40 comments

[ 2.6 ms ] story [ 86.0 ms ] thread
What is possible on these kind of processors at the minute?
We cannot know because if we check this will influence the result of the calculation.
Even when facing specialized quantum circuits classical supercomputers are still more cost effective.

Quantum computers can't solve any problems of sizes that occur in industry. There are huge problems with scaling up quantum computers and there are some reasons to doubt that we will ever be able to scale it up as much to break crypto or solve industrial problems.

How cost effective were microprocessors in the 60s?
Than you. Somebody speaks with some common sense. All major breakthroughs start small and are not cost effectvie immediately. People should be way more enthused that this is possible rather than talking about all the negatives.
I don't think it's true that all major breakthroughs start small, but more importantly, you have no basis for assuming this is a major breakthrough. The issues stated here are not just cost problems of scaling up, but theoretical problems of what this technology can accomplish.
They might have stayed cost prohibitive for a lot longer if it weren't for the Apollo program buying up everything they could get
They became cost effective in a few years. The average price per integrated circuit dropped from $50 in 1962 to $2 in 1967. By then Moore's Law was established. At the end of the decade we had hundreds of transistors on a chip. At the early 70s we already had general-purpose microprocessors with thousands of transistors.

Quantum computing is being researched since the 1980s without a theoretical understanding whether they can be built at a practical scale.

As pointed out by others, they were very cost effective. And they replaced vacuum tubes, which were fragile. But, the most important thing is that they immediately solved a problem (ballistic trajectory calculations in missiles) and therefore had an industrial use which led to a virtuous cycle of improvement.

Quantum computers aren't useful; they're still searching for a use.

I think that NSA and similar organizations would be interested in breaking encryption algorithms. Imagine some old archive encrypted by RSA which contains useful data.

Also that means that encrypting anything valuable with quantum-unresistant algorithm and storing it in an unsafe place is not wise even today.

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 understand it, the biggest problem is getting the manufacturing precision good enough. The more you scale, the more precise the manufacturing needs to be, which was never really a problem with microprocessors.

At the moment I believe they can only solve toy problems (like factoring the number 15 - it's apparently 3 * 5 with high probability) and/or ones that are specifically contrived to show quantum supremacy.
What is the probability that 15=2*7?
Exactly 1, for large values of 2.
Or low values of 15. We must consider that other possibility.
I'd say exactly 1 for very moderate or average values of 2. The very large values might converge on 21 but I'll have to flip a coin on it.
What is 2? What is 7?
now youre asking the right questions lol
We must run enough samples of 2 and 7 through the AI/ML to be sure.
You mean "what's the probability that your computer will say that 15=2*7"? Well, that's nonzero for any classical deterministic algorithm you can run on your computer too.
"like factoring the number 15 - it's apparently 3 * 5 with high probability"

I thought you were talking about deep learning there for a second

The values of 3 and 5 will have to be very precise if you want a good 15 out of the multiplication.

You may also have a high 5, that could compensate for a low value of 2.

Do you happen to have some links about high 5 and low 2? Interested in what these mean.
I imagine they just mean something like how 15/5.9=2.54
15 ≈ 5.8 * 2.6
Supposedly the factorization of 15 is controversial. It is claimed that the process implicitly assumed the correct result.

Not even a single logical qubit has been achieved to date, even with error correction, so it might be a bit soon to talk of factoring anything.

What do you mean? I thought we were able to run programs on these processors already.
"Running programs" is easy. Getting correct answers, not so much. In the case of factoring 15, afaik, nobody has been able to execute Shor's algorithm faithfully -- instead, they simplify the factoring problem using knowledge of the answer. This isn't necessarily about Google -- last I looked, IBM holds the "factorization record" where they were either using a modified Grover's algorithm that incorporated knowledge of the factors. This isn't my area of expertise, but if it was, I'd be hugely embarrassed by the industry.

And then off in the corner, we've got D-Wave, who everybody loves to hate on, doing their own thing with an optimization-based approach to factoring which actually seems to work on (iirc) 10-ish bit semiprimes and zero foreknowledge.

I've come to understand quantum computing problems are basically just specific formulations of things where multiple bad paths will just cancel out due to some sort of probability amplitude summing to 0. I suppose this makes sense then that the answer would be associated with some probability rather than a fixed answer.

Is this theoretically likely to give near certain probability to the right answer on large problems or will it be affected by local minimums and what not?

Is that the issue with scale? That noise propagates through the system and renders the uplift in problem solving moot?

Many quantum algorithms can be iterated to find a solution with arbitrarily high probability. For example, Grover's algorithm has an amplification step, and Shor's algorithm can be repeated (with a different seed guess) creating a new solution with roughly independent probability.
No, quantum supremacy has been displayed already by google[0].

[0]: https://ai.googleblog.com/2019/10/quantum-supremacy-using-pr...

That's an example of what I meant by problems that "are specifically contrived to show quantum supremacy". Scott Aaronson explains this quite well:

https://www.scottaaronson.com/blog/?p=4317

> So, tl;dr, the quantum computer is simply asked to apply a random (but known) sequence of quantum operations—not because we intrinsically care about the result, but because we’re trying to prove that it can beat a classical computer at some well-defined task.

If you want a transistor-count analogy, it's low double-digits. Think 74-series logic.
It will be a while before anything of value can be done on a quantum processor. For now, all that will be done are increasingly convincing demonstrations of 'Quantum Supremacy'. This means that the quantum processor will solve a problem that is believed to be intractable on a classical computer. These problems are very contrived though, no one will actually care about the results of the computations.
If you are interested in the oral histoy of Shors algorithm and error correcting quantum codes. Shor (yes the one who invented Shor's Algorithm) gave a create 30 min presentation here https://www.youtube.com/watch?v=6qD9XElTpCE .
Thanks for sharing, it was really interesting to hear what was influencing him when he discovered shors.
From the linked article, it sounds like this work is getting pretty close to being able to provide a single error-corrected logical qubit, depending on what threshold one expects for the error rate. Is that a fair reading? How close are we to having a realistically usable logical qubit?
> The second scheme, on the right ... Calculations must be discarded rather than corrected when problems are found.

I understand the title is technically correct, and the field of quantum error correction (and error correction in general) encompasses the detection of errors as well as the correction of them. However error detection and error correction are different (but highly related!) problems, and there are are a variety of codes that are good at each. They tried out one error correcting code, and one error detecting (in this instance) code.

It's very cool to see the surface code put to use, and I've always found it intriguing how quantum computer design is motivated by the topology of the underlying error correcting codes. There's a similar proposal for ion-trapped quantum computing where you use a 2D array of trapped ions, rather than a single linear chain, to represent your qubits. You then use different atomic ions for your logical vs error correcting qubits, and you need some different fundamental operations than you do for interactions with a linear chain of ions.