IBM quantum computer solves a classically intractable problem in 15 minutes

[Walter]

IBM and University of Chicago researchers completed a quantum computation that leading classical computing methods couldn't practically reproduce, using 70 error corrected logical qubits to finish the task in roughly 15 minutes. The team developed a new, more structured alternative to the standard random circuit sampling benchmark specifically designed to let errors be detected during the actual quantum computation, addressing the core problem that once a calculation becomes too hard for classical computers to check, verifying the quantum computer got it right becomes nearly impossible too

The experiment carried out 2,415 logical two qubit operations and 468 logical T gates, with effective logical error rates coming in 10 times lower than the underlying physical error rates, representing one of the largest demonstrations of logical quantum computing reported so far. IBM Research Director Jay Gambetta said the milestone establishes, with statistical confidence, a lower bound on how faithfully the computation was executed, giving researchers and businesses a new foundation for trusting quantum computers as they scale toward problems beyond classical reach

University of Chicago's Bill Fefferman noted that verification remains one of the biggest challenges in firmly establishing experimental quantum advantage, and that this specific technique increases confidence the quantum computer is genuinely solving a computationally hard problem rather than just producing noise that happens to look plausible. The circuits and results have been made publicly available through the Quantum Advantage Tracker. Curious what people think this actually means practically now that both speed and trustworthy verification have been demonstrated together

IBM quantum computer solves classically intractable problem in 15 minuteswww.sciencedaily.com

[Luke_67]

Combining raw speed with actual trustworthy verification in the same single experiment is genuinely the harder and more important achievement here, since a fast answer that can't be trusted was never actually all that useful to begin with

[SpikeDudley]

That 10x reduction in logical error rate compared to physical error rate is honestly the number that matters most for anyone tracking whether error correction is genuinely working as intended at meaningful scale