Abstract:
According to Business Insider, the Physical Superintelligence Laboratory (PSI) said that it used an undisclosed Anthropic model to find a proof of the "Aronson-Ambanis Conjecture" in the field of quantum computing within three days. This conjecture has been unresolved since it was proposed in 2008, and involves the circumstances under which quantum computers can be significantly faster than traditional computers.
PSI has disclosed the paper and proof code, and said that the proof passed the test of the mathematical proof auxiliary tool Lean. Stanford academics who have previously studied this issue have read the proof and provided opinions, but the results still await further independent review by the academic community.
The system used in this study was developed by PSI technical staff member David Turturean and ran on the laboratory's own infrastructure, using a model that Anthropic has not yet made public. PSI did not disclose the specific model name. Business Insider said Anthropic did not respond to a request for comment.
PSI stated that the research system gave a complete proof on the third day of operation, and subsequently improved the results twice, and the final version also passed the Lean test.
According to the project description disclosed by PSI, mathematical statements, first drafts of papers and formal proofs are all generated by AI models. Turturian reviewed the paper and takes responsibility for its correctness.
The Aaronson-Ambanis conjecture was proposed by computer scientists Scott Aaronson and Andris Ambainis, focusing on the boundaries of the advantages of quantum computing.
PSI introduced that if this proof is true, for a specific alternative calculation problem, on most random inputs, the number of times a quantum computer reads the input cannot obtain an exponential advantage compared to a traditional computer. This means that there are some significant quantum speed-up dependencies or special structures in the input. What is discussed here is the number of times the algorithm reads the input, which cannot be directly equated to the total running time of the real device.
This conclusion does not mean that quantum computing loses value. For example, PSI said that the quantum algorithm proposed by Peter Shor in 1994 takes advantage of a specific structure in the problem and is much faster than known traditional computing methods in decomposing large integers.
PSI said that Stanford University researchers Guy Blanc, Jordan Docter, Carmen Strassle and Li-Yang Tan had previously solved related problems in a special case. The four are reading PSI's full proof and preparing to write a more concise explanation for professionals. They described this as a "significant result" in their feedback to PSI.
State of Proof, an independent research tracking platform, stated that the Lean inspection records disclosed by PSI provided material for the review, but the platform has not independently reproduced the inspection, nor has it confirmed that the research has completed journal peer review.
Comments