Google recently unveiled its Willow quantum chip, claiming it achieves "beyond classical computation" by completing a random circuit sampling (RCS) task in under five minutes—a task that would take classical supercomputers an estimated 10 septillion years.
While RCS benchmarks lack practical applications, Google argues they are foundational for assessing quantum capabilities.
More practically, Google tries to make the case that RCS performance should be the metric by which all quantum computers are judged. According to Hartmut Neven, the founder of Google Quantum AI, "it's an entry point. If you can't win on random circuit sampling, you can't win on any other algorithm either." He adds RCS is "now widely used as a standard in the field."
However, other companies, including IBM and Honeywell, instead use a metric called quantum volume to tout their breakthroughs. They claim it points to a more holistic understanding of a machine's capabilities by factoring in how its qubits interact with one another. Unfortunately, you won't find any mention of quantum volume in the spec sheet Google shared for Willow, making comparisons difficult.
The true breakthrough lies in Willow's reduced error rates as more qubits are added, marking progress toward scalable, practical quantum systems. However, critics highlight the absence of comparative metrics like quantum volume, leaving questions about its real-world impact.
Read more here
(Score: 3, Touché) by pTamok on Friday December 13, @09:40AM (1 child)
Shtetl-Optimized: The blog of Scott Aaronson: The Google Willow thing [scottaaronson.blog]
He links also to Sabine Hossenfelderon Chitter [x.com]
and has some criticism of her statement above.
Quantum computing will remain 'meh', until it isn't. Much like practical nuclear fusion.
(Score: 2) by driverless on Saturday December 14, @04:37AM
Why does this surprise anyone? This is exactly how advances in quantum work, you manipulate the benchmark or measure in order to make it seem like you've achieved something. This is why you see quantum supremacy announced every six months or so, over and over again.
(Score: 3, Insightful) by bzipitidoo on Friday December 13, @05:35PM
AI, Quantum Computing, memristors, cold fusion, facial recognition, self-driving cars ...
Unless there's some weird physical principle that forbids large scale quantum coherence, seems we will eventually have quantum computing at a usable scale. In 20 years, maybe? Lot of reports of small scale success.
I guess AI that actually deserves the 'I' moniker will also eventually come about, but much later, 100 years minimum. I think we've grossly underestimated how hard it is to achieve intelligence. A considerable amount of SF has intelligence just happening, accidentally. Skynet somehow just became intelligent. V'Ger (and Nomad) too, but at least it was hinted that aliens may have helped them become intelligent. Part of the problem is deciding just what constitutes intelligence. I rate the average insect as way smarter than ChatGPT. How does ChatGPT learn? Can a user teach ChatGPT? No!