What "quantum advantage" actually claims

The claim that shrank by a factor of a million

A machine performs a task in 200 seconds. The announcement says the best supercomputer on earth would need 10,000 years to do the same thing. The ratio is about one and a half billion to one, and it goes round the world.

Some months later, a group of researchers publishes a smarter classical method. On ordinary hardware, the same task takes a few days.

Now the ratio is about 1,300 to one.

Nothing about the quantum machine changed. Nobody found a flaw in it, nobody retracted the run. The number moved by a factor of a million because the number was never a property of the machine. It was a property of the comparison.

This is the single most useful thing to know about quantum computing claims. A speed-up is a ratio, a ratio has a denominator, and the denominator is somebody’s best effort — which someone else can improve on any Tuesday.

Three different claims wearing the same word

Headlines use “quantum computer beats classical” for three claims that mean wildly different things.

One — a demonstration of advantage. The machine did something faster than the best known classical method. Note what is missing: any suggestion the thing is useful. These demonstrations usually pick a task precisely because quantum hardware is naturally good at it and no one needs the answer. Sampling the output of a random circuit is the classic example — it is, essentially, asking the machine to be itself, and asking a classical computer to imitate it.

This is a real scientific milestone. It is evidence that the hardware does what the theory says. It is not a product.

Two — a useful result, obtained faster. The machine did something someone actually wanted, better than the ordinary alternative. This is the claim that would matter, and it is far rarer than the coverage suggests. When you meet one, the question is whether the classical comparison was done seriously — by people who wanted it to win.

Three — a proof about scaling. The famous factoring result is of this kind. It says that as the numbers get bigger, the quantum cost grows far more slowly than the best classical cost. It is a theorem. It is not a claim that any machine has factored anything interesting, and it never was.

Confusing the third for the second is the most common error in the whole subject. A proof about how something scales tells you nothing about where the crossover point sits, or whether any machine will ever reach it.

The four questions

When a claim arrives, these four get you most of the way:

1. What was the task, and does anyone need the answer? If the task was chosen to suit the hardware, that is fine and worth saying out loud.
2. What is the comparison? The best known classical algorithm, or the obvious one? Run properly, on good hardware, by people trying to win?
3. Physical or logical qubits? From the last item: a machine of a few hundred noisy qubits and a machine of a few hundred error-corrected ones are separated by a factor of about a thousand in hardware.
4. How many shots, and what checked the answer? From item 6: if there is no cheap classical checker, the samples are not results.

The asymmetry worth naming

There is a structural imbalance here that explains a lot of the field’s odd rhythm.

To strengthen a quantum claim you must build better hardware — years of work, enormous capital, physical limits. To weaken one you need only find a better classical algorithm, which is a person with a good idea. So classical baselines improve in sudden jumps, usually after an announcement, and the ratio quietly falls.

This is not embarrassing and it is not bad faith. It is what a healthy field looks like: a claim is made precisely so that others can attack it. But it means the honest way to read any single announcement is as a provisional statement about the current state of two racing efforts — never as a fact about a machine.

Where you have seen this before

The pattern generalises past physics, and it is worth carrying.

Any claim of the form “X is N times better than Y” is really a claim about Y, and Y is usually chosen by whoever is making the claim. A drug against placebo or against the best existing treatment. A model against last year’s model or against a careful person. A policy against doing nothing or against the obvious alternative.

The number is rarely the interesting part. The denominator almost always is.