A proof worth examining.
Some of the most useful scientific progress starts with a better question. For Scott Aaronson and Andris Ambainis, one question was how much of quantum computing’s advantage depends on special structure in a problem’s input.
On November 6, Physical Superintelligence (PSI) announced a proof of the conjecture bearing their names. The company says an internally developed research system, using a non-public Anthropic model, found a complete proof during the third day of a research run. It subsequently improved the result.
The development gained fresh independent context on November 8, when Aaronson discussed the proof on his own blog. He says he believes it is correct, pointing both to its Lean certificate and to experts who have examined and understood it. That makes this more than a company’s announcement of an interesting answer: there is an argument for other mathematicians to study.
Two ways to check.
Lean is a proof assistant: software that checks whether a formal mathematical argument follows its stated rules. PSI reports that Lean accepted both the initial proof and its final improved version, and links a manuscript and proof code from its announcement.
Human understanding matters alongside that computer check. Aaronson points readers to an expert exposition by Guy Blanc of the Massachusetts Institute of Technology (MIT) and Jordan Docter, Carmen Strassle and Li-Yang Tan of Stanford. He also says he has not yet understood the proof himself—his confidence draws on the certificate and those experts’ work, rather than his own completed verification.
PSI credits David Turturean with developing and running the research system, reviewing the paper draft and taking responsibility for its correctness. Jim Halverson audited the proof, while Phiala Shanahan commented on its Lean formalization. The published account describes a research partnership, with identifiable human responsibilities.
What the result explains.
The key measurement is how many times an algorithm needs to inspect its input. A classical computer reads individual bits; a quantum query can involve a superposition of input positions. Researchers use this setting to study what kinds of problems permit especially large advantages.
As Aaronson explains, the conjecture’s proof rules out a superpolynomial quantum query advantage for yes-or-no problems that works on most random Boolean inputs. In plainer terms, a classical method can estimate the quantum algorithm’s answer probability for most such inputs without an exponentially larger number of input queries.
That is a result about queries, not a general claim about total running time. It leaves established avenues such as Shor’s factoring algorithm and quantum simulation of physics and chemistry intact. The benefit is a more precise map of where researchers should look for quantum advantages.
A useful research partnership.
For SI-assisted science, the valuable pattern is the combination: generate an argument, make its steps checkable, and let knowledgeable people examine what it means. A confident-sounding answer alone cannot do that job.
Our reading: this is a constructive example of SI helping researchers push beyond routine assistance. The proof builds on earlier human mathematics and leaves a public trail for further study. That gives other researchers something durable to work with—and a clearer starting point for the next question.
