Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

LLMs won't break symmetric crypto

bfswa.blog

Post reply on HN
Print view View original

101–110 of 110 posts

Re: LLMs won't break symmetric crypto

#101

Post by tptacek »

xboxnolifes wrote: post #100
inigyou wrote: post #83

Earlier quoted context omitted.

There is no such thing. There is just more thorough and less thorough bit-blending. Have a look at any symmetric encryption algorithm or secure hash - there's no magic there, they just do a bunch of stuff to thoroughly mix up all the bits. In the case of symmetric encryption, they constrain it to reversible mixing primitives. And cryptographic experience has shown how much mixing is enough mixing. A few basic things…

If there is no such thing, how have people found weaknesses in various cryptographic algorithms over the years? The weaknesses are incorrect algorithms.

They haven't. Excepting RC4, which is wildly different from every other mainstream cipher, the track record of practical block cipher cryptanalysis (of core algorithms) is meager. If you redesigned DES so that it had 16-byte blocks and keys and the round structure to match, but retained the rest of its core design decisions, it'd presumably be viable (if slow).

Re: LLMs won't break symmetric crypto

#102

Post by SideQuark »

arberx wrote: post #13

LLMs will accelerate math research, increasing understanding in areas like quantum which will eventually lead to breakthroughs that will break most standard asymmetric encryption algorithms with the side effect of breaking crypto

There are plenty of algorithms that quantum algorithms give no benefit to. Quantum is NOT a free for all making everything faster. It’s only faster at two very,very restricted things: hidden subgroup problem (which is used for RSA, but most problems do not rely on HSP), and Grover search, which reduces a specific search problem from O(N) to O(sqrtN).

Re: LLMs won't break symmetric crypto

#103

Post by SideQuark »

danielmarkbruce wrote: post #56
insanitybit wrote: post #50

Earlier quoted context omitted.

It's unclear what you are talking about then. Because the idea of training "ciphertext -> plaintext" for language models is absolutely bonkers, so what are you suggesting?

I can't tell if you are serious at this point. I literally say, twice, that I'm not talking about a language model. And I also say the model would predict the key...not the plaintext.

Have you ever built a NN model? Have you ever broken a crypto system, even a small one?

Re: LLMs won't break symmetric crypto

#104

Post by SKYNET800 »

inigyou wrote: post #84
SKYNET800 wrote: post #72

Earlier quoted context omitted.

Have you seen the code and what it does? It’s science, go and take a look.

I did. Looks like technical analysis, which is pseudoscience. For some reason the code also talks about animals and limbs. Very large amounts of the code are also spent on useless details like logging, and monkeypatching matplotlib, that no human would spend so much code on.

Didn’t you realize that logging is needed for “DeepSeek”? “Have you seen the formula, and that this is the alpha version? Are there formulas in pseudoscience? This code was written by “DeepSeek” — you can find a lot of things that he wrote not very well, but for now it works. You’re just a critic, without any real understanding.”

Re: LLMs won't break symmetric crypto

#105

Post by danielmarkbruce »

SideQuark wrote: post #103
danielmarkbruce wrote: post #56

Earlier quoted context omitted.

I can't tell if you are serious at this point. I literally say, twice, that I'm not talking about a language model. And I also say the model would predict the key...not the plaintext.

Have you ever built a NN model? Have you ever broken a crypto system, even a small one?

yes and yes.

You can build and train a model in about 15 lines of pytorch. And you can build and break your own 8 bit xor cipher in about 10 lines of python.

Hacker news is full of software engineers. You are unlikely to find one that hasn't built a model using pytorch these days, and an xor cipher is a common university lab exercise.

Re: LLMs won't break symmetric crypto

#106

Post by insanitybit »

danielmarkbruce wrote: post #94
insanitybit wrote: post #89

Earlier quoted context omitted.

I'm not debating you at all. I'm asking what the model looks like since you've stated (and I've agreed) that a language model wouldn't work. I think it would make sense to explain how a theoretical model could do better than SAT. Otherwise, is the idea here just "magic is possible"?

Yes, "magic is possible" if you defined "magic" as "very large models approximating functions in a way that people didn't think would work". Current SOTA language and vision models, or models used to predict protein shapes are magic by the standards of 2016. As for why could it be better than a SAT? Why couldn't it be? Models are better than deterministic, logically written software for lots of situations. You can cr…

Sure, that seems reasonable enough. I'm pretty skeptical that it will happen, but it's not like it's impossible.

Re: LLMs won't break symmetric crypto

#107

Post by inigyou »

SKYNET800 wrote: post #104
inigyou wrote: post #84

Earlier quoted context omitted.

I did. Looks like technical analysis, which is pseudoscience. For some reason the code also talks about animals and limbs. Very large amounts of the code are also spent on useless details like logging, and monkeypatching matplotlib, that no human would spend so much code on.

Didn’t you realize that logging is needed for “DeepSeek”? “Have you seen the formula, and that this is the alpha version? Are there formulas in pseudoscience? This code was written by “DeepSeek” — you can find a lot of things that he wrote not very well, but for now it works. You’re just a critic, without any real understanding.”

Yes, there are formulas in pseudoscience.

You clearly have no idea how any of this works.

Re: LLMs won't break symmetric crypto

#108

Post by SKYNET800 »

inigyou wrote: post #107
SKYNET800 wrote: post #104

Earlier quoted context omitted.

Didn’t you realize that logging is needed for “DeepSeek”? “Have you seen the formula, and that this is the alpha version? Are there formulas in pseudoscience? This code was written by “DeepSeek” — you can find a lot of things that he wrote not very well, but for now it works. You’re just a critic, without any real understanding.”

Yes, there are formulas in pseudoscience. You clearly have no idea how any of this works.

Are you just so defensive about your beloved psychological Bitcoin, thinking it can’t be wrong? I’ll disappoint you — my system works, and I know how it does it at a scientific level; the formula has been tested.

Re: LLMs won't break symmetric crypto

#109

Post by SKYNET800 »

ande-mnoc wrote: post #74
SKYNET800 wrote: post #72

Earlier quoted context omitted.

Have you seen the code and what it does? It’s science, go and take a look.

You are expecting people to read 25K+ lines of code in a single Python file that is generated by LLM and then translate all the comments written in Russian?

Not 25k+, in the release it’s 39.5k.

Re: LLMs won't break symmetric crypto

#110

Post by SideQuark »

danielmarkbruce wrote: post #105
SideQuark wrote: post #103

Earlier quoted context omitted.

Have you ever built a NN model? Have you ever broken a crypto system, even a small one?

yes and yes. You can build and train a model in about 15 lines of pytorch. And you can build and break your own 8 bit xor cipher in about 10 lines of python. Hacker news is full of software engineers. You are unlikely to find one that hasn't built a model using pytorch these days, and an xor cipher is a common university lab exercise.

When you conflate an 8-bit xor cipher with AES and a 10 line python NN with the complexity of trying to break AES with a NN by having "the model ... predict the key", I can see how you think such a thing is possible. It's nonsense.

For any NN to learn, the function it is approximating must structured enough to admit small set of parameters (i.e., not exponential), otherwise it will take exponentially many nodes to do anything. AES is not differentiable, as are all secure hash and encryption functions. This is a basic test that is used to attack everything.

So you'd get a NN that must be big enough to simply memorize all plaintext, key, output triplets, which is simply a lookup table. With around 2^768 nodes. Good luck.

Post reply on HN