27
0本期嘉宾：1998年菲尔兹奖得主、微软研究院Station Q创始人、拓扑量子计算奠基人迈克尔·弗里德曼（Michael Freedman），因破解四维庞加莱猜想荣获菲尔兹奖，现致力于用代数模型探索数学与AI的深层关系。
【时间线】
00:06 数学的本质与压缩
01:28 压缩在数学中的重要性
05:16 数学库的统计分析：以mathlib作为研究模型
07:00 正式数学的爆炸性增长：指数级增长
15:48 宏观与微观的压缩
19:42 PageRank算法在数学中的应用
23:52 核心定义与常见定义的区别
27:34 压缩与智能的关系：压缩是智能在数学中的核心特征
【核心观点摘录】
“We found that the longest unwrapped statement in the Lean library had length ten to the one hundred and four, larger than the number of a googol. And it came from a wrapped statement of six hundred tokens. So six hundred went to googol.”
“When you see numbers like googol, the factor of a million — if our machines are a million times faster than us — a million is negligible on the scale of googol. So it isn't a question of what humans versus machines will explore. It's a question of what part of the formal world of deduction can be compressed in a way that we and our agents can understand it. And I'm calling that human mathematics, and I'm including our agents as humans.”
“Mathematicians and their agents are actually in the same boat. They have similar limitations to us. They're a factor of a million faster, maybe, but they still can't explore anything like by brute force. They have to have good intuition like we have, and we have to collaborate with them to develop intuitions.”
“If you're going to explore mathematics, you should know roughly what it looks like. Are you in the Appalachians or in the Sierras? What are the mountain ranges? What are the valleys? You should try to understand a little bit of the topography before you go in.”
“Mathematicians work with local compression. Global compression is uncomputable, it's too much, but there may be stuff in between. So it may be by studying compression carefully, we and our agents can explore new modes of thought, where we step a little bit beyond local compression.”
【适合谁听】
- 对AI时代数学教育感到焦虑的家长与学生
- 关心Polymath协作模式与开放科学的研究者
- 思考学术社区与科技产业关系的从业者
- 相信“热情胜过天赋”的学习者
- 所有对数学本质与智能边界好奇的人
