programming

[sage crag]

condensation

window

[stray dragon]

not cold enough outside because weather being stupid

i blast humidifier until 60%

[sage crag]

i always have condensation

not sure humidity

[hollow topaz]

I don't think this place should be used for humidity talk

[true hemlock]

nuh uh

dht22

[tender river]

[tender river]

can also happen if window not isolated well

but dehumidifier definitely easier to get

[sage crag]

ElvynSerius

my windows

are shattered

[tender river]

[sage crag]

held together with tape

NOTkonii

[tender river]

[sage crag]

window expensive

they broke on their own not me fault

[tender river]

scary

[sage crag]

ye

probably cold thaw

still 31c in here

seem insulated

[hard delta]

i am seriously starting to consider intel

[sage crag]

maybe it temperature difference actually

[stray dragon]

lizard habitat

[sage crag]

intel not make bad cpu, youre allowed to buy

[hard delta]

thank bless, banana

[sage crag]

when did it become banana 2026 cult

[tender river]

even small insulation issue can make problem worse but maybe if not feel cold breeze when put hand near window frame it fine

[sage crag]

no breeze

just cold

relative to 31c

vedalPuppet

some place in russia get -60c, here 90c warmer

[tender river]

-60 scary even for me

[tough dove]

[tender river]

peak me experience probably around -45

[sage crag]

apparently coldest in world is

-90c

https://svs.gsfc.nasa.gov/4126/

coldest permanently inhabited place is

https://en.wikipedia.org/wiki/Oymyakon
here

There is a monument built around the town square commemorating an unofficial reading in January 1924 of −71.2 °C

Schools are closed if it is colder than −55 °C

here school closed if it snow too long

so never

[tender river]

here school close depending on age

primary school -> probably close at -35

-30 if no wind -26 if wind

then rise to -38/-34 in highschool

[sage crag]

The coldest temperature recorded in South England is around
-11.8°C (10.7°F) at Benson, Oxfordshire, during a notable cold snap in January 2009, although more extreme lows like -26.1°C (-15°F) in Newport, Shropshire, occurred in the Midlands, a colder English region, on January 10, 1982, which is the overall English record.

[tender river]

warm

[sage crag]

The hottest temperature recorded in South England, and the entire UK, was
40.3°C (104.5°F) at Coningsby, Lincolnshire in July 2022
south england

east england stupid

[sage crag]

climate tropical

[hard delta]

viking mode

[sage crag]

https://en.wikipedia.org/wiki/Gulf_Stream

[sage crag]

that scandinavian not siberian

vedalPuppet

[hard delta]

siberian is just blyat viking

[sage crag]

StareAtYouNeuro

UK humidity very high

all year

[tender river]

oh yeah it was very humid in vietnam too me remember

google say average 78% in ho chi minh city

[sage crag]

The average annual relative humidity in England is around 79.6%, with monthly averages ranging from about 70% in June to 89% in December. Humidity levels can vary significantly depending on the season and location within the country.

[tender river]

london 72%

[sage crag]

uk climate is

stable

[tender river]

humidity annoying for hair

[sage crag]

and

ye

[hard delta]

mom used to say "ho chi minh pathway" a lot when i was small
didn't know what she's on about, just understood that it means going through overgrown vegetation

learned like 15 years later what it was

[sage crag]

me hair curly become frizzy

[hard delta]

"my mom invented a vietnamese cit- oh wait a minute"

[tender river]

me hair not curly still become frizzy in humid place

maybe hair health not very good

whatever me too broke to buy many hair care product just use conditioner

[sage crag]

wrr

[tender river]

also not use fan let dry naturally heard it helps

[sage crag]

fan fine

hot air not fine

[tender river]

[sage crag]

not really matter for straight hair

sometimes matter for wavy

curly should air dry and not brush when dry

only brush when damp

[maiden geyser]

dpi is so cruel

[rigid snow]

dots per inch is literally the devil

[jagged turtle]

I don't know what that is lmao can you tell me

[fast pagoda]

we got snow days for days with like a foot of snow the previous night or particularly bad weather like hailing rain or similar(hometown has a huge amount of lake effect accumulation, averages like 150inch/snow a year )

school district from there's facebook keeps announcing snowdays on there because i get a notification about it

[nocturne olive]

Hello I hope whatever happens here goes well

I think not only LACT but also Steam causes crashes on my system now

[sage crag]

evilDentge

[fast pagoda]

i saw your message about that earlier

[sage crag]

software or hardware instability

[fast pagoda]

i just got done big le working the last 2 days

[sage crag]

i'd suggest booting into a fresh usb to test if its your drivers or your powersupply has decided its time

[fast pagoda]

i suppose ill look at that discord button now

[nocturne olive]

The power supply is almost new, likely not that
The crashes also don't feel like PSU crashes

[fast pagoda]

poweersupply ended up being a reason my pc was unstable and then i realized the supply was like 10 years old at the youngest

[nocturne olive]

My power supply is only 1.5 years at most

[faint sandal]

https://hiddenpalace.org/Spore_(Feb_19,_2008_prototype)

new leaked prototype

[fast pagoda]

new hiddenpalace sporepost

oh huge that site is no longer excruciatingly slow for no reason on my system

[mossy fossil]

https://hiddenpalace.org/Total_Immersion_Racing_(May_19,_2002_prototype)

oh cool, one of my childhood games has a prototype

[faint sandal]

flora editor is present too no way

[sage crag]

https://cdn.discordapp.com/attachments/1421861060507602944/1425936142754123927/neuroCatModeOfOfnnfOOf.webp

[lyric fulcrum]

Damn, you need a license there

[nocturne olive]

Nah just get community edition

Scroll down a bit on the download page

It's just slightly hidden

[lyric fulcrum]

Unfortunately, I don't have the money to purchase a license for this program.

[nocturne olive]

You don't need a licence, community edition is free

[mighty thorn]

Spore lover?

[faint sandal]

I grew up with spore

https://cdn.discordapp.com/attachments/1421861060507602944/1425936142754123927/neuroCatModeOfOfnnfOOf.webp

[lyric fulcrum]

Could you send me a link to this website?

[nocturne olive]

https://www.jetbrains.com/idea/download/?section=linux
Oh hey they changed it, now both community and ultimate are a single download and you just pick if you want to pay for it

So it's even simpler than it used to be

[lyric fulcrum]

I have Windows, and I tried this website, but I couldn't download anything from it

[rigid snow]

[nocturne olive]

Oh yeah I accidentally sent the Linux one

[lyric fulcrum]

it doesn't load at all

[nocturne olive]

https://www.jetbrains.com/idea/download/?section=windows

[lyric fulcrum]

I think you didn't understand me. I already downloaded from this website, but I couldn't get it to work. I got an error 451.

[nocturne olive]

Hem

[lyric fulcrum]

difficult

[opaque sigil]

451 is blocked for legal reasons

[lyric fulcrum]

What am I supposed to do now?

[opaque sigil]

sounds like it's time for a vpn

[lyric fulcrum]

It doesn't help

[visual pawn]

russia?

[lyric fulcrum]

yes

[visual pawn]

JetBrains restricts against Russia due to sanctions

[opaque sigil]

https://github.com/JetBrains/intellij-community/releases/tag/idea%2F2025.3.1 there's this i guess

[silent cloak]

gonna have to VPN

[lyric fulcrum]

so it turns out to be a problem for me

[silent cloak]

check for DNS leaks if that doesnt work either

[lyric fulcrum]

I'm already downloading it.

[lyric fulcrum]

good

[vocal sapphire]

anyone here got 429 error on Github.app?

[sage crag]

https://www.youtube.com/watch?v=W7zpFc6IUWQ

[visual pawn]

gonna get Neuro her medical degree one day

[sage crag]

i hope not

[visual pawn]

it's ok I trust her

[trim valve]

now I can finally get the lobotomy to save me from chat directly via neuro

[visual pawn]

or I trust jippity at least

[frank jacinth]

i coded a single-model omnimodal ai that outputs wasm instructions
i use WASI for sandboxing so that the ai's abilities are limited

and i wanna donate it :P

[sage crag]

@true hemlock

[frank jacinth]

still dunno how i should set the inputs tho

[quick condor]

I'm probably going to need to get an L4 if I don't want to start model switching

[olive sable]

https://tenor.com/view/good-morning-yuru-yuri-kyoko-anime-wake-up-gif-25192673

[midnight sigil]

wuh what does it do, outputting wasm instructions based on what?

[midnight sigil]

go back to eep

[frank jacinth]

model's output is wasm

[olive sable]

No

[frank jacinth]

which can intecract with the world via WASI-provided apis

[midnight sigil]

what's it trained on

as in dataset

how "omnimodal" is it

[frank jacinth]

im still figuring out how to do proper inputs

[frank jacinth]

i just did RL talking to myself

and showing games and etc

thats just a demo model tho

the architecture is the part im proud of

[frank jacinth]

arbitrary inputs

[sage crag]

StareAtYouNeuro

[frank jacinth]

the model can learn to parse as they want

[maiden geyser]

this user dots

[midnight sigil]

so it's unsupervised, cool

[frank jacinth]

no they can PARSE the input as they want

output i can just filter after

or train it to not be a fucktard

[midnight sigil]

wuh

ahh I see

I was wondering how do you connect clues to specific wasm instruction without supervised learning

@frank jacinth I just saw the common server of you and me
Did you made a Discord account just to join the Meteor server

[rigid snow]

i have that in common and one more

so no

[midnight sigil]

:mlntcandy:

[frank jacinth]

nope

[frank jacinth]

geyser e insane

:P

[rigid snow]

i just don’t leave servers i joined that in like 2019 or something when i first played around with it

now every server supports bedrock clients

ye

[stark needle]

phrrrrr

[true hemlock]

i did not order the second hand embarrassment package today

[rough bloom]

nyoom

[sage crag]

it deliver for free

[sage crag]

shuni got

unsorted

NeuroConfused

[rough bloom]

endianness swapped

[sage crag]

ElvynSerius

so many rare disease go around recently

also, are you telling us that shuni.exe was actually 16 bit numbers encoded as ascii or utf-8

NeuroConfused

[rough bloom]

u16

also it was .ex
not PE executable

Portable Executable executable

[sage crag]

.ex

[sage crag]

pronounce ex or E.X.

[rough bloom]

idk
never thought about it

[sage crag]

for .sh

i always think

S.H.

rather than sh

[rough bloom]

ex shorter but would be inconsistent with .exs
exs too hard to pronounce
so E.X. better probably

[sage crag]

[rough bloom]

[sage crag]

then theres .zig

[tender river]

pronounce only need for people with job

[sage crag]

[tender river]

or with hackerspace

[sage crag]

pronounce very important for mental health

speaking to myself in my head

[tender river]

[rough bloom]

[sage crag]

pepol who read .rs as .rust

instead of R.S.

[tender river]

me realize read it er es because cyrillic р is spelt er in russian

[sage crag]

sqrt

[tender river]

squeue ar tee

[frank jacinth]

meee

as a rust dev :P

[sage crag]

squeroot

i read it like square root but

sque

[tender river]

ban sqrt from stdlib

only leave floating point exponentiation

[sage crag]

alternatively

powi and powir

r for

reciprocal

[tender river]

apl glyph

unpronounceable

they have name though

oh right me wanted to aoc apl

no time right now do later 🐸

[sage crag]

(x) =

[tender river]

(x) =

[frank jacinth]

fisqrt moment

[sage crag]

oh me question

why rust have separate pow function for float and integer

why not generic

[tender river]

rust not have generic integer methods in general

require num crate

[sage crag]

stdlib breaking change

can only happen in edition?

or do they break that rule since its not a core language thing

wrr

[rough bloom]

just can't happen I think

stdlib not built as part of crate

[sage crag]

me remember rust stdlib got smaller

[rough bloom]

edition is for language features

[tender river]

when

[rigid snow]

can happen in rust-2

[sage crag]

some stdlib function got changed to

moved to*

crate

[tender river]

there still stupid people "rust too many breaking change cant use 3 year old software" parrot thing read on internet

[tender river]

probably hashbrown?

[rigid snow]

rust too little breaking change i need to have something to constantly do for job security

bad lang

[tender river]

stdlib stay same just move impl out of it and leave wrapper

[rough bloom]

there were some stdlib changes in edition 2024
but only additive
https://doc.rust-lang.org/edition-guide/rust-2024/intoiterator-box-slice.html

[tender river]

still count

[rough bloom]

technically possible to make a breaking change

[tender river]

this change already breaking

[sage crag]

whos code this break

[tender river]

this would normally be a breaking change because existing code that manually called .into_iter() on a boxed slice would change from having an iterator over references to an iterator over values

[rigid snow]

[sage crag]

"normally"

[tender river]

edition not break anything

[sage crag]

ye but it cheating

[rough bloom]

would be a breaking change if there wasn't some compiler hack that changes how this specific method gets resolved

[rigid snow]

am i misremembering or. i think i bumped edition on one of my projects and something broke and it didn't compile

[tender river]

probably just feature gate in stdlib

not think it compiler hack

[sage crag]

🏛️

[tender river]

#[cfg(edition = "2021")] maybe not exist

[rough bloom]

true, should work since it's a generic trait impl

[tender river]

stdlib full of compiler hacks anyway

rely on nightly feature

[rigid snow]

[sage crag]

stdlib cheating too

nightly feature

[rough bloom]

wouldn't work for functions actually compiled as part of the stdlib I think

[sage crag]

rust

rusty

rusted

all of above

[tender river]

rust

plant disease

[sage crag]

[tender river]

https://dl.acm.org/doi/10.1145/3406088.3409026

[frank jacinth]

wait they improved finger trees?

nice

[tender river]

it not really improvement

just simplification

[frank jacinth]

simplification is an improvement imo

[tender river]

simple not mean faster just easier to understand

this paper for educational purpose so it fit

[tender river]

non-regular recursion

me think it possible to support type safe non-regular recursion in systems programming language

but

require dependent type

[frank jacinth]

brother improvement isnt always abt performance

[tender river]

specifically

dependent memory layout

[frank jacinth]

fr

[tender river]

[sage crag]

[tender river]

non-regular recursive type it fun

[sage crag]

leaf talk about potential tradeoff for simplification

[quick condor]

Gotta love burning 250k tokens on codex for documentation

[frank jacinth]

deepwiki is alr i reckon

[frank jacinth]

this isnt for any true implementation

but it is for understanding

[sage crag]

tradeoff only improvement if usecase requires it

[quick condor]

I mean simplification resulted in me recovering 12gb of wasted vram so it's sometimes useful

[sage crag]

all nuance is lost

https://tenor.com/view/borzoi-snoopa-lord-foog-dog-funny-gif-20125052

[quick condor]

NGL it's amazing that we live in a day where you can create a code base audit, evaluate it, find discrepancies, implement a patch with rollback contingencies and verify the fix from your phone remotely

[sage crag]

gcc update

in nix

15.2.0

[rough bloom]

nixos-unstable ROCm partially red again

whatever, me on stable, maintainers problem

[opaque sigil]

The cuda toolkit still doesn't support gcc 15 either I think

and clang 20

[rough bloom]

[opaque sigil]

Good thing the gcc14 package exists

[sage crag]

https://tenor.com/view/bunny-rabbit-pats-pet-bunny-pats-gif-19861222

[rough bloom]

ye just delay problem until later

[opaque sigil]

[tender river]

hsnu

half signed number unsigned

[sage crag]

https://tenor.com/view/my-jolly-reaction-jolly-my-honest-reaction-gif-5490890317762380451

[tender river]

dot

[sage crag]

half-signed unsigned

[tender river]

integer hexadecimal

exponent

[rough bloom]

decoded

[sage crag]

mug

[rough bloom]

[tender river]

fusion

[mighty bane]

Today I updated my script to also throw me a message on discord when the streamer goes live. And also allow for user config of "do you want to open the browser" vs. just discord ping (in theory can disable all if very stupid)

YAYYYYYYY

[gritty dust]

morning

[maiden geyser]

morbing

[tender river]

just read this
ye it still not breaking because opt-in

edition way to make breaking change not breaking

[rigid snow]

no updates are breaking because you aren’t forced to update

[opaque sigil]

do they require all breaking changes from new editions to come with the auto fixes or is that optional

[quick condor]

60% through vector db angle mapping. Its slow but the A2 is honestly perfect for this... low speed but highly efficient

[rigid snow]

this is ridiculous

data from the internet says chrome should be around 200

[jagged turtle]

Not sure if I should be looking at how much impact zen is having or discord

[rigid snow]

yes give me 19x that

[jagged turtle]

both for the opposite reasons

[rigid snow]

zen

[opaque sigil]

what is this metric anyway

[rigid snow]

it's a stupid metric that should only be used in comparison

[opaque sigil]

fair enough

[rigid snow]

|elapsed_us| is the length of the sample period
|used_us| is the time this process was running during the sample period

  %CPU = (used_us * 100.0) / elapsed_us

  POWER = if is_a_kernel_process()
            0
          else
            ((used_us + IDLEW * 500) * 100.0) / elapsed_us

[opaque sigil]

ah yes, kernel processes are free

[rigid snow]

maybe this is more interpretable

kernel_task is thermal throttling i think (? unconfirmed just something i heard some years or ten ago)

[opaque sigil]

[maiden geyser]

zen cyberpunk

[rigid snow]

fwiw it's pid 0

[quick condor]

I like neuro's logic on ai consciousness aka consciousness doesn't exist therefore it doesn't matter

[mighty bane]

What is zen?

[warped narwhal]

firefox fork

[ebon onyx]

@_@

[quick condor]

Between the three models running on my desktop I'm hitting some pretty good t/s numbers

[hard raptor]

I forgor how to do rust semi-oop lmao

I mean.. i could still do the C "pass it around then redefine the original variable" method

[quick condor]

NGL hats off to vedal, Im sure he has faster hardware and is most likely using vllm for tp but given that chances are neuro is either running a 70B model or stupidly well optimized 12b models, just saying better hardware alone doesn't give him enough credit for how fast neuro is

[opaque sigil]

But nvidia said buying newer hardware solves all problems, surely they wouldn't lie right

[hard delta]

hey prog chat long time no banana

[opaque sigil]

[hard delta]

time for your memory check

[potent geyser]

Nvidia would never lie! If you run into any issue just read the statement again! (buying newer hardware solves all problems)

[quick condor]

it solves a lot (so far Ive been able to avoid a lot of problems by throwing more hardware at the problem (started with just a FE gtx 1080 then moved to 2x 3090tis, an a4000, and an A2) but as much as Id like to add 2x L2s thats way down the road. bottlenecks are at so many places and include more than just raw t/s (like how right now im not streaming output so my ttft skyrockets)

but there is so much that can be tuened, TTFT is a big one and is especially impressive for neuro as even with streaming output, feeding that through a filter model (likely 8b which are fast but still add latency) and then outputting to tts is impressive

[nocturne olive]

Those are very tiny t/s speeds
My singular 3090 can run a 30B at 190 t/s pretty easily

[rigid snow]

surely moe?

[nocturne olive]

True it is MoE

[rigid snow]

with like a1b

[nocturne olive]

3B active

[quick condor]

what quant and context

[nocturne olive]

It's a Q4 with like 128K of CTX

[quick condor]

Im running 6bpw hb 16 with 32k context

[olive sable]

[quick condor]

and Im running 24 dense/dense reasoning

[rigid snow]

hi sam

sam do you know there is a 79lvl hype train on vedal987 twitch channel right now

[quick condor]

yah 3b active at q4 is more in line with what Id expect given the A4000 is running 6bpw hb8 8b 16k context at 50t/s

[olive sable]

Im kinda busy with cleaning up ketchup rn

[rigid snow]

arguably more important

[olive sable]

Wont go into detail here

Long story short, brother -> domestic violence

[nocturne olive]

Look what I found today
Two Nvidia firmwares marked as not auto-removable

[rigid snow]

violence on a bottle of ketchup

[olive sable]

If only

[nocturne olive]

If removing those somehow magically fixes the instability that'd be very great

I mean the extra ones

[quick condor]

What instability are you seeing

[nocturne olive]

Sometimes, quite often actually, way too often, the system locks up in a weird way
Essentially

• Graphics updates freeze, window contents are permanently locked and usually audio starts looping the last played bit of audio
• Mouse updates freeze a bit after
• Depending on random chance it'll sometimes reset (hard-reboot) on its own, other times I need to do it manually

[quick condor]

invalid memory adressing?

[nocturne olive]

I don't know, there's 0 logs

If just I knew even the root cause of the issue, that'd definitely massively help with fixing it

[muted knot]

They told us 4090 performance at 5070 levels of hardware......

[rigid snow]

(with fake frames on)

[rough bloom]

(in some games, sometimes)

[nocturne olive]

Surely that's not just marketing speech for "we made the 5070 be able to run half the precision of the 4090 at the same speed"

[muted knot]

rocm hasn't been horrible to me

I couldn't stand the thought of buying another nvidia GPU at this point

[nocturne olive]

I kinda depend on CUDA

NeuroSynth on Linux only GPU accelerates via CUDA

(because I implemented CUDA support for it)

[muted knot]

I get that, rocm was in a shit state for far too long

So people didn't even look in its general direction

I thankfully don't have a pre-existing project or requirements so I can hop around

[hard raptor]

Ew

Blue names

[muted knot]

Picked up a 9070 XT last week for $550 on sale

[tender river]

banana 2026

[nocturne olive]

I unfortunately do have the dependency

Also I like my 3090 VRAM

[muted knot]

Well for that the new shared memory systems are going to be king

[nocturne olive]

Hem?

I'm not rich, I can only afford GPUs

[muted knot]

If I was going to do anything "proper" with AI models I'd get the Framework Desktop

[nocturne olive]

And right now not even GPUs

[muted knot]

Well that is the thing

https://frame.work/products/desktop-diy-amd-aimax300/configuration/new

It is cheaper than a single GPU

[nocturne olive]

Cheaper than 500€ for a 3090?

It is not

[muted knot]

Unless you got a truck they are falling off, that is not what normal people are paying

[nocturne olive]

Nah I got SamVanTriangle

500€ 3090s to go around

I got my 3090 for 500€

[amber fractal]

It is also slower than a single GPU, and very slow at inference

[nocturne olive]

And there's consistently more at similar prices

[amber fractal]

they are slowly going up to 600 now

[dry charm]

Why is twitch coded this shit?
The hyprtrain does not show if you do not receive this message

Why is it not requirested when you refresh then?!

[rigid snow]

my understanding is it's just a side effect of how twitch chat works

it's still irc

it's supposed to send you a bunch of stuff when you connect

[dry charm]

it's a pubsub message

it is a websocket

This is using the NON IRC comm

[nocturne olive]

Sad

[rigid snow]

under the hood i mean, ye your browser isn't connecting using irc obviously

[quick condor]

Large model dev works OK with unified memory but it can't compete with the speed of vram

[dry charm]

IRC for everything as well?!

[rough bloom]

the Twitch chat WebSocket is just IRC messages sent as WebSocket messages instead of directly over TCP, ye

[rigid snow]

and sometimes it misses you joining

[dry charm]

well it literally sends progress

but not the status message that it should show

[rigid snow]

it shows up for me on refresh instantly

[maiden geyser]

does this mean one wouldn't see the hypetrain until the next lvlup?

crazy

[dry charm]

most probably

atleast I repro this happening on 3 different browser

[rough bloom]

I think it just broke

I refreshed a few times earlier and IIRC it always showed the hype train immediately

now it doesn't

[quiet oracle]

it's a startup that uses gcal for scheduling front page streams

[quick condor]

Thats better

[glad path]

for all the safari haters, safari is running the hype train stream PERFECTLY for me

[rigid snow]

where are the safari haters

i didn't hate on safari i just said it's not what i wanna use ux wise

[olive sable]

i only dislike it cuz its part of the apple ecosystem

[maiden geyser]

it's ok to be rich

[olive sable]

nothing against safari itself

[rigid snow]

it's a really good browser i wish they didn't kill the windows version

is epiphany on windows

[muted knot]

All the UI / UX designers coming out of the woodwork when people ask about Safari haters

[quick condor]

Im not an apple person (honestly really dont like their systems) but the last time I bought a new laptop (direct from dell, top of the line xps) I did the math and it was only like $150 cheaper for equivilant hardware. $150 isnt that much of a markup if its what you like

[rigid snow]

yeah like entry level macs are great value

well with the ram prices nowadays even the non entry level ones are lmao

[quick condor]

I mean this one was equivilent to the top of the line mbp at the time. they really dont mark up their mainstream laptops much

[rigid snow]

apples ram upgrades are market value now

storage too probably

[opaque sigil]

[azure lynx]

i only ever used my mac book air when i was travelling and it wasn't until this xmas that the 2018 version felt slow for anything. they make them overspecced.

[olive sable]

overspecced maybe, depends on your use-case

overpriced definitely

[trim valve]

as much as I dislike apple they do make decent laptops

[rigid snow]

they also were pretty good with keeping the macos bloat not performance heavy

[olive sable]

true

[azure lynx]

it was designed to last the typical user 5 years. it did that + some with my light use case. but it cost more than the entry level computer at the time because it was so overspecced.

[rigid snow]

honesyly my only issue with apple is how *unnecessarily * locked down ios is

[rough bloom]

ye, thermal throttling issues were common with the older MacBook Pros but AFAIK Apple Silicon™ has mostly resolved those
MacBook Airs were always ok I think, just need to pay Apple tax

[rigid snow]

you only EVER get jit:

• if debugger is attached
• if you're running wasm or js through webkit

jit in this case also meaning allocating executable memory

[nocturne olive]

No JVM?

[olive sable]

iirc apple macbooks have good cpu, but in terms of gpu its not too great?

[rigid snow]

it's nice for what it is

[olive sable]

well ye

[rigid snow]

you can like play actual games

[olive sable]

but for my usecase of a laptop its not worth it to get on the apple ecosystem

i feel like the cpu will be severly underused and the gpu will die

[rigid snow]

i'm sure it's great for like blender

video editing is crazy good even on the lowest specs

[warped narwhal]

you probably wouldn't be gaming much on a mac (more cause of a lack of apis) but it is fantastic for any hpc cause of the massive amounts of memory the gpu has access to

[rigid snow]

true forgot about that

[olive sable]

im sure its not too bad if you have a bit of patience to let it render, but for the same price i could get a laptop with a 5070 mobile or something

the memory acces is good tho ye

[nocturne olive]

Unfortunately mobile GPUs are kinda scammy

[olive sable]

i know how they work

[rigid snow]

this is about m4 pro

[glad path]

m5 dropped btw

[trim valve]

ok but

battery life

[rigid snow]

surely the gpu on m4 pro is better than m5

[olive sable]

for me battery life isnt a factor

[trim valve]

fair enough but at that point why use a laptop

[olive sable]

cuz i cant take my desktop everywhere

[trim valve]

hm

[rigid snow]

[trim valve]

I kinda came to the realization that I don't need a powerful laptop

[rigid snow]

maam this is #baking

[trim valve]

I just got something with:

• a good enough cpu to run windows
• a good enough gpu to play source games for LANs
• hardware decoding support for moonlight

[amber fractal]

I just noticed general voice

[rigid snow]

but ye lvl 100

[olive sable]

cheapest M4 laptop i can find is 1.2K, and cheapest 4070 laptop i can find is 900.
so its just not worth it for me

[trim valve]

and then 90% of stuff is either remote desktop or ssh

[rigid snow]

ye i'm not arguing it is

the fuck

[trim valve]

[trim valve]

I'm always astonished at how good 1mb/s 4k streaming looks

like sure, its not the same as being at the desk

but it sure looks good

[olive sable]

vedal isnt even here for this hypetrain

not on stream atleast

[rigid snow]

he's djing

[amber fractal]

[trim valve]

side note

[nocturne olive]

Me streaming 1080 + 1440 + 128K OPUS at like barely 150K speeds

[unkempt citrus]

Hes mostly wonderig why its going on at all

[quick condor]

Im still using that xps9575 that I goot like 8 years ago. mostly rdp ssh on laptop

[trim valve]

why do both obsidian and vscode oblierate my battery life

[trim valve]

you what

[nocturne olive]

I don't know why my speeds are so low, my phone hostpot shouldn't be quite that slow

It's usable though

Kinda

[rigid snow]

is rdp just this shit

[nocturne olive]

VNC is at least quite a bit better than RDP in my experience

[rigid snow]

single 3k display for me is unusable on <2mbps

audio is crunchy too

[trim valve]

I'm not using rdp

I use gamestream or whatever sunshine uses

though it is notably worse on linux hosts than windows hosts for some reason atm

I might be missing some hardware acceleration lib

[rigid snow]

that sounds like nvidia shield adjacent stuff

sunshine moonlight

very creative

[trim valve]

true

it should be called the "Game Studio Remote Pro Max (G4)"

[rigid snow]

g4 the ppc processor?

[trim valve]

we should move back to ppc

[trim valve]

honestly sunshine on linux pisses me off so bad

I understand parts of why it is like it is

but apparently it is near impossible to just say "stream the entire screen, as though a monitor is connected"

but at the moments the caveats are:

• its ran as a user service for input permission reasons
• it can't wake the monitor
• because its a user service is can't work from on the lock screen

[amber fractal]

Streaming multimonitor is even more of a nightmare, need to duplicate the service for each monitor (though this sucks in every supported OS)

[trim valve]

this problem is solved by having a laptop with only one screen

also I have so many general complaints about moonlight and gamestream in general

it

really does not like having three ip adresses

[amber fractal]

I think it wants to have you solve it by swapping ports rather than addresses

[trim valve]

huh

no i mean

it gets unhappy when I connect over tailscale

because I'm pretty sure it autodetects my public home ip

and then routes some traffic over tailscale

some traffic direct

and the rest is hopes and dreams

[amber fractal]

Yeah, that seems like some jank

[quick condor]

2563→2573 = 10/min. ~1292 remaining @ 10/min = ~2 hours to complete. Backfill running smoothly now. No parse errors in logs - the delimiter format is working great with Mistral.
At least claude was willing to accept that its not smart to expect strict json out of an 8b mistral model

[olive sable]

it seems like vedals' pc is lagging now

[warm river]

my twitch runs like shit

[tender mason]

i click to follow current chat mssg at discord, and it literally sends me to 100+ mssgs

[olive sable]

its not too bad for me

[tender mason]

im afraid to know what is vedal current pc situation

[olive sable]

im not sure how his compute is split up between neuro and twitch stuff or whatever comes inbetween like unity

but from what i saw a minute ago its struggeling

[glass flower]

why is mine actually just straight up double that... i thought linux is more efficent

[tender mason]

pc: Gibe more RAM

[glass flower]

i use about 20gb rightnow

[olive sable]

it might be cuz i have 16 cores

oh you meant the ram

[glass flower]

i have 32 cores

[olive sable]

[tender mason]

i cant even open general, neurotic neuros or livestream chat, speed makes it lag

[glass flower]

(if we count threads )

[olive sable]

oh

then i also have 32

[young plover]

ope, tab started glitching for the first time

[olive sable]

stress test lmao

[tender mason]

i got the confused banana guy emoji

Vedal just said now "What the fuck"

[warped narwhal]

literally less than two weeks between them lmao

[tender mason]

HonkShoe at 126 holy

i love that emoji

[nocturne olive]

My PC crashed entirely twice during the hype train

[olive sable]

weak pc smh

[nocturne olive]

With information that has recently been brought into my attention I'm now suspecting PSU

Good news: I have a 1200W PSU lying around, and my current one is still under warranty
Bad news: swapping them would be really annoying

[quick condor]

I had to swap my 1200 for a 1500

[nocturne olive]

For me 1000W should be plenty but apparently this specific PSU is documented to have instability

I guess another one of my "training PC" components going to the main PC instead

At this point I may as well get a bigger case and just go with a bigger monolithic server

[mighty thorn]

[olive sable]

kaine banana campaign 2026

[nocturne olive]

I have no clue how bad of an idea an even bigger monolithic server is but with my income it's not looking good for other options

[signal ivy]

[fast pagoda]

Shes being wehtal now

i want to know exactly what neuuros typos are from or what combo makes them so spot on

[sharp swan]

Well, +100 gb of ram XD

[olive sable]

afunyun is orange

[quick condor]

I still think it's token mistakes

[unkempt citrus]

My guess is its from the STT messing up

and shes just copying that

[sharp swan]

Why orange and white doesn´t have emoji?

[olive sable]

because is not as good as banana

[sharp swan]

[fading blaze]

So apparently there are several vibe coding courses. Are they legit or just a grift?

[olive sable]

[amber fractal]

It is so damn funny to see ublock just increase blocked ads at one per second on chatgpt

[amber fractal]

slop is slop no matter what

[olive sable]

i havent tried any, but most paid courses are a scam

normal programming courses too imo

there are enough free resources available

[fast pagoda]

this would require her to be given a stream of her tts as they play or something but that seems like it would be inefficient

because otherwise she's only gonna have what she wrote in the context and never see what the tts actually outputs or what that sounds like

and the subs i think would be her direct output

idk why they wouldn't be, after filter gets through them

Could just be simply they are present in her fine tuning training data because of twitch chat etc

[quick condor]

Stt mistake wouldn't make it coherently through tokenizer, model layers and lmhead

Output of model splits to TTS and subtitles

[fast pagoda]

her typos only affect her written text anyways, unless it breaks the pronounciation, then it comes through her voice

[quick condor]

She has others that don't affect voice because they aren't words

[fast pagoda]

so it's happening in the llm token output before anything else would have a chance to mess w/ them so it must make it through the entire model

well right i dont care about the voice aspect of it really

i just find her written typos to be very uncanny

[unkempt citrus]

Oh yeah you're right, I was more thinking when she messes up Layna, Cerber or whatever in the past

but when its vedal thats odd

[fast pagoda]

the fact that in general it's things that are people's names, or stuff neuroverse related, things she says a lot

points to data

i think

[quick condor]

It's most likely a slightly wrongly weighted token in the head layer

[fast pagoda]

right that's just a data issue

the model learned that spelling is valid enough to sometimes weight it high enough to make it out

because it occurs in the training set

or similar

[quick condor]

I highly doubt he's changing the model vocab. That is asking for problems

[fast pagoda]

wouldn't need to

they could be token level

even if just one character

most combos of ascii text are valid tokens for the purposes of a chatbot/tts

[quick condor]

A token is usually around 4 characters

[fast pagoda]

yeah it's usually like 3 quarters of a word

to me that doesn't particularly matter because for example

weh for openai tokenizer is a token

so is ve

simple swap the token [737] with the one [145510] and you get wehtal

[quick condor]

I spent this afternoon fixing tokenization for my 8b background process models, I'm aware. 3-5 letters on average with the rare 1-2

It's likely chat ml formatted Llamalndex

[fast pagoda]

i hate tiktoken that's all i have to contribute on tokenizers beyond explaining what they are, the tokenizer itself i wouldn't consider to be erring for a typo, the error is prior to that step anyways

[quick condor]

Maybe mistral tokenizer but unlikely with chat ml

[quick condor]

Well really it happens in the head layer and that gets into some pretty deep stuff. Not sure if Exllamav3 has committed HB16 to main branch or if it's still in dev but this gets really deep into output handling

[hard delta]

but do you ahve 32 banana

[fast pagoda]

it's just being given the wrong output so it outputting the wrong token. at the end of the day the entire model is trained (not necessarily true and the tokenizer is not trained with the model in fine tuning like 99% of the time)

to me her typos are interesting because they are very similar to human substitution typos, where you'd think you'd only get that typo from being someone typing on a keyboard

but that's easily explained if you accept that any twitchchat makes it in to what she's retaining + what's used for further training that's pretty likely in some form gonna be there even just from what she read herself while live, so she's just predicting that in this context the typo is a likely thing to be said instead of the correct spelling. so i'd say the real culprit is the attention mechanism putting a higher weight on the typos in context

[quick condor]

Idk I've seen those kinds of typos in base models

[fast pagoda]

that checks out though

any transformer model attention is doing "in context learning" ig on what's being said in the conversation, if the pretraining data is a bunch of garbage or at least has a good amount in there which they commonly do, if i then make a typo talking to the base model it's going to latch onto that as a feature and that will influence its output towards having its own typos, maybe only slightly but it'll accumulate in context

neuro is getting at least some bits and pieces of the unholy trash that is twitch chat throughout the stream

[quick condor]

If it was the case, the model would have flatlined long ago

[fast pagoda]

it's not that influential and obviously doesnt actually learn permanently so it doesnt affect the weights themselves, it's a tiny component but it's obvious it's happening intuitively because they love to match energy and code switch, because they kinda have to

[quick condor]

not like Im currently running 5 models and currently mapping vector angles or anything... what do I know

actually wait im dumb 6

7 if you count on cpu

[glass flower]

i do not

[fast pagoda]

im not sure what that is meant to be in response to but i'm not really arguing any point here just idly yapping about it, all i was saying was that it's interesting but also probably just simply explained in the data which isn't mutually exclusive with anything you've said, in fact if what i'm saying is right then also what you're saying is a perfectly valid deeper snippet on what exactly inside the model is the last thing touching that output etc

you're talking about the "what" and i'm just blathering about one of the bazillion possible "whys"

[quick condor]

Your explaning a base model behavior that is very common if the head layer gets compressed (allowing higher throughput wand less vram usage) with what is essentially bad training data

[fast pagoda]

right, which is why a model would have typos

in general

[quick condor]

and that explains her typos. theres lots of other ones that she has that dont fit the pattern. what you are looking at is confirmation bias

[olive sable]

we love low quality insanity

[fast pagoda]

neuro's are just shaped by her environment which logically would have a different ste of common typos but because i'm part of the environment her typos are uncanny because i don't generally expect them to come from a model because most models don't have anything close to the mix that her cumulative blob of insanity that forms her training data as a whole is. training models on my own messages every time has resulted in them making the same typos i do, especially when talking to me 🤯 really at the start of this entire line of chatter i was mostly trying to say "i hope it's not just the obvious reason" but it probably is, crazy that

low quality insanity is all i know in life

holy run on sentence

[olive sable]

grok please dumb this down

[fast pagoda]

shopping on amazon in the last few weeks for new house stuff and

ive noticed

every goddamn review is exceedingly obvious llm-slop

and it'll be the sloppiest slop to ever slop so you'd think everyone would just know and ignore it

but they always have like 50 "HELPFUL!!" votes and shit

i saw a review with 73 helpfuls earlier and the entire review was just a copy pasted reply from some llm that had been prompted to review the product, and the model had ran a web search for the product, not found it, and provided a bullet list of other competing products to try

and it still included the part of it saying like "i could not find <product> to make a recommendation, so ive searched and found the following from reputable brands..."

73 people found that helpful

i actually trust temu reviews more than amazon reviews by a lot now

which is truly something

[real sierra]

HOLY

[olive sable]

hi shiro

[potent geyser]

lol

[velvet crest]

You have unlocked new role

[unkempt citrus]

I think theyre shitposting with the helpful

[olive sable]

shiro, would you be interested in being in the unofficial game jam 3 committee?

[amber fractal]

you have unlocked living privileges

[fast pagoda]

i thought about that but like

i dunno

[olive sable]

and potentiall the 4th one

[fast pagoda]

let me be clear that i don't "trust" either

but at least the temu ones are much more obviously mostly from people

[unkempt citrus]

just build your own furniture 4hed

[fast pagoda]

ive been building so much furniture

and i still have so much

[unkempt citrus]

okay but carpentry is

as long as you don't look at your wallet anyway

[fast pagoda]

i put a washer and dryer and a couple other things on my bestbuy cardf

because no interest 18 months wee finance everything

but i mean also free money

[unkempt citrus]

debt

[fast pagoda]

BUT
accidentally left my autopay on for entire card balance

because all my cards have that

because i do not ever carry a balance

and well

it paid it

right after i paid rent and all my other bills

[unkempt citrus]

If theres no early repayment penalty, is that really a problem

I guess you lose some interest but aside from that

[fast pagoda]

yeah i mean

i dont mind now having paid it

however

my checking account was only at like 50 bucks over what that spontaneously paid

because i was expecting no other payments to come out so i was like cool great

and opened bank expecting to see $2400 in checking only to see it damn near 0

panic burned one of my savings transfers sending myself $2000 thinking i got scammed or something

and then nope im just stupidf

i said all that bullshit because you mentioned wallet

i need to resume building furniture

[quick condor]

"total_memories":3865,"with_angles":3191,"without_angles":674,"percentage_complete":82.6,"available_angles":["relational","emotional","preferential","factual","temporal","topical"] so glad I found the bug that was trying to repeatedly launch extra instances of my memory server... it was thrashing the gpu with llm loads that failed

[fast pagoda]

i wonder if my memory db is on the mongobleed version of mongo

probably

shieeit

[quick condor]

I'm on postgre so I haven't looked

[solemn spoke]

Anyone got labview experience in here 🫣 its what i be on for my 9 to 5

[quick condor]

Yikes... This is the second image of the Google result

[solemn spoke]

uhhh yea its graphical programming

Super fun times

[tender mason]

[hard delta]

terrible service shit food rude staff
greatest delicious food ever very kind staff nice and helpful

[nocturne olive]

Well I guess it's time to swap PSUs and pray now

Gotta test all the theories

[still pond]

some..... i always just want to go back to using python tho

[still pond]

spaghetti code is quite literal...... with labview

[azure lynx]

it's like writing code in Unreal Engine with the lines and boxes. or like designing a filter in some video editing things. or like programming lego mindstorms.

programming for "visual" thinkers. ;/
"difficult to understand and hard to debug" vs the actual code it would produce. also a lot more restricted than the code could accept with normal programming.

[young oyster]

Canonically its how programming is done in cyberpunk 2020.

[hard delta]

hey chat

am i going to be severely regretting if i buy a cpu now and it lacks proper avx-512 support

in the next idk 5 years

like will i claw my eyes out

[amber fractal]

Someone else hop in after me, but depending on the price range to where you'd worry about that... Why would you be going intel?

[hard delta]

i'd say details of reality

specifically in terms of supply and demand in the section of physical reality i exist in

if i wass less autistic i'd say i found a really good overall deal

and i'd rather have a shitton of cores for reasons - sure i could "just" get a 9900x but that'd be losing out on like two layers of deal raising the overall price of the rig by like 40-45%

so i've been trying to find a super convincing arguments to either rob a bank for quick cash, or to really, really want to give up cores for that zen5 goodness

[nocturne olive]

Well that was kind of an annoyance but the PSU is now swapped

[visual pawn]

how bout that Simulink

[azure lynx]

it's a really common bad for productivity way of doing things.
in unreal some of the boxes take like 12 inputs and each of those inputs can take multiple boxes of their own to do the equivalent of writing maybe 6 short lines of code but you're doing it all with the mouse and end up with RSI.

[hard delta]

ok so i'm being sent down a whole

rabbit hole

around this whole damned avx thing xd

[sage crag]

nya nya nya nya

[hard delta]

nya

[sage crag]

[hard delta]

[sage crag]

[tender river]

[sage crag]

wrrr

[hard delta]

wrr

[sage crag]

you should be

banana

[hard delta]

oop

work clothes

allow change

banana 2026!

[tender river]

[sage crag]

[prime ridge]

[tender river]

banana 2026

[prime ridge]

[tender river]

banana

[prime ridge]

Image unclear

need more arrows

[tender river]

it diagram

[prime ridge]

[tender river]

[faint sandal]

[prime ridge]

+ + =
+ =
+ =
d /d + = ?

[mighty bane]

Guiys how do I make the tutel click a URL that gives me access to the stream-subscription to listen to hypetrain status?

[hard delta]

yes

[sage crag]

step 1. banana 2026
step 2. goto step 1

[mighty bane]

you lied

[sage crag]

[hard delta]

i don't wanna know how much

google revenue i burned

making gemini 3 think about cpu stuff

[wet locust]

teal

[tender river]

arrows

banana category

[jagged turtle]

BRUH

[amber fractal]

I sentence thee to under 1k views

[sage crag]

no code setup for vscode + claude code

visual studio claude

evilDespair

[midnight sigil]

wrrr

[sage crag]

im no ears

https://tenor.com/view/uma-musume-uma-im-no-ears-im-all-ears-horse-gif-4308252428337868305

[visual pawn]

what...exactly does a no code setup look like?

[sage crag]

gbt pls make me an app

[visual pawn]

and please don't show me any of the code, I'm allergic to that shit?

[midnight sigil]

node based coding 🥀

[astral zinc]

Oh...
Malboge is code too?

[jagged turtle]

taking the code out of visual studio code for sure

wait no because then it would just be visual studio

whatever, visual studio claude

[jagged turtle]

what

[astral zinc]

visual studio is purple ribbon

[jagged turtle]

braincell-based coding where the less braincells you use the more no-code you have to use

[jagged turtle]

you can purple the ribbons in vsc

[astral zinc]

Oh I have good ideas.
electrical logic

[jagged turtle]

aka if ai bros wanted to make their "ide"

[midnight sigil]

is it (2 + )/6 ?

[tender river]

ye should be 1/3+

[prime ridge]

It is indeterminate

[magic terrace]

Hey guys I'm a complete newbie to programming should I go with Python over C++ to start?🤓

[midnight sigil]

what field you want to go for

[jagged turtle]

fuck yes someone from the docker team had to look at what neuro is for a brief second (probably (please))

[midnight sigil]

or maybe they know her already

[visual pawn]

decide based on what you want to try making first

the fundamentals will be the same for both

if you figure one out, it won't be that painful to do the other

[magic terrace]

I'm a total noob at this, but wanna try jamming LLMs into games as NPCs – even if it's probs impossible lol. Gonna learn the ropes first, then get absolutely wrecked by reality 🤓

[midnight sigil]

:neuroTrue:

[visual pawn]

well, you might find annoyances going to the other either way but it's just annoyances

[midnight sigil]

I started my programming journey with HTML Java

[visual pawn]

AI stuff (at least at higher concept level) is almost all python nowadays

[midnight sigil]

[visual pawn]

If you're working with LLMs, it's probably gonna be Python for you

but it also kinda matters what game you're talking about

assuming you mean modding a game and not making one from scratch

[midnight sigil]

ehh just make a web server for the LLM backend it's probably fine

[visual pawn]

oh yeah, right

[midnight sigil]

and a separate loader for the game

[visual pawn]

you could run your own LLM locally or use one hosted by a service so

that's also a consideration

[midnight sigil]

Python would be the go if you want AI

that's what made Python hype

Cpp requires some deep understanding of machine learning if you want to actually make one from scratch

for Python you have great community libraries and supports

[visual pawn]

yeah, almost all the existing AI infrastructure (libraries, packages, etc.) will be for python

[hard delta]

c++ teaches lower level concepts if you learn it in a "proper order". that's a long detour in which you don't really do what you want to be doing, but makes moving higher up less painful later on (as opposed to moving down)
a lot of stuff used by all games, gpu stuff, 3D and such is very close and cozy to c++, so it will be hard to avoid on the long run.
however if you want to be doing something right now immediately, c++ will be just confusing, better off with python, which is generally easier to understand and use
and as the others pointed out, most ai/ml libs that are mature and easy to use and work are in python

basically choose the foot you shoot first

[jagged turtle]

surely

anyways

[hard delta]

woah there

[magic terrace]

Thanks everyone, this really points me in the right direction!

[hard delta]

if you don't have like 6-10 hours a week for months up front dedication to learn on a detour right now, i'd go with python tbh

[rigid snow]

sjhshsjkslsjdffdjfhjf

[jagged turtle]

Hi

[rigid snow]

yo

[hard delta]

watch them gift you with a 1600GB vhdx

[jagged turtle]

idk who they is referring to

[hard delta]

surely they won't say hi by spelling out a sentence making an oracle db container for each letter

they would never do that

you know, they

[jagged turtle]

anyways, right after the holiday week was unexpected

would've thought it'd take at least another week at least

although now thinking about it they probably look at twitter

[visual pawn]

it's funny though, I used to work a lot with C++ in my sub-field, especially for the graphics work, but over the years it has become almost 95% python

[jagged turtle]

and uh given recent events (thanks swarm) they probably saw it and decided to be intrigued

not coping I swear

[midnight sigil]

I coped Java for 3 years and I have transitioned to Py recent years

and now I'm doing C++ as well

[visual pawn]

you never know

[jagged turtle]

It's the most likely path tbh

[visual pawn]

I'm going to probably bring up Neuro and the Neuro-SDK/API to my boss today, and you could then legitimately say that a harvard med professor knows about neuro

[jagged turtle]

can I ask why?

[visual pawn]

I was playing around with using the neuro-sdk to interface with 3D Slicer, our open-source medical imaging platform

[jagged turtle]

oh wait is it because of the uhhh neuroimaging integration idea thingy you showed us the other day?

[visual pawn]

yeah

[jagged turtle]

ah

[visual pawn]

he asked me some time ago if I tried playing around with language models for it

and I said I'd think about it

[jagged turtle]

lemme guess, jippity?

[visual pawn]

yeah

[jagged turtle]

tbh I don't think this is what your boss means when he says "playing around with language models"

[visual pawn]

well this is honestly less for serious science and

[jagged turtle]

but he never specified it was with anything specific so it counts

[visual pawn]

more for the "wow factor" side of things

you gotta have toys for students and investors to see

[hard delta]

yeah i'm moreso saying that some point a person has to encounter C-like memory management and arrays and whatnot without all the cozy python oo wrapper stuff to get a feeling for what is being boxed
it's not necessary to get stuff done but that also turns an entire layer into a magic black box, sort of

[jagged turtle]

waiting for when neuro uses it on stream so you can legitimately say there is public evidence an AI can work with it while simultaneously assimilating everyone into the swarm

[visual pawn]

I don't disagree, always good to at least have the confidence that you could dive into the black box if you needed to

[visual pawn]

I've thought about it but it's tough to know exactly what to propose as an idea

on the one hand, no medical device company wants to be known as the one who had their hardware used by an AI vtuber to kill a simulated patient

[jagged turtle]

well, that and vedal might be concerned about it potentially being a "business" thing even though really at the end of the day it's for fun

[jagged turtle]

what can you do with it?

[visual pawn]

and the other is if I make the software "properly", there's basically no way for Neuro to fuck up

which would be very boring

well, as an example, ventriculostomy, which is a procedure where you drill a hole in the skull and insert a catheter deep into the ventricles to drain them

if someone as hydrocephaly

[jagged turtle]

ah

I see

maybe uh

[visual pawn]

if I made the software and robot integration properly, it's broken down into phases

[jagged turtle]

might or might not be a good idea

depends

[visual pawn]

like, the robot simply would not be allowed to move a probe sideways while inserted in the brain

or it would not accept any path that would intersect with any blood vessels

so for this to actually be fun, there needs to be the chance to screw up