zlacker

So can they still do traditional deterministic(?) calculations? Or would that be somewhat akin to using machine learning to do your taxes; possible but just overkill?

I've often heard it said that Quantum Computers can crack cryptographic keys by trying all the possible inputs for a hashing algorithm or something handwavey like that. Are they just spitting out "probable" solutions then? Do you still have to try a handful of the solutions manually just to see which one works?

>>arkanc+(OP)
Here's what Scott Aaronson says in regard to "trying all the possible inputs": https://news.ycombinator.com/item?id=17425474

>>arkanc+(OP)
"Trying all possible solutions" is generally a bad metaphor for quantum computing and will confuse you. (Its more like you start out with all answers being equal probability, and get the wrong answers to somehow cancel each other out making the "right" answer have a high probability)

I am not a quantum person but i once saw a geometric explanation for grover's algorithm which kind of made it all make sense to me. (grover's algorithm is the quantum algo you use for problems where you dont know any approach better than brute force. It can bruteforce stuff in O(sqrt(n)) guesses instead of O(n) like a normal computer). Basically, the geometric version was that you start with all possibilities being of equal probability (i.e. an even superposition of all possible states), negate the amplitude of the correct answer, then reflect the amplitudes around a line that is the mean of the amplitudes (do that sqrt(n)) times. The end result is the correct answer has a higher probability than the other answers. I unfortunately can't find the thing where i originally saw this, but they visualized it basically as a bar graph (of the amplitudes of possible states) and it seemed much clearer to me than other explanations i have come across

>>arkanc+(OP)
Yeah, they can do deterministic calculations. You just avoid ever putting the quibits in a state where measurement gives probabilistic results. it would, however, be a ridiculous use of technology, like using a Neural net to simulate an XOR