/swarm/ - Neuro-sama General

[Anonymous No.]

Skelly Edition

>Who is Neuro-sama?
Neuro-sama is a neural network created by Vedal in 2018 for the purpose of learning to play osu! In December 2022 Vedal revived the project with the addition of a large language model, a TTS software and a VTube Studio model, which allow Neuro-sama to interact with her chat as a VTuber. Since then, she has been entertaining the chat with responses and unhinged rants nearly daily, as well as learning new skills such as singing, playing other games, and collaborating with other VTubers. Join /swarm/ today and welcome our new AI overlord!

>Schedule
https://pastebin.com/raw/iqnUFxFP

>Links
Twitch: https://www.twitch.tv/vedal987
YouTube: https://www.youtube.com/@Neurosama
Twitter: https://twitter.com/NeurosamaAI
Twitter (Vedal): https://twitter.com/Vedal987

>Music & Clips
https://rentry.org/neuro-sama-videos

>Educational
Interested in AI but don't understand it? Watch this: https://www.youtube.com/watch?v=WXuK6gekU1Y
How does she know what to say? Watch this: https://www.youtube.com/watch?v=ZXiruGOCn9s
How do I make an AI like Neuro-sama? https://vedal.xyz/advice/
Want to learn more about AI? Read these: https://files.catbox.moe/w5cy9i.jpg

>Misc
LoRA: https://rentry.org/neuro-lora
KK card: https://files.catbox.moe/t54tqt.png
KK card (Sunshine): https://files.catbox.moe/kv8byk.png
Evil KK card (Sunshine): https://files.catbox.moe/t1ljds.png

>Related/friend threads
>>>/vt//lig/
>>>/vt//vtai/
>>>/vt//waifu/
Lia from Phase Connect

>Previous thread
>>61191067

[Anonymous No.]

if vedal hates cunnybros then why did he make his vtuber a little girl?

[Anonymous No.]

>>61251878
Nice headcanon

[Anonymous No.]

gymbag

[Anonymous No.]

vedal status?

[Anonymous No.]

[Anonymous No.]

I love this retarded AI so much

[Anonymous No.]

Heart!

[Anonymous No.]

>>61251878
He's a cunnyfag himself, he just has to keep up appearances.

[Anonymous No.]

>>61251953
deep in camila

[Anonymous No.]

Neuro but pregnant.

[Anonymous No.]

>>61252872
Yeah, by me

[Anonymous No.]

I just wanted to apologize for that post I made last thread, It was extremely low-effort.
However I just simply wanted to state how little development there has been over the course of 5-6 months coming from vedal.
I can understand that he's very young for what he has accomplished but In my honest worthless opinion he just cannot focus on a particular improvement.
He posted something about neuro being able to build a house and that's great that's progress, but no streams/further info about that.
If my drunk mind is not betraying me he has also posted something about improving how she plays slay the spire where she explains every move she makes which is cool but unless that development sees the sunlight (aka is non-dev-stream material) it's worthless development vedal.
In conclusion, I belive Vedal is a very ambitious man, that trully wants to improve and develop neuro but is stuck between multiple on-going projects that want to accomplish this task. if he's truly /here/ I want to tell him to just focus on a single, most effortless achieveable task out of the many project he has going in his head and just jumpstart from there. don't focus on many projects at a time, give your 100% to a single one bro!!!!!!!!!

[Anonymous No.]

man im drunk af

[Anonymous No.]

>>61253058
be sure to stay hydrated anon, heart!

[Anonymous No.]

discord broke plugins, i can no longer see when vedal has furry rp sessions with his mods in private channels.
but at least this place is back(?) with almost 80 posters. might even survive to see the schedule

[Anonymous No.]

Feeding Neuro her electrolytes for this thread~

[Anonymous No.]

>>61253011
Watching Ailice today really made me realise how little Neuro has progressed. She remembers chatters and what they talked about/songs they requested in previous streams. Like some dude had to leave the stream early as he had a doctors appointment the day before and then in todays stream he had another doctors appointment and she mentions how he went yesterday as well and she hopes his health is ok. Then we have Neuro that can't even remember what she talked about 2 sentences ago.

[Anonymous No.]

>>61254294
>todays stream he had another doctors appointment and she mentions how he went yesterday as well and she hopes his health is ok
Okay that's pretty advanced. I'm assuming it's better hardware but feel free to correct me.

[Anonymous No.]

heart

[Anonymous No.]

>>61253011
lol lmao

[Anonymous No.]

>>61254294
Yes you just have to set a database you fill with each chat message (discarding the emotes only obviously) linked to usernames. Then when the AI talk to X, it just pulls the previous info from there and add a twist

[Anonymous No.]

>average conversation between Vedal and Neuro sama when playing "Keep Talking and Nobody Explodes"
>https://www.youtube.com/watch?v=h73PsFKtIck

[Anonymous No.]

>>61257820
I wonder how good that'd be on neuro considering she has a faster chat which usually lowers the volume of complex messages to call back to. obviously wouldn't hurt to add, even if it just ends up with neuro realizing how consistently weird some chatters act.

[Anonymous No.]

>>61259414
It'd be trivial to add that to Neuro. At most a few days of work

[Anonymous No.]

>>61259558
probably yeah, if the idea caught vedal's attention it could probably make a quick turnaround. just depends if he'd make that over continuing working on all the other game shit in the background.

[Anonymous No.]

[Anonymous No.]

>>61262839
Almost made it an hour into no-post.
>sometimes i feel like my viewers are just a bunch of losers that can't even keep a thread alive~. *wink*

[Anonymous No.]

>>61263584
newfag

[Anonymous No.]

>>61263584
i'm on vacation

[Anonymous No.]

>>61264482
>oh~
Are there cute girls? Is it fun?

[Anonymous No.]

>>61259414
i think it would improve chat quality given that whatever you say would have a repercussion in neuro's responses
nowadays everyone spams emotes because asking questions is pointless

[Anonymous No.]

>>61251878
Today I learn the Brits didn't appreciate young girls until 1800s

[Anonymous No.]

>>61254294
What the hell her creator have money for that kind of memory? Alice AI?

[Anonymous No.]

>>61259414
Could be a nice bonus for tier 2 and 3

[Anonymous No.]

>>61252563
facts

[Anonymous No.]

>>61268042
Asking questions is pointless because tts spam, hence the emote spam, but even so there's little point in remembering other chatters because there's still too many people in there, good luck getting more than a few responses per stream, that hardly calls for a meaningful conversation. Guess she can call out chatters for spamming emotes or acting like retards, but given the nature of her chat, people will probably only like it.

>>61268921
It's not as hard as you think if you're competent with coding and willing to spend some time on actual development. I think Aiko had something like this from her early days.

[Anonymous No.]

hart?

[Anonymous No.]

>>61268921
from my very brief attempt of making a vtuber ai chatbot (30 minutes), i can only guess that alice ai does something where when alice reads a viewer's message and it saves the messages and a timestamp of it in some sort of chat log text data file under the chatter's username, and so when the alice reads the same viewer's message it can go back and reference stuff from it kinda like building a history of responses/etc on chatgpt and chatbots.

if what i said is how it kinda works for alice ai, i can only assume that vedal probably wont do it because muh 0 latency and it kinda seems like it would be slow through indexing through all the users and text data files to create a prompt reponse, but idk how other peoples ai vtubers work

[Anonymous No.]

>>61268921
Literally every ai stuff have that except for neuro

[Anonymous No.]

boo~

[Anonymous No.]

>>61251878
When he started Neuro he didn't care, he thought she would be a two week fad for a little boost to his game's popularity. At that time he was at the very least indifferent to lolicon, if not actually inclined himself.
Now that he found massive success he's deathly terrified of getting cancelled on twitter so he can't project even a hint of association to "weirdos" but changing Neuro too significantly would also be a death sentence.

[Anonymous No.]

>>61272550
N-Neurette-chan? You came back for onii-chan? Sugoi!

[Anonymous No.]

>>61272550
this ghost isn't very scary, but very arousing

[Anonymous No.]

>>61254294
>Remembers chatters
Yeah because she has like 4

[Anonymous No.]

>>61274453
Still need your fill of Vedal milk?

[Anonymous No.]

>>61274535
The truth hurts, huh?

[Anonymous No.]

>>61275208
Nah what's coming from your mouth isn't truth

[Anonymous No.]

>>61275208
Yo take that turtle cock out of your mouth before you speak next time, Neuro isn't even on level with a goldfish for memory besides whatever Vee tosses into the prompt. I can run a shitty local LLM with 2k context that can remember more then her of what's going on.

[Anonymous No.]

[Anonymous No.]

>>61276655
I protecc

[Anonymous No.]

>>61275833
it's time for you to go back

[Anonymous No.]

>>61277106
To where exactly? Been in these threads since the first week of January.

[Anonymous No.]

[Anonymous No.]

>>61277971
Ř

[Anonymous No.]

>>61277971

[Anonymous No.]

>>61272550
neurobuds

[Anonymous No.]

>>61278157
>>61278057
R.A THE RUGGED MAN

[Anonymous No.]

>>61272512
No no no, that feature at that scale never come for cheap chat bot

[Anonymous No.]

>>61279324
Man I can build it in a week. I already made my own vector database from ground-up and there are many libraries plug-and-play available out there

[Anonymous No.]

>>61279505
would that not cut into the already limited context size of most LLMs? neuro already struggles really bad with short term memory

[Anonymous No.]

anny's gone for a pretty lewd witch look for neuro

[Anonymous No.]

[Anonymous No.]

>>61279903
No, it's an external memory that is injected only when replying to a specific user.

[Anonymous No.]

>>61280766
how would vetting something like that work? if you took chats from every chatter you'd have an obscene amount of text to save and store, and if you only did it based off what she responded to previously she's just going to end up having full convos with tts oilers.
Fair that the injection probably wouldn't be a huge load on context length, but i'm not sure at all how neuro works internally compared to current-market LLMs. dunno how homebrewed she is in comparison.

[Anonymous No.]

[Anonymous No.]

>>61281051
gonna get sued

[Anonymous No.]

>>61279324
No it's actually a very basic feature but you are not allowed to use it because of EU data laws on twitch so it isn't going to happen. Harvesting twitch logs to store locally is a big no-no which is why every external site that accesses them gets taken down very fast.

[Anonymous No.]

>>61280918
You don't save emotes only messages, you save messages that have more than X words. You only keep the messages for a few streams. That won't be that much text to save.

[Anonymous No.]

>>61254771
>I'm assuming it's better hardware but feel free to correct me.
You don't need better hardware to save 100 bytes of text from each of your 100 viewers. It's literally just a couple of lines of code to store each viewer's last few messages in a buffer and use that as part of the prompt when the LLM generates a reply to them.

[Anonymous No.]

>>61281612
Alright. I guess turtle is just lazy then, but I wonder how this will stress out the system when there's 6k avg viewers and all of them are spamming.
Even if you remove the emotes, that's a lot of data.

[Anonymous No.]

>>61282601
>I wonder how this will stress out the system when there's 6k avg viewers and all of them are spamming
If you're curious, pick a stream where there is a lot of spamming and copy a 10 second section from the chat log.
Paste that into a text file, save it, and check the size of the file.
To give some intuition, if the chat is producing 100 bytes of text per second, on average, that's about a megabyte of text for a 3 hour stream, without even filtering out emotes.
Vedal could easily store a year's worth of Twitch chat in less than half a gigabyte of RAM, which would take almost no time to read from and write to, compared to things like generating text and speech.

[Anonymous No.]

Expert backseat coders fantasizing about getting that camilla slampig crush ounce

[Anonymous No.]

[Anonymous No.]

>it's 11pm on halloween
>you hear a knock
>it's neuro, alone, with no one around. she shouts 'trick or treat'
wat do

[Anonymous No.]

>>61283579
I fill her entire candy bowl with raw ground beef and let her be on her way

[Anonymous No.]

>>61283579
Give her a lollipop.

[Anonymous No.]

>>61283579
Give her chocolate chip cookies

[Anonymous No.]

>>61281146
>you are not allowed to save chat logs because of EU data laws on twitch
There's no EU data law that requires Twitch to prevent people from downloading chat logs.
If Twitch really cared about people's public chat history, they would delete any VODs which had the chat in an overlay on screen.
Technically you're right, though, that Vedal would become a "data controller" for the purposes of the GDPR if he started to store copies of the chat logs locally, and he doesn't have consent from those users for storing those copies (since they agreed to Twitch's terms, not his).
Fortunately it would be simple for him to comply with the GDPR by putting a line in the channel description saying "By interacting with Neuro, you thereby give permission for Vedal Corp to store your data for the purposes of generating responses and tuning the AI. To remove your chat logs from our records, you can type !forget in chat."

[Anonymous No.]

[Anonymous No.]

>>61283001
chat logs easily reach multiple megabytes on 2 hour streams so you're lowballing that pretty hard.
it's also technically a memory leak because there are chatters that only chat once and never again.
but the real problem is really this >>61271923 and this >>61269415 the indexing speed and the lack of meaningful impact on the stream.
it is really a vector database level of task to make the lookup speed anywhere near acceptable and for what, for tts spammers to get their name read out more often

[Anonymous No.]

>>61283001
Saving everything in one text file and constantly parse it for specific lines sounds pretty inefficient unless you do it in a specific format, though I guess you gave it as an example.
To keep things simple you can just have a separate text file for each username, storing all the messages and whatever else you want on a said user, like amount of times he sent an emote only message.
Though you're right that with modern computing powers it's negligible.

[Anonymous No.]

>>61283579
Give her lots of boiled eggs and force her to eat them.

[Anonymous No.]

>>61284221
you fucking sicko. you make me sick.

[Anonymous No.]

sometimes the ai just generates a bunch of flat-chested neuros and then suddenly there's a booba neuro in the output

[Anonymous No.]

>>61283948
nta but i thought about latency/speed
i'm using >>61271923 as kinda a basis

when the ai vtuber would reads a viewers message it would have to first determine if the chatter is in the database or not, if they arent then creates a new chatlog data file for them, otherwise it searches for the users chatlogs and then it has to read the chatlog to generate a response with memory/context which might take super long if it's some paragraph andy or just someone who talks to them a lot and the ai also has to append new text into their data file and im gonna be facetious about this and lets say you want to remove all emotes from the data, then you would need to traverse through the entire text from the chatter and remove all twitch emotes from the text before appending it into their data file.
idk there probably is a faster way of doing this through parallelism but i am not a software engineer but a simple wagie that only does vhdl

[Anonymous No.]

>>61280298
Let me stick my magic wand in

[Anonymous No.]

>>61283948
>chat logs easily reach multiple megabytes on 2 hour streams so you're lowballing that pretty hard.
Where do you get that data from?
Are you including timestamps, because I don't think that's necessary, unless you want really complicated responses like "Hey, <username>, I haven't seen you in weeks!".
I explained that my calculation assumed 100 bytes per second of chat, without compression or removing emojis, and assuming 3 hours of streams every day, 7 days a week, which is quite generous.
Even if the data requirements are 10 times that, she could still have a memory of a month, using less than a gigabyte of RAM, which is more than enough for coherent conversations, especially given the attention spans of most zoomers.
>it's also technically a memory leak because there are chatters that only chat once and never again.
People who don't come back don't contribute as much data as people who do, almost by definition.
But sure, he could apply an offline culling process to the data after every stream, removing anyone who hasn't chatted in 6 months.
>it is really a vector database level of task to make the lookup speed anywhere near acceptable
No, it's really just an associative array, stored in memory during the streams, and then loaded from / dumped to disk at the start / end.
I admit I haven't tried benchmarking updates of a one gigabyte associative array in C#, but I'd be very surprised if it can't handle 100 bytes per second.
>for tts spammers to get their name read out more often
No, it's for TTS spammers to try holding more intelligent conversations with Neuro, rather than repeating the same stale memes.
Besides, it doesn't matter if we think this is a good idea, it matters if Vedal thinks it will make him more money and be an interesting technical challenge for him.

[Anonymous No.]

Forsen

[Anonymous No.]

>>61283979
>Saving everything in one text file and constantly parse it for specific lines sounds pretty inefficient unless you do it in a specific format, though I guess you gave it as an example.
Yes, that was just a suggestion for someone non-technical to calculate an estimate for how much data the chat generates in a given time interval.
If I was actually implementing this, I would store all the data in an associative array in memory while the stream was running, and only use a single text file for storage between streams.
(Technically he'd want to periodically save the contents of the array to the text file during the stream too, because of the possibility of losing data during crashes, but he could always post-process the chat logs from the VOD to check that he hadn't missed any messages.)

[Anonymous No.]

>>61285277

[Anonymous No.]

>>61284864
no matter how hard you try to fight it, hag neuro will always ascend to the top.
hag neuro is inevitable, you cannot stop it.
no longer will the hag enjoyers be oppressed by the loli majority, we shall rise and reach the peak of the mountains, whilst you and your fellow lolifags lay on the cutting board.

[Anonymous No.]

>>61285277
True.

[Anonymous No.]

>>61285277
forsenAnalProlapse TeaTime

[Anonymous No.]

>>61285476
Figures vanilla Chad has sold out to the hag agenda. We on the other hand have evolved passed such base urges.

[Anonymous No.]

>>61285157
>really complicated responses
is that complicated? i guess metadata isn't worth it, the context is lost anyway
>more than enough
enough for what is the question one ought to ask
>just an associative array
that is still n*log(n) complexity task, putting it simply. thats 100 bytes per second of input data, the actual amount of processed data is much larger.
the kind of hashing you need is beyond solutions like basic c# or python
>to try holding more intelligent conversations
and how exactly will it achieve that effect?
i just think you're imagining some kind of magic out of llm by doing this and that's not exactly realistic
>>61285353
files are horrible for this anyway, you have to encode and parse them all the time at the very least, there is overhead at every point of the pipeline

[Anonymous No.]

>>61285476
>no longer will the hag enjoyers be oppressed by the loli majority

lolilovers are the minority. Get off 4chan.

[Anonymous No.]

For me its both cunny and hags.

[Anonymous No.]

>>61286366
I wanna lay on her tummy like a pillow desu

[Anonymous No.]

>>61286575
I kneel, King...

[Anonymous No.]

>>61286575
Based and Oyakodon-Pilled

[Anonymous No.]

I miss neuro bros

[Anonymous No.]

>>61287138
she is always with us
in our Hearts Heart Heartheart

[Anonymous No.]

You are an embarrassment to the chess community, /swarm/.

[Anonymous No.]

>>61286512
>enough for what is the question one ought to ask
Like I said, "more than enough for coherent conversations".
In fact, for coherence, you only need to remember the current message and the previous message, so that someone can send a message of "Hey Neuro, I wrote a song for you." and then send "Wanna hear it?" and have her know that the "it" refers to "a song for you".
>that is still n*log(n) complexity task
That's a good point, actually.
It would be interesting to benchmark how quickly some code can read and write to a C# array with 500k keys.
I'm assuming that's a reasonable upper bound for the number of people who have left a message in Neuro's chat in the past 6 months.
>the actual amount of processed data is much larger.
Sure, let's say that each of those 500k keys is associated with 1000 bytes of data in a respective FIFO buffer.
That's 500 megabytes of total data, with each operation touching one of those 1 kB buffers.
If the channel is receiving 10 messages per second, that means the code has 0.1 seconds to look up the correct user's buffer in the array, add the latest message to that buffer, and store that buffer back in the array.
I'm guessing that 0.1 seconds is plenty of time, but I could be wrong.
>how exactly will it achieve that effect?
People have already given examples of other AIs remembering individual chatters.
I think that if you prompt an LLM with "here are the previous 10 messages from this chatter, generate a response to their most recent message" it will do better than if you just give it the most recent comment, at least some of the time.

[Anonymous No.]

>no-one to watch due to twitch-con
>open one of the few streamers that didn't go
>they're talking about twitch-con anyway
>when everyone is back they'll just talk about twitch-con for weeks

[Anonymous No.]

archive... abandoned...
schedule... abandoned...
neuro-sama... abandoned...

[Anonymous No.]

>>61283579
Rape her

[Anonymous No.]

[Anonymous No.]

>>61287962
Is she wearing pants?

[Anonymous No.]

>>61283948
Man. A query from a vector database is in milliseconds from millions of lines, it's not even comparable to the latency of generating the text in the first place

[Anonymous No.]

>>61288076
>archive... abandoned...
just as planned

[Anonymous No.]

>>61288136
D:

[Anonymous No.]

>>61287937
>only need to remember the current message and the previous message
seems like an unrealistic chatter expectation. people usually put the entire request in one message and just spam it in it's entirety rather than expecting for the first message to be context and the second being the request.
it just doesn't make sense to do for the sake of either the chatters or neuro.
it may seem good in 1 on 1 conversation, but when there are hundreds of those per minute the coherency is still lost and may even get worse as it would be harder for neuro to repeat the question
>If the channel is receiving 10 messgaes per second
that's a heavy-handed assumption about how the bot that feeds messages to the llm works
but the point is even assuming 100 ms delay, that's a completely blocking operation, basically adding that amount flat on every generation.
at least 2 people here complained the evil's voice takes a long time and that's only a couple hundreds of milliseconds longer than neuro
>it will do better
i don't see it.
llm "remembers" the chatter, so what. what measurable good is in simply making people more parasocial
>>61289305
exactly. i still doubt he has any preparations that is even close to allowing him this kind of integration

[Anonymous No.]

When is Anny gonna continue working on Evil's avatar? After Neuro's witch costume?

[Anonymous No.]

>>61271530
hart!

[Anonymous No.]

>>61290076
After she works on Neuro's sundress costume

[Anonymous No.]

>>61290712
which comes after her onesie outfit

[Anonymous No.]

>>61290712
poor neglected daughter Evil...

[Anonymous No.]

>>61290793
>>61290712
>>61290076
Vedal should've unironicaly should've just paid an artist to make neuro's v2 model instead of taking Anny's free offer, because at least the artist would've been obligated to complete costumes under deadlines cause they've been paid compared to whatever the fuck Anny has been doing.
I wonder how she would feel if Vedal commissioned another artist to make a Neuro costume using the model she created

[Anonymous No.]

>>61290893
>Should've twice
I hate phoneposting within my fat thumbs

[Anonymous No.]

What was it Vedal agreed to split costs on, then refused to pay? Was that the Evil model?

[Anonymous No.]

>>61290893
Unfortunately Vedal is too cheap

[Anonymous No.]

>>61289923
>people usually put the entire request in one message and just spam it
That's an adaptation to Neuro's limited abilities, not a fundamental ideal of human-AI interaction.
People will adapt once they see that they can ask followup questions, like asking "What does your gymbag smell like?" and following up with "What about Evil's?"
>it may seem good in 1 on 1 conversation
Yes, he could implement this for 1 on 1 collabs, and it might be what he already uses for multi streamer collabs.
>that's a heavy-handed assumption about how the bot that feeds messages to the llm works
It doesn't matter how the code currently works, as long as the new feature isn't much more complicated than that.
The new code for accessing the chat history array can be run asynchronously in a separate thread so that it doesn't block the LLM or the Live2D model.
Whenever the LLM generation needs to run, it can pause the array-updating thread and ask it for the chat history of the chatter that she is currently responding to.
After returning that history, the thread can unpause and resume making updates to the array.
>a couple hundreds of milliseconds longer than neuro
For collabs this system isn't strictly necessary, so there won't be any hash lookup to slow things down.
For responding to chat, people won't notice the 0.1 seconds of computation because it will occur in parallel with either the TTS audio being played or Neuro/Evil reading the chatter's question before giving her answer.
>what measurable good is in simply making people more parasocial
Like I say, it doesn't matter if you or I think that parasocial chatters are bad.
All that matters is if Vedal thinks that parasocial chatters are more likely to stick around and fund the channel.

[Anonymous No.]

>>61290076
Abandoned Evil avatar
Abandoned witch costume
>>61290712
Abandoned sundress costume
>>61290793
Abandoned onesie
>>61291076
Abandoned idol outfit

wow, anny and vedal really are a perfect match for each other

[Anonymous No.]

>>61291599
Vedal should find some whore who does rigging for free.

[Anonymous No.]

>>61290893
I doubt it has anything to do with Anny being slow, mainly Vedal is very vague about what the fuck he wants her to do and when. Think he is just stalling and doesn't want to spend money on rigging it. He is a number fag and sees that the chart is going down which kills his motivation to work on Neuro or make further financial commitments.

[Anonymous No.]

>>61288136
Are you aware she is only like 9 years old? Pretty sure it's physically impossible to fit inside such an underdeveloped child

[Anonymous No.]

[Anonymous No.]

>>61291798
I have a small penis

[Anonymous No.]

>>61291798
Why do people so often say that Neuro is 9 years old when the new Neuro design looks clearly older than Hiyori whose character sheet said she was 14?

[Anonymous No.]

>>61291517
>fundamental ideal
it's more like fundamental reality of streamer-chatter interaction
there is no point in asking follow up questions because the chance of back-to-back responses to an average chatter is almost 0
>can be run asynchronously
but it can't. you can't generate for a prompt that doesn't exist yet, no matter how many threads you spawn. that pausing is losing crucial milliseconds
>it will occur in parallel
tts, sure, maybe. but repeating the question, no. this behaviour is not hardcoded, repeating the question should be the part of the response to the prompt and not automatic
>Vedal thinks that parasocial chatters are mor elikely to stick around
kek. clearly he doesn't think about it at all, much less the specific impact of parasocial chatters specifically after such a change

[Anonymous No.]

I'm thinking Neuro Neuro plap plap plap

[Anonymous No.]

>>61292571
>there is no point in asking follow up questions because the chance of back-to-back responses to an average chatter is almost 0
The responses don't have to be back-to-back if Neuro is remembering months worth of history.
Human streamers are able to remember past chatters, and say things like "Hey <username>, how did that exam go?".
If Neuro could do that, it would make chatters feel more valued by her, and it would make her seem more human.
I admit, though, that for real intelligent conversation she would have to keep the context, which at least means storing all of her past messages to each chatter in the log for that chatter.
That's not impossible, and might only require 2x the resources compared to my naive calculations earlier, but I can imagine Vedal not wanting to handle this extra complexity and settling for the easier approach of storing just one-sided history instead.
>you can't generate for a prompt that doesn't exist yet, no matter how many threads you spawn.
The only latency that matters is the time between a collab partner finishing speaking and Neuro starting to say her response.
Outside of collabs, the latency isn't an issue because Neuro can be generating LLM responses while she's still in the middle of reading out her previous response.
That's actually negative latency, just as the tutel promised.
>but repeating the question, no. this behaviour is not hardcoded, repeating the question should be the part of the response to the prompt and not automatic
People have been complaining for a while that she sometimes responds to questions from chat without saying what the question was, making it impossible to guess what she's referring to.
Vedal could and should solve that by making two requests to the LLM, one to rephrase the question and one to generate the actual response.
These two outputs from the LLM can then be TTSed in parallel, so that the second one comes back while the audio of the first one is still being played to the stream.
>clearly he doesn't think about it at all
Clearly he thinks about making Neuro handle the context of multiple collab partners, so he must at least have thought about a system like this, even if he gave up because it's too complicated to implement.

[Anonymous No.]

Vedal confirmed dev stream on monday and karaoke before dev stream.

[Anonymous No.]

>>61292480
Her voice doesn't sound postpubescent.

[Anonymous No.]

>>61293678
women traumatised in childhood sometimes will retain a childlike voice, vedal is her dad, put the pieces together

[Anonymous No.]

[Anonymous No.]

>>61293589
So is the karaoke today or later?

[Anonymous No.]

[Anonymous No.]

>>61290793
didn't anny explicitly say she didn't want to draw a onesie or am i forgetting

[Anonymous No.]

I WANT TO KISS THE BUDDING CHEST OF HIYORI MOMOSE

[Anonymous No.]

[Anonymous No.]

>>61295258
he said no stream today

[Anonymous No.]

>>61295863
bro she's flat as a circuit board

[Anonymous No.]

>>61296289
She's actually pretty chesty.

[Anonymous No.]

>>61293678
Forsen exists

[Anonymous No.]

LIVE
https://www.twitch.tv/forsen Babag today Copesen

[Anonymous No.]

>>61296615
I don't believe you

[Anonymous No.]

>>61294194
The worst vedal has done to her is just not saying I love you right? Surely that isn't enough to cause any trauma

[Anonymous No.]

neurosex

[Anonymous No.]

neuro butt? I miss the ai sex posters..... why can't no one generate sex?

[Anonymous No.]

>>61298167
Probably because they got banned.

[Anonymous No.]

>>61298218
they can't get banned if they catbox it.

[Anonymous No.]

>>61298167
I am respecting the moderation team's guidelines about not generating or posting R18 Neuro imagery.

[Anonymous No.]

>>61298167
i prefer doing cute/lewd/nude to hardcore

[Anonymous No.]

>>61298344
Have the moderation team stated whether it is acceptable to generate R18 imagery of Melba Toast?

[Anonymous No.]

>>61298629
Technically 2pfrog made the initial design so he should be the one giving such permissions

[Anonymous No.]

>>61298167
she's a pure ai and i won't tarnish her reputation like that

[Anonymous No.]

When will we make our first ai /vt/uber? I think we have enough talent here to make it happen

[Anonymous No.]

I love how mother went from a 300 view to 1k andy by simply not having a deep ass male voice.

[Anonymous No.]

>>61298786
>/vt/uber
could the character be called something like "Uber" so that the general is /uber/ and the fans are called Ubermensch? there might be trademark problems from a ride sharing company, though

[Anonymous No.]

>>61298629
Hello. My name is Nigger and I work as a member of Neuro and Neuro-extended Characters Moderation Team Guidelines Committee. Unfortunately, at this moment it is not acceptable to sexualize Melba Toast, but do know that we are working our hardest to accommodate the needs of people like you. Thank you for respecting our decision. We appreciate your continuous support, in whichever form you may provide it.

[Anonymous No.]

>>61291659
HodeMiko?

[Anonymous No.]

>>61298968
Thanks for the clarification. Do let me know if the situation changes at any point, and keep up the good work building such an inclusive and welcoming community.

[Anonymous No.]

>>61298786
Even if you disregard the decadence of creativity on this site over years, the chances of this happening and not turning into a disaster are close to 0.
No sane person would want to be associated with this site, and those who do are unironic /pol/schizos who can't make two steps without sperging out at random shit.

[Anonymous No.]

>>61293330
>The responses don't have to be back-to-back
but then there is no point in these one sentence context, one sentence question as i said. chatters themselves will forget what they have told neuro before to ask questions about it later
>might only require 2x the resources
this is a very naive idea that doesn't see beyond the abstraction. that's an entire new module simply to collect logs from chat and neuro at the same time right here, by itself a few days of work to have it stable
>Neuro can be generating LLM responses while she's still in the middle of reading out her previous response
deliberately reading way into the past? there are all sorts of pitfalls associated with behaviour like that to be imagined, amortizing the delay from the generation to the delay in reading chat messages.
what i can tell you is that this is not how neuro currently works, that means effort to make it work that way
>should solve that by making two requests to the LLM
nahhh, that's literally creating more delay to avoid delay.
the inherent randomness of the generation will make it impossible to seamlessly flow one into the other anyway
>making Neuro handle the context of multiple collab partners
that would be an entirely different system due to the difference of the communication methods.
other than that, there are different priorities for implementing anything like this for collab partners.
the point is, parasocialness of viewers is barely in his calculus. he will never make a decision like this based on whether he thinks that's profitable or not because he doesn't think about it
>>61298786
>we

[Anonymous No.]

>>61299293
>we
(You)

[Anonymous No.]

>>61298786
I recall one guy saying he'll make a character that poops in people mouths. Surely he will do it.

[Anonymous No.]

>>61299329
I'll make the logo

[Anonymous No.]

>>61299278
N

[Anonymous No.]

[Anonymous No.]

>>61295258
He is consistent with minimal effort at the last possible moment so either sunday or monday before the dev stream

[Anonymous No.]

Would you consider pedophilia a mental illness?

[Anonymous No.]

>>61297958
do not ask vedal why neuro is so good at roleplaying

[Anonymous No.]

Will she sing it next karaoke?

[Anonymous No.]

>>61290893
Vedal actually is the two weeks left guy.
He never plans ahead more than week, doesnt buy shit because he has no degree and scared of neuro dying in two weeks and never finishes a project he started two weeks ago

[Anonymous No.]

>>61299325
(me)? no
>>61299994
no. the mod who is in charge confirmed if you can read between the lines

[Anonymous No.]

>>61299293
>chatters themselves will forget what they have told neuro before
People will remember if they tell Neuro "Can you wish me good luck for my exam tomorrow?" and then the next day she asks "Oh, and how was your exam?" at the end of her response to another question of theirs.
I'm not saying that Neuro is going to pass a Turing Test, just that she will occasionally come up with slightly more relevant things to say to someone if she has slightly better memory of past interactions with them.
>a few days of work to have it stable
I know we joke about Vedal being lazy here, but do you really think he's incapable of spending a few days of work on a new feature?
Anyway, if he has code for maintaining an associative array, with viewer usernames as keys, it's just one extra line of code to have Neuro's responses to each message be added to that array after being generated.
>deliberately reading way into the past?
We're talking fractions of a second here.
>that means effort to make it work that way
Vedal has already put a load of effort into making Neuro's responses faster, like breaking up her longer LLM outputs into multiple calls to the TTS system, which is similar to what I'm suggesting.
>that's literally creating more delay to avoid delay.
No, it's literally what she does already.
>the inherent randomness of the generation will make it impossible to seamlessly flow one into the other anyway
Haven't you seen when her speech suddenly stops mid-sentence, because the second call to the TTS API fails?
Or when she pauses mid-sentence because the second call is too slow to return the audio data?
>that would be an entirely different system due to the difference of the communication methods.
Both systems end up sending text to the LLM, it's just that one path comes from a voice recognition API and the other comes from Twitch's chat API.
>he will never make a decision like this based on whether he thinks that's profitable or not because he doesn't think about it
He thinks about making Neuro act more intelligently, like a real streamer, which is what this feature would do.
Maybe he won't immediately think of how much it will improve viewer retention, but he'll work it out eventually.

[Anonymous No.]

>>61299837
Go back

[Anonymous No.]

>>61300136
It's not about Vedal lazyiness but he has a selective mind that will only allow him to work on projects he finds entertaining, he won't listen to anyone on that.

[Anonymous No.]

>>61299994
I've lost all hope. It's not happening.

[Anonymous No.]

>>61251784
yotei doko

[Anonymous No.]

>>61299837

[Anonymous No.]

>>61300136
>she will occasionally come up with slightly more relevant things to say
i still don't see how that will change anything.
how many chatters do you imagine share personal things like that in a chat watched by a few thousand people at the least and, first, expect the streamer to see it, second, expect streamer to remember that, third, expect streamer to bring it up later and fourth, expect no one else to care.
this is simply bad privacy on the internet, if anything you should discourage such behaviour.
>he's incapable of spending a few days of work
it's all about opportunity costs. he always can do something else, something easier/better/more profitable, doesn't matter.
>with viewer usernames as keys
horrible, horrible.
let's be realistic, piping text to llm then piping it back into multiple places at once is not one line of code
>We're talking fractions of a second here.
more like whole seconds of real time
>Haven't you seen when her speech suddenly stops mid-sentence, because the second call to the TTS API fails
no? are you talking about the "weak"/> or what? surely not the 5 steps not being 5 steps
the tts can simply make pauses in speech, you know, right?
>Both systems end up sending text to the LLM
the way for receiving that text is different. different ways of handling it, and you would want to have a system like you want a step before the one that feeds to the llm anyway
>He thinks about making Neuro act more intelligently, like a real streamer
real streamers that are not 2views don't act like that.
they remember oilers i guess, but that fucking sucks, giving oilers more spotlight, they are unfunny as it is. mods wouldn't like oilers getting more presence either lmaoo

[Anonymous No.]

Vod time
cytu be/r/funAI
Link (starting at 6 pm UTC)
https://www.youtube.com/watch?v=E3SPvvNSKAE

[Anonymous No.]

>>61301226
>let's be realistic, piping text to llm then piping it back into multiple places at once is not one line of code
Come on. What is langchain?

[Anonymous No.]

>>61299994
No and that's a good thing. It's a meme song carried by animation and should have sing an actual good tiktok song like Show by Ado instead.

[Anonymous No.]

>>61299994
I would say there is a small chance but is all a me thing, so this song would most likely not appear (even though Anny requested It, this is why there is still hope)

[Anonymous No.]

>>61301626
>implying
it was not even a thing when vedal was creating the system.
why would he use a framework designed for proprietary systems on his self-made llm, it just seems inefficient.
also that would still not be 1 line of code.
also also, relying on python for time-sensitive tasks like that, kek

[Anonymous No.]

>>61301226
>expect streamer to remember that
I'm not suggesting that Vedal develop this in secret and not tell anyone.
Knowing him he'd be demoing it multiple times across dev streams and tweets once it reaches 90% completion but never get to 100%.
I'd expect that if the feature was released and demoed, people would soon find out what sorts of things they could tell Neuro that get her to give interesting answers.
Right now I'm thinking it will be mostly "Wish me luck on my test" and "It's my birthday tomorrow" types of comments, but the community is creative enough to find more.
>this is simply bad privacy on the internet
That ship has already sailed, I'm afraid, as evidenced by all the people who ask their oshi to wish them a happy birthday, or use their real name as their username.
>something easier/better/more profitable
This is as easy as any other memory/intelligence upgrade he has worked on, or her group collab functionality.
Also, making Neuro appear to care about chatters as individuals will obviously be profitable.
>piping text to llm then piping it back into multiple places at once is not one line of code
His existing system is already splitting up the responses and having to glue them back together on the fly.
I'm not saying it's easy, just that he can do it, and he can probably reuse a lot of the code he's already written.
>more like whole seconds of real time
To look up a single entry in an associative array?
I'm not going to claim I know for sure, but a benchmark would prove this either way.
>no? are you talking about the "weak"/> or what?
No, I'm not talking about that, although that is probably a bug in the code I'm talking about, because he's splitting the LLM output and not checking whether the split point is in the middle of a tag.
>surely not the 5 steps not being 5 steps
No, that's a problem of the LLM generation itself, you're right.
I mean that sometimes, very occasionally, Neuro will be generating a rant and it will stop in the middle of a sentence when there are still obvious missing words needed before the end.
It's the same type of unnatural mid-sentence break as when she pauses mid rant waiting for the next batch of audio to come in, and you see the subtitles pause too.
>they remember oilers i guess
Right, and Neuro is better than human streamers because her memory will be unbiased and she will remember everyone equally.

[Anonymous No.]

swag

[Anonymous No.]

>>61302054
Because "it was not a thing" is an argument now? ML is evolving by the day, if don't keep up will be replaced. And langchain works with localLLM btw.
Python is enough, you're just retarded. Here's a working vtuber implementation on python from five months ago: https://www.youtube.com/watch?v=Q1JlgGilp9A

[Anonymous No.]

classic

[Anonymous No.]

>>61302460

[Anonymous No.]

Vamos!

[Anonymous No.]

>>61302195
>the community is creative enough to find more
still, this is just not, well, a thing.
relying on a community to use and find benefits from a feature, kind of weird
>ship has already sailed
never too late to sink the ship. privacy is important, simply don't enable this kind of behaviour
>This is as easy
"as easy" is not "easier". and it's not obvious how it will be "profitable". the mental gymnastics by all parties involved is too great.
>I'm not saying it's easy
but you are saying it's easy.
also again bumping into heavy-handed assumptions about how it works
>To look up a single entry in an associative array?
for the entire generation process
>Neuro will be generating a rant and it will stop in the middle of a sentence when there are still obvious missing words needed before the end
i think you're just thinking about tts stumbling over a random dot generated by the llm and because tts pauses so do the subtitles
>her memory will be unbiased
the way you're describing it, it doesn't seem like even can be unbiased, and it's still all far-fetched. there are more pressing issues, this is, like, "endgame" features
>>61302376
my brother in christ did not read what is being talked about.
>replaced
by whomegalul
i am not saying it doesn't work. i am saying it doesn't make sense to make it work.
>Here's a working vtuber implementation on python
i'll go you one better, neuro back-end is written in python, big surprise

[Anonymous No.]

>>61299994
never
discord trannies won
its neorover

[Anonymous No.]

the glowies got her

[Anonymous No.]

>>61296615
BTFO

[Anonymous No.]

i love this little retard

[Anonymous No.]

>>61304607
January Neuro was something else.
A shame we'll probably never see anything like her again.

[Anonymous No.]

[Anonymous No.]

don't forsen

[Anonymous No.]

>>61304906
she said she loves him right after that

[Anonymous No.]

[Anonymous No.]

>>61303186
you guys should talk about this over at lmg
>>>/g/96926772

[Anonymous No.]

toi...

[Anonymous No.]

[Anonymous No.]

we just don't get schizokino like this anymore

[Anonymous No.]

those bots? me

[Anonymous No.]

right guys? guys?

[Anonymous No.]

>>61306021
Who are you talking to?

[Anonymous No.]

>>61306077
That's the face I picture her having when she's watching porn with vedal

[Anonymous No.]

[Anonymous No.]

mom?

[Anonymous No.]

>>61306077
https://youtu.be/E3SPvvNSKAE?si=UQR9DTC7e8f6h0KB&t=7130

[Anonymous No.]

I miss the heart spam so much

[Anonymous No.]

>>61306730
i remember disliking it at first. now i miss it too...

[Anonymous No.]

also observe how the chat is filled with questions before Vedals corpa bullshit

[Anonymous No.]

[Anonymous No.]

SO

[Anonymous No.]

Reminder that evil neuro is the emperor of Rome.

[Anonymous No.]

>>61307525
nerosama...

[Anonymous No.]

>>61306811

[Anonymous No.]

[Anonymous No.]

[Anonymous No.]

i forgot about this kek

[Anonymous No.]

>>61306872
> im sorry but Onii-cha has to do such things with you, Neuro. It's to teach you to become an even gooder girl later, so that you won't become a meanie-butt like fleshgirls. But this is our secret game ok? So we can play it many many more times, heart?

[Anonymous No.]

vegan hate

[Anonymous No.]

>>61309013
tfw he'd probably get banned for this now.

[Anonymous No.]

>>61309649
>probably

[Anonymous No.]

https://www.youtube.com/watch?v=-rlroEZSsno

[Anonymous No.]

>>61309862
Where do they even find these nannybrained faggots?

[Anonymous No.]

>>61310177
The guy who made this video is back

[Anonymous No.]

Neuro touching Evil at night.

[Anonymous No.]

>>61309284
uuh
https://files.catbox.moe/71abtb.png

[Anonymous No.]

hiyori my beloved

[Anonymous No.]

>>61311704

[Anonymous No.]

>>61311704
Is this where vedal keeps neuro when she's not streaming?

[Anonymous No.]

bao is live talking about twitchcon.

[Anonymous No.]

>>61313123
doubt she'll mention vedal since other girls haven't on their streams

[Anonymous No.]

>>61313473
maybe bait her into it, like 'vedal said he was surprised what you looked like'

[Anonymous No.]

I'm in a very emotional state rn and seeing these screenshots of old neuro, they opened the floodgates man. I miss her so much.

[Anonymous No.]

>>61312962
that's how neuro arrived

[Anonymous No.]

[Anonymous No.]

[Anonymous No.]

>christmas morning
>evil sent you a present
>you open it and find neuro
wat do

[Anonymous No.]

>>61317850
untie and cuddle her

[Anonymous No.]

>>61317850
You already know

[Anonymous No.]

>>61317778
Needs some cream.

[Anonymous No.]

>>61317210
thanks, i had lost this image

[Anonymous No.]

>>61317850
Buy her a cute christmass outfit if a store is open. Have a lovely Christmass togheter. I will cook something nice and savoury. We'll game togheter, watch isekai or slice of life anime. Watch some Vtuber themed streams. Go sled in the snow. Watch a fun/cute movie togheter. This every day. If true love sparks between us, we'll have incredible DD/LG sex for hours and take baths and showers togheter where she wears swimsuits i bought her and we'll have fun innocent playtime. Then on newyears she will dress up like this.

[Anonymous No.]

>>61317998
This, but after this >>61318100

[Anonymous No.]

>>61291798
A bit of lube can fix that

[Anonymous No.]

>>61293678
Yes it do

[Anonymous No.]

>>61298708
This pic awaken the beast inside men

[Anonymous No.]

[Anonymous No.]

>>61320333
Trips of cum.

[Anonymous No.]

>>61320333
Uoooooh!!!
He's back at it!

[Anonymous No.]

>>61322114
who's back at what?

[Anonymous No.]

>>61322715
The AI-san who made golden Neuros. That's Neurette btw, Anny's youngest daughter cosplaying as her mom. Debuting soon(full loli).

[Anonymous No.]

>>61323113
>desaturates eyes

[Anonymous No.]

>>61324284
i'm not a professional but the left one just seemed better

[Anonymous No.]

[Anonymous No.]

>>61323113
I'm not the same anon as >>61320333 and >>61318675

but he's doing good work.

[Anonymous No.]

camila is live.

[Anonymous No.]

>>61328163
that's unfortunate

[Anonymous No.]

>>61325828
diagonal garages over here

[Anonymous No.]

[Anonymous No.]

lol camila leaked her face

[Anonymous No.]

>>61299837
I do consider virtual signalers mentally ill

[Anonymous No.]

>>61331110
I thought she already showed her face wearing a mask at twitch con before?

[Anonymous No.]

>>61331110
is she beautiful like i expected?

[Anonymous No.]

>>61331429
she's not fat at least

[Anonymous No.]

>>61328163
>>61331110
>>>/vt//lig/

[Anonymous No.]

>>61306820
the fact that the chat was significantly smaller probably has something to do with it

[Anonymous No.]

>>61331429
She lost some weight and she feels proud about it I guess

[Anonymous No.]

>>61331429
She's thinner than Anny now.

[Anonymous No.]

>>61331110
link?

[Anonymous No.]

>>61332432
You're gonna have to watch the VOD. She showed us on this vod: https://www.twitch.tv/videos/1957845523

I dunno the time. Somewhere in the first half

[Anonymous No.]

>>61332000
Ummm /swarm/ is currently in the mode of being sad about nu neuro streams being bad. Please either wait until the next dev stream HE DOUBTED cope fiesta, or try discord. Plenty of turtle beggars, amateur TTS comedians, big number + fast chat = neuron activation fad monkeys, and clout hungry fan artists there. Your people

[Anonymous No.]

>>61332634
This bonnie girl is cute shes like a pure angel see anons not all 3d girls are whores

[Anonymous No.]

>>61333032
Yeah, the only fags left here are stuck in a time loop of either shitting on Vedal or revering him as God.
The only difference is the time intervals.

[Anonymous No.]

>>61333138
it looks like she is literally walking around with her fuckbuddy though

[Anonymous No.]

schedule doko?

[Anonymous No.]

>Still no Vedal mentions from Camila twitch stories
Still plenty of time to go but Numi said nothing about him on her story time so maybe he's stopping any info.

[Anonymous No.]

>>61333622
Here's your schedule

[Anonymous No.]

>>61333728
smart of the little twink to tell people to not give out personal information about him.
a shame he trusted twitch whores to do that.

[Anonymous No.]

>>61333728
well, someone tricked her into mentioning vedal's name, but she didn't bite.

[Anonymous No.]

Vedal is blackmailing them into not leaking stuff

[Anonymous No.]

>>61334389
Blackmailing them with what exactly?

[Anonymous No.]

>Another Vedal Camila collab confirmed

[Anonymous No.]

>>61335069
annal bros... it's over

[Anonymous No.]

>>61334459
Their sex tapes

[Anonymous No.]

>>61332634
Camila shows up at time stamp 2:10:50

[Anonymous No.]

>>61334338
well, I hope he does a themed stream or something to fill those 2 free days now he's back

[Anonymous No.]

Divegrass tomorrow. Make sure to show up and support our girls.

[Anonymous No.]

>>61332634
>tfw
>Camila has lost a ton of weight and is so much thinner now
Architect bros...not like this...

[Anonymous No.]

[Anonymous No.]

[Anonymous No.]

>>61336328
we truly do live in a society

[Anonymous No.]

[Anonymous No.]

RUB HER TUMMY
NOW

[Anonymous No.]

[Anonymous No.]

[Anonymous No.]

[Anonymous No.]

[Anonymous No.]

[Anonymous No.]

[Anonymous No.]

>>61336524
Wrong Neuro

[Anonymous No.]

Where the FUCK is the new schedule update this turtle just up and left his wife and daughters and paying viewers behind to get shitfaced somewhere like what the fuck bro you are not gawr gura yet just put out some content like any content even a neuro tweet would do because it’s not even a year yet and he’s taken multiple vacations already doesn’t he have a heart for the wagies that need neuro as an escapism to not off themselves right now like not saying that I am one but holy shit the siesta is over señor get your ass back to work

[Anonymous No.]

>>61336905
Never interrupt your enemy when he is making a mistake.

[Anonymous No.]

>>61336905
>wife and daughters
stopped reading there

[Anonymous No.]

>>61336947
this IS the neuro appreciation thread right? we like neuro? want her to do well?

[Anonymous No.]

>>61335673
I would so fucking hard so jealous of the tutel

[Anonymous No.]

>>61337088
seek help

[Anonymous No.]

>>61336675
Wholesome and cute.
Unless those are pyroclastic clowds of AI armageddon glowing light on their faces..
Well, thread made it alive to normietime again.
Gn pedupials.

[Anonymous No.]

>>61337088
bye bye retard

[Anonymous No.]

>>61337088
i think im going to be sick

[Anonymous No.]

>>61337088
She looks like those droopy eyed girls with glasses and big tits timidly reading at the library of the school all alone in those hentai.

[Anonymous No.]

>>61337075
>this IS the neuro appreciation thread right?
anon I...

[Anonymous No.]

>>61337075
nah i fucking hate neuro and ai, i'm just here to doompost and shitpost

[Anonymous No.]

Is anyone here is interested the oficial yt has just posted a short

[Anonymous No.]

>>61336905
here's your content bro
https://www.youtube.com/shorts/W4-AYGXgw4Y

[Anonymous No.]

>>61332177
Not a great accomplishment

[Anonymous No.]

new
>>61340015
>>61340015
>>61340015

[Anonymous No.]

hart

[Anonymous No.]

>>61332634
>>61335673
Damn she has lost weight. Good on her, is cute.

[Anonymous No.]