> Prerequisites: You need episode 005 (the exponential misunderstanding) and 004 (interference). > Next episode preview: 010, "Drawing a Decision Tree for Quantum Computing." > On-site examples: Jiuzhang 4.0 and IBM's 70 logical qubits are samples of two different "advantage" narratives — come back to them after reading.
Every few months, a headline goes viral: "Team X announces: quantum computer beats world's most powerful supercomputer!" The comment section instantly splits — one side hails the revolution, the other denounces yet another scam. Both sides are arguing over one term — quantum advantage — which is actually a grandstand holding three different tickets. Once you tell these three tickets apart, you'll know where to stand in the next news battle.
Three Tickets: Three Meanings of "Advantage"
Ticket one: theoretical advantage. Mathematically proven: for some class of problems, a quantum computer has an algorithm that a classical computer *in principle* cannot catch up with (the complexity class BQP, covered in episode 081). Note the wording — "in principle cannot catch up" is a theorem-level statement about the mathematical structure of the universe, regardless of whether any machine has been built. Shor's algorithm for factoring large numbers belongs here — but note: its experimental demonstrations remain at toy scale (factoring 15 into 3×5); the big loom hasn't been woven yet.
Ticket two: experimental advantage. On one specific, well-defined task, a real quantum machine beat the strongest classical simulation of its day. This is the protagonist of the last decade's news. It's more grounded than ticket one, but has a fatal weakness — the opponent is alive: classical algorithms keep improving, and the "quantum won" record keeps being snatched back by better classical simulations (episode 082 covers this tug-of-war).
Ticket three: practical advantage. A quantum machine is faster or cheaper than classical approaches on problems the real world cares about (new drugs, new materials, finance) — enough that someone is willing to pay for it. As of this writing, nobody holds this ticket. All the "quantum computing will change your life" news describes expectations, not the present.
The tickets cost progressively more, and are worth progressively more. Identify which ticket a headline holds, and most of the argument ends.
The Man-Made Track Called "Sampling"
The default track for ticket two is a task called random circuit sampling: let a quantum chip perform a random sequence of operations, then ask a classical computer to "please predict / reproduce the output distribution of this random device." Photonic machines of the Jiuzhang series and Google's Sycamore have both claimed supercomputer-beating advantages on this track — Jiuzhang 4.0's claim that "a supercomputer would need hundreds of millions of billions of years, it takes minutes" is a stopwatch reading on this track.
Why call it "man-made"? Two reasons:
- The task itself is useless. The output of the sampling task is a pile of semantically meaningless random numbers that nobody needs. Its only reason to exist is precisely that quantum devices are naturally good at it and classical simulation is naturally bad at it — a gymnastics move designed specifically for the quantum contestant.
- It compares both sides' strongest versions. The classical score isn't carved in stone: simulation methods (tensor networks) advance in leaps every two years, repeatedly compressing "billions of years" back to "hours," and the quantum side can only push the record back with bigger chips. Every advantage announcement you see is a dynamic scoreboard, not a verdict.
- Disanalogy one: Bolt's speed is a stable physical fact; the stopwatch never upgrades. But the classical side's "stopwatch" is a living algorithm that gets faster every generation — in this race, the timer itself is evolving.
- Disanalogy two: The track is an international standard; the sampling track is a problem chosen by the quantum side. A win on someone else's home turf, on their rules, deserves a discounted trophy.
To be fair, this track isn't meaningless — it's like doping tests in sports: every advance by the opposition (classical simulation) stress-tests the home side (quantum hardware) and squeezes out the exaggeration. But between "beating a supercomputer on a useless task" and "solving useful problems" lies an entire unconquered continent of materials science, chemistry, and finance.
Why "Proof" Is So Hard
The difficulty of pinning down advantage lies on the classical side: proving "classical can never catch up" requires hard theorems from complexity theory — and for sampling tasks, such theorems are exceptionally elusive (you're fighting the set of all the strongest simulation algorithms, and you cannot enumerate every clever idea of the future). So in reality, every "experimental advantage" is a scoreboard advantage: the lead is a fact; a permanent lead is not.
This also explains a seemingly odd industry phenomenon: responsible teams announce advantages with careful wording ("on a specific task," "against the strongest classical simulation as of now"), while marketing copy simply writes "crushed" — the same fact, two phrasings, differing only in whether you've been given the three-ticket knowledge.
Taking the Analogy Apart
Where the 100-meter-dash analogy breaks down:
What to Take Away
1. Advantage comes in three tickets: theoretical advantage (theorem-level; Shor belongs here but experiments are stuck at toy scale), experimental advantage (a real machine wins a specific task, but the classical opponent evolves and the score is dynamic), and practical advantage (real problems plus real money — as of writing, nobody holds it). 2. The sampling track (Jiuzhang, Sycamore) is the template for ticket two: a useless task, tailor-made for quantum, with a dynamic scoreboard — only by understanding the track can you understand the news. 3. To judge any advantage claim, ask three questions first: which ticket? what task? which version of the classical competitor?
Hook for the next stop: the concepts are all in place, so it's time to hand you a portable decision tree — for any "can quantum computing do X" question, which forks to walk down. Episode 010, the finale of the first batch.