一万代理 88 小时攻 Navier-Stokes 问题
称上万代理并行交换 2.7M 消息、产出 130B token，88 小时给出候选证明，再由 Astra 用 Lean 花 17 小时核查部分论证。
10,000 AI agents reportedly attacked a 90-year-old Navier-Stokes problem in just 88 hours The scale of the experiment is the real story Thousands of agents worked in parallel inside one research environment, exchanging around 2.7 million messages and generating roughly 130 billion tokens The system reportedly produced a candidate proof in 88 hours GPT-6 Astra then spent another 17 hours checking parts of the argument programmatically with Lean The claimed result concerns whether certain fluid conditions can produce singular behavior, where velocity grows without bound That is the kind of problem mathematicians have struggled with for decades Whether the proof ultimately survives full expert review is the part that matters most But the workflow itself is already significant AI is moving from helping one researcher write or calculate faster to running entire parallel research organizations that can explore thousands of ideas at once
这条是引用着另一条写的：x.com/egorjioo/status/2099496479895310403