AI Model Solves Long-Standing Quantum Computing Conjecture
Researchers at Physical Superintelligence (PSI) report that an unreleased AI model developed by Anthropic has solved an 18-year-old conjecture in the field of quantum computing. According to reports circulating on social media platforms such as Reddit and X, the AI research system produced a proof in just three days, addressing a mathematical problem that has remained unsolved since it was first proposed by two mathematicians nearly two decades ago.
The proof, which has been formally verified using Lean—a mathematical software designed to check the correctness of complex proofs—establishes a fundamental limit on quantum speedups. Specifically, the findings suggest that quantum computers cannot achieve significant speed advantages when processing information that is essentially random.
Implications for Quantum Advantage
The resolution of this conjecture provides formal confirmation for a long-held hypothesis in the scientific community. While it was widely suspected to be true, the mathematical proof had remained elusive.
Current quantum algorithms that provide exponential speedups typically rely on exploiting specific structures within data. In contrast, unstructured searches are already known to offer only polynomial speedups, as demonstrated by Grover’s algorithm. The new proof essentially generalizes this principle, confirming that without underlying structure, the massive computational advantages typically associated with quantum computing are limited.
Verification and Technical Context
The technical details of the breakthrough were shared via PSI’s research channels, noting that the proof was validated through formal verification. This process uses software to ensure that the logical steps of a mathematical argument are flawless, a method increasingly used to verify deep research results across various scientific disciplines.
While the results have gained significant attention in online technical communities, it is important to note that the specific Anthropic model used in this research remains unreleased. Consequently, the full capabilities and the exact architecture of the system used by Physical Superintelligence have not been independently verified by third-party benchmarks.
The breakthrough highlights a shift in the application of large language models, moving from general text generation toward specialized scientific reasoning and formal mathematical verification. As researchers continue to explore the boundaries of “physical superintelligence,” the ability of AI to resolve complex conjectures in a matter of days rather than decades marks a notable development in computational mathematics.