#programming
1 messages · Page 250 of 1
how do i optimise this token-wise
it feels like there should be a solution.. but i dont see it
my fuking cloudflare worker project i JUST got my environment reconfigured for, got delayed on stuff i said i was gonna fix because of IRL weddings and such, and then my cpu exploded so i've been absent a bit from that (i am terrified to look at the thread and explain myself)
I see a
transposition_info.int3==X&&transposition_info.int1
but otherwise I got nothing
https://blog.quanticdream.com/detroit-a-vulkan-in-the-engine/
oooh
i love blog posts about engine stuff
found ur token problem right here officer (kek)
nono bwaa} appears only 3 times
if i macros it, i will spend 5 tokens for macros, and then use 2 more macros by its name
so i will get 7 tokens total and now its 6
no bwa macro today
what if u alias transposition_info to like ti (idk if that will save tokens or not w/ toknt because ive been neglecting chess bot even harder than everything else)
i would think the compiler would alias this automagically at build time and it wouldnt actually save any tokens but idk if it would for just the count toknt does
and i suppose you could yank the logic for checking your transposition table entries into its own function but depending on the magnitude of this logic being used in the thing it may or may not save anything
entire transposition searcher
this final iff is bad
but idk what else to do
what happens if i do this 
i think the runtime will become angy mostly
surely nothing is being discarded
You do in fact not need to declare main with parameters indeed
Huh why am I pink now
you are now patrick star
pinku
ohh
i could
round all parenthesis to 2 or more

so i have places where i have (( or )) and its supposed to be like that
but i can replace normal ( with (( and close it on the other size
waits but i will gain nothing then
i can replace (( or )) only
did you not know
you dont have to accept args
does it see (( as 1 token or 2 (id assume 2 but) cuz i think you could save one(1) token at the last iff maybe

(( is two tokens but i dont have many of them
i have a few duplicates after restoring things oh my
macros for 2 token replacement is only viable for 6+ occurences i think
macros itself is 5
you replace 5x2 tokens with just 5
it becomes 0 gain
yeah it needs reuse for sure i just cant see the rest of the code obv so if it was common i figured i'd mention
???
yeah idk my role got replaced with the osu one out of nowhere

If this server has external applications support
o7
it doesnt
it doesn't
guten tag and guten nacht (1:31am)
its not 1.31am in germany rn?
shouldn't be
ye it isnt
(20-10-2025 time : 12:03 am ) or 00:03
so sad that owc has to be so late
clearly, it's after the ball
we love when 2 people in 2 different time-zones arguing with each-other about what time it actually is
may time.windows.com be our judge
SLEEPING TIME : I WILL STILL WAKE UP 5 AM
tale as old as time (and it's exactly as old as I perceive it to be, you unfortunately are incorrect in your assumption)


photo evidence says otherwise
is there some command for c++ to stop declaring variable types
like
"guess yourself man"
My phone wake since
402 hours
auto but that doesn't help you
yeah i meant without auto at all
then 
what happens if i actually dont free the board 
surely shiro can handle a little of memory leak 
Bye 👋
auto& we out heyuh
enjoy a well earned rest after 402 hours
wasn't there a new version that doesn't allocate a new array every time
or did that not make its way into the header
auto& avoids alloc/copy
dllexport means dont need to free?
no that's just for making it available on windows
structured bindings that's the name
you should be able to allocate an array with enough moves and then just reuse that
218 is technically the absolute max number of legal moves you can have
i am getting new list of moves many many times
its not being reused most of the time
decltype returns the type of a templated function which automatically makes the return type the exact same type as the expression in the function
you can do decltype(auto) lol
but delctype(x) is 4 tokens
huh
i should really get toknt set back up so i can test
all my type declarations are 1 token except structures
step 1 is to install python
i love how they added a new way to declare functions specifically for this yet nobody actually uses it
not even considering JS anymore?
maybe i need to figure out a better way of getting -1 and 1 out of 0 and 1
idk how often you'd be using this but you could do like a using dt = decltype(auto); and then just use dt everywhere if it's super prevalent ay? i think you'd need to use dt 3 times to overcome what i am assuming to be 8 tokens in the declaration
im using this formula now (1-x*2) where x is bool
maybe i dont get what it does
int int1
dt int1
whats the point
i have bad news, this doesn't work btw
see that's what i was hesitating over
use js if you want, i just dont like js
it will get expanded and be the same thing when processed
i have been so afk that i havent even had a chance to actually do my own chess entry so idk what actually is counting as a token saving 
all you're doing is hurt compile times 

there's not really a better solution than doing * 2 - 1
this is how we do it in shaders even for our normalmaps
oh, you inverted them
aslo works but then you're getting 1 and -1 out of 0 and 1
not -1 and 1
oh yeah i can do x*2-1
maybe it saves ! somewhere let me seee
wait no it doesnt
wait no it does
am i stupid
cant i replace this with just %2
hmm wait i cant
even tho the alias thing doesnt save tokens, the decltype thing could if you have weird ass types that are super long that you need to reference such as some goofy type shit like
std::vector<std::pair<int,std::string>>::iterator foo() {
return bwaa.begin();
}
gets turned into
decltype(auto) foo() {
return bwaa.begin();
}
saves a lot in that particular instance but idk what you're dealing with, this is why i brought up decltype(auto) though
4/4%2 is 1
4%2 is 0
can accomplish these things with auto too depending on the case tho
which means decltype becomes a waste in those
We're getting a new NeuroSynth peak soon, I just got an insane transcript
only weird types i have in my bot
when life throws you lemons, say "bro wtf stop throwing lemons at me" and proceed to throw them back

mhm my head
evil neuro cover of styx's too much time on my hands would be dank (i am definitely currently listening to this song so i am bias) but i presume that's a new regular nwero not eliv unfortunately
It got both

Not a model, a song
ah
evil neuro
evil neuro 
1049 tokens 
Finally figured out what was causing the weird issue where my PC failed to POST when rebooting from windows
It was fast boot
when life throws lemons you tell your engineers to build a catapult that makes the lemons compustible and launches them flaming hot right back at life
as my good friend cave johnson says
"my engineers" i wish i had those
1034 
1035
1032 
1033
your linter is just broken
stupid sort needs [0] 😠

1029 

🥹 and hes still got it
im so afraid to break something while doing this
im doing backups way too rarely
1027
surely i'll find 3 tokens 
alright its done, i just had to remove couple of outdated macros which became useless
1023 

you can do it if you want it 
There's a lot of work to be done to make it even remotely fast enough
i don’t even have an invalid one
i=0
out = ['!']*((9+8-7)*(6+5-4-3-2)+1)
outs = ''
while i < (9+8-7)*(6+5-4-3-2*1):
if i<1:
out[i]=chr(72)
elif i<2 or i/3==9%5:
out[i]=chr(90+(17%10))
elif i<4:
out[i]=chr(pow(10,2)+pow(4,2)-pow(2,2))
elif i==pow(i-2,2) or i==ord(out[8])-ord(out[11])-1:
out[i]=chr(int(0.5*(pow(2,2)+pow(3,2)-pow(4,2)+pow(5,2))*(pow(2,2)+pow(3,2)-pow(4,2)+pow(5,2))/2))
elif i==5:
out[i]=chr(int(pow(8,2)*1/2))
out[3*i-1]=out[i]
elif i%2==0 and (1/i)>.16:
out[i]=chr((1+2+3)*(5+i))
elif i==1+2+3-4+5:
out[i]=chr(pow(2,2)*pow((i-2),2)+(i-2))
out[(1+1)*(i+1)+1]=out[i]
elif i==ord(out[5])/4 or ord(out[5])/4==i-pow(ord(out[11]),(1/2)):
if i==ord(out[5])/4:
out[i]=chr(int(pow((6*i/4),2)-(i+2)*3))
else:
i=int(i-pow(ord(out[11]),1/2))
out[i+10]=chr(int(pow((6*i/4),2)-(i+2)*3))
i=int(i+pow(ord(out[11]),1/2))
elif i==ord(out[0])/(i-1):
out[i]=chr(int(pow(i,(1/2))+((pow(i,(3))-pow((i+i),(2))-(i*pow(ord(out[4]),(1/2))/(i))*(i+i)+((i)/i))/(2))+i))
elif i==((pow(2,2)+pow(3,2)-pow(4,2)+pow(5,2))/(2))-((2)/(2)):
out[i]=chr(ord(out[8])-i)
out[int((ord(out[6])/pow(ord(out[4]),.5))+i)]=out[i]
elif i==ord(out[9])-ord(out[7]):
out[i]=chr(pow((1+2),(3))+4*5*6-7*8+9)
elif i==ord(out[3])-ord(out[2-1]):
out[i]=chr(int((-i+((1337)/(420))*60+9)-ord(out[10])+2))
elif i>18:
out[i]=chr(-(ord(out[7])-ord(out[16]))+ord(out[2]))
i+=1
i=0
while i < 3+1*4+1+5+9-2+6-5:
outs=outs+out[i]
i+=1
print(outs)
lil puzzle for you guys- can you figure out what this outputs without running it?
-# i spent far too long on this meme thing
That is NOT token efficient 
and the 2nd option seems better for my time
How are we going to figure it out
hint: the last while loop is just a formatting thing
21 exclamation marks?
oh i messed it up 1 sec
i made a small change before sending
and messed it up
frick
second line should be
out = ['!']*((9+8-7)*(6+5-4-3-2)+1)
What does it do
fixed it in the main block too
Is it like python
['!']*21 = ['!','!','!','!','!','!','!','!','!','!','!','!','!','!','!','!','!','!','!','!','!']
So to do it i have to know all ascii chars
just dont run it, you can google an ascii table
I'll pass for now im in bed 
lol
power
not related but
imagine not having an ascii table saved on your phone tho... SMH my head
mhm?
Does it accept float power
methinks
Can i get sqrt with 0.5 power or so
yes
Cool
chr is int -> char
ord is vice versa
spoiler tag this for backreaders
but yes
poozle
oh i see
3D modeling industry is being held up by guys with weird obsessions
we love the fucking "patio umbrella blueprint"
Man, these are a lot of operations...
if only I had some type of machine that could perform all of these computations at once
Some type of... computer...
just wait until you hear about cad
you can become someone who gets paid to make models of your weird obsessions
i get paid to sit and watch neuro streams (and occasionally check off assignments and help ppl with cad) on fridays! (CAD lab assistant)
I keep convincing myself I'm gonna learn cad properly then forgetting to
surely this time
does blender count as cad?
cuz it is cad, but when i hear people use the term cad they're doing autodesk or smthng
fusion 360
One Frame Man

I can only think about how I use an umbrella which is to undo the latch and pull it down until the wind pressure moves it to the other latch
usually cad implies non destructive editing and parameters, among other things
blender is much more of an art tool
non destructive editing and parameters 

Similarly, the majority of 3D design tools that revolve around meshes/triangles are destructive or only allow very specific subsets of their functionality to be modeled non-destructively (e.g. Blender). A notable exception to this is many of the newer parametric design tools in CAD such as Grasshopper and Fusion360. In such tools, everything is defined in terms of dimensions and constraints — e.g. this angle is 30 degrees, these lines are parallel — and you can go back at any point to alter those previous dimensions, which then recalculates the entire 3D object.
i sent it in a few other servers im in's programming channels and ive just realized imma get so many HBD shiro screenshots now
also, line 36 should be out[i]=chr(int((-i+((1337)/(420))*60+9)-ord(out[10])+2))
i fixed it in the block but for those who have already copied
What the....
yes? 
i mean it is intentionally made as difficult to comprehend as possible
from hehe import *; aW5wdXRudW1iZXI = aW50(aW5wdXQ(VA+eQ+cA+ZQ+" "+YQ+" "+bg+dQ+bQ+Yg+ZQ+cg+":"+" ")); b3RoZXJtb2RpZmVy = Y21GdVpHOXQucmFuaW50(0x0, aW5wdXRudW1iZXI);aW5wdXRudW1iZXI -= b3RoZXJtb2RpZmVybW9kaWZlcjE = Y21GdVpHOXQucmFuaW50(0x1/MTA, 0x67/MTA);cHJpbnQ(KA+c3Ry(aW5wdXRudW1iZXI/bW9kaWZlcjE)+Kg+c3Ry(bW9kaWZlcjE)+KQ+Kw+c3Ry(b3RoZXJtb2RpZmVy));
oh wait its just weird variable names
i straight up spent almost 3h making the meme program above. abt 30m i almost stopped but got sunk cost fallacied
I'm back
Total uptime
Otg has failed an and I need ANOTHER OTG again BUT WHY KEYBOARD IS USELESS WITHOUT OTG IN PHONE
tf is hehe
python file where the functions are defined
Should have just done it in one file
well then it would be much more understandable
Are you sure about that?
yes
i mean more understandable in my current system of just changing all the fuctions to random text, there is ofc better ways to mask what a script does
Trust
I need 7000 cpu core 4.ghz units all same
In one chip
Shame as GPU is just nog independant enough sometimes
Let me ask a question?
heh heh
How much knowledge about cpu and quantum machine
Do you even have a pet for these discounts?
Alright one base64 in the terminal later now lets begin the actual fun
i know what a quantum computer is and have what i would call a shaky understanding of how they function, mainly cuz qubits r cool
basically, not at all outside of general awareness of their existence
aka useless
Alright, sanity check works, now lets begin the cursed part of this
I should make this setup slightly more reproducable actually
The last part is a pretty good explanation of quantum computers 
bwaa and awa have been defined, of course each in only one line
because not even the generation code is safe
You know how functions have a fixed point? where for some specific value f (x) = x
like let add(x) = x + x; let fixed = (add(0)) == 0;
there are fixed points in hashes
where a hash routes to it's own value
Iterate function application of hashing, the results of hashes path until they loop, or they will hit a fixed point and terminate.
so like hash(3) = sha1(sha1(sha1(seed)))
so you keep iterating through repeated function application until you get a loop or a fixed point
Alrighty done
then you draw the graph such that every value is represented only once
for all the values you have explored
from base64 import b64decode as bwaadow
exec(bwaadow("CmZyb20gcmFuZG9tIGltcG9ydCByYW5kaW50CgojIGNvaW5fZmxpcApkZWYgWTI5cGJsOW1iR2x3KCk6IHJldHVybiByYW5kaW50KDAsMSkKCiNDT05TVDEKUTA5T1UxUXggPSAiU2FtVmlNaWxlIgoKI3NwZWFrYQpkZWYgYzNCbFlXdGgoY2hhdHRlciwgZmxpcCk6IHByaW50KGYiIiJ7Y2hhdHRlcn0gZnsiaXMgZWVwaW5nIiBpZiBmbGlwIGVsc2UgImlzIG5vdCBlZXBpbmcifSIiIikKCg=="), globals(), (captured := {}))
for k,v in captured.items(): locals()[k] = v
ZG90aGU1MDUw = Y29pbl9mbGlw()
c3BlYWth(Q09OU1Qx, ZG90aGU1MDUw)
-# could have done base64 encoding not hardcoded but what is the fun in that?
sha1(seed) -> sha1(sha1(seed))
sha1(sha1(seed)) -> sha1(sha1(sha1(seed)))
and so on
Not a virus, just a 50/50 between print messages
coin_flip
def Y29pbl9mbGlw(): return randint(0,1)
#CONST1
Q09OU1Qx = "SamViMile"
#speaka
def c3BlYWth(chatter, flip): print(f"""{chatter} f{"is eeping" if flip else "is not eeping"}""")
eeping

Yep, simple but at least in one file
thats exactly what i was trying to avoid with my multifile system
Yeah it is easy to reverse, nothing that can be done there. If I were to send code in a channel, I'd at least want it to be reversible.
Not that it is the only method I could think of, just the one I felt like doing
by golly you're right
i am ironmaneinstein (who was decently uneasy about quantum spookiness for his part)
I'll do that another time, I don't want to spend all night on this tonight
i gave this to 3 claudes (opus/sonnet/haiku) to watch them argue about it like battledroids because it's funny to me and i do not value my own time.
it pissed off 4.1 opus because it couldn't figure out i=12 or i=13 and then started roasting sonnet which DID get it through handwaving - it figured out the answer without needing to do the full step-through process due to llm spooky big number fuzzy intuition from training data or luck or some combo w/e
it got so bothered though
LOL
Did you give it the one with correct line 36 or the broken one? Either way you should try giving it the opposite and see its response
^
Before it was out[i]=chr(int((60+9*((1337)/(420))*60+9)-i-ord(out[10])+2))
the one it got was: out[i]=chr(int((-i+((1337)/(420))*60+9)-ord(out[10])+2))
Ok u had the right one
yeah it ran fine in my ide i checked to be sure because it made me doubt the first time wasnt a fluke or something, it was NOT figuring 12 or 13 out
Since triple parens are illegal I originally moved the 60+9 to the left side forgetting that would make it be 9x instead of 60x
Oof lol
I spent way too long on that program
both opus 4.1 and haiku 4.5 got tripped up, on different spots
sonnet just YOLO'd its way through and figured out the result through gigabrain big numbers high parameters woo woo and then pretended to explain it
Also when I initially sent it it got automodded (cuz the triple parens) but also I didn’t have the 20 and 21 obfuscated so Shiro figured it out 
lol
Try giving it the version with wrong line 36 and ask it to find the error
New chat prob cuz otherwise it’d just look at history
haiku got thru 12/13 and then got stuck thinking i=15 evaluated to an out of bounds array position of 21
boutta do this now for science

Oh everything after i=14 got funky when programming
🏛️
classic

loooool
opus 4.1 thinks the true intention was a much more powerful, heartfelt message indeed
NAHHHHHH
Happy Birthday Ohio!
I still don’t know why it returns ƪ ngl
sonnet thinks it's "Happy Birthday yo!"
they fall apart and then nothing makes sense
Ngl it’s impressive we’ve made computers that are terrible at computing
all 3 of them actually got stopped on i=15 this time before the error at line 36 even happens lmao
haiku went back to the
'le 21stdhth array slot' copium
Here’s the key btw
The *1 is confusing it so much lol
I think it doesn’t know indices start at 0
i posed this question to glm 4.6 btw
about 7-8 mins ago
it is still thinking
on the right track tho it looks like
the environmental impact from posing this cipher puzzle to computers that are too dumb to compute
is astounding
it's so over
Holy
glm4.5v stopped at 15 and declared the rest to be bullshit and the answer is just "happy birthday!"
Give to the highest ChatGPT model you have access to, i wanna see if i just discovered a new AI poison
Ok imma disappear for a min or two cuz i need to restart my iPad cuz it doesn’t want to charge
you've currently ruined z.ai's entire yearly compute budget i'm pretty sure
4.6 has been thinking about it for 13 minutes
-1 lake :LULE:
extinction of koalas is unfortunately now guaranteed
Have you shot it at ChatGPT yet? I don’t even remember what the newest model is nowadays
90,500 token response w/ openai's tokenizer so off by likely some amount
Holy
yeah i have he's thonkin
I just hope these data centers are running on renewable energy
Otherwise this is going to spike my carbon footprint
they are unironically still 
What’s the time at?
i gave one gpt5 w/pro reasoning with the same prompt as the caludes/glm got with the bug
one gpt5 w/ pro reasoning a slightly "enhanced" prompt with a little more guidance
and then a gpt4.5 got the aforementioned enhanced oone as well so it's 3 of them rn grinding
Holy
Ngl we’ve prob used like an average house worth of electricity with these prompts by now
How long has 4.6 been running now?
14 minutes
is chatgpt just guessing next letters
i mean when you dont say prompt that are "specific" it will start hallucation
this one actually is giving the appearance of having mostly figured it out
can neuro sama ai used in roblox studio gng
it doesnt know shiro though so it guessed that the solution was meant to by "Happy Birthday[,] Hiro!" it thought the S was meant for a comma after birthday
I mean, closest we’ve got so far
they have all recommended as a fix to ditch all this crazy math and just set the values as constants kek
i meant "can neuro sama control roblox character"
she's generally not fast enough to do realtime games on her own
stuff like minecraft there's a model helping her that's handling the movement part
she kinda directs her char like a meat titan
i wonder why that equation specifically is so difficult for them to handle
yeah ig cus realtime game has action+event that happened every fps
4.5 is just killing it, ultra insight into the pure raw math has arrived:
clearly it was meant to be
OOF
Happ< Birth!!! !!i!!((
yeah that's probably it
one of them is still workin on it and it's clear theyve reached desperate measures as they appear to be googling basic equations
NAHHHHH
how does that related to python
this one said fuck your puzzle it's easy
oh okay make sense
bruh really said easy flex
WAIT HOLY
the code theyre trying to figure out is python
5 Pro is just too powerful
the models are attempting to manually calculate this step by step without being allowed to just run it in a terminal: #programming message
the sad thing about 5 pro is that the reasoning trace is hidden completely
i can't see how it got there
Dang…
this is all i got:
easy flex
Output: Happy Birthday Shiro!
how I got there. no runtime. just eyeballing the math
array size = (10)*(2)+1 = 21 slots → all '!' to start, so last char stays !
loop runs i = 0..19, fills spots via those goofy conditions
quick fill order:
i=0 → chr(72) → H
i=1 or i/3==4 → a (also hits i=12)
i=2,3 → p, p
i=4 with that 22²/4 thing → y
i=5 → space, and mirrors to i=14 → space
i=6 → (1+2+3)*(5+6)=66 → B
i=7 → i and copy to i=17 → i
i=8 via the 6i/4 bit → r
i=9 the big sqrt/power mess simplifies to 116 → t
i=10 uses ord(out[8]) - i = 114 - 10 → h, also sets i=16 → h
i=11 from ord(out[9]) - ord(out[7]) path → 100 → d
i=12 (from that i/3==4) → a
i=13 from ord(out[8]) - ord(out[11]) - 1 → y
i=15 from the 1337/420 gag → 83 → S
i=18 hits the sqrt branch (temporary i=8 calc) → r
i=19 final formula → 111 → o
concat all 0..20:
H a p p y _ B i r t h d a y _ S h i r o !
done.
i deserve a neurochatting for that one
That’s actually basically the process i used when making it LOL
making them talk like this makes their responses align better with a "person" to me
seems to also follow w/ logic (when they actually have sound logic by accident)
Lol
try letting chatgpt solve codeforces problem
H. Beautiful Problem
time limit per test2 seconds
memory limit per test256 megabytes
For an array a
of length n
and three integers x
, l
, and r
(1≤l≤r≤n
), define:
f(a,x,l,r)={0,1,ifif(x−minrj=l(aj))⋅(x−maxrj=l(aj)))<0(x−minrj=l(aj))⋅(x−maxrj=l(aj)))≥0
You are given an array a
of length n
(1≤ai≤n
), and m
intervals [li,ri]
(1≤li≤ri≤n
).
For each x=1,2,…,n
, answer the following question independently:
Does there exist a rearrangement a′
of a
, such that for all 1≤i≤m
, f(a′,x,li,ri)=1
?
Input
The first line contains a single integer t
(1≤t≤2⋅104
) — the number of test cases. Description of each testcase follows.
The first line contains two integers n
and m
(2≤n≤2000
, 1≤m≤2000
).
The next line contains n
space-separated integers a1,a2,⋯,an
(1≤ai≤n
).
The next m
lines each contain two space-separated integers li,ri
(1≤li≤ri≤n
), each denoting an interval.
It is guaranteed that the sum of n2
and the sum of m2
over all test cases does not exceed 4⋅106
, respectively.
Output
For each test case, output a binary string s
. For x=1,2,…,n
, sx=1
only if there exists a rearrangement a′
of a
, such that for all 1≤i≤m
, f(a′,x,li,ri)=1
. Otherwise, sx=0
.
Example
InputCopy
4
4 2
1 1 3 4
1 2
2 4
3 2
1 1 3
1 2
2 3
3 1
1 1 1
1 3
9 3
4 5 9 1 1 1 2 2 3
1 6
3 7
7 9
OutputCopy
1011
101
111
100100001
Note
In the first test case,
For x=1
, one valid rearrangement is a′=[1,1,3,4]
.
For x=2
, there is no rearrangement a′
of a
satisfying f(a′,2,1,2)=f(a′,2,2,4)=1
.
For x=3
, the only valid rearrangement is a′=[4,3,1,1]
.
For x=4
, one valid rearrangement is a′=[1,1,3,4]
.
i thought codeforces was mostly defeated in the llm space at this point
yk only 38 ppl out of this world solve this
Wait wait wait hold up i just realized did you give it a screenshot not the text itself
that's not a screenshot
Cuz the box makes it look like it
it just highlights it with a white background for some reason
will chatgpt get past this question
yeah idk why it does that lmao
Oh it’s just formatted to look like text message style
yeah idk but it's def text
ill ask the oracle
yeah welcome to optimized prompting 
i cant test if it fail or pass
soemthing of a prompt engineer myself or whatever the current most money generating thing to call it is
there alot of testcases in codeforces
also the code must not exceeded time limit
and memory limits
that very hard
time limit per test2 seconds
memory limit per test256 megabytes
ok so:
can't actually run to see pass/fail
2min runtime limit overall
2second runtime limit per individual test case
256mb memory constraint
is that everything
yeah
you must not have any testcases "fail", codeforces it's a comptitveve programming platform for nerd
the feeling is like, you're coding a working projects, and someone walk in and ask "can you inverse binary tree"
first time hearing that, just humbled you, cus you dont know what to do, there a difference between programmer who does projects and programmer who solves problems
thoughts on binary space partitioning trees?
idk about that
In computer science, binary space partitioning (BSP) is a method for space partitioning which recursively subdivides a Euclidean space into two convex sets by using hyperplanes as partitions. This process of subdividing gives rise to a representation of objects within the space in the form of a tree data structure known as a BSP tree.
Binary spa...
i do not know what does this mean at all
took me half a week to grasp it, don't worry
i don't exactly remember how it works
but it's like really complicated
and also quite cool
you cant memorizes everything, so wikipeda exist
basically same as math formula, there alot on google
with specfic uses
for reference I spent two weeks trying to write doom for a graphing calculator
with nothing but the ability to fill polygons and draw pixels
and a few graphics buffers
I got as far as placing walls which would render in 3d
blud that was googling all math came back with an absolutely groundbreaking answer
and texture mapping
you gonna make your game engine
except some walls got inverted and I still have no fucking clue why
it's a nightmare when it's all just numbers and coord and positions, vec3 placing a block manaully
Wait W A T
if i ever meet sam altman in person he's going to be shocked to find a have this pdf printed out and ready for him to sign immediately
oops name got randomized
AGI is here
-9???
WHERE DID IT GET -9 ANYWAY
this is just 9 pages of pure chaos and has almost nothing to do with the question posed LOOOOLLLL
Ikr
22 minutes straight of 
and just drops the stinkiest turd of an answer
technology is amazing
Midwestern AI
it gets that from me unironically
it’s impressive that humanity has managed to make computers that suck at computing
we'll see, brochacho is still PONDERING that codeforces question rn
I wish this included the prompt
3 = 3.000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000213059871049587934875082734765029347580263408756252345234598273450982734807562034875
apparently
floating point errors
i should have realized this at the time but he decided to go into research mode for this report & hallucination fiasco
https://chatgpt.com/share/68f5bb7b-164c-8006-a2d7-c2221e2e2e9f here's the chat
i had to say this because this lazy mf just stopped after this message "ill drop the output"
lol
i have to share the output too i guess
of the "tool" which is just the PDF thing
Ah
brother how are you not maybe taking a step back at this point and saying huh that seems uh wrong (i know it can't and doesnt think so that's why but it's so funny to look at to me)
+1 birthday wish
is that a continuation of the conversation?
Yea
dang it seems like my style prompts and memories dont carry over which makes sense for several reasons but damn he's back to talking like an asshole
I’m no prompt engineer so idk how to effectively get it to “reflect” or anything but that’s what I’d do if I could
ill see what he has to say for himself
well he's still goin and has opted to hid e the detailed thinking about halfway through so it must be devastating
btw i restarted him
he's chuggin
it appears that the one that evaluates in the thousands, it has opted to incrementally increase the fraction over and over by one(1) each time so it's gonna be a while
off by a galaxy

and now I see exactly why my earlier “hello/world” fantasy was dead wrong at basically every branch. poor fellow is taking it hard
At least it admitted it was wrong
Content error
Lemme try pasting the link into browser instead of just clicking it and getting sent to the app
Yep browser paste worked
https://chatgpt.com/share/68f5bb7b-164c-8006-a2d7-c2221e2e2e9f the full convo idk if the link updated
i think there's a solid 30 seconds where it wont load
We have successfully done too much for the ChatGPT app
personally i enjoy that it shortened the rubric
step
what I did
correct
why wrong
principle
adjustment
i have done this in the past if i messed something up and had to write things about it
shorten the rubric or whatever it is explaining what i did wrong as much as i can lmao
Read that as extra wife 
can't forget about the 2nd wife
Ngl great strategy
andy is not finding this very exciting whatsoever
ill take his reaction as my cue to actually do something slightly productive so i'll bbl
Fair fair. I’m literally in bed at this point so gnight and good luck with being productive
Aww kitty!!!!
Also if you have time could you get that reflection bit as a PDF? I really liked that bit
He's like me fr fr
this is eerily similar to opening my phone every time i step away from my desk at work for over 30 seconds
yea i gotta wait for the depressionbot9000 to put it together though since it wasnt giving a pdf export option so give it like 45 minutes to 5 business days
ive personally set back at least one yacht for altman this evening
nice
Depressionbot9000… dang man that name makes me feel bad for him
at least it doesnt start saying it's about to go jump off a building like gemini does if it gets frustrated
._.
and since gemini is pretty ass at tool calling in comparison to its actual logic capabilities, it threatens that ALL THE TIME
once i tried to make chatgpt write something using pyobjc and it got ccaught in an infinite cycle of debugging
Real
it couldn't quite figure out how to spontaneously get the system to generate a PDF like it does for canvas items so it went ahead and pulled some shitty python lib called reportlab and custom generated this scuffed pdf just for you
LOL
beautiful:
Holy scuff
another datacenter falls to the follow up "fix ur pdf brah" request
@final tinsel alright there - no dinner till the pdf at LEAST figured out what word wrap is so he touched it up
80% less scuff
It’d be great if it acknowledged the existence of markdown but I’ll take it
brotha
why you no say anything
it's default md
they pretty much shit out MD inherently
Aight who at apple decided that iPads can’t natively view markdown
The complete GPT-5 vs My Crappy Program saga:
i=0
out = ['!']*((9+8-7)*(6+5-4-3-2)+1)
outs = ''
while i < (9+8-7)*(6+5-4-3-2*1):
if i<1:
out[i]=chr(72)
elif i<2 or i/3==9%5:
out[i]=chr(90+(17%10))
elif i<4:
out[i]=chr(pow(10,2)+pow(4,2)-pow(2,2))
elif i==pow(i-2,2) or i==ord(out[8])-ord(out[11])-1:
out[i]=chr(int(0.5*(pow(2,2)+pow(3,2)-pow(4,2)+pow(5,2))*(pow(2,2)+pow(3,2)-pow(4,2)+pow(5,2))/2))
elif i==5:
out[i]=chr(int(pow(8,2)*1/2))
out[3*i-1]=out[i]
elif i%2==0 and (1/i)>.16:
out[i]=chr((1+2+3)*(5+i))
elif i==1+2+3-4+5:
out[i]=chr(pow(2,2)*pow((i-2),2)+(i-2))
out[(1+1)*(i+1)+1]=out[i]
elif i==ord(out[5])/4 or ord(out[5])/4==i-pow(ord(out[11]),(1/2)):
if i==ord(out[5])/4:
out[i]=chr(int(pow((6*i/4),2)-(i+2)*3))
else:
i=int(i-pow(ord(out[11]),1/2))
out[i+10]=chr(int(pow((6*i/4),2)-(i+2)*3))
i=int(i+pow(ord(out[11]),1/2))
elif i==ord(out[0])/(i-1):
out[i]=chr(int(pow(i,(1/2))+((pow(i,(3))-pow((i+i),(2))-(i*pow(ord(out[4]),(1/2))/(i))*(i+i)+((i)/i))/(2))+i))
elif i==((pow(2,2)+pow(3,2)-pow(4,2)+pow(5,2))/(2))-((2)/(2)):
out[i]=chr(ord(out[8])-i)
out[int((ord(out[6])/pow(ord(out[4]),.5))+i)]=out[i]
elif i==ord(out[9])-ord(out[7]):
out[i]=chr(pow((1+2),(3))+4*5*6-7*8+9)
elif i==ord(out[3])-ord(out[2-1]):
out[i]=chr(int((-i+((1337)/(420))*60+9)-ord(out[10])+2))
elif i>18:
out[i]=chr(-(ord(out[7])-ord(out[16]))+ord(out[2]))
i+=1
i=0
while i < 3+1*4+1+5+9-2+6-5:
outs=outs+out[i]
i+=1
print(outs)
when i saw "helloo, w" as the declared correct translation of the first part i think was the least AI hype ive ever felt in my life
Optimistn’t
who at apple decided the apple watch browser can't load images and breaks when you load literally any website
the last thing that copy of gpt5 sees
a tortured existence
also my vscode theme is so absurdly broken it's unreal
Man you are making me feel bad for it
i thank chatgpt because why not am I weird for doing that
its not for any particular reason
i dont really care for chatgpt
i do too, i cant treat something that can reply to me badly
or things that cant repy to me for that matter
but still it is genuinely helpful so theres no reason not to thank it
all my messages w/ any llm look like this it's so cringe
Human brain go empathy on anything vaguely human
i dont feel empathy for chatgpt
i find that being uber nice gets better responses anyways 
it does seem to have that effect
it's trained on data where the high quality data with good responses generally is going to be at least cordial
some of the worst goddamn work product that comes out of me at actual work is dealing with adversarial people
you'd respond better if given thanks
but yeah they mirror the overall """"""vibez""""""" of the data because that's literally all they are
the spooky ghost personification of a fuckload of text
they literally fall for pyschological traps that humans fall for
also why the middle of the bellcurve wall is so powerful
becuase that's the sum total/net result of the data
if you fit a function to that data
it's gonna be right there in the middle unless you adjust which is what all post training and dataset prep does on either end obviously
Chatgpt announced now verfied user can generate a adult stuff
Have u try python 3.14.2 version
Still don't use python 13 , 14 ,15 version
Use 8,9,10,11,12
And why would one not want to use Python 3.13+?
Lack of support
And why not use it for stuff that supports it? You can have more than 1 Python version installed after all
good morning!
Guess il do some 3d modeling
I just woke up about 2 hours ago too
I was thinking, if rust allows custom compile targets.. maybe someone already made a target for JVM
i remember there being a one for clr 
and sure enough this exists too https://github.com/IntegralPilot/rustc_codegen_jvm
nostd though
anyone have a good template for a docs website?
Signal is down 

Docker hub is down as well and I need to push to prod a bugfix 
AWS is having an issue in us-east-1, maybe it's related
npm is also pretty much down, takes forever to do anything (even moreso than is usual for it)
Yeah something central died again
but its not cloudflare this time since discord works 
first thing I did was check the cloudflare status page 
why is there a random ass network outage for random shit rn
something global died most likely
bc everyone and their mother apparently has all their stuff in aws us-east-1 for some reason
did aws go down 
https://health.aws.amazon.com/health/status they're having some issues
well the entirety of signal is down because of it
funnily enough, Signal's status page says everything is fine 
not surprised
on their end, they arent wrong

real and true
supposedly mcdonalds app is also dead because of it
so much for an anti-surveillance app
smh no geographic redundancy
you'd expect the big guys to have at least some, not to rely on us-east-1 to be invulnerable...
one would expect some system redundancy from anything that should be taken seriously tbh
That's why McDonald's is a franchise. The US app is down, but it's a completely different app from all others.
The issue is that some of these other apps just also happened to rely on that server. Responsibility distribution gone wrong
`export type ObserveMap = {
__index : ObserveMap,
new : () -> ObserveMap,
Observers : {
[string] : {
() -> ()
}
},
Add : (self : ObserveMap,
Eventname : string,
callback : () -> ()
) -> () -> (),
Call : (self : ObserveMap, Eventname : string) -> (),
Clear : (self : ObserveMap) -> ()
}`
`local ObserveMap = {} :: ObserveMap
ObserveMap.__index = ObserveMap
function ObserveMap.new()
local self = setmetatable({
Observers = {}
} :: any, ObserveMap) :: ObserveMap
return self
end
function ObserveMap:Add(eventName: string, callback: () -> ()): () -> ()
if not eventName then
warn("No eventName provided")
return function() end
end
if not self.Observers[eventName] then
self.Observers[eventName] = {}
end
-- Store the callback
table.insert(self.Observers[eventName], callback)
-- Return a cleanup function to remove the All callbacks
return function()
local observers = self.Observers[eventName]
if not observers then return end
task.spawn(function()
for i, storedCallback in observers do
if storedCallback == callback then
table.remove(observers, i)
break
end
end
if #observers == 0 then
self.Observers[eventName] = nil
end
end)
end -- oh my god
end -- keep going
function ObserveMap:Call(Eventname : string)
if not self.Observers[Eventname] then
warn("No event named : " .. Eventname)
return
end
local observer = self.Observers[Eventname]
for _, callback in observer do
task.spawn(callback)
end
end
function ObserveMap:CallAll()
for eventName, callbacks in self.Observers do
for _, callback in callbacks do
task.spawn(callback)
end
end
end
function ObserveMap:Clear()
self.Observers = {}
end`
they (supposedly) do have redundancy, each region is split into availability zones that aren't all supposed to break at the same time
doesn't always work out though, since deployments can still be made to a whole region
gotta love seeing graphs like these 
cross-region redundancy tho
wait what- really? I was waiting for starlight to install its dependencies, but npm create worked for me tho?
I feel like npm would use azure
since yk
npm -> github -> microsoft
-> azure
but idr if they actually use azure
that would be ideal, and most AWS services AFAIK are per-region like that
but for some services it's just not worth it I guess 
like, you can't just split stuff across regions as you want, bandwidth and latency and management becomes a much bigger issue
Iunno, they might be using multiple providers? hard to say, but I'd say it's related
yeah, that's true
I mean it would be smart to use multiple providers
so that we don't have the domino effect
I mean, npm still works, it just takes 5x longer than it should
so I'd wager they do have either multiple providers or at least are in more than just us-east-1 and the rest of the infrastructure is picking up the load, therefore being slow
probably the former
DynamoDB in us-east-1 is also not fully down, it just has "significant error rates"
so it might just be retries
compare with docker hub being fully down tho
worse error handling?
possibly
idk, hard to tell without knowing how horribly cooked DynamoDB currently is
and then there is docker.io who just used US-East-1 and full down 
funny how this just exposed who put all their eggs on the US-East-1 basket
turns out it's half the internet bc nobody bothers to change the default region...
it dies the worst time 
nice ci
would be great if it didn't pass, ain't it?
I have 0 idea why the base image is not in the cache
It has an important fix for a bug that is in Prod rn even so I'm pressured to push it 
is it possible that it is cached but it still checks if there's a newer version and it doesn't use the cached one if the online version check fails?
iirc some things can fail like this if they can't check the version at all, instead of using the cache
it's just how some stuff behaves, but usually there's config options you can pass to tweak it...
yeah okay, cache fails on master now as well
Great!
Oct 20 2:01 AM PDT We have identified a potential root cause for error rates for the DynamoDB APIs in the US-EAST-1 Region. Based on our investigation, the issue appears to be related to DNS resolution of the DynamoDB API endpoint in US-EAST-1.
it's always DNS
of course 
always dns 
absolute :classic:
what's so hard in turning strings into ipv4 addresses
small billion dollar company pls understand
🥀 should have a local DNS fr
cloud explod
Too big to fail
Does anyone know the python neuro sdk enough to help me with something?
The stacktrace might be big, but it is useful
No way our VERDAL bro post 15 hours video
I was leaving in basement soo long
CHATGPT CANNOT ProceD 4 MINTUS MAX
STILLL MY AI LIMITING MAX SET 6 HOURS
time is up
Time for class
what this mean
verdal bro
i like resonating
feel funny

chat i am so cooked i had an interview and
the mf just told me
i have 2 more to go if i'm lucky
HMMM
she told me one is a ML interview and one is a coding interview
i'm scared for the coding interview
cause i have weak DSA
Atleast I'm school students tho thank god
I just follow my passion
Not school will
section .data
msg db "Don't be scared bro just think I will crack it ", 0xA
len equ $ - msg
section .text
global _start
_start:
mov eax, 4
mov ebx, 1
mov ecx, msg
mov edx, len
int 0x80
mov eax, 1
xor ebx, ebx
int 0x80
haven't seen any new bots with weird names join the server today
???
wtf is he saying
Assembly jumpscare
"possibly related"
well
tomorrow's computer science class will be interesting
-# it's basically cs nerd hangout class
Did we get NN containment leaks into #programming?
it's still not recovered

Nope
Latest update
Why is there even a default aws region in the first place
Yo
True
Ronald McDonald has fallen it's so over
If wacdonal can't manage to keep their monopoly scam going then you know the Internet has fallen
If this is how they have their statuspage or even just how they put out this particular update
Based
Aws is down?
Beige comfy
No that’s my STARSS and I added the AWS status RSS feed
Frig
Local
Is good
I was mind blown thinking aws has a federation approved ux for anything
why is the enterprise giving me a network error
Was boutta log into AWS for the first time in 3 yrs
Picard ordered tea and the replicator gave him cold earl grey
damnn rip
The entire alpha quadrant is at risk
Lol, Picard bouta go on a rampage for his earl grey
Build gpu farm in home but I have extra place to store
Pipe unit almost installation
This was unplanned
https://tinygrad.org/#tinygrad
Really really hope that tinygrad is not blowing smoke with their claims
If they do what they say they are then itll make life so goooooood for training
STARSS is a news reader featuring a futuristic interface experience. STARSS provides convenient access to all your favorite news and blog sites via RSS feeds with classic design, animation, and sounds.
STARSS is a Universal App, which runs on iPad, iPad Pro, iPhone, and features full support for R…
What is that
they aren't, they've been cooking and they're heading in a pretty good direction I think
Bro have u sacrifices over performance of gpus
It is an iPad app it’s not like I’m gonna have this open as well as 12 other things simultaneously
300 mb left on mem poool
PAGE IS HIGH
Hey Google hire me please I will help in QUANTUM ai
they also have that Llama 405B contract with AMD so they don't have much a choice except to make tinygrad very fast 
Want me to connect you with my friend who does research in quantum computing?
New update
At least some things are back online- I’m getting canvas notifications but still can’t actually view course pages
AWS is down here as well, so uhhh. It isn't just the east side
US West just hasn't woken up yet
What !
I thought there was more discussion about other regions in that
I would have also thought so, weird
I'm lucky I did my channel reading so I could answer why amazon shopping didn't work
One sanity check later and there is a US-WEST region, wonder if it isn't known yet or? idk
Hellos 

What a great day to start my new job in tech support 

deadass this is actally my life right now
is wherever you work also down cuz of this?
did it get fixed already? cuz i odnt notice anything
US regions os AWS
My dad works in IT, he’s having to deal with issues on this rn 

so its still going on, just only in the usa?
Yep
i had a chat with amazon customer service literally yesterday and they were so nice and helpful. now they're probably in anguish 
Oh I just realized it’s prob cuz whatever ur trying to access is located at the east servers despite your physical location being west
aws explod
:awsflipexplode:
That'd make sense
You'd think at least amazon itself would have west services
Well yeah but I’d imagine not every customer has duplicates on both sets of servers
Also are u directly pinging an AWS site or just something that uses it or like…
I was asked on amazon shopping not fully loading
Oh yeah Amazon shopping is unaffected for me
I just wanted to submit my statics homework…
idk what aws is even responsible for
Like a third of all websites
all i know is prime video works perfectly fine here so i dont really care
to iggly tho
The name is Amazon Web Services, so hosting a few sites
not a single site i use it seems
a guitar can be encoded to 12 bits per action 
3 bit per finger
for the vertical at least, i have no clue how the horizontal works yet
what?
6 strings
- 1 for a neutral
3 bits is 8 combinations
you use 4 fingers to do the note stuff
12 bits
you use 5
not the thumb no?
yea, you wrap it over the top sometimes
oh fuck
and you can use your other hand as well
why cant guitar be easy?
i need a way to make the animation proceduraly generate based on the input
and you have to encode how long open strings ring out for
alright
i think thats jsut too much effort for animation
time to try to polish my old bot 
they'll hear it in the audio, im not making the strings animate perfectly
wait what are you doing
I read that as polish like as in from Poland

data -> poses
im trying to find a way to turn data into indices for poses
but i dont know how to play guitar
Hi! My band just released this single called "Shibuya" and I thought I'd make this guitar playthrough because people expressed interest in learning it! I purposely kept still and didn't move a lot so you can watch my fingers and see the frets :) This is my first time editing this thing by myself so I hope it helps!
Thank you to my bass player Da...

HMMM, AWS said next update was at 10am their time… which is now… and no update…
no i mean like uhh in general, animating a guitar for a game?
base64 encoding is basically a way of data compression so we send less things over cable?
ye
i could pregenerate the entire sequence manually, but thats a lot more lame
no, it increases the size
what is the source data for the positions or whatever you're encoding into those 13 bits
none
im still trying to invent the method im gonna use

to encode data into a printable string

make it procedural from midi 
also, isnt that just a guitar tab?
that sounds like even more of a pain than asking for the notes
is what jsut a guitar tab?
there are 2 E's 
yes, standard tuning has two Es
interesting
different octaves
bruh
wait, is ALL data going through internet being encoded?
I need data bro's
the point of base64 is to encode arbitrary data into an alphabet of size 64 for the purpose of mapping it to printable characters. it increases the size because you're basically mapping each 6 bits (because 64=2^6) onto 8 bits (a char).
if encoded to utf-16 it's gonna be 10 extra bits instead of 2
this doesnt answer why
"printable" - for printer???
anyway i googled and they say its because some part of data can be interpreted as some sort of command
and no command consists of purely printable characters so it becomes safe
and i was wondering before "how are the commands just a sets of characters, what if my jpg has this sequence of characters"
now i know 
because there are plenty of text medium. What if you wanted to share an unprintable byte sequence via a text message. Or encode an image in a html file
integer
4 fingers with movement over 6 strings + tumb over 1 string. you can have the fingers next to eachother, on the same fret, or 1, or 2 inbetween, and the whole hand moves over the frets too.
so 4 * 3bit + 1bit + 4 * 3bit + 5bit
12 + 1 + 12 + 5 bit
30 bit
i think that works?
this is suddendly becoming a lot more complicated



maybe tomorrow


didnt even notice aws being down

