#programming
1 messages · Page 339 of 1
wasm multithreading 
ill be waiting for cybernetic eyes
no
not happening
well
maybe in c++
i’m not suggesting that
not in python
unless the newest wasm spec made it way better
i don’t remember
inconceivable
ye red is orange to you
and the embed perms color actually is what it looks like
brownish orange
soundcloooooud let me login 

waht the fuck
or maybe i’m brainrotted by the bright ass role colors here
wrrr
banana isnt normal yellow
mods how do i get red?
Its orange to me 
this is normal yellow
maybe you have nitro basic
you used to have regular nitro then
when did regular nitro get replaced again?

grrr
Reject GUI, return to terminal
I see SamVanTriangle has become orange
i now have a sliding bean, idk how the fuck it works but it works ill get to learn it after hand writing all these codes again and again fr
its a rigid body i know but ill get to it to learn the one that uses complex movement

The classic sliding bean, you using apply force to the rigid body or setting position manually?
i think its the latter?
oh im dumb
it literally says there
(apply_central_force) lulw
Its the better way, directly setting position doesnt collide with targets and would have given you headaches later.
whats the difference between the two exactly?
so, setting position (using transform.position=new vector3(x,y,z) ) will put the object at that position, it's simple but it completely bypasses the physics engine. applyforce applies a force to the rigid body which propells it forward (or in whatever direction you pick)
although i just noticed now...your code isn't unity...
you tell the physics engine to push you instead of just moving
godot
the principle is the same, but my example of setting the position might not apply.
oh so if you want to make ur own physics u do use the latter?
explanation still works just treat code as pseudocode
when you are talking moving around rigid bodies the real code is physics, any programming language is just implementation
tho better word would probably be teleporting not moving
for this analogy
i think apply force is better in this context then
ye it’s the only correct way basically
so, rigid bodies are a concept from physics, your game engine has a physics engine if it implements rigid bodies. applying force is certainly the better way to move things. only time you would wanna use direct movement is for something like an animation and you wanna be real careful with it
if you accidentally stick a rigid body inside another rigid body they tend to explode
if you want items to behave in world then apply force is better and if what you said is right (teleporting) everytime you move is pain
shit, need to learn another code then, im thinking of making an animated player, but thats for the future me, im just learning how to make a testing world atm
nah you don't need to worry about that too much. the most common thing is to have the mesh be animated and then propelled...most characters are just floating beans
it's why you will often see clipping happen in video games, the character is just a bean and the mesh is cosmetic
and the visuals of like grass flying when you walk into grassland and sand is put ontop of it?
i see i see
best to have fun being a bean for a while
all the rest of the stuff will follow in time
what does this mean? , you mean when ur playing the game and you just walk upo to another rigid body?
if you clip into an object
nah, this usually happens with teleporting, if you teleport a rigid body into another rigid body the physics engine freaks out
yea playing with this bean helped me actually know how the world light is and how the camera positioned
sometimes it can happen when applying forces if the force is greater than the physic engine's time step (basically if you have a LOT of velocity)
now make camera movement
happens in AAA games and indie games alike...so don't sweat it too much
yea thats the next step onto this tutorial i was watching
im just taking a break
Is there any reason why the hack solution of sending Neuro a "[x]'s [bodypart] touches your [bodypart]" prompt on collision detection wouldn't work for giving her a sense of touch?
reacts too slowly for it to work at all + whatever the movement model is needs to be trained to react in response to that kind of collision + you need some sort of collision detection that detects specific body parts (which is probably something else that would need to be made) + it seems like such a dumb way to implement collision
ye i can't think of one, except maybe vrchat constraints
api constraints i mean
also why is it a "hack" solution, how else is an llm supposed to know something is touching
I could come up with it, which makes it a hack solution by definition
Is there an alternative in Vrchat though? I dont think it has a system for collision detection so you'd basically have no choice but to have some kind of ML detect collision from the screen no?
that sounds awful to implement...
Like yeah it's gonna suck but what are your options?
I guess you could build your own vrchat but that's an insane amount of work too
for neurosama to have touch i think she will need another AI that integrates with her. this is actually how your brain does it, you have a small brain in each vertebrae that handles communication with different parts of your body. what people call "muscle memory" is actually vertebrae memory, and that is also what handles reflexes.
in the context of neurosama, you would probably need an AI that translates pressure (determined by force applied between two rigid bodies with a collider placed on the fingers) to something neurosama's LLM can understand and translate
i don't know anything about vrchat besides the frontpage summary, but maybe you could mod it somehow?
iirc there were some haptic suits that would let you "feel touch"
so should actually be possible
There are? In that case vrchat must already have a system to handle and export it. What was tutel complaining about on stream then?
i don't think vrchat has one is the problem, it's either a mod or a plugin or something like that
the tricky thing would be that the body itself isn't physical, it's animated...going back to the moving bean discussion from before, so adding a collider to the fingers of something that isn't being handled by physics (even if the collider is just a trigger) would...be tricky.
i have 0 experience with vrchat, even playing it
Well as long as it works and doesn't break their TOS somehow it should be fine
yeah i can't speak to VR chat either, best i can speak to is unity. if you wanted to make an environment for neuro that she could move around in.
they added AC at one point which is the main problem, the haptic suit integration could have existed as a mod before then
they added air conditioning, no way
can't exactly mod a game with anticheat
not alternating current?
I really don't think neuro needs simulated pressure data lol. That's setting the bar way to high for a one man dev team
otherwise it should be easy vrchat is just unity
she just needs to know what part of her model is touching what other object/model
i'm a one man dev team, i've done similar things as part of the game i'm making.
object seems like a bad idea
a binary "touching"/"not touching" would be simpler and at least a step up...not as nice as analog touching but better than no touching
sure but the turtle is building a streamer, not a deep simulation game
Don't get me wrong it would be cool, but the returns on investment would be basically none
the returns? you get to program a simulated touch for an AI...that is a return itself
better to just slap together the simplest possible solution
:evilDentge:
But from a business perspective you're just introducing more expensive to maintain code for basically the same result
Used to be involved in space station 13 and the single biggest problem we had was people adding an endless number of MORE GODDAMN SYSTEMS that made every refactor break a hundred more things
can't really speak from a business standpoint...i just write code.
granted, it being an open source project had a lot to do with that. Lots of people merging complex stuff only they understood and then leaving forever lol
I think we had like two guys who had any clue how the atmospherics subsystem worked
what was your PR procedures like?
Anyone could open one and then maintainers yelled at the PR author until they were happy with it
I would describe it as "not very rigorous"
Seriously though... There is no UI for the models, user input is passed through VTT, evaluated for pipeline path, prompts generated, sent to models, returned for TTS. None of the features that are used by tgwui are needed so I cut 6gb overhead per model
we had bi-yearly player revolts because one of the more colorful personalities among the maintainer team decided to make the game unplayable for shits and giggles
How many levels of inheritance did the project have (if it was OO)?
did you have any kind of automated testing?
Don't remember exactly but something like 8 i think?
a bunch
Oh wow, that's alot :o
hang on i have it somewhere on my computer still
8 levels...wow...my classes usually have like 2 or 3 maybe...not counting whatever monobehavior inherits
Running task specific models is actually often much more effective than one big model.
I run an 8b model for conversation trigger queues/thinking gating/emotional updates, a 24b instruct main model, 24 b reasoning "thought" model, a lightweight embedding model, a post run 8b enhanced embedding model, whisper v3 large, and will likely add a vision model as well as others.
Not only that but its written in BYOND DM
AKA the worst language on the planet
(it really isnt but it sure felt like it)
that's what i'm saying. the human brain has like multiple brains piled on top of each other and one in each vertebrae.
Never heard of it until now
because it's literally only used for space station 13 lol
It's actually better that her main model isn't doing most of it. Larger models != better
how good would you say this whole setup is? this sounds like it takes something 24GB+ VRAM
i've often wondered how she would do with a unitree g1 robot. i seen someone had hooked an AI (claude i think) up to one of these to control it, so it should be doable.
are you trying to beat silicon valley to the singularity or something?
Cant think of a practical reason you'd build that so I'm assuming it's purely for the love of the game
the neuro singularity!
I'm running a 24b with 32k context on 3090ti #1, reasoning 24b model with 32k context on 3090ti #2, 8b gate model with 8k context and whisper v3 large on an an A4000, and the embedding models on a A2. Currently sitting at 6/80Gb free vram
The man behind the RAM shortage
here i am on a gaming laptop from 2019
im just trying to do something similar to neuro but i can only hold like a single 3b instruct or 7b quantised lmao
have you considered installing more RAM?
💀 damn
ayy same buddy
I'll get this bad boy to last for 7 more years
when i get enough money for that, tho i am currently laptop user
Was already in the LLM space and was watching the Minecraft stream and thought "I could probably make something in the same general vein as Neuro".
What I found so far:
- yes I can if I keep at it
- Vedal has every reason to be an alcoholic
i got like 16 gb of ram.
still got that 90s mentality where having more than a megabyte is a lot.
true and real
You should yoink the twitch name dark_vedal987 and start up a rival vtuber
I'm running Exllamav3 (exl3 quantization) 6bpw HB 16 for 24b models (18Gb base + f16 context) and 6bpw HB8 for 8b models (7.3Gb base + f16 context)
That's not including everything else that isn't gpu on the main pc + memory DB server
well i do get that neuro is more than well engineered inference, pipelining and a really good memory setup but it so hard to get back of it
damn, i would take it also costs quite a bit for it to run too
See that's the funny thing, if I were to get to a point where I actually ended up doing some kind of streaming, I wouldn't say it's "competition" there is a key difference in my methodology that heavily shifts things: Vedal has designed Neuro as an entertainer first and everything else is a byproduct. I have zero intended goal. The project will go in the direction it goes
recently i saw someone in yt made a model like very old version of neuro
When Vedal mentioned that his heating bill was cheap now but his electricity bill was high yesterday I started laughing
he did mention his bills before when neuro revealed twitch income for a month, cant say how truth it is
The solution is moving next to a power plant and mooching off their grid btw
Put a massive coil nearby, fluctuating magnetic field might let you steal some current.
Alternatively you can get a job at a larger company and set up your server tower in some dusty storage room. Bribe the IT goblin and everyone else will assume it's supposed to be there
There is definitely things that I know are different in how I approached them vs Vedal. I'm like 95% sure I'm using a different model family and I probably am using a different memory DB setup because I have the luxury of dumping a massive DB into ram but other things (like discord voice libraries having aweful documentation) are the same pains
discord has voice libraries? I did not know that
What language are the libraries written in?
There is a good chance that he's using at least some cloud compute which would explain it. He also probably has slightly more expensive electricity but power alone doesn't explain 7k if that's the case
Python
you need them for speech, kinda hard to setup but chatgpt helps there a bit
Are they dlls or source code?
The voice libraries are the implementation of voice chat audio streaming
Oh. I'm dumb lol, thought you meant the actual text-to-speech
could be yeah, he also has to deal with voice model billings
I'm not exactly sure what he's using for TTS (especially for evil) but TTS isn't too bad. He isn't using the same TTS I use though as I specifically went with one that wasn't multilingual
discord is very restrictive on how it wants the bots audio stream to work so thats a bit of pain
Neuro uses a stock Microsoft Azure TTS "Ashley" but with pitch turned up
It's poorly documented, they expect you to handle any invalid data they send, and the libraries have aweful voice documentation
i guessed it would be something like ElevenLabs
Interesting, yah I've been running piper for now but at some point I want to move to something else. I don't really like the cloud options.
As it is there is so much more I can do with tweaking modulation based on emotional response so that's a problem for later
I would wanna use the merlin voice from win 98
I can see why he is procrastinating voice work
Audio ML is just not really trivial and it can never really be perfect
I specialize in singing personally, really fun
idk where chatgpt got it from but it can help build audio streaming which works with discord from scratch, tho dont think it will be as good as built libraries but you would know errors better
i see
The problem is that the libraries have non existent error handling and discord will silently drop the websocket. Debugging errors is a pain
i see, will check some library and compare with my current code could be worth a shot if it can help with improving debugging
I think in theory I could sequence the vods into DNA base pairs and use CRISPR to append it to my genome 🤔

https://x.com/__Tkat__/status/2006403626709999784 but does updating a link get an embed randomly
nope
lloool
yuri is my sleeper agent activation word

Not for me
You have not Unlocked New Role
Ngl, idk what that means.
you will find out 
Ominous...
eat leaf 
I'm so glad I saw this coming and got in on the ground with a 4090 at the start of 2025
If you were looking for the character from Pokémon: Let's Go, Pikachu! and Let's Go, Eevee!, see Green (game).
Leaf redirects here. For other uses, see Leaf (disambiguation).
Leaf (Japanese: リーフ Leaf) is the female player character in Pokémon FireRed and LeafGreen. Her male counterpart is Red. As such, she—alongside Red—is known thr...
official localization say 茶リーフ
0.7 interesting. Id have expected her to say 1
ye that name source 
because
in fubuki chat
use 茶葉 instead of 草
new chayleaf lore reveal 
im trying to think of an equivalent to this in english and
grass (lol) -> tea leaves (lol??)
probably no equivalent
ye learn new language unlock new pun
it even funnier when you're bad at it because take you longer to get
is an array of colors a colorful one
this has to be scientifically proven
tuple[int, int, int]
one definition of colourful (colorful)
Full of color; abounding in colors.
trait system
pun too complicated hard to translate
aacefhly <- cant translate
i give up
no keep trying
chayleaf <- 01234567
aacefhly <- xy057143, { x, y } ≡ { 2, 6 }
this required more effort than i thought it would
mentally
multilingual (spanish and python)
multilingual (spanish and python) (also works in english and rust)
i have not confirmed if it works in other languages

but its provable that it doesnt work in all languages
list[Tuple[int, int, int]] <- rust
(python)
ye need translation
work after translation


Tuple[int,int,int] surely thats a triple not a tuple

frog nog
twople

people.iter_mut().for_each(Person::ok)
people who not ok will be reduced to none 
problem solve nobody panic

why are people enum
my rust is rusty it seem
ye it definitely doesn't return that 
Result<Person, _>::ok <- return Option<Person>
but cant infer _ in this position 
ok so
so use |x| x.ok() instead
add generic
fn banana<E>(p: Vec<Result<Person, E>>) -> Vec<Option<Person>> {
p.into_iter().map(Result::ok).collect()
}
rust generic bad
no full program type inference 

wrr
might as well be using java
fn banana(p: Vec<Result<Person, impl core::error::Error>>) -> Vec<Option<()>> {
p.into_iter().map(Result::ok).collect()
}
still not inferred 
double you arr arr
close enough 
fn banana(p: Vec<Result<()), impl std::any::Any>>) -> Vec<Option<()>> {
p.into_iter().map(Result::ok).collect()
}
fn main() {
banana([Ok::<(), ()>(())].to_vec());
}
rust

that's crazy

Cleanup complete
Space freed: 15.21GB | Items cleaned: 3997 | Categories: 41
15.21kg 3997 flies 41 traps
A typical housefly weighs between 0.00007 and 0.0016 ounces, which is about 2 to 45 milligrams.
kg unit still too big


how the fuck are they so light
chatgpt generated
with future ai technology
humanwill be able to harness this power
use to create
light material
funding please
15.21kg of flies is 638820 flies
volume 3.5mm^3 per fly
maybe multiply by 1.2 or 1.5 to account for space between fly
me use 5mm^3 for nice number
it something like
3 square meter?
square 
probably very brittle
we need to add superglue weight to equation
wait i have giant pile of dead flies near a window should i start supergluing them together
probably not 100000
fly ball
no need for glue
just buy 1 square meter box
put in there
pour glue from above once reach 5kg

1 hypercubic meter box 
fill three dimensions with flies and the last one with superglue
what if i modify the C compiler to make C fancy?
so like:
#include <stdio.h>
abyss MyFunc(int MyParam) {
when (MyParam == 0) {
printf("The number is 0\n");
reconvey;
}
otherwise when (MyParam < 0) {
printf("The number is lower than 0\n");
reconvey;
}
otherwise when (MyParam > 0) {
printf("The number is higher than 0\n");
reconvey;
}
otherwise {
printf("Error: THIS CODE IS SUPPOSED TO BE UNREACHABLE");
reconvey;
}
}

can just do with preprocessor 
but
languages made for keyword or syntax reasons is like the PL equivalent of brainfuck clones
Yeah you're right.. i should base it on objective-c instead
abyss 
typedef void abyss;
#define when if
#define otherwise else
#define reconvey return
"fancy" c
Funny thing... i tried this before actually... it doesn't work like that appearantly... that's why modify the compiler
why wouldn’t it work 
it does actually work like that 
for better or worse (for worse)
it does in fact work like that 
Language wise i think objective-c is alright... it's just that it's designed in a way that to properly use it you have to do some wierd import things
Then why it don't work for me?
objective c is probably the most dynamic language i've ever come across
swift (objective c if it was good)
toast not use smalltalk

Swift is actually good yeah
i didn’t try macros after all
you should its a fun language
even if maybe not very useful in practice because it live in its own bubble
one day maybe 
Dam... well.. w3schools online environment just broken af then
are there even usecases left for it with swift being a thing for years now
legacy 
legacy like
probably some cursed low level integration
rust
true
when(x<20)
{
reconvey "less";
}
otherwise
{
reconvey "more";
}
rust legacy rewrite in haskell 
I second this
haskell
haskell best language 
i still have that exact nightly build i installed to fix lsp crashing
liquid glass
looks pretty solid to me
I still remember learning it back it in the day. Was cool to learn, but never worked with Haskell again
wrr its fun but me usually work on low level code 
gaseous glass when
i would maybe use it for server backends over rust 
gaseous glass isnt glass
assembly
Last time i checked he indeed is made of flesh and not assembly code
the electrons are speaking binary
wrr
wrrr
night night 
what do you mean then?
hblang

no actually more like
js

js low level lang when me write it
anyway ye bye 
wrr
js still so popular
i will reveal answer if you become banana
This does not look right when scrolling by quickly
looks the same to me
they say you are what you eat. I don't have bananas
banana 2026 campaign
I EAT GRASS
browsers should switch over to haskell
wasm
Imagine where we could be if the engineers who made V8 fast worked on something actually good

How does she make typos?
LLM trained on human data -> can make common human mistakes
it's like the metal pipe for Evil.
Its an odd typo, iirc she said "myt throne" i wonder what the token for that would be, iirc llms predict forward, I could see myt proceeding throne if you were starting with throne and predicting backwards, but how would you come to use myt?
prediction as good as the weather forecast
Maybe it's a bug originating somewhere else
Iirc she IS multiple AI in a trenchcoat...so...
I wonder if the filters have any influence on the output besides banning whole words
reminds me of the english language
somehow it works
My brain is a bunch of neural networks in a trenchcoat (not literally...I dont own one) so I cant say much
its an error in the lm head/tokenizer
maybe you can rent one in the cloud.
How does that lead to typos? (I don"t know much about llms)
the LM head is the last model layer. between it and the tokenizer, the numeric optput of the other layers is converted to text
Ah I see
She is. there is no way even a frontier model alone would be able to behave the way she does as even those use model switching
she often subs single letters like a person would, it's endearing
single letters is an interesting one. Tokens are not words. On average a token is 3-4 characters. In theory two tokens that make up a word could be chosen that would correctly form the word other than one letter. That is the most likely situation. The other is that Vedal intentionally injected corrupted training data with inaccurate spellings. The second option sounds unlikely and dangerous to attempt for a quirk
AMD Ryzen 5 5600x
RTX 5060 Twin Edge 8GB
32 GB (2x16) DDR4 3200Mhz RAM
1 TB SSD
Monitor: LG 27" 180Hz 1 ms
Can y'all rate this setup for gaming? Is it good enough?
The GPU kinda on the weak side
The CPU could probably drive a much more powerful GPU
Depending on how much the GPU costs could probably get a better one used
I think I'd go for the 5060 Ti (or an equivalent previous gen card) if possible
8 GB of VRAM is not that much, and the 5060 is also just on the weaker side as far as GPUs go IIRC
Does Neuro's 3D room (non-vr) have any official reference file for scale? For people who want to make more 3d for her?
Case: Corsair 7000D
Motherboard: Asus X299 prime a II
Ram: 8x16 crucial balistix sport 2400 (128Gb)
CPU: Intel i9 10900x
GPU 1: Nvidia RTX 3090TI Founders Edition
GPU 2: Nvidia RTX 3090TI Founders Edition
GPU 3: Nvidia RTX A4000
PSU: Corsair HX1500i
Drives
C: WD Black SN750 1Tb
D: Samsung 870QVO 2tb
M: Intel DCS4600 960Gb
N: Intel DCS4600 960Gb
P: Crucial P1 1Tb
Cooling
Pump: Ekwb D5
Radiator 1: Alphacool NexXxos XT45 480 (push pull)
Radiator 2: EK coolstream SE120 (push)
Radiator 3: EK coolstream PE240 (push)
GPU 1 waterblock: Alphacool Eisblock Aurora
GPU 2 waterblock: Alphacool Eisblock Aurora
GPU 3 waterblock: Alphacool ES
CPU waterblock: ek supremacy
Fans: 11 Delta PFC1212DEs
Can I play Minecraft
no
thats me

im sourced from what?
Crap, in that case would 3 5090 astrals be enough
why do you even ask?
You use Windows on that? Why?
Linux would probably save like an extra few GB of VRAM

but get those 5090s anyway, please
i just cooked a full bag of popcorn without burning any or having any unpopped

im now going to eat all of it
your hardware has been seized by the triangle orginization for feeding the television


What if feed the NeuroSynthProject instead
you can have the crumbs

maybe a leftover gpu the tv didnt eat
feeding the tv is of the upmost importance tho
Unfortunate
I would really like having those GPUs, NeuroSynth would train way faster and be like 3x higher quality
feed the machine 





koniiiiiiiiiiiiiiii

16 i

🟨

yellow?
Various NeuroSynthing, not sure how many saw this but I am not sure how good it is
i will kill the next person comparing me to mcdonalds btw


I wouldn't have noticed if you hadn't said it now
Streisand effect already in action

not a single word

i said "the next" cuz it already happened before
First time I did anything finished with this dynamics style on tuning
No way Sam is the color of banned user now
stop or i will actually hire a VA isntead of using neurosynth
i wasnt joking when i said i was gonna crashout
OhISee
I've forced a power limit onto my primary GPU and a super agressive fan curve on my bottom GPU so that I can actually play games without thermal issues
Also I keep the window open to the -13 or so degrees air
I see you've chosen the colour of mayonnaise
we are not friends
:(
He changed it from white to green

Looks more like the inside of a lime or kiwi now
Are we tho?
we should be fine
Yay
for now
nah thats fine
Shouldn't you be like working on your new home?
Siwwy

No fonyon
And his middle name

Tbf.. he's the one that gets all my demos and "finished" versions

I had to do something like this on the 3090tis when I had the stock coolers. It helps significantly
Sily
me read something about rust
@
me click some random link
@
Handleable failures are a conspiracy by C programmers who can't guarantee success with the type system. 
should i have a pause icon when its paused, or a play icon when its paused?
play when paused, pause when playing.
the button should show what will happen if you click it
okay
im replacing some bad scripts with modules that do the same thing and such
but mpris is killing me
the {position} thing also increments when paused
I think if it's a button it should be "what will this do if I click it" whereas if it's just a status icon it should be "what is the state right now"
its a button ye
or well
the whole thing is clickable
so at first i thought of it as a status icon
but its easier to think of it as a button since im mostly clicking the icon
like the clickable section encompasses more than just the pause/play icon?
ye
hmmm
the title and such are also clickable
does it need a persistent pause/play icon then?
change the text from || to |> and back and not have an icon?
can take a cue from youtube fullscreen UI basically
could also use a font which has the icon
where the bar will have an explicit button but it will display the opposite Pause or Play icon when you actually do it

i am ye
what is the button actually meant to convey? perhaps you could show a red light or green light under a rectangle that you can press to represent stop or go?
I dunno if I missed it but yeah I have no context for what you're making lol
not very colorblind good though
if it's meant to be a player of sorts, then I think effectively copying some popular player interface is basically the way to go
assuming the intended audience fits that
not literally copy, just the basic concepts
and the flow
"a switch which changes appearance when toggled" could be any animation sequence, playing from either the start to the finish or backwards from the finish to the start depending on the toggled value. go wild.
could even have completely different animations
google glasses?
AR glasses yeah
Always listening AI RAG trinkets
https://youtu.be/dGA8-CkaSOk?t=42
"one of the brightest and widest fields of view"
Pickle 1 is the first soul computer that lives with you. Order now: https://pickle.com
Connect with Pickle
Website: https://pickle.com
X: https://x.com/pickle
Instagram: https://instagram.com/pickle
Just with google glass shit this time
ai powered google glasses
man, I dunno how many people actually used google glass back in the day but it was ass
it wasn't AR at all
what do they mean by soul
it was a sticky note in the very corner of your field of view
That's what AR glasses are now tbf, the FOV isn't that impressive
God knows
at least AR glasses now can have some level of tracking; well depending on which
This is such a scam lmao
like I'd call the Hololens legitimate AR, along with any VR/MR headset
Fair, though with the exception of Meta most of the current glasses are waveguide which isn't too different
Larger and sharper but still monocolour and projected without tracking
well, with a form factor like that, I assume it's that yeah
Personally the main thing for me with smart glasses is I don't see a practical reason to have them at the price point they sell for
the tech is cool, the glasses look stupid
Then there's also the privacy concerns which aren't helped by the fact the three leading offerings are Meta and two Chinese companies
I've tried doing daily tasks with a quest 3 on and it's doable
but I don't think I'd ever go outside like that
I will say the even reality G2s look pretty nice
In terms of actually being kind of fashionable
those are alright
They have a rounded one too which looks decent
But again £600 for a pebble watch on your eyes
Not great value
I think we're at a stage where if it looks decent, then it doesn't offer much more than a smart watch
ye
and for it to do something actually neat, it needs to look a bit dumb
need to assemble a team to make AR waifus before these can get popular
I did my thesis on VR in radiology but I could never get to the stage of actually justifying it yet
at best I could reach non-inferiority
but the AR waifus tho
The third option aside from Meta and Even Realities is Rokid, but while they have a healthy open software ecosystem they look dumb as hell
And like Even Realities you have to contend with the fact you're wearing a Chinese made camera device on your face all day
a bit too thick
but yuo need to put the battery somewhere so idk
well most people don't really care about stuff like it being from Meta
they just care about it being convenient
They're great as assistive tech, blind people also get a better value prop since they can use the ray ban wayfarers instead of buying the screen one 
That's like a quarter the price
battery would be a lot beetter too
Gemini 3 Pro throwing in a (yes, really) 
I love how the Gemini models sometimes just make off hand comments like that
Rendahhhh

you're paused, stop incrementing the time
That sounds like such an easy to fix bug
it is not 
Wait do you have source
playerctl just reports it wrong
Then that's a bug in the app your getting info from
oh
Mpris guarantees that when the player is paused the time does not go up
The standard guarantees it, the app implemented the standard wrong
What happens when you unpause?
Can it become something like
10:22/4:10
Or does it restore it back to actual timestamp after unpausing
nope, it just continues
So if you pause
Let it become 4:00/4:10
Unpause
20 seconds pass
4:20/4:10?

And we need the app to provide the info because it can be played from any moment of song not just from 0:00
ye
anything playerctl provides
Does it say if it's paused
Are there other sources of timestamp
Can we calculate it ourselves
Tracking current situation and taking time from other source
ye, so in theory i can save the time it got paused and keep it static, but when it continues it will need to use the one provided still
unless we increment it by 1 ourselves
but then it wont work if we skip 5 seconds
you can track weird time changes
Changed more than +1 = was rewinded
Add to our own timer
Is this the actual YouTube music player, or the normal YouTube player? Try it on the other one that you are on right now and see if it behaves differently
ye but if the incorrect timer reaches the end of the song that stops working
the yt music one
ill test the normal one
Cause I remember using eww with playerctl and it worked fine on normal yt, so it might just be yt music that's broken
hmm. the normal youtube one seems to pause fine
So yt music is fucked as usual
Can't you do AR as an entire room fitted with projectors?
or does it have to be via small screen only?
There would be no depth to the effect
They should invest into directly projecting shit into the brain
Fuck eyes
What if my vision is not perfect
Does it account for that
Get AR eyes 
If we keep the screen at 144p will you there even be a difference 
Oh this is clever
Its a handheld gripper thingy for collecting robotics data without teleop or master/slave setups
Might have to try my hand at making a scuffed DIY version
I haven't started building it 
Oh 
you need stuff to be shifted slightly for each eye to create the depth perception
True true
For now this is my data collection method
Ancient vr technology
How does it work? It actually moves things and records the results? Or its some kind of simulation?
You remote control the arms using the joycon gyro and buttons while filming either from the arms POV or from a third person perspective, then use the servo data and visual feed combined to fine tune a VLA on the task you want completed
Anyway for now this is all theoretical, I won't have my robot arm built until probably around the jan 8th
I'm midway through it right now but a part snapped so I had to rush an express printing order on a replacement
Typical PLA
Do you train it for 50h so it can pick up a can of beer
Why ai? What does ai do exactly
Mostly for visual stuff?
You know what, there's a good if slightly outdated video that'll give you the general idea here
I have summarized my experience with the SO-ARM100 robot from the general overview to the training process and the results I got.
Links:
- Lerobot repo: https://github.com/huggingface/lerobot
- SO-arm100 repo: https://github.com/TheRobotStudio/SO-ARM100
- DOT-policy repo: https://github.com/IliaLarchenko/dot_policy
- My Hugging Face page: https...
AI is used since it can self correct during the execution of actions and generally provide a degree of autonomy, and takes text input -> outputs physical actions
if you can go for PLA pro, go for it, it's only a bit more but it's a lot more robust
Yeah I'm using PLA+, tbh it might just have been a shoddy print on the suppliers part
Poor quality filament or something
if you're getting into the hundreds, it might be more worth it to get your own printer
at least for PLA
I think the SO-ARM100 parts were like, $10 of material total for me
on a $150 printer
For me it was around £90 but I'd also ordinarily just use my unis printers for this stuff
Only reason I got these shipped was because they don't like you using your own filament at the uni and they only offer standard PLA, Resin or PetG
you can get into the dilemna where your home printer is better than what's at uni/work
My uni has top of the line printers so it defo wouldn't be
We're talking ultimakers
I have an ultimaker method at work, and it's just so much worse than my home printer
it's a lot less reliable, I've had to fix so many individual parts on it
the heat blanket came off and started hitting a limit switch early for one
the enclosure just doesn't retain heat very well in general as well
and I used to rely on it when I needed a dual extruder printer for separate support materials, but I can do that at home now too
Other thing is I lowkey wouldn't want a 3d printer in my regular living space
Nanoparticulate and all that
it's why I built a sealed enclosure for mine
Its like microplastic stuff?
just a filter seems to work fine for an FDM printer unless you're printing a ton of ABS or something
oh yeah, the $9000 Method looks like it'd be enclosed but
it actually just vents all the nanoparticulate out
without even a filter
Vent it outside the window 
and you need to buy this ridiculous add-on if you want a filter https://www.matterhackers.com/store/l/makerbot-clean-air-system-for-makerbot-method-3d-printers/sk/MKSWKCUJ
The MakerBot Clean Air is a smart HEPA filtration system designed to allow you to breathe easy and keep your working environment free of ultra-fine particles which might be emitted by your Method 3D printer during the fabrication process. This add-on is ideal for customers who may use the Method line of 3D Printers in classrooms or engineering s...
I vent my resin printer out a window
that shit is really bad filter or no filter
Yeah 3d printers create an insane amount of not even micro but nanoplastics
so I modded mine with a duct and in-line fan to keep negative pressure at all times
it does depend on the material though
They literally bypass the blood brain barrier and bioaccumulate on your brain
Really nasty stuff
like ABS tends to be worse by a lot than PLA
True, but none of it is stuff you wanna mess with if you don't have a dedicated filtration system, enclosure + washer or separate space for maker stuff like a garage workshop
negative presure? i thought positive pressure was preferred?
positive pressure if you want to keep a room sterile
but I don't want anything in the 3D printer enclosure to get out
so I want negative instead
oh the enclosure

if you're that afraid of it, you probably shouldn't have it because the stress of the possibility that filter might fail will affect your health more than any potential harm should it actually occur.
You have unlocked new role
I just mean my DLP resin printer has a negative pressure system
my FDM printer I'm fine with a filter
but I also almost never print ABS at home
I print ABS at work because we need the temperature resistance, but otherwise ABS isn't much of an advantage
also, I've never had a good experience with a resin printer in a shared space
i mostly print pla and tpu
people get the resin everywhere and never clean up
yeah does it really matter when you live in a city
Sometimes i go out smoke at night and it smells funny because we got factories not so far away from me
i love eating and sleeping in microplastics
I'd say it really depends on the material in regards to whether it matters
but smoking is gonna be a way bigger issue lol
ye
might leave a good fossil. preserved forever for future scientists to study.
Like someone said
"I am microplastic"
if I understand right, the issue with PLA is rarely ever the PLA itself unless accidentally heated to crazy high temperatures
but rather the additives that can be in some PLA filaments
i think so ye
PLA itself produces way less toxic fumes but the nanoparticulate issue remains
The biodegradable label is quite misleading, it only breaks down under very specific conditions
i heard carbon fiber filaments are a scam too because none of the stress gets transferred to the CB cuz its jsut in there and not properly attached with an interface material
the distinct smell when you print PLA comes from additives to make it not break when on the spool (afaict). Coz after it's printed, that smell either isn't there or is really really faint.
it depends on what you actually want the carbon fiber for iirc but
it's not gonna make it particularly strong or anything yeah
ye
Its not an automatic win but if you design around it then it can improve structural integrity quite a lot
Have you guys considered metal
but it won't be like, a carbon fiber composite material
I've considered metal but I'd probably sooner get a CNC machine
Yeah at that point you might as well go for CNC
the rare times I need a metal part that can't be CNCed, I'd 3rd party order it
same goes with stuff like PEEK plastic
its not really stronger in the direction of the fibers because the fiber isnt properly interfacing with pla.
it might be a bit stronger perfendicualr tho
i've made combat robots (3lb class) with aluminum a couple of times but it's too expensive to get machined and can be bent relatively easily for its weight. I mostly use TPU for the body now, with steel armor strips. (but I've also lost almost all of my fights.)
or so ive heard
some of the parts I have to print for work need to be sterilizable under either autoclave heat and/or hydrogen peroxide plasma but also survive being thrown around by the people doing the sterilization
and I still have yet to find a material I'd actually call perfect, at least nothing that doesn't cost a stupid amount we don't have the budget for
almost as in you almost lost all of your fingers but you still have them all, or you lost a decent amount of them but you still have some?
I mostly use PA 12 Nylon now, with the note that it can't be autoclaved
I have so much respect for combat robot builders, used to love Robot Wars
I can't use metal either because of imaging artifacts; and doubly so if the metal is ferromagnetic
I could never put in the time to design and build a robot just for it to potentially be smashed like that, I'd get too attached
Takes nerves of steel
I don't want to be the person they use as an example in every MRI safety course
chris broad was on robot wars as a kid
in 4 competitions i've won like 3 fights out of about 9 or something with 3 different robots. i'm too stupid to give up.
The victor is not victorious if the vanquished does not consider himself so
@sick owl yeah with his dad
This is every single fight from series 4 of Robot Wars and series 2 of Robot Wars Extreme that includes Killertron.
Series 4 -
Heat H:
Killertron vs Maverick vs Suicidal Tendencies 2 (1:40)
Killertron vs Wheely Big Cheese (5:20)
Celebrity Special:
Killertron vs Gemini (11:11)
Extreme I -
Episode 4: (Mayhem)
Killertron vs Aggrobot 3 vs Splinte...


leaf









so then its firefox's fault?
