Google 让 LLM 智能体团队用 Lean 证明量子程序正确性，量子算术成本降低 290 倍
Google 把 LLM 智能体和 Lean 形式化验证结合来做量子算法优化，不靠模拟打分而是让 Lean 内核直接证明正确性。量子算术成本降了 290 倍，聚变模拟提速 11 倍，做量子计算或形式化验证的都该看看这套工作流。
Google 发布一篇量子 AI 研究论文，让多组 LLM 智能体编写量子程序并用 Lean 定理证明器验证正确性。经典计算机只能在小规模实例上测试量子算法，模拟打分的搜索循环无法触及量子计算真正有用的规模，因此研究团队改为把规格说明和成本目标写成 Lean 定理。
工作流程是：智能体团队提出候选程序，Lean 内核检查正确性与成本，另一个独立的比较器对所有通过验证的答案做二次复核。各团队在共享 wiki 上记录结果和失败尝试，并定期开会交流想法。
结果上，验证过的程序把量子算术的成本相对此前最佳基线降低了 290 倍；与核聚变、光化学和药物代谢相关的模拟分别提速 11 倍、8 倍和 2.5 倍。
Banger paper from Google on quantum AI.
They gave teams of LLM agents the ability to write quantum programs and prove them correct in Lean.
Classical computers can only test quantum algorithms on small instances. A simulation-scored search loop can't reach the sizes where quantum computing becomes useful.
So researchers write a specification and a cost target as a Lean theorem. Agent teams propose programs, the Lean kernel checks correctness and cost, and a separate comparator re-verifies every accepted answer. Teams share results and dead ends on a wiki and meet at scheduled conferences to exchange ideas.
The verified programs cut the cost of quantum arithmetic by 290x against the best prior baseline. Simulations relevant to fusion, photochemistry, and drug metabolism improved by 11x, 8x, and 2.5x.
https://t.co/DCG6qWftVM
Chat with Paper: https://t.co/8Ftdbl74oe
This report is categorized under AITOP's "Papers" desk, focusing on US frontier AI labs and Silicon Valley technology leaders. Within our cross-regional analytical framework, it directly maps to AI developer toolchains, coding agents, and engineering workflows. Key entities and technical topics under continuous tracking include Google, LLM智能体, Lean, 量子计算.
Source attribution: curated from DAIR.AI (@dair_ai).
