a little too based but damn
#programming
hard delta
flint dome
any software is hackable do pen and paper
complicated microphone setup and only one monitor?
split steppe
can't be infected if the content comes from the cd, right? those damn hackers can't infect my cd
actually, didnt the og xp come on like 100 floppy disks, i don't remember anymore
flint dome
where h shifter :catAsk:
young plover
I'm a proud American automatic driver 
I do use the paddle shifters for fun occasionally, but the only racing game I play is Forza and when I need to actually win I use a controller. 
flint dome
I'm a proud American automatic driver <:SMILE:1173553648534568960>
I do use the ...
you should be shot
flint dome
mugged in broad daylight too in the middle of the fucking street
flint dome
violence in #baking
violence? what violence? we're simply cleaning up, my friend
cant let the brownies get bad now can we
young plover
Hard to see in this picture but the peddles are just sitting on the cardboard box that the secretlabs footrest came in. Needed to get them off the floor so I don't strain my ankles.
flint dome
as long as the box dont break its acceptable
you could have your pedal on the god damn dash and nobody would say shit
young plover
It do be a sturdy box. Need to bolt the peddles down or something so they don't slide.
flint dome
slap some double tape and call it a day
patent shard
proven merlin
I'm a proud American automatic driver <:SMILE:1173553648534568960>
I do use the ...
You see, i use a Keyboard if i have to win, and a wheel when i want to have fun. We are Not the same.
young plover
keyboard?
bro I'd rather use tilt controls
proven merlin
keyboard?
bro I'd rather use tilt controls
Yes, im best on Keyboard. Its what happens when you play Racing games for 15 years on kb
I also occassional do use kinect fake wheel
But mostly to troll
fiery girder
Yes, im best on Keyboard. Its what happens when you play Racing games for 15 yea...
15 years¡
sick owl
Apparently Kimi K2.6 is more sycophantic than 2.5
That's a shame, I really liked how 2.5 was willing to be downright harsh if the situation called for it
silent cloak
I have more than your lifetime of stealing
silent cloak
opaque sigil
the authentic linux experience for windows users 
silent cloak
Devs implementing new methods into the linux kernel to break the sound more optimally
split steppe
the authentic linux experience for windows users <:neuroPogHD:145885055997614495...
windows is finally catching up with linux 
sage crag
the authentic linux experience for windows users <:neuroPogHD:145885055997614495...
Pulse For Windows when
opaque sigil
technically exists
sage crag
fiery girder
<:neuroD:1134963217676906606>
Yo bunny, u using my emoji? No way. Who did that to you?
sage crag
Yo bunny, u using my emoji? No way. Who did that to you?
ive been using this emoji before you joined the server
fiery girder
But it's mine and I had stolen from u. So you can't use that. Hmph
fiery girder
Nooooo
[And he died with that. Though the bunny doesn't know that fact. But He was bitten by a rabbit at a very young age. He has PTSD from rabbits.]
sage crag
warped narwhal
bunny or rabbit <:neuroCatUuh:1131629092475768884>
Food, food, or food? 
sage crag
nrr
fiery girder
Food, food, or food? <a:KuryuminSmug:781502710926540840>
Rabbits
stark needle
🐇 
fiery girder
Yë
stark needle
@sage crag this u?
sage crag
fiery girder
Revenge. You bit my finger. Now I bit upon your leg piece!!!
fiery girder
Wait, I never meant you
I never could
That's way too absurd
Nooo
Nooo
Nooooooo
sage crag
fiery girder
sage crag
brain damage
fiery girder
[2nd death. You only have 1 more death point left]
[Bunnies Special Skill: Brain Damage levels up from lvl 3 to lvl 5 for double kill]
stark needle
☆♡《♥bny🐇lop•idol~★》♡☆
kind fable
glass flower
copilot fix npm, no mistakes
glass flower
also kinda ironic when they just changed the way copilot works on github..
jagged turtle
throw vscode marketplace with npm
altho they did release staged publishing to npm, finally
sick owl
Yeah, turns out LLM can actually do some rudimentary math <:enub:116391223058523...
It certainly can
opaque sigil
actually
is google's math-focused gemini available to the public at all
hmm looks like deep think is available but not the finetuned model they used with it 
opaque wharf
hmm looks like deep think is available but not the finetuned model they used wit...
Yeah, I used the deep think gemini flash
amber fractal
Perfectly aligned
opaque wharf
https://tenor.com/view/what-are-you-challenging-me-scooby-doo-shaggy-matthew-lil...
This is how I imagine your face btw @fast pagoda
mighty thorn
Why do people feel the need to pretend they are all knowing and use it to justify absolute nonsensepilled opinionslop
fast pagoda
This is how I imagine your face btw <@146797401720487936> <:enub:116391223058523...
lololol
opaque wharf
Giving those shaggy and scooby doo energy
hard delta
Why do people feel the need to pretend they are all knowing and use it to justif...
the template of the argument this guy used is correct, but they managed to fill it up badly
mighty thorn
the template of the argument this guy used is correct, but they managed to fill ...
LMAO
maiden geyser
Why do people feel the need to pretend they are all knowing and use it to justif...
don't know how to write a resume? don't worry, you won't be accepted anyway
restive cloak
don't know how to write a resume? don't worry, you won't be accepted anyway
Boom
fickle rain
One turn BTW
restive cloak
Why do people feel the need to pretend they are all knowing and use it to justif...
I as an Ai user myself unfortunately wanna say that if i face a real programmer i will do “it” in myself def gross and it = “pepe”, print(it)…in short the most that i can do alone is some small python stuff with no libraries import, I’m no match to any kind or programmers over there alone
hard delta
10 + 9 = i'm having dinner with ur mom tonite
restive cloak
A wild response
amber fractal
https://arstechnica.com/ai/2026/05/study-ai-models-that-consider-users-feeling-a...
Wow, imagine the dataset where being emotional is not always logical
fickle rain
Legendary slop-driven debugger
(unfortunately I have to run it as sudo because I'm using qemu:///system...)
proven merlin
Hmm guys, is almost 5 Minutes for a pragmata grid answer fast enough? 
(to be fair, 4B qwen 3.5, 11300 tokens, at 42t/s.. not much hardware to push it with lol
Shaved off a minute.. for a worse result. They truely do struggle with this huh
hard delta
according to those prices my pc has a z890 mobo, an u265kf, an 5070ti and 2x16gb of ddr5 ram (and some other stuff) for $910 
but those might be free cause it's a 990
restive cloak
The room is becoming hot today
nocturne olive
fast pagoda
if todays prices are believed my pc is worth almost 9k
which is nowhere near paid amount
we are so cooked
restive cloak
Hey guys aside of that, if you want to fine tune an ai to be near human personality (not too warmth, not too aggressive or pg13+, able to say yes or no when needed) how will you do that, i mean how will you adjust the data set for that?
mighty thorn
Hey guys aside of that, if you want to fine tune an ai to be near human personal...
Like
Like make it act less ai assistant like in general
Or like
Have it act like a specific person?
restive cloak
Have it act like a specific person?
Imagine it as your coworker rather then an assistant
mighty thorn
Imagine it as your coworker rather then an assistant
Uhhh
restive cloak
Not the usual “yes dave i can help you with that just bla bla bla”
mighty thorn
I’d just make my own dataset
Like
Get the prompts from no_robots
Then answer them all yourself
Then train on that
restive cloak
no_robots?
mighty thorn
no_robots?
Is poggies dataset
nocturne olive
Surely there isn't some sneaky race condition hiding in here ```kt
if (!isSettingUp) {
isSettingUp = true
// Don't block GUI thread for environment setup
Thread {
EmbeddedTerminalTtyConnector.terminalPrint("Validating environment")
val environmentLocationCpy = environmentLocation
if (environmentLocationCpy == "Unset") {
// TODO: Tell the user to set the environment location
isSettingUp = false
Thread.currentThread().interrupt()
}
val environmentFile = File(environmentLocationCpy)
if (!environmentFile.exists()) {
// TODO: Send warning that the selected location does not exist
isSettingUp = false
Thread.currentThread().interrupt()
}
val pythonCommand = if (pythonVersion == "No Python found") {
TODO("Popup prompt for a String")
// TODO: Here be a popup asking for a Python path
} else {
PythonConfiguration.pythonCommand!!
}
EmbeddedTerminalTtyConnector.terminalPrint("Creating new virtual environment")
val process = ProcessBuilder(pythonCommand, "-u", "-m", "venv", "DiffTrainer-Kt")
.directory(environmentFile)
.redirectErrorStream(true)
.start()
hookProcessOutput(process)
process.waitFor()
// TODO: Download tools, initialize environment
localPythonConfiguration = PythonConfiguration.initializeNewEnvironmentConfig(
configurationLocation,
Configuration(VenvType.Python, environmentLocationCpy)
)
isSettingUp = false
}
}
I wonder if it can actually do fine with good debug flow (OFC not good enough for actual driver that'd run on more than 3.5 VMs, but good enough for my usecases)
true hemlock
if todays prices are believed my pc is worth almost 9k
my net worth deadass octupled
but i don't even benefit from it
silent cloak
true hemlock
because i can't sell my 768gb ddr5 rdimm ecc
i feel the feeling of a property owner
silent cloak
true hemlock
when their property price skyrocketed but they can't sell their house because they live in it
fast pagoda
when their property price skyrocketed but they can't sell their house because th...
sell it -> price go up -> destitute
fast pagoda
just sell it and buy a 24gb unified memory macbook
this happening would be a punishment never deserved by anyone
maiden geyser
when their property price skyrocketed but they can't sell their house because th...
same person who voted "no" in the "is someone with 10k$ rich" poll btw
high brook
Forth German Evil test
Halp this song is stuck in myhead
-# AND I CANT EVEN UNDERSTAND IT
fast pagoda
sorry i just realized i never responded to this. it depends but you want to use something that is not too incredibly different to the style you are targeting. no robots is good for avoiding it killing ur model's style, so is probably a good start. ideally you use hand written or highly hand curated data that is tailored specifically to the model and use case you are looking for so it teaches the model the intended behavior without blowing ueverything else up w/ noise
opaque sigil
Forth German Evil test
if you told me this was half german half english i would believe you 
proven merlin
Forth German Evil test
The accent lmao xD
It definitively isnt a clean Cover but if i dont pay too much Attention i can decypher it mostly
opaque sigil
I guess it's the accent huh
proven merlin
Accent or pronounciation or perhaps slurring? Something like that
silent cloak
https://i.redd.it/vsxldcuri1se1.gif
So true
high brook
The accent lmao xD
lmao. i need more German Data for it.
I'm using a very small set of german samples
but trust me this is progress
proven merlin
Yeah its definitively useable/listenable and understandable
high brook
Xoda disagrees
opaque sigil
Yeah its definitively useable/listenable and understandable
idk about that one, though i also don't know the song in the first place so maybe it just is like this
proven merlin
Hm.. i wonder what works for datasets, would Kraftwerks songs work? Perhaps Something from Die Printen or other old german hit songs? Or what kind of data you need?
proven merlin
idk about that one, though i also don't know the song in the first place so mayb...
Well i can understand like.. 70-80% of the Lyrics so like.. decent.
proven merlin
oh the issue is the data has to be labeled
So like, as example, lyrics with audio timestamps?
high brook
sec
like this 0 10512472 SP 10512472 14521542 AP 14521542 15519275 j 15519275 17532880 eh 17532880 17922904 d 17922904 19691610 er 19691610 20798186 m 20798186 24163264 o 24163264 24979592 er 24979592 25678004 g 25678004 27582766 ex 27582766 29587302 n 29587302 30539684 sh 30539684 31192744 t 31192744 32199546 a 32199546 33188208 er 33188208 34213152 t 34213152 34802720 ex 34802720 35700680 q 35700680 36244896 m 36244896 37061224 i 37061224 38086168 t 38086168 38739228 n 38739228 39809524 e 39809524 40789116 m 40789116 41551020 l 41551020 43591836 a 43591836 44471656 ng 44471656 45024940 ex 45024940 45587300 n 45587300 46721088 k 46721088 49242632 u
NNVS style lab files
silent cloak
Buh
high brook
proven merlin
Oh, so even smaller, prefferably phonemics?
high brook
i uploaded the dictionary file
silent cloak
high brook
but yeah honetly neuro and evil synths are solid POC at this point i just need more data labeled in german english and chinese
fast pagoda
i love oil skelly
silent cloak
i love oil skelly
Hell yeah
I used to play the game back in the day and this channel does a lot of source code dives
proven merlin
Hm .. as a uneducated person, im surprised about the lack of Umlaute
silent cloak
So its cool to see
fast pagoda
oh this is about the game heh heh
im unc i forgor
back in the day when scp 1-1000 was being created
opaque sigil
Hm .. as a uneducated person, im surprised about the lack of Umlaute
this is a level below I think, training on the actual sounds
fast pagoda
luv me oil skeleton man
luv me shy guy
luv me indestructible alligator
simple as
proven merlin
Hm .. as a uneducated person, im surprised about the lack of Umlaute
Tho ig u do have ae instead of ä
And so on
high brook
Hm .. as a uneducated person, im surprised about the lack of Umlaute
oh. this is basically an arpabet style system
fast pagoda
arpanet
opaque sigil
arpabet is just ipa but ascii iirc right
high brook
no. that's X-sampa
opaque sigil
silent cloak
oh this is about the game heh heh
The game is ancient too atp
fast pagoda
containment breach huh
proven merlin
oh. this is basically an arpabet style system
Id love to understand this more but im completly clueless and cooked rn.. imma sleep- tho if theres Something German happen feel free to Ping
fast pagoda
scp wiki started in 2008 
silent cloak
scp wiki started in 2008 <:wilted_rose_4k:1495545866751049768>
Time flies
silent cloak
Must have a ton of stock in nvidia
fast pagoda
i wish i had no soul
maiden geyser
i wish i had no soul
have you tried applying to raytheon
fast pagoda
Is a prerequisite, not a, uh, side effect
silent cloak
Do what the guy in the vid above does
fast pagoda
still a prerequisite to get started
maiden geyser
forget about it, son, a slap is all you need
fossil bay
maiden geyser
<:neuroWave:1090851734450077787>
viral oasis
real quail
https://cdn.discordapp.com/attachments/961105387019456544/1455262183431864320/ki...
REAL
kind nimbus
Forth German Evil test
Lol 
olive sable
opaque wharf
I have finished the horror of a website that needs to use a lot of curves
reef ingot
I quit programing
umbral wigeon
Y
amber fractal
I quit programing
F
sage crag
olive sable
opaque wharf
I quit programing
The pipeline from programmer to farmer is real
reef ingot
The pipeline from programmer to farmer is real
I am going to do gunsmith stuff
high brook
any german speakers with IPA knowlage want to look this over?
trying to map some stuff for phonemes
floral hawk
We gone from "how to set up whisper" to "fine tuning vision models" for minecraft just before twitch launch. What a journey
I suspect Zero will be live within the month if all goes well and final debug isn't too meaty
Very little technical debt so far, so I'm feeling cautiously optimistic.
alpine cloud
I am going to do gunsmith stuff
if i werent a programmer, i'd probably be a blacksmith or carpenter. gunsmith does sound appealing too though
minor trout
Multiple LLMs: The "Brain Registry." Instead of one AI doing everything, you route tasks to specialists. A massive, smart model (like Claude 3.5 Sonnet or GPT-4o) handles complex logic and coding, while a smaller, lightning-fast local model (like Llama 3 or Kimi) handles casual chat and quick filtering.
Emotion System: The "Internal State." This adds a layer of synthetic empathy or persistent personality. The AI tracks its own mood, sentiment, or relationship dynamic with the user. Its prompt instructions dynamically shift depending on whether the conversation history indicates stress, excitement, or casual banter.
Symbolic AI: The "Anchor of Truth." Standard LLMs struggle with perfect logic, hard math, and strict rules. Symbolic AI integrates classical, rule-based logic (if/then parameters, graphs, or knowledge bases) to keep the LLM grounded. It ensures the AI follows strict boundaries and doesn't "hallucinate" critical details.
World Model: The "Simulation Room." A world model allows the AI to predict the consequences of its actions before it takes them. If you tell it to organize a folder or write code, it simulates the outcome internally first, ensuring it won't break your system or cause unexpected errors.
VLM (Vision-Language Model): The "Eyes." This allows the AI to parse images, look at drawings of your custom characters, or read charts. When connected to tools, it lets the bot "see" a desktop screen or a user's uploaded art and interact with it intelligently.
OpenClaw: The "Central Nervous System." OpenClaw acts as the background daemon runtime that links everything to your daily life. It handles long-term memory (storing context in plain Markdown files), runs background loops (proactive heartbeats), executes system tools, and handles the messaging pipeline.
does this plan look fine to you people?
floral hawk
This is pretty much the standard, but running like 4 models is going to cause more harm than good imo
minor trout
i have pretty good hardware
floral hawk
That's not why
minor trout
MoE, not running all of them at once
Mixture of experts
calling the one that would handle the task the best
floral hawk
You have the right idea, but if you're set on such a convuluted setup the codebase for this is going to be atrocious to write
But you also don't need essentially "subagents" for everything
true hemlock
floral hawk
A robust codebase can make one model extremely powerful if written correctly.
true hemlock
"symbolic ai"?
floral hawk
We did, it doesn't make sense in any practical way.
minor trout
why does people not know about it
stark needle
I think your plan is built on a shaky foundation and a wrong premise
true hemlock
its just not being implemented anywhere aside from being mentioned in papers and such
also why would you emulate emotion
olive sable
wtf is symbolic ai
switch cases
minor trout
thought it might be better to have a proper system than tellling the llm to act
true hemlock
also why would you emulate emotion
(yes, using current ml models to "simulate" emotions are just emulation)
floral hawk
Zero can play games, read, tool call, speak, control OBS/Twitch and do web searches with 1 main model, a vision model, and a micromodel for "emotional" interpretation.
fast pagoda
symbolic ai lead to the ai winter
minor trout
Zero can play games, read, tool call, speak, control OBS/Twitch and do web searc...
what is "Zero"
true hemlock
Zero can play games, read, tool call, speak, control OBS/Twitch and do web searc...
you could do function calling on just one model yeah
floral hawk
My own stack. Just an example
true hemlock
pretty much
olive sable
just start by explaining how exactly you are gonna possebely implement a symbolic ai system to check if the MoE is hallucinating or not?
its just not possible
floral hawk
Its also an 8b model capable of all of that through literally just a strong foundation. You dont need 4 models.
minor trout
the llm uses it, the symbolic ai doesn't check the final output of the llm
olive sable
how does it use it?
stark needle
the llm uses it, the symbolic ai doesn't check the final output of the llm
Do you know what symbolic ai means
minor trout
Do you know what symbolic ai means
if else statement ai that has only logics and symbols
it does not know metaphors
or common sense
olive sable
so how are you planning to make a chain of switch cases make sure the llm doesnt hallucinate?
fast pagoda
neither do llms
minor trout
neither do llms
they know metaphors at the very least
rough bloom
switch cases
this is like saying that LLMs are "just addition and multiplication", it's completely missing the point
symbolic AI still shit though
minor trout
shty things + shty things is either good or an even worser sht
floral hawk
What is your plan to even coordinate all of this as a stack?
minor trout
the MoE generates candidate answers in raw JSON formats, an independent rule-engine validates those tokens against an immutable relational database of world constraints, and it automatically triggers an internal self-correction loop if a logical contradiction is detected.
stark needle
they know metaphors at the very least
Llms predict the next index of a specific tokenizer hence a number, they don't know metaphors
minor trout
Llms predict the next index of a specific tokenizer hence a number, they don't k...
they don't truly know
they just predict
olive sable
the MoE generates candidate answers in raw JSON formats, an independent rule-eng...
i see the idea now, but how does this prevent hallucination?
stark needle
Yes then they don't know metaphors
minor trout
It doesn't stop the MoE from thinking a hallucination; it stops the system from accepting it. The Symbolic AI acts as an un-gaslightable boundary controller. If the neural network hallucinates a fact that violates our hard-coded database logic, the symbolic system catches the error, blocks the output, and forces an internal rewrite before the user ever sees a single token.
stark needle
the MoE generates candidate answers in raw JSON formats, an independent rule-eng...
What would a relational database of world constraints look like? How would it match raw text into a structured Format and perform validation
olive sable
if your LLM generates a bunch of false statistics, how is yout self-correction loop going to validate whether its correct or not?
floral hawk
What is the use case exactly
minor trout
We use Structured Generation tools to force the MoE to wrap its narrative actions into a precise JSON schema. Our Python backend extracts those fields and runs a standard SELECT query against an SQL database housing our world rules. If the database query returns a logical contradiction—such as a character profile violating a foreign-key constraint or a level requirement—the script throws a validation error, blocks the message, and triggers an automated prompt-retry loop
stark needle
—
rough bloom
—
olive sable
—
split steppe
—
faint sandal
erm dash
stark needle
We use Structured Generation tools to force the MoE to wrap its narrative action...
Do you know what a foreign key constraint is
jagged turtle
what in the hell am I reading
split steppe
man is reinventing symbolic ai, without ever having read on the history of it
falling for the "i'm just going to have a fact database" trick
jagged turtle
man is reinventing symbolic ai, without ever having read on the history of it
relax he's clearly inventing a new branch of it
jagged turtle
falling for the "i'm just going to have a fact database" trick
oh boy
surely that works
and won't hallucinate details anyways
split steppe
if you have infinite monkeys it works i guess, we've been trying to do it since the invention of AI
there's simply too many rules, relationships and facts in the world to write them down
jagged turtle
how do you define something as "fact"
split steppe
we stopped at IBM Watson and gave up
jagged turtle
what we think are "facts" can change
olive sable
We use Structured Generation tools to force the MoE to wrap its narrative action...
let me put it this way.
if your LLM generates that X% of the population has y condition.
how is your box full of if-else statements going to verify whether its correct?
it cant
split steppe
that as well, it's a never ending hell of recording trillions of items and relationships and keeping them all up to date
jagged turtle
why does this read like a self-bot/AI response with the emdashes, overly formal writing and "we"
split steppe
theres a reason we ended up at LLMs, it kind of sort of does that semi efficiently from random text
olive sable
the "we" is annoying to read ye
woven temple
there's simply too many rules, relationships and facts in the world to write the...
There surely aren't more than like 20
rough bloom
why does this read like a self-bot/AI response with the emdashes, overly formal ...
because it is 
rare bridge
why does this read like a self-bot/AI response with the emdashes, overly formal ...
i think they're just copypasting from chatgpt or gemini
jagged turtle
i think they're just copypasting from chatgpt or gemini
so nailed it with the AI response
also hello cloud
bitter phoenix
Retrying prompts at runtime is what you wanna avoid
rough bloom
cloud
rare bridge
floral hawk
You guys get people like this alot huh
olive sable
"gemini help me argue about why my idea is the best eventho it fails to work propely in practise"
floral hawk
Im newer here so idk.
jagged turtle
does this read like "immutable db of facts"?
split steppe
have you noticed how many people are showing up here that think they're about to create superintelligence and are also like "my first day programming guys, wish me luck! chatgpt says i'm so smart tho"
floral hawk
I was thinking that too
jagged turtle
You guys get people like this alot huh
at least once a month on average
rough bloom
You guys get people like this alot huh
the consequences of #baking being moved to right below #general-network 
floral hawk
I mean I came here for AI help and because Im a fan or neuro but I certainly wasnt trying to reinvent the fuckin wheel
"Gender Calculus"
jagged turtle
what I'm reading is basically AI trying to course-correct AI
rough bloom
olive sable
Does the database just have the whole world or sum
you need to understand, he has the hardware nexccesary he siad, so he has 2 trillion H100s needed to calculate planet earth to the accuracy of each electron
woven temple
have you noticed how many people are showing up here that think they're about to...
You just don't get them. They are innovative free thinkers, like Elon. Meant to lead others to do the work on their very special and unique ideas 
jagged turtle
bro could make more money selling those H100s if true
floral hawk
you need to understand, he has the hardware nexccesary he siad, so he has 2 tril...
I was simply
Unfamiliar with his game
rough bloom
you need to understand, he has the hardware nexccesary he siad, so he has 2 tril...
if I understood the ChatGPT response correctly it's just for narration? so when some story event happens you just save that in a database and check future events against that
jagged turtle
if I understood the ChatGPT response correctly it's just for narration? so when ...
I don't get this
rough bloom
still useless and probably impossible to do properly
but more possible than the other thing 
floral hawk
I'm all for inventive thinking. That's how this field moves forward. But people who just like, don't really understand the technology trying to overextend is kinda kek. I'm pretty fluid at programming/tech but came completely fresh to AI stuff yk. Gotta stay humble. Someone can't learn what they think they already know.
tight tinsel
clearly this just means we should throw another 2 billion to datacenters
after all more compute mean we will surely get a result
olive sable
hi breadperson
tight tinsel
<a:neuroWaveA:1095012855285293067> hi breadperson
hii
stark needle
hii
jagged turtle
I'm sure the key to unlocking the answers in the world is just throwing more raw compute at the AI companies
tight tinsel
noway its the shadowlilac
floral hawk
heya
stark needle
noway its the shadowlilac
no it's GB10 lover OpenClaw overlord 🦞
Chat propose even more cringe names i can use
floral hawk
Also, can I get some thoughts on something
split steppe
I'm all for inventive thinking. That's how this field moves forward. But people ...
he literally just prompted chatgpt to ask "hey how can i make ai street smart"
the whole plan is entirely generated
jagged turtle
he literally just prompted chatgpt to ask "hey how can i make ai street smart"
actual
rough bloom
I don't get this
basically, if a character is at a certain location, and the character moves to another location, you could make the LLM output some event in JSON for that
then later when that character does something else you can check if they're at the correct location for that or whatever
mostly useless and certainly doesn't prevent hallucinations but is somewhat possible at least
split steppe
afterall
olive sable
discord spellcheck doesnt work 
floral hawk
I'm trying to make a more robust minecraft setup, how would you guys recommend fine tuning an existing vision model? I've never done fine tunes of existing models, just training of some smaller LMs, but even then I've never done so with a vision model.
I'm basically tryna make minecraft eyeballs
woven temple
<:Processing:1464001170673766450> basically, if a character is at a certain loca...
Isn't it way easier and faster to make a simple RAG system for that?
stark needle
I'm trying to make a more robust minecraft setup, how would you guys recommend f...
For what exactly
tight tinsel
Chat propose even more cringe names i can use
"the next sam altman"
floral hawk
For what exactly
Zero's minecraft gameplay
stark needle
Zero's minecraft gameplay
Yes but finetuning for which exact purpose
"playing minecraft" is a very wide scope
split steppe
minecraft is not that hard to automate
floral hawk
It's kindof an enrichment of an already fairly robust bot setup. It already has a state machine, event busses, etc etc
Also, her being able to react to how things look visually.
Content
But realistically, she won't be able to tell to keep away from a creeper for example, without vision. She can, but not nearly as well
split steppe
I'm trying to make a more robust minecraft setup, how would you guys recommend f...
i wouldnt use vision as the primary input, i would only use that as some reinforcement maybe. i would instead hook into the actual data
make a mod
floral hawk
Its not the primary
And yeah I already have a mod
Like I said, its more for enrichment than any baseline of anything
split steppe
then she would know from the data where the creeper and terrain is and keep away from it just fine
jagged turtle
<:Processing:1464001170673766450> basically, if a character is at a certain loca...
that's also what's called, uhm, validation
jovial remnant
I'm basically tryna make minecraft eyeballs
depends heavily on how do you want these "eyes" to be used exactly. Is it identifying what is on the image first and then after that making some actions, or is it doing actions directly based on image tokens, etc.
floral hawk
then she would know from the data where the creeper and terrain is and keep away...
Maybe so
floral hawk
depends heavily on how do you want these "eyes" to be used exactly. Is it identi...
Essentially, yes
jovial remnant
Essentially, yes
yes to which one, cause I listed two different approaches?
rough bloom
that's also what's called, uhm, validation
that's what was described
olive sable
anyone here on nixos, how do yall make the spellcheck work on vesktop?
split steppe
i'd instead use the vision to cover for stuff you can't do from raw data at all, like react to the "quality" of the scene as a whole kind of like neuro commenting on something that looks pretty or scary, which is not really something you can do from the data until it's been rendered out
floral hawk
yes to which one, cause I listed two different approaches?
Identify then react
rough bloom
anyone here on nixos, how do yall make the spellcheck work on vesktop?
should already just work™
olive sable
rough bloom
mabe make sure that it's not off for some reason
floral hawk
i'd instead use the vision to cover for stuff you can't do from raw data at all,...
This is one pathway I could look into
rough bloom
right click message input text field thing
olive sable
i need it to give me the correct word
split steppe
that's not a thing on discord
split steppe
discord overrides context menu tho, i've never seen it work
faint sandal
opaque wharf
Have you installed hunspell?
olive sable
ye
opaque wharf
The language pack?
split steppe
it does show the misspell so surely
olive sable
hunspell hunspellDicts.en_US
split steppe
opening the console?
minor trout
I just couldn't be bothered and told gemini to write it for me so not my fault, and it was just an idea, no need to roast me for that
rare bridge
olive sable
split steppe
I just couldn't be bothered and told gemini to write it for me so not my fault, ...
we wouldn't roast you if you used your own brain to think and respond to people, most of your messages are straight out of that fuckin clanker
jovial remnant
Identify then react
The simplest way is to just gather tons of different images of the gamplay and then for each create the identification you want the model to output. This could technically be made automatic by having minecraft mod that on screenshot (or some other more simple method) gathers what you want to identify from player's view. But at that point no reason to train a vision model since you could have the mod feeding that directly to the model instead, and so training a vision finetune for purely that kind of identification likely is not the best approach. You'd likely want to do vision finetuning for stuff that you can not automate gathering from player's view. That would likely mean manual labeling of images for what you want to identify
minor trout
we wouldn't roast you if you used your own brain to think and respond to people,...
copy and paste, haven't even read what it said
opaque wharf
damn fair point
This is like the moment vedal asks Neuro to swear
split steppe
with great effort
floral hawk
The simplest way is to just gather tons of different images of the gamplay and t...
Heard, appreciate.
olive sable
the same way closed source llms are made
split steppe
usually they're not really "opensource"
minor trout
usually they're not really "opensource"
wdym
split steppe
but open weight, you dont get the training harness nor data, only the resulting weights
rough bloom
i'm a busy man what do you think
so incredibly busy that you're here yapping about symbolic AI
floral hawk
I just couldn't be bothered and told gemini to write it for me so not my fault, ...
You wanted advice but couldnt be bothered to explain your own idea?? Then how are you even qualified to build it
split steppe
You wanted advice but couldnt be bothered to explain your own idea?? Then how ar...
exactly
opaque wharf
but open weight, you dont get the training harness nor data, only the resulting ...
To be fair, even if the dataset is made open, I don't think the result would be similar due to inherent randomess used when learning. Unless if everyone started doing learning with a seeded rng
minor trout
You wanted advice but couldnt be bothered to explain your own idea?? Then how ar...
being bothered is not the same as not knowing what my idea is
split steppe
To be fair, even if the dataset is made open, I don't think the result would be ...
yup, not to mention the hardware requirements to train the whole thing
floral hawk
Insane redirect attempt
split steppe
floral hawk
I have legitimately no words
split steppe
well the problem with ideas is that there's this tiny problem with all of them, you have real world on the other side of them
i have an idea that will turn me into a trillionaire, sadly it requires magic and we dont have that
ideas are fine, but dont get deluded by the llms that you suddenly have the know-how to materialize them
jovial remnant
The simplest way is to just gather tons of different images of the gamplay and t...
Then also depending on how you inference the model, the training data would either be own standalone context completion job, of which results you return to actual running context where the model has the history of actions it has been doing. Or alternatively you could have identification and then action be part of the same completion generation by the model. (tldr you either identify image in fresh standalone context, or together with everything else)
split steppe
but the biggest issue is that of basic human respect, don't give us LLM generated responses, it speaks volumes about you that you can't even be bothered to talk human to human and just dumped what people said into a bot
rough bloom
To be fair, even if the dataset is made open, I don't think the result would be ...
even with the same RNG, some minor kernel optimization can already cause the training runs to diverge
you'd really need to replicate the whole setup, including the hardware, and that's assuming the training setup is even built to be reproducible (so probably nothing asynchronous allowed)
minor trout
but the biggest issue is that of basic human respect, don't give us LLM generate...
yep, i am sorry about that
floral hawk
Then also depending on how you inference the model, the training data would eith...
I genuinely appreciate the outside perspective.
minor trout
will reply with respect human to human now
opaque wharf
even with the same RNG, some minor kernel optimization can already cause the tra...
Wait, I thought the algorithm for training is numerically stable?
floral hawk
I ran into a roadblock with carpetmod fake players not working with baritone, so I'm going to be derailed custom building pathfinding for a few days at least anyway.
minor trout
i have a question to you guys
floral hawk
On the bright side, I get more control out of custom pathfinding. Cons: Gotta learn how to do pathfinding
opaque wharf
Wait, I thought the algorithm for training is numerically stable?
IT IS NOT?!
minor trout
why are you guys in this specific server talking about programming?
split steppe
Wait, I thought the algorithm for training is numerically stable?
the starting point is literally random noise
rigid snow
IT IS NOT?!
as non deterministic as it gets
opaque wharf
floral hawk
Alot of us here were attracted here due to Neuro and a mutual interest in LLMs, it makes sense if you think about it
But that doesnt mean thats all we work on
minor trout
Alot of us here were attracted here due to Neuro and a mutual interest in LLMs, ...
true
minor trout
But that doesnt mean thats all we work on
what else do you work on?
floral hawk
LLMs and programming do go hand in hand though
split steppe
why are you guys in this specific server talking about programming?
vedal is a programmer, so naturally the culture around him will be significantly saturated with other programmers
floral hawk
I build UIs, game mods, and from time to time I automate workflows for remote jobs
opaque wharf
the starting point is literally random noise
Yes but what I meant is that if the rng is seeded then kernel optimization wouldn't make the training diverge too much
floral hawk
To name a few
minor trout
I build UIs, game mods, and from time to time I automate workflows for remote jo...
cool,
hard delta
why are you guys in this specific server talking about programming?
notice the cute AI vtuber twins 
that are all over the place
opaque wharf
If kernel optimization difference would yield a different result then that is not numerically stable is it?
split steppe
Yes but what I meant is that if the rng is seeded then kernel optimization would...
the random noise determines which of the billions of local maximums you land into, and the learned features end up in completely arbitrary locations and arbitrary linkages in the resulting network
minor trout
i realized most of the people i'm talking to right now are the same people i talked to yesterday
minor trout
notice the cute AI vtuber twins <:nwero:1104980808185413652> <:eliv:112259514752...
yes, neuro and evil
olive sable
why are you guys in this specific server talking about programming?
you're in #programming
minor trout
hard delta
that's how most peopel got here, and then "oh hey there is a programming channel"
and then in there peple actually said some interesting stuff here and there
it's the pipeline
floral hawk
Yeap
minor trout
you're in <#1071784467036913664>
yea but i was asking why in a server about two anime bots
split steppe
next one in the pipeline is thighhigh socks and a skirt
jovial remnant
I genuinely appreciate the outside perspective.
No worries, I haven't done any vison fine-tuning runs myself yet (tho I will be doing a ton soon) so my suggestions are likely not the most efficient, but those are the first approaches that come to mind, since finetuning generally follows very similar ideas no matter the modality you want to finetune. The technical details of training itself are usually a separate concern to the gathering of the data itself that you figure out as you go depending on what code you use for training
opaque wharf
you're in <#1071784467036913664>
rigid snow
#baking 
minor trout
you're in <#1071784467036913664>
also i can read
split steppe
yea but i was asking why in a server about two anime bots
vedal talks a lot about the technical side of em and also used to livecode, a lot of programmers watch him and participate
it's technically interesting as well
floral hawk
No worries, I haven't done any vison fine-tuning runs myself yet (tho I will be ...
I did find a good dataset for minecraft vision training, so now I just need to decide how to implement
olive sable
also i can read
you're one of the few, we get a lot of new people in here
rough bloom
vedal talks a lot about the technical side of em and also used to livecode, a lo...
used to 
split steppe
neuro does quite a lot of interesting things, marvelously optimized
rough bloom
used to <:catdespair:1087521982817509426>
bullied too hard
split steppe
minor trout
vedal talks a lot about the technical side of em and also used to livecode, a lo...
didn't know that, but would be interested to join if he does it again
opaque wharf
On the bright side, we got server build stream today
minor trout
bullied too hard
i'm concerned about what you mean by that
split steppe
he probably won't anymore, too much at stake and the audience is too normalized away that most people won't attend
jovial remnant
I did find a good dataset for minecraft vision training, so now I just need to d...
I'd also recommend making sure that data in that dataset fits well with your intended usage, cause it might be the case that's it worthwhile do modify/adjust it to better fit your needs.
rough bloom
i'm concerned about what you mean by that
Vedal code is sometimes
of questionable quality
olive sable
true
rigid snow
vedal talks a lot about the technical side of em and also used to livecode, a lo...
does he talk a lot about the technical side? are we from different universes or something
split steppe
he does have over a million followers now, it's mostly non-programmer crowd
minor trout
kinda sad, would have been helpful to learn from him
split steppe
does he talk a lot about the technical side? are we from different universes or ...
well, not too much of the very internals
minor trout
what does he do now these days?
opaque wharf
Let's bully him even more if he over torqued his threadripper
rigid snow
oh right today is irl stream day
split steppe
what does he do now these days?
just game, collab and twin stream, and all the development is now hidden behind the scenes, even integrations, we never get to see it now
minor trout
What do people think the definition of AGI is because i hear different perspectives on when an ai becomes an AGI
minor trout
just game, collab and twin stream, and all the development is now hidden behind ...
sad
olive sable
AGI is a slop definition without any real substance behind it
rigid snow
qwen 3.5 9b is agi
split steppe
agi is a never ending goalpost of whatever marketing team is in charge now. if you listen to em agi was 10 years ago when gpt was made
hard delta
yea but i was asking why in a server about two anime bots
probably precisely because of that
most purely topic-focused servers are full of idiots trying to look smart + channels are moderated into misery
meanwhile here we have banana dog and fox🐇
minor trout
probably precisely because of that
most purely topic-focused servers are full of...
and bunnies
rough bloom
just game, collab and twin stream, and all the development is now hidden behind ...
we do see game integrations, not the old ones, but a lot of new ones are public 
(because they're either from the community or are commissioned)
split steppe
but originally, agi was an idealistic machine that can do anything a human can, at the same level as a human. just nobody ever agrees which human that is. probably sam
opaque wharf
probably precisely because of that
most purely topic-focused servers are full of...
No more banana dog
opaque wharf
but originally, agi was an idealistic machine that can do anything a human can, ...
To be fair, most human are just average or shit at what they actually do for a living too
minor trout
To be fair, most human are just average or shit at what they actually do for a l...
agreed
split steppe
we do see game integrations, not the old ones, but a lot of new ones are public ...
yea i mean more what vedal develops himself
rigid snow
To be fair, most human are just average or shit at what they actually do for a l...
yes that’s what average means 
split steppe
To be fair, most human are just average or shit at what they actually do for a l...
speak for yourself 
rough bloom
yea i mean more what vedal develops himself
yeah that's all still very private
olive sable
im bringing the average down, you should be thankful
opaque wharf
yes that’s what average means <:mlntcandy:1455273822625333375>
Then by that definition, surely current LLM has reached most of the field average no?
rigid snow
yeah? kinda
minor trout
out of billions of people, how many do you think actually has some useful skills? like less than half the population
opaque wharf
im bringing the average down, you should be thankful
split steppe
Then by that definition, surely current LLM has reached most of the field averag...
can't pour a coffee tho
opaque wharf
yeah? kinda
AGI achieved
split steppe
llms are only a tiny fraction of human capability
agi is general capability across the board
rigid snow
yeah? kinda
i mean at coding specifically ye
opaque wharf
can't pour a coffee tho
We do current started to explore more machine learning stuff for real time robotic
minor trout
i mean if ai has to do what everything humans can do, they're gonna need a body
olive sable
i think the time has come for nix flake update
opaque wharf
Which is what I'm actually interested in than LLM
split steppe
LLM is specialized intelligence
opaque wharf
i mean at coding specifically ye
They are pretty average at math too tho
split steppe
they need embodiment as a bare minimum
rigid snow
They are pretty average at math too tho
i don’t math i wouldn’t know
minor trout
and that's gonna be hard because engineers can't get the energy efficiency and replicate every muscles of the human body
opaque wharf
i don’t math i wouldn’t know
I do, just yesterday I need to do a ton of it
split steppe
you don't need it to perfectly match a human body necessarily, it just needs to be able to perform like a human
look at boston dynamics
those are fairly good at mobility
opaque wharf
They are now also quite alright for personal translation
minor trout
so they don't necessarily need to look like a human
split steppe
robots struggle with fidelity still, like fine control of pouring or not cracking eggs n shit
opaque wharf
look at boston dynamics
I prefer bosstown dynamic more
opaque wharf
robots struggle with fidelity still, like fine control of pouring or not crackin...
Yeah, that's more to do with the physical layer stuff tho
rough bloom
i think the time has come for `nix flake update` <:catdespair:108752198281750942...
NixOS 26.05 soon
olive sable
34GB??????????? 
split steppe
so they don't necessarily need to look like a human
yea not at all, it could be an octopus for all its worth, just needs to be able to reasonably do most things an average human can
minor trout
robots struggle with fidelity still, like fine control of pouring or not crackin...
i mean if a robot doesn't look like human, can they put up the middle finger?
opaque wharf
Which is why the hand sensor thingy that someone unveil st CES is actually amazing
split steppe
i mean if a robot doesn't look like human, can they put up the middle finger?
true that, gotta give it one human hand :V
rough bloom
34GB??????????? <:welpsagiri:1303972670127214662>
ye mass rebuild probably
split steppe
34GB??????????? <:welpsagiri:1303972670127214662>
how are you enjoying nixos
olive sable
very well timed screenshot i guess. pyopengl, havent heard that name in ages
minor trout
34GB??????????? <:welpsagiri:1303972670127214662>
huge
minor trout
sudden question, where are you guys from?
split steppe
a good chunk from europe or adjacent
rough bloom
neurope
split steppe
vedal is br*tish afterall
split steppe
throw a dart in the subcontinent the size of 1/3 of the world 
rigid snow
i tried 9b models that are tuned for openclaw like agents and they’re aight for shuffling docs around and doing generic stuff. i wanted to try and merge an rp tune with that and look at what monstrosity comes out. maybe an actually fun personal assistant, because like, i like how gpt 5.5 performs but its sooo boring in how it plays its own soul.md out
olive sable
asia
you said you have the hardware needed to run your ai idea. but what hardware exactly would that be?
do you have 2 4090s or something?
split steppe
hes from china or smth, got an entire lab prolly
bitter phoenix
i tried 9b models that are tuned for openclaw like agents and they’re aight for ...
gpt 5.5 gives much less effort to going beyond the exact task than 5.4
minor trout
you said you have the hardware needed to run your ai idea. but what hardware exa...
no but enough and optimized
olive sable
hes from china or smth, got an entire lab prolly
its john blackwell 
minor trout
like i'm not gonna make it gpu heavy if i want it to be quick and a daemon
rigid snow
gpt 5.5 gives much less effort to going beyond the exact task than 5.4
i still like 5.5 more
olive sable
no but enough and optimized
"optimized" doesn't work well once you need 16GB for your context window
minor trout
hes from china or smth, got an entire lab prolly
i hope bro, i hope but i'm not old enough for that
olive sable
the deamon is running
minor trout
OHHHH
split steppe
not old enough
mom, can i get a lot of 8 B200, and AMD Epyc, 200x 32TB SSDs and google fiber, i need it for school
minor trout
i misread that as a demon not a daemon
minor trout
> not old enough
mom, can i get a lot of 8 B200, and AMD Epyc, 200x 32TB SSDs an...
like not old enough as in i'm not over 30 years old but out of school
rigid snow
not over 30 seconds
split steppe
hes born not even yesterday :V
split steppe
minor trout
they grow up so fast
fr
olive sable
split steppe
nah you just need to be some rich guys child and nepo your way into owning a datacenter
baby musk
that sounds wrong
minor trout
I'm not experienced as you people and like an amateur so i probably won't care if you see me as pathetic
split steppe
it's not pathetic to learn, it is however pretty pathetic to outsource your smarts to an llm, gotta remember that
opaque wharf
Oh, konii is here
split steppe
llms rarely teach
minor trout
who's konii
minor trout
right now i only know how to do python and a little bit of javascript, css, and html
tender river
who's konii
rigid snow
you
split steppe
i konii remember
split steppe
right now i only know how to do python and a little bit of javascript, css, and ...
try some C for a change of pace
minor trout
try some C for a change of pace
will try
minor trout
will watch thank you
opaque wharf
https://www.youtube.com/@Tsoding
minor trout
i love silence the most
minor trout
the hell
split steppe
umbral wigeon
Claude, rewrite this video in rust
minor trout
not gonna lie he's acting a bit autistic
split steppe
aint we all a bit autistic
minor trout
aint we all a bit autistic
yes
split steppe
he's just european tho
minor trout
we all are
minor trout
he's just european tho
i thought you said you're european too?
split steppe
ya we're all like that here
umbral wigeon
Claude, Summary chat above, edit: No mistakes
rigid snow
who’s european 
minor trout
who’s european <:glueless:1282337396230328425>
you
minor trout
Claude, Summary chat above, edit: No mistakes
never tried claude yet
split steppe
ur a peein, am a peein
umbral wigeon
never tried claude yet
I like opencode, because I'm homeless open source maintainer
minor trout
how did you know
profile looks like typical european
rigid snow
:ohidontsee
umbral wigeon
wishing luck
I kinda like chinese AI
split steppe
typical european profile, saw it from a mile away
minor trout
I kinda like chinese AI
and you're from thailand
is there a reason why you like chinese ai?
umbral wigeon
Yes, but I like to glaze china
minor trout
Yes, but I like to glaze china
aura
umbral wigeon
China best technology
minor trout
China best technology
partly yes
split steppe
chinese ai is free and easy to run :V
minor trout
chinese ai is free and easy to run :V
i wouldn't trust it if i were you
umbral wigeon
minor trout
and i recently realized the qwen series were made by china and trained on claude data
minor trout
https://tenor.com/view/social-credit-credit-social-%D1%83%D0%B2%D0%B0%D0%B6%D0%B...
fr
split steppe
i dont trust ai period, still, it's pretty nice to be able to run one for random vague questions i sometimes have
minor trout
i dont trust ai period, still, it's pretty nice to be able to run one for random...
like what?
rigid snow
i dont trust ai period, still, it's pretty nice to be able to run one for random...
i trust ai to make my codebase into an incomprehensible mess
split steppe
like what is this one weird movie i only remember a single frame of that i also remember incorrectly
i dont use ai for any work, it's brainrotting if used that way
opaque wharf
i dont trust ai period, still, it's pretty nice to be able to run one for random...
I trust them but still know if their answer is bullshitting
split steppe
never gonna pay any of them either
umbral wigeon
i trust ai to make my codebase into an incomprehensible mess
I trust ai to write boilerplate for me cus my fingers hurt, so I'll use my eyes and brain instead
opaque wharf
never gonna pay any of them either
Same, this is the only hill I'd die on
split steppe
I trust them but still know if their answer is bullshitting <:enub:1163912230585...
until your skills degrade to the point you dont anymore, because you've stopped refining your own
minor trout
one of my friends say clankers should all be exterminated
split steppe
based, i like your friend
minor trout
because they're stupid and slop makers
jagged turtle
yea but i was asking why in a server about two anime bots
oh gee golly gosh I sure wonder why people in a server about AI anime girls are talking about programming in the programming channel
opaque wharf
until your skills degrade to the point you dont anymore, because you've stopped ...
Naah, I don't do that
rigid snow
oh also on vibecoding, the funniest thing happened recently. i was vibecoding a neuro clone because i wanted to try the stack out and i asked for lipsync. i fucking died when i started talking to it for the first time and it started to lipsync ME instead of the tts
minor trout
oh gee golly gosh I sure wonder why people in a server about AI anime girls are ...
didn't know much about neuro evil and vedal until 3 days ago
minor trout
oh also on vibecoding, the funniest thing happened recently. i was vibecoding a ...
i would be interested to see how it ends up
rigid snow
oh also on vibecoding, the funniest thing happened recently. i was vibecoding a ...
gpt 5.5 btw, supposedly frontier
olive sable
PBO is set to 200W. but it still doesnt want to go higher than 110W 
opaque wharf
Naah, I don't do that <:neurOMEGALUL:1097297318119743638>
Can't degrade my chinese skill when I don't have any 
split steppe
PBO is set to 200W. but it still doesnt want to go higher than 110W <:bwaadow:13...
brother your cpu is 90C
opaque wharf
But I know if their translation are BS
olive sable
its fine
opaque sigil
compiling isn't that heavy in the first place
olive sable
the fans arent ramped up fully yet
jagged turtle
yea i mean more what vedal develops himself
you mean copilot
olive sable
i need to change the fan curve
split steppe
you can cook eggs on that, fix your fans lol
minor trout
Can't degrade my chinese skill when I don't have any <:evilWheeze:11313162603538...
i know some basics
大王
jagged turtle
we do see game integrations, not the old ones, but a lot of new ones are public ...
sts2 is comissioned but it hasn't been oss'd yet (soon™ )
minor trout
who's the 大王 here, it's vedal!
opaque wharf
i know some basics
I don't even know a single character
opaque wharf
That's why I said AI won't degrade my skill
jagged turtle
speaking of game integrations the video I wrote about neuro api was released today and never have I felt more stressed in my life when I woke up
minor trout
speaking of game integrations the video I wrote about neuro api was released tod...
why
can i see the video?
split steppe
That's why I said AI won't degrade my skill <:evilWheeze:1131316260353867887>
ai can't make me stupid if i'm already stupid!
every AI bro ever
olive sable
every AI bro ever
me running 4B models on a 4090
nocturne olive
PBO is set to 200W. but it still doesnt want to go higher than 110W <:bwaadow:13...
Thermal throttling
Mabe
90C is sus
olive sable
i should try setting a more aggresive fan curve
split steppe
100% thermal throttling, he's cooking his cpu and complains it doesn't pull more
minor trout
what's the best image generator the world has right now
split steppe
throttling already begins at 80C usually
opaque wharf
ai can't make me stupid if i'm already stupid!
Correct. I am thankful for their many pages of translated Chinese technical document
rough bloom
<a:YES:1162976978278809641>
90°C is tjmax for that generation
split steppe
thats the first thing i change on a new rig lol
minor trout
why does mac air don't have cooling fans
split steppe
i hope your fans are plugged into the right spot too
olive sable
lemme wait until it finishes compiling this shit before i change the fan curves
split steppe
why does mac air don't have cooling fans
super efficient low power chip and the whole metal body used as a heatsink
olive sable
its compilign blender rn
jagged turtle
can i see the video?
find it yourself
olive sable
i hope your fans are plugged into the right spot too
they are ye
split steppe
i should get out of bed and continue my engine development
opaque sigil
why are you compiling blender 
olive sable
hip
minor trout
find it yourself
not a video but i'm guessing it's https://github.com/VedalAI/neuro-sdk
jagged turtle
not a video but i'm guessing it's https://github.com/VedalAI/neuro-sdk
what
olive sable
why are you compiling blender <:neuroCry:1110064189839523881>
(blender.override { hipSupport = true; })
jagged turtle
why are you compiling blender <:neuroCry:1110064189839523881>
are you not allowed to compile blender
olive sable
split steppe
`(blender.override { hipSupport = true; })`
lol i thought u meant its hip to compile blender nowadays :V
proven merlin
Id probably undervolt (pbo curve optimize) the cpu, youre more likely to hit better clocks that way.. (if youre tdp or temp Limited)
olive sable
[2/566/587 built, 1987 copied (34098.5/34106.3 MiB), 7352.7 MiB DL] building blender-5.0.1 (buildPhase): [ 45%] Building CXX object source/blender/blenlib/CMakeFiles/bf_blenlib.dir/intern/array_store.cc.o
olive sable
github
split steppe
the repo he sent is the api/sdk
olive sable
opaque sigil
i think you want pkgsRocm.blender , supposedly cached from 2 days ago
olive sable
https://github.com/VedalAI/neuro-sdk
how is this not pinned?
opaque wharf
are you not allowed to compile blender
From nixOS it is usually a sign of jank
And you know how Sam is 
olive sable
https://github.com/VedalAI/neuro-sdk
@real sierra maybe pin the neuro sdk
jagged turtle
how is this not pinned?
because there's a dedicated #1336777692070023271 thread for it which is supposed to be the place for its discussions: #1336984506489376779
olive sable
From nixOS it is usually a sign of jank <:neurOMEGALUL:1097297318119743638>
im litteraly jsut enabeling HIP support. for cuda you would have to do the same
opaque sigil
someone needs to sit this dude down and explain to him the concept of a binary cache
olive sable
because there's a dedicated <#1336777692070023271> thread for it which is suppos...
i feel like people come to here to talk about the sdk a lot still
projects forum is a bit hidden away
minor trout
https://github.com/VedalAI/neuro-sdk
so this thing allows ai to play games
split steppe
this channel should prolly also have a pin for it i feel like
jagged turtle
i feel like people come to here to talk about the sdk a lot still
I think recommended action should be to redirect to the discussion thread
if it's like, questions specifically
or leave it here idrc and then whoever can answer could answer
olive sable
then pin a shortcut to the channel
jagged turtle
this channel should prolly also have a pin for it i feel like
yeah that also works
olive sable
pin a link to it
for neuro-sdk game integration see https://discord.com/channels/574720535888396288/1336984506489376779
olive sable
for neuro-sdk game integration see https://discord.com/channels/5747205358883962...
pin this
split steppe
pin sam to the wall
olive sable
stark needle
shiro
olive sable
shadowo
stark needle
⚪
split steppe
dont mods have pin perms, clouds on
stark needle
shiro
nocturne olive
90°C is tjmax for that generation
Makes perfect sense
olive sable
rare bridge
sure
split steppe
olive sable
thanks cloud 
opaque sigil
blender from 25.11 relies on hipSupport while 26.05 blender relies on rocmSupport and the former is not cached
sage crag
nocturne olive
New legendary #programming pin
opaque wharf
<:NeurOhISee:1136209034966483084> blender from 25.11 relies on `hipSupport` whil...
So, sam is downloading older version?
opaque sigil
unstable is at 5.1.1
olive sable
but my nixos is old
olive sable
oh ye im on stable
opaque sigil
next week i guess
rough bloom
do we know the codename for 26.11 yet <:FOCUS:1168267148523737239>
don't think so
should be decided some time after 26.05 release I think?
sage crag
processing
sage crag
vetoed
olive sable
opaque sigil
sage crag
wow that logo is awful
olive sable
sage crag
it looks like it fell apart
opaque sigil
wtf is that 
sage crag
earthquake(P)
olive sable
disgusting
rough bloom
very unique
sage crag
look at the 4 logos at the top of the page
split steppe
why do people insist on making their shit generic and unrecognisable these days, i'll never get it
minor trout
i feel like this channel only has the same people talking over and over again
opaque sigil
did you see the disco icon theme they added to pixel phone
opaque wharf
i feel like this channel only has the same people talking over and over again
A lot more are MIA
@desert plaza the human compiler please come back
stark needle
i feel like this channel only has the same people talking over and over again
Very likely
opaque wharf
did you see the disco icon theme they added to pixel phone <:FOCUS:1168267148523...
What disco icon?
opaque sigil
minor trout
the first pinned message was you saying smth on 2024 november
stark needle
<:enub:1163912230585237655>
Wtf what
opaque wharf
<:enub:1163912230585237655>
WHAT THE ACTUAL F*** IS THAT
opaque sigil
a joke they for some reason turned into an actual icon pack after spotify changed theirs for a bit
split steppe
WHAT THE ACTUAL F*** IS THAT
it's called fashion, baby, look it up
stark needle
a joke they for some reason turned into an actual icon pack after spotify change...
Where are icon packs even available
Bruh
split steppe
spotify's new logo is definitely awful though
opaque wharf
split steppe
low contrast, low recognizability
opaque sigil
Where are icon packs even available
idk i don't have a pixel go find it
tender river
p5.js is cool
split steppe
Where are icon packs even available
on android its just apps, but you need a launcher that supports em
opaque wharf
p5.js is cool
It is
I am not talking about the software itself
Like, processing is goated for what they are
minor trout
<:enub:1163912230585237655>
ㅣㅐㅐㅏㄴ ㅕ히ㅛ
opaque sigil
we need more whimsy in the world
minor trout
op, no i meant looks ugly
opaque wharf
we need more whimsy in the world
I feel like there are some line in the middle
Like scratch logo
tender river
it looks like it fell apart
it really is all over the place 
like at least make it 45deg
opaque sigil
I feel like there are some line in the middle <:enub:1163912230585237655>
no line 
opaque wharf
it really is all over the place <:vedalCry:1234565975492264018> like at least ma...
I think it would look even more hideous on 45 degree
tender river
i think it could be better
also the other logos share nothing in common
other than
minor trout
are you trying to type korean using individual consonants/vowels :V
my keyboard was on korean and i tried typing an english sentence
minor trout
i can speak in korean but unless there is someone korean, no one would understand
split steppe
local koreans raise your hand
olive sable
Local to who?
kind fable
olive sable
sam really?
I didnt raise my hand
leaden crest
olive sable
Raising hand looks like this "🙋♂️"
leaden crest
I raise both hands
leaden crest
i can speak in korean but unless there is someone korean, no one would understan...
correction you can speak in korean but shiro would descend and smite you
split steppe
korean spotted 👆
leaden crest
uhh who else is korean in here holdup
olive sable
Rule 10 ye
minor trout
한국어 할 수 있는 사람 손
olive sable
minor trout
샘드위치
split steppe
minor trout
ㅋㅋㅋ
opaque wharf
Local to who?
"What do you mean the LLM is local? That's Claude!"
"Local TO ME!"
split steppe
supikip has jp tag, are they korean 🤔
leaden crest
joayo joayo hobagi joayo
opaque wharf
🤣
https://fxtwitter.com/hd_nvim/status/2058177537679782139
It's satire right?
kind fable
It's satire right? <:catdespair:1087521982817509426>
Yes altho based on real facts
split steppe
olive sable
Idk
opaque wharf
Yes altho based on real facts
It's pretty hard to spot satire vs reality for a project that does rewrite experiment without finishing the main task first
minor trout
what about sandwiches
i said "sam"dwiches
leaden crest
i said "sam"dwiches
i assumed so but idk what a samdwich is
olive sable
This is the first time i tried to search something and it thought it was an ip. Normally it does the opposite where it puts an ip into normal google
minor trout
well it's @olive sable stuck between two piece of bread
leaden crest
and “m” and “n” are pretty close apart on korean keyboard
olive sable
leaden crest
@olive sable
minor trout
and “m” and “n” are pretty close apart on korean keyboard
true
minor trout
<:neuroBread:1245345798296703096> <@635476518692454420> <:neuroBread:12453457982...
yum
opaque wharf
and “m” and “n” are pretty close apart on korean keyboard
So does in qwerty keyboard
minor trout
So does in qwerty keyboard
funny coincidence
leaden crest
So does in qwerty keyboard
true!
minor trout
true! <:neuroHypers:1135051812722982922>
how fluent are you in english
leaden crest
im native english
I cant speak korean well
or well
I speak it fluently but i cant spell for shit
opaque wharf
This is the first time i tried to search something and it thought it was an ip. ...
What browser is that anyway?
olive sable
Firefox
minor trout
so like i looked through the list of people and so far i've found one people with a korean username
2 now
and every other korean people are at the bottom
olive sable
opaque wharf
and every other korean people are at the bottom
UTF-8 Sorting at its finest
olive sable
I guess il send this here too
Ac pump repaired
Condensation pumps seem to be the most fragile parts in an ac system
This is the 2nd one we had to replace
opaque wharf
Condensation pumps seem to be the most fragile parts in an ac system
It's the fan usually