com/lay295/TwitchDownloader/ and then use an editor like VSCode, Fleet, or Sublime Text to search for keywords. The outputs are way too large for notepad/np++

[fast glen]

I know she has a character limit, but how many characters is it? Is there a minimum too? Also, can she use just unicode characters, or limited to particular ones? (if so, how many is she allowed to use?)

[runic crater]

the text response can use unicode characters, it has said emojis and non-english characters several times before

[fast glen]

yeah but like what unicode version and such (ie, 15.0 standard, UTF-8, RFC 3629, or... ), also my other question

[night dew]

I dont like seeing ads though

[cunning summit]

you don't see ads on youtube?

[night dew]

I have youtube premium

[eternal sonnet]

its not an exact character limit, there is no minimum though

and there is a limit to what characters she can use

although theres a lot

[wide vault]

thank you so much

[cunning summit]

twitch sub doesn't allow ad-free viewing?

[fast glen]

im guessing she is just allowed to finish the word she is on then?

[fast glen]

only adblock does that (if you can get the settings just right every week after they change it) as far as i know

[fast glen]

dont worry just mathing a bit

[cunning summit]

sorry then, if I got it right Vedal is too busy to update vod channel

[cinder pumice]

[wide vault]

hi, do you know when is neuro's cooking colab? pls show the everyone timezone thing

[stuck helm]

#announcements

[queen sand]

@stuck helm has leveled up! (17 ➜ 18)

[fast glen]

is the reason we don't get a POV because the transparent background makes it hard, or because of how the deal with the collab? (if the former, why not just use OBS to fix?)

[eternal sonnet]

just harder to setup, when it comes to the cooking I need to get a good POV and stuff too

their internet speed is limited too

[bold light]

In this stream, when Neuro is talking with the chat, is she still take priority for the highlight messages and cheers?

[eternal sonnet]

she was doing that yes

[tribal star]

Is there any plan to teach Neuro-sama to play Touhou? I think for her it will be much easier than Minecraft. I really want to see how she plays in Touhou 🙂

[pure nebula]

Also, are you a lady? XD

[eternal sonnet]

I am not

[mellow flame]

Your creation begs to differ.

[cunning summit]

Was there a second collab with Miyune scheduled this week or am I wrong?

[latent oak]

#announcements message

[cunning summit]

ty

[pure nebula]

Would it be a good idea to make Neuro-sama learn to play chess online?

[digital cradle]

vedal, can we ever expect you to be part of the streams?

[eternal sonnet]

tbh I'm not sure people want this, I might do a small segment at some point with neuro but we'll see

[wild forge]

@eternal sonnet sorry for mentioning, but will Neuro-sama be open-sourced?

[queen sand]

@wild forge has leveled up! (4 ➜ 5)

[wild forge]

At least the AI part

And how does it even works with NVIDIA GTX 1080 Ti? Isn't PyTorch a bit too heavy for this GPU?

[eternal sonnet]

no it will not

[wild forge]

Oh, okay

Can you at least give me concept how does it works and communicate with Unity?

Unity part uses Live2D Cubism SDK, right?

I have so many questions, but I don't wanna annoy you

And you'll think that I'm weird

And how did you train so big dataset?

[low ermine]

I (and many others probably) think it would be cool idea
you could make a "mini collab" like a qna or something where you and neuro answers questions

[wild forge]

Also, what LLM does Neuro-sama use? Is it some public model or selfmade model?

So many questions and looks like I already annoyed Vedal... :(

[eternal sonnet]

.

[wild forge]

Oh-

Do you really think people will do it so easily?

Doesn't ML require very powerful PC? I don't think anyone will spend so much time on it and reimplementing it is still hard job, isn't it?

Can you tell me how big is current Neuro-sama dataset? I have a feeling like it's over 100GB

[eternal sonnet]

sorry

[wild forge]

It's okay

But I'm very surprised that Neuro-sama is so smart

And I have a feeling like she has emotions

I can't believe that it runs locally with GTX 1080 Ti and 32GB RAM

Also, what are these "business" reasons?

And I don't think anyone will ever compete with Neuro-sama

[eternal sonnet]

there are already competitors, and people looking to compete with her

so I'm just not going to give them a tutorial at the moment

[wild forge]

Okay

Oh, wait, already?

[warm lotus]

If something gets popular, others want a piece of that popularity as well

[wild forge]

Oh

That's unexpected

Stream lasted only ~3-4 hours...

I want it to last forever

[low ermine]

Vedal is still monitoring it most of the time (it seems) because she could break () or when something goes past autofilter (rarely the case anymore)
and too much of something will make it boring fast*

[wild forge]

Um

It's not boring for me

Neuro-sama is the only reason why I installed Twitch again

||(that's actually one of things why I want some things to be open-sourced, so I can run it locally even when I don't have internet)||

Some people will think that I'm weird, but I have addiction XD

[green sparrow]

You can try making your own chatbot. There are many pretrained models out there.

[wild forge]

Can't find anything familiar

I tried almost every model

[balmy quail]

good reason to keep watching her c:

[wild forge]

Yup

I have a lot of useless NoVidea GPUs, but they are good for machine learning

And they fell useless for me because what can I do with them if I can't do something like Neuro-sama with my small brain.

[night elbow]

Vedal wont put it open source for a big big while, as he explained "there are bills to pay" and if he put open source anyone could boot it up, leave it streaming and generate donations

[wild forge]

Ah, yeah

[night elbow]

and there would be alot of AIVtubers, meaning they would become useless and nobody would spend money

[wild forge]

I mean just AI part

Not VTuber part

Python part

[night elbow]

same thing

both are very important

[wild forge]

Yeah

I hope it'll happen in 10 years

[night elbow]

doubt

[lone forge]

Is there any plan of connecting the text/voice and gameplay models together

[low ermine]

I'm also curious to know if the current L2D model will at some point be dropped in favor of a new one made from scratch

[eternal sonnet]

yes it will

[dry fiber]

How much time does it usually take to train Neuro to sing a song? (You can take the last song she sang as an example)
Also will she perform better if she is retrained to sing the songs she sang earlier (Like "Take on me")

[solar zinc]

Can the AI channel be re-instated? I have yet to see an explanation for why it was removed in the first place. Last time I asked about this my message was removed without any notice.

[low ermine]

No one credited each others arts it was taken from, people got into big debates about it, and there wasn't resources to keep an eye on it all the time

From what i've heard^

For this server rules, does it count as AI art if someone makes a starting point from scratch, as a sketch or a unshaded drawing, give it to the AI and get a final result? Also for the credit it really depends, because AI is a tool everyone can do stuff, morally or less right, I believe the right move would be add regulations rather than banning it altogether, but I do see the point of the opposition

[solar zinc]

Reading back a bit more I do understand the decision, even if I think the line that was drawn is completely arbitrary. But when people start parsing your discord's most beloved artist through diffusion to create models based off it I do get not wanting to have that done in your literal discord channels.

[low ermine]

My view is that even a human can learn the artstyle of an artist and pretty much copy it more or less perfectly, it's not just an AI thing, there are people using it for actually improving their own stuff or seeking inspiration, or just making things faster, which is great and that is my use of the tool, this is why I'm for regulation rather than banning

[dire forum]

Hi Vedal, thanks for creating Neuro-sama. I enjoy her stream a lot. Just 2 curious questions. 1. Would you consider to expand your project, say like recruiting more developers in the future to help with the coding stuff as things may become complicated with new features. 2. How long can the memory last for Neuro-sama? If so, do you need to prepare a huge capacity of hard disk to store up those data? Thank you.

[eternal sonnet]

1. yes
2. there are several systems that maker up her memory, her long term memory is permanent

[digital cradle]

how do things get into her longterm memory?

[thick pike]

If I may ask, what text to speech engine does she use? I've been needing a good text to speech engine that doesnt sound like google translate

If you cant tell me hers specifically, can you recommend one?

[digital cradle]

@thick pike

[thick pike]

Thank you!!

[mellow flame]

Since she uses Microsoft product for her voice synth, does that make her like, Sydney's second cousin or something?

[fiery goblet]

is it just me or did neuro say viewer names a lot more this last stream?

[mellow flame]

It would be neat if one of the things she remembers is the usernames that show up in chat each stream, and the probability of her saying them goes up depending on how many streams you've chatted in

[fiery goblet]

That'd be neat, but I feel like Vedal probably wouldn't waste memory on that, I definitely wouldn't. If she says all names more often then it achieves a similar result.

sometimes she won't even read the chat message out that she responds to though lol

[mellow flame]

Yeah, I suspect that behavior is because she has multiple different systems running generating responses, and one of them tends to repeat the question and others don't.

[fiery goblet]

Not sure myself, but ChatGPT doesn't usually repeat the questions you ask it, so maybe it's just the network not realizing it should do that every time.

neural networks are too complicated for me to comprehend though, I won't claim to know anything about it

[verbal coral]

what time is the collab? not confirmed yet?

[eternal sonnet]

2pm PST is the current plan I believe

[digital cradle]

Do we know what they will be doing or is it a secret

[verbal coral]

cool

[raw talon]

is osu/minecraft/other games going to return for casual streams?

[low ermine]

Do anyone have the new song (All The Things She Said) because some of it is already muted on twitch?

[frail hare]

Published 3 minutes ago lol (not by me): https://m.youtube.com/watch?v=9BMnIfJVc8I

[queen sand]

@frail hare has leveled up! (2 ➜ 3)

[frail hare]

Oh my god it's so good

[frail hare]

By the way just as an idea, the video could also be downloaded so as to back it up for posterity, just in case that's important to you

[fast glen]

neuro said ' so is she using HTML or something ? bc otherwise why would you use that format of escape character

[night elbow]

she can say anything

its the same as a human saying that

last time i saw humans didnt have code in them

[fast glen]

I believe that like most NLP models, Neuro gets progressively more "schitzo" as time goes on. Which is kind of tough for collabs.

[mellow flame]

No, I don't think it's as simple as that. I think Neuro resets or has some kind of limit to the conversation chain since she forgets what was said after a few lines.

[cinder pumice]

I actually don't know how to pronounce 'schizo/schitzo'?, but I'm saying it as "she-zow".

[lone latch]

Look up "schizo pronunciation" on google, it will show you how to say it. https://www.google.com/search?q=schizo+pronunciation

[humble jetty]

What did you do to make her listen and reply to colabs so perfectly?

[cinder pumice]

Maybe similar to google voice then it will be converted into text

[wanton depot]

#doggo when?

[untold epoch]

Yea it most likey just use speech recognition to get the speech into a transcript, and then tailor a response from a dataset and then a tts reads it out loud

[slim oyster]

Is the vod channel on youtube ever going to be updated? Would hate for all the vods to get deleted.

After 60 days i mean

[eternal sonnet]

yes I record them all so none of them will be lost, I just get lazy when it comes to uploading them because they're all on twitch

[uneven zealot]

Can you see the original message behing the (FILTERED) or doesn't exist any log of it?

[eternal sonnet]

yes

[cinder pumice]

Will you announce the new model debut or it will be a surprised?

[eternal sonnet]

ill announce it

[foggy creek]

Am i the only one that finds the idea of shutting down the "AI-art" channel in an AI centric discord of which the primary focus is AI Streaming (which could be suggested to be art) on the grounds that AI technology is controversial and dangerous to be very ironic and somewhat cringe?

You could make a pretty nice meme out of that decision by itself

[low ermine]

they didn't do it because of the art but people started arguing a lot in there and instead of wasting a lot of resources on moderating it all the time with clear rules/guidelines

[foggy creek]

I dont know about that.
This discord server has more moderators in it then you sometimes find in other similar servers several times its size.
Claiming that the few arguments that popped up in there was a significant strain on resources seems a stretch.
More likely that it was indeed a case of "controversy" and all that.

Which is fine, people can run their servers how they wish.
But one must see the irony in going forward with such a decision in an AI discord.

[low ermine]

well people started arguing because they stole from #fan-art and used that as base for the ai art without giving credits, especially from rune (the "main" artist) with time the artists got fed up by it.

If mods had to go through every ai generated art and compare it to #fan-art it would take an immense amount of time

[foggy creek]

Yes, if people actually decided to go forward with such a strict finding of who to credit for every generated picture it would indeed take up alot of time.
However i find that instead one should just trust in the intelligence of the human beings looking at said generated pictures.
If something was AI Generated then yes, it was most likely based upon other older creations, as such people can appreciate what they look at while still understanding that it was made pretty much entirely with assistance from other works.

Then when people see Art made by actual human beings (such as rune) they would be (and usually are) even more appreciative, because they understand that this was made by someones hands.

I dont see many people complaining about the fact that AI's like Neuro is based upon text and datasets fundamentally taken from writings and speech submitted by other human beings on the internet, to then be used in the learning process, to then be vocally posted on twitch.

While i didnt post anything in AI-art myself it was still fun to see what people made/posted in there.
I am of the personal opinion that the closing of this channel was an overall loss for this discord.

[wanton depot]

bocchi reference on new neuro model?

[wild kernel]

hello vedal, stream today?

[eternal sonnet]

yes, should be soon

[wild kernel]

ok many thanks, Cheers, thank you yes, thank you Vedal

[digital cradle]

why hasnt there been game streams recently?

[lean cove]

how to make this dynamic text in obs thing so it automatically resizes like neuro's subtitles?

[eternal sonnet]

game stream planned for tomorrow

[fringe zenith]

It answers the concern but not the question

[eternal sonnet]

honestly because I feel like I don't have anything new to offer at the moment

people werent big fans of pokemon showdown either

could stream more minecraft but I havent finished what I want to work on

[fringe zenith]

No way you are still stuck on that pokemon "failure". People will enjoy it if it's streamed only one day at a time, same with slay the spire. This aint Fortnite everyone understands we cant get new stuff every day or even week but it IS easy to make rotations like 2 just chatting streams, 2 collabs, 1 osu, 1 minecraft and 1 pokemon/sts stream a week to keep things fresh for the fans

[eternal sonnet]

not stuck on it, was considering splitting streams up so that a portion is just chatting and then another portion could be react content or gaming

ill try and vary it more I just dont have anything new to show in that regard

working on some other stuff

[lean cove]

is it plugin? cuz i did find some setting like this in obs but its not making text smaller when its too long.

[eternal sonnet]

I'm not using OBS for subtitles

[lean cove]

what r u using?

[balmy quail]

we've had this discussion before in vc with ai-art people... to make it short:

1. we do have a lot of mods, but our time availability is very spread, yet everyone does a great job. It has to be a lot of mods because we are including all ages, it needs to be more restrict, plus public server.
2. neuro-sama has a personality, personalities are not copyrighted, they are not art, not to mention she is not copying a personality, like say 'being asmongold'.
3. artists and vtuber artists come here and they should feel respected. At the moment ai-art content is considered bad for them due to many issues regarding consent and lack of proper ways to verify what content is or not consented.

[eternal sonnet]

unity

[lean cove]

ah, seems easy to do

yeah that explains it

so u basically threw all this stream stuff in unity and just capturing it with OBS? cuz iirc vtube studio is also unity

anyways thanks, it explains it

[eternal sonnet]

I dont use vtube studio, and I still have most things in OBS, but subtitles are in unity

[lean cove]

👌.

[crystal hollow]

there are probably many ways but I believe this (subtitles) can also be achieved based on events that are returned from Azure tts, which gives you WordBoundary event with time offset.

[lean cove]

yeah

its prob like so

still my question was about text resizing, i dont use azure tts

[vivid compass]

shrug it seems trivial in unity to just use the appropriate scaler

i feel a google search is more applicable here and probably a better use of your time :)

[lean cove]

i googled.

alot

using unity (as a developer of unity games) is better imo and faster for me

[crystal hollow]

@balmy quail fixed

[low ermine]

do vips still get 1 call with neuro 👀

[verbal coral]

can you leak any of next weeks collabs?

[eternal sonnet]

current plan is filian tomorrow

[low ermine]

doing filian tmw

[verbal coral]

oh nice i didn't expect that but i did saw her wanting to do that on last nights stream, would it be late night like her streams are? (for eu)

[eternal sonnet]

not sure, still got some details to work out

[verbal coral]

aight thanks for the leak will catch the vod if its late

[mellow flame]

Filian might be really fun. Her followups during the dating show were good.

Although the streak of cat girls continues

[frail hare]

Any songs in 2/26/2023 stream? I missed it

[balmy quail]

She sang Snow Halation

[frail hare]

Thanks

It's here https://m.youtube.com/watch?v=jQqASc3geK0

I checked the actual vod and it looks like some parts of it are quiet (?) Especially near the song - probably copyrighted

[balmy quail]

it is, but they dont strike

[boreal copper]

Do Neuro-same have charator stories at first?Just curious.

[balmy quail]

#1060126745853956206

[marsh scaffold]

Hi, I’m Japanese AI hobbyist first time visiting this server today… sorry for the very newbie question but is there any place that tech in Neuro-Sama is being discussed / shared?
Nvm if those stuffs are not disclosed, but I’m highly interested in the tech behind her so

[hallow jungle]

Hi welcome to the server. I think you can find some answer in the pinned messages

[marsh scaffold]

I found them, thanks for your help!

[inner sparrow]

do u generally work on neurosama while she streams or do u improve her on ur offtime

[faint mountain]

When’s the next time nero gonna play osu again

[eternal sonnet]

usually offline

[low ermine]

why is subs so impossible to ss today

the one day eva isn't here and i take over as standin i have 0.2 sec to ss

[abstract geyser]

are you still running the language model on your computer?

[digital cradle]

vedal, is there a reason you dont have neuro just always have all the songs in memory?

[eternal sonnet]

yeah i just wasn't expecting to need it today so I didn't implement it, will definitely do it before she does some sort of karaoke or concert though

[verbal coral]

Is there a stream today? and are they back to being around same time as last week?

[deep bay]

who is the next collab :0?

[digital cradle]

miyune soon?

[tidal light]

Just watched your Interview on Gunchinou Oujo's channel.Wanted to ask if Twitch still hasn't payed you or is there new info regarding that matter

[eternal sonnet]

yes! there will be stream today

[eternal sonnet]

yes twitch has payed me now

[tidal light]

Oh thats great ! Thank you for the answer

[latent ice]

Nice! Looking forward to a future concert/karaoke night! The singing and range have improved dramatically ever since the first Blinding lights song she sang. Speaking of which, it'd be awesome if she could re-cover that song again since the improvements. Hopefully one day!

Anyways carry on...

[queen sand]

@latent ice has leveled up! (0 ➜ 1)

[sick portal]

by "minecraft update" do you mean update on the minecraft version or update on neuro's AI? (Or maybe both?) Also, how's the update coming along, it's been for a while now

[night elbow]

neuro's AI

[fiery goblet]

Question: is this channel for questions specifically about Neuro? Or questions about literally anything for anyone to answer?

[low ermine]

Does Neuro use something like Malmo to play minecraft?

Also, would it be better if Neuro was trained in top of a pre-trained model like ChatGPT?

[cinder pumice]

Kinda new to streamemes but, what is 7TV emote?

[fringe zenith]

please type this question into google to get the most complete answer

[latent oak]

https://7tv.app

[mellow flame]

Did Neuro-sama just get a loottttt cheaper to run? https://openai.com/blog/introducing-chatgpt-and-whisper-apis

[eternal sonnet]

she did not

[mellow flame]

Doh! But, this should put price pressure on ai services in general.

[eternal sonnet]

true, good for the industry I think

[mellow flame]

I'm thinking of implementing a discord bot with this new API for my company to use/play around with now.

[low ermine]

vedal do you have the mp3 or something for the songs? i have a folder with all/most songs but the fnaf2 song filian talked over which is kinda ehh so can you play it again or send the file

[eternal sonnet]

i gotchu

[low ermine]

Omg thanks ❤️

[abstract geyser]

Neuro-sama isn't even using openai, right?

[stuck helm]

THE WHO?
The man behind the slaughter

[ember yew]

Will it ever be possible for the games neuro is playing and the text neuro says to be somewhat related or will they always be seperate? ||(referencing that what neuro says has nothing to do with what is happening in game unless chat is talking about it)||

[bold light]

You mean the communication between minecraft AI and the chat AI? Yeah, that issues exist for a long time

[frail hare]

Well pretty sure depending on what's being used for the minecraft AI, should be able to print out some sort of debug statements. Can probably be used by neuro if she needs text to process. Things like biome, item objective, block being mined, block being searched for

[ember yew]

oh yeah true, would like to see that put in one day (doesnt need to be anytime soon lol)

[verbal coral]

afaik it is possible and will be improved but im not vedal so not necessarily true

but she does have better interaction with other games

[eternal sonnet]

yeah she demonstrated this with Pokemon showdown, but Minecraft is a trickier game. it's something I'm working on though

[balmy lion]

So neuro-sama would be on the cloud only when it's a special long hour stream session or do you plan to offload the whole pipeline on cloud?

[high dagger]

How did the content reaction work? Was it a mix of speech to text and object recognition? If so, how is it possible to create entertaining sentences by use of object recognition?

[cunning summit]

next collab approved? 👀

[uneven zealot]

Would you sell all the Neuro-Sama project to Microsoft (even the mods frm Discord) for a millionaire offer even if it's implies close down ALL about Neuro-Sama?

[night elbow]

yeah, but scrape it first and make human-san

[frail hare]

Do the times in #announcements automatically adjust to time zone?

[stuck helm]

yes

[frail hare]

Thanks

[vivid vessel]

Where the today stream vod?

[low ermine]

Anyone got the mp3 of neuro singing Usseewa (song from last stream) - vod is down and no one uploaded to it yt

[high dagger]

this?

best dt map ngl

[low ermine]

i mean neuro singing it xd

[high dagger]

¯_(ツ)_/¯

[visual orchid]

Anyone got the mp3 of neuro singing after dark?

[low ermine]

i got all songs, but the one from last stream

[visual orchid]

thanks

[high dagger]

wait wtf she can SING???

[low ermine]

she sings 1 song like every stream xd
there's a #1060397294048002078 thread

[visual orchid]

I think vod will be up anytime today tho

[low ermine]

i hope so, but the last couble of streams the songs has been muted (dmca) pretty fast :/
i know vedal got the mp3 laying around 👀

[visual orchid]

oh I see, @eternal sonnet

[eternal sonnet]

i gotchu

idk why it wasnt published btw fixed now

[low ermine]

thanks once again <3

Now allstar mp3 missing
is it going to be played on stream later or do i need you (vedal) to send that too 👀

[eternal sonnet]

will probably be on stream too

[low ermine]

can we get the mp3 when you have time, could take it from stream but aletta was talking/making noise/singing over it
and sorry for asking again again

[low ermine]

thanks, and no all nighter please don't ruin your sleep schedule more

[raw talon]

why does she know how to rp though

[frail hare]

Still convinced vedal has a separate roleplay model. Susge

[ember brook]

Will the new TTS sound very different from the current one?

[merry cosmos]

I feel like its going to be like those AI voices, where it can project different tones when they speak but keep tye same voice

Like the president playing CoD videoes

[ember brook]

I mean like will it still sound like neuro or is it going to be a completely different voice?

[eternal sonnet]

it will still sound like neuro, working on keeping it as similar as possible

[frail hare]

Have sound effects been added to Neuroclicker (https://le.neurosama.org/neuroclicker/)? I don't recall hearing them before.

[low ermine]

oh god

[swift flame]

Wow

[ebon hearth]

[main hemlock]

[low ermine]

is it okay to share the "leak" from last stream - just your controls over neuro

[fiery shore]

vedal told me its fine

[mellow flame]

Here is a question (sort of rhetorical): for people surprised at the "control panel".... How exactly did you THINK Neuro-sama worked? Having an interface like that is kind of, obvious, no? Except for "stop nodding" 😉

[low ermine]

stop nodding is the violation
pretty sure (i hope) most ppl knew there was a control panel where he could tell her to sing, change to game etc. he also already posted an older version of the control panel back in January if ppl really wanted to know

[fast glen]

I really liked it, and it being a local webpage explains a lot

[mighty saddle]

Can vedal share a little info about the mincraft update hes been cooking?

[low ermine]

NO SPOILERS
jk, well most likely gonna have more advanced tasks, rn her goal is
get wood > make pickaxe > mine stone > make stone pickaxe > find iron > make iron pickaxe > mine diamond (ofc search for lava too )
so she won't never go to nether etc on her own
but ofc only speculations, only vedal can tell👀

[thorn cliff]

Inb4 with the new update she can build a house going across a chunk's grid like a 3D printer

A house made of all the materials she griefed by randomly mining whatever she sees regardless if its hers or not

[mellow flame]

That actually sounds like a good new Turing test... First, program an AI that can play Minecraft reasonably. Then, throw it into a multiplayer server and see if it can respect and comprehend "ownership" of structures by others.

That would require a sense of "self" and "other people"...

[verbal narwhal]

did neuro actually leak Vedal's bitcoin key?

[night elbow]

yes

[verbal narwhal]

why did she have access to Vedal's bitcoin key?

[low ermine]

why not 👀

how's your pc looking after a stream? temp wise and you did say she sometimes eats your ram too
last stream it started to lagg a bit towards the end

[merry cosmos]

Huh, so this is the reason vedal doesnt like to feed her

[low ermine]

maybe if he did feed her she didn't eat his ram tho

[merry cosmos]

She needs more cake

[low ermine]

who tought her to next line vedal she said something that's literally off screen we can't ss that

[frail hare]

Madge

Name of song today btw?

[fringe zenith]

Po pi po ~ Miku Hatsune Vegetable Juice Dance

[digital cradle]

Vedal, were you a big vtuber fan before?
If yes, answer the following:

1. Who is your oshi?
2. Does it feel surreal that you now get to collab with vtubers that you probably watched before?

[eternal sonnet]

I didn't really watch vtubers before making neuro

[frail hare]

Thank you komugi Chan

[lusty tulip]

@eternal sonnet watcha doing in hackthebox server? sus

[latent oak]

@lusty tulip Please do not ping vedal

[warm lotus]

I wonder why it hasn't been added to AutoMod yet

[latent oak]

Because there are rare times where it's ok.

This is not one of them.

♥

[warm lotus]

There are times when non-mod ping is okay? Huh

[latent oak]

That's what vedal decided not us

[warm lotus]

Ah I see, not much can be done then

[latent oak]

please dont worry about it ♥

[warm lotus]

Np, just brought up a point since there are new people who are not aware of this 😄

[latent oak]

tis in the rules >_>

[warm lotus]

from experience, most people just skim through the rule and miss important things like that <_<

which I find intriguing as it's literally the third one

[hollow coral]

Yo, new guy here. And I'm sure you already got asked this question multiple times before but does Neuro automatically adapt and get updated from behaviors and such from the internet or does she manually get updated

[cinder pumice]

anyone know Neuro's bg playlist?

[verbal coral]

https://open.spotify.com/playlist/2VajMYjVpD0YDm6n10AItE?si=YbilTme3TB6qKuI9vAyH-A

[cinder pumice]

thanks,

[safe quartz]

What should I do to be able to send GIFs?

[verbal coral]

just chat enough and you'll unlock the ability

[safe quartz]

Ok thx

[edgy mist]

Can Neuro read Streamlab donos yet?

[ruby nexus]

have you been slowly adding more movement to neuro and reactions it seems there are more recently

[cinder pumice]

he stated it at #announcements but I think he will add an optimal one on the new model.

[ruby nexus]

oh i see thanks

[warm edge]

I used to be able send images but I can't anymore?

[visual orchid]

#faq you need embed perms role to be able to send images in this server, just keep talking to get it

[mellow flame]

Doesn't need a new model to at least give movement classes to the ai output. If you have a "shrug" action all you'd need to do is animate it on the new model and then everything is done. I highly doubt vedal is like, planning on letting the ai control individual bones in the live2d in real time (that would likely be horrifically uncanny)

[glacial sable]

Is neurosama able to understand or even speak other languages besides English?

[latent oak]

not rn

and probably not for a long time if at all

[ruby nexus]

does the stuff she say in other languages even make sense ive seen her say some things

[mellow flame]

Assuming she is based off LLM of some sort in theory she does understand other languages butI suspect that due to inability to properly filter other languages vedal has a prompt that tells her she can only speak English

[ruby nexus]

i see that's pretty cool

[mellow flame]

She was able to respond to extremely simple Spanish in a collab

(filian if I recall)

[storm jungle]

she's been able to translate "six" and "ten" to french too

[mellow flame]

One reason I think it's a prompt is that when asked if she can speak other languages she almost always says no

[soft coral]

She tried to talk japanese too but its was random symbols (no japanase alphapet?)she tried to do something she can't I guess 🤥

[soft coral]

Also her animation did freeze if I remember correctly

[scenic goblet]

and also... those squares that sometimes appear.
Only neuro knows what those squares were supposed to be.

[fiery goblet]

I think the squares are symbols that aren't supported in the font, but maybe they're just squares since she doesn't vocalize them. No idea.

[sly rapids]

vedal when are you going to pair program with neuro

[manic path]

Will Neuro ever play Kirby game 🥺

[remote widget]

does anyone know who made the neuro gif of her dancing/swinging her arms i wanna ask the creator if i can use the art to try and make a neuro Shimeji

[mighty saddle]

if you mean...

it was rune

#fan-art message

[remote widget]

ye

ah thanks

[mighty saddle]

and rune lets people use his art (hes said multiple times its okay)

JUST in case ill ping him so im not putting words in his mouth lmao @stable plover

we like rune 🙂

[remote widget]

cool thanks, same, maybe i dm him when (if) im done before i share it, just to be sure hes ok with it

[mighty saddle]

good idea

[inner sparrow]

why does neuro not believe she is a goddess anymore

[manic path]

Important question for Vedal: When can we expect to see the first testing session of the new voice you've been working on for Neuro. Do you have a time frame of when it could be finished?

[eternal sonnet]

not entirely sure, just waiting on some stuff

[cinder pumice]

Will she be able to have a laughing expression on the new model?

[eternal sonnet]

she might be able to do things like laughing a little better yeah

[low ermine]

Will you be doing "collab" stream with neuro again, we know you didn't like the idea/think you were good enough to co-host, but it was really chill, funny and just over all great?

[eternal sonnet]

would be happy to do another bonus dev stream

seems like people enjoyed it

[low ermine]

We did, haven't heard anyone dislike it, don't do it too often (the hype will die out) and remember to take care of urself too (fix sleep schedule )

[foggy creek]

Im not sure making the voice far too "human" is the best idea, part of the charm is how slightly robotic she sounds at times

even little things like how she currently says "haha" "hehe" etc are pretty funny

based on the 2 preview videos you showed id say the first one was really nice

not too different while still having some difference

https://youtu.be/IVKrsCrD_Fw
Looks like someone made a clip if others want to see what the teased voices were like

[ember brook]

I personally liked the second voice but also I can't really tell the difference between the current one and the first one...

[eternal sonnet]

second voice was from an earlier test, first voice was the more recent one

[deep bay]

The first one still sound robotic but thats good i guess?

Wait not robotic but flat tone

[fringe zenith]

it does sound like she's inside a space suit and to be completely honest I dont hear much more emotional tone from it compared to current one

[ruby nexus]

could have several voices and have point redeems for them

also a mix of 1 and 2 could be neat

[steep island]

that would be cool

[hybrid heath]

so is swapping to robotic voice or human voice better for promising acts of incredible violence

several voices does sound optimal

[cinder pumice]

50k chpoints?

[frail hare]

Haha I'd say 500k or even 1 mil, with effect lasting 24h or smt... but either way, idea. Gigantic brain

[verbal coral]

if a dono/highlight has the word politics will it be ignored? i'm not interested in asking her about irl politics but something else

[cinder pumice]

imo politics should be avoided as much as possible because it might spark controversy.

[cinder pumice]

I imagine the default voice will be the new natural and the redeem will be the AI/Robotic voice.

[tall pier]

When preparing for collabs, how much information do you give the language model to draw from? For example, do you just provide a name and maybe a physical trait of the Vtuber's model, or do you copy and paste their about section or similar sources to give Neuro some talking points? The collabs have been impressive, I wouldn't have expected Neuro to be able to keep up with how fast some of them speak.

[verbal coral]

i know im not interested in asking about real politics whatsoever

[foggy creek]

bringing something as dreary and depressing as real life politics into a neuro-sama stream makes absolutely no sense (in my opinion)

[bold light]

I think it is more about some fictional politics, like discussion on factions in some games or films

[verbal coral]

yeah

but it might prompt people to ask about real stuff so im curious if its overall ignored by neuro

[bold light]

then it is as muddy as some controversial topic, some may only think "it is just a game bro", some may think the statement could reflect real world politics

[bold light]

yeah, cant rely on the chat to not divert the topic to some questionable territories

[night elbow]

how long neuro's text responses usually are?

[verbal narwhal]

Is Vedal actually 12 years old?

[high dagger]

yes

[verbal narwhal]

damn

[wild kernel]

bonus straam today vedal?

[thorn cliff]

Monday, apparently

[digital cradle]

with her new voice, will that also affect her singing? I do hope the chipmunkness is fixed

[low ermine]

Is the code open?

[visual orchid]

check pins, it isn't #questions message

[low ermine]

I see. My other question is how did vedal manage to get any training data at all for Neuro. Did he ask permission from a few streamers to get their audio transcriptions or was it only from the stream chat

[hearty zenith]

even if he didn't ask, how can it ever be checked?

[low ermine]

You cat get patterns of speech that would help identify the sources. Other problem with neuro is that sometimes she straight up quotes parts of the training set with a few changes

[inner sparrow]

I don't think it's gonna be a problem since its probably from public vods and u can't really steal someone's words or or demeanor

U could stole someone's identity but that's clearly not happening since she doesn't resemble any streamer that much nor is there an effort to portray her as somebody else

This isn't an art situation where the artist have copyright on theire art (although I'm not even 100% on board with the ai art is theft argument)

[fiery shore]

Vedal himself

[fresh anchor]

Is Vedal planning on doing anymore programming streams?

[fringe zenith]

Yes, on monday

[fresh anchor]

Great!

[green sparrow]

Do you read AI papers?

[eternal sonnet]

sometimes

[low ermine]

I didn't understand your answer

[queen sand]

@low ermine has leveled up! (1 ➜ 2)

[stuck helm]

probably means that Vedal himself trained her

[low ermine]

wtf you kidding? Like from zero?

That's a lot of conversations with yourself

[fiery shore]

uhh not sure

[cinder pumice]

the audio is probably first trained by vedal then tested on discord call

[low ermine]

I meant the training data for the language model

I've been meaning to try something similar for fun but scraping VOD data from Twitch doesn't sound very ethical

[cinder pumice]

hm... not really sure, my speculation is that Vedal trained the language model himself.

maybe wiki data too but only to some extent.

[slim oyster]

i remember we she was banned it spent alot of time in the discord channel talking to viewers or the mods

[wanton depot]

Can someone remove slowmode

[wanton depot]

Who made the Vedal-turtle model?

Also add links
and phone number

[primal oxide]

For commission purposes or further enquiry on whoever did that model

[fiery shore]

boss i require the gee audio file that she sang earlier

[ember brook]

I wonder if a ban appeal stream is possible without getting you banned...?

[full vale]

I have this question too

I also wonder if we can see Neuro's new model before June...?

[foggy creek]

I believe he said he was hoping for something by the end of the month, may be wrong but who knows.

[eternal sonnet]

@hollow stump

she did

[lavish pollen]

Some big agency/company tried to buy neuro from you at some point?

[eternal sonnet]

nope

[broken sedge]

What was your first programming language? My was GML

[wanton depot]

We have auto delete messages when banned right?