abetusk

31 ms·

511.
FTCC: Fast Text Classification with Compressors Dictionary(github.com)
3 points by abetusk 3y ago|0 comments
512.
by abetusk 3y ago
Fez - beautiful art, beautiful music [0], nostalgia with a modern take Journey - beautiful game that does not center on combat Braid - good art, great puzzles and a game mechanic that is fully realized and explored (time rewind) Cuphead - g
513.
Gradient Descent More Efficient with Periodic Long Steps (For Convex Functions)(twitter.com)
12 points by abetusk 3y ago|0 comments
514.
by abetusk 3y ago
Interesting idea. I wish there were some visualizations for some real world graphs. For the lazy, in a control flog graph (CFG) or a directed acyclic graph (DAG), a node, A, is said to "dominate" another node, B, if all paths from
515.
by abetusk 3y ago
The appearance on Johnny Carson where Uri Geller failed to perform his magical ability is available online [0]. Johnny Carson used Paul Randi as an advisor to set up the experiment so as to counteract what tricks Uri Geller was using to ded
516.
Scalable Spectral Packing(news.mit.edu)
2 points by abetusk 3y ago|0 comments
517.
by abetusk 3y ago
I think I see. The economies of the internet are changing. Now with LLMs being accessible at an exponentially cheaper rate, we're seeing old models crumble and new models rising. The era of moderated user content is changing drasticall
518.
by abetusk 3y ago
The intent is to prevent exactly this type of possibility. They may be able to put it behind a paywall and copyright future work but not work that's already been published. Obviously not legal precedent, but there is some discussion on
519.
by abetusk 3y ago
As a reminder, all the SE sites have content under a Creative Commons, By Attribution, Share Alike license, allowing for, among other things, commercial re-use [0] [1]. Yes, it sucks that the SE sites are getting more draconian about allowi
520.
by abetusk 3y ago
I'm no expert but here are my opinions: One goal is to find an 'efficient' annotation of a superposition. That is, "what is the minimum superpermutation string length of of N symbols", which itself is a sort of opti
521.
by abetusk 3y ago
The article doesn't do a good job of explaining the constraint. The sequence being generated is called a superpermutation [0]. From the Wikipedia article: """ ...a superpermutation on n symbols is a string that contains
522.
by abetusk 3y ago
"... it is commonly referred to as the Nobel Prize in Economics." [0] [0] https://en.wikipedia.org/wiki/Nobel_Memorial_Prize_in_Econom...
523.
by abetusk 3y ago
Just a note about applications, there is a pretty direct application of aperiodic tiles with quasicrystals [0]. Diffraction patterns of quasicrystals can have symmetry that can't occur with periodic tiling which explained some weird di
524.
by abetusk 3y ago
The recently discovered 2D "hat" monotile tiles aperiodically but allows reflections (and must allow reflections to tile?). The paper discusses a (2D) monotile whose shape will allow an aperiodic tiling without reflections.
525.
by abetusk 3y ago
Yes, I think I understand, both why you believe P=NP and what the 2XSAT construction is. You're basically saying "Look at this polynomial problem and this other polynomial problem, if we combine them in a natural way, suddenly it
526.
by abetusk 3y ago
> I think your intuition is wrong here. The number of variables in the clause is not the reason why NP is difficult. XORSAT also has arbitrary number of variables per clauses, and is doable in P. Furthermore, the 2XSAT reduction shows th
527.
by abetusk 3y ago
> I am philosophically very close to finitism, and I think from the perspective of the Skolem's paradox it's not clear whether different infinities are not just a "mirage", so to speak. The distinction between finite
528.
by abetusk 3y ago
Godels incompleteness theorem and the Halting problem are very nearly equivalent. I believe Turing, very soon after hearing about Godels' incompleteness theorem, used it to show the non-existence of the Halting problem. The consequence
529.
by abetusk 3y ago
P != NP is more likely akin to a fundamental mathematical law, like Noethers theorem [0]. No amount of self improving AI will alter the fundamental laws of physics. More provincially, in our solar system, there's a cap on energy so tha
530.
by abetusk 3y ago
It's pretty old (3+ years) but I did some sophomoric analysis on some HN data [0]. I just added a frequency-karma plot and it looks power law, which should be no big surprise. Using a MLE from John Cook [1], the exponent looks to be ab
531.
by abetusk 3y ago
Thanks, I should have searched for the original article.
532.
by abetusk 3y ago
The leaked document from Google: https://www.semianalysis.com/p/google-we-have-no-moat-and-ne...
533.
“We have no moat, and neither does OpenAI”(twitter.com)
8 points by abetusk 3y ago|3 comments
534.
by abetusk 3y ago
The linked paper isn't rendering so well for me. Here's another version that's better? http://languagelog.ldc.upenn.edu/myl/Shannon1950.pdf
535.
by abetusk 4y ago
I'm not sure I understand it well enough to say but watching a video on it [0] I think there were a few key points: * "Attention is all you need" introduced positional encoding which allows you to keep context of the word, al
536.
by abetusk 4y ago
You continue to argue in bad faith. Copyright law is often used to prevent people from copying work. The GPL and its ilk are legal mechanisms designed to allow people to share their work.
537.
by abetusk 4y ago
That's an unkind reading. The implication is that GPL circumvents some relevant restrictions of copyright law in question by creating a legal framework to do so.
538.
by abetusk 4y ago
This was inevitable. Copyright law has always used a "color of your bits" argument [0]. GPL and other libre/free/open licenses were a great hack to circumvent draconian copyright laws but the laws themselves are not desi
539.
by abetusk 4y ago
My apologies, I've only skimmed, but maybe this encapsulates the main idea of your post? """ How MLPs could be trained was an open problem at the time, I think, but the field could easily have chugged along and we’d have
540.
by abetusk 4y ago
I don't think this is a correct characterization. Another content creator is getting takedown requests for vanilla gameplay [0]. PointCrow also mentions takedown requests for other videos that have nothing to do with the modded multipl

More ›