« first day (4863 days earlier)      last day (278 days later) » 

att
att
02:07
is there a vanilla bijective-base en/decode challenge
@lyxal I think I can't pronounce half of sh
02:34
@lyxal he sussy
(I assume that was the joke) (you're welcome for pointing it out anyway :P)
shes fussy
@lyxal [ussy]
 
1 hour later…
03:44
@hyper-neutrino it's in the same spirit as sbeve
I figured lol
03:57
14 year old code golfers be shaking rn
 
1 hour later…
05:07
> Austin Robert Carlile (born September 27, 1987) is an American singer, songwriter and musician from Pensacola, Florida. He achieved prominence as formerly the lead vocalist of Attack Attack! and Of Mice & Men. After leaving Of Mice & Men, he began coaching youth baseball in Costa Rica.
Such a random career change lol
06:07
I just made a catastrophic edit to one of my answers by mistake and for some reason I see no way to roll it back? codegolf.stackexchange.com/a/273239/97916
I thought you could do that if you click on the 'activity' button
I rolled it back for you, but for reference, the rollback button is on the edit history page
you get there by clicking "edited <time frame> ago" next to the "answered by" box
aha, that's what I was missing. Thank you
nw, happy to help
Personally I'm on the side of the skibidi toilets. The cameramen are too authoritarianism-pilled for me
I know it's a controversial take, but I think the skibidi toilets were here first and have the right to stand up to the well-organized, mind-controlling forces that want to keep them down
06:22
unrelated, but I find it interesting that when MDN partners with OpenAI no one cares or makes noise about it, but when SE partners with OpenAI, you get protests, a thousand downvotes and displeasure from the community
Far less people contribute to MDN
even then, you don't hear of news articles quoting people quitting contributions to MDN and encouraging vandalism of posts
@Adám Sorry for the delay in responding. Yeah, I'd be down, looks like an interesting podcast! For full clarity, Jam never really got off the ground (uni kinda took focus at the wrong moment :P) but I did publish the fork of Jelly that's been used for unofficial updates to Jelly
I thought the fork was Jam
Nope; Jam was the planned successor to the fork, as an official and independent language
06:33
Ah
mb
I suppose, technically, Jam is the fork, as the entire content of Jam atm is just the fork :P
But conceptually, Jam != Jelly (caird fork)
@lyxal There is much less expectation of ownership on MDN. Posts don't even list their authors. So it makes sense people would be less upset. Also you can't deface because someone needs to review the pull request
I put caird's jelly fork on DSO a while back
There's also M which is jelly but with arbitrary-precision rationals
@mousetail fair enough
 
2 hours later…
08:13
@Bubbler I like this idea! The reason I ask is I want to multiply an array of numpy.float128 more quickly using cython or numba. But I don't really trust my expertise so I want to have some way I am still handling things with the right precision.
numba doesn't support np.float128 so I have to implement a type extension which I haven't done before
@Bubbler I get 16444 for np.float128 and 1074 for np.float64
is that expected?
sounds about right
it's all very confusing
y = np.float16(np.random.random())
c = 1
while y != 0:
    y /= 2
    c+=1
print(c)
1071
what on earth is going on?
log2 plus a bias
float64 = C double = 11 exponent bits; about half of the space is used to express numbers lower than 1, so ~1024 divisions to get close enough to 0
yes.. but float16?
08:27
the exact number depends on things like subnormal numbers
so y eventually convert into float64?
shouldn't float16 give a very different answer to float64?
what if you try with y = np.float16(1)?
(shouldn't significantly change the result but just checking)
1076
>>> np.float16(2)**-1024
5.562684646268003e-309
08:30
ah I think numpy casts it to float64
>>> np.float16(2**-32)
0.0
Is there a typeof in python?
In [39]: y = np.float16(1)

In [40]: np.finfo(y)
Out[40]: finfo(resolution=0.001, min=-6.55040e+04, max=6.55040e+04, dtype=float16)

In [41]: y /=2

In [42]: np.finfo(y)
Out[42]: finfo(resolution=1e-15, min=-1.7976931348623157e+308, max=1.7976931348623157e+308, dtype=float64)
so it seems like you need to divide by np.float16(2) instead of just 2
I then get 26
>>> np.float16([1,2,3])*1e-8
array([0., 0., 0.], dtype=float16)
>>> np.float16([1,2,3])*1e-4
array([0.0001, 0.0002, 0.0003], dtype=float16)
08:32
numpy has its own implicit casting rules seems invented without any reference to what had been agreed in other more famous languages before
IEEE754 float16 has 5 exponent bits so it checks out
thanks.. no so see if I can get numba to multiply an array of np.float128s correctly!
gulp :)
Is there IEEE754 float8 or even float4? (I know float8 and float4 exist)
no, smallest float type classified under IEEE754 is float16, according to wikipedia
@Bubbler what was "no" in reply to?
08:35
@Simd I
s/no see/now to see/
but I didn't say there was a smaller float type
Are 0xD6 L1OM opcodes only available in 64-bit to avoid confusion with SALC?
so ML people kinda arbitrarily invented 8-bit and 4-bit floats to train their models faster
...and I hear now that they reduced it down to just booleans
float1 doesn't make much sense but int1 do
float2 can handle {0,1,2,4} but doesn't look like useful, though can't just claim useless like float1
08:40
that makes no sense!
they are supposed to be doing stochastic gradient descent
but what is that with a 4 bit number??
do you have any reference to them using booleans? I would be really interested in that
I just heard from a person who happens to work with such an AI model
ah ok. I would love to see that in practice. It's not a float!:)
lol yeah.
AI researchers are so good at creating buzzwords for their papers
08:49
:)
0
A: Sandbox for Proposed Challenges

l4m2The mail sender There are N houses and some paths in a town and. Some mails need sent from a house to another. You are now in house 1, and it's your job to send these mails as soon as possible. Output the shortest time. Test cases: (Path P1, P2, cost)... [Src, Dest]... => Cost (1,2,8) [1,2] => 8 ...

09:04
@emanresuA My fork also uses arbitrary precision, to a degree. More than M does, but less than I'd like
Also, I either genuinely didn't know, or fully forgot, that it was on DSO :P
Thanks for adding it!
0
Q: Is this a solvable Numberlink?

badatgolfChallenge Per Wikipedia: Numberlink is a type of logic puzzle involving finding paths to connect numbers in a grid. The player has to pair up all the matching numbers on the grid with single continuous lines (or paths). The lines cannot branch off or cross over each other, and the numbers have t...

 
6 hours later…
14:58
Live Nation (Ticketmaster) just got hit by an anti-trust suit :D
15:17
@lyxal SE has a rebellious side, and more places to coordinate as a commubity
"Commubity" lol
My favorite part was when he said "it's commubing time" and commubed all over them
rydwolf, check TGC
NP posted a question six times
15:39
Does anyone else constantly press control-shift-C instead of control-C in places where that doesn't work?
16:36
I don't, probably since I'm used to using terminals where highlighting text automatically copies it
It might be a bit late for this, but do y'all think Construct a uInt from an array of bits would benefit from a community wiki answer for languages where the solution is a builtin?
In skimming through the answers so far, I think I only see one that's a straight builtin. There's a couple more that are "join into string and then use builtin."
 
1 hour later…
 
1 hour later…
19:00
@cairdcoinheringaahing can i give your email to the host?
 
1 hour later…
20:18
Hey @Ginger, boy do I have good news for you
(CC @Seggan)
Ugh the image is too big
We have ceiling boot 😎
@Adám feel free :)
20:33
@RydwolfPrograms I believe Pippi Longstocking had some ceiling boots at one point
20:46
@RydwolfPrograms :D
my appreciation is redundant because I saw your Discord ping first, but the sanctity of the transcript must be upheld
22:15
@RydwolfPrograms ooh
22:27
@emanresuA Let him cook, he's clearly onto something
If Calvin's Hobbies were still around, we'd have a black-hole-themed challenge within 24 hours
22:55
Fun fact: I've used the Writer monad in Haskell, but I've never used the Reader monad nor the State monad.
Should I try using them, or even try to use them?
what's the difference :p
The difference between what and what?
"Try V-ing" is a mere attempt. "Try to V" is an effort.

« first day (4863 days earlier)      last day (278 days later) »