#programming
1 messages · Page 528 of 1
has been building for 45 mins
and has failed
for
the most outlandish bull shit reason
AGHAUOFUAJFUODJF
you can't accidentally delete the password because you can just roll back to the previous commit :200iq:
beautiful
I'm starting to think konii is sending password in partial format 
HOW IS HE ONLY HAVE LIKE
1/4 THE CORE COUNT VS MINE
1/6 ram

also threadripper
quack you're not even a fair comparison
he is richer....
far richer
even when considering my grants
trie actually
If he plans to run game VMs on it the Threadripper is better for performance no?
would you give your idiot child a $1 million body or a $20k body they can crash a few times within budget 
now what about two idiot children identical twin daughters that also fight

pa

I am not your dad 
Here, go learn more math instead 
https://youtu.be/triNEat764M
Want to restore the planet’s ecosystems and see your impact in monthly videos? The first 100 people to join Planet Wild with my code ABIDE4 will get the first month for free on me: https://planetwild.com/r/abidebyreason/join
If you want to get to know them better first, check out their Yellowstone to Yukon mission: https://planetwild.com/r/abi...

what do i have to type into vscode to suddenly become motivated ⁉️
you open the vscode temrinal and type sudo poweroff
i don't feel motivated
and then you stop programming for a 3 month break,a nd when you're back from your break you'll hav emotivation
maybe
if not try again
nothing on the outside nothing on the inside
i'm at month 8
press enter of course
step one: leave your house
step two: don't come back
colors are changing but otherwise i don't feel much
you open youtube, you procrastinate for 23 hours, you sleep for 14 hours, and then you have a decent meal before you work
Me when "AssertionError: all notes are rest in {name}" ```kt
val restRegex = Regex(",(rest ?)+,")
fun cleanRests(csvPath: String) {
val file = File(csvPath)
val backupFile = File(file.parent + "/backup.csv")
file.copyTo(backupFile)
FileOutputStream(file).writer().use { output ->
backupFile.forEachLine { line ->
if (!line.contains(restRegex)) {
output.write(line + "\n")
}
}
}
}
ah yes the mythical 37 hour day
*23 years
didn't know anyone used kotlin seriously
i guess having the infinite java ecosystem handy helps
And Compose is 
is that the ui framework?

is it native?
Well either way it's really amazing to use
So far the only actually seriously usable GUI framework I've found
looks like it's copying swiftui, which does have glowing reviews atleast for simple stuff

same style https://developer.apple.com/swiftui/
Okay, just tested the STT logic for my stack chan firmware
We are good to go
Runs entirely on my servers CPU at a couple times faster than realtime
Based on Parakeet TDT V2 0.6B
speaking of, i found https://en.wikipedia.org/wiki/Software_Automatic_Mouth
which will suit my early needs perfectly, with some espeak-ng default voices too as an alt
Software Automatic Mouth, or S.A.M. (sometimes abbreviated as SAM), is a speech synthesis program developed by Mark Barton and sold by Don't Ask Software. The program was released for the Atari 8-bit computers, Apple II, and Commodore 64. Released in 1982, it was one of the first commercial all-software voice-synthesis programs.
Don't Ask Softwa...
SAM is surprisingly understandable for being purely algorithmic and from 1982
If you want a good low resource TTS engine you could just go with Kokoro 82M
Its pretty robust for what it is
Concat synthesis is great
nah i want the scuffed sound for part of what i'm doing
well
SAM is exactly what i was after
i just wish there was more concat synthesis programs i could use with my own voice, seems most resources have been taken over by AI/NN approaches
UTAU
i don't think they've said anything good for the past 5 years at io
spark seems cool
new search seems god awful
i looked it a bit but it seems a bit convoluted to get into and most tools not meant for realtime, couldn't find a nice guide. i mostly want to: record a handful of phonemes and drive it through code in realtime
ask youtube seems god awful
has anyone here tried antigravity?
is it as shit balls as it looks?
You could most likely just write one yourself, doesn't sound too complicated
i probably will
not that i'm complaining, most of the stuff i want for my ARG is kind of niche
also found out SAM is perfect for creating glitch sounds that i wanted
❯ rlwrap -cAR nc -nvlp 53
Connection from 172.16.18.14:65365
Windows PowerShell
Copyright (C) Microsoft Corporation. All rights reserved.
Install the latest PowerShell for new features and improvements! https://aka.ms/PSWindows
PS C:\xampp\htdocs\samurai\wp-content\plugins\malicious>
Catching a reverse shell is the best feeling in the world ngl 
a new version was released but it's mvsep exclusive
damn ruined my evening

how can you see into my eyes like open doors
ngl the chorus never went as hard to me as the rest of the song
Wake me up
me when my alarm (that is literally for deaf people) doesnt wake me up
Try this
I feel like this would give me a heart attack
2DTok has been going well
In the meantime I made a multi user live vc transcriber
I’m only missing a few components for the first proper version of my Neuro clone
have to use C++ 🥀
or
Roasted forever 🥀
power went out 
which means cold boot and full speed fans for about a year
need case so bad
noise pollution at home so i can tell them they arent allowed to build datacenters
it's already too noisy
Legendary Python LLVM pass?
Fellow Python user?
Get a thinkpad 
i have enough shitty laptops
Idk how I got a .DS_STORE despite being on Linux and windows
How long did it take to train Neurosama?
Ngl - cuz training anything (no matter how small) requires like 1.5 months of train time-
😭
(Nothing happened trust)
What kinda thing are you training to take that long, my longest training run of all time was 3 days or so
Wait how many parameter were u on?
It was something around 100 million
WHAT
Surely you haven't been training on a 20 year old CPU right?
im at 6 million params 😭
Training an LLM from scratch
And my eta is 1,200 hrs
Surely you're using GPU acceleration
No its the MacBook M3 with mps acceleration
...
Oh that makes sense
Mac will not do for AI training
Inference will work fine, but for training you will need Nvidia usually
Ahh yes I too was once hopeful like you
back in my day
Nothing but a cpu and dreams
As long as you've got an Nvidia GPU, you can actually train reasonable big models reasonably quickly
My 3090 can do a 100M parameter DiffSinger in about a day
Essentially, go GPU or go home
You can try use kaggle, they lend out 2 T4’s or 1 P100 for 30 hrs a week
If u don’t wanna spend a cent
ha
im doing 175k
(that’s ~34x less, for the mathematically challenged)
neurosama existed since 2021 and it's been constantly iterated upon
so try 5 years
Early Neuro was just the osu bot not the llm
She is a fine tuned model
So honestly not long
Unless he trained it on desktop
Technically 2018
The era of the only decent guides and sources being academic research papers and tenorflow/keras
Counter of all the things my macmini 8 2018 has done and still survived:
Dropped 3x from my desk (somehow)
running at 100% cpu for multiple days at a time
had a broke usb adaptor put into it and accepted it (also somehow)
I onve needed to iterate over two ynamic list in python in tandem. I used the zip function and just broke the tuple back apart
scratched me bcus the edges are oddly sharp
also had a diff laptop placed on top of it
made odd creaking noises
placed in the sun
My 2011 Dell laptop survived getting tossed out a second story window like a frisbee
how....
also survived the closet of doom and despair for 5 years
actually 7 now that i think abt it
and somehow
nothing is broken on it
I managed to somehow catch it by the ethernet cable i had plugged into it. I somhow grabbed the cable with the laptop only a couple inches off the ground
except for the sd card slot but im 99% sure thats actually a speaker
dam
Very angry roommate. I left that day
he tossed all my stuff in the yard by time o got back to get it
wait i just read you gotta train llms or does that include openai and llama or do you just have to fine tune those?
he later asked me to help him out. Told him to fuck off
just finetune if ur using base models
or use LoRA if ur lazy
oh okay thanks there was 1 more question i had and thats integrating into discord i tried using python since most of my code is python but i read that discord.py is experimental and very fragile should i use a server and a node bridge instead?
uh no
discord.py isnt experemental
it works fine
probably old logs
it does, i was having so many issues with it i even tried downgrading the version and still couldn't get it to work
???
well are u just using it for back and forth
like discord bot integration
audio receiving and audio playing in a discord call is what i was using it for
oh thats a bit experemental
the audio send is easy but the recieve is hard
you can work around it using virtual audio cable and routing the audio from ur system or discord window into the program
but that means that you have to be in the call too
yeah my old project used that but it corrupted somewhere and i lost everything, well thanks for the help ill start the vac import now

The computer is almost done! In this update I explain the mechanisms required for instruction decoding, which translates instruction bits into control line movement. This enables the computer to perform certain operations with a custom-defined instruction set, which I explain while writing the first sample program. While running the program, som...
what the fuck
a computer out of knex

babbage would be proud
that's actually genius... using BMP and zipping for lossless images
I'd probably rather just use avc or hevc cause it should get pretty close (or better in some cases)
webp works well too
isn't that mostly just png but worse?
I don't think so
I think it heavily varies but it can get extremely good compression
I assume it compresses the combination of files together, right?
so if it's a video, it would compress extremely well
yeah.. but I mean for like a video
not completely separate images
Bro, using specific compression algorithms for the task usually yields better results than using general purpose compression algorithm

zip is deflate
lossless png is also deflate + some extra stuff
you can also do
qoi + deflate
that typically works better than default png encoding settings

yeah, zipping seems to be around the efficiency of webp, eh
I just use oxipng for pngs
We are investigating unauthorized access to GitHub’s internal repositories. While we currently have no evidence of impact to customer information stored outside of GitHub’s internal repositories (such as our customers’ enterprises, organizations, and repositories), we are closely monitoring our infrastructure for follow-on activity.
if encoding video then
probably use lossless video encoding
like FFV1 or H.264 with lossless settings


ye
I think apng is pretty competitive
I remember trying avc and it was higher file size
or was it hevc, I don't remember
wrr
used sometimes
apng 
I don't wanna have to write my own audio protocol


try avif
avif is good for lossy but I don't think it's super competitive for lossless
is quic too much
it av1


Intel® Inside™
not sure if good for audio
WHALES
but, really, I do think apng is really efficient if you want truly lossless
also, avif doesn't support rgb I think
do try a proper lossless video codec 
hevc is pretty good but I think apng was better for rgb or 4:4:4 yuv encoding

av1 is known to not be great for lossless
I'm doing discord vc -> code (i.e. python library)
ye then QUIC is fine probably
because I keep seeing chatters making increasingly terrible solutions
encoder issue
what about
send VC audio through
ICMP echo requests

I didn't think so either cause you know hevc can do interframe encoding
but idk
pings
maybe it just depends
send audio over cookie
build robot that sits in Discord VC and types audio data using keyboard
Look at mumble
what
Unless you really want to roll out your own

it spits out the
the the
frequency data
as text
so you can imagine the sound
unironically
I have considered writing a radio <-> slack bridge
because it would be funny

🐇 
This cat is a terrible driver 
wait wtf yeah

discord thank you so much for obliterating colour profiles from images
but only sometimes
Is it really worth it
yes

how tiny


evilwajaja
ye snack 
NOT make carpet
ye no make carpet
is this the bnuuy named snowball
non-lop lop

dont squish it
i wont

no 🦊 
yes :D

scope crepe
me see food
food make hungry
me eat random pantry snack now


shadow you're a nerd right

hypothetically, if you wanted a library to pull audio from a vc
how would you want it to look from the consumer end
Wym like user interface or?
yeah
To do what recording or?
literally just take a discord vc and give it to some arbitrary code
i.e. for people who need to give discord vc abilities to their discord gf ai
Just use virtual audio cable no? Or do u mean a discord bot or
Maybe an input where copy link/copy channel id can be used and then expose the channel audios as a stream in a protocol buffer
oh true protobuf could work here
Yeah way more efficient than sending shit via json cause serialized
I never said anything about json lol
moreso that I was gonna fully roll my own protocol

Or create webrtc streams
webrtc feels wrong
I don't need audio to get dropped
like the audio isn't going to a human's ears
I might just use grpc
I probably would roll my own here
but ye mabe gRPC or something also work
idk if I'd rather roll my own or grpc
since latency isn't important it doesn't seem that difficult
just have TCP connection and then send the samples with simple frames containing any necessary metadata
hm
yeah honestly that's probably best
also tempted to figure out how easy it'd be to link to davinci resolve
which is made harder by my davinci resolve install not working

must fight the urge to modify my minecraft packet macro to work for this protocol
that thing was majestic
wait chat I can just use websockets

side rant, why does obs have like no support for timecode
I just tried github copilot's cli just as a test
safe to say I am not touching that ever again
i used it quite a bit... it does seem to be a fine harness.. but it eats tokens like crazy any small change and its at like 2 million cached tokens. And i don't like how it doesn't show you what has changed
oh yeah speaking of harnesses, apparently they decided to update antigravity to the new one that isn't actually an IDE instead of the one that's an IDE and i don't think there's a way to move past convos back 
well done google

also gemini cli is being replaced by antigravity-cli so that's cool, who needs an open source harness anyway
Mumble uses grpc
it's a good harness in aesthetics only it seems
it couldn't figure out a simple docker build issue
I lied. They used the protobuf only 
weird.. i didn't really have issues with it not being able to fix things..
I don't really wanna build this on mumble
because:
- the target consumer is code, not a person
- I have a pre-existing dislike for mumble due to lack of functional apps on mobile
You can use their protocol and library, not needing to roll your own
If you don't need to do anymore processing then technically the best way is to just open a raw socket then make it stream the data there
issue is
need multiplexing
for multiple channels
Not on the requirement 
surely ffmpeg can handle this 
I mean
But the bot will listen to all through discord proprietary way no? The problem is the sending it to other code
I have actually had this setup pipe to ffmpeg in the past
it worked kinda well
haven't touched it since
died though
it can read/write from/to arbitrary sockets right
So best way is to listen to multiple port to handle multiple code listening 
I thought this is all a problem to solve AFTER the bot managed to get data from VC 
ye it is I think 
getting the vc data in the first place is also more difficult because encryption but has surely been solved already
See, then just listen over multiple port 
do I look like I live in 1995
ye



How do you have that font?
stole it 
from
the cia
ye
GitHub got compromised. Entire source code and private user and org repos are for sale.
Microslop is slopping again
wasnt me
I would advise everyone to change your passwords as internal dbs including hashes most likely got compromised as well
sauce?
GitHub revealed unauthorized access to about 3,800 internal repositories after a poisoned Visual Studio Code extension infected one employee device, likely from a supply chain attack on Microsoft's Marketplace. The hacking group TeamPCP claimed responsibility, boasting of stealing source code and tools worth $50,000 on a cybercrime forum. GitHub...
So I've ported my tokenizer to rust and it is some amount faster (idk how to check) so its now time to port the MLP thingy (technical term)
Also the luau tokenizer used integers to store tokens but the rust one uses 1 byte per token
How is this possible? Is each token 1 single character?
Why not just implement BPE or what's wrong with it
From my limited testing my custom embedding magic works better with 1 token = 1 letter and I've got the compute to do it
So I'm just doing it that way
Also rust is mad at capital letters and that makes me mad at rust
Training a llm with this is very suboptimal cause
U would need extreme learning density
To break down everything into single letters
Yea I know its suboptimal
But it works better (in my testing) and I've saved enough compute in other places for this to be fine
So I don't see a reason to not do it like this
?
Sorry I was talking about the topic above
😭
byte level modal collapse machine
beep beep
inb4 dude feeds back whatever tf the model spew as a training data and call it "feedback loop learning"
innovation
dare to be different
dudes just doing everything thats an obvious "DON"'Ts" in machine learning field
fail in new and better more exotic ways
https://github.com/dmaivel/ntoseye
I wonder if this breaks if Hyper-V is enabled in VM
(it probably does)
Maybe if you use the COM port backend
Big fan of being able to just vibecode new "features" for it tho
I like how people assume my unconventional magic is just insane / won't work yet I implemented the system before and have had it working just fine with a 1.5B model
It isn't like I'm reinventing the wheel by using a byte level model instead of normal BPE
if people wanna be sceptical of my methods then why not be sceptical of me using meta learning instead of backprop
But you managed to get it from the cia
im getting this modulenotfound error and im completely lost as to how to fix it
Requirement already satisfied: numpy in C:\Users\ninte\miniforge3\envs\ece105\Lib\site-packages (2.4.4)
PS C:\Users\ninte\Code\More shit i need for school\Lab5\starter code> & C:\Users\ninte\AppData\Local\Python\pythoncore-3.14-64\python.exe "c:/Users/ninte/Code/More shit i need for school/Lab5/starter code/pso_harness.py"
Traceback (most recent call last):
File "c:\Users\ninte\Code\More shit i need for school\Lab5\starter code\pso_harness.py", line 14, in <module>
import numpy as np
ModuleNotFoundError: No module named 'numpy'```
this is probably gonna be some virtual environment stuff
someone else recommended that but i have no idea how to set one up
moreso that in, you're already using it
installed numpy globally but not in the venv
if you're using some big smart ide like pycharm etc, they might have their own little interface somewhere to handle installing dependencies
ive deactivated every env that im aware of
im using vsc and i dont like it
oh yeah also idk if working directory affects it
not sure otherwise, never spent more than 3 days using python at a time xD
python is terrible and i hate working with it but i have to for school
based
python is Perfectly Capable™
me when im handed a 1000 line python file and all the variable types are ambiguous:
yeah but thats like... work
ah yes the python typecheckers which barely work
python is a scripting language, its a valid criticism of scripting languages but not of python in particular imo

strong typing is convenient for
systems programming languages
python is particularly bad at it though, same as ruby
at least its better than bash
the extreme dynamism means you barely know what you're getting, and it's exponentially worse in larger codebases
string or integer
yea but you aren't writing 1000 lines of bash
python is mostly used to tech new programmers the fundamentals before moving on to other languages
i don't understand why people decided python is good for beginners, it's absolutely not
python is the language that anyone who wants to make something, and isnt a dedicated programmer, will use before other languages
due to its huge ecosystem
^
usage sure, but it's a poor choice for education though
i mean its kinda good for really small scale 1-off projects
i would pick C over python, anytime
but even then
python is amazing for data science and ml despite it being a scripting language
i should learn more programming languages
rn i only know python, java, and C++ and i havent used C++ in ages
such that its not uncommon to prefer python over things like julia or R for data science
literally only because people chose to use it for ML, none of the actual ML code is written in python
cross platform file-system:
the python side is only bindings to C++ libraries
yes, the shared libraries are written in cpp or c or rust or whatever
but

the programs are written in python
yea because one corp decided to use python for the bindings of their proprietary ml libs
the ml is written in python, it calls to libraries like torch because they exist, work, and are fast
its because mathematicians arent programmers
mathematicians use R
r or julia or python or matlab or c or
but python is simply the most convenient
R and julia are good, ive used them both for datascience
they arent as flexible as python though
sure, but python is only a recently popular thing, mainly because of the 1 or 2 proprietary ml libs and the gold rush after them
in DS/ML specifically
python has its place
yh python is very flexible and easy to learn
but honestly it should be more for quick scripting
its not a bad language, and its not useless
rather than building huge codebases
my point is it's a bad language to teach programming
tru 
i mean it is the easiest to learn when u r starting
but when u actually wanna learn programming seriously
it doesnt really matter what language you use to teach programming
initial learners wont find a difference
and then you have to unlearn literally everything you "learned" from python
foundational computer science is the same in every language
rly?
as long as youre happy
my uni makes learning C mandatory to understand low level programming concepts
ye it is
what kind of education you got that they're teaching foundational CS in python
if you want to do something in any c-family language, good chances are the code will look very similar no matter which one you choose, as long as youre writing from first principles
its a serviceable language, the slander is crazy
sure it wont teach you about memory management, but its value as a learning tool is not to be understated

"serious" programming exclusive to C 
low level != good for everything
i love being stuck on
if you're not writing microcode you're not serious
1 version of a cpu's microarchitecture
ye
1 SKU
breaks on firmware update
different version different microarch
actually just make ASIC
tru python is rly good for beginners starting out
its js that well
if u wanna go deeper to low level systems programming
then u phase out and switch from python
if you want to do systems programming in python, its like telling a horse to fly
surely this is obvious

Alr my dad found the problem and apparently python was installed more than once
yh 
"python was installed more than once"
statements dreamed up by the insane
its very possible on windows
unless it's embedded in some yamls for the test cluster 
dont forget one of the 12 different venv solutions
wrr
whatever the hell python people are snorting at this point
the distribution of any programming language on windows looks like this
its not just python

its far worse for c actually 
nah, grab tcc and compile anywhere anytime, stick it on a usb stick
dont write c on windows its not worth it
i think it's good for baby's first taste of success
like you sit down a guy who never wrote a single line of code, you have to explain less things for them to do their own personalized witty version of hello world or calculator than in idk java
and since a lot of people especially today run on those instant gratification successes to not give up 25 minutes in
it became the beginnerlang™
honestly i would use Lua and Love2D for an absolute beginner
easy graphics, also reads like english
ye
love.graphics.print("hello")
someone new will be confused by the existence of love in your program
practically no boilerplate
code at least looks fairly intuitive for small programs
someone new will see and appreciate that hello appeared on screen
tutel graphic
overestimate the cognitive power of new programmers
the first thing theyll do is give up and ask chatgpt
nah, i dont treat people like morons

can confirm
if they use chatgpt they would fail my class day 1 never to return
Ahh yes, the language with famous mismatch between the spec and implementation. Which ended up being a bug in the spec 
What about Scratch for a first programming language?
scratch is another good option
bless dad
restricting the number of things a grubby little programmer can do
whole language visually there, impossible to assemble wrong, immediate feedback
Pegassus 
is good for teaching them how things work
python people do nixos on the filesystem level
comenius logo 
@tender river chaygpt
decipher
output from markov chain
Insane statement 
why can't i find an english screenshot
if they did that it would be the bare minimum of competency
Having parents ftw
are you telling me this thing was only used here
the statement doesnt make any sense to begin with 
makes perfect sense to me
wrote a line of Python
ye
brain forever damaged

o7
did you know that writing a line of python can get you banned from governmental jobs

here are 10 dangers of writing python which we will go over in our article today
look out! #6 is the most terrifying
clowm
what in the expanding snak
baboon enamel
Look where numpy was installed and where you’re running Python from
Venv moment
Whoa
there was more than 1 python installed

Oh lmao doxed my email
I've had that since I was like 7 years old
extra free claude code credits
(effectively)
Too late, I've signed you up to 70019374 different newsletters and I've received £0.12 for my work
didn't know you could do that with PowerShell
i always wondered how to use or what even PowerShell is
when it came to like windows 10
Well it’s a powerful shell
fair
Turns out that cellular automata are perfectly viable for image encryption/obfuscation
powershell.exe -ep bypass
okaayy
8 trackers with lsm6dsr now
and 4 trackers with lsm6dso that i decided to use for my arm and its somewhat kinda ass in comparison 
waiting for my lsm6dsr batch to arrive is pure agony
also pushed 4th rev of my custom pcb for printing. going to add magnetometer for calibration mechanics
man i need to move all the SMD to that board later god i hope the lsm6 batch arrives first
else, desoldering is
pain
honestly might decide to build from scratch with the new board and just make the previous one my spare
went too deep into vr i spent a lot of time perfecting FBT now without spending any much
also compatible with slime vr so poggers, though i noticed some issue with mounting gotta look into that one 
i wasn't expecting it to also be pretty much comparable in terms of battery life compared to the official slimevr set afunyun got despite having smaller batteries
also tracking quality hmmmm
afunyun skill issue with mounting and calibrating/tuning them lmao can't test that yet
when can i buy the V2 quack trackers? 
i was just about to cancel my plans on mass producing my trackers when i realized the official slimes are like way better
but ehhh
might reconsider
after spending like 14 hours playing with afunyun and his scuffed trackers lol
i can jsut buy the slimes if need be
i jsut thought it would be funny to get some third party jank
jank 
honestly

slime vr rn way better than vive/tundra trackers or any other lighthouse trackers
apparently you get the same quirks with having to calibrate shit anyway
their ICM-45686 implementation is pretty damn good there's practically no difference with the vives
Surely if
designs trackers they will work with Linux
it does

i said what i said
Using already pre-existing drivers
🦞ro 🐇 🤨
6
6
2
☆♡ 《❤️bny🐇lop•idol~★》 ♡☆
🐇
i initially used my own firmware
oh the tracking vibrates... i thought you installed like haptic motors into the headset and it vibrates on your head
well, still using it because im more familiar with it than the slime vr software
but both works
honestly it does that sometimes... i had a similar issue even on windows when i first got the headset like 10 years ago 
sometimes it just dissapears...
but the tracking quality was rarely the best (might have been because one of the tracker sensors only worked with usb2? idk it never worked with usb3 for somereason)
it works fine on windows for me, only linux vibrates
also IMU based trackers are not obstructible
no base stations needed
you can strap the trackers under your clothing
i think its something to do with the open source drivers doing a calibration when i start steamVR, cuz how badly it shakes changes on each boot
and play under the blanket
you need usb3 for both base stations and the headset
one of the base stations never worked under usb3
so if yo used usb2 it might have sucked cuz you limited the bandwidth
I will make you send it for 100 times more 
computed_fn_65643 wants everyone to see me vibrate? 
usb 2 should be already more than enough for tracking
doesn't need much bw
The Oculus Rift (CV1) external sensors and the Rift S headset require USB 3.0 primarily for power delivery rather than raw data bandwidth
huh


USB 2.0 Ports: Typically limit power output to 500mA.
USB 3.0 Ports: Can output up to 900mA (or more, depending on your motherboard), which ensures the optical sensors do not brown out, disconnect, or suffer from signal degradation
You'd be surprised at how efficient binary data stream can be for position only
Time to undervolt trackers 
If you have 16-bit per axis, that's like what? 6 byte?
6-byte with polling rate of 500Hz is 3kbyte
wdym position only? they're cameras, they provide a full 1280x720@60 over USB

this is why lighthouse based trackers are inferior now tbh
prone to occlusion
so you gotta face only in few sides
honestly i might try recording myself dancing with my trackers
idk man
inside out was terrible when i had it
especially due to lighting
lighthouse is far more accurate when its setup right
especially if ur doing something like beatsaber
im shilling the IMU based trackers
occlusion is far worse for inside out too
thats why the steam frame will fix it. it can see in the dark
idk about inside out
since i just built one
idk why they didnt go for IR just like the lighthouse ones
it does use it
why waste the power doing visual processing
a big issue still is if u move ur hands behind ur head
idk if they have that solved
idk why IR is not a standard most photoreceptors can do IR
the only decent tracking is a radioactive isotope in each place you want tracked and lighthouse trackers with xray cameras
based
idk about that...
literally never need to charge it
idk their tech
the rift is IR
so ill have to read into it
tldr
- people in vrc can't differentiate between the two (slimes vs vive/tundra)
- tracking itself just on par rn
- not prone to occlusion, you can fit the trackers inside clothing and blankets
- better suited for mocap since you can add more tracking points flexibly
- far cheaper
idc about that stuff though i want accuracy for high speed
ignore certain reddit post where they tell you to not get slimes when they themself got the shitty third party ones with shit IMU
yeah
ideally you'd use lighthouse trackers cuz you dont need ot charge them, but you need to make sure they can see behind objects and shit which jsut makes it not that easy.
what were those sony cameras that were banned cuz they could see through clothes? i need some of them
its pretty accurate and high speed now
uh
you still need to charge them
slimevr is dedicated more for like the vrc chronically online ppl
nah, its outside in
idk how it performs in max level beatsaber or pavlov style games
haven't tried, though i read its also pretty good now
though i suppose lighthouse could be more convenient
but its apparently not really plug and play either
yeah
and still doesnt like occlusion
so yeah not for accuracy
IMU are literally independent?
Testing if ai can reconstruct them or if they are secure
Among several other tests
not affected by occlusion at all
For other things
drift is a thing though but well, they came up with magnetometer design aswell now
the drift is probably a big reason why this hasnt caught on
to actually actively calibrate it



WebRTC more reasonable if over internet probably


