programming

[kind nimbus]

Kernels or config generations?

[tender river]

na me only have 9 kernel

[kind nimbus]

sudo nh clean all

[olive sable]

I geuss im deranged?

Goodmorning

[rough bloom]

deranged together

-# also waking up so late that you're almost back to a normal sleep schedule

[silent cloak]

big ben scrambled ur PC

[quick condor]

Thanks flash attention

it would be worse but Im just now realizing I grabbed my xeon silver server and not the one with golds

[tender river]

im fine with nix-collect-garbage

[olive sable]

Thats how i roll

Ma days are longer than 24 hours so my sleepchedule drifts

[rigid snow]

i did that yesterday and i'm very broken

what do you mean you live like this

[olive sable]

Uh

I just do

[rigid snow]

dead by 30

[olive sable]

Ye probably

Oh well, wont be missing much anyways

[rigid snow]

don't say that

[olive sable]

Was meant to be joking but the lack of tone i can send through text doesnt help here

Maybe the neuro arg will be solved in 2038

[rigid snow]

ye unfortunately you wouldn't have been the first or last person to say that and mean it, i am for some reason surrounded by people who live by that "i'll die by 30 anyway who cares if i drink 14 red bulls a day"

[olive sable]

[rigid snow]

i'm fine

they just dumb

[pure nest]

yummers

[olive sable]

I dont really have any big goals i want to reach, but i think by the time im 30 ill have a good life

[rigid snow]

i have absolutely zero idea of where i'll be by then

so much can change so fast, not having any well articulated goals doesn't help either

[olive sable]

I think ill be doing the same exact stuff as im doing now, just a bit more employed and a bit more back pain

I just do what i enjoy

Within reason ofcourse

[rigid snow]

are there things beyond reason you enjoy

[olive sable]

I mean

I could spend 4 weeks binge-watching adventure time and completely ignoring everything else in my life

But i dont because i have other thing i should be doing

I enjoy my work so its not too bad

[pure nest]

what's your wokr

[olive sable]

Im a "technical artist" or whatever its called

So i do programming of graphics and 3d modeling

[pure nest]

ooohhhhh...

[rigid snow]

there are too many things i enjoy and i have huge issues doing other things i should

because usually the thing i’m supposed to do isn’t what i want to do in the moment

but i can do so many things i never feel unproductive i just end up fulfilling that urge with the wrong thing

strawberry donut

[kind nimbus]

sure i just like nh as a nice wrapper since it has some good shortcuts like the clean all one e.g.

[tender river]

do something often -> will be in shell history anyway

[kind nimbus]

well yea ofcourse

[ivory tiger]

Mhmm

[rigid snow]

[ivory tiger]

I struggle with this 2 but it's a bit different. basically I can't plan and carry out a big goal. I know the way to do it is just focus each day without worrying bout the big picture but it's hard to execute

[quick condor]

I have to break big things down into smaller tasks. For example today I got two things done, the discord bot api working for my AI to handle voice chat properly and the ministral 8b memory recall model model set up on the back end server

[languid escarp]

Anyone seen a YT vid describing the process someone took to make an AI inspired by Neuro, named Rose?

[olive sable]

no i dont think i have

[amber fractal]

not interested in neuro clones personally, I'd rather see the tech that comes after LLMs.

[jagged turtle]

yes, why so

(keeping in mind it's kinda weird to talk about it here)

i wonder if anyone outside of tech understood any of that

[olive sable]

they probably understood the windows update breakign neuro stuff

[quick condor]

I think there is a pretty big difference between Neuro clone and inspired

[olive sable]

but i doubt many of them know what nixos even is

[jagged turtle]

or x11 tbh

[olive sable]

ye

[quick condor]

That rant made my day

[jagged turtle]

lmao I need that rant clipped

[quick condor]

Maybe not wayland users because there is already something wrong with them

[jagged turtle]

om

[olive sable]

"my sliiight penchant for windows has nothing to do with me being a microsoft spy, by the way."

[quick condor]

Hacks Microsoft to steal Tay source code

[jagged turtle]

🔎 "where does satya nadella live"

[quick condor]

Installing a kernel level root kit on Vedal's pc

[umbral thorn]

[amber fractal]

[olive sable]

[amber fractal]

I wish I was sponsored by big tech, maybe then I could spend more time coding AI (the NN repo) because I'd actually get paid for it.

[quick condor]

I think there is an important distinction, especially in programming: works similarly/takes inspiration from doesn't necessarily mean clone. In general clones are going to be worse but there is countless permutations that someone can take in development that can result in a similar outcome... It reminds me of something a professor told my class years ago: I'm not going to prevent any of you from cheating but I'll know. There is next to zero reason any of you should have the same code and it's not hard to tell who is the one being copied off and who is copying

[amber fractal]

My brother, you know what I mean. The only type of LLM based persona I'm interested in is neuro and evil. That is it

I wouldn't even be phased by something better as long as it still has a LLM under the hood

[quick condor]

That's like saying you're only ever going to be interested in one vtuber because they are just a person behind a model.
Not saying that how you feel now will change, especially since everything elae I've seen does fit the "bad clone" label pretty well, just that taking an overly harsh viewpoint isn't generally great

[hearty notch]

a lot of people are just sick of llm tech because of an understandable reaction to the feeling of forced overhype and ubiquity

[amber fractal]

God forbid I have opinions diffrent from yours

[hearty notch]

I still personally find llm tech exciting but I also think integrating models into everything is uninspiring and 99% slop

[quick condor]

But I can also see what you mean as even removing the persona skin, people have pretty stiff opinions of which model responds the best

[amber fractal]

I just don't find LLMs personas as a general concept interesting full stop, that part of me died out a long time ago. I'm just interested in Neuro and Evil because of them, not the tech behind them.

That's my opinion

[jagged turtle]

most useful youtube ui

(I clicked on the clip button)

[quick condor]

What? Lol

Nvm, I can already tell by the logical fallacies there that there's no point in attempting a discussion

[hearty notch]

its not fallacious even if you think 'the tech behind them' 'is a part' 'of them' because they arent mutually exclusive concepts

it is valid to think the tech behind them is involved in making them interesting without that being the center of why one finds them interesting

[quick condor]

Right but if that's the case then the possibility of something using the same tech "part" but that acts completely differently being possibly appealing later on to you should be considered. If its not what makes you interested in them then someone else using it doesn't preclude you from being interested in them

Idk

[hearty notch]

there is only room for 2 ai in iggly dot npys cold cold heart

also like neuro and evil are a large part interesting because of the media and social ecosystem thats developed around them

which any llm as clever as them cant easily attain

because it takes a lot of cultivation

[quick condor]

I don't think the LLM is even relevant to that

[hearty notch]

the fact that it is an llm is interesting and idiosyncratically shapes how people feel about it and also the ceiling and dynamics of attahcment

at least thats how i see it

its interesting in how singular a phenomenon it is

[quick condor]

I think a part of it is that it wasn't planned, it was mostly organic. The problem with developing something like Neuro and Evil is that you can't just decide one day to build it and get the same outcome. As with anything that is in part popular due to its novelty, it's hard to create something similar unless it is unique in some way

[hearty notch]

youd need some unique twist to not be seen as utterly derivative and trying to ride on coattails

[quick condor]

Yes. And I don't think that it's impossible to do with something like this, the thing is that unlike something like say general vtubing, the barrier to entry is pretty damn high.

As an example: the project I've been working on is pretty bare bones and it requires something like 3-4k in just GPUs. Sure you can rent cloud compute but that gets stupid expensive fast

[mighty thorn]

[quick condor]

The other thing is that even with the gear, it's not simple to get to work even remotely correctly

[hoary oyster]

Make it a parasocial train wreck with the creator slowly losing rheir mind

[olive sable]

the discussion above here is stupid
thats like saying, "why is this person your friend? at the end of the day they're just made of flesh. you should actually go befriend a cow because its basically the same thing and im right"

[hearty notch]

i said unique twist...

[quick condor]

I don't think it's possible to come close to what Vedal has created without losing your mind

[mighty thorn]

I’m currently struggling to understand the point trying to be conveyed

[olive sable]

iggly said they dont care about different llms other than neuro and evil that much

[hoary oyster]

It could be much muuuuch worse

[hearty notch]

i think hascrack got disheartened that iggly.npy is not excited about the general idea of llm persona training and tried to construct a logical reason why they should be (even though thats not how it works)

[hearty notch]

yea ik i actually thought about like played up main creator personas that could be valid that are very diff from vedals persona

tbh kinda valid, the human in the loop being a diff kinda compelling could make it

[quick condor]

No, that wasn't the point

[hearty notch]

thats what it felt like to me, what was your point then

[quick condor]

The idea that nothing similar could possibly be interesting but yet that isn't the reason you find them interesting.

Either you like them for some other reason (in which case you could like something similar for a different reason) or you do actually like them for that reason (and you are just saying that you don't)

[mighty thorn]

I too like saying words

[quick condor]

I'm not sure where the disheartened came from

[true hemlock]

he's implying that he only likes neuro and evil because of their character, not the llm or whatever

simple as that, because i view the neuro twins the same way.

[hearty notch]

oh i see it was the appearance of fallacy that bothered you

[olive sable]

iggly said they don't find LLMs personas as a general concept interesting. not that they dont find evil and neuro as personas interesting

[hearty notch]

idk i just try to steelman and figure out how their view can be logically consistent and its usually in some alternate view on dependency or relevancy

[mighty thorn]

[true hemlock]

it doesn't have to always be logical and technical

[hearty notch]

i love words

i hate words

i love and hate everything

[olive sable]

[quick condor]

At the end of the day my point was simply if you aren't sure why then it's better to keep an open mind, not you have to like something else

[true hemlock]

"i like neuro and evil" doesn't have to be "i like anything llm". the llm behind is just being used to power their interaction, some people already knows the tech behind and find it boring but neuro and evil themself are a separate character that you can love separately from whatever is behind them

[olive sable]

[mighty thorn]

[quick condor]

Right that goes both ways though

I'm not saying that if you like them you must like anything like them.

I'm saying that if it isn't what interests you, yet you like them, then there is a chance down the road something different comes along that you like for a different reason

[hearty notch]

yea but the prior antecedent was like being interested in llm tech itself (on a programming level)

and thats what they seem firm on

and programming is a huge world so like

its a disinterest one could easily sustain (on the programming level)

[olive sable]

i thought the original message was "not liking llm persona stuff, but liking neuro and evil despite that", not "liking only neuro and evil's persona"

so from what i understood iggly doesnt like llm perosna tech at all

[true hemlock]

he just likes neuro and evil as a character

that's all

[quick condor]

Idk I think it's kinda a pointless conversation. And there is like 4-5 different ways every piece of this could be taken and they are all completely valid and not mutually exclusive

[amber fractal]

Quack has the correct answer, end of thought

[hearty notch]

i can yap infinitely because i love words dont let me keep you here

[quick condor]

As far as the programming side goes, from what I've learned from my own testing, the model is only a piece of the whole pie.

[true hemlock]

dude

we all knew

[quick condor]

Did I say you didn't, I Just said that because I've been finding working through those parts interesting

[olive sable]

you know what parts i ifnd interesting?

[amber fractal]

It was implied

[olive sable]

raid shadow legends

[hearty notch]

neuro no

[olive sable]

sign up now to get 2 free bonusses

[quick condor]

I think it would be funny if neuro just kept berating Vedal until he caves and does a raid shadow legends ad in the most defeated monotone way possible

[olive sable]

[quick condor]

In the same way that he gave up fixing his headphones and gave in to using the razer ones

Fun fact (idk if this is still the case) the razer software that Vedal mentioned popped up when he plugged them in use to be a method for privileged escalation attacks

[olive sable]

erm

[hearty notch]

tbh i mainly love neuro because she is basically me i have the exact same personality and im a narcissist

[quick condor]

Basically the software launched with administrative privileges and when you chose where to install it, the file explorer window it pulls up is an elevated window

Even for a standard user

[olive sable]

i see

[quick condor]

https://adcy.io/zero-day-vulnerability-in-razer-devices/

It should no longer work but it was funny af

[olive sable]

i have prob never used a pc where the user account wasn't an administrator, so to me it sounds like a non-issue ngl

[quick condor]

The problem: in a corporate environment with standard accounts, you could plug in a mouse, a normal mouse, and gain local admin rights

[rigid snow]

i'm so confused, does it somehow bypass uac? is it because the synapse installation thing gets pulled in with the driver by windows and gets launched with admin privileges

[quick condor]

Yes

[rigid snow]

because it's a driver

well, "driver"
ye makes sense

[true hemlock]

also 4tb nvme arrived

[quick condor]

And it doesn't check if the user account has the rights to do so because it's handed by the system

[true hemlock]

and the price just increased again

holy fuck

[amber fractal]

Good call then

[olive sable]

Driver: San Francisco

everytime i look at ssd prices i get a miniature heart attack

[true hemlock]

considering that 1TB was already $130

$300 for a 4TB is nice

[olive sable]

[true hemlock]

now its up to $350 wtf

[amber fractal]

and every time you look at ram prices, you have a deadly heart attack

[olive sable]

what ssd did you get?

[true hemlock]

also i might able to snatch a 5000MB/s one for $220

but its QLC

:(

[true hemlock]

EN870

[olive sable]

i have no frame of reference for that one so everytime i look its jsut more number

[true hemlock]

4TB TLC

[olive sable]

they dont sell those here

so the only lsiting in amazon is 690 euro

[true hemlock]

damn fuck

[olive sable]

samsung 990 pro 4tb is 380

[true hemlock]

yeah about there

[olive sable]

oh

directly on the website that ssd is only 311

[true hemlock]

hm?

[olive sable]

elodica is the brand no?

they have a website

[true hemlock]

its $450 in here

for 990 pro

[olive sable]

damn

[true hemlock]

and is increasing...

[olive sable]

this one right?

[true hemlock]

so about $360?

damn

[olive sable]

ye

[true hemlock]

not even a day and that's another $60 increase

[amber fractal]

not me mistyping

[true hemlock]

i'll uh

fuck it

i'll contact my connections

see if they have 4TB in batch/bulk

[olive sable]

im considering getting 2 990 pros

[true hemlock]

4tb?

[olive sable]

im debating howe much capacity tho

either 2 or 4

[amber fractal]

$356.99USD

[true hemlock]

single 4tb is better than 2x 2tb

less prone to fail

but uh

the advantage with 2 drives is that you get to use it separately from the bandwidth

[olive sable]

1tb is 140 euro, 2tb is 206 euro, and 4tb is 380 euro

[true hemlock]

christ

[amber fractal]

yes

[true hemlock]

140 euro already??

[olive sable]

ye

144 wihout heatsing, 147 with

the 2tb is 209 without and 206 with

[true hemlock]

yeah, great

fuck you, sam altman. fuck you

[amber fractal]

sam triangle is good tho, is a real bro

[olive sable]

the 4tb is 380 without heatsink, and 430 with

fucking 50 euro price difference

[true hemlock]

one rise, the other fall

[olive sable]

[true hemlock]

those bulky heatsinks are dirt cheap off ali

[olive sable]

yep

and my mobo has its own heatsink so it would only get in the way

lemme check if i got paid already

[amber fractal]

Capacity of yes is just 500gb to 4tb

[true hemlock]

$2 heatsink

[amber fractal]

no infinite storage

[true hemlock]

this thing is usually meant for gen 5 drive

[olive sable]

ignore the 1 star review

[true hemlock]

kek

[amber fractal]

1 star means it is good

[true hemlock]

no way the speed is 4TB

[olive sable]

Advertised as a 4tb ssd at what appeared to be a very good price, but in fat it was/is a 1tb ssd at an inflated price. Basically fraudulently advertised to catch buyers. Thankfully i was refunded after complaining

its a 1tb at an even more exorbitant price

140 was enough

[true hemlock]

could've been worse.

[amber fractal]

how did I not catch that

[true hemlock]

the usual firmware scams

that reads out

"20TB"

but is actually 8GB

[olive sable]

20TB per second? per year?

[true hemlock]

hey surely 20TB meant the TBW endurance

-# 8GB flash having over 20TBW endurance would be crazy tho

also preordered some 22TB sas hdd

[amber fractal]

surely the 4TB read speed is just an option

[true hemlock]

i need to expand my 88/132TB raid 6 lol

[olive sable]

lookign through aliexpress. the legit listings are even more expensive than jsut buying on amazon

and the reasonably priced ones are like this

[true hemlock]

only things worth buying from there is usually CPUs chinese people sell on bulk or OEM CPUs

and motherboard sometimes

or if you're looking for niche old stuff

[olive sable]

i ussualy get parts from there

for 3d printing or old tv parts and such

[true hemlock]

don't you love it when certain grey haired someone bombard you with lga 3647 xeons in bulk shipped to your address from ali

-# not that im complaining

[olive sable]

grey haired?

[true hemlock]

who else im talking about

[olive sable]

uh

give me a sec

im bad at rememberign faces

[amber fractal]

https://frame.work/products/storage-expansion-card?v=FRACCFBZ0A Oh how slow this is I grabbed the 250gb with my initial order, hasn't been through much use though.

[olive sable]

chirp???

[amber fractal]

om

[olive sable]

thats the only pewrson i know you know irl

i have no clue what colour their hair is

[quick condor]

I just have a pile of these as well as some 1.92tb ones

[olive sable]

let me see what i need to buy fdor storage before these prices get even worse

[quick condor]

Mood

[true hemlock]

im laughing my ass off rn

[amber fractal]

I'd rather just scrape together storage than deal with these prices

I not saying anything

[olive sable]

an nvme for linux since im still running on sata.
a big hard drive to store a fuckton of data on.
and maybe a new ssd for windows so i can put the old one in an external enclosure

[true hemlock]

6x 22tb sas coming soon yippie

$210 each

also im planning to snatch like 3 3090 during my visit at my home town

[amber fractal]

I just really don't have the infra to get decent hardware yet. I'm just going to stall this out as I thankfully don't need it for my projects.

[true hemlock]

$400 each probably

[olive sable]

nice

[amber fractal]

letsgooooooo

[quick condor]

My desktop has a mishmash of ssds in it, i think currently I have a 1tb WD Black sn750, a 1 tb crucial P1, 2 960Gb Intel DC S4600s and a 2tb 870qvo

My servers have dual 960gb Intel DC S4600 for the boot drive and the DB server runs 8 1.6tb SAS 12ssds while the vm host has 8 1.92TB ones

[olive sable]

at this point its actually either 4tb ssd or 24tb hard drive. both are 380

[true hemlock]

that would've been a headache

to have a lot of small drives

[amber fractal]

Time to balance out everyone else with my own setup, which is 3(7*) laptops and a mini PC.

[olive sable]

rn i have 1tb nvme for windows.
2x200gb sata ssd, one of those being linux
and a 500gb hard drive

[quick condor]

Ehh depends. Desktop ones each have a use and servers are either hardware raid or ceph

[true hemlock]

all my sata are hdd

[olive sable]

you know how there are these external hard drive enclosures? how do those even work?
are they slow compared to straight sata?

[true hemlock]

uh

depends

iirc they usually use Thunderbolt

[amber fractal]

framework 13 for most work, some gaming (including VRC), RTX 3060 laptop for VR titles, GTX 970 laptop that is being used as just cold archive, and N100 mini PC just as extra IO devices.

[true hemlock]

but some probably uses Ethernet

[quick condor]

Depends on the controller

[olive sable]

usb 3.0 supposedly

[amber fractal]

forgets a laptop

[true hemlock]

wouldn't recommend

[olive sable]

[amber fractal]

some USB some thunderbolt, depends on which

[true hemlock]

i don't use a dedicated external hard drive enclosures either

[quick condor]

Most are usb 3.0/3.1. I've used an external usb3. 1 one that works well enough for large transfers

[olive sable]

i dont think its worth it

[quick condor]

Mostly use it when shooting astro all night

[true hemlock]

5Gbps

[olive sable]

thouht it might have been handy for using the files on my laptop too as some sort of central server thing without actually getting a server

[true hemlock]

i mean, that's plenty of bandwidth but i still wouldn't

especially when you can set a caching system, just use pcie sas controllers imo

[amber fractal]

the 10Gbps nvme usb adapter treats me well, but I wouldn't use it for anything major.

[olive sable]

ill keep it to the nvme enclosure then

[amber fractal]

we love the gen 4 rocket drive being used inside of it as a flash drivr

[olive sable]

its been in my house for like a week already but i have no drive to put in it

[true hemlock]

im tempted for a trx50 system

[quick condor]

For me I treat it as a flash drive because flash drives suck

[olive sable]

depends on the drive

the 5mb/s ones suck, but the usb 3.2x2 super plus ultra pro rtx ones are nice

[quick condor]

Compared to a mid range nvme in a 10gbps usb enclosure, all flash drives suck

[true hemlock]

gen 3 nvme is all you need for 10gbps

even then the gen 3 is still way faster lol

[quick condor]

Nothing comes close to the time I configured a zfs pool with a 512Gb dirty cache ram buffer

That was fast but sketch af

[amber fractal]

Man this drive is overkill https://www.amazon.com/SABRENT-Rocket-2230-Performance-SB-2130-1TB/dp/B0BQG6JCRP

sidenote for anyone who checks link, 3rd size option

[quick condor]

Actually the nvme enclosure I use is a sabrent

[olive sable]

the one i got is anker

[true hemlock]

aight i should still have about 32 free pcie lanes on my sp5

[quick condor]

Anker is solid. I'm currently using one of their 100w Gan wall adapters

[amber fractal]

I have some noname enclosure, it was free so no compaints

[olive sable]

anker is afaik just a drop-shipping company but with decent quality

[true hemlock]

shopping list rq

• 6x 22TB sas HDD
• 1x 4TB nvme (arrived)
• 4x V100 32GB SXM (with SXM to PCIe adapter)
• 3x 3090

[olive sable]

yall got a fuckton of money jesus

[true hemlock]

i am NOT affording another sp5 system, my current one should do fine

[amber fractal]

I should figure out what funds I have

[true hemlock]

i could get the V100 + SXM adapter for $300 from another dealer

[amber fractal]

I was going to ask on that actually

[true hemlock]

$3540 total

the V100 and 22tb sas is for my research server so

[quick condor]

BTW that A2 is working perfectly for my DB server

[amber fractal]

I only maybe have 1.5k only thing I'd want to do is get a gpu upgrade

[true hemlock]

my sp5 should support bifurcation

[true hemlock]

quick lmao

50 series prod getting cut by 2026

[olive sable]

with your current storage situation im suprised you think using a riser cable with a better priced gpu is such a sin

[quick condor]

Way better than cpu for a system that needs a little gpu compute

[quick condor]

That server is using Dell enterprise ssds, my desktop has the scuffed storage

[true hemlock]

i find it a bit more amusing that dude thought risers are "scuffed"

[amber fractal]

I honestly might just accept the jank and try a V100. I'm not getting anything better while staying decently cheap. Don't really use gpus for video out anyways so no loss

[true hemlock]

why though

eh

honestly

[quick condor]

I was using a riser in my desktop before I went watercooled. Caused issues

[true hemlock]

you can still try configuring the V100 to P0 state

and overclock it to near match 3080 performance

volta is hella overclockable

[olive sable]

Toshiba X300 PRO 20TB for 357 bucks?

[true hemlock]

that's fair price

new?

[olive sable]

ye

amazon

[true hemlock]

yeah

nice

[quick condor]

Not bad

[olive sable]

shit. import fees

nevermind

[true hemlock]

oh F

[amber fractal]

1, I don't mind the jank and am just curious. 2, 3080s are $300 in the used market here.

[true hemlock]

the hbm memory is nice

just sucks that its hard to overclock them lol

the major reason im opting for V100 is because they're the cheapest gpus that has good fp64 rates

[amber fractal]

I'd also like 3x the memory capacity compared to a 3080, that is also a factor if I'm looking at cost.

[true hemlock]

huh, where are you getting the 32gb ones for $300

[quick condor]

I've got hbm in the jank that is my laptop

[true hemlock]

you're not me

[amber fractal]

Nowhere yet

[quick condor]

The 8705g is a cursed cpu

[amber fractal]

I doubt I'd be able to get any at that nice price anyways, so it'll stay on the backburner for awhile.

I am way too patient

[true hemlock]

you can def get the 16gb for $300 though

[amber fractal]

Yeah, that is possible

[olive sable]

Seagate Exos X18 Enterprise 16TB, 237 euro free shipping with prime

[true hemlock]

nice

[amber fractal]

I'm better off waiting honestly, it's better for purchases like this to be able to swing a lot more cash around to do it.

[olive sable]

quack approved. aight imma buy it

[true hemlock]

uh

gpu supply gonna get fucked in 2026 lmao

[amber fractal]

Oh I'm aware

I'll just simply not need it for many years

as far as I'm aware, I can stall until the AI bubble collapses

[true hemlock]

ah

whats yours rn

[amber fractal]

Still the same 4x P40 cards on thunderbolt egpu docks

[true hemlock]

OH LMAO

based

[quick condor]

Idk why but I just had a thought that should work: using my desktop as the heat source for a lava lamp

[true hemlock]

i wanna do that aswell now

you tempted me

[amber fractal]

I would be test booting this but I have cable issues

[amber fractal]

Nah, I have a better version for you

[quick condor]

It should work

[amber fractal]

I know it does

[quick condor]

Now I need to do it

[olive sable]

how warm does a lava lamp need to be?

i thouhgt it was more than 100C

[true hemlock]

god fucking damn it

[olive sable]

[true hemlock]

a bit over that yeah

its a PTC heater

[amber fractal]

let me propose what I would do if I were to redo this first, alright?

because it keeps the same spirit, but more cost efficent and useful personally

[olive sable]

what is the cursing baout? what happened?

[true hemlock]

12v 1.95A

[quick condor]

They apparently run at 55-60c which is what my loop runs at under full load

[true hemlock]

lack pcie lanes

[olive sable]

[true hemlock]

that's the internal bottle temp, not the heater.

the heater itself reach 110-130C

its a 12V 1.95A PTC, max power draw around 23W but overtime goes down to 20W

[quick condor]

Depends on the heater source, that's just to heat up quickly

[true hemlock]

max you'll see this thing run on is 140C, but the minimum is definitely above 100C

[true hemlock]

no lol

uh

have you read the thread

[quick condor]

I'm saying lava lamps in general

[true hemlock]

the thread for lava lamp parts and replacement

[true hemlock]

ours is using that PTC heater which just averages at there

the average lamp uses hotter ceramic heater.

there

[true hemlock]

this is why the neuro lamp took quite long to heat up

because we didn't have it shove 200+ C

your 60C ish PC won't do anything realistically due to heat transfer limitations and heat dissipation

[quick condor]

What? There are 15w incandescent lamp powered lamps

[amber fractal]

if I were to redo this, I would instead of using thunderbolt because lack of bandwidth, I would instead take the extra cash for this thing https://c-payne.com/products/pcie-gen5-mcio-switch-100-lane-microchip-switchtec-pm50100?pr_prod_strat=e5_desc&pr_rec_id=062baceb0&pr_rec_pid=10242691465483&pr_ref_pid=8642442035467&pr_seq=uniform as the all of the thunderbolt equipment costed close to the amount of this board anyways. 2k for the chip board, and 50 for each of the pci gen 5x16 slots. Instead of the $300 (tbf $200 ones exist) thunderbolt docks and $200 for each of the two thunderbolt hubs.

[olive sable]

those get hot

very hot

[true hemlock]

and they run hotter than our neuro's lamp heater

[quick condor]

They don't make direct contact

[olive sable]

15W lamp can get to between 200 and 260C

[true hemlock]

and even if its not direct contact you're still looking at about above 120C being radiated lol

[quick condor]

They produce 45btu of heat. Period

[amber fractal]

the meme in this one is 4 of the x16 slots for gpu, 1 x16 for a quad nvme, and the final x4 to the host system

[true hemlock]

honestly

you can try it yourself

and see for it yourself

[quick condor]

15w is a fixed input power it doesn't magically make more heat

[amber fractal]

surely it is a valid configuration

[olive sable]

15W lamps for lava lamps are not the same type as the ones used in lighting homes

lava lamp lamps are made to run a lot hotter by having less light output

[true hemlock]

bro throws out the heat accumulation in the equation

[olive sable]

that too yes

[true hemlock]

also the total area of heat dissipation/spread

don't underestimate 15W lmao

[amber fractal]

In conclusion, whoever wants to redo the thunderbolt thing. what about instead of only being limited to pci gene 3x4, you get a full pcie 5x16 connection?

[true hemlock]

given different function the 15W is going to produce higher temperature than your 200W pc or something. thermal density also has to be considered

[quick condor]

My PC pulls 1400w

[olive sable]

so?

[true hemlock]

i know one way that could work for what you're thinking. put the lava lamp bottle directly on your CPU IHS

[true hemlock]

doesn't matter, heat density also has to be considered

my server pulls above 3000W

doesn't make it ideal to heat up a lava lamp

[olive sable]

im sure if you turned of the cooling in your pc at 1400W it too would easily reach 200C easily if it didnt melt and thermal thorttle

a lot faster even

lava lamps take hours to heat up

[true hemlock]

and pls don't blame me if it explodes thx

[true hemlock]

when he said 1400W, i can only think of him either redirecting the whole airflow to the lamp which just straight up won't work, considering that ambient bottle base needs to be above 180C if its not a direct contact, or him putting the bottle on one of the chip and ends up exploding it instead

[olive sable]

[quick condor]

I quite literally have run full power for 36 hours batch processing data. Airflow is mostly running through a large low speed rad into the case and a higher speed 240mm rad on top.

[olive sable]

thats the thing tho

airlfow

you're not letting it build up

[young plover]

The semiconductor properties turn funky long before 200°C

[quick condor]

Exhaust temp is 50c when I keep my fan speed at 30%

[true hemlock]

yep that wouldn't work

[quick condor]

Dropping it to 15% would spike temps significantly

[olive sable]

if you have a 100W cpu without the cooler on it. that thing gets to 100C in seconds or even in miliseconds during full usage

so it isnt that far fetched to think a 15W energy source could get to double that temperature over 60x60x3 times the duration

that is 10800 times the duration

ten thousand times

[true hemlock]

you wouldn't want to heat the lamp in just few seconds either

you need to let the heat build up

which is why heater under 25W are used

[quick condor]

You are aware you can literally calculate the btu output based on wattage input

[true hemlock]

this generation is so cooked

[olive sable]

you are aware that eventho wattage is a unit that takes time into account, we're discussign different periods here?

[true hemlock]

with this kind of reading comprehension

[olive sable]

as in, there may or may not be a small difference of ten thousand times the length

that lets the heat build up

[quick condor]

The heat doesn't build up. It reaches 60c inside floats up and cools. It shouldn't get hotter or it wolnt flow

[true hemlock]

also 50C airflow, i can't stop thinking about how thermodynamics is also being thrown outta the window if he wanted the internal bottle to be above 65C with just that kind of airflow

[true hemlock]

and what happens when it cools?

heat dissipation

[amber fractal]

[true hemlock]

there's a reason we make 15W build up to above 180C on the base

heat transfer is slow

[quick condor]

50c is the temperature if I am running at 30% fan speed. Temperature goes up the lower the speed is

[olive sable]

lava lamps are designed to insulate tho. the whole point is to let the heat build up?

[true hemlock]

and heat dissipation has to be considered

therefore

even if you make your PC exhaust 80C

it still wouldn't melt the wax

what makes you think it could, when the ambient temperature on the bottle base is literally above 180C

[olive sable]

i dont even get what we're arguing anymore

[rigid snow]

if you immerse it in the exhaust maybe it will in like a day

[olive sable]

ive had enough

[true hemlock]

same

[true hemlock]

and it wouldn't flow up and down because by then you made the whole bottle have the same temperature

[rigid snow]

ye

[quick condor]

Well kinda... I could then point a spare case fan at the top

[true hemlock]

ours is at 130C for the neuro lamp, and that's direct contact

[true hemlock]

then that'll turn it into "never melt" because you're also dissipating all the heat lol

my argument is simple, even if you have 80C from your exhaust it wouldn't be enough

[quick condor]

If you get the bottle temp up enough to melt it then it's hot enough

[rigid snow]

this is so silly. have you ever tried to heat anything with warm air, it takes forever

[quick condor]

It does yes

[true hemlock]

heat gun time

[quick condor]

But I'm not running it to heat the lamp

I'm running it in the first place

[true hemlock]

for what purpose lol

[quick condor]

Batch processing of data

[true hemlock]

so the lamp shouldn't even be argued about in the first place

[rigid snow]

[olive sable]

1

[quick condor]

It was quite literally a dumb thought I had and yall are focusing in on a very specific lamp with one of many wax compositions and reading off its specs and over analyzing a dumb idea

[olive sable]

are 1TB ssd's even worth considering? they're 75% the cost of a 2tb while being 50% the capacity

[warped narwhal]

Welcome to #programming

[quick condor]

Fair

[true hemlock]

shittier endurance too, no

[olive sable]

im thinking about getting a 2tb for linux and getting another 1tb cuz they're cheaper for windows. butt he pricing on 1tb is so ass

[true hemlock]

ssd endurance scales with capacity

[olive sable]

[quick condor]

In case yall forgot I don't do smart... That is 11 delta pfc1212DEs in my case

[true hemlock]

damn soft tubing

[rigid snow]

pedantry central fr

[quick condor]

Your questioning soft tubing when I have 550w of fans

[true hemlock]

honestly? doesn't seem like 550W

unless if for some reason you push them beyond 8000RPM

no, more

[quick condor]

12v 4.8A fans

They aren't 20/28mm fans they are 40mm thick

[true hemlock]

oh 11 of them?

then that's fair

[quick condor]

Yes

[true hemlock]

i have some of those aswell

[quick condor]

Yah, tbh they are pretty quiet below 40%

[true hemlock]

was also given an industrial 6000rpm 120mm fan, though that one is 28mm thick

and loud asf lol

[quick condor]

I have some 93mm hp 18k rpm fans but they need a bldc controller and pull 17a

[olive sable]

550W of fans seems excessive but i guess you're never actually ussing that much power

[true hemlock]

lmao that

[quick condor]

Yes those are metal blades

I'm actually scared of those ones

[true hemlock]

fair

my 6000rpm made a cut on my finger

[quick condor]

Why do you think the Deltas with exposed blades have guards on them

Been there

[quick condor]

Final dumb thought, neuro can write some simple code but how bad do you think the outcome would be if she was given unfettered access to something like claude code in a non sandboxed environment

[olive sable]

i think these are the best prices im gonna find

[jagged turtle]

How about we test that?

(not the unsandboxed part tho)

how did I get 3 different versions of esbuild in my project

[void mortar]

Have you tried Bun?

[jagged turtle]

yes. i dont particularly like it

but I will give it some more time to smooth out

also the docker node sdk doesn't work with it anyways lol

[void mortar]

What is it you don't like about it?

From a DX perspective it was so nice to be able to massively cut my dependencies list, not having a ts-compile step, built-in bundler, etc.

[jagged turtle]

I don't know, it just doesn't particularly vibe with me

Probably because I use pnpm rn and it saves so much storage space and is better at working with monorepos

[void mortar]

Ah! Don't have experience with Bun in monorepos. Would have to do research how well it behaves.

But given size of Bun I wouldn't be surprised if they have fixed support for that

And I so wanna talk "monorepo vs separate" with you, but don't have time right now

[jagged turtle]

I see value to both

I do really like the fact that Bun has a compiler so you can make EXEs, though

[jagged turtle]

anyways, do want to have a discussion about it because I find it so interesting

[rigid snow]

just realized the grass being fucking green isn’t normal for the end of december

it’s that bad ye

[sage crag]

what do you mean green grass not normal for end of december

ive not seen it snow more than once a year

[rigid snow]

not normal where i live

[sage crag]

it will be

[rigid snow]

blobal worming

[sage crag]

[rigid snow]

🪱

[rough bloom]

deploy bird

solve blobal worming

[sage crag]

[sage crag]

autismu

i keep hearing きのこ in random sentences and thinking "mushroom"

pattern recognition is dangerous

now i think everything is about mushrooms

[olive sable]

uh

kinoni?

[sage crag]

no

こ <- ko

[olive sable]

im bad at japanese

oh

ni is with that extra

に

[sage crag]

its not the same, the shape of the bottom part is different

[olive sable]

uh

looks pretty similair to me

but extra vertical line

so, kinoko?

[sage crag]

ye

[sage crag]

in a more traditional font

or handwriting

you could see it

[olive sable]

fair enough

[sage crag]

here's one font

[olive sable]

uh

[sage crag]

here's another font

[olive sable]

the right one is more curved sure

[olive sable]

but on htis one it looks like they jsut smooshed it to be smaller

[sage crag]

it completely different shape NeuroCatUHH

[olive sable]

why you embed

emote is here

[sage crag]

because i enjoy

embed

enub~1

make emoji larger

[olive sable]

looks same size for me

[sage crag]

compact mode

[olive sable]

mobile moment?

oh

i see

is this twilight in the marketing material for this thing?

[faint sandal]

I just got my first copilot generated slop PR on my GitHub project

it's clear the submitter did not know what they were doing

[jagged turtle]

I wanna see

[thorny fulcrum]

Yo guys, any solid app recommendations for remote access to CasaOS? Just set it up and I'm a bit lost on how to get it working outside my home network

[faint sandal]

https://github.com/Still34/AffinityPatcher/pull/13

[languid escarp]

@jade kindle wanted help making their own chat bot, and I thought the vid would have useful info.

I'm working on something inspired by Neuro-sama using my Python knowledge, and so far I have a chatbot with short-term memory. Do you guys know where I can find some material to learn? Sorry if my question is bothersome
OH. I copied their message text. Not link. Ah.
And I couldn't find the vid again because I had bad search-fu.
Found it. Just needed to search "made an ai" and was like 3rd result. https://www.youtube.com/watch?v=muhbxvJxfQc 😤

[faint sandal]

they closed the PR after I questioned them LOL

[sage crag]

@rigid snow

[rough bloom]

at least they're admitting that they have no clue what they're doing

[sage crag]

reopening the PR is so funny though

[rough bloom]

yeah

[faint sandal]

i feel like they don't know github very well

[rough bloom]

reopening just to comment

[sage crag]

it'd be funnier if they made another issue to comment about their previous issue

[rough bloom]

there's still time

waiting for the sequel

[sage crag]

veryNeuro

[true hemlock]

hmm

B68S

micron?

iirc they have B68S 8Tbit chips

that means TLC, though. nice

[hard raptor]

https://youtube.com/shorts/sWz1soDjfQI

The only true use.for these

[regal hare]

I've been thinking of making a website that's like a radio station for neuros covers. Anyone's down for that idea?

[sage crag]

@jagged turtle

[jade kindle]

Thanks for taking the time to find the video, I'll watch it now!

[true hemlock]

flight tomorrow so i need to pack my whole briefcase PC in my checked luggage ig

that thing survived being dropped to concrete thrice already and it broke the concrete, no doubt its surviving the flight

[quick condor]

Problems I see:

1. The visual model isn't the best point to start on something like this.
2. Immediately going to streaming responses before the LLM finishes it's response without first getting non streamed responses to work is a nightmare to debug
3. Most critical: this is a basic chat bot. It's missing everything but the basic llm pipeline

[true hemlock]

2 highly depends on your back end

just wrap llama.cpp or something lmao

or any supporting compute backend that you want

[nocturne olive]

Unfortunately the llama.cpp Java bindings kinda suck
Sure they do all the basic functionality
But it's unfortunately unable to unload a model

[quick condor]

I guess true, I found it easier to implement later mostly because STT and LLM currently share a gpu for me

Currently I just use openai api to my existing api for the primary model. For the memory recall/deep embedding model on the memory server I went with Exllamav3 with a custom api wrapper because I'm trying to strip assistant naming conventions from calls

[true hemlock]

ai "devs" nowdays smh

[nocturne olive]

Hopefully that model unloading issue will be fixed some day because I want to make an app around those bindings

[quick condor]

But then again I don't use ggufs so that means no Llama.cpp

[nocturne olive]

Why not use GGUF?

It's free VRAM savings

[true hemlock]

and bandwidth efficiency

you get scaled blocks

[quick condor]

Fine control and performance with exl3

[true hemlock]

how big of a model and how fast of token per second

on what hardware

exl3 also has quantization though

[sage crag]

https://cdn.discordapp.com/attachments/1072697081443131476/1395669923774337116/8b9145324db46a5e510f6c3e9fa3fb8d28331253.gif

[quick condor]

Depends:

1. mistral small 24b 8bpw HB16 on a 3090ti and A4000 with 32k context - 25t/s
2. mistral small 24b 8bpw HB16 with 72k context on 2x 3090ti and the A4000 - 20t/s
3. Llama3.1 6bpw HB6 (I forget the context, most likely 32k) 2x 3090ti and the A4000 - 15t/s
4. 2x mistral small 24b 6bpw HB16 and 1x mistral small 24b 5bpw HB16 all with 6k context (batched together) - 90t/s
5. Ministral 8b 6bpw HB16 on an A2 with 8-16k context (unsure of t/s)

[nocturne olive]

25 t/s 24B on 3090Ti?

Hem

[true hemlock]

wdym on 3090Ti and A4000

don't tell me...

[quick condor]

It's split between the a4000

[nocturne olive]

Meanwhile my 3090 (not even Ti) 30B 190 t/s (though MoE)

[true hemlock]

😭 why the fuck?

no but

why the fuck?

😭

[quick condor]

Context on a slightly above Q8 equivalent model

[nocturne olive]

Why use a model above Q4?

[true hemlock]

get Q6, the A4000 is seriously bottlenecking it + overhead

[tender river]

can only see in september here

[true hemlock]

A4000 is less than half the bandwidth of the 3090Ti

[nocturne olive]

I think it was something like IQ4 something something

[quick condor]

At 32k+ context q4 is a joke

[nocturne olive]

Is it though?

[true hemlock]

q6

[quick condor]

Q6 still introduces some errors and conversations are being fed into training data

[true hemlock]

????????

"conversations are being fed into training data"

😭

this is so wrong in so many ways

you'd NEVER feed its own output back to training data

RL is a separate case

[quick condor]

Well not raw but ideally with minimal editing. Mostly selective data selection based on classification after running it through several filters

[true hemlock]

you still wouldn't, it'll just overall slightly degrade the model because the thing it output is almost always lower quality of its initial training data

ideally, also do not finetune the model with small data often

the model's generalization would be fucked up

[quick condor]

Initial training data is multi lingual and shaped for an assistant role. I'm not doing fine tunes rapidly, I wait until I get enough good training data from the backend crony job to fill a specific training type dataset before training

I think my last fine tune was 25mb of raw text

[true hemlock]

that's... relatively small

how many epochs?

[quick condor]

But that was a different project training prose. Not alpaca formatted

[true hemlock]

i need to swap the aio with my sff cooler and pack this thing

[rigid snow]

[true hemlock]

don't mind the random cpu cooler, that was used for my vrm cooling for my overclocking setup last time

yep the sff cooler would do fine

4.9W at 5GHz

for basic usage

[quick condor]

Also BTW from last night's conversation, this is what full speed sounds like

[true hemlock]

sounds like my 7u

[quick condor]

Probably because the fans came from a 4u

[true hemlock]

my beloved

gonna visit my home town for few weeks so im bringing this bad boi with me

[quick condor]

Wait how does 24/32 work (I'm not use to the last couple gens of consumer cpus)

[true hemlock]

8 p cores with SMT + 16 e cores with no SMT

[quick condor]

I figured that was probably why

[true hemlock]

its an i9 13900, just that its an engineering sample

except this one is cherry picked so a -225mV undervolt somehow works on it

same CPU i've used to get #1 in this leaderboard alongside with some samsung b die https://github.com/samvanmaele/vulkan

[quick condor]

I'm running a 10900x in my desktop and second Gen scaleables in my servers

[nocturne olive]

2 epochs on an LLM seems like a lot but also I don't know how differently epochs behave on finetuning

[true hemlock]

not 10980xe? smh

2 epochs is fine

[true hemlock]

@quick condor should give a try

hint: its more of CPU single thread bench technically, the fun part is figuring out the optimized compiling flags on your machine

[quick condor]

It was cheaper and I didn't really need more cpu performance on my desktop, I was just replacing the 7820x that it had since I built it since intel filtered you on pcie lanes

[nocturne olive]

I guess my triangle score was somewhat limited by me not messing with the compile flags too much, but I'm also no C/C++ dev so I'm not super familiar with the compiler

[true hemlock]

nah

yours makes sense

i could get that high just because the overclock + some pipeline config + memory overclock and timing tuning

[nocturne olive]

I guess that makes sense

[true hemlock]

extreme overclocking gets you that far lmao

[nocturne olive]

Though I'm also pushing my mem to its limits
CPU not so much though

[true hemlock]

but i can't do it for long

else im degrading my poor chips

im using samsung b die

that is meant for overclocking

[nocturne olive]

I sure don't have that

I have I don't even know what

[true hemlock]

its a 2x8GB but iirc its gone up to over $600 for that specific kit rn

[quick condor]

I have a pretty conservative oc on my cpu since it's the last thing before the 480mm rad. The 120 and 240mm ones just take the edge off the GPU heat dump

[true hemlock]

undervolt

do an undervolt

then clock offset for overclocking

[nocturne olive]

My CPU has 50mV undervoltage and 200MHz extra clocks on both P and E cores

[true hemlock]

you don't get extra heat that way and in fact it gives you less heat becase the voltage curve is being tightened

[nocturne olive]

I should try to undervolt and overclock more after subathon or something

[true hemlock]

iirc on very specific scenario mine can handle over -400mV (its how i can hit 5.2GHz at just 800ish mV) but that's for really short period of validation

[quick condor]

I need to get around to undervoltin but it's kinda finicky with mixed windows/wsl2 workloads

For the cpu though I just haven't really bothered much since it isn't my bottleneck in most cases

[true hemlock]

i love it when engineering sample are actually cherry picked

[nocturne olive]

Funny
My -50mV 5.2GHz is fully stable I think
I wonder if I can do -100mV or run higher clocks

[quick condor]

Depends, cpu eng samples are normally, gpu not so much... This was a system I hated dealing with because it kept chewing up cards

[true hemlock]

my piles of 50 series ES

(they overclock somewhat pretty well)

[quick condor]

(the p41 doesn't exist officially)

[true hemlock]

one hell of a clusterfuck

[void mortar]

You still have the warranty Ad sticker on the monitor?

[true hemlock]

i forgot about it lmao

[quiet abyss]

lol

everyone forgets it

[rough rover]

Does anyone know how the Neurosama vr architecture works or is it just smoke and mirrors?

[tender river]

its a combination of models is all i can tell you

vedal works on neuro full time he has lots of time to experiment and find something that works

[dreamy dune]

MEOW

[rigid snow]

[olive sable]

watched "1922"

it was alright

there was a scene where they showed a doorknob where the paint was peeling, and then i had to hear a rant about how thats inaccurate since they didnt do that in the 20s

[opaque sigil]

[olive sable]

[opaque sigil]

i watched the new avatar today, it was one of the movies of all time for sure

[quick condor]

My guess is something simple (bassicly a way to translate her normal movements into vrc inputs) but thats just a guess... Its what id do, simple and effective

[olive sable]

avatar blue people or arrow?

[opaque sigil]

the former

[olive sable]

ah

[opaque sigil]

story kinda eh but very pretty, would watch again (ok that's a lie i won't it's like 3h15m or sth)

[dreamy dune]

meow

[olive sable]

my 3D teachers say the same thing about the new star wars movies

bad story, good CGI

[opaque sigil]

i wouldn't say it's bad per se, it's just the same again

[olive sable]

[opaque sigil]

sometimes all you need is some good cgi

[amber fractal]

Man it was fun to play around with it, I believe bred actually had better results with like 3 core LLVM software driver on their system, so some curveballs exist.

I say this as the highest ranking laptop result. Then you look to see somehow it beat some interesting systems despite being only going to 3.5Ghz during that benchmark.

If I had decent cooling I would have ranked higher trust

[olive sable]

neuro guitar

[tender river]

https://imgur.com/a/CO5Mscw

[sage crag]

eneuro

guitar

wrr wrr

[olive sable]

wrrr

[sage crag]

https://tenor.com/view/wrrr-cat-angry-gif-10783708059666180688

[olive sable]

[sage crag]

enub

[olive sable]

i think

i might

play hollow knight this week

[sage crag]

minyanstare

i have a feeling

[olive sable]

its been actual months since i bought the game

[sage crag]

that neuro might as well

[olive sable]

i need to play before her then

dont want spoilers

[sage crag]

MiyagiCroco

[olive sable]

welp, its time for the final stretch in 3D modeling the hair

[tender river]

wrrr

wrrrrr

wrrrrrrrr

https://www.youtube.com/watch?v=1iPWt1gvT_w

[olive sable]

52 minutes?

grok summarize to 20 words or less

[tender river]

not agree with all interesting anyway

[tender river]

me take back after all

https://cdn.discordapp.com/emojis/1450560641411715165.webp?size=48&name=MiyagiCroco

[sage crag]

croco

[tender river]

me need

[languid bough]

Anyone know of a Greasemonkey script to auto-block people in Twitch chat, who spam repetitive stuff? The old AntiKappa script is from like 9 years ago.. doubt that still works.

[tender river]

entire chat blocked

me recommend use #livestream-chat instead

[languid bough]

@tender river ya, probably, but never know. I remember using something liek that, that worked quite well, at one point.

[tender river]

no idea sorry

[rough bloom]

not Greasemonkey but FrankerFaceZ has a few addons to help with this

Declutterer and Emote Less Chat mostly I think

I don't use them though

(this technically doesn't block people but it does remove the messages)

[languid bough]

oh okay, I'll look into that too. Thnx

[quick condor]

yah I honestly dont understand the spam... it makes chat completely non functional.

[languid bough]

herd mentality maybe. people just wanna participate, and the culture of twitchchat has just sorta devolved to spam lol

[sage crag]

why need

my croco

[tender river]

share

[sage crag]

ok

you can have

half

cut in half

[olive sable]

[quick condor]

I mean I get it when there is something big but like... all the time?

[tender river]

work for me please confirm lack of warranty of fitness for a particular purpose 🔪

[olive sable]

thats just how it is for big streamers. you cant expect anyone to hold a conversation at 200 messages per second

[sage crag]

me not give any confirmation

[tender river]

me not breach contract law you can cut instead

[sage crag]

its cake

[quick condor]

Claude buit this pretty quick. its nowhere near perfect but thats like 2 min since starting it and chat ground to a halt

up to 1200 now

[languid bough]

Oh neato. I also just found a script called "Twitch chat cleaner" that seems to do somethign similar.

[tender river]

who bake

[sage crag]

croco

[quick condor]

nice. Yah for something simple like this NGL, claude threw together something functional enough

[languid bough]

That's awesome!

[tender river]

https://media.discordapp.net/stickers/1349755103040503868.png?size=160&name=AtorinConfused

croco bake croko

[quick condor]

3500 removed in 7 minutes... thats sad

[sage crag]

not paradox just recursive

[tender river]

does recursion continue or

stop at croco

[sage crag]

recursion limit reached

[tender river]

ye cut 🔪

[sage crag]

MiyagiCroco

[tender river]

https://cdn.discordapp.com/emojis/1450560641411715165.webp?size=48&name=MiyagiCroco

[quick condor]

I think that the message filtered stat isnt updating correctly as its updating too fast

this is actually readable

[glass flower]

wait... you don't have 7tv? no wonder you have issues reading chat

[quick condor]

I dont usually watch twitch. and emotes are just a part of why

[patent shard]

I have FFZ, it's kinda buggy
wondering if I should switch

[amber fractal]

It reminds me of someone saying that all big streamers have this issue, it is just emotes. Smaller streamers have a readable chat because it isn't just reactions all the time.

[amber fractal]

Considering someone was trying to use AI to filter down chat, seeing this work well makes me question why that approach is even being taken. Just void all of chat and show a collection of the words that were said in that time frame.

[silent cloak]

There are also projects to do the opposite

[amber fractal]

The real question is if any project just moves them to two separate windows.

[silent cloak]

Plenty

Chatterino even has that

[amber fractal]

That makes sense actually

[languid bough]

hah if that ends up working well enough, might be worth posting somewhere

[sage crag]

you should try and remove messages that are

punctuated with emotes like this

[amber fractal]

also fetch the 7tv emotes to add to the blocklist

[quick condor]

I added that

That was 1 hour

[faint sandal]

do you not like my jah lerolero0 EDM juh EDM

[sage crag]

not as many as i was expecting from 15k concurrent viewers

[quick condor]

Im like 80% sure it didnt count right... It sometimes struggled (blank gaps in chat)

[fierce light]

henlo

any c++ people here?

[young plover]

henlo

am c++ people

[fierce light]

can you compile stuff for me if I give you the instruct. pls

[young plover]

eh, depends

[fierce light]

and then upload the binaries to mega or smth

I have no clue about cpp, but I have no more disk space to get the entire VS cpp package to try all of it

and I have to have some cpp components in my python venvs somewhere that are interfering somewhere because VS can't fucking find any compilers

[young plover]

Ah, running Linux so I can't anyways.

[fierce light]

dang

[jagged turtle]

[jagged turtle]

thx 4 the ping

[jagged turtle]

https://github.com/VedalAI/NetCord another discord .net library fork

[tender river]

@sage crag i implemented like 50-70% of the new vm (in js), for fun asked deepseek to implement the other part, it started writing a step by step plan and just gave up after writing 100 steps

just like me frfr

[jagged turtle]

(why vm in js)

[tender river]

because i need circular structures

and also js is like the least boilerplate when you need an imperative language

i dont really care about performance at this stage i care about understanding what i'm even doing

ye anyway this attempt is promising js made it easier than c++ but i still need to sleep on it i think

[silent cloak]

i dont know why the inlined while loop looks so cursed to me whenever i see it in a language

[tender river]

c# has it

[tender river]

wrrr me think lazy evaluation make thing harder

i guess i'll just give up on lazy if it wont work not big issue since it only for interactive mode which may benefit from eager evaluation anyway

for batch non-incremental processing me not need any of this

laziness good for performance but at least incremental computation mean only have to do extra work once and conversely eager eval mean better performance on certain incremental changes (because it's essentially partial evaluation)

[silent cloak]

i know but using it in C# is vile

its like putting a whole class on a single line

[tender river]

idiomatic c# take lot of vertical space me not like

[silent cloak]

honestly i prefer vertical space to horizontal

[tender river]

horizontal -> can find thing you want by just looking
vertical -> have to find thing you want by looking AND scrolling
more work

[silent cloak]

idk my brain just runs faster vertically

part of why i dont like var either in a lot of cases is seeing the type lets me infer things before i finish reading to the end of the line

[tender river]

it because

me not care about low level details me care about high level overview

[silent cloak]

ah

yeah that makes sense then

[tender river]

i dont really have to know how a particular loop or if works if i already know what it does

and if i really do i can just look, its just less useful to me

[silent cloak]

my brain just works like stepping with a debugging tool i suppose

[tender river]

<msg above was deleted>