MODULE 1 · LESSON 1
Free — no login requiredSign in to track progress, save quiz attempts and enrol in the full course.
Sign in to track progress / enrolThe Two Hundred Second Argument
In October 2019, Google published a paper in Nature announcing a result its authors called quantum supremacy.
Their processor, Sycamore, had 53 working qubits. It had performed a specific calculation in 200 seconds. Google estimated that Summit, then among the most powerful supercomputers in the world, would need approximately 10,000 years to produce the same result.
The framing was irresistible. Two hundred seconds against ten thousand years. It went around the world in a day.
IBM's response
Within days, IBM published a rebuttal, and it did not dispute that Sycamore had done what Google said it did. IBM's objection was to the comparison.
Google's 10,000 year estimate assumed a particular method of simulating the problem classically. IBM argued that a better method, using the vast disk storage available on Summit rather than memory alone, could produce the same result in about 2.5 days.
Ten thousand years against two and a half days. The same experiment, the same classical computer, a factor of roughly a million and a half between the two claims.
Subsequent work by other researchers narrowed the gap further, with improved classical algorithms bringing the simulation down further still.
Why both sides had a case
Here is the resolution, and it is the most useful thing in this lesson.
Google was accurately reporting what its machine did and comparing it against the best classical method known to them at the time. IBM was pointing out that the classical baseline was not fixed, and that a cleverer classical approach existed.
Neither was being dishonest. They were making different claims.
A quantum advantage claim is never a statement about the quantum computer alone. It is a statement about the gap between a quantum machine and the best classical method anyone has thought of yet, and the second half of that comparison keeps improving.
This pattern has repeated many times since. A quantum result is announced, and within months a classical algorithm appears that narrows or closes the gap. The phenomenon is common enough to have a name in the field: dequantisation, the discovery that a problem thought to need a quantum computer can be handled classically after all.
That is not a scandal. It is how the field makes progress, and it is why the vocabulary shifted. The term quantum supremacy fell out of favour, partly for its unfortunate connotations and partly because it implied a permanence the results do not have. Quantum advantage is now the preferred phrase, and careful people qualify it further: advantage on which task, against which classical method, with what error rate.
The task itself
One more detail matters, and it is routinely omitted from coverage.
The calculation Sycamore performed was random circuit sampling. In plain terms, the machine was given a random sequence of quantum operations, ran them, and reported the distribution of outputs. Verifying that the distribution was correct is itself extremely hard classically, which is precisely what made it a good demonstration.
It is also useless. Nobody wanted that distribution. The task was chosen specifically because it is hard for classical computers and natural for quantum ones, which makes it an excellent benchmark and not an application.
This distinction runs through the entire field and is worth holding onto:
- A benchmark shows that the hardware does something measurable that classical machines find difficult.
- An application solves a problem somebody actually has.
The 2019 result was a genuine and important benchmark. It was not a quantum computer doing anything useful, and no serious participant claimed otherwise. The gap between those two things is most of what the rest of this course is about.
This field produces a steady stream of impressive sounding results. Four questions separate the substantial from the promotional, and none of them require technical expertise.
What was the task, and does anybody want the answer? Random circuit sampling is a benchmark. Simulating a molecule somebody is trying to synthesise is an application. Both are legitimate, and press coverage frequently blurs them. If the answer produced is of no interest to anyone, you are looking at a hardware demonstration.
What is the classical baseline, and how hard did they try? A comparison against a naive classical method is much weaker than one against a well optimised specialist implementation. The Sycamore case is the canonical illustration. If the announcement does not say what it was compared against, the comparison is not yet meaningful.
Are the qubits physical or logical? This is the single most important distinction in modern announcements and is covered fully in Module 4. A thousand physical qubits and a thousand logical qubits differ by orders of magnitude in capability, and headlines rarely distinguish them.
What was the error rate, and how deep was the circuit? A processor with many qubits that can only perform a few operations before noise overwhelms the result is limited in a way a qubit count does not convey. Depth matters as much as width.
Add a fifth for commercial claims specifically: was the quantum part actually doing the work? Many announced results use a hybrid approach where a classical computer does most of the computation and a small quantum step is included. Sometimes that step is essential. Sometimes removing it entirely changes nothing, and researchers have repeatedly demonstrated exactly that by testing the classical portion alone.
Google reported a task taking 200 seconds that it estimated would take a classical supercomputer 10,000 years. IBM argued the same task could be done classically in about 2.5 days. What is the best interpretation?
Quantum advantage
Click to flipA quantum computer outperforming the best known classical approach on a specific task. Always relative to a classical baseline that keeps improving.
Click to flip backThe famous 200 seconds against 10,000 years claim was answered within days by a classical method taking 2.5 days, and both sides were defensible, because a quantum advantage claim is always a claim about a gap whose classical side keeps moving. The task itself, random circuit sampling, was chosen precisely because it is useless: an excellent benchmark and not an application. Ask four questions of any announcement: what was the task, what was the classical baseline, were the qubits physical or logical, and what was the error rate at what circuit depth.