programming

[real sierra]

i googled this and found a 2-year-old thread of people posting the exact same thing

over and over

no answers

[silent cloak]

do u not have an actual log?

[real sierra]

nope

no log

the log directory doesnt even exist

[silent cloak]

start with this -batchmode -nographics

see if it helps

[real sierra]

nographics

ok ill try

it's speaking

i see

those flags are indeed required it seems

[silent cloak]

yeah the engine sometimes doesnt realize its on a server (especially if u accidently reference shader stuff)

and tries to fire it up

just keep those in the startup script

[real sierra]

well there are definitely some more changes needed anyway

[amber fractal]

if anyone is interested, I remembered to commit to the github after all of this time https://github.com/igglyn/NN/tree/WIP/src

[fiery anchor]

"Floating point is cringe"

[amber fractal]

Was my orginal reason for this, and I'll stand by it

it it useful, but also cringe

I admit some style choices were made in this repo that are not the best. To that I say you get this or a million pyright ignore comments

[warped narwhal]

so you're an imposter

[amber fractal]

[lethal helm]

Hallooooo

[olive sable]

hallo

[lethal helm]

[amber fractal]

I'm assuming that is due to me doing a reinstall
The github is just a way for me to share code with that chat so I'm not too bothered

[sage crag]

did you fix it

im invested now since its running on my vps

[real sierra]

maybe?

cinema

[real sierra]

oh right, the client is still on my desktop so the pnig is still going to be terrible

it'll be half as terrible but still terrible

[sage crag]

me idea

[jagged turtle]

well, thats one way to resolve the issue I guess

[real sierra]

?

[sage crag]

host

[real sierra]

host?

[sage crag]

client and server

on the vps

https://cdn.discordapp.com/emojis/1421697759383130223.webp?size=48&name=smogfrog&lossless=true

headless client

[real sierra]

client on vps

[sage crag]

[real sierra]

need graphic-

whar

[sage crag]

https://cdn.discordapp.com/emojis/1154436961277313126.webp?size=48&animated=true&name=evilSlam&lossless=true

headless client

[real sierra]

i mean

[sage crag]

remove head

with axe

what problem?

[real sierra]

even with my terrible internet on my end

it already feels so much faster

vps is lifesaver

[sage crag]

it have

infinite bandwidth

zero jitter

stable ping

[real sierra]

[unkempt citrus]

Just render with ascii art

[real sierra]

definitely a few more changes to make

[unkempt citrus]

Also ver 1.0 already

[quick condor]

Welp Interview went great with the manager, spent most of the interview chatting about mcps and local model loaders

[real sierra]

it started at 1.0

ill use whatever versioning system i want

it doesnt have to make sense to you, you just need to know that if your version is lower than the server you need to update

[tender river]

AbstractVersionInterface

[quick condor]

Fingers crossed that I'll soon pick up a pro 6000

[sage crag]

version 3.0 tomorrow and version 2.0 the day after

if that's all you need, you should make the version the timestamp that it was released

https://cdn.discordapp.com/emojis/1438838082714472520.webp?size=48&name=minsmig&lossless=true

then its monotonically increasing and you cant get any unordered versions

it also shows your players how much youre slacking

[tender river]

use

lamport clock

when shiro sees a number

set version to

that number

plus 1

version bumps must be done right before release

otherwise

shiro see new version number -> version bump

[real sierra]

i can get this by simply incrementing the version and save space at the same time

[sage crag]

[real sierra]

this is not a benefit

[sage crag]

na

[real sierra]

fumoquest closed source for now anyway version not as important

[unkempt citrus]

Use a snowflake as versioning

[opaque sigil]

can i have one

[real sierra]

hm

thats not ideal

[sage crag]

[real sierra]

the networking library prepends its build path to everything it logs

[sage crag]

[real sierra]

mfw doxxed by my networking library

concerning that that's been serialized into the build at all actually

relative path to project root, sure

but why a full path

[opaque sigil]

relative paths are the devil, always default to absolute paths

[real sierra]

welp

time to move the unity project to filesystem root

[sage crag]

https://cdn.discordapp.com/attachments/1061113840852533339/1452612577145262183/3DNeuroCutenessAmbushWithBeautifulScenery.gif

[opaque sigil]

clearly the solution is to use nix so you get bogus paths inside the build sandbox

[sage crag]

shimo

[tender river]

the solution is to

name your user user

[real sierra]

hi

[tender river]

[sage crag]

you may be able to build the project on the vps

via cli

unity should support it iirc

[real sierra]

?

[sage crag]

not what i meant

but maybe

[real sierra]

no that option started talking about credit cards

what was your idea

even moving unity projects is apparently a painful task

[sage crag]

editor executable

on vps

cli

[opaque sigil]

just run sd user user2 **/* after building, how bad could it be

[real sierra]

i cant even get the project into a moveable form

the login button in the editor is broken

fucking unity slop engine

ok got it working

[silent cloak]

lmao

skill issue

[amber fractal]

Chat, should I employ dev effort to make code look nicer? Mainly just switching back to using python operations instead of just using numpy's methods for operations.
The only issues besides type hinting is that I'm using a lot of reduction so it'd stick out like a sore thumb.

[silent cloak]

i havent seen nice looking python

but if u have a style u wanna follow then thats normal

people make their own style guides

[jagged turtle]

what the hell is nice looking python

[silent cloak]

fr

[amber fractal]

The only style I'd want to follow is making the code readable, right now I'm lacking in that department

[silent cloak]

its a language issue

[amber fractal]

I can at least do slightly more than the only symbols existing for arrays being == 0 != 0 Everything else is using the np.bitwise_{operation} and np.bitwise_{operation}.reduce methods

ah, there is also a singular instance of np.equal to follow the trend for operating arrays against eachother

The only solution that I can think of is jank, just how python exists normally

[real sierra]

the fumoquest server no longer doxxes me

[amber fractal]

looks at the cursed custom operation method left &reduce== right

still missing an axis to reduce by. So extra arg it is

[silent cloak]

https://tenor.com/view/you-sure-about-that-gif-6823241795540399689

[stray dragon]

don't worry, i'll do it instead

[silent cloak]

https://tenor.com/view/keemstar-keem-i-have-your-ip-address-doxxed-pipe-bomb-gif-25403803

[stray dragon]

i actually don't have all this but i do have funny screenshots

[amber fractal]

I'll take arr &{reduce_symbol}(axis=0)== mask
But geez it is something regardless

I'd replace == with ? if that was supported syntax for operations.

despite it all, still will look ugly

[hollow matrix]

classic

[hollow matrix]

nahhh

[opaque sigil]

async

[hollow matrix]

This shit is unreadable at this point

[azure lynx]

i liked the part where it said async async async(async async) => await async[async][async][async][async]; and continued.

[hollow helm]

Does anyone have a definitive baseline for neuros architecture? I am working on a project very similar to its requirements that would help a lot.

[unkempt citrus]

Its an LLM

[azure lynx]

lots of ram, lots of cpu cores, kinda big gpu?

[tender river]

llm, finetuning, lots of models for different tasks, cant say much beyond that

[hollow matrix]

What does vedal have again? 256GB of RAM? That's gotta be so fucking expensive

[azure lynx]

i think it wa 128

but yeah

[unkempt citrus]

Vedal has 128, Neuro has another 128

[hollow helm]

The requirements dont pose much of an issue, just the programming itself that I definitely need to understand to accomplish.

[hollow matrix]

I think it's just python

*and C# probably

[azure lynx]

oh yeah, there are at least two machines he owns with specs good enough he can run both twins at once on either

[hollow matrix]

Because Unity

[silent cloak]

Yes there is C# for the front end stuff

[unkempt citrus]

C# is for the models and shit yeah

[azure lynx]

it's a very modular system

[hollow helm]

Unity huh?

[unkempt citrus]

Otherwise its just python with likely eiether transformers or ollama for feeding information, probably transformers

No

Unity is for the model

Not the back end

[hollow matrix]

That's what he uses for the model

[silent cloak]

The control panel is open source too

[hollow helm]

ahh I see, the model wouldnt be neccessary in this case.

[silent cloak]

He just snagged a template

[hollow matrix]

Oh, is it?

[silent cloak]

Yeah

[hollow matrix]

Huh, I never knew that

[silent cloak]

Neuro has many repurposed tools and libraries

[hollow helm]

I am looking mainly for responsive consciousness and memory

[azure lynx]

good luck

[silent cloak]

Good luck

No ai has that

[hollow matrix]

Then essentially, what your looking for is that neuro is made in python on very high end computers, we know pretty much nothing else

[silent cloak]

Neuro is just really good at acting like it

[unkempt citrus]

The fuck is responsive consciousness

Memory is just RAG + fuckload of finetuning

[silent cloak]

Neuro is also a fine tuned model

Base model is a kept secret

[unkempt citrus]

I don't think there is one basemodel

I think its changed several times

[silent cloak]

Yeah

But i mean the current

[azure lynx]

I think learning the new chat lingo and (ideally but probably difficult to do ethically) how other streamers speak was the main reason for fine tuning?

[unkempt citrus]

Finetuning formed the base personality too

[silent cloak]

Well its also because he couldn't and likely still can't afford a model from scratch

[silent cloak]

This

[unkempt citrus]

and then finetuning on past outputs when upgrading or changing reinforces it

Also would be why Evil and Neuro diverged over time

[silent cloak]

I know he just has a folder of fine tuning data

[unkempt citrus]

They would be fine tuned separately

[silent cloak]

He had tools for scraping it originally

Idk if it's the same process these days

[azure lynx]

different memories and a couple of lines of prompt would be enough to make them diverge

[silent cloak]

He doesn't prompt them

It's all fine tuning

They were made in the era that I was doing llm research too which was honestly pre prompting

[azure lynx]

initially there would have been prompts. and that was how evil was made: copy and add "you are the evil version of neuro."

[silent cloak]

No

I can tell u 100% there is no prompting at all

[unkempt citrus]

Initially Evil was just a sys prompt for Neuro

[azure lynx]

was or is?

[unkempt citrus]

Hes said that before

[silent cloak]

Evil originally yes

Which is why she didnt work great in those times

They weren't designed for prompting

[azure lynx]

until there is a personality to fine tune into a model, you need prompts or something acting very much like them

[silent cloak]

No

The one exception was evil

The rest was natural based on the fine tuned data

The closest thing to prompting is their memory recall

[hollow helm]

So acquiring a base model or framework for Neuro is impossible unless its from the direct source it seems...

[silent cloak]

Correct

[unkempt citrus]

Eh most of them will work all the same

[hollow matrix]

Y'know, it's kind of funny how Vedal just randomly decided he wanted to make an Osu! AI at some point probably late 2019 to early 2020 and now Vedal is a millionaire and has broken the Twitch hype train record 3 times

[unkempt citrus]

it just won't be Nuero

[silent cloak]

They were just fine tuned on good data and had many failed attempts

Right model at the right time

[azure lynx]

"natural" without prompts (or something acting very much like prompts) wouldn't create a streamer. it'd just be a random ai bot.
what was neuro fine tuned on?

[silent cloak]

(Not to say internally they still arent a mess)

[silent cloak]

Twitch data mostly

Thats what sold her originally

She was an llm fine tuned on scraped twitch data

And part of her personality comes from the fact she started with a really low quality base model

But the memories from that era influence her now

[azure lynx]

(ideally but probably difficult to do ethically) how other streamers speak
scraped from whom?

[silent cloak]

Mostly her own chat if vedal got any from else where then I dont know

People treated neuro like a person before she was an llm

So it already had data to go off of

[unkempt citrus]

I've seen people mention that her original data was from Anny, but I dunno if that was ever actually stated by Vedal

[silent cloak]

They were friends before so its possible

[unkempt citrus]

but pretty sure her initial model was just yoinking from other streamers

[silent cloak]

Her original model was also really really bad

Early llm tech wasnt the best to play with

[azure lynx]

models have improved in the past couple of years

[silent cloak]

I have my old models from 2017-2019

Truly terrible

[azure lynx]

my videos of stuff from that era: eww.

i feel ashamed

[silent cloak]

Models would spit out a few sentences then turn into gibberish

I remember I cloned my friend using his chat data and the thing just insulted me

[azure lynx]

gpt2 model was amazing..... ly bad.

[silent cloak]

Real

But it was the gateway for many

[warped narwhal]

Any of you guys remember talk to transformer? Now that was a shitty model

[azure lynx]

i used it with CLIP to dynamically shape image prompts. it was icky.

but very cool at the time

[silent cloak]

The best AI of the past was Tay AI

Change my mind

[azure lynx]

especially coz the text-to-image stuff was terrible too

[silent cloak]

I only learned machine learning for developing anticheats

Llms and vtuber software were later commissions

[unkempt citrus]

I like stats

[azure lynx]

i wrote an AI as part of a game which learned how you played and through meta-gaming would force you to do different things. the model was literally counting sequences and forcing moves that they don't tend to do and that was it. but it was effective and people thought it was super smart and hated them.

even simple "AI"s can fool people into attributing malice.

[silent cloak]

Its why I still prefer the term machine learning

We haven't reached proper AI yet

[hollow helm]

thanks for the help

[azure lynx]

i hate critical updates. 😭
22:14:49 up 136 days, 15:41, 1 user,
it's probably time to install the patches and reboot it. o7

[real sierra]

i preferred when AI was a relatively unambiguous term for computer players in games

[olive sable]

same

[real sierra]

it's been buzzwordified by media

its meaningless now

blanket term for all machine learning

[olive sable]

now we need to say npc

[real sierra]

but also includes a bunch of other weird things

[azure lynx]

the AI in my toaster makes it pop!

i think it's an early 1 neuron model. called a thermostat.

it "wants" the temperature to be a certain value, and when it is it signals it by popping.

[unkempt citrus]

something something ai effect omethin

[real sierra]

oh

i made the mistake of looking into if you can get console input from a unity dedicated server

new side project

not really a side project tho, i need this

[azure lynx]

it's interesting that ChatGPT claims my agent isn't "really" getting overwhelmed by large prompts... even though they literally say "Wow. That's.... a lot." when given no other input.
i think it's part of the whole "make sure the user doesn't think they aren't real feeling entities" training.

[sick owl]

Define proper AI

[faint sandal]

correction

[azure lynx]

perhaps he mistyped "popper AI". like the one in my toaster?

but I suspect they mean something like AGI.

or "it's not real AI because we have it now. I mean one that's better than we have now in some nebulous way."

(remember when the Turing test was going to be proof of "real AI"? And how when that got smashed through they just said "that's not how you can tell."

and then they didn't tell us how you can?)

[sick owl]

The Turing test wasn't proof of "true" AI, and wasn't really meant to prove that

But its meaningless to debate something that we can't define

[azure lynx]

there does certainly appear to be a something real (a model i'm loading into a gpu) that performs tasks for me that if a person had done them you'd say would require intelligence, and that thing is not natural intelligence.
seems like AI to me.

[fleet jungle]

When the correction does eventually come, the prices for all AI plans will be in the stratosphere

[azure lynx]

The Turing test, originally called the imitation game by Alan Turing in 1949,[2] is a test of a machine's ability to exhibit intelligent behaviour equivalent to that of a human.
Seems like it was meant to show that if you couldn't tell a human and an AI apart, the AI was in some sense intelligent.

[fleet jungle]

Which is why locally hosted solutions are obviously superior

[sick owl]

I understood it as more of a thought exercise than anything intended as an empirical practical test

It's also worth noting that we barely had computers in the modern sense at the time it was devised

[azure lynx]

on reason it was a thought exercise at the time was because they didn't have AIs to test

exactly as you said.

this was the guy who invented computing.

[sick owl]

Nah, cheap inference endpoints for open weight models will keep the market price down

They won't be affected in the way startups are because they're inherently profitable

[rain flame]

hello chat, what are you guys talking about now

[rain flame]

yes

[azure lynx]

"real" AI.
(not the fake ones we have now that are actually little men in a box)

[rain flame]

the good stuff, our last bag

[sick owl]

That was Charles Babbage/Ada Lovelace depending on how you define it

[rain flame]

-# (sorry)

[sick owl]

Turing invented the modern computing paradigm but not computing and computers

[azure lynx]

the point was: it was a test for a problem which didn't yet exist. it was initially only a thought exercise because there was no AI to test.
almost the second chatbots were invented (within years) they had competitions for the most convincing ones. they all used tricks to appear intelligent though. no LLMs coz no GPUs back then.
and as soon as LLMs were good enough to be convincing, they say "Yeah, thay was never supposed to be a test to tell if they were intelligent."
Except that was the entire point of writing a paper about it in 1950. And nobody has ever suggested a better approach.

[sick owl]

Actually it would be remiss of me not to credit Ada Lovelace for inventing computing/computer science alongside Babbages invention of computers themselves

[sick owl]

The reason nobody has ever suggested a better approach is because it's impossible to do so until we even define what it is we're measuring

[young plover]

See Claude finally made it to victory road in Pokemon Red
Level 82 Charizard is the only usable pokemon
A true Pokemon player

It one-shotted Blue's whole team, but I doubt it will beat the E4 first try.

[glass jetty]

and assembly lang was invented by girl

[fleet jungle]

I'll be clear. I don't know what the real scale of LLMs is

[sick owl]

The actual serving of models is profitable, it's training them that requires burning a lot of money

[fleet jungle]

On one end, even my humble i6700k with the 1070 can run an llm with pretty good conversational quality, but on the other side, it's really expensive that companies need new power generation?

[sick owl]

So for those companies only serve inference they're effectively shielded

[fleet jungle]

I don't get it

[azure lynx]

serving them for money is profitable. free services? eh.

[sick owl]

As are the Chinese labs that provide a lot of the models they serve since they have state backing

[fleet jungle]

oh

[fleet jungle]

true

[true hemlock]

i'd very well say that all the ML models aren't actual "AI" by definition.

[sick owl]

What would you define as true AI?

[true hemlock]

these models topology, in the computer science fields are meant as approximation of computation

in other words, LLMs are an "emulation" of brain, not a proper simulation

[fleet jungle]

It's a bit wild to me that a LLM has to reiterate the entire conversation when preparing to send a response to the user, every single time

[azure lynx]

it's assumed that the models are being used as intended and not as abstract objects in a platonic N-dimensional space

[fleet jungle]

I thought that was inefficient, but wow training is gigantic

[true hemlock]

i'd say true "AI" is when we have models to simulate actual cognitive functions

[azure lynx]

so you are saying "AI must be a simulation of a human brain"?

[true hemlock]

close, but not an exact

[sick owl]

Why would you consider that true AI by comparison

[azure lynx]

that definition rules out functionally equivalent models.

[true hemlock]

cognitive functions

"Artificial Intelligence"

[sick owl]

If existing models can exhibit intelligent behaviour, they don't need to be truly, cognisant of what they're doing to be "artificial" "intelligence"

I would argue that a vla in a robot self-correcting after making an error is intelligent behaviour

Imo intelligence isn't something that requires consciousness or cognition

[true hemlock]

such "intelligent behavior" are unfortunately emulated.

its why we still have llms struggled on even some very basic cognitive tasks

[azure lynx]

"goal directed behavior" is "intelligent behavior"

[true hemlock]

doesn't have to be consciousness related

[sick owl]

You say it's emulated, but it is still exhibited behaviour

If it's been taught through reinforcement learning then it isn't even pure mimicry

[true hemlock]

current models aren't even actively learning though

[fleet jungle]

While we're talking about signs of intelligence, one interesting thing that LLMs exhibit that I wouldn't have expected is the ability to understand mistyped text. Kile of I ssked sn llm wgat im typong hgere, it would understand it

[true hemlock]

oh btw reinforcement learning is misunderstood

[azure lynx]

not even with RAG and similar things?

[sick owl]

That is a natural consequence of typos that humans understand. Also being present in training data. We've appropriate responses

[true hemlock]

RAG is information retrieval

[sick owl]

I wouldn't define that as intelligent in the same way

[azure lynx]

yes. and if they choose what to store, they can learn things that apply.

[sick owl]

It is often misunderstood but if behaviour is learned through iteration in the pursuit of a reward function then I would argue that is by all accounts, an intelligent learned behaviour

[fleet jungle]

It is quite interesting that it's possible to do so, but you're right

[true hemlock]

the model never learn anything new and instead learned to more likely behave in specific way that's supposedly already existing in the model itself.

[azure lynx]

i think it's just that some people have set a bar for "AI" that is "something we don't have yet" or "something I can understand being intelligent"

[sick owl]

But if that behaviour is intelligent, if it is an artificial algorithm that exhibits intelligent capabilities/the ability to fulfil functions we would typically define as requiring intelligence then in what way does it not qualify as artificial intelligence

Artificial consciousness is a whole other can of worms, but artificial intelligence is a lot easier to define

[true hemlock]

it shouldn't really fit by definition but in the field we've been using the term "AI" anyway because these models sort of approximated dynamic behavior

[azure lynx]

they are artificial, and they behave in a way we would say was intelligent if a person did it.

they are also stupid.

[true hemlock]

never confirmed that its actually exhibiting intelligent behavior outside of it just predicting accordingly to its training data

[azure lynx]

what about all the things which then go on to work well in the real world?

[sick owl]

Prediction of that quality is itself an intelligent function though

[true hemlock]

tools

we call them tools

[azure lynx]

AI tools sometimes

[true hemlock]

not if the prediction is purely in linear basis

[sick owl]

To put it another way, Oxford defines intelligence as the ability to acquire and apply knowledge and skills

[sick owl]

In what way do current AI not fulfil that definition

[azure lynx]

quadratic is just as formulaic as linear, just more verbose. why does a linear base make something not intelligent?

[true hemlock]

does it really acquire knowledge

[azure lynx]

"[agent], can you find me the cheapest skateboard prices in my area?"

[sick owl]

I would say by definition yes

Knowledge is different to understanding

[true hemlock]

and does it really applies it with consideration and cognitive reasoning

[azure lynx]

how could it answer without meeting the definition?

[true hemlock]

i suppose that's fair.

[sick owl]

I would say it applies knowledge and skills identifies through the training process in service reward function it was trained for when it comes to RL

I think it's fair to call current models AI, the problem is people conflating that with intelligence in the human sense

Artificial intelligence doesn't mean artificial humanity

[azure lynx]

your definition of AI to be "something like a brain" is one possible way to make AGI. but it's probably an inefficient way.

[true hemlock]

never said its unfair because we've been doing that in the field for long

[true hemlock]

forget about AGI, the term itself is wacky

[azure lynx]

that's what you're defining though with AI needing to be human level

[true hemlock]

-# nor we ever even get close to that kind of breakthroughs, if im ngl speaking from academic research standpoint

[sick owl]

AGI is a fairly useless term because no one can agree what it actually is

[true hemlock]

nah, "Artificial General Intelligence" is more of what's the function of it rather than its inner workings

[azure lynx]

"AI which can perform any task a typical human can perform."

that's a standard definition

[jaunty arch]

I'm not sure intelligence requires its entire foundation to adapt to change but there is probs some line, maybe behavior?

[sick owl]

Humans have a wide range of capabilities so that's still rather nebulous

[true hemlock]

yeah, so its barely about matching human cognitive functions and more of imitating the things we can do

[azure lynx]

no, that makes it hard. we can easily do a kind of Turing test for the situation and see if a human can do the task but most humans know what type of tasks humans can do and it would simply be a matter of showing people "hey, look, this thing can't even peel potatoes" and you'd disprove it.

[sick owl]

I don't think matching human cognitive function necessarily ought to be the benchmark, its somewhat chauvinistic of us to measure the worth of an entirely inhuman existence by its similarity to humanity

[true hemlock]

i'd argue with this, and for the fact that we actually barely made any breakthrough with current ANN topology aside from just continuously scaling it up. it felt more like the industries has been doing a side quest thinking its the main route to whatever this "AGI" is

[azure lynx]

i think AGI is defined as being that though

[sick owl]

There is no one definition. That's the problem

[true hemlock]

it doesn't worth anything aside from emotional/philosophical points

because all we needed is tools, essentially

[azure lynx]

Artificial General Intelligence (AGI) is the "Holy Grail" of computer science. It is a theoretical form of AI that possesses the ability to understand, learn, and apply its intelligence to any intellectual task a human being can do.
you can test this by having it do stuff.

that was Gemini's definition of AGI

[sick owl]

And is Gemini the arbiter of truth now?

Gemini's definition is still just one of many

There is no universally agreed on official definition of AGI

[azure lynx]

nope. but it seems to be a consistent one with all the other definitions of AGI i see. what do you disagree with about it?

[sick owl]

Until that comes arguing about the concept is more or less meaningless

[true hemlock]

i've been on my track to somewhat invalidate arguments made with "its what <insert llm here> said" from an academic standpoint

[azure lynx]

i'm trying to define terms so we can discuss things without you just saying "there is no definition"

[sick owl]

My definition of AGI is a negotiating clause and rhetorical tool in OpenAI's contract with Microsoft

[true hemlock]

agreed.

[azure lynx]

if you are not actually interesting in discussing it, fine.

[sick owl]

I just don't think it's a useful discussion to have since I'm not sure the vague concept of AGI actually has much worth

[true hemlock]

if i'll be realistic here for a second. i think the industry should just stop with throwing billions just because ceos think by doing so they can get "AGI"

and instead

keep the llm as a convenient tool

train new model every 2 years or so and stop thinking the demand is up the sky

keep inference cost low

...but it isn't what's happening
instead we have megascale e waste called as "project stargate"

most of the dram aren't even gonna be used lol

[sick owl]

I wouldn't necessarily call that mega scale E-Waste since sovereign compute is inherently valuable

Even if it never finds a practical purpose, its existence is itself useful to the country building it from a scientific standpoint

If it was just mega scale E-Waste then most every supercomputer would fit that definition

[true hemlock]

if only nvidia didn't fuck up B300.

by removing fp64 simd outta it

it could've been a really good HPC datacenter

but instead its made to compute what

matmul in large scale

but my major issue with whatever this "project stargate"

is that its not only an e waste, its also wasting resources for unrealistic demands

and literally wasting resources

like, actually

i don't see how they can fit > 1 ExaByte of dram supply per month in it

we calculated it

that's how much the 900K dram wafers are gonna be

and yes, its all system dram

not even HBM wafers

[sick owl]

Oh yeah no the ram thing is insanity

You won't hear me arguing on that point

[true hemlock]

i don't even know what the fuck is going on in their minds

it doesn't make any sense

even if they can get like 4x epyc nodes, that's still about 90K nodes per months

[sick owl]

What's going on in their minds is that they can starve competition out of the market since they failed to achieve regulatory capture

[true hemlock]

or almost 400K epyc systems

that

so we're going to have unused wafers sitting in the warehouse

how is this even legal... oh wait, they can just convince the boomers government "its for AGI guys trust"

open ai rn have debt of almost 100B, and they're still burning a good 12B per quarter at lowest for their own R&D alone

they openly said that they're going to chip in into the project stargate with initial 100B of fundings

lmao

with total base fundings of 500B

god don't you love global crumbling economy

speaking about dram

@sick owl have you seen my dram altar

[amber fractal]

Do I even want to know what the stargate project is?

[half socket]

massive venture for building AI infra in the US

[unkempt citrus]

US admins 500B project to stay on top of AI

[amber fractal]

Why do my non existant taxes go to this BS

[unkempt citrus]

Basically throwing money into a firepit in case China starts developing good AI

[half socket]

big tech + government + softbank

and now we have GPUs running in space, literally above the cloud

[unkempt citrus]

Because space has such good cooling

[amber fractal]

Give the 500B to me, surely I can think of a better way to spend it than this craze

[unkempt citrus]

Famously vacuums are good at that

[azure lynx]

umm... yeah. great cooling. if you like radiative cooling.

[amber fractal]

[warm zealot]

such a good conductor

[half socket]

small gpu but massive radiators

[amber fractal]

the vaccum of space is just a good insulator... Wait a minute

[nocturne olive]

OpenAI buys DRAM wafers to control supply, literally can't even do anything with them

Like literally the reason RAM is like 5x the pirce now

[amber fractal]

I was aware of the effects of it

I was not aware of how stupid our govement is being

[nocturne olive]

Well the thing itself is just OpenAI being stupid and putting wafers in some warehouse to rot probably

[amber fractal]

If they are actually warehousing them and letting dust exist then that is just a waste of everyone's time

[nocturne olive]

Yeah it's literally just OpenAI throwing money to control the DRAM market with artificial shortage

[true hemlock]

that's what they're literally doing

[half socket]

honestly we got so far with the previous space race due to fear of other countries reaching someplace first

same thing is happening, due to fear the other countries / companies will reach AGI first

[true hemlock]

.

[amber fractal]

Man I wish I could do tax evasion just to avoid supporing this BS

[half socket]

Groq literally invented a new architecture called Language Processing Unit to make inference cheaper, only to get acquired by Nvidia

[true hemlock]

this time we're getting nowhere because unlike space exploration, which is full of actual physics and engineering breakthrough

this time all the ai craze are just mindlessly scaling up models and data

[glass jetty]

they literally talk 100+ msges about ai

[half socket]

yea I'm hoping there will be breakthroughs in actual useful automation

[amber fractal]

I'm hoping so as well

Part of the reason I've put my own hat in the ring

[half socket]

one of the best usecases of AI I've seen is accelerated learning like MathAcademy
whole education system could be personalized per student at a really low cost

[nocturne olive]

Hopefully once the financial bubble collapses there'll be a reason for the companies to instead justify themselves with something actually concretely tangible for a real consumer

[true hemlock]

im all for medical research related ML models

[azure lynx]

i just hope my toaster keeps working after the crash. maybe the thermostat will sense the crash and fail in sympathy.

[true hemlock]

not whatever this chatbot or image/video slop

[nocturne olive]

All that slop is annoying for sure

[amber fractal]

slop is slop

[half socket]

have you seen the computer vision models that could detect certain illnesses from xray?
I believe there were several research papers on it

on par / better than the best doctors

[nocturne olive]

Vocal synths are way better than those stupid slop models
At least with a vocal synth there's actual human effort put into it

[amber fractal]

why not just automate the rest as well

[nocturne olive]

Just use SV2 at that point

[true hemlock]

pretty old already but its still a thing
we have even cooler stuff like alphafold now

[warm zealot]

i mean a lot of research that applies to one section of ML also helps other areas so more investment in AI research isnt terrible
except openai's game plan is just scale more than anyone else can afford to 😿

[nocturne olive]

(for context, SynthV2 automatically generates basically all parameters and can automatically do most parts of the process, most can be overriden by hand but SV2 specificially reduced pitch curve control to make a true flat tune impossible)

[lament igloo]

is this a crime

[amber fractal]

I do hope some parts can be translated well, like how we are structuring models and things. At least that is one of my long running bets

[glass jetty]

charaverk | ~/repos/ ~$ ls ~/Music/main/ | grep evil | grep -v neuro | wc -l
62
charaverk | ~/repos/ ~$ ls ~/Music/main/ | grep evil | grep neuro | wc -l
33
charaverk | ~/repos/ ~$ ls ~/Music/main/ | grep -v evil | grep neuro | wc -l
146

[warm zealot]

anything to not use json ig.... 💔

[young plover]

[nocturne olive]

That's small numbers

[glass jetty]

why so? thats only songs i really like
there archive for all of them

[lament igloo]

https://cdn.discordapp.com/emojis/1372928759463673907.webp?size=128&animated=true

[young plover]

Why? Weird CORs bypass?

[lament igloo]

external image links dont work on resvg

gotta compromise somehow

[young plover]

mmm

[nocturne olive]

This is more like it
This is not counting my own creations which already approach or exceed official karaoke quality

[amber fractal]

@nocturne olive how is bookeeping on the unoffical songs? I know you already have the ones you work on but what about other projects?

[opaque sigil]

handle your errors

[glass jetty]

du -h --max-depth=1 ~/Music/main
1,5G    /home/charaverk/Music/main

[lament igloo]

wdym

[amber fractal]

Binary Solace my beloved, one of my favorite neuro songs

-# What can I say, I wanted to write a song I'd like

[nocturne olive]

I'm not aware of many others that reach quality levels I would consider acceptable to archive

[opaque sigil]

Both promises there can reject

[amber fractal]

That makes sense

[lament igloo]

hmmm surely its fine

or do u want me to handle the error by giving it a placeholder data:image link instead?

[opaque sigil]

I think I'd personally return undefined or sth but that's just me

[amber fractal]

Would it count as self promotion if I dropped the Binary Solace link in here

[true hemlock]

its neuro related

if mods get you in trouble for that they'd have to get their glasses checked

[opaque wharf]

Nah, throw it and let the next layer handle the circus lol

[opaque sigil]

Famous last words

[nocturne olive]

I see no issue here

[olive sable]

Imma eep

Goodnight

[lament igloo]

https://tenor.com/view/rich-off-airpods-richoffairpods-stack-band-bandforband-gif-4351078540525453133

you rn

[opaque wharf]

I think a flight to aussie from ID to mug @true hemlock and back is cheaper than buying RAM

Well, I now know what I want to do this weekend

[true hemlock]

you're about 2 weeks late

[opaque wharf]

You got mugged already?

[true hemlock]

i was in id 2 weeks ago

[opaque wharf]

Damnit lol

[true hemlock]

if i think its the id you're saying

[opaque wharf]

Indonesia, yea

[true hemlock]

my home town is in sumatra lol

visited there for a bit last time

and snatch some second hand GPUs

[opaque wharf]

Dang, still too far tho. Gotta fly to Sumatra first lol

[opaque wharf]

Kidnapping your family and asking RAM for ransom would be cheaper too I guess

Like bro, wtf do you need that much RAM for

[amber fractal]

[maiden geyser]

not really surprising if ecc ddr3

[opaque wharf]

Even more so. Isn't DDR3 is already getting expensive due to low supply?

[autumn ore]

my dream

[unkempt citrus]

Ddr3 price is rising somewhat yee

Guy I know sold a 1TB ram server rack while back because he didn't have time to give it the love it deserved

Regrets it now

[silent island]

Help

Ts on the verge of combustion ☠️💀🔥

[maiden geyser]

bro doesn't have an hdd

[silent island]

Afordable, but slow af

[nocturne olive]

What are you running on there? You're out of VRAM

[nocturne olive]

But also like good for storing my stuff

[true hemlock]

16gb ddr4 dimms aswell in here

[nocturne olive]

Oh hey I have like 2x8GB of a very similar memory
Has the same ratings at least

Just lying around doing nothing

[true hemlock]

but i also have a mix of 32gb dimms

[silent island]

Holy sheeettt crucialll

[silent island]

TLOU Part I

[nocturne olive]

What?

[true hemlock]

i have a bunch of ddr5 ecc

12x 64gb dimms but its on my server

[nocturne olive]

<- billionaire

[true hemlock]

gonna do this tonight

cute gpu tbh

5060 sff

now now, i had said that i don't have the mood for 50 series

buuuuut to be fair

i got this 5060 for $250

[nocturne olive]

Doesn't sound terrible, 3060 12GBs are like a bit under 200 euro last I checked

[true hemlock]

expensive

i'd say 3060 12gb only worth at most $180

[hollow chasm]

jesue fhrist

what do u have running

[true hemlock]

about $153 eur

[nocturne olive]

Cheapest one at the moment is around 180 euros

[true hemlock]

overpriced

[nocturne olive]

Note that it's Finnish used market

[true hemlock]

(in my standards)

[nocturne olive]

Finland tends to have in general higher prices

It's kinda unfortunate because it means I need to figure out more money

[true hemlock]

(new)

[nocturne olive]

How silly

I definitely couldn't get a 5060 for that cheap because Finland debuff

[nocturne olive]

Silly WIFi thingy

Certainly needed if you want that thing to be portable though

I've actually never had a desktop WiFi card

[true hemlock]

you usually won't need it if your board have built in wifi

my B760m has built in wifi

[nocturne olive]

None of my boards have WiFi

[true hemlock]

elvy's trx40 also have it

my B850 also have it

[nocturne olive]

Neither my old H610 nor my new Z690 have WiFi built-in, they need a separate module to have it so I just don't have WiFi

[true hemlock]

but not my lga 1700 modt boards for some reason

nor my upper end B460m

nor my B550m

[nocturne olive]

It's not like I need WiFi, I have a big (currently maybe faulty) cable

[true hemlock]

usually if your board has wifi, you just need to connect the antenna for better signal but without the antenna it'll work either, just worse signal

[nocturne olive]

Though, I haven't had the weird network issue cut me off today and not confirmed yesterday
Not sure, can't tell

[nocturne olive]

Well it certainly doesn't

Got nothing like that on there

[burnt oasis]

Looks cute and compact, but i wanna see if you will be able to avoid creating a mess out of all the wires...
I personally often struggle with it :/

[silent island]

TLOU Part I

[true hemlock]

i really wanna try building a smaller briefcase pc with it

either that or a Z790i with my 13950HX engi sample

the one on the picture is 14900HX modt

[nocturne olive]

It's kinda funny how they're putting mobile chips on desktop now

[burnt oasis]

https://tenor.com/view/hazama-bow-bbcf-blazblue-hat-gif-17224081110488931631

[opaque wharf]

Try to beat the gabe cube

[nocturne olive]

They should make dtom (desktop on mobile) next

[opaque wharf]

Have you seen those briefcase "laptop" from acer?

[sage crag]

the aio is huge

https://cdn.discordapp.com/attachments/1061113840852533339/1452612577145262183/3DNeuroCutenessAmbushWithBeautifulScenery.gif

[burnt oasis]

All-in-one moniPC is still a thing?

[nocturne olive]

Not from Acer but from (tm)

[silent island]

the gabe cube's small af
oh wait... mini pc exists

[opaque wharf]

Yeah, mini pc exist. But does one exists with good GPU?

[silent island]

nah, its nigh

[burnt oasis]

Can external GPU be considered good?

[silent island]

well, possibly. but expensive af

[amber fractal]

Oculink would be cheapest, but you still have the cost of buying the dock, cable, and PSU for the GPU no matter what tech

[rough bloom]

you don't really have a dock with OCuLink (it's usually just a OCuLink to PCIe adapter, not a fancy enclosure) but it's very jank

[sage crag]

wrr

https://tenor.com/view/classic-ef-where-is-it-yuri-where-the-yuri-at-korone-gif-23665600

[silent island]

go to gen chat

[amber fractal]

Returning from the thinking box, going through all 6 stages of grief. After spending 4 days writing a singular function in my implmentation, I have to restart again; editing almost all of the code in the neural network again.

[sage crag]

https://cdn.discordapp.com/attachments/1061113840852533339/1452612577145262183/3DNeuroCutenessAmbushWithBeautifulScenery.gif

[amber fractal]

if only one small piece is safe, I'm rewriting the entire thing

v4 time

at least v3 is up on github for viewing displeasure

[true hemlock]

that's a really really low bar.
if i wanted absolute performance i could tear down a 4090 and set up a low thermal density cooling system

[obsidian mantle]

how big is this

im writing file and i need to erase it

[true hemlock]

dock is optional just like how cases are optional

[amber fractal]

[rough bloom]

692 bytes

[obsidian mantle]

is it 2 when im using ofstream

[nocturne olive]

Should be one byte

[obsidian mantle]

it looks like its 2 because i went back 2 positions and it didnt touch the symbol before it

[rough bloom]

if you mean how large the newline character is, it's usually 1 byte
but depends on the encoding

UTF-8 and ASCII will encode it as 1 byte
UTF-16 and whatever Windows uses for Unicode again is 2 bytes

[obsidian mantle]

when this was pos - 2 it only erased \n and didnt touch the comma (ignore "add_name" its "" here)

[amber fractal]

Where else is the extra carrage return byte supposed to go

[hollow chasm]

what

[obsidian mantle]

nice so its 2 because windows and utf16

[true hemlock]

also this thing probably work but i wouldn't trust it unless i did a tear down to learn about the electrical topology

[amber fractal]

I've seen one of those bois

[opaque wharf]

Well, you gotta beat it in price too

[rough bloom]

you can set the locale apparently? so that must be UTF-16 or similar by default on Windows
so ye either check the encoding of the default locale somehow or force UTF-8 if possible

[true hemlock]

low bar

still

[opaque wharf]

Brainfart sorry

[glass jetty]

i ran driver and monitor

[true hemlock]

steam machine ain't that small

[glass jetty]

but forgot that server and driver are different thing in nut

xD

just wasted 20 min

[amber fractal]

Didn't think of buying it at the time because I wasn't entirely sure on what it was. It was in one of those amazon rebate stores so never saw it again afterwards.

[opaque wharf]

Nah, I mean price

[true hemlock]

easy, i'll just put together some of the parts i have and its a guarantee to be lower than $800 if its counted with how much i paid for the parts initially

nah but

[amber fractal]

That is not wrong

[true hemlock]

its fairly easy

to actually build a better price to performance pc than the steam machine

[amber fractal]

I trust you can, I doubt that many others can

[true hemlock]

depending on which oem 7500f distributor you can probably get that for $120

ram is probably the hardest

because you probably wanna match the specs so 2x16 ddr5 usually cost like $300 nowdays

$420.
for motherboard, you can get a fairly cheap matx with 2 dimm slot. those tend to be the smallest

pair with 9060xt and a decent 750W psu, should still be under the $1000 budget

idk how much does sff matx cases cost though

yeah aight

[opaque wharf]

Small case is usually pricey

[true hemlock]

its harder for you peasants

[amber fractal]

a cardboard box doesn't cost much

[opaque wharf]

Which is weird af

[true hemlock]

if i do it

i think i'll go with that 14900HX modt

i got it for $300

CPU + board already

more expensive than 7500f + decent B650m but oh well its a bit more performant in games.

[amber fractal]

I know your deal hunting skills well, so I'm not even suprised anymore by this whole breakdown

[true hemlock]

that board supports ddr4 btw

i can probably just hook it with the 2x16gb ddr4 dimm i got for $50 last year

buuuut that ain't fun

so i'll do 2x32 that i got for $90 just weeks before the rampocalypse

$390

i'll use that 5060 i got for $250, its brand new

$640

proper voltage tuning and PL profile on the 14900HX, i'd like to use the RM650 i got for $19 last year but that ain't fair

[opaque wharf]

How do you even got all of those deals man

[amber fractal]

Your existense is unbalenced, but I ain't asking for it to be patched. Because then I won't get to hear about insane deals

[true hemlock]

that rm650 was second hand

but such second hand unit is generally still more reliable than new B tier PSU

[burnt oasis]

Speech 100

[true hemlock]

i think you should still able to get it for $60 used

[mossy fossil]

ngl, i think i nailed this desktop with that wallpaper now

[true hemlock]

$700

ssd...

ah fuck it

[amber fractal]

activate linux my beloved

[true hemlock]

its so unfair

a new 2tb nvme for $180, leave $120 for the sff case

there you go

[mossy fossil]

my favorite joke program

[true hemlock]

$900 build that beats the steam machine

potentially is more compact(?)

[amber fractal]

maybe?

not good at sizing as haven't really dealt with desktops at all

[true hemlock]

this but replace the cooler with sff one

you can probably get the sff cooler i used for my briefcase for $60 at most rn.
vertical mount the GPU on the sff case, PSU positioned on top

im ngl

there's still budget surplus of $40

so fuck the $60 psu

[amber fractal]

2tb hdd

[true hemlock]

do a sff psu instead for $100

should give you a compact build comparable to steam machine

[mossy fossil]

i still hate the fact that steam depends on steamwebhelper now

[true hemlock]

for faster performance

[mossy fossil]

and i'm particularly mad about that because VGUI IS STILL ON THE CLIENT, IT'S THE ERROR MESSAGE WHEN WEBHELPER FAILS TO LOAD

WHY VALVE

[true hemlock]

nah i'd make this unfair

use the $19 psu

[ivory plinth]

death to low end pc by 100% cpu usage

[true hemlock]

use the 1tb nvme i got for also $19 from the same exact guy

$798

what do we want extra

oh right

the sff cooler

$858

[obsidian mantle]

its a bit fun to watch as it scans all nodes with speedhack

[true hemlock]

specs is
14900HX (tuned)
B660m modt board
2x32gb ddr4
RTX 5060
1TB nvme gen 4
650W psu

steam machine is using 6 core zen4 cpu

7600 basically

ryzen 5 7600
chipset unknown
2x8gb ddr5?
28 CU rdna 3 (bit slower than rx 7600)
idk about storage

[amber fractal]

my bets are on 512/1tb nvme in a "special form factor"

[true hemlock]

if its 2230 it'll be funny

[amber fractal]

if it is 512+2tb valve will have my respect

[true hemlock]

steam machine isn't the best for value but eh
kind of want it but i don't have any good justification for it

[amber fractal]

grab the headset and not suffer with networking?

it does have a dedicated wifi for streaming and steam controller dongle

[true hemlock]

oh i do have justification for the vr headset though

[nocturne olive]

I you want to compete with Steam machine, you gotta be on Linux
Which means you'd significantly benefit from an AMD GPU

[true hemlock]

elvy is making a number of 3d models including mine

[amber fractal]

[true hemlock]

that makes it even easier

[nocturne olive]

The Steam machine is just a funny Linux box with valve branding

[true hemlock]

i'll just hit a deal with my dealer from my hometown

get rx 6800 for $250

mod it

[nocturne olive]

Sounds about right

[true hemlock]

for low thermal density cooling

[nocturne olive]

And there you have a GPU that probably works better on a Linux system than a 5090

*when accounting for worst cases of driver issues

[true hemlock]

man its genuinely so hard to justify for a steam machine

[amber fractal]

I can't wait to be able to meet more people in vr

[true hemlock]

i still have like at least 9 hardware i haven't played around

[true hemlock]

heard from her she's making one for moneka

maybe she'll also make one for paccha

[nocturne olive]

Elvy the 3D modeler

Elvy models are peak, at least the couple I've seen

[true hemlock]

gonna test the 14900hx + 5060 system

is elvy awake

[amber fractal]

seems like on DND but alive

[true hemlock]

i promised her we'll do the overclocking together

target is to push through a burst of 400W from a single EPS

[nocturne olive]

Hem

400W??

[amber fractal]

I await those results

[true hemlock]

you read that right

[nocturne olive]

That thing's gonna pull more wattage than my 3090

(My 3090 pulls 370W)

[true hemlock]

for some of you that were concerned

[amber fractal]

that'll pull more than my two most commonly used laptops

[true hemlock]

yes, 400W is very very possible on a single 8 pin EPS

official specs is 334W iirc

but 500W sustained is what we extreme overclockers usually do on sTRX systems lol

[amber fractal]

spec only matters if the cable is the issue

[nocturne olive]

12VHPWR but with 8 pins I guess

[true hemlock]

500W sustained is unofficial

but its been done so often

[nocturne olive]

Who needs 4 extra power pins right?

[true hemlock]

and somewhat safer than 12vhpwr drawing 684W lmao

[amber fractal]

334W is the "cheap PSUs will explode" limit

I expect anything decent to be able to sustain higher

[true hemlock]

connector

cable are absolutely no issue

[nocturne olive]

By the way it's kinda interesting how the 3090 I got apparently happens to be one that pulls 20W more than a stock 3090, kinda silly

[true hemlock]

oc model?

probably just linux driver lmao

[amber fractal]

I'd thought cables would be overloaded first, but too much nvidia slop has been fed to me ig

[true hemlock]

the connector are the one being overloaded

that's why everytime you see the 12vhpwr burning issue article

its the connector that melted

not the cable

[nocturne olive]

No it's actually from the factory set up with a power limit of 370 rather than 350
It's this https://www.techpowerup.com/gpu-specs/msi-rtx-3090-gaming-x-trio.b7955

[true hemlock]

proper 16 awg can sustain 12A fine per rail with not much temp hike

[nocturne olive]

If it was the Linux driver my 4070Ti would also be set to pull some crazy wattage

[true hemlock]

4 rail EPS cable can do 576W absolutely fine

just that its the matter of if the connector can handle it or not

[true hemlock]

same with your 8 pin PCIe cables if they're 16 awg

[rough bloom]

ye, there are instances of individual wires getting quite hot too (due to poor connector contact + bad circuitry from NVIDIA), but what fails is usually the connector

[true hemlock]

shunt resistance imbalance

the cable hit 200C iirc?

[nocturne olive]

It just is some sort of 3090 designed to pull 370W

[rough bloom]

hot

[true hemlock]

nvm misremembered. was around 80C

but that's still a big issue

[amber fractal]

80% of the way to water boiling is a lot more that 100% of what the human body is meant to handle

[true hemlock]

christ

[true hemlock]

23 amps

yeah the connector IS the one failing

[amber fractal]

now that I think about it, this isn't the most possible as it'd probably desolder itself first and the plastic cover would melt off, making it possible for the cable to short itself

[true hemlock]

and most good PSU would immediately kill the shorted rail

not most

all

because if it doesn't, its not a good psu

[nocturne olive]

The magic of protections

[true hemlock]

there's a reason i said that a good unit, even second hand, is still better than a new but mid unit

https://docs.google.com/spreadsheets/d/1akCHL7Vhzk_EhrpIGkz8zTEvYfLDcaSpZRB6Xt6JWkc/edit?usp=sharing

used A tier (for no longer than 3-5 years) is still going to be better than a new B tier for spec reasons

though i'd not recommend for normal peoples to get second hand psu in any way

but i have technical justification if i did

[sage crag]

second hand psu

https://cdn.discordapp.com/emojis/1447953342083891271.webp?size=48&name=xddFlower&lossless=true

[frozen hollow]

Anyone used PixiJS to develop small browser games?

Just kinda curious how you test on a phone without absurd amounts of development friction

[rough bloom]

same as any other website I think, you run a dev server on your machine and then open that from the phone
you can connect DevTools too, so it's basically the same as on desktop except it's running remotely

[frozen hollow]

ok connect devtools? thanks

I'm running a dev server and just going to like 192. on the phone but it doesn't tell me much.

[rough bloom]

yeah, having the console + a debugger is very convenient

[frozen hollow]

we will get there we gonna make it

[frozen hollow]

Thank

[opaque wharf]

Console.log based debugging is still the most widely used debugging method lol

[jagged turtle]

well yeah, nothing beats a good old console.log

[burnt oasis]

I mean, how else you would log values from variables from runtime?

[opaque wharf]

A debugger could display the value that you track and more

But ain't nobody got time for that

[amber fractal]

no one can keep a debugger working with their personal projects. Stuff just moves too quickly

[silent island]

damn... is this normal?

[faint sandal]

https://fxtwitter.com/suyog41/status/2014691305692266744

Yogesh Londhe (@suyog41)

Amos Stealer
︀︀
︀︀NeuraVision2.dmg
︀︀3921a213f60eceecdcebdd9cf466bf27
︀︀
︀︀#Amos #AmosStealer #Stealer #MAC #IOC

**👁️ 37 **

FxTwitter · Jan 23, 2026, 1:26 PM

neurovision

[amber fractal]

Glad to hear that I'm not the only one that thought of that name connection

I was planning on using Neura as an alias for in a writing project I really need to work on

[obsidian mantle]

oookay skill tree navigation done.
that was not easy

now i need to extract the reason why a perk is unavailable
they provide IsTrue function on condition but the condition itself is looking like a very bad thing

[sick owl]

Your what

[sick owl]

[rigid snow]

20% of laptops are macbooks

[maiden geyser]

20% are apple's arm laptops

[rigid snow]

which happen to be macbooks yeah

[maiden geyser]

macbooks didn't exist before 2019 thoughever

[sage crag]

would you look at that

https://cdn.discordapp.com/emojis/1455273822625333375.webp?size=48&name=mlntcandy&lossless=true

[real sierra]

koniii

[sage crag]

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

[real sierra]

the vps works really well

thank u again

[sage crag]

[knotty obsidian]

https://youtube.com/shorts/9Bu2O2VrgGg?si=fycAWQ8kh8FoH60E

[real sierra]

the 8gb ram did scare me a bit

[knotty obsidian]

Look at this.

[sage crag]

as long as you arent using it for cybercrimes you can use it for whatever you need

[real sierra]

all it doing is fumoquest

[sage crag]

wrr

[knotty obsidian]

Hey did you look at it ?

[sage crag]

no

?si=fycAWQ8kh8FoH60E

explod

[real sierra]

what does this have to do with programming

[knotty obsidian]

https://youtube.com/shorts/9Bu2O2VrgGg?si=fycAWQ8kh8FoH60E

Yeah,uh....come on,memes.

[rigid snow]

[sage crag]

https://cdn.discordapp.com/emojis/1424073798692507813.webp?size=48&name=NeuroConfused&lossless=true

[stray dragon]

why here in this channel

also ew shorts link and ew tracker at end of link

[rigid snow]

maybe their dad is a programmer

[knotty obsidian]

@stray dragon What's wrong with shorts ? Shorts are sometimes good or sometimes bad.

[sage crag]

https://cdn.discordapp.com/emojis/1464001170673766450.webp?size=48&name=Processing&lossless=true

annoying

[stray dragon]

annoying

• no skip forward

[knotty obsidian]

@sage crag Okay

[stray dragon]

• bunch of shit on top of the video

[sage crag]

wrr

[rigid snow]

wrr

[stray dragon]

you can convert any shorts link to a normal video link by changing it from youtube.com/shorts/abcdefg to youtube.com/watch?v=abcdefg

better

[knotty obsidian]

Really ? I never knew that ?

[stray dragon]

yeah it works for any video

[sage crag]

wrr

[knotty obsidian]

Oye.Sunno yarr.

[stray dragon]

the ?si=zyxwvut at the end is tracker also

tells youtube where video was watched from

can remove with no issues

[knotty obsidian]

So,can I use a tracker to locate a device ?

[rigid snow]

[knotty obsidian]

Yo.

[rigid snow]

gurt: yo

[knotty obsidian]

@stray dragon you haven't answered my question

[stray dragon]

yo gurt

[knotty obsidian]

Hey,does anybody know about JavaFX ?

[stray dragon]

you have reached the end of your daily free responses from t, please subscribe for just $9.99 a month for more fun t responses

[sage crag]

https://www.youtube.com/watch?v=F50bCrlTbIs

the chrome logo is blushing

https://cdn.discordapp.com/emojis/1421763041074217092.webp?size=48&animated=true&name=MyHonestReaction~1&lossless=true

this implies that the chrome logo has blood vessels

and therefore a heart

wrr

[knotty obsidian]

https://youtube.com/shorts/arB42hjUAII?si=VMqdfdpeqNsZx9zO

[sage crag]

https://cdn.discordapp.com/emojis/1421763041074217092.webp?size=48&animated=true&name=MyHonestReaction~1&lossless=true

dentge

[quick condor]

Waiting to hear back but I'm pretty sure I got the position I was interviewing for. Will mean I can actually afford 2 of my hobbies at once instead of 1

[silent cloak]

Its a broad term but I generally mean neural plasticity. Not even AGI levels but literally anything other than a highly inefficient slop transformer

[sick owl]

I think you're looking for a different term in that case, since it doesn't really square with the definition of intelligence

[silent cloak]

Current ANNs dont have intelligence

You need neural plasticity

Otherwise it is just an auto correct (ANNs only learn during training)

[opaque wharf]

I'm gonna do a reverse uno
#1460386243601764382 message

Epic fail

https://images-ext-1.discordapp.net/external/05qlR3SKNOg-AP-XcmGWRhkM_uB0oAZse_zxDCSd5PA/https/media.tenor.com/vcYOSa1pAcQAAAPo/30000-usd.mp4

I give up lol

[sage crag]

https://tenor.com/view/30000-usd-gif-13674633028161831364

[opaque wharf]

I saw that

https://tenor.com/view/neurosama-evil-neuro-embed-failure-gif-15456327638408323033

[glass jetty]

there literally no way to omit link, only if drop video itself

[opaque wharf]

https://tenor.com/view/but-i-don't-have-30000-30000-dollars-arknights-endfield-gif-7119074259363027863

[glass jetty]

dont even had 3k

[knotty obsidian]

@glass jetty why are you after his/her life ?

[glass jetty]

what?

also, why all of this vids/gifs cant load for me

what a discord failure

[stray dragon]

you can omit link if you post gif link instead of video link

:)

[glass jetty]

but you have to convert it first

also it loses sound and fps

[stray dragon]

it didn't have sound in the first place

it's a gif

https://klipy.com/gifs/30000-usd

[glass jetty]

anyway, i really cant play this vids on discord, i can play it after download tho...
not for the first time - massive discord failure

[stray dragon]

true

mobile?

[glass jetty]

nope

browser

[stray dragon]

damn not a mobile issue for once

[opaque wharf]

That's weird, I have no problem here using firefox

[sage crag]

i have no issue

[stray dragon]

maybe do a speedtest, your internet might be slow

[opaque wharf]

I do however, need to hover it