AI
Superposition: The Exponential Advantage
Superposition lets a quantum system exist in many configurations simultaneously. Hartmut Neven explained this with a thought experiment: the audience sitting in one configuration could switch seats, creating different arrangements. Quantum physics says all these configurations are realized and evolve together. A qubit is a quantum bit placed into superposition. The Willow chip holds 105 qubits. In one clock cycle, it touches 2 to the 105th power different bit strings. That computation would take a classical supercomputer 10 septillion years. This exponential scaling is the source of quantum computing’s power.
Error Correction Crosses Below Threshold
Google’s Willow chip, unveiled in 2024, demonstrated two breakthroughs. First, it completed in minutes a calculation that would take the world’s fastest supercomputer 10 septillion years. More important was the second: below-threshold error correction. Quantum error correction uses redundancy, multiple physical qubits form one stable logical qubit. Earlier attempts introduced more hardware and raised error rates. In 2022, errors fell for the first time but by only a few percent. Willow cut errors by a factor of two. The field recognized that error correction works.
From Benchmarks to Useful Computation
Critics asked for useful computations, not benchmarks. Quantum echoes answered. It is a quantum machine learning algorithm that learns from nuclear magnetic resonance data. NMR machines are used by pharmaceutical companies and hospitals for MRI scans. The algorithm extracts more information from these devices. Researchers at Google and Berkeley applied quantum echoes to biphenyl, a molecule with two benzene rings. Chemists did not know the rotation angle between the rings. The quantum algorithm computed it, answering an open question in chemistry.
Quantum and AI Accelerate Each Other
The team is named Quantum AI for a reason. Hartmut Neven argued that future AI needs the most powerful computational substrate available. Quantum computers can produce training data that classical machines cannot. AlphaFold needed 50 years of protein database collection. Quantum computers could generate valuable training sets for materials science much faster. Last week, the team shipped the first quantum-generated data sets to DeepMind. Conversely, AI accelerates quantum progress. DeepMind’s AlphaQubit helped improve error correction. The relationship runs both directions.
Cryptography Deadlines Move Closer
Quantum computers can crack RSA and elliptic curve cryptography. In 2019, the consensus required 20 million qubits to break RSA-256. A year ago, algorithm advances lowered that to one million qubits. Two months ago, Greg Kitney showed that elliptic curve cryptography needs only a few hundred thousand physical qubits. The road map originally set 1 million qubits as milestone six. Now Google believes 100,000 qubits could produce commercial impact. The company advised moving to post-quantum cryptography by 2029.
Notable Quotes
this tiny chip can do certain computations that the largest factorysized data center would need an inordinate amount of time 10 septillion years Hartmut Neven · ▶ Watch (7:43)
a tiny wormhole was actually created Hartmut Neven · ▶ Watch (28:03)
we need to move to more powerful postquantum crypto schemes already by 2029 Hartmut Neven · ▶ Watch (31:56)
quantum AI is a more powerful form of AI Hartmut Neven · ▶ Watch (37:55)
Key Takeaways
- Superposition lets 105 qubits process 2^105 states in one clock cycle.
- Willow chip cut error rates by a factor of two below threshold.
- Google advises migrating to post-quantum cryptography by 2029.