PYTHON · NUMPY

Dequantize an advantage yourself, then find the one that survives

Build a speedup that looks like new-hardware magic, destroy it with a classical algorithm, and then measure the one advantage that no algorithm can take away.

01Challenge

Reproduce in miniature what has happened to quantum computing repeatedly since 2022, then apply the same attack to an energy claim and watch it fail.

Predict firstA new machine reports a large speedup on a real workload. What is the most likely outcome within three years?

02Model

A novel machine reports a result. It is impressive. Then somebody finds a classical algorithm that reaches the same place on ordinary hardware, and the advantage disappears. This has a name, dequantization, and it has been unusually productive: through 2026 it removed the claimed advantage from short-path algorithms for constraint problems and from quantum simulation of short-time local dynamics. The lesson is not that the hardware was fake. It is that a speedup is a claim about computational complexity, and complexity claims can be attacked by anyone with a laptop and a good idea. So the question worth asking of any new machine is which of its claims are of that kind, and which are claims about physics instead.

03Guided practice

1. Build the advantage. Estimate a property of a large matrix by sampling rows in proportion to their weight. Compare against reading every entry. The sampling method wins enormously.

2. Now dequantize it. The advantage came from privileged sampling access, not from exotic hardware. Give an ordinary classical algorithm the same access and it matches. Watch your advantage evaporate.

3. Attack the energy claim the same way. Count the operand fetches each version performs, and price them. Then notice what the smarter algorithm did not change.

04Feedback

PASS WHEN PASS when step 2 shows the classical method matching the 'novel' one, and step 3 shows a further energy gain that persists after the algorithmic advantage is gone.

If your run fails, check:

The point of step 2 is not the arithmetic; it is that the advantage lived in the access model, not the hardware. If that did not land, re-read what the two methods are each allowed to do.
In step 3, notice which factor survives a smarter algorithm. Anyone can reduce the number of operations. Nobody can make a fetch that happens cost less than a fetch that does not happen.

05Retrieve & space

What kind of claim is vulnerable to dequantization, and why?
Where did the advantage in step 1 actually live?
Give one energy advantage that a better algorithm cannot remove, and say why not.

06Mastery & project

You can classify a performance claim as algorithmic or physical, and predict which kind will survive contact with a cleverer opponent.

Take any efficiency claim you meet and sort it: is this fewer operations, or is it work happening somewhere else? The first is provisional. The second is a statement about the arrangement of matter.

← The receipt, and why grades never blend Back to course →

PYTHON · NUMPY

Dequantize an advantage yourself, then find the one that survives

You will construct a speedup, break it, and then find the claim that does not break.

You read

the arrays and values already in scope

You change

the code you write in each cell

Fixed

the dataset and the checks that grade you