Report: GPT-6 and Claude Assist Claimed Sompolinsky Lyapunov-Spectrum Derivation
A Chinese-language report published by 36Kr says theoretical neuroscientist David Clark used GPT-6 Astra to work on the complete Lyapunov spectrum of the Sompolinsky 1988 model. The report says GPT-6 ran independently for about 100 minutes before producing an initial derivation, whose solution it says matched supercomputer simulations. Claude Opus 5.5 then assisted Clark and the AI models in analyzing and reformulating the derivation into a human-readable result. The report says the final work reproduced the original Schrödinger-equation result and links to an arXiv paper.
The report presents AI systems as collaborators in a claimed analytical advance on a longstanding problem in theoretical neuroscience. If the derivation is validated, the work could provide a new way to characterize chaotic dynamics in the Sompolinsky model, a model used to study cortical activity and recurrent neural networks.
最近，理论神经科学家David Clark宣告成功解析了Sompolinsky 1988模型的完整李雅普诺夫谱。
链接：https://arxiv.org/abs/2610.12426
如果你对这些干涩的学术名词感到陌生，没关系。你只需要知道两件事：
第一，这是理论神经科学领域近四十年来最难啃的骨头，甚至被普遍认为是「解析上绝对无解」的死局。
第二，这次完成破局的核心主力，不是人类，而是GPT-6 Astra。而负责打辅助收尾的，是Claude Opus 5.5。