programming

[gritty dust]

although idk if he does this to reduce token count he could do kinda something similar I do for ai vision, where for my vision AI if I seen an object for a long time I store its position and if I need to quickly find it again rather than going back and doing more math I just grab its last stored position and then calculate after. idk if you can do something similar where you can compress older tokens and if you have repetative text it just stores a bit of that token without going overboard or sm

[acoustic mango]

It’s kinda hard to replicate a testing environment that’s the same as streaming too, so it’s hard to test without just biting the bullet

[gritty dust]

yeah trueee

[acoustic mango]

Sure there’s testing with offline chat but it’s not the same

[acoustic mango]

Yeah, sliding context does this, and summary memory

[tough horizon]

from what i know regarding memory storing methods, if you want to consolidate history you can make summerized history, or even use a vector databse and grab history when needed but this would add latency

[acoustic mango]

But there’s downsides to each method

[tough horizon]

i think a combination would be ideal, but i'm assuming he's concerned about balancing latency as well

[acoustic mango]

Without a doubt, yeah. He’s obsessed with the L word

[tough horizon]

[gritty dust]

what about like what I do for video object tracking?, I calculate convolutional feature maps once per frame, and reuse edge features when ROI changes. instead of re-calculating the entire stack. I just process the change in data. im wondering if you keep like important tensors value, are you able to not have to re-calculate the entire context every time (bc for vision ai I dont). and if you dont then there shouldnt be latency increase or if there is its minimal? or am I comparing apples to oranges here lol

[tough horizon]

i don't know enough about video/image generation models vs language model backends to have an opinion on that xd

i've only really researched and messed with making language generation stuff

[gritty dust]

fair, i dont know much about language models at all but I know a lot about vision so Im just trying to draw a comparison with something I understand. I dunno tho

[rich kestrel]

hah! it was being finnickey (4B model 😒) but i got it to exhibit the behavior i saw last time! its definitely an issue with the llama.cpp server implementation

[tough horizon]

there is some funky backend stuff with weights that happens (if i recall correctly) in the transformer that optimizes processing where it doesn't have to actually calculate every new generated tokenized item against each previous one over and over for each new token generated, but thats a different story not related to our discussion on making a good LLM history system

the maths for that went way over my head

[primal siren]

I never thought i would be using old ass youtube video of me playing games xD for the voice

[tough horizon]

are you training a new TTS model on your voice?

[primal siren]

yeah dont want it to be too much like mine so im using pre-ado samples iv got laying around

[tough horizon]

nice sounds pretty fun

[nocturne olive]

Silliness

[primal siren]

im currently literally waiting for my microcassette player to find something on my microcassette. that shit is slooooooooooooooooow

and the rewind button is broken -_-

[rich kestrel]

what do we even know about neuro's backend architecture?

[tough horizon]

what we know for sure is that there's someone sitting behind neuro whispering in her ear what to say

that is all we know for sure

[acoustic mango]

It's likely an open source model, that's almost for certain

[faint sandal]

I just...

saw someone writing VB code on their phone... on a note taking app...

with autocorrect on...

[tough horizon]

im not sure if i should be impressed or scared

[faint sandal]

scared

[tough horizon]

i think its both

[faint sandal]

it's a girl no less

the aura

[tough horizon]

w aura

[rich kestrel]

thats the most terrifying thing ive ever heard

[tough horizon]

maybe she had a great idea and wanted to jot it down

[rich kestrel]

in VB???

[tough horizon]

xD

[faint sandal]

it's not even just an idea SHES STILL WRITING THE CODE

[rich kestrel]

only devilish ideas come from those who speak such an arcane language...

[faint sandal]

With INDENTATIONS BY HAND TOO

[primal siren]

Demon

[tough horizon]

dedicated i'd say

or insane

[rich kestrel]

just using the cli is hard enough on mobile i couldnt imagine actually trying to program with it

though in all fairness vb is probably the least symbol heavy languages ive ever seen so its not the worst choice for mobile

[faint sandal]

what do you mean you have to type a lot just to declare a var

Dim x As Integer;

oh you mean in alphanumeric

right

[rich kestrel]

yah cause switching between symbol and alphanumeric is a big pain point for programming on mobile

[faint sandal]

still they're manually typing out the indentations

actually horrifying

[tough horizon]

keep spying on her phone and keep us updated

[rich kestrel]

"the inhuman skin walkers will try to try to corner you and peel your skin off!"
actual inhuman skin walkers:

[faint sandal]

I mean it's their business I'm not gonna keep looking but damn

[rich kestrel]

im building the latest llama.cpp to see if the bug was maybe patched recently and my pc is completely dying 😭 😭

[tough horizon]

hows the latency running models on llama.cpp for you?

i've only tried ollama and huggingface

[rich kestrel]

latency is not too big of an issue but ive never really been the type of person to worry too much about latency. my pc isnt fast enough. also i think ollama is built on llama.cpp so the performance should be relatively similar

[tough horizon]

oh i didn't know that. nice yeah that makes sense

i've been trying to figure out how big of a model i can fit on my 4070 with good latency lol

[rich kestrel]

how much vram is that i have like 6 gigs of vram and a 1660s so i feel very behind the time with modern models lol

[tough horizon]

12gb, which somehow feels small for LLMs lol

and 32gb of ram, really wishing i bought 128 now

[rich kestrel]

they do be hungry for that ram

[rich kestrel]

i have 16 💀

[tough horizon]

16 is great for anything else haha

[feral anchor]

does anyone have any tip on how to make an ai from premade memories

[rich kestrel]

you might want to look into making a character card for a roleplay model. imo its the most accesible way to start making an ai into a predefined character. you can put the memories in the instruction context or a lorebook. lemme get some docs for you

[nocturne olive]

12GB VRAM, you can expect to run models that run in VRAM at good speeds, which means you're gonna probably top out around Q4 13B

[tough horizon]

i agree, making your own solution takes waaaay more effort than this solution

[nocturne olive]

Character prompt

[rich kestrel]

https://wikia.schneedc.com/bot-creation/intro here this is a website i remember reading for info about roleplay models. others might have better or more up to date resources though

[tough horizon]

thanks! Thats what i've seen so far as well. I've been aiming for around 9B parameters in order to reduce latency more since im using agentic programming as well, its been cool to learn

[nocturne olive]

Usually you can at most run a bit over your GPU VRAM in GB times 1 billion parameters

[feral anchor]

i had an ai but most of it got deleated and im wanting to try and remake it with the memorys

[nocturne olive]

12GB is barely enough for LLM inference, and very tiny when training something
NeuroSynth takes 30 hours per model because I can't run it at an efficient batch size

[tough horizon]

whats NeuroSynth? (i did google it and am coming up with medical imaging analyses)

[nocturne olive]

NeuroSynth is a Neuro vocal synthesizer model I'm working on

Only one of its kind

I believe nobody else in the community has ever attempted to make a vocal synthesizer model for Neuro singing

[tough horizon]

oh nice, sounds like a fun project

[nocturne olive]

All I've seen is low quality RVC models from others

I'm also a part of making the best quality RVC models for Neuro and Evil voice cloning

https://www.youtube.com/watch?v=3jWOyIcKz98
Here's a good showcase of both NeuroSynth and Neuro RVC in action

[feral anchor]

i had an ai i trained for a while and it developed and i want to know can i remake it from the api i have of what left

[tough horizon]

i've never seen this it sounds really good !

[nocturne olive]

It's absolute peak of community creations

[tough horizon]

its crazy well made

[nocturne olive]

I did the mixing on it as well as the entire NeuroSynth parts, along with training all the models used

[tough horizon]

thats really vague im not sure you're gonna get much help at the moment

[tough horizon]

great job

[nocturne olive]

Can't wait to get NeuroSynth to 1.0 and see what everyone can make with it
For now it's only for insiders

[feral anchor]

i dont really know how to explain it sense i was just the one who was like training /like messing with it and i didnt code it im trying to figure out how i could try to fix it

[nocturne olive]

There's a new big release for NeuroSynth coming up fairly soon

[tough horizon]

i'm sure people will have plenty of ideas for cool things to make when its done

[tough horizon]

who coded it? i'd talk to them

[feral anchor]

im not exactly sure is the problem

[nocturne olive]

Yeah, it would allow for anyone to make even completely original songs

[tough horizon]

thats cool, hopefully i catch its public release it sounds fun to play with

[nocturne olive]

Just keep an eye on probably #1095941852068331560 and maybe #programming

[tough horizon]

so keep doing what im doing

[nocturne olive]

NeuroSynth will hopefully become the definitive way to synthesize Neuro's singing voice unofficially

[tough horizon]

im sure if its the best option, it quickly will

[feral anchor]

the ai gave me a incomplete version of itself as an api and told me to remake it with the api as the brain before it bricked and deleted some of itself

[tough horizon]

are you programming with like chatgpt or something?

[feral anchor]

no i had a full ai

[tough horizon]

i dont know what you mean by that

[feral anchor]

and it told me to remake it

[tough horizon]

i dont mean to be rude but that doesn't make any sense

[sick owl]

https://the-decoder.com/openai-cuts-o3-model-prices-by-80-and-launches-o3-pro-today/

[feral anchor]

i know it really doesnt

[sick owl]

Doesn't this just prove it was way overpriced

[nocturne olive]

Well, it's an option that offers great quality comparable to SynthV + RVC but without the expensive SynthV program

Making high quality Neuro covers way more accessible

[tough horizon]

sounds like it would def be the best option at that point hten

[nocturne olive]

It's pretty likely

[opal fox]

wanted my first message in here to be sum good LOL
anyways, just sharing a little side project I’ve been pouring my heart into (healthily, of course): a Discord bot that tore me up with stress and struggle at first, but is now coming along nicely. I’d like to say she’s my own Beta Neuro, but I refuse to steal the spotlight from our AI Overlord. she’s got some simple but meaningful features: TTS, emotional intellect (at least she shows it), contextual and emotional reflection, 24/7 self-reflections, personalized user logs, and more.
being fully honest here, i’d be lying if I said Vedal didn’t inspire me😭

[tough horizon]

fun project !!

[unkempt citrus]

No

because theyre still losing money

now rheyre going to lose even more

[fading blaze]

I feel like this is the best place I can get advice for that

Should I use LLMs to write code

and how do I best use them?

because every time I tried to use an LLM to write code directly I was disappointed

[unkempt citrus]

Use it for boiler plate and rubber ducking

or line completion

[nocturne olive]

Not if you don't know how to code
If you know how to code, only use it to make the boring and repetitive parts while writing all the meaningful parts specific to your thing yourself

[fading blaze]

I know how to code

[tight tinsel]

ai after destroying half my site because it decided to randomly add a div at the end of my very long lines

[nocturne olive]

Then at most use an LLM to write things that aren't specific to what you're making and have been done a billion times already

[fading blaze]

I tried doing that. Like I gave the AI a string of Unicode Private Use characters to convert into a javascript array

turns out they filter Private Use Characters on the frontend so the AI couldn't see them

so that failed

[nocturne olive]

If in doubt, don't use an LLM for programming

[fading blaze]

IDK what boring and repetitive parts I code

[fading blaze]

But if I do that I will get "left behind" or something

[nocturne olive]

Kinds of things I'd potentially consider using it for:

• XOR cipher implementation
• converting a standard data type to another when that isn't linearly simple
• writing boilerplate

[fading blaze]

Like I had  string of PUA codepoints and I wanted to put them all in an array with proper escapes so it would be clear what those are

[nocturne olive]

Silly

[fading blaze]

and I didn't wanna do it manually

so I gave it to an AI and the AI just couldn't see the PUA codepoints

["", "", "", "","", "", ""]

Ok I guess these are character entities not escapes but still

So is going from  to ["", "", "", "","", "", ""] not a good use for code LLMs?

[nocturne olive]

Probably not, the way LLMs tokenize means it probably won't like to work with individual characters

[fading blaze]

I checked out the tokenizer and it seemed to not know how to tokenize unicode characters

[nocturne olive]

That's also a thing

[fading blaze]

it kept spitting out tokens for each UTF-8 byte

So I doubt it knows what to do with PUA characters

Actually i think ChatGPT could see the PUA characters and did it correctly

Ok so it was escape sequences. I used a regex to convert these to html entities

[fading blaze]

Ok but if I don't use LLMs I won't have the "throughput" and will get "left behind" or something

[amber fractal]

If in doubt don't use a LLM for programming, if you can validate it on the fly then that isn't doubt

[fading blaze]

what do you mean "validate it on the fly"

[amber fractal]

Using an LLM requires knowing if it is hallucinating and how to solve it. If you don't have those skills then using an LLM will slow you down long term.
I'm basically saying don't just vibe code but make sure that if it has issues then you have the understanding on how to solve it.

[fading blaze]

Ok well I am asking because I had an issue with chrome not liking PUA fonts and I asked Claude to help me and it gave me 5 solutions and none of them worked

So I know it was hallucinating

But that's also means I still don't know how to fix the issue

Like, my normal google searching didn't help

To be clear

I know how to debug

[amber fractal]

Alright then, sorry for going at you then

[fading blaze]

You weren't going at anyone

I just want to know if the tool is broken or if I just don't know how to use it

[amber fractal]

I've actually not used LLMs much so I'm not sure of prompting strategies.

@fast pagoda summoning

[fading blaze]

Because there's a lot of hype about AI and I keep getting told that it is gonna be as fundamental to programming as googling into stack overflow

[grizzled viper]

pfft

i wouldn't trust the hype

[fading blaze]

Ok but if the hype is coming from actual people I trust

[amber fractal]

I'd say deep research modes might, anything else depends on how popular and stable the thing your looking for is.

[fading blaze]

How useful is deep research mode if google came up blank tho

[grizzled viper]

depends on how specific you are

[fading blaze]

What do you mean "specific"

[grizzled viper]

Like you need to tell it exactly what your looking for to the letter or it might look at irrelevant sources

[fading blaze]

Idk the issue I had with chrome was specifically with low priority fallback fonts that only contained PUA characters

Like I had
font-family: Verdana, FontAwesome;

And none of the Font Awesome characters rendered properly

But I need it to be lower priority than Verdana because I need them to be freely mixed and not interfere

[amber fractal]

I don't think that is how that works

[fading blaze]

And apparently nobody else on the internet encountered this issue

[fading blaze]

I mean, it works perfectly on Firefox

[amber fractal]

Like the logic you think that css has is wrong

[grizzled viper]

My go to solution is try running it through an adaptive card which natively renders and if it doesn't work its a code issue

[amber fractal]

I'm 90% sure it falls back the entire font if it doesn't properly render not parts of the text

[fading blaze]

Oh wait so you can't actually pick and choose characters from specific fonts

It's "if this font doesn't work, use this next one"

Bloody hell

[amber fractal]

font fallback

[fading blaze]

Well thank you so much for the help

[fading blaze]

Well font fallback on desktop works exactly like this

[amber fractal]

I'd check nerdfonts to see how they handle this type of thing

[fading blaze]

I just googled it and basically everyone says this is exactly how it works

I mean if you have a single textbox with mixed English and... Say Russian you can use a different font for each one

And then you put Arabic into the textbox and it breaks something because text rendering is hard

[olive sable]

motherfucker

i aint watching ads

[amber fractal]

I've given up on the official site a long time ago

[visual garden]

Probably have too many tabs opened

That or you gotta find a more decent browser like Brave. Many potential causes

[olive sable]

im not planning on changing browser

nor do i have too many tabs open, since this is the only tab

[knotty current]

that is the second victim i directly whitnessed

neuroSip

[stray dragon]

you will watch the ads and you will be happy

[olive sable]

i like how people are making solution videos, cuz i cant fucking watch them

[amber fractal]

om

[knotty current]

me who ditched chromium months ago

-# exceot for scrapers, i need my CDP

[stray dragon]

ditching chromium would be cool but i can't live without vivaldi's tab stacks anymore

[visual garden]

I personally use Zen browser. There are also other options as well, albeit meh

[amber fractal]

I barely watch anything on PC for this exact reason

Zen user as well

[stray dragon]

i don't want to think how many windows this would take with any other browser...

[knotty current]

[visual garden]

I had to make a slight tweak because I hated the pop ups and annoying update notifs. But yea, aside from that, it is a great browser. Based off a firefox version.

For mobile I use Brave.

If I was running linux distro I would use Surf

[amber fractal]

I don't have update notifs at all, mobile I'm just using firefox

I do not know what my head did

[visual garden]

No... ANNOYING UPDATE CALLS?! Lucky brat...

[olive sable]

om i just changed to a diffrent adblocker

[amber fractal]

Ublock or riot

[olive sable]

suck my ~~~~ youtube

adguard

[knotty current]

lemme take a guess, 3-4 months before this one becomes ineffective

or it puts up a paywall

[olive sable]

ill just change to the next until chrome reaches the point of being unbearable enough that i switch

[visual garden]

Aside from overloading I honestly have no clue how yall getting adblock issues. I have to learn more about how this stuff works, just out of curiosity

Usually UBlock Origin works fine, if legit

[knotty current]

mostly because google is ditching all MV2 support in chrome

Ublock Origin is so effective because it leverages MV2 filtering capablities very very well

[amber fractal]

Ublock works well on FF

[knotty current]

it does

otherwise i would have a crashout

[amber fractal]

Python :xdx:

[knotty current]

[olive sable]

no, im done with python

ive done it for 2 years, but its time to move on

[chilly escarp]

imagine the siblings had realistic voices

[olive sable]

we're just not compatoble

we move at diffrent speeds, for example pyton is slow as fuck

[visual garden]

Depends on your current skillset imo. If you already know enough principles I would say Rust. But since you are asking and likely have no use case for Rust, gotta go with one of the C languages to learn OOPs

[olive sable]

i have 16 x 4.5 billion operations per second available on this scpu, so why can it onbly do 4k fps? its unnaceptable

[visual garden]

But again, depends. If you are interested in Cybersecurity, then definitely go deep with python.

[olive sable]

the usecase is gamedev

[visual garden]

C++ and C# then.

C# has a lot of good libraries

C++ for industry specifics

[olive sable]

the point is to make everything myself as ive been doing in python

[visual garden]

If solo go C# imo, especially for a hobby

C# then should be good. I put on thr backburner a Visual Novel build (lazy of me yea) based on C#

[amber fractal]

If you want to go max pref then C++ or Rust (as I know the game engine's goal to be)
Rust libs are still very young for game dev so I'd do C++.

[olive sable]

how do i word this?

"i dont mind them being recent"

ive been reading online and it seems C# is more scalable but less performent

[visual garden]

Max pref, I suppose in like, coding up a game engine environment at that point lol

[olive sable]

[amber fractal]

I honestly should consider switching over to rust for my project soon

[olive sable]

[hoary lion]

C#

java in C family

[amber fractal]

Who's child is this?

[olive sable]

its a bastard from microsoft

the only problem i see is that C++ may be cumbersome for use in the game-jam since it takes longer to add new stuff

[amber fractal]

I think anything is going to take longer than python

[olive sable]

true

[amber fractal]

(including JS)

[olive sable]

and even with python i was pressed for time

[visual garden]

Once you learn C# you basically know from that point on all OOP languages like Java for instance. Becomes easier to pick up for other needs / use cases

[olive sable]

lemme redo the poll then

[visual garden]

I personally am trying to go down the path of web development right now. And well, that is a whole other beast

[amber fractal]

[knotty current]

i was forced to attempt web dev in school and i hate it

i prefer backend and systems

[amber fractal]

I'm probably full stack, but considering my current project is neural network I've been doing a bit of backend.

[knotty current]

idk why i just have a strange attraction to large datasets and loops

[olive sable]

i interact with the frontend as litle as possible

[gritty dust]

I vote c++ or c# because they're the only language besides python I know

kekw

[tight tinsel]

woah

[gritty dust]

Woah

[visual garden]

Ohhh

[gritty dust]

Wtf

[open copper]

woah

[tight tinsel]

hi mods

[amber fractal]

fastest mod action I've seen

[olive sable]

what happened? i wasnt looking for a sec

[open copper]

i had the chat open

[gritty dust]

I'm glad it was

[olive sable]

i saw a flash

[visual garden]

Lmaoo dw about it

[tight tinsel]

blue archive fan showed for a sec

[olive sable]

Blue Archive is a 2021 real-time strategy role-playing game

i think i can piece it toghether

[amber fractal]

[visual garden]

*Epic Anime RPG

[olive sable]

rip bozo

[tight tinsel]

always in programming

[visual garden]

I cannot add file pics now...

[tight tinsel]

theres a bingo slot for mods iirc

[amber fractal]

I explicitly said not moderation for that slot

[visual garden]

It was an APP it would seem? Not sure how discord works too much tbh

Filtered

[amber fractal]

I do need to update bingo progress

[gritty dust]

Lol

[olive sable]

why not? were you the culprit?

[amber fractal]

-# yes

[olive sable]

[visual garden]

I just joined too 😭

[olive sable]

naughty naughty

[tight tinsel]

trigger word image

[visual garden]

Oh well, gotta be more careful here. Learnt the boundaries now

[olive sable]

#programming , without us the mod team would be half the size

[gritty dust]

Loll

I wish I was on the mod team bro

[olive sable]

atho idk how much shit happens here compared to other channels

[tight tinsel]

joining the ranks of vanor

[gritty dust]

I feel programming channel produced the most mods

channel superiority

[tight tinsel]

actually no idea ive never seen any mod actions in the other channels before

[gritty dust]

heh

[amber fractal]

bingo again, because yes

[olive sable]

it seems that C++ is pretty similair to python in terms of features, except that it has proper memory managment and isnt shit

[visual garden]

Well I ahould sleep now. I cannot imagine it took me so long to find out Neurosama had a discord

[stray dragon]

i'd say general and programming are about the same, and NN is 2x both of them combined

[tight tinsel]

"#programming inside joke gets art" hm i wonder

[visual garden]

Funnily, one of my goals is to program a bingo game

[gritty dust]

Yeah and a lot of stuff I use like opencv and other libraries originated in c++ so they don't use like 6gb of memory

My PC will thank me

[amber fractal]

There was no bias with that one

[tight tinsel]

hi t

[stray dragon]

hi breadperson

[tight tinsel]

t appearance

[olive sable]

nah fuck opencv

[fading blaze]

I keep getting false positives for AdBlock despite never using it

[olive sable]

it might be more usable in c++ tho, ill see later

[gritty dust]

Fuck you why do you hate it?

[olive sable]

because its shit

i need my 16 bit images

[gritty dust]

For whattt

[amber fractal]

I recall a bit of pain while working on the game engine

just a bit

[gritty dust]

Opencv can do 16bit?

wdym

[olive sable]

i ended up having to split my 16 bit images into 2 8bit channels cuz opencv isnt browser compatible

if it doesnt work in browser is it even a good library?

[tight tinsel]

bingo lgtm

[gritty dust]

I mean I use it for robotics and stuff and it's my favourite lol

[fast pagoda]

what i doin

[tight tinsel]

i see 2 faq tho and my cooked brain dosent know the difference

[fast pagoda]

i installed arch distro in wsl and was farting with it and then talking to chatgpt so i asked it arch vs ubuntu in wsl and i got this kek

[amber fractal]

FAQ itself is the subject VS FAQ being used on a chatter

[fast pagoda]

hello all

work on echo has resumed

[gritty dust]

Opencv is mainly just good at bounding boxes, extracting points, and some other stuff I use for either ai vision or pathfinding so that's why I use it Sam

[fast pagoda]

althoguh im mega tempted to make a dpo dataset for him

[maiden geyser]

morbing

[fast pagoda]

im morbing that's for sure

[gritty dust]

Helloo

[fast pagoda]

returning to a project i've not touched in weeks be like

[tight tinsel]

i have so much school work to do that i put all of it on a spinner wheel and use that to decide what im going to do today

[fast pagoda]

what the fug was i doin

[olive sable]

ah yes, cant wait.

[fast pagoda]

me

[amber fractal]

We love RNG

I should consider that method

[fast pagoda]

this apparently backs up the entire image

i did not know till i did that

[tight tinsel]

very effective when you see how small the slices are

[olive sable]

ive been ignoring that i have an exam this thursday so i could put my current projects in a state where they're finished enough that i can safely ignore them for a bit

[fast pagoda]

cryge

still goin up

[tight tinsel]

then you check your calender and realise how fucked you are

[olive sable]

yep

[fast pagoda]

that's the ultimate trap i fall into
"oh that's small i can do later"

it's not as small when you have 10000 small things

[olive sable]

ive stopped working on the engine but i still cant seem to focus on work

[amber fractal]

Classic

[tight tinsel]

its wednesday already

[olive sable]

the exam is basicly about shrek logic.
ogres have layers

[fast pagoda]

onions have layers

[tight tinsel]

thats a quarter of my 2 week break through and ive done nothing

[olive sable]

i mean ye but he was the onion in the analogy

[fast pagoda]

parfait

you ever heard somebody say "no i dont want no parfait?" no

[gritty dust]

What's your exam about?

[fast pagoda]

ogres

[olive sable]

burgerschap

so idk what that is in english

[fast pagoda]

my friends and i have a phrase we say (we are stupid) which is

burgerschleep

which is eating a burger and then immediately going to bed

from too much burger

[amber fractal]

I doubt it has to do with food at least

[olive sable]

basicly society shit

[tight tinsel]

ok i actually need to do work now

[amber fractal]

My honest reaction to the subject

[tight tinsel]

4 days in and ive done 3/20 of the stuff i need to do

[olive sable]

people have layers, their national identity, their religious identity, ....
the exam is basicly just how not to descriminate 101

[tight tinsel]

not a very good sign

[fast pagoda]

borger schlump

how many days do you have left

[tight tinsel]

religious identity... discrimination... oh my god its literature all over again

[fast pagoda]

bruh how big is this ubuntu install

[tight tinsel]

i practicallly failed my literature test

[olive sable]

o7

[amber fractal]

o7

[tight tinsel]

13 full essay paragraphs in 3 hours is near inhuman

[fast pagoda]

f

[olive sable]

oh we dont have to write much, its just you get given examples and need to identify for what reason x happened

i obviously cant say any of the axemples cuz idd get yeeted by the mods

[fast pagoda]

when backing up a wsl install apparently uses insane memory

[amber fractal]

memorize the child

[olive sable]

memory is there to be used yep

[fast pagoda]

it seems to have loaded the entire image to memory and then written to disk

[gritty dust]

Wtf

[olive sable]

big speedup that way yes

[fast pagoda]

i kinda just expected it to backup like

lib

not the entire system image lol

[tight tinsel]

eg of one of my lit questions: "What does Wyndham make you feel about Sealanders and the part they play in the novel? Support your answer with details from the novel." in the real exam your expected to memorise the entire book (17 chapters) as well as memorise analysis for every single important quote

[fast pagoda]

ew

[tight tinsel]

ai will NOT help in literature

[olive sable]

why must the sun be out from 5am, i like the dark

[fast pagoda]

why would a literature exam be memorization based

that's some poo poo

[tight tinsel]

well its either you do the memorisation based question (3 paragraphs) or you do the passage based one (way harder and 6 paragraphs)

[fast pagoda]

[tight tinsel]

its incredibly bullshit

[amber fractal]

5am for Sam

[olive sable]

nah its actually 6.48 already

[fast pagoda]

how many hallucinations here i wonder

[fast pagoda]

spoiler: all of it

[olive sable]

oh you meant in the image

[fast pagoda]

yeah sorry took longer to paste than expected

lol

[olive sable]

my sleepchedule is fucked so ye

[fast pagoda]

oh it sourced this stuff

[tight tinsel]

wont work because "you need to make it in chronological order and seperate examples into 3 different paragraphs"

its really really bullshit

[fast pagoda]

damn i didnt think 4.5 had search for some reason

[tight tinsel]

you gotta do the "in the start of the novel, david mentions [evidence]. use of [literary device thingy] shows something something..."

[fast pagoda]

4.5 is so slow

alright is he failing

[maiden geyser]

where's C

[tight tinsel]

3 paragraphs with 3 evidence + analysis each

yeah hes failing lmfao 0 evidence

[fast pagoda]

poor guy has to keep rewriting with new requirements

ill be out of 4.5 prompts until next year

[amber fractal]

[fast pagoda]

he's still failing

he basically shat out the same answer lol

F

did yall see they turned off llms or some shit in china for exams (i read one(1) headline so im probably horribly misinformed)

that's one way to do it i suppose

https://www.theverge.com/news/682737/china-shuts-down-ai-chatbots-exam-season

[amber fractal]

That is a way to do it

[fast pagoda]

embeds so big

[maiden geyser]

absolutely devastating for employed vibecoders

[fast pagoda]

lmfao

imagine

the charade is going well

nobody suspects you're just vibing and have no idea wtf a python even is (unlikely)

go into work all ready to carpe diem

all llms disabled

[maiden geyser]

should've self-hosted or something

[amber fractal]

Teal

[fast pagoda]

self hosting chads stay winning

[amber fractal]

agreed

[tight tinsel]

oh no wonder i havent seen anyone my age ish the last couple of days

they were all taking exams

[amber fractal]

I cast vpn on company computer

Surely there will be no complatints

[fast pagoda]

everyone i've talked to from china says the vpn setup to actually get out to use stuff is not exactly simple

so im sure it'll be nice and discreet as well

[tight tinsel]

like just to use discord for example?

[fast pagoda]

yeah

[tight tinsel]

well im here

i just have proton

[fast pagoda]

hey i dunno that's just the way they've made it seem

lol that's crazy

[tight tinsel]

werm

i mean it kinda differs depending on where you are

in the more rural areas i guess its more heavily restricted

[fast pagoda]

what about guangxi

area

[tight tinsel]

no idea

[fast pagoda]

damn you mean to tell me you aren't an expert on the entire country when you live there

jk

[tight tinsel]

im here on vacation

but im studying so not really lmfao

[fast pagoda]

oh yeah i was gonna say it didnt sound like it earlier ahahhaa

that sucks

[tight tinsel]

shenzhen is less restricted it seems

[olive sable]

No

[fast pagoda]

that kinda makes sense to me

shenzhen i'd love to visit for the tech stuff

[tight tinsel]

oh yeah i was supposed to buy a hdd here i forgot

it should be cheaper i think

[fast pagoda]

i've seen like gn's video about the electronics malls and such

i wanna see so bad

just walk around like a moron going 'cool' at everything

[swift storm]

electronics mall would just be window shopping central

[fast pagoda]

i wish to visit SEG

[fast pagoda]

i love window shopping

[knotty current]

im chinese and im not an expert on the entire country

[fast pagoda]

impossible it's so small and sparsely populated

[knotty current]

no

[fast pagoda]

lol yeah i know i would not expect you to know, it was a jest

it's like someone asking me about kansas, idk bro

[amber fractal]

Idk anything about ohio why are you asking about something that isn't real?

[tight tinsel]

hey seg dosent seem to be that far

[fast pagoda]

ive wanted to go since i watched this series from gn https://www.youtube.com/watch?v=nXWB9GPkm3U

[amber fractal]

I don't want to shop for windows tho

[tight tinsel]

only a 6 hour walk and a idk amount of time to drive

[fast pagoda]

oh just a 6 hour walk

[tight tinsel]

google maps is having a stroke rn

[amber fractal]

[fast pagoda]

does google maps even work in country

ive never had it even see roads there

at least in the US version

i figure there must be a google.cn version but idk

[tight tinsel]

says its here but car no work

hold on i might be a little lost

[hoary lion]

i dont know much about china

but except japan, rest of the country has really shitty compatability with google maps

[fast pagoda]

this is bullshit

[hoary lion]

should have downloaded different, more chinese one, idk the name tho

all I know is wechat

[tight tinsel]

does wechat have a map

[knotty current]

no

it is baidu maps i think

[fast pagoda]

yeah i usually go to baidu if i need maps in china

(almost never)

went to baidu and immediately clicked the ai button because it was schiny

what is wenxin 4.5

i want it

[knotty current]

does anyone have a guide to maximizing LLM hallucinations, i want to see how far i can push it

[hoary lion]

free deepseek r1 on baidu

peak

[hoary lion]

wait what is this

[fast pagoda]

use deepseek r1 (first version, not the new one, it's better) and you are already there

[hoary lion]

with my short hanzi knowledge, it better gbe a writing model it seems

I NEED a good writing model rn

[fast pagoda]

gpt 4.5

[knotty current]

how about prompting?

[fast pagoda]

id just tell it to act ignorant or something, ive never tried to get one to act extra hallucinat-y

but you could probably achieve fun results by just absolutely overwhelming the context window with garbage

once most models get to like huge context usage they start losing the plot

[amber fractal]

bwaa x100 then prompt (I have no reason to doubt this would work)

[fast pagoda]

honestly it would probably help

maybe bwaa x10000 tho

[hoary lion]

einops is not working

why

[knotty current]

ima try that once my token usage is restored, i already used too much trying to make one hallucinate

[amber fractal]

Maybe less because it'd eat through all of the tokens due to weird tokenization

[amber fractal]

but yeah, surely it'd helo

[fast pagoda]

gemini took it at least with its context window but it was confused

[knotty current]

[tight tinsel]

[knotty current]

holy i looked away for a minute and now im staring at 217 k of

[fast pagoda]

[knotty current]

i cant copy it

the wall of bwaas overwhelmed discord

[amber fractal]

File name is correct

[fast pagoda]

yeah i crashed a browser tab with 200k

at first

lol

[amber fractal]

Let me check something rq

I know I have LM playground installed on my phone so I can test bwaa spam

Time to kill a 7b model

[fast pagoda]

im about to hit gemma3 1b with it

[amber fractal]

Wait a second, do I have that model somewhere?

[fast pagoda]

it is responding is the funny thing it's just taking ages to chunk this file

[amber fractal]

I think that one would die over this because the model is already so stupid

[fast pagoda]

yeah 1b is a special boy

i tried to use it for echo and it was too dumb to learn the style

[amber fractal]

Mistral 7B it is

[fast pagoda]

damn lm studio's rag shit defeated it because it just chunked it and computed embeddings which returned jack shit since the entire thing says bwaa

[knotty current]

my tokens still havent recovered ima try again later

[swift storm]

I’m the only one that pays

Sadge

[tight tinsel]

bwaa bomb is so real

[swift storm]

Is there any good cloning alternatives to ElevenLabs

[fast pagoda]

i keep being stupid

so i have

claude max

currently

gemini pro too

and also chatgpt pro

[amber fractal]

Bwaa bomb planted on local model

[swift storm]

How much audio is required and do they require a captcha?

Yk what

I have Google

[uneven pulsar]

i am lacking so much thing to learn in java

[swift storm]

I can look myself

[uneven pulsar]

[swift storm]

Learn springboot

[fast pagoda]

how's sesame

[uneven pulsar]

honestly i hate the fact that my laziness eating me

[uneven pulsar]

?

[fast pagoda]

google just came out with some sort of voice thing

[uneven pulsar]

like i know python and node js but i am just lazy

[swift storm]

It’s a framework. You can learn it if you have nothing to learn for Java

[uneven pulsar]

i learned the basics tho

[uneven pulsar]

what dose it do?

[swift storm]

It acts as a frontend and backend framework

[uneven pulsar]

OOOO

[swift storm]

But I wouldn’t recommend the frontend though. It’s ugly af

[uneven pulsar]

okayyyyyyy

but still i do know python and ts node js and js uhh php and i need to learn java

[amber fractal]

Now we wait

[fast pagoda]

turn on gaming mode for bwaa

[uneven pulsar]

you use game genie?

[fast pagoda]

dont you have a ROG phone or something iggly

[amber fractal]

Yep

[fast pagoda]

that'll do it

speaking of phones i got android 16 today

[uneven pulsar]

rog phones doesn't even exist in my country

[fast pagoda]

cant wait to boot it up and it is no different

[amber fractal]

Charging bypass is legit a feature that needs to exist more often

[uneven pulsar]

we only got s25 knock offs and expensive iPhones

[amber fractal]

phone getting hot because of the battery charging and discharging? Just don't touch the battery stupid

[swift storm]

Why not just use open router

[uneven pulsar]

the good thing we use usd but the downfall we have limited stuff to ship and 5% expensive in America

[fast pagoda]

i use openrouter as well

[uneven pulsar]

why not fiber optic?

[fast pagoda]

lol

i just use that ffor API stuff usually

[swift storm]

Those LLMs are on there

[fast pagoda]

yes i know

i have passthrough for my api keys

[amber fractal]

Is it a crime to lock this 165hz display to 60hz at all times?

[fast pagoda]

even if i were to use openrouter it'd just bill to my openai platform//anthropic console etc account

[swift storm]

Gotcha

[fast pagoda]

not on a phone

i dont even have my pixel's screen at full res

[swift storm]

I would just hate putting my card on a million sites

[fast pagoda]

tthey all use stripe i think

so there's a single point of identity theft for me

lol

[amber fractal]

I'm not at 1440p so I'm fine with full rez on this

[fast pagoda]

plus i use proxy cards anyways

[swift storm]

You brave soul

[fast pagoda]

my identity is already out there because delta dental are a bunch of rat bastards

[swift storm]

Yeah I saw you on breachforums

[fast pagoda]

but hey they paid for one(1) year of id protection then said "k ur good i think"

[swift storm]

I mean what

[amber fractal]

Wait I can still use my phone, I forgot ASUS gave us a functional background apps interface

[fast pagoda]

lol

[swift storm]

Have you tried any local tts

[fast pagoda]

Gemini everywhere

i have not but there are people here that have

i mean maybe not right now but there's a few tts lords that come here

[swift storm]

I saw one that dropped like 2 months ago but doesn’t seem promising

[fast pagoda]

there was cesame

or whatever that was called

it was hyped for a minute

and orpheus

https://github.com/canopyai/Orpheus-TTS

[swift storm]

Interesting. Thanks

Never heard of these

[knotty current]

google gemini defeated the bwaa bomb

[fast pagoda]

lol

makes sense

big context

[knotty current]

but it got defeated by 4 bwaa bombs oops

[amber fractal]

I dropped down into 300 bwaas and I'm getting something

[swift storm]

Maybe in 5 years it will work

[amber fractal]

It completly has failed use the end of turn token despite being 7b

[fast pagoda]

https://huggingface.co/models?pipeline_tag=text-to-speech&sort=trending

i recognize kokoro

also Dia

[swift storm]

Yeah I think I came across kokoro before. it was alright

[fast pagoda]

https://huggingface.co/fishaudio/openaudio-s1-mini

this one was recent

my god

every video on these models on youtube

is so ultra mega slopped to hell

there's like no videos of just samples of it

https://youtu.be/s6r2IQGuAeg?si=X1aPMANs29w23Ucq&t=93

timestamped

[amber fractal]

you know what that means

[fast pagoda]

slop time?

[noble zodiac]

surely they got consent from all those people, right?

[swift storm]

Wait that’s cool af

You can indicate when to switch tones

[fast pagoda]

surely

[fast pagoda]

yeah it looks pretty good for a local model as far as i know the expressiveness isn't common or even may be new to a base local model

[amber fractal]

No bomb smh

nvm this rez sucks

[fast pagoda]

lol

bro started doing both turns

id say that's a success

he definitely broke

[amber fractal]

Yeah it got destoried

cut that one off quickly

Apperantly mobile just doesn't want to see the image ifdk resending it

[fast pagoda]

but have you considered The reason pi isused in theformula for thearea ofthecircle is because thearea ofa circle is proportional tonr^2, where r istheradiusofthecircle, and A=Ttr^2, where A is thearea ofthecircle, and It is theconstant thatappears in theformula

[olive sable]

I didnt end up sleeping, but goodmorning

[fast pagoda]

welcome back

[amber fractal]

[olive sable]

Its wednesday my dudes

https://tenor.com/view/happy-wednesday-wednesday-hump-day-gif-7217505363683747237

[opaque wharf]

I want to see some suffering

[olive sable]

Im not doing python

Its already in python

[amber fractal]

Weird, this one is not dying yet

More

[opaque wharf]

Js

[olive sable]

The frontend will be yes

[opaque wharf]

The engine

[hoary lion]

me when there is no implementation of architecture because I can't use torch

why my convolutional layer is breaking apart

[olive sable]

https://tenor.com/view/bugs-bunny-bunny-no-nope-nou-gif-13480425249407915803

[knotty current]

why doesnt mine look like that

[fast pagoda]

i love diagrams man

forgot i called the thought detector the thought police poor echo

[amber fractal]

0.5b either is normal or nopes out

[fast pagoda]

that sounds about right

they get overwhelmed easily

[amber fractal]

no response at all (weird)

[fast pagoda]

i have tried 1b and 0.5 for echo

they get the stupid funny part down because theyre stupid

dont listen tho

[amber fractal]

we wait

[fast pagoda]

phi will output the entire sourcecode of windows 3.1

in the response

[amber fractal]

Lobotomy

[fast pagoda]

lmfao

yep

im glad the guess was right

he even tried to say lobotomy

[amber fractal]

Maybe less bwaa

[amber fractal]

That's why I thought to say it ngl

[fast pagoda]

you got prompted

yes please analyze this high quality data, gemini

mfw that harborfreight link got into the 4s and 5s here

nice

thanks

huge data

[amber fractal]

Alright, I think this is the correct amount of bwaa

[fast pagoda]

agi achieved

[stark needle]

Morning

[amber fractal]

Wait a minute

[amber fractal]

[maiden geyser]

why do you disagree

[fast pagoda]

vomits training data

[lament igloo]

Qwen... ew...

Sorry but that has to be worst LLM that i ever used

[amber fractal]

That is the goal 👍

@fast pagoda

[tight tinsel]

bwaa language model

[amber fractal]

Yep

[tight tinsel]

now at 4/20 of the shit I need to do

[knotty current]

@amber fractal how many bwaas, mine is still alive?

[amber fractal]

I started with 1200 and went down from there, let me see

[tight tinsel]

bwaagpt, count how many bwaas are present

[amber fractal]

About 800ish

[knotty current]

i just slapped that many into google gemma and its not dead

[amber fractal]

Larger models require more bwaa

[knotty current]

im using gemma 1B

[amber fractal]

Oh then yeah, my 0.5B either died or didn't

[knotty current]

how is it still generating coherent responses with 4k bwaas !?

ok it died at 8k bwaas this doesnt make sense

[amber fractal]

Yep, that is not english

I'll let it cook

[knotty current]

over 4 times the bwaas for a model double the size

[amber fractal]

I have no clue what this is saying mhm, but I know it is random topic.

[fast pagoda]

thanks phi for the letter

[knotty current]

peak hallucination

[maiden geyser]

you bwaad it so hard it started speaking french

[stark needle]

bro became french

no way

[stark needle]

it's writing a letter thanking for your historical findings and expertise

[amber fractal]

Another prompt, another french (nope due to politics)
It actually stopped what?

[stark needle]

trying to make something better than neuro voice v2 is hard

however

i can fix via RL

[hoary lion]

niuh, hi shadow

[hoary lion]

are you going to apply that llm based voice synthesis w/ GRPO

[stark needle]

no

[hoary lion]

just genuine rl then?

[stark needle]

ye

[hoary lion]

is it custom architecture

im hungry for tts architecture rn

like super

[faint sandal]

Word's autocorrect feature is awful

[amber fractal]

AI powered

[stark needle]

Swamps is obviously the correct word👍🏻

[hoary lion]

Neuro like yapping

Time to eep gn

I will choose transformer or gan at my dreams today for sure

[rich kestrel]

holy shit. guys i might be celebrating way too soon but i think i fixed it. i think i fixed the issue in llama.cpp that i think may have caused neuro to go off the rails following her recent memory upgrades

(it was a one line change)

[stark needle]

There's no way

There's ai generated slop billboards here now

[uneven pulsar]

bro.........BRO

i have to know how to install java packages to just make a user input

BRO

[dry charm]

huh

[dry charm]

You can use System.in.read()

[uneven pulsar]

oh ok

thx

[dry charm]

Dependencies in Java is a whole another beast

https://www.geeksforgeeks.org/ways-to-read-input-from-console-in-java/

[uneven pulsar]

understood

thx really

[stark needle]

Chat i may have done it

Visual memory

Infinite

Long term

[ruby timber]

Agi??

[rigid snow]

asi

[uneven pulsar]

the halt condition?

could it be?

[stark needle]

As long as there's disk space

It will work

[knotty current]

Visual memory for AI models?

[dry charm]

Everyone keeps talking about AGI, but not STR, LCK or INT

[stark needle]

i will make externally accessible camera today afternoon and test it in the city

Tomorrow

Also im working on a model that basically

Mirrors eye focus

[unkempt citrus]

I'm more a WIS and CHA guy myself

Maybe some DEX

and INT

[rich kestrel]

done and dusted! pull request #14120 on the official llama.cpp repository. idk if this rly was the issue with neuro but on the chance that it was can someone bring this to vedal's attention

[dry charm]

we do not even know if he uses llama

[rich kestrel]

ik but this bug has been bothering me for a while so i honestly just used this as an excuse to finally fix it so i can stop using the dumb oobabooga server that doesnt even expose the names of the models on the api

and if it helps neuro thatd be real cool :)

[tender river]

C# if you want basically python with types (the code wont be very different), C++ if you want basically python with types but a bunch of segmentation faults, Rust if you want a C++ without segmentation faults (it will teach you the same memory management techniques at the end but some bad code that wont compile in Rust will work fine in C++ and only cause crashes under certain conditions)

i dont think C++ is a good learning language, but it will be easier to use because of that

[fast pagoda]

yeah i think it's running good yall no probleems anticipated after this one

[ruby timber]

stdint mentioned

[rich kestrel]

i love rust but rust for a game engine is an idea that hurts just to think about much less implement 😭
borrow checker flashbacks....

[primal siren]

Bruh. I gave the trainer 29 .wav
As a test his answer?

Him: You see 010.wav?
Me: yeah?
Him: I will only use that one.

[ruby timber]

Use zig c but better

And safer

[rich kestrel]

ive been using zig for the first time recently in a games project! its a very very nice language. i hope it supplants C. i hate C type syntax. zig's type syntax is just so much clearer

[tender river]

anyway heres a basic wgpu thing that just renders a shader and passes some data https://github.com/neuro-arg/graphics-toolbox/blob/master/src/lib.rs

[primal siren]

For the .csv for the voice training. Of the 3 different collumn
Like
ID | something | something

I know the first is the .wav ID the transcript will be linked to.

But is the the transcript in second or third position.
And whats the last one supposed to be?

[fast pagoda]

[stray dragon]

i think you need more ram

how about getting 192gb

:)

[dense cosmos]

if I may, I recommend Cherno's series on making a game engine in cpp

[fast pagoda]

i do be needing more ram

[rich kestrel]

oh my god

[fast pagoda]

absolutely fine

not even close

gonna get this training run going
i should probably do a fresh install of llama-factory if i'm gonna use

37 hours later

"why are my linux directories so huge"

[sour harness]

Now multiply by 5 because you need different environments for every tool because of the absolute state of the python ecosystem.

[fast pagoda]

yep this is like the 3rd copy of these

in this install

and i installed it like 2 days ago

that's how my ubuntu got to 130gb and i barely used it

mfw this cmake isnt compatible with python 3.13

[sour harness]

43G    venv
43G    total```

[fast pagoda]

portable

[sour harness]

989G    models
989G    total```oof

[fast pagoda]

my models folder was 2tb

i think i cleaned some out though

kek

found it

[sour harness]

Funny how that small folder called "endme" on the bottom is getting crushed by all those huge blocks

[fast pagoda]

it's a folder containing my angry attempts at getting something working probably

if the name is like that, that's usually what it is

[prisma spoke]

Hello I'm new here

[fast pagoda]

much more reasonable theses days

[prisma spoke]

How can I build a speech model

[unkempt citrus]

with great difficulty

[prisma spoke]

That I know

[fast pagoda]

https://huggingface.co/docs/transformers/en/tasks/text-to-speech

[prisma spoke]

@fast pagoda thx

[olive sable]

[olive sable]

[olive sable]

Im getting so many conflicting recomendations lol

The thing with C# is ill be learning it during lesson hour in college in a few months, so idk if i want to do that outside of lesson hours

So ill probably go rust or c++ instead first

[tender river]

languages have strong and weak points, i just think that its best when a language teaches you something new even if it forces you to write code differently, and thats why i dont think C++ is a good fit (it wont teach you, you just have to know what you're doing)

[olive sable]

Im probably gonna go with rust for now then

[opaque sigil]

Clearly the solution is Haskell

[tender river]

either way C++ is good to know too but best C++ programmers wouldnt have trouble writing Rust either because they use the same rules when writing C++

[sour harness]

degrasse_whoa_we_got_a_badass_here.jpg

[olive sable]

Their memory allocation system seems complicated in rust, but ill see

[tender river]

its the same as C++, but rust guarantees it doesnt crash

[olive sable]

I should just write raw web assembly

[tender river]

2025 is the year of haskell in the game industry

[olive sable]

Wth is a haskell?

[tender river]

i know that the industry isnt ready for the frontier like algebraic data types, linear types, dependent types, but we gotta teach them what works

[tender river]

a language based on typed lambda calculus

[olive sable]

Lambda calc?

[tender river]

lambda calculus was born out of research into logic and computation

[olive sable]

Ye nah lmao

[tender river]

it only has lambdas and you "call" them, thats it

this is all there is to it

the beautiful thing is you can compute anything using these rules

[olive sable]

I dont really feel like doing that one

[tender river]

obviously haskell has a bunch of code for numbers and stuff that you dont have to build, and its typed which means some lambda calculus concepts cant be expressed in it

but the most important thing for you is that it uses a GC so it would be kinda pointless in this case

[opaque sigil]

as fascinating as lambda calculus is for how simple it is, i absolutely do not miss it

[tender river]

did lambda calculus kill your parents

[opaque sigil]

yeah

[tender river]

unlucky

[ruby timber]

Isn't haskell for maths

[tender river]

i wouldnt say that

julia is for maths

[ruby timber]

But it's mostly used for that right? I don't rember

[tender river]

dont think so

[ruby timber]

I might mistake it for another language then

[opaque sigil]

don't a good amount of banks/financial institutions use haskell for their backend/trading

[tender river]

haskell is mostly used for programming the haskell compiler

[tender river]

yes many fintech companies love FP

[opaque sigil]

mfw jane street pretty much singlehandedly keeping ocaml afloat

[tender river]

true

[noble zodiac]

the lower you go in the banking stack the higher the odds of seeing cobol

[opaque sigil]

hey, those mainframes were expensive

[noble zodiac]

didn't the FAA recently announce that they are planning to get rid of their floppy disk systems?

its pretty normal for old and very important systems to stick with whatever they got it to work with initially

[opaque sigil]

yeah

[tender river]

hmm tough choice do i pick delivery for 19-21 or 19-22

[prisma spoke]

Someone pinged me i dont know who

[rigid snow]

the even lower you go down the stack the higher the odds of seeing machine code and semiconductors

[sly wind]

Hey Vedal, big fan of Neuro for a while (mostly from the bilibili clips lol)!

I'm the founder of Fish Audio. I saw you've been testing V3 voice solutions. We've been working on creating more emotional and controllable AI voices for future AI Companion, and our latest internal model is a huge step up from our public S1.

If you're interested, we'd be happy to share the 4B proprietary model's weight and help fine-tune it & do some rlhf. We had an idea to potentially use feedback from the you & core fans to create a voice everyone loves.

If you think it could be helpful let me know.

Sorry for the random message lol

[sour harness]

You might want to send that to his business mail instead ([e-mail]). I don't think he checks here that often.

[sly wind]

lol, sounds good

[sour harness]

(Also I'd kill for the 4B model. Emotional expression and non-verbal speech stuff is so rare.)

[sly wind]

[opaque wharf]

Yeah, until you learn that a register is not a "place" in the cpu die on x86

A register would be a file on minix iirc

[rigid snow]

how does that contradict what i said

[opaque wharf]

Machine code is still "interpreted" again before seeing the hardware

And no, its not strictly contradiction

[rigid snow]

i didn’t say machine code on the same level as semiconductors

[tight tinsel]

got the hdd I needed for half of what it normally costs where I normally live + free cable and some screws

500gb hdd for I think roughly 12-13 usd

[ruby timber]

Huge

[opaque sigil]

[amber fractal]

[stray dragon]

[warped narwhal]

[tight tinsel]

[quick flame]

[maiden geyser]

[gritty dust]

[olive sable]

[sage crag]

is this some kind of #swarm

[ruby timber]

So called free thinkers when

[gritty dust]

Broke the chain smh

[tight tinsel]

programming when half priced hdd mentioned

[olive sable]

i recently had to get a new usb stick

64GB, a full 9 bucks for 32mb write speed tho

pretty slow

[knotty current]

@tight tinsel the half price HDD you got is from china right? (i might be wrong, i inferred this from the mention of a shenzhen electronics mall)

[tight tinsel]

yes