thxg

6 ms·

31.
by thxg 4y ago
Wordle works with a 12k-word dictionary, which is very much comprehensive (that includes "words" like "grrrl"). However, the hidden/secret words are picked from a much smaller set of 2k reasonable words (i.e., frequ
32.
by thxg 4y ago
"Cheating" is subjective, but you are correct. To put it in more neutral terms: Using the known 2309-word list of solutions, hard mode can be solved 100% of the time within 6 guesses. Not using it, you would need 7 guesses to guar
33.
by thxg 4y ago
(The article is from 2018.) I think the article is strongly from the point of view of what to do "in production". If you have a bunch of servers with known specialized workloads, I can believe that enabling swap is good for effici
34.
by thxg 5y ago
Guesses come from the 12972-word list in both cases.
35.
by thxg 5y ago
Just to make things extra clear, the post we are commenting on has - 99.4% in 6 guesses or less - 3.64 guesses average - no attempt of any formal optimality result, while the one you posted has: - not just 100% in 6 guesses or less (so 100%
36.
by thxg 5y ago
Ok, I found this one [1] with 100% wins on the whole 12k-word set. But it optimizes for worst-case, not average. [1] https://www.poirrier.ca/notes/wordle/
37.
Mathematical optimization over Wordle decision trees(poirrier.ca)
3 points by thxg 5y ago|0 comments
38.
by thxg 5y ago
This [1] solves it in 3.4212 average guesses 100% of the time and claims optimality, which is 99% of the difficulty it seems. For the limited 2671-word set of possible words, I think the question is now settled (for the whole 12k-word set,
39.
“Fused-off” AVX-512 can be enabled on Alder Lake by disabling E cores(phoronix.com)
8 points by thxg 5y ago|0 comments
40.
by thxg 5y ago
You're right, I didn't know that. In a quick "long double" vs "double" microbenchmark, I see a negligible penalty for additions and comparisons, and 37% penalty when multiplications are involved.
41.
by thxg 5y ago
The article covers ways to obtain smaller error bounds, mainly for testing algorithm stability in practice. The methods proposed are very varied in nature: blocked algorithms, x87 80-bit arithmetic, AVX2/ARMv8-A fused multiply-add (FM
42.
by thxg 5y ago
Code using "long double" on x86_64 will compile to x87 instructions (at least with GCC on Linux). This can be triggered unintentionally. For example, string-to-double conversions in Boost use long doubles. (I have been bitten by t
43.
by thxg 5y ago
> FMA is a godsend for writing accurate arithmatic. True, and the small performance boost (on some platforms) is nice too > never reordering floating point computations unless you explecitly give permission Yes, requiring explicit cod
44.
by thxg 5y ago
Yes, indeed it is the case with fused multiply-add operations, as I wrote below in another thread. By default at -O3, unfortunately, gcc and icc both take responsibility for deciding whether or not to use FMA instructions when they see (a x
45.
by thxg 5y ago
This work essentially allows one to go beyond the IEEE754 standard which mandates correctly-rounded elementary operations +, -, x and / : they provide implementations of libm functions (cos, sin, log, exp, etc.) that are also correct
46.
by thxg 5y ago
Nowadays, all major platforms follow IEEE754, which specifies exactly the results of elementary operations (+, -, x, /). You must get bit-for-bit reproducible results across different runs, compiles, libraries, OSes and hardware. Any d
47.
by thxg 5y ago
The short papers accompanying the submissions are not out yet, but apparently the winning team did not use ML at all: https://mobile.twitter.com/alexkontorovich/status/1423259110...
48.
by thxg 5y ago
General introduction to the TSP problem video by a member of the winning team: https://m.youtube.com/watch?v=vQMFVLa7yYQ
49.
TSP solver wins $100k prize at Amazon's “Last Mile” ML Challenge (using no ML)(amazon.science)
3 points by thxg 5y ago|2 comments
50.
by thxg 5y ago
Video explanation by the very entertaining Bill Cook from the winning team: https://m.youtube.com/watch?v=vQMFVLa7yYQ One peculiarity of the winning submission is that, in an ML challenge, they did not use ML at all: https
51.
TSP solver wins $100k prize at Amazon's “Last Mile” ML Challenge (using no ML)(amazon.science)
4 points by thxg 5y ago|1 comments
52.
by thxg 5y ago
Yes, although the trick is used in the first place in order to replace a 6-cycle division by a 1-cycle multiplication (on x86_64, assuming no CPU stalls). Roughly, in order to still make it a net positive, we have a 4-cycles budget for addi
53.
by thxg 5y ago
Hmmm... maybe Zoom devs too use Stack Overflow? https://stackoverflow.com/questions/3376679/qt-how-to-detect...
54.
by thxg 5y ago
^ This is the correct answer. I installed the Zoom client just to have a look for myself. The syscalls in question emanate from freshly forked processes that immediately execvp() the command `pidof` (on my system it finds it under /usr
55.
NSF funds $20M grant to merge AI and Mathematical Optimization(ai4opt.org)
2 points by thxg 5y ago|0 comments
56.
NSF funds $20M grant to merge AI and Mathematical Optimization(ai4opt.org)
1 points by thxg 5y ago|0 comments
57.
by thxg 5y ago
Sorry, we can write iterations <= 2N. We don't even need O() here.
58.
by thxg 5y ago
I'm not sure we should answer homework questions here, but this one is cute so... At first sight it seems like O(N log(N)) at worst, because the outer loop does log(N) iterations. But you can actually get tighter. The number of inner l
59.
by thxg 5y ago
Also, the article concludes that FMA hurts accuracy, and this is at best misleading (FMA is better in the worst case). But the example given does highlight something important: Depending on the case, the accuracy gains can be tiny (so much
60.
by thxg 5y ago
Okay I got nerdsniped by this, reimplemented and reproduced it. As others mentioned, the article does not show a flaw in the M1's FMA. The results are the same on x86_64. It is just due to how this particular formula interacts with flo

More ›