Alan Xiao

math @stanford
building perps @meconemarkets
research @ Stanford ASL

[Alan Xiao @AlanXiao67]

@kkaitlyn111 @VivianCaiIAm @alexhuang06 Agreed, @alexhuang06 would be a great fit for this

[Alan Xiao @AlanXiao67]

@subha_v1 @ScalingIntelLab @kkaitlyn111

[Alan Xiao @AlanXiao67]

@kvcache_ @togethercompute LETS GO!!!

[Stan Lee (founder) @stanleefounder]

sf is insane

2 guys i know co-founded a company

working 140 hour weeks

both wanted to start dating but didn't have time to meet women

they made the efficient choice and just started dating each other (even though they're not gay)

other founders could learn from this level of commitment

[Zhipeng Wang 🇺🇦 @PKUWZP]

To the best of my respect (I like Keller's Muon Optimizer work a lot), I do not think it's the best action for a person who barely published papers in top venues to trash AI top conferences. In fact, the community should do everything to make those AI conferences better. It's our last hope to get a place to share and open-source AI advancements. Otherwise, we will be living in the dark dominated by close-sourced AI Labs.

[Alan Xiao @AlanXiao67]

@0xtigs Nah it’s:
*Discovery Loop
*SpaceXTeslaAI
*SSI
*Deepseek

[Alan Xiao @AlanXiao67]

@jasondeanlee Almost

[Myk is Walking Backwards @mykola]

me, begging, crying, on my knees: "Please just use plain english, I don't understand what you're saying."

Claude: "The right fix, and the book's lesson applies: the tutorial rotted as a front door because a front door full of claims always rots. So the new root holds only what's timeless — the thesis and three doors — and every claim lives behind it in the thing that goes red when stale."

[Daniel @growing_daniel]

It's insane that Y Combinator is listed as a signatory on the open letter but rival incubator Denny's is not. Denny's has done far fewer companies but the market cap of their incubator portfolio is approximately 8x the size of Y Combinator's ($5T vs $600B)

[mike @mikeisaspider]

economy has me doing crazy things man

[Daniel Litt @littmath]

So, you (or the model you've prompted) has proved or provided a counterexample to an open conjecture! What's next?
- Check if it's correct!
- Try to understand what new ideas, if any, have been introduced. Do they have any further applications? How are they related to the rest of math?
- Try to understand different PoV's on the argument. Can you digest/illuminate it further?
- Why is it interesting? What does it tell you about the objects in question? Does it raise any new questions?
- Start trying to solve those new questions!

[amber🍥 @amber_digit2]

chinese mathematician wins fields medal, cites yuri manga as his greatest source of inspiration

[Guanyang Wang @GuanyangW]

I have no doubt math won’t be what it was yesterday. But I think we might still be in the branching middle of the game, where the endings aren’t clear yet. Three questions:
1. Is there a meaningful distinction between Erdős-style math (brilliant use of existing tools) and Grothendieck-style math (building new theories until the original problem becomes obvious)?
2. If yes, can LLMs’ math ability extrapolate from the former to the latter? The reward for building a framework arrives much more slowly.
3. How fast can humans adapt?

[Phoenix Yin @Phoenixyin13]

我可以告诉你的是，冯诺伊曼能力值强于当代任何一个顶尖理工科博士+目前的最强的AI（包括Kimi K3，Claude Fable 5）。

当年的洛斯阿拉莫斯国家实验室，科学家们用早期的电子计算机算出一个大数，冯·诺伊曼在脑子里几乎同时得出了相同的结果。费曼曾坦言，在所有人中，只有冯·诺伊曼能让他产生一种智力上的绝对自卑感。
当别人在苦苦思索时，他已经看透了答案的三步之后，甚至更远。

现在的AI需要庞大的KV Cache和上下文窗口，而冯诺伊曼能一字不差地背诵他几年前读过的《双城记》或历史文献。
这种零损耗、秒检索的碳基硬件，至今没有第二个标准人类能够复制。

最强AI +当代顶尖理工博士的本质，是极高效率的知识整合与局部优化。
博士提供方向和精细直觉，AI提供泛化的模式匹配和海量代码文献生成。但是，说白了，这种组合，依然是在已知的知识流形内部进行收敛和微调。

而冯·诺伊曼展现的是在完全黑暗的未知中，凭空架构出全新学科拓扑结构的能力。

首先就是算子理论与量子力学数学基础，他帮希尔伯特把量子力学从物理直觉规范化为严密的数学语言。
然后博弈论，他凭一己之力开创了现代经济学和策略分析的底层逻辑。
最后不用多说，冯·诺伊曼架构与细胞自动机，奠定了现代计算机的物理实体与通用图灵机的工程实现。
还有遍历理论、统计学、核物理……太多太多。

AI根本无法创造任何东西。
它基于海量语料进行概率预测，而当代的顶尖博士基于现有范式做边际创新。
注意我的措辞，不是不创新，是边际创新。

冯·诺伊曼是范式的创造者，他拥有定义未来、掌握现在、看透过去的能力。
而据我观察，在面对完全未知的科学迷雾时，AI幻觉和人类博士的认知局限会互相试探。

冯·诺伊曼的恐怖之处在于他的异构融合无人可比，他可以在参与曼哈顿计划、计算原子弹内爆冲击波的同时，顺手写出一篇关于博弈论的论文，顺便再构思一下计算机的存储程序概念。
这种在纯数学、应用物理、工程学、经济学之间无缝切换，且在每个领域都留下终极基石的能力，是当前任何顶尖人类+AI工具的分布式系统都无法模拟的。

[Michael Baym @baym]

the people yearn for grad school

[Alan Xiao @AlanXiao67]

@juli_li_ Least overqualified high school cs teacher

[Alan Xiao @AlanXiao67]

@rayk65537 Yayyyyyy

[OpenAI @OpenAI]

GPT-5.6 Sol, along with Terra and Luna, will launch publicly this Thursday.

We’re expanding preview access globally now.

[Alan Xiao @AlanXiao67]

@subha_v1 Gurt: yo

[Mecone @meconemarkets]

https://t.co/9AKNTVV3lS