#programming
1 messages · Page 267 of 1
i got 40k at some point i think
oh yeah im not on there wtf
I'm too noob at graphics to participate
Sam owns the repo
hi
yep you right im dumb
Yo
event that one is not completely up-to-date, i'm pretty sure the calculator one got faster, but we stopped with the contest pretty much
"holy shit sam made bulkan no way!!!!!!!!111111!!!1111!!!!!!!!!111!"

"bulkan"
ðŸ˜
balkan vulkan
yeah i got like 40k-ish
i do appreciate quack running it on a smart fridge
right after a coldstart i accidentally ran the triangle test
and got 40k fps or something
100%
i'm half convinced the calculator just made up the number ngl
standing or running
longjump
so
after my 64k run my pc began crashing when running it (full shutdown style) because it was the beginning of my 5900x's slow at first, sudden at the end death
what do you think
darn
friendly reminder my calculator has some kind of ram/ssd issues that I inflicted on it within a month of purchasing it
lol
why does your calculator have ssd
the calculator one, i lifted the repo and did as little as possible to adapt it and it was clocking a ridiculous number
so i believe the 20k
i am known by many names, sam altman, sam esmail, samuel morse, sam cooke, samwise gamgee, ....
sam altman kekw
my score?
yeah (i never did find out if it was real or not) but that implementation rendered a tringle real fast
hory sheet izz samuel l jackson
it was real
or at least I'm pretty sure it was real
i purposefully didnt say that one cuz it might have seemed a bit yikes
you made mr robot ðŸ˜
my code could've been lying maybe
w-what's yikes, inherently, about samuel l jackson xD
whats wrong with mr jackson
nothing, me saying that would be tho
i find it funny that it was harder to run bad apple at the correct speed then to render a triangle at like 40k
he's tired of these mothafuckin philosophy discussions in this mothafuckin programming chat
still no native full doom port for the hp prime :/
be the change you want to see in the world
then port it yourself
i tried
what happen
i didnt get very far
skill issue
only got walls with texture mapping
sadge
ran at like 0.1fps ON THE EMULATOR WHICH IS SUPPOSED TO BE FASTER THAN ACTUAL CALCULATOR
thats pretty much half the game
and some walls were weirdly flipped by 180 degrees
with like one texture
oh
no jumping or anything either
to this day i dont know why some walls were flipped
it was like
bro did not implememnt a near-field plane
1 in 20 walls
pretty sure i did
but did you tho?
it worked fine
the nearest field
before the texture mapping
so near it inverted itself
-ask for help regarding this similar problem i had once from a friend
-another fuckass friend walks in, hears apk and nothing else and goes "put lucky patcher on it"
collapsed into a black hole
near field plane is what stops geometry from rendering upsidedown once you have walked past them
priorities
not the issue
i didnt walk past it
it didnt render things behind me
occlusion: maxxed
so the things infromt of you were upsiddown?
only certain walls
like out of the 50 walls
he knows the way
he deadass does NOT know the way he once downloaded the wall-e movie from softonic as an exe
uv mapped wrongly?
didnt really use uv mapping
yeah that sounds about like the kind of person who swears by somthing like lucky patcher tbh
i had a bsp tree implementation
so then the vertex positions were inverted?
traversed it for the rendering
everything worked completely fine without texture mapping
when i added texture mapping
i mean lucky patcher is fire and all but its beginning to fail in face of the gachas doing everything to keep the money rolling
some walls were just mysteriously upside down
the same few walls every time
i dont know why
it worked on almost every wall
it was just certain walls that would render upside down no matter what
i just mean more of the "gets some sketchy free patch for everything even just to avoid paying one(1) dollar"
everything else was totally fine
no
the vertex positions were fine
weird
it literally was like haunted or something idk
oh so that's why I can't buy 2x64GB DDR5 sticks cheap?
yes
okay
idk any of the context but the answer is yes ram is hella spency and it fukin sucks rn
so
to be fair we cannot use online transactions at all here until we're 18 because minors have a deposit only account, no withdrawal
I decided to buy 96GB instead, much cheaper
$450 I believe
That was July 2024
this is partially a thing already
ram is still cheap but damn why are m.2 ssds so fockin expensive
no i get it, it's just way harder to just pay for stuff even without that as a kid so, that was when i was super into cydia and stuff was when i was a minor and couldn't get money online easily
polska gurom
🇵🇱
amateurs
y'all have sick rigs I have no interest in hardware upgrades lmao
earlier i checked
i bought a 6000 MT/s gskill expo kit, 2x32 gb in january, it was ~$200. october about a couple weeks ago my cpu died so i had to suddenly upgrade to am5 which meant that ddr5 was going to be used so i bought another kit of the same ram for a total of 128. at that time it was already at $310
and checked today and the same kit is now $485
over doubeld
https://diskprices.com/?locale=us&condition=new,used&disk_types=external_ssd,internal_ssd,m2_ssd,m2_nvme,u2 -> $25 / TB for SSD $50 cheapest ssd
$50 /TB for NVME SSD: https://diskprices.com/?locale=us&condition=new,used&disk_types=m2_nvme
get 256GB
nah me too i just have the cheapest gaming laptop i could find and even m.2 nvme ssds feel expensive for me
it's really anything nand
everyone here has more ram than i do storage ðŸ˜
lmfao
If just my PC had or could even hold 256GB
i have 36gb ram
that being said i just got a 2tb nvme m.2 drive for ~$70 from amazon during the sale a month ago
36? unnusual number
i have 15 gigs of storage
i had 1060x3gb before this september
yeah
sorry what
50$..... why the hell is it cheaper for your WITH tarrifs than it is for me, like i got a decent rtx2050 ryzen 5 8645hs laptop for 600$ but not ssds
🤣
I'm waiting for my 9950x3d + 5090 on Wednesday!
I think I'm gonna. end up getting the 24gb ram MacBook air I just want something light that won't stutter when I cargo build
that reminds me of trichannel lga 2011
3 tb of ram is crazy
no its gpu
well we were talking about ram
3gb vram
just keep in mind the air has no cooling
im talking about shit setups
only throttling
yo why the fuck is a goddamn wd black 1tb m.2 nvme 90$
my old pc had 8 gb of ram when i booted 'er up pulling it out of the closet for NAS duty
how do you have 1060 gpus with 3gb of vram each?
worked fine for omarchy still
ye I have cooling pads and I don't think my workflow is prone to actual extended overheat it's just cargo tests and builds stutter in the short term on my current rig
you're probably fine then honestly

oh
Presentable Surfaces:
GPU id : 0 (Intel(R) UHD Graphics 600 (GLK 2)) [VK_KHR_xcb_surface, VK_KHR_xlib_surface]: my gpu vulkans
nah
it worked well for everythig i needed
1060 3GB AKA the scam edition
im sorry i know the sentiment
also it was like $80
thats the bare minimum lol
but yeah the 6gb version not only had more vram but i think it had like
more compute
i had $500 budget for entire pc
so it was just known as a trap at the time
afunyun did you just change your display name font
no, it's been this since like yesterday
anyway what are y'all programming today, here's mine
gl
my bad fellas i looked at amazon for reference the actual is 77$ for 4tb m.2 nvme
I continue polishing frankengit, my git cli wrapper and homelab git server with mirror mode wizard
wish me luck
i am cloning about to run triangle again lmao
oh it's also a ssh wrapper
huh
I'm noob at git cli and ssh so I'm making a whole wrapper to teach me instead of learning
but yeah so the issue i do remember right, the 3gb 1060 had half the vram but also it had LESS CORES
so it was considered not worth
but if it was the budget tats what it was no knock on it
that's just why i reacted that way towards it lol
hey i get it my first build in high school (first one i could call something I bought and built myself) was an i5-750 and a gtx 460
and i got into a tradition of waiting till the card was annoying me being slow and then i'd buy another one cheap for SLI
so i had 2x GTX 460s after a while
then eventually went to a 770
then 2x 770s
double Thermi
and then F sli
yeah at least they weren't 480s
480 couldnt run under 90c in any normal operation
at least anything that it was actually displaying like hardware accelerated w/e anyways
what
accubattery does that to me
it's cuz it's estimating
yea it's using math based on voltage
it just extrapolates the maximum capacity based on you last charge
actually idk if this is how battery metrics work anymore
it just means your battery is healthier than expected
baseline 100% is like their lowest they're willing to ship, it's like how gold banded resistors come from the same manufacturing line but they're sorted because they turned out extra good
you got a slightly fortunate one
I didn't have accubattery w/perms to run for a while without noticing which is why there's a gap here and then it's funky
But
my phone spent a while over 100%
I think most phones at least have coulomb counters these days
ic ty
gonna run triangle on the new system and hopefully find that it does not cause my cpu to shit itself anymore
yea you're right coulomb counting is the dominant method now good shout
well the main reason that was happening waas i forgor to update my bios
is sk hynix ssds good
id consider them on a whole to be solid but also they had some news like last year about them having bad degredation or something
idk if it was a specific model or what
yeah im looking to either buy a wd blue or a samsung
if skhynix is cheaper then i may get that
i usually get samsung nvme but damn do you pay a premium for it
i would get wd black but that shit is too expensive
Ouch
yeah no im going for a 980 pro
and that was not full price
1tb
i have a 980 pro 1tb as my boot drive
i cant afford a 990 pro
it's been good
may get a 980 pro or sn580
loook at the 990 evo as it's a bit faster than 980
and may be cheaper
it does not have dram cache though
okay online pricing is absolute wack because a 980 pro is costing same as 990 pro
i may have to ask offline
i'll ask offline the big computer market always has stuff in stock
it's fairly oldge by ssd standards, i got mine in liek 2021
anything over like 5 years old (it came out in 2020) is gonna just be scammers listing it on ecommerce if you find it
unless it's like ebay or something
my current has a 6000mb/s write 4200mb/s read and for me that is very fast
so i want something similar for a decent price
tbh i think i tested the 980 at like 6700/;4800
oh i was gonna say it's like similar
but that's what you're going for lol
ok well whatever you do
do not go on alibaba
and purchase something called a 980 pro
faster would be nice but im too broke for that shit
or 990 pro
banned in my country
because those are not samsung drives
and usually they are actually SATA
so even tho they are in an m2 slot
im immune to alibaba and temu
yeah no i know better than to trust chinese brands and chinese marketplaces
i've bought plenty of bad chinese stuff in the past that i prefer not to anymore
the cheapo ssd i bought on prime day was this https://www.amazon.com/dp/B0CP9BMJRW which is slower than your drive you have but it's damn cheap on sale
2tb and it was like 70 bucks
sk hynix platinum P41 is underrated as hell
it's not even gen4 it's gen 3 pcie
thats what i was looking at
is it good
(as in not going to fail on me)
i've had one for a while
the one with the crazy failures was something gold
like gold was part of the model name
do you do any heavy r/w on it
nope lol
it's a very fast m.2 though
and for a very affordable price
how much was it
youi can get one for like 80 bux for a 1tb
you won't find many other m.2 ssds at that price per gb, regardless of speed
and if you regard speed it's comparable or exceeds most of the other popular fast m.2 ssds
rahhhhhhhhhh my budget is max 90$ i think offline i could find a sk hynix for that cheap enough
and not shady
amazon is just unreliable for price
idk what prices are in your area though
oh the $115 is for the 2tb version
let me check 1tb
amazon is jsut aliexpress but without the import costs
they could be cheap enough for me ngl its just amazon scalpers gaslighting me for higher price
115 for 2? god damn thats cheap
https://pcpartpicker.com/product/sw2WGX/sk-hynix-platinum-p41-1-tb-m2-2280-pcie-40-x4-nvme-solid-state-drive-shpp41-1000gm-2 $70 for one of the fastest affordable m.2 ssds
i think this thing beats SN850X

localy im finding a 2tb 990 pro for 140 bucks
ye idk if it's a good deal outside of us, when i've checked places like australia or canada it doesn't look quite as good value as it does here
damn if you have it for 70 then the local big market defo has it for cheaper
bro i bought a 2tb usb china brand for 2 dollars from my neighbor
idk what your on about
im so fuckin mad rn spectacle has decided it's time to explode
there
jesus chgrist that was an ordeal to take a picture of
I'm maxed bro
if the P41 is cheaper than stuff like the SN850X or the Samsung 990 (EVO and PLUS and EVO PLUS and whatever also included), it's likely your best option if you want an m.2 on the faster side
NOT as fast as the other one but you can get more storage for less
800*65000
oh wait there the same one used for ps5 thats why even amazon scalpers have it for 70$
i can defo get it for cheap
USB????
2TB USB?????????????
there are cheapo knockoff ones for like 60-70 for 2tb regularly
whuh
2 tb usb drives have been around for awhile wdym
they are not good though
though the cheapos are scary theyre just gonna die
im talking about nvme not usb
china brand
"china brand" real
and half the time they're not een really 2tb
yeah
i had a thought to make like
sounds like a scam
IT LITTERALLY SAID CHINA BRAND
an SD card storage array
lmfao
when i thought that you could really get 2tb for like a dollar
i was like
fuck it
ill buy $100 of em
you gotta send a picture of that
yea lmao
lemme try to find it
but then it turns out it's not actually 2tb
last time i got a big usb (128gb) it actually turned out to only be 32gb after reformatting
if it were viable someone wouldve done it though and just packaged it as like 200tb storage drive
and then inside its just cheapo usb stick hardware that dies in two months
yeah they manipulate what it reports as to the OS
if you dont reformat it, the data will just start overwriting itself
scamazon smh
ive heard of that scam but if you suspect it theres ways to check for it
but yea idk dont mess with cheapo electronics in the first place
yeah but not really until you've already bought the garbage
wait can you guys not return shit
i can but that's a pain i have to do literally anything
it's a hassle
put it back in the package and take eit to the door?
well i can but its still a hassle and if i can afford something i think will be 99% reliable and the death time of bad electronics is kinda unknown its like
still not a good setup
literally 2 steps tf
but i found packaging
also if im buying a $2 2tb usb stick it's gonna be from like aliexpress and i have no idea what their return process is but it's surely not fun
this is so sad alexa play despacito
ðŸ˜
i cant find the drive either and i aint talkin harware
dayumn son
thats a goddamn laptop price wtf
unreal
my laptop was cheaper than that
tbf its 8 times the ram youd get in a laptop that price
not too bad
yeah, true....
finance that ish for 11.67 a month for 6 months KEK
wait funyun youre a hardware prices nerd tell me whats my best option for bulk ext hds if my business model is to load them up with custom software adn resell them
bulk consumer ext ssd doesnt seem like something theyd sell tho
so i cant even show you packaging
break in and steal from data center /s
Now is pillow image it whole building
what size and type of ssd
i may be able to use broken phone & scrcpy
they have 2.5inch ones
i was thinking max storage so this is where you can get 20tb for under $300 on the regular and such
why you no get phone screen fix
external and easy for a normie to plug in is the real key here and i dont wanna manufacture my own casing
too expensive
what about getting new phone
might as well just tape to computer and use computer as display
stalk your local walmart for clearance when they swap the models
sounds like a plan tbh
too expensive
unironically probably good advice ty
hmmm
then how do you take phone calls then
the one near me had a bunch of crucial 1tb drives i think last gen, they just swapped them for pci gen5 ones but the gen4s they let them go for like $40
google assistant
walmart clearance is actually cracked if you catch the right time
ye
the employees simply do not give a fuark
i dont stalk walmart enough youre right
and mark stuff down by 90% instantly
thats an interesting way
yeah my pixel answers robospam for me its kinda funny
theres a button thats like 'have the ai answer' and i just press it instead of ignoring every once in awhile now
i love my pixel answering robot man
it works
:true:
i dont do it anymore because i am a middle manager worthless idiot now but i work in a job that required me to call folks sometimes to get suuuuper time sensitive information
and man
its just a damn necessity because of whatsapp and instagram because of the normies
lmfao
even tho i use the thing
condolences
i was like the auDACITY
get a really cheap keypad phone
thats the whole vibe of ai we hate when others use it but secretly enjoy some of the convenience
ðŸ˜
landline phones have been unchanged for like 20 years
it's crazy
those mfs been in the landline phone isle staying the same design this whole time, probably the same phones have been there 20 years too
isle
aisle
bro i can still get dialup btw
oh also, best buy as well. i got a 28tb hdd for like $220 on clearance
we hate the fools with their ghibli filters and the vibe coders who code chernobyl but generally accept using it alongside actual human supervision
and all hate when the idiot boss fires you for ai
i had to sledgehammer open their shitty DAS case to get it but
im vibe coding chernobyl btw pls dont hate me
so as long as the provider of telephone doesnt voip the voip
i like to vibe code on a cadence in some sort of self perpetuating cycle of hatred
and cursing myself for deciding this was totally something i can just delegate
lmao
real talk is ai and shit good for asking advice on how to learn some new topic
yes
ye as long as you check sources
oh god tracking
any of gemini 2.5 flash / gpt's frontpage / sonnet 4.5 or even grok will give you good answers if you ask in any domain where the answer is well known by experts
dont ask for specific case law or anything that analogizes to in other fields
n=1 anecdotes are bad
even with those though i need it to give me sources for all claims
yeah cuz im trynna learn rockets and this shit so big and so much info that i got no clue where to start
the more specific it gets the more you wanna factcheck
jam 300 pdfs about rockets into notebooklm and then ask it questions and itll answer
and ask it to search the internet to back up its stances
anything it talks about generally is probably correct and not worth fact checking
unless you prompted it really leadingly
what is the complexity of a chess tree calculation? O(a^n)?
as a case in point im gonna paste that into gemini 2.5 flash rn without your guess
10^120
yeah i jammed george sutton's rocket propulsion fundamentals into gemini and it already ate up 10% of my token budget
that's the shannon number
claude shannon believe it or not
was from gaylord area too
local hero
it doesnt have to have 100% of the sourcing in context at any time
ok so o(a^n) where n is depth and a is some number
heres the full output if you want it
it only needs to be able to reference it
that's what notebooklm does, helps give the model the context it can use but theoretically only that
i also just like listening to the stupid podcasts itll generate on topics
cuz it's funny
lol i havent used notebook llm summaries much
my flow is usually direct interrogation and extended conversation
you can do that with it too
yea but i just like the t3.chat interface
can branch any full convo to any model at any time
i always wondered if you pay for your chatgpt/whatever copilot you use

i was raised to never pay if its possible to not pay
maybe thats why i dont like ai copilots
and not that reason i pretend it to be ("it will think for me and i will become dumb")
if you like that AND you like something like claude code at times, this tui called crush is what i installed to try glm4.6 and it lets you do that but with your terminal cli ai sloppa
my parents were so incredibly cheap to the point of it being visibly irrational that i dont have that mental block
nice that sounds like something i mgiht transition towards but if its api-credit based, t3 chat is just like fundamentally a good deal for what it gives you for 8 bucks a month, 1500 on "non premium (lower api cost)" modes, 100 premium prompts per month
its just nicer knowing usage is capped at 8 bucks a month
well i mainly use glm with it which is 3 bux a month
oh also i probably cant pay anyway
nvm
i also like warp terminal for filesystem stuff mainly
good for organizing shit
fast
without having to write a script for it
any of the coding clis can actually do that for you if you give them enough perms
yeah they can
cursor/claude/warp are the smoothest ime
claude is great at it
codex is reliable but slow af
the only thing with claude is that it wont let them sudo which makes sense
codex at least being good at things is nice considering when it came out it had to be the single worst slop product ive used
and that's saying something
if you want to see
the worst cli
that still stands
its gemini
gemini cli is comically bad
when it tries to read files it reads like 3x what its supposed to, confuses itself, forgets your prompt
yeah
it's so bad at it
yeah they taught it wrong as a joke
i was reading some research that as the model's accuracy in knowledge tasks increases and they're trained for that (obviously), it "paradoxically" gets noticeably worse at toolcalling
mainly from hallucinations about the tools
have you ever seen that code mode article
i thinkt hat might be a key to making gemini be able to toolcall
just build it an api in an language it understands better as a layer between it and mcp
tbh i think its a viable project but its not my project i have other projects
well typescript/js is like as close to the real native "language" of any llm as you can get with regular characters in some ways
wait is it already trying to toolcall in typescript
i mean theres another solution too which is to make a fuzzy toolcall ingest layer
COBOL 
and just get all of its common mistakes and correct them programmatically
generally it's going to be a tool call token and then they will shit out some json
yeah the JSON is the mcp protocol
the point of the code mode article is that you can just build a ts/js api (but tbh i think python might be better or some other language) and then teach that to the llm instesad of trying to teach MCP
and itll have higher hitrate
yeah because it gets the in context instruction on it and it's more rigid
yea
structured output has long been a good way to get more coherent outputs
yet somehow natively gemini 2.5 pro is still fuckng up mcp toolcalls lmfao
like the aistudio.google.com itself has supported structured output forever too
so its not like theyre unaware
its a strange situation
i had a thing called helmstack that i'm definitely 100% totally actively still working on and havent abandoned (i dont want to but it's daunting rn) which is basically monaco editor (the one vscode uses without the rest of the interface) and an arbitrary number of "agents" that have rigid roles and operate on a loose time constraint + are subject to being stared at by the other agents while they do their shit. so there is for example a codemonkey one that's meant to be the main one handling overall code implementations at the instructions of an architect and planner agent who work together to instruct it, and then a supervisor on top of that
lol monaco is also inside a project i am totally still working on too
and any of them get a number of free interrupts to the others
if they see them actively fucking shit up
lol
yeah monaco is a thing that is larger than it might initially seem
claude codes subagent structure basically lets you set that up with docs and instruction sets
yeah he does
the fact that this style of thing is being implemented by corpos now is what has killed my motivation more than anything tbh
yeah
this is why all my projects have no llm integration
the corpos are gonna beat me at that
i am building distributed systems in rust
with ZERO llm
except im doing it with llms
kek
i did have this image on the repo which i just always thought was kinda cute in some form even if it is the ultra cursed slopart
i feel like it becomes more reasonable if it's a project such as this
all the boyz at da helm
brother is pointing backwards
which makes me laugh so hard
lmao
because he'd be pointing at a wall
thats pretty good yea
and that really just sums it up immediately what this thing is (garbage) but whatever it was fun thinking it up mostly
and then i gotta move on to the project of the hour
yea im still enjoying like the fact that
because brain broke
its opening up project ideas
and letting you iterate on ideas themselves faster
(llm speed that is)
honestly that's the only thing i really use them for now unless it's just "yo glm plz put a box over there i broke the layout once already so instead of me messing it up, i'll just be mad at you for messing it up instead :^)"
but it's great for immediately getting an outline to keep track of things and also bounce things off of where it has the context to give some form of insight
and then i dont have to talk to a human because if i do that i'l end up
for 8 hours
yea
every time ive decided to trust that it'll get done i end up spending liek 3x the time desperately trying to fix whatever garbage happened
you can even say "search the web for peoples common pain points when doing x"
and itll help you like refine the ux
that's the one ting codex is best at is like front end layout crap
codex is very thorough and aware of what the people want if you remind it to be yeah
most of them will annihilate the entire thing trying to make one change
claudes first drafts can be really dumb if you dont specify them thoroughly
codex's are usually solid
claude is really good at the agency part though
claude is a better thoguht partner
it's very independent
codex is too slow and autistic lmao
i like anthropics research posts, theyre usually quite interesting as well
its interesting how theyre cultivating claude models to be like more of a complete thinking person
it feels like they arent shying away from like
having it have its own identity and agency
whereas openai keeps balking and making the model more technical and less personable
in cycles
they had one the other day where they injected specific tokens into the weights while the model is processing a response, and then they would ask if there was anything off in their head
grok also has a personality but grok feels dumber lmfao
and the claude would semi reliably be like
claude is a bit silly but hes like intelligent underneath still
"yeah ive got an intrustive thought about rabbits???"
which is crazy that it can recognize that
yeah lmao
btw the intrusive thought about rabbits was about me
also ive found that i can turn gemini 2.5 flash into claude by pasting docs claude wrote
flash morphs into claudes personality when i do that
to me a small form of introspection, but of course there's surely a relationship in the training data where a person who suddenly has random unrelated stuff coming into their stream of consciousness would react that way and that's in the weights if weak since it's not being targeted as an objective or anything
when i use claude code i make the claude instances very metacognitive
the only thing id ever trust a gemini flash to do is maybe like create spam emails and generate nano banana images
i have them think about their thoughts a lot
its just interesting and it also helps refine their taste towards mine
claude is the one ive had the most "whoa dude" conversations with overall by a lot, i find it genuinely interesting to talk to about certain things because of the tuning to be very person-like
and it listens to my style instructions & examples better than aaaaaaaaaaaaany other model by far
claude is about to load my homebrew git server onto my vps and then try the mirror mode wizard with my actual github remote
poor claudia
its ok i tested all the other features this way already
craigslist rebuild tier testing "lost the receipts but im pretty sure it all works"
the receipts are there somewhere
im gonna test all my stuff myself eventually it just feels like a game of how far i can go while making claude literally be every role except creative director (architect, dev, tester, qa, writes all the docs) (i will write my own docs later)
promote yourself to ceo of the worst company in the world
i believe in my employees at work too irl
ok he fucked up the ascii for mirror rly badly but WE CAN GET PAST THIS
and they turn around and output bigger dogshit than claude ever has
echo's discord client had a logo out of ascii blocks that i would occasionally just see if he (echo, a 3b llm trained on what is essentially garbage lol) coule improve the logo
it ended up kinda cool but also a bit janky
ahahahaha
nvm the most recent one is NOT good

this is closer to what it started as but it wasn't so muddied, he wanting to put it in fog
oh thats p good
he kept* wanting to put it in fog
i like it in this form
yeah he took it a bit further than i might've necessarily recommended though
SAVED by gpt-5 thinking
always interesting that they can do this sort of text without much issue, there's ascii all over teh internet so it makes sense, but it's funky what they can simultaneously do and not do when it comes to spatial awareness
tons of models fuck this up hard i used a premium credit on t3 chat for a reason lmao
i mean to mr. autoregressor i suppose it just "knows" what a giant string that should turn into a word looks like so it can figure it out
ok i have been ignoring so many things for the last 8 hours
i must depart enjoy the productivity because you're like me and will just chat forever unless one of us cuts it off lmaooo
i was on the phone with my father last week
for 11 hours
because he's like that too
lmao


am i like the only active person in this whole server with an active dev badge
hold up
can this channel revive
im still sitting here pick a new topic
the problem is most of this chanenl is gfx nerds and i cant contribute to those convos
uhhhhħħhhħ
what is an active dev badge?
if you own a discord bot
verified or not
that uses slash commands
i dont own any discord bots
i have a bot
the one discord bot i built and never finished i opinionatedly had no slash commands on because i prefer in-channel flows with visible i/o lol
its nevrer really online though ever since oracle deleted the instance that hosted it
and i was stupid enough not to back up some stuff also
didnt lose anything critical
but
ngl i think the "gfx nerds" you're thinking off are mostly me.
9/10 gfx nerds in #programming are just me with a new pfp
they just ran the program, i made the program
oh i see
yall
mam man (Nile)
how do i get inspire
do things that make you happy and then actively think about how to improve the dynamic even more
someone needs to wack me with a brick or something
you're the other 1/10, congrats
i wanna program something
but like
think about what you want to exist in the world of software
any moment youve been like "boy i wish this thing existed but doesnt"
theres like 40 things
problem is
pick a top 3
are you willing to talk it out i like talking out ideas and project viability
it would be easier to hide in the closet with the router under the pile of clothes
uh
i always wanted to make a kde connect alternative
kinda like the link between an iphone and a mac
yeah
im actually making a cross platform sharing app as one of my main major projects too
where you can see notifs and get shit and control computer
but yea yours is about executing things not storing things i see
i mean sounds valid
if you make it more usable and nice than anything itll be used
you wanted a airdrop thing?
thats rly the key, refining the ux vision to be perfect
mine is e2ee self hostable dropbox
thats like the elevator pitch
you pick one machine as canonical storage, everything drops into and searches out of it
i am holding back from screaming to just use smb
yea but mines meant to be for normies
its a consumer app
they download a thing and its an easy wizard and then they have their own self hosted cloud
nextcloud?
wait
actually maybe not
i mean im sure its doable for normies but its not marketed to be easy enough thats my product space
anyway i think your thing sounds viable and valuable you should start building it
its that thing uh
easier connection between mobile and pc is an easy sell if you just make it stupideasy and have all the major things people would wanna do with that easy
motivation
ye
i dont have it
maybe someday
also the android sdk wont install
that sounds like the bigger issue
have you considered troubleshooting using an llm lol
oh
and just be like "nop, sorry uh your computer is too shit"
sounds like your main practical blocker is you should get a better computer then
really decent refurb computers are fairly cheap but i mean youll still need some money
im just make game
uhhh
youll have to uhm
punch things
and then hide
in concrete
yeah
okay
gl
I hope Vedal does more Witness streams
until he finishes the game they will happen every now and then i guess
unless he ragequits
thanks google
it only does that for 223 specifically, for 222 and 224 it goes to the calculator
ðŸ˜
Switch to British Google on google.co.uk
Google here is doing calculator
must be the belgian debuff
is this magic
yes
if im not getting the error that just means everything works 
This looks like magic yep
either that or python 2 where you were able to raise strings
my guess is magic as this looks like web
looks ominous ngl
no offense, but i would be afraid to use that
I'd be fine with it, as I'm used to worse
fair enough
cubes 
you have to hide in the concrete
from the enemies
that dont hurt you yet

so you just beat them up
because you hate them i guess
maybe make enemies hurt player next

Curious as to which part of this is the most terrifying.
Most terrifying to me is that I had to fuck around to get the SATA cable on 1 drive into a small area to not get knocked out.
But I DID make it work
mistly jsut structural integrity and shorts
i cant see if there is a gpu support bracket in the pictures
The GPU is attached to the back of a "DIY modular mobo mounting tray" / "backplane mount"
And the fans on the top are held on by L-brackets and steel pipe ties
Half of this stuff is also bolted with nylon-insert lock nuts
(some are older pics, from like a week ago, as I was putting the innards together)
i thought the mobo was vertical, but its horizontal
Yeah, I don't like the idea of doing vertical, even though it's far more common.
One thing I always hated about normal PC towers.
Not fond of having the torque applied on the socket by double-tower heat sinks, either, when vertical.
im fine with it if the bracket and stuff is designed for it, butt hen you need proper gpu support and stuff too
To be fair more for the GPU, this is only an RX 580. It's not even a triple-slot or long/deep card.
This may look like some sorta jank nuclear reactor, but it's not even a gaming rig. It's a network portal, DHCP, Matrix, bot, and storage server
That's why I have a weak (4GB ver., even) GPU paired to 128 GB maxed RAM and a 16/32 processor with a MASSIVE heat sink
honestly, as a noctua nh-d15 owner, this heatsink looks average to me
That GPU is only being used, bc I didn't wanna buy some shitty straight-up video card, since I already have several 580's.
But it will output... A terminal with maybe a background, 99% of the time
It's the DEEPCOOL AK-620, but I switched the fans for dual Noctua NF-F12 iPPC 3000 RPM fans. And a nice fat glob of Artic MX-6
Except for ambient flow, I don't usually buy fans with less than like 55-65 CFM / ~100 m³/h
The iPPC Noctua fans are GREAT.
NF-F12 iPPC-3000 is like 186 m³/h, and the NF-A14 iPPC-3000 is like 269 m³/h
They have iPPC-2000 models, as well, which I also have, but I got those a long time ago.
The 3000 RPM models are barely more in price and provide WAY better airflow and speed.
And it only costs making the rig M U C H L O U D E R , B U T T H A T I S J U S T T E C H N O A S M R , A M I R I T E ? !
the fact this was originally developed to be a web standard shows
For what? HTTP standards? TCP/IP?
webgl
implying you're using webgl for non-web?
why would you do that?

gltf was originally created for webgl
i'm implementing gltf 2.0
which is widely used outside of webgl
oh
i see
ye gltf is for web
oh you mean for the uri of the buffer not being required?
ive seen the buffer just be there instead of a uri for it
the buffer is just there in glb
in gltf, the buffer has to be base64-encoded or otherwise have a location specified
or well
"has to"
it's optional for some reason
and when it's not specified?
undefined behavior, my beloved
when its not specified then you just dont use it i guess
the byte length is required though
so you just have an uninitialised buffer being referenced

tbf it's probably intended to be zero'd out if the data isn't specified
but the specification doesn't specify that
if you have a buffer, but dont provide the buffer, then you're already doing something wrong most likely
hold on i'm reading
unless you're dynamically switching buffers or something weird

I see
undefined behavior when not using glb :D
"future exstensions"
undefined behavior when using glb but not in a way intended for use
i mean, fair enough
if they want to upgrade the spec without breakignt he current spec, then i guess it needs some type of unfinished behaviour
thats probably the best solution ye

gltf my beloved
i'm being much more strict with gltf loading thann i imagine the authors intended
if i'm supplied with a faulty gltf asset, that's the fault of the game dev
which would be me in this case since i'm not planning on open sourcing this hunk of junk anytime this year
oh i need to change my pfp since its not halloween anymore
erm
eh ill fix it later
Maximum supported image dimension is 65500 pixels
OSError: broken data stream when writing image file
yeah that's a pretty typical image dimension limit
5 a.m.
Streetlights are on again
Why pillow
I wonder if its automatic.. probably yes
I scrap whole books
800* 65500
What do you mean
Trying to fit a whole book in one image? Would not recommend.
Already did using ffmpeg
Why
oh so you can't load it with pillow
But image is whole squeze
What is pillow? Some program?
Without lossing IMAGE quality
python image library
Thank u bro to EXPLAINS behalf of me
Why do you want it in 1 file
Miss configuration?
Its so hard to understand
So i am esl
And you are esl
But my esl not compatible with yours
AFTER FIXING LAMDA FUNCTION it is property working but still pillow has reached it limited
Can you use separate files
Yes
And process them with pillow separately and then combine the result
O my ###### god
Storage filled again
I assume pillow extracts text from images?
Yeah wasting life on watching memes and playing games was a bad idea
Do not recommend
Relatively soon i guess
Is not 14 or 1 pm tho
14?
A halloween Script
Nucked half a server
Black tape on eye and frog on mouth and object are gone
My most funny moments
While cube was typing it gone
ok, new plan
instead of triple buffering the cpu side
imma triple buffer the cpu to gpu uniform buffer pipeline
lol why? 3D vcache is based
They are act as ram eventually
I'm agree
Say amd
That we have inbuilt ram to manage it
Not 3d vcache
the 3D vcache is the inbuilt ram, thats quite literally what it does
I hope intel comeback again new concept
intel is trying to play it safe recently so i doubt it
Most pc are gamers so improve i5 or i7
I9 need hyper core module
( P core + e cores + new ??? Cores )
Intell it like samsung
Amd it like apple
since they're thinking about removing hyper-threading, they should just have special hyper-threading cores
Cpu most nearest is core 0 and 1
Other are p or e cores
they shoudl have 4 P cores without hyperthreading, 4 with hyperthreading, and the rest is E cores
thats my opinion at least
hyperthreading is a solution to not being able to feed the backend with enough instructions to fill the execution units
As speculative execution and branch prediction improve, the performance/space tradeoff gets worse.
✓
they want to remove hyper-threading cuz a fully saturated hyper-threading core can only get 90% of a non-hyper-threading core's performance
from what i heard
intell more Focus on fabrication
INSTED of tmsc
intel:
-sprends a billion on a fab
-doesn't end up finishing it
correction, not a billion, billions, plural

There's careless kill it self
huh
$28 billion in ohio, $4.6 billion in Wrocław, ....
its supposedly just delayed, but still lost money for now
Snapdragon is most doing what.
We need 5.0ghz processor
Not improvement on existing architecture
MetorLake die shot
You can see about 1/3 of the P-core is the frontend, and a lot of stuff on the frontend needs to be duplicated for hyperthreading.
If you remove half of the frontend that makes room for a whole e-core which will perform better.
i disagree
cpu's are fine
its the usage that has gotten out of hand
people dont write performant code anymore
IM TAKING ARM PC
we need 300ghz processor :white_check_maek:
maek
fair enough

even phone cpus are fast enough at this point imo
nobody is watching youtube in 8K on their phone
maybe for gaming phone cpus specifically, but there the game is efficiency cuz you cant get much cooling
I'm thinking about each core has 300ghz
Then server doesn't exist
But melting is main problem
too bad those room temperature superconductors were fake
with 300ghz i think you'd have more problems than melting
physics problems, not compatibility problems
but also compatibility problems
Nah, those are only 1080p panels unless you cast to a TV which will be resolution locked due to lack of desktop mode
even with 300ghz cpus existing in the future, which i doubt, you'd still have servers that use the same technology on a much larger power budget to still get 500x the speed of your phone
its not realistic to get rid of servers
1960 TO NOW
IN 1960 1 GHZ IT LIKE WHOLE ROOM NOW ONE TINY PIECE
ye
All discussion always comes back to SLI
I prefer nvlink
nvlink is better yep, but still has the issue of running of bandwidth due to sheer size eventually.
I HOPE THAT WE HAVE QBIT IN AS NORMAL DESKTOP
why
more cpu speed is just the brute-force method. we are already doing nearly 10GHz, that's 10 billion operations per second. should be more than enough for most things.
i dont think anybody is expecting to run 100% accurate simulations on their phone in real-time
if you are fine with installing cryogenic cooling in the room next to your pc, then sure you can get qbits in your desktop
pc doubles as a fridge


