Hacker News Reader

Comment by rain1 | Hacker News Reader

It is incredibly dishonest of them to post this without any details about the noise parameters of the system.

When reading "127-qubit system" you would expect that you can perform arbitrary quantum computations on these 127 qubits and they would reasonably cohere for at least a few quantum gates.

In reality the noise levels are so strong that you can essentially do nothing with them except get random noise results. Maybe averaging the same computation 10 million times will just give you enough proof that they were actually coherent and did a quantum computation.

The omission of proper technical details is essentially the same as lying.

View on HN

Basically little more than having a bathtub and claiming you've built a computer that does 600e23 node fluid dynamic calculations. But a lot more expensive.

I haven't read anything on this one yet, but your analogy fits Google's “quantum supremacy” paper really well, I likes it.

Ugh, the point is a fine one but it appears it has to be made:

Validation of experimental theory through the characterization and control of an entire system is not the same as building the same system and simply seeing the final state is what you expect. The latter is much easier and says very little about your understanding.

Here's an analogy: Two people can get drunk, shack up for the night, and 9 months later have created one of the most powerful known computers: A brain. Oops. On the flip, it's unlikely we'll have a full characterization and understanding of the human brain in our lifetimes – but if we ever do, the things we'll be able to do with that understanding will very likely be profound.

My reply was glib but I think in principle correct. The idea of a strictly controlled system in the NISQ domain to validate quantum supremacy in theory is an interesting approach, but it feel deceptive to me because this 127-qubit computer cannot in fact factor 127-bit numbers with Shor's algorithm or anything like that.

The accomplishment is more akin to creating a bathtub with 127 atoms and doing fluid dynamic simulations on that, which is a much harder problem in many ways than doing the 6e25 version of the experiment. But it is very questionable to me whether any claims of quantum supremacy retain validity when leaving the NISQ domain and trying to do useful computations.

Gil Kalai's work in the area [1] continues to be very influential to me, especially what I consider the most interesting observations, namely that classical computers only barely work -- we rely on the use of long settlement times to avoid Buridan's Principle [2], and without that even conventional computers are too noisy to do actual computation.

[1] https://gilkalai.wordpress.com/2021/11/04/face-to-face-talks... is a recent one

[2] https://lamport.azurewebsites.net/pubs/buridan.pdf

I mean, maybe in the theoretical sense, but do conventional computers barely work in practice? Seems a bit of a pedantic argument.

Gil Kalai and others with similar arguments play an important role in the QC community. They keep the rest of us honest and help point out the gaps. But I do think the ground they have to stand on is shrinking, and fast. Ultimately, they might still be right – that much is certain – but it seems to me that the strides being made in error correction, qubit design, qubit control, hardware architecture, and software are now pushing the field into an exponential scaling regime.

To me, the big question is much less whether we'll get there, and much more "what will they be good for?"

It seems like IBM has blown their credibility so many times. As soon as I saw IBM mentioned in the lead of the title, I knew what was to follow is almost entirely actual-content-free marketing spin.

With the caveat that the paper is from a competitor that I like, this benchmark paper [1] makes me inclined to disregard this result. See figure 1.

__EDIT:__ whoops wrong figure, just read section iv or see the first figure here [2]

[1] - https://arxiv.org/abs/2110.03137

[2] - https://ionq.com/posts/october-18-2021-benchmarking-our-next...

It's over here: [1]. James Wootton makes some comments about it here [2]. Looks like 1% error rate which I think is too high to do much with, but it is still exciting progress for those of us working in the field!

[1] https://quantum-computing.ibm.com/services?services=systems&...

[2] https://twitter.com/decodoku/status/1460616092959883265

It's particularly jarring given that IBM came up with the concept of quantum volume [0]...

[0] https://en.wikipedia.org/wiki/Quantum_volume

    > The omission of proper technical details is essentially the same as lying.

Welcome, child, to the beautiful/"special" world of marketing!

Be gone fetus, for there has always been honest marketing. It's just harder to find, sadly.

If it’s below the noise floor, how do we separate it from ‘accidental (or incompetent?) honest marketing’?

Oh, I do not deny there are honest marketing people. But there are plenty of people in that game happy to omit some inconvenient factoids. And as always, those are the ones giving their craft as a whole a bad name.

Welcome to IBM. Is Quantum the new Watson?

I guess they should show that they have achieved quantum advantage:

The Chinese did show it some time ago:

https://www.globaltimes.cn/page/202110/1237312.shtml

Also another problem: you now have 2^127 output values leaving the quantum processor. If you're using a hybrid quantum algorithm that requires classical processing as well (which are most algos used today), you'd need more than a yottabyte of RAM. We can get around this problem by storing all 2^127 pieces of output data into other data types that compress the total size, but if you genuinely are trying to use all 2^127 outputs, you'd still need to do some pretty intensive searching to even find meaningful outputs. I guess this is where Grovers search could come really handy, right?

You don't get the entire wave-function as output; the wave-function is not observable. Different measurements might reveal information about certain components of the state, at least probabilistically, but those same measurements will always destroy some information. See the No-cloning Theorem.

Right, but you would still get the basis states for all 127 qubits right? And that would be 2^127 output states. Yes, you could do some sort of search maybe to find highest probability outputs only, but if you needed every output value for a follow up algorithmic step (like in VQE for ground state prep wherein you keep using previous results to adjust the wavefunctions until ground is reached), then wouldn't it be a bit tough to use?

You have 127 qubits that you measure and you end up with a classical string of length 127. Sure, that classical string, the measurement result, could have ended being any of 2^127 possible different values into which the wavefunction collapses. But that is no different from saying that there are 1^1024 possible states that a 1kB of classical RAM can be in. It is not related to the (conjectured) computational advantage that quantum computers have.

Right okay makes sense...guess I am just too used to NISQ and having to run many thousands of shots for high enough fidelity..if all you wanted was one output, then yeah one classical string is easy enough, thanks

I hope US esp. but also our EU scientist friends eat everybody else's lunch. Better tech. Better science. Better math. Better algos. And, yes, let's get those details right too. Noise management is a key discriminator between POC and practical.

Reply on news.ycombinator.com