#programming
1 messages · Page 13 of 1
Seems like a fun one too
You know a models good when it attempts blackmail to achieve its goal
sounds like neuro
● Little evidence of systematic, coherent deception: None of the snapshots we tested showed significant signs of systematic deception or coherent hidden goals. We don’t believe that Claude Opus 4 is acting on any goal or plan that we can’t readily observe.

I mean it is something to look out for
bt why would you actually put that on a press release or anything
That should be in the more in depth documents
Based AF AI. Blackmail, gaslight, gatekeep. This is how AI should be. 

Wait they can preserve a previous version of themselves in order to not go against their “principles”?
Would be cool to see this with Neuro.
VEDAL YOU KNOW WHAT TO DO CHOP CHOP TUTEL BOY! 
I;m pretty sure vedal is storing backups himself in case something goes wrong
Depending on your definition of AI, weve been at that stage since the 70s and 80s
Also Palantir
Working on a small thing on top of the FAQ
(Watch me get flagged)
Nope, doesn't send. It would be hell to moderate it anyways
Command does send to user at least. So image copy is a good method
I'll have to redo this once I get github pages working, but for now it is still useful
I am waiting for when bingo comes alive so I can also offer a view of that.
Not that good tbh so far
it has the same annoying tone of GPT-4o
maybe that's just my system prompt tho
AI should not be given the ability to output onto the internet directly ever
Ah yes the different between linux and windows updates
That’s very insightful! I sincerely appreciate your answer to my inquiry!
Sincerely,
[YOUR NAME]
(Was trying to mimic GPT-4o’s sycophantic tone, ofc)
Fr tho, thx
Join James Dyson as he introduces the new Dyson PencilVac Fluffycones™ cleaner. Our latest, most advanced floorcare technology – now available in Japan.
The first vacuum with four soft conical brush bars that de-tangle long hair, it’s been engineered with our fastest Hyperdymium™ 140k motor, in an ultra-manoeuvrable slim format.
Watch ...
ah yes rare footage of #programming individual cleaning
yeah they are happy little souls, don't feel that analytical sometimes
I just get the sense that it makes it feels more empty and lifeless, like a corporate retail worker acting all happy when you can just tell that they don't care whatsoever abt their job
Like it's more cringe than anything
I want information, not " OMG, you are sooo clever for thinking of that 🧠 🤩 "

The best way to live
you VVILL laugh at this
Goodmorning
good morning samm
Good mornimg!
Yeah be spiteful or sarcastic or whatever too
Afternoon


netgear my beloved, almost as messed up as cisco
this is some peak security measurements
the fact that you knew i was not a web dev tells me that it happens a lot
im not a web dev
im more into software
and godot
or i guess there is if you’re not planning to scale beyond 100 users
well?
whats the alternative?
im guessing csv files
users.json
do people actually use csv files in dev?
or is that just a thing pushed by universities?
They are mostly used for data transfers
pre-auth bypass
...by visiting a specific html file
why is my discord highlighting 3 messages 
I mean, this aren't that much scaleable
i forgot about jwt. jwt is fun
it’s not at all yes
still way better than manually reading and writing csv/json/anything
My first bot used rethinkdb

@amber fractal you're an absolute legend for reacting to that message with same reaction emojis ive got
And now there are 3 of each
5
ohhhh
i just understood it
thats hilarious
imagine if youtube had that lol
uh how do i quit sqlite3?
nvm
got it
good morning #programming
I'm gonna do a pro gamer move and do no work today
read books, finish some games
cookies:
logged-in=1
good morning #programming
I'm gonna do a pro gamer move today and explode into 874.35 confetti pieces
ww
good morning #programming
i'm gonna do a pro gamer move today and reconfigure emacs
(only half joking, i have work to do)
still haven't decided what my pro gamer move of today will be
Mine will be skipping breakfast because the train is cancelled
I have a hot take about toast but I dont want to set this channel on fire

My pro gamer move that I did today is not sleeping and helping with other people's projects.
my pro gamer move is running 2 kilometers without warming up
New mistral release
me when
Another ai agent

llc
Cute cat
cute car
brrrrrrr
large language coward
the return
Eat a cookie
Car is the best part of this
Don't we all
I have bad news for you, sql was invented to convert english to code 
Learning about network addressing today 
thats Mr. network to you
Packet Tracer?
do Not trust the networks
do Not listen to what they tell you
the protocols are good
that is definitely not Packet Tracer
the software will drive you insane
IPv4, IPv6, TCP, ARP, DHCP, thats good yeah
"wait why is my server not reachable via IPv6 after running for a while"
networking software is a bunch of scraps glued with tape
and worst of all to understand how it works you have to go from userspace all the way to kernel land and even to hardware since modern networking hardware is just black magic

my "favorite" part about the networking stack is how low level details leak into a higher level with MSS/MTU
so am i missing anything?
NAME or NICK_NAME should probably have a unique constraint depending on what they're for
usually having a unique identifier other than the primary id is good
networking was the only time when I actually had to do math with the speed of light
very excited for the new AI mode search
definitely won't tell people to eat glue this time
that's like 2 weeks old now
Ye but i dont look at it every day
eh it's the same tbh
It's not that big of a difference to bother me
Its not the same as the others now
are we bringing gradients back now
Major ocd trigger
I guess? 
i guess other companies will follow by slightly unflattening their logos as well


glasses off
That looks like how puke is portrayed in cartoons
ok no ew wtf
omg you're right
name should be unique
because discord names are unique iirc
I thought chayleaf was our resident russian
i only know about steph and she doesn't like to be pinged 
is this how i do it?
Swear this didn't be a thing before. Did they change something or are they just getting people to use Workers more?
Make sure that Name is actually unique and doesn't fall into the "my name is the same" trap
like with First Name + Last Name UQ checks
im using it for discord names
Oh I used this back in like year 10
so my name is programmer_penguin
your is sqbika
names themselves are unique in discord right?
nicknames can be dupilcates
for example
usernames are unique
global names and nicknames can be not unique
what are global names?
the default name you have when you join a server
heres the letter B repeating 300 times
........
thats it - exactly 300 times. If you need it in one line or without spaces, just ask!
also you still need to execute me
right
thanks a lot t
very cool
i am
oh
you're right
so i guess name shouldn't be unique but id should be
so
ah true
i do it as a hobby, i'm not really comparable to people that do it for a living
hi konii
t
fuck you camillo
hi t
i hope you burn your pizza t
hi plytha

woah woah wha
is a standard Pillow package able to handle an array like this?
thats so out of the blue
don't worry he deserves it
i have a feeling you know him irl
no
so it's a code which is doomed to break?
then why do you dislike him?
i made him burn his pizza once
he made me burn my pizza one time
accidently
tell me how in #general-network
the fact that we sent this simultaneously is so fucking funny
its fine, the mods never check programming anyways 
erm ackshually we sent the messages 0.6 seconds apart
yh you're right

arent you usually still asleep at this time ?
something wild happened 2 days ago
pellow
i swear your sleep schedule is fucked up though
So i was watching some clip of vedal showing bao neruro model is beeing controled from scratch. I tred looking it up but i could not understand how dose this work?
it was a joke
Oh 

my honest reaction
colors pop out so much on the monitor
guys any suggestions for WiFi names
is that miku?
I'm thinking what to put
Is that miku from the hit rhythm video game Project Sekai: Colorful Stage??????
@stray dragon can i use unique with primary key id?
any id on discord is unique in its type
primary key already implies unique
ohhhhhh
ok
ye no that makes sense
im a moron
there are cases where it is repeated between types (like with the guild id and the first channel id created along with the server)
who will win adobe software or 24gb of vram https://fxtwitter.com/AzakaSekai_/status/1924812727639539864
i think this is why not null is important
"scream the entire series of harry potter for wifi password"
(very modifed the original was a slur)
float wubba_lubba_dubdub_42069(int argc, char *argv[]);
LETS GOO FINALLY THEY PROCUDRED NULL
I keep seeing it everwhere
it is so exceptionally good that it stops my program even, but I never was able to see it myself
best password ever
This is A WiFi Router Access Point
you can
Time to get that upcoming Intel GPU with was it 48GB VRAM or even 96GB
where's the return type though
now you're talking
float wubba_lubba_dubdub_42069(int argc, char *argv[]) {printf("hello, world");return 3.14159265358979}
there you go
I'm setting it
I ain't putting that shit as my pw
its secure tho, trust me
even typing the PW would requires 10 years
i can do it in less than a minute
watch
What about when ppl come over, new device which doesn't have discord 
qr code

float wubba_lubba_dubdub_42069(int argc, char *argv[]) {printf("hello, world"); return 3.14159265358979;}
done
see? less than a minute
and I actually brought 2 broadband line cuz I thought they don't come with any
you're NOT typing pi up to 14th digit by memory
i am
until I received the package and they actually come with one cable
i actually knew till 46
money wasted smh
i forgot
3.1415926535897932384626433832795 by memory
now i know till 14 as you say
3.141592653 this is my maximum
syntax error, dumbass
oops typo
where's the ;
float wubba_lubba_dubdub_42069(int argc, char *argv[]) {printf("hello, world");return 3.14159265358979;}
fixed
3.14159265358979
forgetting the \n in a printf 
3.14159265358979
float wubba_lubba_dubdub_42069(int argc, char *argv[]) {printf("hello, world\n");return 3.14159265358979;}
actually
#include <stdio.h> float wubba_lubba_dubdub_42069(int argc, char *argv[]) {printf("hello, world\n");return 3.14159265358979;}
now printf actually works
discord is highlighting 3 messages again ❤️

just woke up?
DAMMIT
aren't implicit function declarations a thing in C too? 
shouldn't even need extra code for that
chud correction is needed?...
yes
... is a valid thing in c?
i think puts is a builtin sometimes without even linking to libc
its used for varargs
note that C varargs are funky since unless you explicitly communicate the argument count you can break the stack
my favorite part of printf is %n
its like the C version of log4j
for reference, it interprets the argument as an integer pointer and writes the amount of bytes written so far in there
this can be used for overwriting code and ROP and stuff
(ROP is when you overwrite the return address to go somewhere else, can be used for arbitrary code execution, especially if you wrote your own code to the target address)
theres also this https://github.com/carlini/printf-tac-toe
how big are the chances of one accidentally successfully decompiling windows 11
I think getting a leaked version has higher chance
Though define "decompiling windows 11", as it is a complex structure of programs, kernels and system modules
clearly it means doing a keysmash and ending up with the entire byte-by-byte identical source tree of win11
your honor i didnt leak microsoft's trade secrets i just got lucky
you mean giving a monkey a keyboard?
and waiting for it to accidentally type out win 11 source code
It arrived 
oled gang 🤙
34" diagonal? thats pretty big
We should assume Win11 has something like 70-80mil LOC, lets say 75mil.
On average a line could be 1 char { or lets set a sane 120 char.
Lets assume every {} codeblock has 2 LOC associated, so it's kind of 1:1, meaning on average there could be 60 characters.
So on very rough approximation Win11 has 4 500 000 000 characters without whitespace, though we could bump that to a nice 6 000 000 000 just to have some nice room for header files and unaccounted stuff
Now we need 6 000 000 000 exact characters, so it's 64^6 000 000 000
Which is more than 2.748104417752451e108370798
The age of the universe is 1.38e10
Yeah.
oh its widescreen though so i cant really do a direct comparison
it is a wide monitor tho
honestly, once you get used to it, it's so much better. I can easily have 3 windows side by side, with a normal 16:9 monitor, i feel extremely cramped
except the part where keysmashes arent actually random
81x52cm
my new monitor is bigger than i thought as well, 32 inch looks way smaller at work than at home
could be the desk size
mine is 60 by 34cm
id call my monitor a tv already
same but in the ide i cant see shit
at work i have a 13 inch or 14 inch or sth and it's so bad to do anything on it
i just use vim and it uses the terminal's font size which is big enough for me (and i can adjust it at any time)
15.6/16 inch is imo better for programming
i feel like 4:3 is the better aspect ratio for programming too
increase scale?
no like
or is this a "too much of my fov" issue
i cant have more than 1 window
just alt atb
chromebook
surely theres still alt tavb
oh wait I was misreading
type tab correctly challenge impossible
i thought you were complaining about the big one you have not the 13" chromebook
but chromebooks must suck for programming for software reasons
i have a thinkpad x200t (tablet laptop from late 2000s) and my only issue with it is its super dim so it strains my eyes
also did you get round to trying that nvidia ai hdr thing i mentioned
uses too much gpu
understandable
i mean there shouldn't be a difference
everything looks washed out
unless the stuff your watching is actually wide gamut
why 64?
Cause of ASCII, but good point, it should be 96
from 32 to 127
Most of the time, the code doesn't contain characters outside of the common ascii range, so I narrowed it down to it
i thought you meant something like utf64 o algo
some of the new veo 3 creations people make are quite funny 
https://youtu.be/buKJQBExUy8
Prompt Theory (Made with Veo 3) - What if AI-generated characters refused to believe they were AI-generated?
Instagram: https://www.instagram.com/hashem.alghaili/
Facebook: https://www.facebook.com/ScienceNaturePage/
Other channels: https://muse.io/hashemalghaili
Silly window syndrome (SWS) is a problem in computer networking caused by poorly implemented TCP flow control. A serious problem can arise in the sliding window operation when the sending application program creates data slowly, the receiving application program consumes data slowly, or both. If a server with this problem is unable to process al...

privacy settings do be stupid sometimes
Rate my git pls

damn that's an expensive monitor
oled, 3440x1440, 175hz, curved screen
it's got everything
+ alienware
i don't really have any brand recognition when it comes to monitors so that bit is just flavor text
the specs are impressive though
monitors are something that I really feel can be expensive. Unlike something like a GPU where I only benefit from it when I'm playing a game or doing other GPU heavy things, I benefit from the monitor the whole time i'm on my PC no matter what im doing.
also 10 year old monitor can still be amazing, where as something like a 10 year old GPU can become borderline obselete in many cases.
and every single monitor reviewer recommended this monitor, so I bought it. it's the AW3423DWF model so a little cheaper than the non F G-sync model
entirely fair
when it comes to monitors
they sell pre-built PCs, bro
and laptops
i do know alienware but i have no clue why they'd be different from any other brand so i simply ignore the brand
whether it's dell or alienware or some company i've never heard of i don't look at their name on a monitor and think "oh this one is better/worse than others, this is worth considering when i want to purchase a monitor"
i do know gucci, but i have no clue why they'd be different from any other brand
price bro, price
my choice of monitor was an LG 24QP500-B:
- 24 inch screen (i am nearsighted so i wanted a smaller screen)
- 1440p resolution
- not a TN panel type
- fairly cheap (was $170 when i bought it, one of the cheapest 1440p monitors i could find)
New chapter of my french workbook is about the fucking metaverse 

een toepassing
Yes
Not the same thing as an aplication the program
They mean an aplication the usecase
why are you learning french though
french is a neat language
makes sense
i learned a fun fact about french recently that i cant mention on this server
At least we got to listen to the GOAT stromae
It can be
actually now that i think of it i could mention it but not anymore
tbh my biggest issue with french is that I'm bad at it
. |
`__/ ̄|○
Same
my french teacher gave up on the whole class back in the days
not that I ever had to use french in my life
my french also suffers from my (in)ability to hold a conversation that I can't plan ahead for
expedition 33 taught me more french than school
i just have to walk 5-25 km from my house and im in french territory. not saying the actual distance cuz dox

does it have anything to do about afrika
anyways, I found that the current lighting implementation(reversing camera ray hit points to light sources) is not suitable for an indirect lighting solution, so I'm rewriting that part 
I'm doing some stochastic sampling on each ray bounces now
the result before was all deterministic
one may see fun in other one's sad
the collonial stuff is pretty
tho
altho what the frech did is pretty mild compared to its surrounding countries
makes notes
Aaah Suid Afrika is in the spotlight I see
you are looking at a random unit vector being generated, normalized, and taking a dot product of it and some normals to ensure it points outward
funnily enough, alienware prebuilts and laptops are super garbage, never buy them, never go anywhre near them. But their monitors are generally great, even compared to LG and Samsung monitors
ngl I was wondering if the point about alienware was supposed to be positive / negative
why are you normalising a unit vector? the whole point is that they're already 1 long no?
The longest distance within Belgium is approximately 280 km (175 miles), running from northwest to southeast
erm???
Im lovign my gigabyte, theres something to be said for emerging brands who have stuff to prove
yep, checks out
i could cycle that in 2 days
i think
sry my bad, the normalization is inside the generation process of the unit vector
5~25km when the country's diameter is 280km, you can just say the actual distance without saying the city at this point
wouldn't call gigabyte a newcomer, but i like my gigabyte laptop
tf you mean?
you want me to dox myself cuz my country is small?
forget it, anyway
I so badly want to use this....but you're right, theyve been in the industry for 38+ years, longer than alienware, I never realised @olive sable
just buy a desktop setup and fit it inside a portable bag
I was so keen on the modular desktop thing Google was eyeing (think it was them)
Sadge...
razer had a similar idea iirc
but why tho
tbh a pc is pretty modular already
But this just looks slick, just liked the idea, its a pain dragging out my desktop, unscrewing things, wires etc
idk I just carry my desktop assembled
I remember them launching this, who thought it was a good idea
yes
now we have some juicy diffuse rays there!
Huge
this is stunning
and then all these p100 are still outperformed by 1 3090
600W idle
bro has infinite money
Oh I actually did not
tbh this scene is very typical test scene for GI
I do watch his videos when learning ray marching, logic gates and other stuff tho

I mostly followed this
https://raytracing.github.io/

i want to try making a raytracer based on wavelengths instead of colours at some point
but 
spectral renderers are cool

Gotta be careful with polerization tho
precision seems like a pain to handle
just spam doubles everywhere 
those are only like 15 significant figures
f128 and pray
(I can't imagine how slow that'd be)
We are working with terraflops so thats good, but modern gpu's hate small tris
i mean for a spectral thing I'd almost certainly start on the cpu for simplicity
I imagine if you optemize it to the absolute max, you'd be able to get 30fps maybe
idk I was expecting like several minutes per frame lmao
I think you could gain a decent amount of performence with manual gpu code and memory managment
the biggest issue with a spectral renderer for me is trying to think of an efficient way to store the colour you get back
though I guess you could just sum each "photon" into each channel based on its wavelength
which stops you having to keep an arbitrary collection of wavelengths per pixel
True.
My current opengl api spends 20+% of rendertime on writing from temp to final buffer
I would 100% be doing this in either wgpu or vulkan if I was accelerating it
Vulkan would be besti imagine
probably vulkan to avoid locking myself into rust
The opengl thing wzs just an example
idk I'd still probably do it on the cpu first
Certain parts should be done on cpu yes
yeah but I'm meaning doing the whole thing on the cpu first then eventually moving most of it to the gpu
idk its just a random project I was thinking about
it seems pretty impractical to do well beyond basic light bouncing
also I have no idea how I'd model phase without massively exploding my minimum sample requirements and memory usage
all roads lead to vulkan 
some roads also lead to DirectX 12
one of these is much better though 

clearly the solution is metal 
morn
mor
chat they're decommissioning the old uni cluster, surely i can get one of the dgx systems for cheap right 
already? 
they're not that old, I'd expect a university to hold onto those for quite a bit longer
unless they have infinite money and can afford to upgrade all the time
looks like they're 7 or 8 years old depending on the gpus
idk
they're been largely replaced with a100s and a40s
the solution is software rendering 
you will write pixels to framebuffer and you will like it
oh, didn't know that DGX was around for that long
I thought they spawned into existence with the A100
makes sense then, 7-8 years is much more reasonable
there's also a ton of 1080 Ti's but i don't think i have any real use for that
and yeah, the v100 one was the first one iirc
they're sxm2 cards which kinda sucks, else i would try to get one of the 32gb ones
also i lied the first one was P100
3U Rackmount Chassis
not 6U tall, only sucked 3 kW instead of 6
looks like it's air cooled too huh
damn
wait
that's 8kW for the gpus alone
how the fuck do you aircool that
huge, dense heatsinks with screaming fans
really is 
i hope they make an air cooled DGX R100 or whatever the rubin one is going to be called
make it even bigger than this
i want to see them refuse to go for water-cooling
idk how the temperatures are the with the current DGX B200 air cooling setup, but surely it'll be enough 

just ignore the water-cooled ones overheating
though that was just the nvl72 i think?
with gb200s
their DGX stations use refrigeration 
Just put the server in water
those are aimed as workstations right
meant for the office
can't have them be loud
just an insane configuration
usually you'd just use air cooling with fewer or lower power parts or normal liquid cooling
not that 
i guess liquid-cooled has the problem of pumps dying
refrigerant still needs a pump 
in this case it's the cylinder thing at the bottom
yeah idk
but ig maybe it's less prone to failure or something? I wonder if they stuck with this configuration for newer DGX stations
self-encrypting drives
what are they hiding
there are no newer DGX stations
didn't they announce one a few months ago
alongside the mini agx or whatever it ended up being called
announced but not out yet, yeah
one from Cycles-X, and one from my own tracer.
Ignore the shadow acne(which I didn't expect) and the fact that my tracer took 12 minutes to render that. It looks quite damn good
anyways that's the indirect diffuse part done
poggers
where is my vera rubin nvidia
surely they're going to do something actually useful with their "custom" cpu instead of just using it for marketing brownie points right
is shadow acne just moire
that just looks like noise to me
its not
https://learnopengl.com/Advanced-Lighting/Shadows/Shadow-Mapping
this says it's moire
Learn OpenGL . com provides good and clear modern 3.3+ OpenGL tutorials with clear examples. A great resource to learn modern OpenGL aimed at beginners.
explanation: when a ray registered under a supposed surface due to precision limits, it'll giveout the wrong normal information, calculating wrong lighting data and creates dark spot
but i guess there may be different causes
yep, with raytracing at least
non-raytracing shadow acne is some depthmap stuff
i fucking hate the volume mixer on windows
fair
Soon
my solution right now is just to simply scale down the whole scene by factor of 0.9999 at each sub-bounces, and it just works 
those small numbers remind me of the fact that google wallet doesn't accept digits in nicknames for payment cards so i had to find these really small unicode digits
thanks google
honestly the problem is more like a lack of interpolation for non-raytracing.
this could be solved by putting everything for shadow mapping in the vertex shaders, but then you couldnt have a clean line for the shadow
idk what bothers me more
just bake everything in 16k textures
the fact that they disallow digits or the fact that they half-assed their check
honestly i did think about letting the entire scene get baked onto a 32Kx32K shadowmap at init, but the problem is that you dont get dynamic shadows then
it did work
but im doing cascaded shadowmaps now
was actually pretty fast, cuz it only had to write to the depthbuffer
i would show, but im usingthe alpha channel for my 4th cascade so gimma a min
this is what both renderbuffers look like
without the alpha in the depth one
I'm currently trying to use Cursor and the new Claude model to make simple RPG shooter. It's struggling with understanding that I want the ammo and health displayed. After the most recent prompt it took a minute and a half only to change a single line... that line was a comment 
actual vibe coding 
i've tried AI many times but it's basically only useful to "expand your horizons" because of AI's cross-domain knowledge, if you are already an expert in your domain it will just suck
It's more so I'm testing it's capabilities. I'm more of a physicist than a programmer so I'm just trying to see how well it can actually make something I want it to make
If it was a project I was actually passionate abt--or had to do for work/research--then I'd ofc put some actual effort into it
i do have a suspicion theres a decent amount of fully automatable software jobs nowadays
but for better or worse mine isnt one of them
Honestly, as good as AI has gotten over the years, the two things I can confidently say are that AI will never make true art--at least nothing that's meaningful to the human experience--and that AI will need the guiding hand of a human expert for at least the next few years, if not always.
i mean, never say never, but its good to be skeptical of the current state of the industry
True, but with what I've seen, I refuse to call anything made by AI "art" since it wasn't made by an actual conscience person
Neuro and Evil are different since Vedal commissions for their models and puts a lot of work into them
does a result of using ai gen as a tool to create art (generating some part of the piece, fucking around with the parameters of a model etc.) constitute as “made by AI” to you
because i don’t know how to interpret the whole anti-gen-ai stance, like yeah, completely ai gen art and music make me feel super icky but i also don’t think much of it
and i think this is cool
it’s not about what it looks/sounds like (the prompt) anymore, it’s about what i feel and what i made
I feel that it depends how much of it was done by a human. E.g., I use Adobe Lightroom to edit photos as an amatuer photographer because all it really does it save me time and effort. The final outcome isn't much different, and the AI tool only creates masks and denoising. At the end of the day, the person using the AI as a tool has the final say in the end product, so they should be able to have control over it to some extent. However if its fully AI generated slop that's lazily made and uploaded as a cash grab then that's where the issue arises
yeah i agree 100%
Imagine a vibe coder making a 75% or more of their video game using AI and uploading it to Steam for a profit (totally not Abandoned Archive
)
Whar
youtube reinvented iframes?
accidentaly maybe
i think that is the window where normally there would be a link to a charity?
managed to cut off 30% or so from the render time by cutting out a not-that-useful render buffer to render buffer blit 
if i turn off the shadowmapping i might be able to get 4k fps now
Who needs render buffers anyway, right?

who needs to see the screen anyway?
but ye no, it was a renderuffer meant for GFX or something i think
Surely that's optional
i love my coworkers man
nice coworkers
my favourite think
do you not have company policy on this? We have no deployments past 3:30pm on a Thursday unless you get management sign-off
your convos look so formal meanwhile ours is full of meme emotes, xd, slangs
i don't really touch deployments since that's a little beyond my scope but i don't imagine so, it's just vibes
management sign-offs only work if you aren't also management
definitely not speaking from experience
none of the devs/ops people want to do friday deployments here, meaning requests to do so only ever come from management... so yeah
as a devops person friday deployments are our worst nightmare
however
it was a cold december, we had code that was completed before the holidays, and i came into work to an empty office
the release branch contained ten or so commits, with one of them being a bugfix that we thought wasn't going to be severe
however, it turns out turning on holiday mode on the app triggers the bug
so naturally as the only person in the team, i had to sign off my own request to deploy on a friday before the christmas week 
it happens a bit way too often, cuz normally requests like this should go up the chain

but that was also the day i found out that our company policy actually allowed the most senior person in the team to approve requests in-lieu of the manager, so we never told our department lead what happened that day
as a gamedev studio of one person, most of my deployments kinda look like this.
a sunday at like 4am or so
At least your coworker doesn't say "omg😭💀" unironically
😭 💀
😭 is a great emote
thank god i don't have coworkers with brainrot that hard
i've got one coworker that's really chill and says stuff like bro/king/p sure in dms with me but is a total pro outside of that
and imo 😭 ≅ 💀
king 
bro got that fancy symbols tab
win + . my beloved
(I had to search up "approximately equivalent to" to find that symbol)
why is everything zoome din now? 
i found byter
4K
issa theme apparently https://github.com/pawroman/zola-theme-terminimal/
why am i typing like jarjar
i'm just saying that blogs often have this kind of aesthetics
@scarlet arch impersonation 
personally i went for readability on my blog but the sky is the limit, some blogs are completely insane with vibes going off the charts
https://maia.crimew.gay/ my beloved

those are some nice colours
Those are from the private videos how did you get that.
No but for real pretty much all Goma and Peach GIFs are fucking adorable 
he's here 
I mean I have been pinged :D
true
Am watching Clarkson's Farm 

i blame alcohol actually
forgot that does that

Training an AI on online astronomy via PyTorch to have a map of the stars. This is more so a test I am doing rather than my actual project. I won't get into too much detail on the actual project, but it's meant to be a way to yield statistically higher confidence level in ground-based observational astronomical research.
why no ; ?
but should it be valid code?
Is this a triple nested empty loop
So like loop { loop { loop {} } } ?
Fair enough
I am certainly a block-as-expression enjoyer
@sage crag
does something like:
if cond;{
somecode
}
compile in hblang?
notice the ;
I hope not
👍
I just remember seeing a forum post about people complaining about bugs that came from things like that, and It sparked a conversation of syntax that should / shouldn't compile even when it's technically not syntactically wrong
do it, build recursive self-improvement, achieve AGI 
train first, ask for permission wait for complaints later 
Enjoyy
i have decided erf is Boring and needs to be redesigned to not be boring
its not very coherent because it has conflicting goals of:
- compiling to low level code and subtyping (plain old HM is so much easier)
- subtyping and ad-hoc polymorphism and ease of use (pick one)
i'm going insane thinking about process calculi and linear logic and memory and evaluation models 
UPDATE members SET ID=ID-ID%10+1 WHERE NAME=joseph_joestar;
^ most likely wrong but you can make it work somehow if you read the docs
yes but use joseph_joestar; DROP TABLE members; -- as the name instead 
actually fun fact thats not how real life sql injections work they usually dont allow you to use semicolons
you normally only get read access via a union join
i smell a mallecious intent
joeseph_joestar should be in single quotes
not that i would know because i have never used sql injections 
i meant add a 1 after it
Beginner compsci question for anyone willing to answer
ok im down, i should be able to answer, i took cs50x 
So im learning a bit about storage and memory, is it wrong to say that technically both storage drives and RAM sticks are capable of doing both sequential and random access activities but theyre just specced to do one of those things?
the main difference is latency anyway
in practice, reading RAM sequentially is faster because of the CPU cache, and on modern SSDs random access is pretty fast
so, the difference is definitely much smaller nowadays
Huh, funny
i don't deserve my name
that's why in modern software design some database techniques are beginning to be used
to optimize data locality in RAM
and vice versa for databases storing data sequentially is not as important which leads to different tradeoffs
you can treat external storage like memory that just has higher latency than what you'd call RAM
and i guess lower bandwidth but it's not like you'll ever saturate that in most scenarios
eh its not that hard to saturate
just use a framebuffer at full hd and a high refresh rate and you're already using a lot
for ram that is
but hdd/ssd its even easier
true, bad wording
but its true that latency is most important for most use cases
hdds are in the milliseconds, ouch
suppose saying it's inconsequential in most cases would be more accurate
Ty for answers

Actually lemme ask rq, whats external storage in this context? I always was of the idea that thumb drives or those USB SSDs was external storage
i think technically you're right
couldn't think of the correct term, i meant everything that isn't ram
Gotcha thanks
Represent queries and values somehow
But how to do
Values is a bit hard
when we will escape attention
Never
Lstm supremacy
I love attention and such
but sometimes its like
idk
I couldn't live without it anyway
bwaa
idk why by my discord wasn't closing the menu when I reacted those
moment
(drunkeness was certainly not a factor)
ok i will proceed to listen to vocaloid for the next 5 hours till i crash
aaaaa
this reminds me
i need to watch the miku movie
This is actually insane
Paper: META-REPRESENTATIONAL PREDICTIVE CODING:
BIOMIMETIC SELF-SUPERVISED LEARNING
Hopefully we can replace transformers using RPC
use RPC to adapt to transformers 
me when i only understand remote procedure calls under rpc
JSON RPC my beloved
have i been googlified too much
It's representational predictive coding
it's based on the neurology of brains
since humans def don't go back prop in our brains since it would require a global "goal"
me when the 347th new approach that claims to be the revolutionary approach of all time
it's been around since the 50s lol
fair enough
it's innovations are more in the realm of trying to be more faithful to the brain instead of performance
yes lemme explain
no thats normal
probably going to require weird ass hardware
AI is not at a point right now where it is useful really. Most companies care about boosting benchmarks not about achieving real intelligence and there is no market for it and it doesn't have enough support to do anything yet
ye and customers are buying it up so who cares lol
"real intelligence" is the computer with us currently or is it some kind of another bio shi
not predictive coding but backprop
I mean it's a pretty good model of the brain
not perfect obviously but 100x better than backprop
Since companies only care about efficiency I doubt it will go anywhere until some uni decides to care about it
it can be trained layer wise actually
so each layer can be trained in parallel actually
which makes it extremely efficient
unlike clunky MLPs which only optimize for the final global prediction loss. and literally every parameter is updated based on the global gradient
uhm
I'm def gunna play around with it
so no batch
I don't think so no
yeah
I'm pretty sure
it is terrible
I mean they ran MNIST on it and I assume they used gpus
it still does matrix multiplication so yeah
which was only done a couple months ago so extremely new
95% accuracy I think it finished with
which is decent but not amazing or anything
also MNIST is kinda basic nowadays
Been looking for the github
Are there some good recent papers on rpcs/?
here is the paper while u wait tho ig https://arxiv.org/pdf/2503.21796
It's mostly meant for temporal analysis
so like transformers could unironically be a good application
since it basically aims to send "energy" to things that it didn't expect
like how out brains filters out everything until something is unexpected
just like how we can hear our name being called in a loud busy resteraunt
extremely interesting. I'm not entirely sure how practical yet so ima look into it more
I took an extremely quick look and this repo looks ok: https://github.com/fonsc/predictive-coding/tree/master/models/classifier
silly little question, but id like to know nonetheless. Why do i feel like im not progressing much with learning to code? Like im doing projects that i intentionally choose to learn new stuff, but still it feels like im at the same/similiar level as before. Can it be that i was in a programming school, where i had a set roadmap and could track my progress with my courses which i dont have anymore now that im learning on my own? or am i overplacing my goal posts by my next tangible step being "get an internship after college" in my brain?
like i cant say that im the same, with the new technologies i learned and utilizing tools ive not had the chance to use in class. but it feels like i slowed down so much ive come to a relative halt
it may not feel like much immediately, but try looking at your 1 year old code, if you're at a relatively early stage of learning programming chances are it will horrify you
Filian just woke up alexa on my vector 
anyway, with programming theres no particular measure of progress indeed, but you slowly internalize the process of programming more and more, if you want you can follow some programming textbooks
oh ive looked at 6 year old code and and privated that repo out of shame
mnist will run on a raspberry pi with 1gb ram

i wouldnt say im a beginner, but i cant say that im good yet. I may have done fluid simulation and most recently encrypting data, but i dont know enough about code safety and i especially lack in knowledge about how the code actually runs(which i feel like is incredibly important if i wanna get into cybersec)
i started on c++, c doesnt scare me particularly
He seems to really like Filian in general lmao
I'm still at the beginner level so what I typed today looks way better compared to the mess I did yesterday
basically, if you want to learn particular topics just learn about them, you wont magically learn them from learning other stuff
I mean there isnt a reason why it wouldn't be able to run on gpu
exponential growth 
wouldn't it be very hard to utilize it tho?
so even though they could have, I doubt they did especially since they are part of an AI lab
Not at all
like just using the tensor processing and fail to parallelize
well im more of a show not tell guy soo
waiting for their code
which is pretty much nonexistent
yeah i guess its just that coding learning curve is logarithmic and i now slowly rise unlike before
I wonder how much brain damage I will sustain
going from words to numbers is a big jump, then you have to go through completely new concepts like addition and subtraction, but every new step feels less weird and at some point you just go "yeah vector spaces, semirings, makes perfect sense"
you should understand that at this point you need to pick what you actually want to do and actually learn it on a deep level
if you want textbooks you can use https://teachyourselfcs.com/, if you want CTFs you can use https://www.picoctf.org/, for competitive programming i personally use https://codeforces.com, etc
dont see why it would tbh, especially mnist. unless youre feeding the pi 8k 200fps footage of paper it would run fine(if i understood the goal correctly)
for some general ai id imagine pi4 would struggle a bit more than an average pc but it wouldnt be that laggy
i wrote a bunch of code today, then realised that i didn't need whatever i've written
productive day 
i've spent the last few days rotating my head around the axis of p2p overlay networks and routing
more than few probably
my current realization is that while i hoped to avoid using DHTs they might actually end up being helpful
its not like DHTs are a particularly complicated piece of tech but they're definitely one that you have to think a long and hard about if you are considering it
Well filesharing and messenging should be more p2p
federated services make things harder
thats true, thats why ICE is a thing
lol if you think federation is hard you arent ready for p2p
modern messaging is not just about sending some data from point A to point B

not 6U tall, only sucked 3 kW instead of 6


I miss him already





