Meta's AI boss, Yann LeCun, made a splash after saying quantum computers are not that great. Speaking at a media event to mark ten years of Meta's AI team, he said the technology is "a fascinating scientific topic". Still, he was unsure of "the possibility of actually making useful quantum computers."
LeCun is not a quantum computing expert; other big names in the field also raise doubts. Oskar Painter, head of quantum hardware for Amazon Web Services, says there is a "tremendous amount of hype" in the industry right now and "it can be hard to tell the hopeful from the hopeless."
A big problem for today's quantum computers is that they make many mistakes. Some have said these so-called "noisy intermediate-scale quantum" (NISQ) machines could still work well. But Painter says that's not likely, and quantum error-correction tricks will be needed to make practical quantum computers.
The main idea is to spread information over unreliable qubits to make "logical qubits." But this could need as many as 1,000 dodgy qubits for each good one. Some have said that quantum error correction could be impossible, but that's not popular. Either way, making these tricks work at the size and speed needed is a long way off, Painter says.
"Given the remaining technical challenges in making a fault-tolerant quantum computer that can run billions of gates over thousands of qubits, it's hard to say when it will happen, but I would guess at least ten years away," he said.
In May, top Microsoft boffin Matthias Troyer penned a paper saying that quantum computers could only do better than regular computers in a few areas.
"We discovered over the last ten years that many things people have suggested don't work. And then we found some straightforward reasons for that."
The main point of quantum computing is to solve problems much faster than regular computers, but how much quicker depends. There are two things where quantum tricks seem to give a tremendous speed up, said Troyer.
One is breaking big numbers into smaller ones, which could crack the codes that keep the Internet safe. The other is copying quantum systems, which could help with chemistry and materials.
Quantum tricks have been suggested for optimisation, drug design, and fluid dynamics. But the speed-ups don't always work out--sometimes they are only a bit faster, meaning the time it takes the quantum trick to solve a problem is the square root of the time taken by the normal one.
Troyer says these speed-ups can quickly disappear because of the enormous amount of work quantum computers need. Running a qubit is much more complicated and slower than flipping a switch. This means that for more minor problems, an average computer will always be faster, and the point where the quantum computer takes the lead depends on how fast the normal one gets more challenging.
Troyer and his mates compared a single Nvidia A100 GPU against a made-up future fault-tolerant quantum computer with 10,000 "logical qubits" and gate times much faster than today's machines.
They found that a quantum trick with a bit of a speed-up would have to run for hundreds or thousands of years before it could beat a normal one on problems significant enough to matter.
Troyer said quantum computers will only work on small-data problems with huge speed-ups. "All the rest is nice theory but will not be useful," he said.
All this is pouring cold water on the idea that Quantum computers will be here soon or that the Internet is in danger of having its codes broken by thieves or spooks using the technology.
It would appear that, for now, the cat is still only potentially dead or alive.