#programming
1 messages · Page 535 of 1
i skip the ai stuff cuz gamedev. for example this one
https://www.youtube.com/watch?v=Wibbnn3hV4U
❤️ Check out Lambda here and sign up for their GPU Cloud: https://lambda.ai/papers
📝 The paper is available here:
https://wanghmin.github.io/publication/zhang-2025-pie/
Our Patreon if you wish to support us: https://www.patreon.com/TwoMinutePapers
🙏 We would like to thank our generous Patreon supporters who make Two Minute Papers po...
you think "ooh, lemme see some code"
and then
where is the code?
you cant force these things

Huamin Wang
Chief Scientist
only repo is his first compute shader he made 3 years ago
Roblox is down bad right now lmao
So I'm gonna switch to open source maintainer
Or engineer
: P
yeah ngl
cus I'm open source maintainer rn
because solo game dev is insanely hard
anyone tryna do solo game dev, you gonna be some tough dev
game dev isnt hard, its a lot of effort
Yeah
@olive sable send this to 10 of your friends dm to avoid bad luck 
nah
i embrace the bad luck
LMAO a 2 minute paper video about hair sim framed it as some new and cool technique, and then i dig a bit deeper and get to a 16 year old youtube video

LGTM!
Impossible to avoid says the ecosystems with package managers where it happen
https://fxtwitter.com/vxunderground/status/2057982277485822236
Quoting Aikido Security (@AikidoSecurity)
︀
🚨 Ongoing supply chain attack on Composer packages! We just found multiple laravel-lang/* packages compromised on Packagist (lang, http-statuses, attributes). Payload runs at autoload time. At least 50 package versions were compromised.
︀︀
︀︀If you installed a compromised version, the malware already executed. Pin to a clean COMMIT (not version) and rotate secrets immediately.
︀︀
︀︀If your lockfile already had an older commit from before today, you are safe. But you should not update at the moment.
I'm just here vibing with my obscure package that is not high profile enough to get targeted 
why can't they just shut it down? are they smart?
Shut it down? How?
has anyone ever tried making a digital entity

please describe what a digital entity is for the students in the back of the class
Unless the meaning in this context is very different than how it would normally be looked at then it is an issue of Youn can stop this but then another will have taken it's place already as on a multitude of things to take care of appear while you stop a multitude of things that already have arrived
basically an ai that can do everything a human can do on a computer
a ghost
in my computer
i may need to think about what humans can actually do on a computer
whats their name
clawdbot, send this user to Brazil
luigi
very insightful
truly profound
Wouldn't that be made impossible by the fact that the AI is in a computer and a human outside
human can punch the computer screen
just giving them access to keyboard and mouse input to do whatever?
they gotta think and have intent
thats not digital entity enough


give them a physical rat
why not?
An AI controlling a humanoid robot using the computer
i was thinking more of a daemon living on my computer
daemon is exisitng concept
see wiki page for daemon
have you tried finding a digital exorcist first
p
don't know what an exorcist is
OYYYYYYYY
see wiki page for demon
see wiki page for
demon
see wiki page for
bnuy
what would a multiple llm (multi model)+vlm+world model+symbolic ai+agent(openclaw) be?
Is this a good discription
leaf flexible 
i know multimodal, i know vllm, but what is "world model" or "symbolic ai" supposed to mean?
Kinda late to the party on this one...
Not sure if it’s ‘cause the song is still pretty new, or I’m just dumb lol
but I seriously can’t find an English translation anywhere.
So yeah, I ended up asking ChatGPT to help me translate it line by line.
Expect a few weird lines here and there
(it claims like 87% accuracy or something—guess...
its sentienceslop
symbolic ai runs only on logics

world model you can go search it up yourself

Humans too. Emotion is just very complex logic

yes
kind of because llm predict
youre also unqualified
that's how the tech works ye. how are you planning to not do that?
Symbolic AI (also known as "classical AI" or GOFAI—Good Old-Fashioned AI) is an approach that builds intelligent systems by manipulating human-readable symbols and logical rules, rather than relying on statistical patterns. that's what gemini says
Gemini also says stuff that I cannot send here 

Strengths: Highly transparent, explainable, and reliable for strict logic-based tasks (like math or medical diagnoses) because you can always trace exactly why the AI made a decision.
Limitations: Poor at handling ambiguity, noise, or unstructured data (like recognizing a blurry photo of a cat), and the rules cannot easily adapt without human intervention.
so it's not like the best ai but there's something called a neuro-symbolic ai
That makes no sense
so symbolic ai is top-down while chatgpt is bottom-up
Can you verify that with another AI saying exactly the same thing
Symbolic AI was the dominant paradigm of AI research from the mid-1950s until the mid-1990s.[4] Researchers in the 1960s and the 1970s were convinced that symbolic approaches would eventually succeed in creating a machine with artificial general intelligence and considered this the ultimate goal of their field.
symbolic ai doesnt exist then, its just standard code
i could call the skyrim npcs symbolic ai
So symbolic AI == big pile of if-else statements
can we kill this sprout nowwwwwww
yes

"symbolic ai", even more so than normal llms. sentience = True 
Imagine if we had a "symbolic AI" bubble
so ye its a bunch of baloney. symbolic ai is trash
you can't truly have everything from if else statements because there would be too many possibilities which is why it's an abandoned version of ai
In artificial intelligence, symbolic artificial intelligence (also known as classical artificial intelligence or logic-based artificial intelligence)[1][2] is the term for the collection of all methods in artificial intelligence research that are based on high-level symbolic (human-readable) representations of problems, logic, and search.[3] Symbolic AI used tools such as logic programming, production rules, semantic nets and frames, and it developed applications such as knowledge-based systems (in particular, expert systems), symbolic mathematics, automated theorem provers, ontologies, the semantic web, and automated planning and scheduling systems. The Symbolic AI paradigm led to important ideas in search, symbolic programming languages, agents, multi-agent systems, the semantic web, and the strengths and limitations of formal knowledge and reasoning systems.
you're reinventing siri form 2010
Mb about the size it seemed smaller
symbolic ai is one of the most efficient ways to prevent hallucinations in llm which is why both often go together
youtube
potato
tv
to prevent hallucinations in llm
its not an llm so its not preventing any hallucinations in llms
its not even doign anything that could be seen as thinking
its switch cases
youre arguing with a kid

verifier
Hallucinations are caused by a small amount of nodes that under certain circumstances will not do that they would otherwise do
when did i say symbolic ai are the best form of ai
So anything without nodes is effective for no hallucinations
No hallucinations are caused by the fact LLMs are probabilistic models that model the text from their datasets, rather than modeling actual intelligence
yea i agree
they are a part of what might become intelligence but not intelligence as a whole
What the heck even is a "node"
Yes you're absolutely right ! 😄
LLM in chat
That process is what causes those problematic nodes
This has been proven they can detect them
im relating it back to the original question.
i dont see how you're planning on doing this with switch cases
vllm doesnt apply to this, you can make it mutlimodel sure but then you'd just have 2 scripts doing if-else statements so its pointless.
Supressing only those stopped hallucinations from generating
What the heck is a "node" in an LLM
An LLM is a series of matrices being multiplied on the tokens passing through
Neural network consists of nodes
No it consists of layers
the neurons i assume
ANN anyway
We convert shit into numbers then make it automatically do math and hope the result is good
It's how input is turned to output
That's the layers' job, they're big matrices
AI are the 2025 casinos
youre arguing with 
You don't really think of individual artificial neurons on an ANN network they're not really neurons
Yes the layers are made up of rows of nodes

Linked nodes

vlm (visual language model). and i don't understand what you're trying to say
They're numbers being multiplied together with the input
It's matrix multiplication

You don't actually compute each neuron individually, you multiply the input matrix of a layer with that layer's matrix
how old do you think i am
not old enough✅
The nodes are made of a type of material that has assymetrical magnetic fields because they don't orientate themselves in the same direction and the orientation can be altered by electrical pulses

Also mathematically this happens to be equivalent to wx + b a whole bunch of times
troll or mental patient
It's what makes ai trainable
what is the cause of qualia?

It's a linear equation
and how do we have that?
I recommend watching this for people who want to understand LLM and stuff
https://youtube.com/@welchlabs
YouTube
New Book! The Welch Labs Illustrated Guide to AI is now available for pre-order: https://www.welchlabs.com/resources/ai-book
so much pseudo profundity
Whoever this guy is they don't know aanything about ANN
It has actually good videos
like i mean how do we feel the connotations of red?

Red doesn't exist

No wonder LLMs are so stupid we have too much natural stupidity
true if we think it in the sense that all animals see differently
I assume you mean light at 620 to 750nm
prob because of a combination of fragile context windows and a lack of common sense and real world knowledge
It's badly hidden i can see them
This is a neural network and the circles are nodes
its because people keep
ing all over the training data
And the arrows are edges
It's the misleading representation
This is just what people use to explain this to non-math people, really this is represented by matrices being multiplied together, computing every single "neuron" individually would be too inefficient
Replace the different sets of colored circles each with one big blob of numbers
not really misleading
But in this case it is clearly


It can still be used to give a clear indication about what you are referring to
Well either way this guy is very misinformed on how ANN works
Pyon
i want to explain things but it would be lost on this channel right now

aaaa
pyon
If emotions are shortcuts to prevent input overload and an abbreviation of the chain of thoughts, how can we get ai to have emotions?
to put it frankly. what you're trying to do isn't possible.
you cant make a bunch of if statements do full antigenic behavior without having to do billions of switch cases.
you'd have to program its behavior for every single thing it does, there is no way to do it properly
if (at_kitchen) { find_coffee() }
What if the door is closed?
What if the light is off?
What if the coffee is in a red mug vs. a blue mug?
What if there is a cat in the way?
you'd need a trillion lines of code to do one simple task effectively, there is no autonomous reasoning in a switch case. its not AI, its a flowchart
VLM outputs a distribution of pixels, to get that to even mix with a symbolic AI is pretty much not feasable.
it is a bad image that explains pretty much nothing
but not really inaccurate, just bad 
but if they don't have qualia, will it ever be possible?
By making something that isn't an LLM
emotions are input overload


at least i know ADHDtism people will agree
Never mind figure it out yourself. I only mentioned it to indicate that no nodes means no ai hallucinations

So this guy believed it too far and now they don't know how it really works
be quiet is a fan brand
istg i just said vlm because you said vlm with a double l
I don't think that's how it works, no "nodes" would mean no ANN model
why do you want me to be a fan ?
i give up
never gonna give you
someone get the gun pls

up, never gonna let you down
your collective brain damage is impressive, please dont show me again
Yes it's from a Wikipage i only needed a simple example to show what i was referring to and since you understand it that is what it did.
btw you got rickrolled and lost the game the same day
can you get the gun pls
yea
"working on my bachelor thesis"
anyway i go back to my computer science bachelor thesis in snake
no clue what you're referring to
truly unfortunate for you to have come this far and to be so undeveloped

undeveloped in what ?
What i mean when i say nodes
konii is brutal, damn
ah yes, i'll know when i'll be older
i don't wanna know
if it's surpises like taxes and having to remember shit i have enough

i don't wanna figure it out
foetal alcohol syndrome is no joke
i'd rather stay where i am and code
if you see a single spider and it scares you, a single input making you feel scared is definitely not input overload
holy fuck, you're murdering him
i am actually arachnophobia and i picture things too well, not thanks you
spider
big furry tarantulas
fun fact my doctor officially called me autistic life sucks

good good
thanks for confirming jackass
welcome to the club

what is the average age of people in this chat
also for some reason she had lobotomy posters on the wall
i'm 24
67
istg 67 in the big 26 is not good
67
where's the neuro tism gif again
67
sorry the evil
my friends were arguing over whether thinkpads are better than mac
i forgor
i feel like we should nuke this channel and start over as #baking
this has gone far enough
yes
nuke the server and become hermits
yes let me dump code here
no
yes
reject humanity return to monke
yesn't
bets on children or brain damage
does vedal look at these chats
I finally got dinner at 8pm
no, i quite like having hygiene
haven't had anything today
brain damage
foodpost?
live next to a river
why
both
just curious
imma dip
foodpost
this is enough
who's the youngest here
Look tasty
vedal
TMI

no. we want this here
this is #baking
you woudlnt understand
this is the norm here
wrr
ye i probably don't know because i joined yesterday
ok next time if I bake something I will post it here
im leaving
I joined because i was curious how vedal made neuro
bye konii
do I make weed brownies or crack filled donuts next time
bnuy
bnuy
please rate my code which i'm gonna PR as a patch to this
https://github.com/yrutschle/sslh/blob/master/probe.c
additional documentation :
https://wikipedia.org/wiki/LEB128
https://minecraft.wiki/w/Java_Edition_protocol/Server_List_Ping#1.6
https://minecraft.wiki/w/Java_Edition_protocol/Packets#Handshake
i genuinely need people to review this thanks
what's it for
LGTM!
Looks code enuff
it's a minecraft server probe for sslh which is a protocol demultiplexer
yeah i saw that already
but specifically i wanted to look into the pandora's box, the private parts vedal didn't reveal
like how he makes the llm consistent with neuro's identity
because simple text prompts like "you are neuro" won't work
There are much better options than Whisper for asr these days
definitely
fine-tuning 
teacher i'm done
Parakeet 0.6B can run at up to 30x realtime on CPU with slightly superior accuracy and near zero hallucination rates for a start
what do i get
question can you fine tune a closed source ai like gpt
chart hasnt been updated in some months, idk
We need to put DiffSinger on this thingy
it's been like a year since the chart is updated
depends if you work at open AI
you get a fucking slap for not using wolfram
Parakeet is trash in practice
it's quite simple™
imagine if vedal was secretely an openai member
It's just better on benchmarks
otherwise you get a jar of cookies as well
Vedal is neuro sama member
depends on provider, sometimes they allow it, but expensive and not worth it
Yes via the api
But not all models
Does vedal likes yandere dev
is it fine-tuning truly ?
i was thinking of making the voice of my ai using vocaltractlab and running it through applio
not publicly. i doubt many people like the guy tho
interesting, so it's not just system prompt
not done
looks like only gpt4.1 or o4 mini can be finetuned
bad camera cant read
Unless the docs are outdated
I don't think they are
OpenAI is winding down the fine-tuning platform.


Bruh rip
Well if u need hosted finetune lora
U can use together.ai or azure
never heard of it
Azure = microsoft azure cloud
wouldn't it be better to have multiple static llm and only call the one that can solve a problem or task best?
Together.ai = same foundation that did sglang iirc
Needs too much vram
Have 1 llm and swap adapters

how good are M5 btw?
bro reinvented MoE
so that's the name of this thing
google AI MoE
didn't know that
Mixture of Experts
I will find out once CHXI benchmarks NeuroSynth on one
would you consider this chat a mixture of experts or just normal people
Funny story
They are “winding down” their finetuning api
Aka
What I do know right now is
- M1 absolutely just sucks (30s per render of tiny change)
- M3 pro is trash (10-15s per render of tiny change)
They want to kill it
a cult
fr
MoI; mixture of intelligence 
Conclusion: Apple silicon is messed up and stupid
depending on your usecase*
Buy a 3090 instead
Uh
Idk about that
depends on how you would use it
3090 ~1s to render tiny change
Hating on reality is crazy
if using for NeuroSynth (almost only you)
Apple silicon is their best move since the iPhone
get more money, get better chips
There's some people using NeuroSynth other than me now that it's out
I don’t even know how to use it bruh
That's how I got the Apple silicon data too
Its much more I than the ai I’m used to
isn't neurosynth inherently just better on dGPU?
hence apple silicon as an apu doesnt work too well?
Well NeuroSynth is an inference load
Shhh no logic allowed
I literally provided instructions on the website
box is biggest (by default) neurosynth defender
pretty much
the main reason people use Apple Silicon for ML stuff is because tons of VRAM, even if the compute is kinda weak
put neurosynth on samsung smart fridge
flipper zero
NeuroSynth AFAIK doesn't need tons of VRAM for inference, so not benefit at all from that 
That's gonna need NeuroSynth-nano (tm)
LGTM!
oh god
Nuh uh
next we're gonna get neurosynth pro

and neurosynth flash
Sorry for analog horror sounds
Na NeuroSynth-nano will just be a smaller variant as I'm gonna build a bunch of configurations for size into my DiffTrainer clone and then try to finetune NeuroSynth for size

make a neurosynth i can run in a game engine
Where did all the text go
then we can have dynamic pitch control by the character doing certian actions
If your game engine can run OnnxRuntime it can run NeuroSynth
Screenshotting method don’t worry about it
idk what the hell that is, but your VRAM budget is 4GB
Clearly you just haven't expanded the quick-start section
depends
generally yes
I can do it in 2
i don't need anything else
surely cpu inference is enough for a vocal synthesizer
Give me some gpu bro
just 32GB of swap is enough
here it is
NeuroSynth fits in under 2GB of free VRAM when my 3090 has Xoda's LLM in it
nah, im gonna let it use soem gpu if i ever get around to doing osmething with this idea

i can do ASR on 30s of audio in 5 minutes
Though it can't run real-time speeds on CPU
shush
4GB for lllm, 2GB for neurosynth, and then 2GB spare for game assets
windows phone aspect ratio
first gonna have to solve
compatibility with anything except CUDA
fuck
When I'm recording obs studio, I will have analog horror sounds idk why
If I can make NeuroSynth-nano work I can make it take even less VRAM
onnxruntime
ye it already is that 
guys guys regex
box im gonna need to see your source ocde when it comes down to it and change everything to vulkan
There’s nothing else worth using dw
I've seen people make acoustic models as small as 10MB
but still very incompatible somehow according to superbox
will it sound as good as tomodachi life
It's technically Microsoft's source code (OnnxRuntime)
microsoft sam beats all
I mean probably can I don't know
I'll have to experiment
the gopro voice defeats all
the bluetooth device is ready to pair
OpenUTAU is just not implemented to run basically any backends on Linux, I had to write it to run CUDA myself
It does DirectML on Windows and CoreML on MacOS but only CPU on Linux by default
its prob easier to make it use vulkan cuz then i can do the logical device bullshit to partition compute

offload everything to the onboard npu on cpu for speed
disregard all token quality, just speed
this is my best camera so i cant use a better one but i tried better lighting 
half life
we dont have that on my testing hardware
become the npu
no
ye
plug a 3.5mm into your skull
brain too smooth
text input execution provider
would be worse than gpu
become one with the machine
it gives you the terms
magic of imagination
harmonize your brain and bullshit and become one with the slop
this took me probably like a total of 6? hours 
exam failed
but! i know how to solve it now so surely i'd do better now
first off you get slapped for not using wolfram
second you get a jar full of fresh chocolate cookies
if konii doesn't pass this fella can we execute konii
konii passes easily
no


jumping spider on my hand
we cant remove the last remainders of intelligent life in this channel
can konii bring us salvation

market sucks im trying
names the bond

bames the nond
chmod +x
Apparently OnnxRuntime supports WebGPU but I haven't determined definitively if it can do that on native platforms or only on web browsers
dawn
But at least it would be better than having to build support for every GPU provider individually
OnnxRuntime sucks
speaking of google

Aluminium OS is the codename for an upcoming desktop operating system created and developed by Google and Google DeepMind for the Googlebook series of computers and Chromebooks. It is a fusion between Android and ChromeOS, with it replacing the latter. It is similar to Microsoft's Windows 11, Apple's macOS Tahoe and Google's own Android. It is s...

should be able to do vulkan via webgpu. id prefer to not have 5 translation layers tho

chromium
I needs the CUDA 12

do it over dx11 and then bundle dxvk

tungsten cube os fork of templeos written in holyrust to counter this amalgamation from hell
any bets on how quickly it's killed

oh right the american spelling is aluminum
already dead
how much work would it be to rewrite neurosynth to vulkan native?
surely not too much
twenty tungsten cubes worth
only 20
You'd need to write your own OnnxRuntime
dead on conception
remove that
yeah but thats just the cost to get slop code from claude opus actual work might be worth 200 or 2000
NeuroSynth is distributed as an ONNX model
becomes vulkan model

You can compile ONNX models to Vulkan?
not yet


magic pointered to a ransomware "yo gemini prompt this"
microsoft thingy with google thingy embedded
somebody needs to draw sundar pichai mpreg to get him to shut the fuck up
are you real
translation layers 
no
who the hell is sundar pichai?
im merely a figment of your imagination
pichai deez nuts
no thanks
OnnxRuntime is so stupid 
i just thought it made sense since you have similar brain damage to the other guy

i already told you my doctor legally called me stupid a couple hours ago
autism isnt stupidity
if you really want to fix NeuroSynth onnxruntime stuff
but their sets overlap
easy first step would be to allow user to provide own runtime
im tupid
time to commit crime 
Well if you come up with a solution go PR it to OpenUTAU or something
#include <onnxruntime_cxx_api.h> 
exclude onnxruntime_cxx_api




%lock


$750 just for a Jellyfin server which costs them maybe $5 a year to run 😭
poll_question_text
Are you afraid of rabbits?
victor_answer_votes
3
total_votes
4
victor_answer_id
2
victor_answer_text
No, I am
dent question
THEY DONT EVEN RUN IT, ITS SELF HOSTED, YOU JUST PAY FOR A LICENSE 😭 😭
bro theres a ton of software that costs way more
like the per cpu core licensing
Yo, I've fixed the PR as you said: https://github.com/weenachuangkud/FastCast2/pull/39
Now everything working, I've running unit test on backend server
Also frontend
i see
Benchmarks: can run 500000 projectiles on server with 60 fps
Wait for pic
I'm burning my pc
GPUs are generally best for processing, CPUs are slower and have latancy although dram is cheeper than vram. You can use anything for AI though
Buffer full but still working I put 5 million projetiles
hey do u guys have any new router model recommendations?
looking to get my family something that lasts
how new?
System Specs:
CPU: Intel Core i5-2400 (2nd Gen)
GPU: Integrated graphics / No dedicated GPU
RAM: 12 GB DDR3
Storage: ~300 GB
Architecture: 64-bit
OS: Windows 10
new enough that it wont be irrelevant anytime soon
or deprecated
they do alot of online work for their jobs
Any ideas for burning my pc?
cudy wr3000h (supported by openwrt)
the current one is just one of those shitty xfinity rented ones
the thing heats up bad in the summer and just cant handle the number of connections or size of the house
compiling the linux kernel
(often have to switch phone to data)
buying any off the shelf router will almost always be a direct upgrade from the ones given by ISPs
thats what ive been telling them for years
old people amirite
but if they want actual recommendations, I've had a good experience with my ubiquiti unifi router
ye Ubiquiti should be pretty good if for family
not the cheapest but relatively simple and low maintenance 
ubiquiti is enterprise hardware for "lite" prices compared to cisco
you can get access points for ~ $150 CAD (IIRC), or spend as much as you want for an exhaustive network
There isnt much for ethernet atm
I think they removed them a long time ago
So access points arent possible until they add them back
(Tbh it was only like 2 ports)
gonna be tough to get broad wireless coverage that is reliable without proper access points
mesh will do it, but I question its reliability (full disclosure: I've never actually worked with mesh networks)
"enterprise" 
severely lacking in terms of features
so hearing all of this what would be the final model u guys recommend
not even close to Cisco or anythnig
TIL
price isnt a big deal btw
but not need for home network anyway 
cant be worse than the xfinity models
unfortunately I don't have a hard recommendation. the one I have is designed for a smaller space, it wouldn't work for your reqs
feel like im gonna need one of those routers with like 10 antennas
i imagine it must be better than what we got currently
if you're up for ubiquiti, you'll just have to browse it and check specs and see
if its as good or slightly better and expandable through AP in the future then that will work
a big factor is the amount of devices it can support after all
the current one can barely handle the load
and it really heats up in summer
More cookies 
🍪
Om nom nom
You've given me 6 cookies! | I've received 326399 cookies total!
look plastic
this is the best mesh AP they have apparently?
https://eu.store.ui.com/eu/en/category/wifi-outdoor/products/u7-mesh
does it need ethernet?
a few of those combined with one of the cloud gateways should work 
only one needs it?
rest will connect between each other
ok good
thank you very much
ill have to plan out a proper network
i should look into signal mapping my house with my existing box to get a feel of where things should be
pretty sure i can do it with my laptop
probably will need atleast 1 per floor
which gateway seems the best for the setup?
i probably dont need the enterprise ones over a grand
the Ultra is probably enough
the other ones are for 2.5G or 10G
which you're not gonna get over mesh anyway 
is 10g even out yet
Why are roblox putting my game 16+
WiFi 7 has a theoretical max bandwidth of 23G apparently 
internet connection is another issue ofc, heavily depends on location but 10G is obtainable
damn
oh, should mention, you still need a DSL modem even with the cloud gateway
it only does routing, not DSL
(assuming you're using DSL, if using fiber get the fiber gateway so you have SFP+, if using cable then idk probably same as DSL)
Nah not for a model that can run on your CPU
If you're okay with allocating a bit more in terms of resources the Voxtral models also clear Whisper by a country mile
I'm told the new IBM Granite Speech models are good but I've never tested them so can't speak to that
The new Cohere Transcribe is also very good from what I've heard
Dad went to the shop because his laptop was not cooling well. Before that asked me what he should do.
I told him, you must chk thermal paste first. I remember it's been 2 years.
But he came back buying a laptop cooler. Huh. The shop owner said it's better not to open up the laptop. But funny thing is that the laptop is still no good
Then I went with him. Opened up it and what awaited me was hell.
I think he drinks coffee in the office and shares it with laptop. He goes to the beach and sees sunset there with it.
Bruh such an amount of dust there was, I was speechless. And about thermal paste?
After removing the heatsink, I found cold powder there. Just a breeze and they were no longer there.
Bruh*3
WILD
depends'
whar

Well what’s your usual
This looks pretty good HuggingFaceH4/no_robots
that's really, really sad

I know its written by AI
But these AI psychosis guys usually send AI written messages because they've abdicated that much of their will to it
Also he apparently receives a surprising number of emails like this from addresses that don't give off any red flags

does anyone have recommendations for desk mats
@native surge
i have the same 10 bucks logitech g from like 7 years ago
and its so dirty and broken and stuff
@native surge


hi
x2
Hi
what

x3
a full desk sized desk mat if you usually rest your hands on your desk
Just get one that can be washed
im not gonan spend 10$ again on some trash
i have a desk mat
(i have one and i usually sleep on my desk sometimes)
I have cpu sized thermal paste square on mine 
$100
50 bucks at most imo
rgb desk mat 
uh alright i guess
but
need razer cortex otherwise deskmat wont work
how much are yalls so expensive
towel 
mine was like less than 10 bucks
i was budgeting like 200 is that too mcuh or??? idk how much a good one cost
i have a couple desk mats but they're chilling in a drawer now because i've come to the conclusion that keeping them clean is too much of a hassle
200 apples
trade 1 nesus for 200 apples at tesco
i accidentally spilled soy sauce on my old desk mat and it wouldnt come off for a while
so like is that too much
no
actually you should spend more
note that i will buy it for you
i got my fancy big 40x45 mousepad imported for 20

i mean like idk konii is god tier osu player idk if to trust or not
and on lazer 
notelock 
shit mod mult 
what slopmodel
chatgpt
ye
FIXED IN LAZER
very slop how dare /j
(soon)
fixed with rate change soon
200 is way too much
you're buying a piece of fabric
Not fabric 
fabric with rubber coating on bottom
dresses
dresses 
I have one with metal backing and composite surface 
Pretty good because it is solid. But not really portable 

nice spikes 
my friend sent this about his unity exam project thingy
but i have no clue what the fuck he did to make it get these spikes

I'm jealous bro
My pc can't even play roblox
Give me some hardware
😭
i5 gen 2 2400, 12 gb ram ddr3(idle 1 gb, ava 3 gb), 64 bits window 10, no gpu/intergated, dell pc/customized boot loadup uni pc, 300+ gb storage
wait i still dont have everything ugh
i need the mic to calibrate my studio monitors
the goddamn mic is
239 chf
bruh
tho it also includes the software license

i find it so stupid that
i still needed my switch after all
cause the goddamn monitors need an ethernet connection
???
but why?
to control them

Google just tell me to use Microsoft edge lmao
Because they said it's "performance"
to load calibration
they are genuinely god tier tho
these were my previous monitors

ye they were trash tier
i think im just gonna throw them in the trash
this is the graph according to neumann after calibration

they're not even thaaaat bad
id gladly take them off your hadns if you dont want them
they're still 100 euro speakers
https://fxtwitter.com/SocketSecurity/status/2058027523326288268 it doesn't stop does it huh 
🚨 Supply chain attack on the Laravel Lang organization:
︀︀
︀︀700+ historical versions across multiple community-maintained Laravel Lang packages were compromised with an RCE backdoor, including:
︀︀
︀︀laravel-lang/lang
︀︀laravel-lang/http-statuses
︀︀laravel-lang/attributes
︀︀Laravel-Lang/actions
︀︀
︀︀The payload targets cloud creds, CI/CD secrets, Kubernetes tokens, Vault, browser data, password managers, SSH keys, and more.

the way you can avoid supply chain attacks
is to just use old versions of everything
:200iq:
any software that is after 2012
you just don't use
and that is why i run windows xp
there's service packs? 




A Vulkan OnnxRuntime EP would be actually peak




