_hl_

17 ms·

91.
by _hl_ 5y ago
My reading of it is that it's essentially one-to-one (up to minor constant factors), they took the binary circuit for xorshift128 and replace the xor gates by functionally equivalent NN blocks, and then trained the weights.
92.
by _hl_ 5y ago
The proof accounts for that. You can even allow for the ANN being totally wrong on "almost all" seeds except for a non-negligible fraction, the definition of a CSPRNG allows for that. If it is wrong for all but a negligible fracti
93.
by _hl_ 5y ago
> That's somewhat equivalent of partially seeing the secret internal state Not just partially, xorshift128 is designed such that the output depends only on the last 4 results. So this is "just" learning the output function
94.
by _hl_ 5y ago
The argument is fairly straightforward: Let n be the number of seed bits of the CSPRNG. Assume there exists an ANN that approximates the CSPRNG so that the number of possibly wrong output bits is at most polylogarithmic in n (otherwise you
95.
by _hl_ 5y ago
What they are doing is essentially encoding the binary circuit of the xorshift128 PRNG as a neutal network. The fact that you can encode abritrary binary circuits as neural networks is well-known, so it's not surprising that it is poss
96.
by _hl_ 5y ago
The underlying assumption of all of modern cryptography is that one-way functions exist (implying P != NP), and further that the algorithms we use in practice are actually instances of these theoretically hard problems. So there provably al
97.
by _hl_ 5y ago
Done. Thanks for being so generous.
98.
by _hl_ 5y ago
Is that conditional on donating via some cryptocoin? I'd be up for donating a well-deserved $100 (can't really afford much more right now sadly) but prefer traditional methods.
99.
by _hl_ 5y ago
The French tax office for example has created a programming language in which the legally binding tax code is defined.
100.
by _hl_ 5y ago
It sounds like they are not even able to serve ads, on any property. So while far from perfect, it's probably a decent estimate without doing in-depth analysis.
101.
by _hl_ 5y ago
That's exactly the line of reasoning that these privacy invading companies want you to follow. The law doesn't mandate obnoxiously huge popups that slow to a crawl when you try to actually use them to disable cookies and whatnot,
102.
by _hl_ 5y ago
It means that the correctness properties that were formally proved are not necessarily what you would intuitively understand as "correct". E.g. you can formally prove that a buggy factorial(n) always outputs (n - 1)! by incorrectl
103.
by _hl_ 5y ago
I remember seeing that (regex CTs) and immediately thinking "wtf, why would anyone want to do that". Took me quite a while to understand that it's actually a pretty clever way to write fast state machines in browserland. So t
104.
by _hl_ 5y ago
Tangentialy related: Faster training of Neural ODEs is super exciting! There are a lot of promising applications (although personally I believe that the intuition of "magically choosing the number of layers" is misguided, but I&#x
105.
by _hl_ 5y ago
I think the point they were making is that LAPACK et al are a pain to use directly (they really are!), and abstractions are limited by the capabilities of the language or heavy to compile (not sure I agree, Eigen is pretty okay).
106.
by _hl_ 5y ago
The two aren't really related. It's a bit like saying "we were able to build computers, surely we can also build <insert arbitrary tech thing>". Still waiting for that high-temperature superconductor. As far as I u
107.
by _hl_ 5y ago
That's impressive! Are you willing to talk a bit about how this all works? I assume you host the hardware yourself somewhere, which in the days of AWS et al must be pretty tough to pull off, especially with these specs. Where do you ge
108.
by _hl_ 5y ago
Which is why, for this type of research, you use an ensemble of multiple competing models, which also helps with extracting probabilities. If you think their methodology is flawed, then have the decency to respect the enormous amount of eff
109.
by _hl_ 5y ago
Aren't parts of servo still part of Firefox, and hence those parts will continue to be maintained within the context of Firefox?
110.
by _hl_ 5y ago
That is... profoundly idiotic and short-sighted.
111.
by _hl_ 5y ago
One of the interesting points about SCION is that it achieves today's required forwarding performance on commodity hardware . No more expensive and inefficient TCAMs, no more limits to scalability. This, among other features of SCION,
112.
by _hl_ 5y ago
I'm not from the US, so my knowledge on the death penalty is quite limited. But I clicked on a link posted below listing executions in Texas, and I was shocked to see that the most recent execution was last year for a crime committed b
113.
by _hl_ 6y ago
This is only the preliminary concept mockup, not the final note.
114.
by _hl_ 6y ago
Indeed, but there is a much longer history of people trying and failing to break the schemes, and we have come to understand the hardness assumptions in classical crypto as probably quite reasonable.
115.
by _hl_ 6y ago
"Secure against quantum" doesn't really mean much because too little is known to make that claim confidently. AFAIK the term generally refers to algorithms that don't rely on factoring being hard, but instead make some d
116.
by _hl_ 6y ago
The hope is that quantum computers will give us an exponential speed up on some problems, which could (again, hopefully) allow us to solve some NP hard problems. This can be extremely useful for scientific computing (e.g. protein folding) a
117.
by _hl_ 6y ago
A quantum computer is defined as a machine that can implement an arbitrary "unitary evolution" (up to arbitrary precision). If you don't know enough math to understand what a unitary evolution is, think of it as a quantum com
118.
by _hl_ 6y ago
On the other hand, as someone who knows a bit of math, I love those Wikipedia articles that get straight to the point, and hate those where I first have to dig through a pile of (to me) useless "intuition". I'm not sure if it
119.
by _hl_ 6y ago
manim works by generating a "partial movie file" for each animation, i.e. a single mp4 file for each scene.wait() call and so on. The final output just stitches these together. I did exactly what you want using a small reveal.js p
120.
by _hl_ 6y ago
I think it's funny how the blog post is essentially an architecture documentation of the rust analyzer's architecture documentation.

More ›