Physical Superintelligence says its AI research system helped solve five open problems in quantum information theory
Today Physical Superintelligence PBC is announcing solutions to five long-standing problems in quantum information theory. t.co/LjmLz0mCtF Thread 👇
Context
It adds a second lab claiming AI assisted results on recognized open problems on the same day as OpenAI, a sign that AI driven discovery is spreading beyond the biggest companies. The results are self reported by PSI and have not yet been reviewed by outside physicists.
Physical Superintelligence PBC, known as PSI, describes itself as an AI native laboratory for physics research. Its research page, dated October 23, lists five results on problems from a public list of open quantum problems, each with a paper, a Lean formalization and verification code. PSI announced the batch on X at 7:44 PM Eastern on October 26.
Deep dive
Physical Superintelligence PBC announced on October 26 that its researchers, working with the company's AI research system, obtained new results on five long standing problems in quantum information theory. PSI says each result comes with a paper, a formalization in the Lean proof assistant and verification code, and that its release repository documents how AI was used in each case. In its thread, PSI wrote that its scientists directed the work and verified the results.
The five problems come from a numbered public list of open quantum problems. For problem 2, PSI says Bell pairs cannot be distilled from three copies of certain Werner states over a stated parameter range in every dimension of three or more, while the question for arbitrarily many copies stays open. For problem 12, it says two fixed regions in the vacuum of a massive free scalar field cannot violate the CHSH Bell inequality beyond a finite separation, even though they stay entangled.
For problem 27, PSI says Fourier phase measurements give the largest CGLMP Bell violation for a maximally entangled pair, and that every optimal strategy is equivalent to one of them. For problem 38, known as the PPT squared problem, it says any composition of 2(d minus 1) PPT maps on a d dimensional system breaks entanglement. For problem 43, it reports a counterexample showing that some causal processes have no finite dimensional reversible description.
PSI's page notes that OpenAI had earlier reported, in a preprint, that two PPT maps are not enough in general.
What to expect next
The results now need checking by quantum information researchers, who can rerun PSI's Lean formalizations and verification code. Watch whether the maintainers of the open quantum problems list mark any of the five as solved, and whether the papers are submitted to peer reviewed journals. Because OpenAI announced its own physics and math claims the same day, expect more debate over how to credit and verify AI assisted proofs, and whether labs like PSI publish more batches.