#programming
1 messages · Page 493 of 1
but now I am suffering from the lack of docs as I expand the networking setup from test rig to production
if you structured your project well it should be easy to swap the lib
to at least test if it will work better
what
no tests and bad code structure? I'm taking away your keyboard perms
fumoquest is MMO I making 
eventually need box for host and test cerber
does nixos-hardware support server stuff
I thought that was for like
laptops
why dont you check nixos-hardware

I wouldn't call it bad
shiro want to put it on my server, but is afraid of being doxxed, so now it goes on your server or something
but it was not written with gutting the entire networking library in mind
i'll see what i can get done for you 
https://github.com/NixOS/nixos-hardware/tree/master/common
at the very least you get all of these common things
not urgent like at all
but would love to know if open to idea eventually if has spare box
shuni also has spare machine it seems
that cheesi
and ofcourse my server
officer cheesi
maybe I ask vedal for spare box 
it's just a good practice to abstract away dependencies
unless it's impossible to ever find a replacement which I doubt is the case
my informant says
playit
basically
is tailscale but only for games
I can test the neuro vscode integration for him with fumoquest
vedal new server is actually for fumoquest


ye anyway nixos-hardware can configure graphics for you automatically

very convenient
strongly recommend
I don't like this very much, too bloat for me, so I don't use it 
if the problem is doxxing i can change some of my server settings to make it closed or whatever. my tailscale guy says he can just add you but limit your access to my stuff too
i like it very much 
i think if you go through the tailscale guy i cant see your ip
external dependency seems unnecessary for 3 options that are already in the default config anyway 
but I can see why you would like it 
Tailscale not hide IP 
can still see it
I am always worried about doxxing
if i ever get a cat im going to name it nix flake
i wouldnt know how

it unfortunate that I need to be worried about it
how does a vps work? jsut another server to be a mailman?
have a public vps as a gateway i guess to route things through
or maybe libc if it's not a black cat

VPS is just a VM hosted by some provider
I mean if I'm renting the vps I'm not gonna just use it as a proxy
for you to do whatever with
I'll just put fumoquest on it lol
virtual private server 
ye
i thought tailscale hides the ip by assigning it to a 100.x.y.z
you can sometimes still see it
what about screeps
it not do that 
in the admin panel for example
connection is (usually) still directly between hosts
creep is a good cat name but
you cannot take cat into public
FJCVTZS
it shows the endpoint addresses of every device
also if the cat is actually a human it might be disgruntled
that too, yeah 
name cat "; DROP TABLE Pets; --
ive seen lots of chinese yuri stories where the main character is transformed into a cat (and transforms back later), so you need to be careful
cat neuroPogHD
idk how much of this needs to be hidden, so i just did most of it
ye so
ping a few of those hosts
see what happens in that column on the right
name your cat
where it just says "offline" or "-" right now
macintosh

I should just not be cheap and buy or rent my own box
?

whgat should i name my terminal theme
makes me think of cherry
ok ye, i see
because pink and black
hmmmm

if you want really cheap there's like https://lowendbox.com/ and similar, usually with yearly contracts
but would recommend random 5€ VPS from Hetzner or OVH or something instead, more reputable and flexible 

so you're trying to hijack the system?
im pretty sure i can just use ilo5 to get it back
gotta find an ILO bypass
since i have direct mouse and keyboard control of the active session nothing you do would work
you can probably kill the iLO from the host OS if you want 
flash bad firmware or something, idk
Snip snip

iirc in the bios there was a backup firmware or something
I'd also be surprised if that server didn't require some level of bios update auth
free heater
i bless you
"global descriptor table"
names for pets 2026
Even if you disable iLO software-wise, it can always be revived using the Physical Security Override Switch

it likely does, yeah, should all be signed
still lots of other settings to mess with but not sure how destructive you can be with it
name your cat "trusted platform module 2.0"
kettle 
name your cat "Filesystem in Userspace" (FUSE)
{
while(p‐‐!=0);
}```
wait will it be p - 1 then compare p or compare p then p -1
--p <- before
p-- <- after

oh ok noted thanks
--p--
Is that your doing or your mother?
do we have enough photos of Sam's house to map it out yet
me
while (p --------> 0);
I remember I got bored once and figured out vedals kitchen layout
ig itll be easier than using
{
unsigned int x;
for(x=0;x<p;x++);
}```
who would do that
you can just do while(p--); btw
but
compiler may just optimize it away to do nothing
compiler doesnt like you writing useless code
you have most of my room, and the theatre room. you're missing about 5 bedrooms, 3 attics, the basement, the living room, the hall, the 2 desks/studies, 2 bathrooms, and the outside
the famous "goes to" operator
But compiler didn't protest 
how???
while (p-->x) {
p->x--;
}
disastrous typo
void delay(unsigned int p) {
if (p) delay(p-1);
}

but what methodology diod you use? just extrepolate from the random noises you heard in the background of the cooking stream?
no longer in the server where I sent the image
what the fuck is this huh????
iirc I was specifically height checking vedal
recursive implementation
so I had to figure out some scale
now your delay function has extra overhead and can cause stack overflows
you're welcome
recursion
recursion
to fix stack overflow, jsut need more ram. 256gb now, but can expand up to 1.5TB
recursion

recursion reminds me of
recursion
inscription
the division algorithm when I was working on that virtual circuit board computer
This is what happens when you forgot base case. Smh
the cursed division algorithm that randomly failed for some reason
i rember
I still don't know why it failed
also compiler smarter than shr
I don't know what went wrong

if it just called delay(p); without the if it might not even have the ret 
tailscale funnel to public ip, i have one of those
cake
new species
ye it is
though you're using Headscale
so you don't have that
i can do a seperate tailscale i think probably
good name
@real sierra
how come playit doesnt hide the ip?
Which register holds the output data for a port?
CTRLx
???
playit not for remote access to server
it for game server users
i do have ipv6 i think
surely that helps 
modified theme a bit because i prefer it slightly warmer 
isnt there some free oracle server with arm cores?
just use those as a vps or as tailscale exit node
Always Free
Compute
Arm Compute Instance
Arm-based Ampere A1 cores and 24 GB of memory usable as 1 VM or up to 4 VMs
Always Free
3,000 OCPU hours and 18,000 GB hours per month
hmmm

i like warm colors but i think i prefer the higher contrast in this case
Unity not support building for Linux on arm I think
unless you talk to sales and give monies
ye but just use it as vps
wasnt the point of the vps so shiros ip dont get leakesd at my server?
no
background #16100f
foreground #fbf2eb
cursor #ff9973
selection_background #ff7fa9
selection_foreground #130d0e
color0 #1f1315
color8 #6d474d
color1 #c95c30
color9 #ff7d50
color2 #e57345
color10 #ff9d6e
color3 #ef9f61
color11 #ffc995
color4 #cd607d
color12 #ff7fa7
color5 #a53a71
color13 #ff5da8
color6 #dd799a
color14 #ffb3d1
color7 #f7ddd8
color15 #fff8f4
updtdate ye
mine is this
its ported from my nvim theme
which is a very specific version of the nvim vscode theme that doesnt actually at all resemble original vscode theme, plus colors replaced by my base16 colors

wrr
this layout is horrible
like every possible anti pattern here
would work but would be unnecessarily complicated
if already have VPS might as well run the thing on it
if you think this is bad then you havent read any of my code
transparent 
welcome to k, it c but konii
wide code is good because its easy to follow the code's high level structure

It's not THAT bad tho
cloudflare tunnel?
But anyone who has access to the vps can see the log anyway 
log(log(n))
no reason to do that 
actual k is much more terse 
Konii playing something other than OSU!? The world is ending
does cloudflare tunnel hide shiro ip or only mine?
oh yeah i have the server ready with discord bridge and everything
just need to wait for emi
konii play shaw, osu, tetris, balatro, terraria
many
me play balatro yesterday
Look sam, anyone who has access to the infra can always just see the logs 
hence the cloudflare tunnel no?


You'll see "connection from ###.###.###.###" 
cat ~/.cache/.../log_file_containing_ips.txt
so what you're saying is that its not possible

Look at how TOR do it 
wrr
idk who TOR is
onion
oh
john tor
The onion 
idk how TOR do it

They uhh, depends on 3rd party that controls other node
routing traffic through three random volunteer nodes
then you might aswell just use those volunteer nodes

unless its like some random phone
@real sierra host fumoquest server on your phone using termux


The thing is, it's not the usual TOR. It's specifically the onion service that is doubly anonymous
You have like, rendezvous and stuff happening
cant you jsut route ssh through tor then?
You can 
i imagine it wont be too fast tho
Ye, the latency due to geographical separation
Yeah, that could work I think
But whoever wanted to connect with you would also need cloudflare program 
im pretty sure on shiro's side for 0 trust he can jsut type int he website
and cloudflare does the ssh stuff
shiro you should host fumoquest on github actions with ddns
you can
you can
cloudflare tunnel on pc
ssh over cloudflare tunnel on pc
work
i trie before
ye
that work
if shiro wants to do it that way i can make it work
This was me before
Host the.host-that-you.chose
ProxyCommand /run/current-system/sw/bin/cloudflared access ssh --hostname

i also have domain
To be fair, I also do use arch for my media server
apparently works, but not yet in nixpkgs. need to use flake
way to confirm the stereotype
-# 1 unused because only exists because funny
cuz i dont need any
Lmao, I meant that as a consolation to sam who uses nix for his server os
from konii i would expect yuri.dev or something
mine was cheap
I have .xyz 
"ublock origin: bad tlds list"
Haha, you all can't have .id 
(it requires Indonesian national identity card)
not available 
but ye would be very expensive probably

Btw, does the eu have tld?
im guessing mine was cheap cuz its my name and nobody likes my name
ye
Oh, that's interesting
.eu tld
Actually, 3 letter domain are premium
They can go as high as thousands of dollar 
What is belgium cc tld?
.be
But the city! You own the city! 

there's still a casstle with the original name spelling

not exactly 200 years ago, it was a gradual change cuz people got too lazy to spell it with all vowels
Krechek. Indonesian food
i see
removing nvidia prime did not break it, so i wonder why i had to add it int he first place
lets see what happens when i remove mgag200
did someone say domain
hi
Domain expansion 
i have too many
not much. i shoudl prob reboot for changes to take effect? 
Why hoard it then 
because i'd like to have them regardless
you know what. ANC on my headphones doesnt even filter the higher frequency of these fans


Yeah, ANC is pretty bad for high pitch sound iirc. I think skoda uses that fact to make the bell 
ok ye it broke the 4090s hdmi output.
i wonder if thats due to the mgag200 driver or the nvidia prime

it's not that it's particularly bad at it, for the bell it's just that particular frequency they seem to let through clearly
Which means the anc didn't do its job properly 
i'd say with my airpods the lower frequencies like being in a car is far worse
you're using both the output of the 4090 and the iLO?
than high ones
ilo to my pc, 4090 as output that isnt 1024x768 cuz i cant see anything 

i mean yeah it's difficult tech you'll never have perfect anc
would just not run Hyprland on the iLO display personally
ye im not
-# is that even exposed like that, idk, I never ran any GUI app on my server
But my 4090 hdmi looks like this now so i cant use it
maybe
i removed prime and now its liek this
it worked with prime
but then had weird issue with windows being choppy when moved
ok, adding the mgag200 driver back didnt fix it. so prime is the reason i had dispaly out in the first place

can you disable the integrated graphics entirely?
not a good idea permanently but would be nice for testing
then ilo window stops worlking i think
yee
intentional 
to avoid any issues with stuff using the wrong output or the weird mgag200 driver
it might be high cpu usage cuz its copying display out from the 4090 into the ilo
i guess jsut disabel sync?
reboot time
i can feel the air from the fans all the way at my desk
buket
bmc works like 25% of the time if i connect after the os is loaded
just accepted that it's unhappy about doing that
clearly a skill issue but doesnt really do anything when rdp is rigght ther

32 bit libs? (steam moment)
thats fine you dont actually need an executible in the 64 bit world of the future
or to spell executable right
we're past that
i actually ended up using less than 50% of the weekly limit on codex, ye just use 5.5 honestly
E2B at Q8 would be better than 4B at Q4-5
The Gemma 4 series is highly sensitive to quantization, doubly so for the tiny models
Thanks. But it turns out they both are not ready for my shitty laptop 
Idk man, I tried running it using lm studio and when I chat it up it crashed 
bmc jumpscare
That might just be an lm studio issue
Maybe the model is just afraid of computing heresy that I'd commit if they run 
The amount of people in here who have had issues with it and then switched to llama.cpp and been fine is crazy
To be fair, it could be. But I can't be arsed to install llama.cpp
you'll never guess which inference engine lm studio uses
I know it does use llama.cpp on the back end but LM studio seems to introduce a bunch of weird bugs somewhere in the pipeline
Oh my God I am so glad I caught the announcement before the stream so hype dreametech <xiaomi> are launching a new line of vacuums in just 14 hours live it's going to be an honour to attend and purchase as least 2 of each
and here i am thinking 10tps on qwen3.5-9b is very slow
Finally
I have iGPU. So I probably should use the vulkan backend right?
I can buy more slightly different vacuums
Why do you need so many vacuum man
What are you making
If it supports rocm then use that, if it's an old Intel integrated graphics or Vega GPU then use the CPU instead
Openvino?
Got it, thanks
Idk if they have an openvino backend
Openvino on amd sounds cursed ngl
Oh it's AMD
Like trying to run rocm on nVidia 
My CPU is AuthenticAMD Ryzen 7 5700U with Radeon Graphics 
I believe that uses Vega 7 for the igpu so I would stick with CPU inference
I wonder if there exists NotAuthenticAMD. Chay probably would have it 
Alright
Idk I think my highest openvino score in geek bench is on amd on cpu
It's like running games on Linux over WINE yield higher FPS moment
You would think intel would have very specific instruction for their CPU for openvino
in the old days cpu clones used to say "intel" or "amd" for compat purposes, GenuineIntel and AuthenticAMD is what they came up with to ident cpus actually made by them
Yeah my highest quantized openvino geek bench ai score is my 9900x3d which defeated my r9700 GPU using onnx
Kekw
Is rocm that bad?
Directml 

GPU won in fp32 and 16 by a lot tho
Windows should patent this technology. Useless loading bar 
The rest

They managed to crack anti gravity technology
this codex feature is cool even though this is a false detection, "draft out" also brings up plan mode the same way. ux masterclass by openai, never thought i'd say that
I can't imagine this is so massively slower than my 8745hs that you have to live this way
Over VM 
yeah cpu inference should be tolerable on that i think
This is a joke btw
No sane person would do that
Competing with my company

this has been there for 3 days now
Microsoft is redirecting their slop from windows to github it seems
Because who cares about github right. Layperson wouldn't even know about what the site is about
ghostty's moving off because of this
So is everyone gonna migrate to Codeberg eventually?
Probably not
https://fxtwitter.com/NyarchLinux/status/2049772942586740744 time to switch to nyarch 
We have benchmarked Nyarch against @CachyOS and Windows 11. Here are the results:
︀︀- On the gaming benchmark, Nyarch totally destroys both CachyOS and Windows
︀︀- Playing 4K anime, Nyarch ties with CachyOS
︀︀- On RAM usage we beat CachyOS but lose to Windows
Not unless if forgejo managed to finish the federation
codeberg is too unreliable and restrictive for a lot of people 
Top tier shitpost lmao

*Nyarch
*Looks inside
*NixCult
I only just looked closer at the charts
Lmao
it's not a cult 
nah
just like banana toast!
Now show me Fortnite FPS
But the server is literally named like that lmao
theyre speaking to me directly
every time i go look at nyarch it has become more self deprecating
LFG
lol
And honestly that's fine lol. People can also use their pc to have fun
Mother of steady hands
I had a bizzare notification this morning that led me to this article. While reading ot it reminded me of Evils 'Location' bug. I wonder if its related.
Holy mackaroni, deskflow can be used on windows lock screen
im so glad my parents are talking about ai replacing people's jobs /s
Was it prime
Tweeting
was what prime?
I got a notification this morning, tweet from the primagen about not seeing 5.4 of 5.5 talk about goblins lol
Super randomly
oh. no this was the notification
The goblins are invading
So it would seem
Question for you guys. Is 6-bit depth monitor on a laptop common?
Did you hear about the ai game IQ test for AI that was made public last month. The one that every human who tried it beat 100% while all the big AI like chatgpt and grok and Gemini were all between 0%-10%. Wouldn't it be interesting to see how Neuro and evil would do if they tried. I feel like the result could be surprising but unsure.
Not really anymore. Unless you go for ultra cheap
Then I'll go with ultra-cheap. This laptop is so ass lmao
Latitude 3420
Knowing Indonesia, they'll jack it up to 400-500
arc agi and humanity's last exam are this flavour of benchmark
the standards there are some shit like gtx 1650 and 3050
economy so screwed
Indonesians considered those high end already
i had one of these as my work laptop
It is a work laptop, yeah
You totally could lol
I think a vast swath of land in the mountain will cost you no more than 5000 usd
Like, a really fuckass huge land
They don't have 343 Guilty Spark 
that's stackchan
that'll be my destiny ghost//forerunner monitor
i wonder how quickly my workplace would notice my laptop's ram suddenly decreasing to 16g if i were to do some custom maintenance
Kek, checked in here, and that's the first thing I see LOL
7^3
how many rp
I think if this keeps up it will be 100 mill IDR
Currently it's still just around ~86 mill
6-7 give or take

I thought this benchmark was years old.
I saw it at least like 6-8 months ago.
(End of last June, actually)
So they just updated it?
i was really slow on this one
Doesn't this test require that the AI in question be either open sourced to them, or like that they need to register in a way that de-anonymises the test-taker?
where elvy?
400% on all of them and also somehow everyone died
they have to observe it in some fashion
i dont know off the top of my head exactly how
but im pretty sure they require alike the full CoT etc
actuaslly less than that but they "can" require it
New benchmark gets announced
Model makers start overfitting for that new benchmark too
Nearly 100% gets reached on that benchmark
Repeat
genius inference there
Very "overfitting to the new benchmarks when they come out" shaped graphs
Eventually the benchmark is just to become human
And the AI overfit to the point it is not human
also the same shape of relatively linear progress over time
They're 100% faking the "linear progress" trend
HLE
The difficulty of making anything work scales exponentially, not linearly
According to this, they don't seem to have any sort of actual AGI testing methods, as an AGI is extraordinarily unlikely to be an LLM.
So a prompt-less AI without a curated dataset (learn on the go) would likely not be testable.
there's a reason they hedge these things all over the sites
the whole point is that the test should not be hard
you're generally supposed to be able to pull a meh human at random and have them do pretty well
it doesnt indicate anything other than how they do on that subset of stupid things
LLM benchmarks aren't actually for measuring performance, they're for farming investors more efficiently
i was merely mentioning to other things to something else that got mentioned
i have no stake in the benchmarks being good lol
you dont have to prove it to me
the truth they don't want you to know
Is this reply towards me?
It doesn't sound like it should be a follow-up to my comment.
My comment was saying an actual AGI could not be tested, and could not win the prize.
As far as their methods appear to indicate
And is also extremely obvious

investors might have room temperature iq idk

the thing you posted i didnt reply to but it's just them hedging sayiing they cant actually measure agi yea
I think that's a little optimistic, but I don't wanna be too mean.
They're already quite dumb and wasting most of their cash. They don't need a full medical diagnosis, too
Ah, ok
Me when every model since 2024
All the LLMs are benchmaxxed
can they get benched instead?
So it turns out I can run Gemma 4 E4B 8-bit quant. However only for chat mode
When you realise... 
https://www.benton.org/headlines/ai-can-cost-more-human-workers-now
And find out they also... 
So, the most logical decision is obviously to... 
https://www.theverge.com/ai-artificial-intelligence/920378/oracle-openai-datacenter-buildout
We do be livin' in one HELLUVA Society™, fams, no cap, fr 
Lemme whip out the performance mode on power profiles daemon
My face when the ai bubble pops and, following the historical trend of how after the dotcom bubble popped we gave up on the internet forever and never expanded it beyond colleges ever again, ai disappears forever
She's going alright
What other mode is there
AW
Agent and autocomplete?
what the red bar of death
Super high tech CIA redaction technique
My face when like 15,000 companies go bankrupt, and I get to buy mega ass loads of RAM for hay pennies on the paperclip
No
They all use eccslop ram
It wouldn’t work in ur pc
a drop in demand should still lower strain on ram production and result in price drops
"Strain on RAM production" lol. Lmao even.
also it's not just ram
it's gpus too that spiked because of the bubble
@fast pagoda
Who said it's for a PC?
Who said it's not for my own AI cluster?
Who said I believe AI would disappear, instead of just NOT being used as corpo slaves?
yes that's how supply and demand work
And CPUs... And SSDs... And even fucken HDDs...
cpus are like fine
storage idk
Has everyone truly gaslit themselves into believing the whole "AI is the reason for high RAM prices" and actually think that they will go down?
everytime i see discussion about "oh my god <insert model name here> hit <insert percentage> score in <insert benchmark>"
fun fact all those parts aren't even used
all companies do is "sign off contract" of "reserve me this percentage of your total supply"
then supply gets halted
then those said companies back down
now suppliers/manufacturers realized they can charge whatever tf they wanted
absurd
then ai bros cope
"nooo ai demands is real they really need all those hardwares"
uhm yea maybe the SSD i got for the server was a bit too big 
Btw, this is another vim exit meme but how do I quit llama.cpp-cli? 
Ohh ctrl+c multiple time does the trick
But why multiple times
surely you are qualified to tell us the real reason and aren't just a tinfoil hatter
Storage DEFINITELY bad
CPUs are starting to take a hit, bc new announcements of investments nore heavily in Xeon and Epyc, instead of regular Intel or Ryzen processors
Not to mention that increasing production would cost a fraction of what they're making from the exhorbitant prices
I am very aware of this.
I mean I know storage is expensive but idk how it was in the past
It's almost like there is no incentive to reduce the "shortage" when RAM manufacturers make more money from big corpo than selling RAM widely like they used to
In the past, 2x 18 TB WD Red Pros was $700
Today, ONE 18 TB WD Red Pro is almost $600
The shortage is definitely real and not artificial trust
(At least from WD themselves. The aftermarket pricing is different, but still worse)
Maybe not the price part but it is true that production (not of ram, of the components needed for the ram) has expanded basically zero since the bubble started.
It is more beneficial for RAM manufacturers to keep the "shortage" going because it makes them more money
which didn't stop them from artificially inflating it in the past?
and yet it was more affordable
you are just tinfoil hatting
However the actual issue is a mix of 25% geopolitics and 85% that these factories require billions of dollars and several years, sometimes close to a decade, to build and spin up from scratch
So it’d be a gigantic risk to build more then finish when there is actually lower demand overall
uh, the issue isn't even the supply imma ngl
nah bro, jack1702 knows better
increasinf production is instant and basically free
we have enough ram
just that these companies are holding their supplies
in hope that ai companies would pay for far more for the dram chips
if it's not purchasable it doesn't count towards market supply
supplies and production itself is more than enough.
well yeah, it was enough for years
yeah, they're just artificially limiting supplies at this point
until bubble
The prices arent going to go down all that much unless serious political barriers are pushed past. Current US economic and political climate is driving oil price inflation whoch has spilled over effects onto everything.
and its gonna make snackies more expensive
there are shit tons of dram wafers sitting in some warehouses rn
unused
sam altman may you rot in hell
you fucking moron
It can be both
no, dram wafers just sitting unused uncut
because open ai back off outta their 40% dram supply
raw uncut sticks of d..ram
Frog
now these companies are coping hard still wanting to hold all these wafers

who cut it in 2
No you dont get it hes playing 5d chess. He's driving up the price to accelerate the crash and then hes going to but them up for super cheap after the crash.
See this is exactly the cop out manufacturers want

By pushing all the blame to AI corpos for hogging the supply, they keep heat off themselves so they don't have to take any responsibility
what's he gonna do with all the cheap ram? build a house?
yeah

its both the ai industry and the manufacturers fault.
not just the ai industry.
greedy bastard
I have no clue where these frogs keep coming from, they just appear in my backyard randomly
elvy invasion
dont think ai hype will fall anytime soon
the frog spawner
https://x.com/moonlake/status/2049549673900503215
buildings instantly collapse
Introducing Moonlake's 3D Agent.
Our agent acts like a technical artist that can build and reconstruct articulated assets and large-scale editable scenes with hundreds of objects from a single image and can improve its generations continuously.
Learn more in the thread below.
by soon i mean like few years
they keep raising token prices lately
which means investor money has been drying up
and LLMs are so highly inefficient they won't pay for themselves
i think more it just means that they think they've captured who/what they will and can begin squeezing
price increass will always result in losing part of the customer base
and the way they do it they also suffer reputational damage to speed that up
these fat fuckers of corpa kept throwing cash and funds at each others
yea, but not a lot of people understand that part and fall for ai trap, considering a lot of new gen programmers are very accustomed to usage of it for most purposes too
the vaasstt majority of use and spend are from under 30% of the users most likely
as is the case with almost everything
they wont leave
these new gen programmers are always stuck onto js projects lmao
most common ones are
SaaS
Grimbo
oh yea
that why js is now backend front end, inflight, injected in ur veins with air bubbles, somehow its 3d game engines
then you have vibe coded sites
imagine writing a funny scripting language for netscape navigator in 95
and then seeing it now
Also a thing that people are overlooking for the AI bubble. Look right now at what these companies are doing. OpenAI, xAI, Anthropic, and Google are getting themsleves deeply involved in the US military industrial complex. AI companies are becoming critical for decision making and operational secuirty. So when these companies are starting to go bust from the big bubble popping. What will happen? They are going to get MASSIVE bail outs. Im talking 2008 recession level bailouts because they have become so entwined with the military that the US government wont let them fail.

Yeah thats the game plan.
yay we're giving statistical and probabilistic models control on weaponaries.
ok time to debug the server 
why are you hating dual gpu with one being mobo gpu?
Thats what i want to do in order to do dual gpu lossless scaling on the cheap
its not igpu, jsut like one built intot he mobo
Nah
we crossed that bridge when claude was involved in simultaneously blasting 1000 targets or whatever the hell they said it did within maven (palantir)
Bubble wouldn’t affect Google enough to shut down their ai operations
They use their own tpus which they buy from themselves
they buy from intel now
suddenly
Oh yeah no Google wont but they will try and get the AI bailout bux
they make it inhouse, which is arguably way better than other AI companies and their reliance on other fabless companies.
Exactly
uh oh
my calorie counting golden retriever AI
Google played the extreme long game with ai boom
Damn you samsung where is my charger?
Weren't they the one involved in attention is all you need?
Yes
I mean they essentially are the paper
I mean, they have been playing this game since 2014.
yes deepmind is the one actually making anything useful anyways
they put out a ton of non-llm actually useful research and actionable product
of course they are limited by corporate now.
i said this recently here but my brother is a meteorologist for noaa and weathernext2 is a major part of their forecasting at this point alongside the traditional ensembles
they have alot of things that nobody cares about because theyre "boring"
boring is relative.
that's the quotes job to convey
your brother sounds cool.
deepmind is the only cool one.
if you wanna see enthusiasm get to know a meteorologist
the rest are trash in the scientific community.
theyre all gigantic fucking nerds for weather
Google is the only American lab putting up a fight against China in open source
"fight against china"
in very limited scope
why is this so cringe sounding
because it shouldnt be a fight but here we are
Fuck IBM
D:
?
ibm has some cools tuff tho
And fuck Dead Hat by proxy
Because I said it
And only because you quoted it out of context
no its really generally cringe
If you had “putting up a fight against China” it’d be better
"arms race between us and china"
"china is nanoseconds behind us"
bla bla
financebros and techbros talk
granite is the single most transparent and responsibly maintained llm family
for some reason
apache 2.0, entire process is completely open
including all data
"only because you quoted it"
why is this so cringe sounding
This guy yawned constantly during history class
Open source ai
Next you’ll tell me space race wasn’t a competition and they both just wanted to advance science
eh, competition is good.
Smh
duopoly > monopoly in a narrow sense
this one is cringe
space race had/has tangible benefits
^
SSNbro uselessmaxxed opinionslop
that actually aren't just "CAPITALISM FASTER"
Thinkpads my beloved
true, the entire space race was funded in less money than an order of magnitude less than the entire AI industry entirely.
Tho,
yep
lemme fact check that
space race actually gets you cool breakthrough in applicable astrophysics in engineering
this one gets you what
"mOrE mAtRiX mUl"

AI does have massive benefits the issue is that corporatism and active attempts of mass regulatory capture weren't behind putting a man on the moon.
it was so peak even the soviets went "LETS FUCKING GOOOO"
reboot
Good hob. How is the server coming along btw.
we should've funded nasa more
HELPPpppp
-# ppppp
instead we get
death
military having 40x more fundings
decent. it hates dual gpu with one being mobo tho. and its loud
then ai bs gets 2.5T
jet engine loud
i hear space chat
Bruh who tf is we
and we have slop all over instead
You live in Australia
Did you tape over the intrusion detection sensor?
But e got these cuties though
all this shit for image/video slop and deleted codebases 
these cuties are the only good one coming out
real

genuinely it can't be hard to make an among us plugin for creating a programmatic bot API.
need to go get my passport now, server debugging go later
believe it or not? 3 billion doollaars
maybe if i sacrifice a ram stick, I can do it.
There is a RL Among US i am pretty sure but Neuro cant play since her API is basiclly Zork
i cant imagine being sundar out here directing my fucking search engine company to outspent 13 years of apollo program on blobs of matmul
that must be a weird feeling
It’s not
It’s the single thing Vedal has open sourced
So it’s actually instant
nvm it just phased through the wall, time to make a path finder.
agentic af
I've unfortunately burdened myself to make the API from scratch, its a weird bug with me.
bold step for botkind
we could've thrown all those trillions to research institutions that work with subatomic particles aswell
https://blog.google/innovation-and-ai/products/dolphingemma/
semi cool use for llm
honestly as someone who's been into the astrophysics rabbithole. antimatter is a big deal.
theres cool uses; very few of those uses come from the chatbots
"So long and thanks for all the fish."
much better frontiers out there
the chemical work is amazing; we can sequence novel organisms in 24 hours; we can learn how to speak with elephants
since its the only known way to store energy at the mass-energy conversion rate of exactly e=mc²
nuclear fission is about 0.07%
fusion is about 0.1%
please tell me someone got the reference
search algos useful and cool
so i decided to train a bit
gas are way below
and i got a sub one
ye s
hello hhgttg
there are A LOT of other uses of neural networks like these
All the cool stuff is being done behind scenes or by hobbyists. Consumer facing chat bots just take up the news.
but instead we get clanker llms
hitchhiker's is unfortunately historical nonfiction
there are a lot of uses for diffusion models especially on the robotics field
instead we get...

me when i love diffusion models and if you say that you get tarred and feathered
which i mean fair giventhe exposure
this makes me sad.
we also have alphafold
yes, Actually fr.
most of it is gatekept.
theres lots of work done with the opensource models
you dont tend to hear about that tho
and Nvidia do some decent work; so do ibm and indie orgs
splats are kindda ai

this is why i shill granite
my son
Level1Tech had a good video on it. There was a company that is making an AI system to help scientists make synthetic medications because its such a delicate process that taking it out minutes to early or to late completely screws up everything.
biomolecular sciences be like that sometimes
yeah if only it didnt suck balls......... (/jk its not that bad)










