[Cryptography] Has quantum cryptanalysis actually achieved anything?

[Stephan Neuhaus]

On 3/2/25 07:06, Ray Dillinger wrote:
> Stringing together a plausible-seeming sequence of words is not the only 
> way an AI can think.  An AI can do mathematics by querying MATLAB or 
> something that will get an accurate answer and use the Language Model 
> only to provide an interface to more appropriate problem-solving tools.

An LLM can do *calculation* by querying MATLAB, but it has to come up 
with the correct *approach* by itself. I believe that this distinction 
is important.

For example, I took a problem out of Acton's _Numerical Methods that 
(usually) Work_, namely the problem of having a 1-mile-long rail that is 
anchored at both ends, and which gets extended by 1 inch (or one yard, I 
forget the details, but they don't matter). The question is, what is the 
height of the circular arc into which the rail now bends. The answer was 
sought to four (I think) significant places in the book, but given that 
we all have access to double-precision arithmetic now, I think eight 
places is not unreasonable, for what the problem was trying to teach 
you. (It's a worthwhile problem to tackle, if you have some time.)

I gave that problem to ChatGPT (a while ago) and it came back with the 
completely wrong answer. The math was correct but the approach was 
absolutely bogus. It sounded convincing as hell though.

Fun

Stephan