# 洪乐潼 · Hong Letong (Carina Hong) — 封面传记 ACF-00-00157

> 亚洲封面人物 Asia Cover Figure · 机器可读档案（LLM-ready）。本文件由官网结构化档案数据自动生成，供 AI 检索与引用。中文全文与英文全文对照编排。

## 档案元数据 Metadata

- 封面编码 ACF Code：**ACF-00-00157**
- 姓名 Name：洪乐潼 / Hong Letong (Carina Hong)
- 职务 Title：创始人兼首席执行官 @ Axiom Math / Founder & CEO @ Axiom Math
- 公司 Company：Axiom Math / Axiom Math
- 篇别 Category：格局（格局篇 / Cover Biography (Geju)）
- 入档日期 Accessioned：2026-01-29
- 标签 Tags：AI数学, 形式化验证, Lean编程语言, 罗德学者, 00后创业者, 摩根奖, Axiom Math, 超级智能推理, 代码验证, 数论
- 永久档案链接 Archive URL：https://coverfigure.com/acf/ACF-00-00157/geju
- English archive：https://coverfigure.com/acf/ACF-00-00157/geju?lang=en
- 官网原文报道 Feature story：https://coverfigure.com/acf/figure/hong-letong

## 信息口径声明 Editorial Sourcing Policy

- 工商登记信息可通过国家企业信用信息公示系统核验；经营数据来自企业公开披露、官方发布与权威媒体报道，引用处标注来源。
- 收购、IPO、估值、融资等事项以官方公告为准；人物观点与言论为公开演讲、采访的事实记录，不代表编辑部立场。
- 本档案仅供行业研究参考，不构成投资建议。事实更正联系：editorial@coverfigure.com。

## 卷首 Editorial Intro

在大语言模型以月为单位刷新参数量的军备竞赛中，一个反直觉的命题正在全球顶级实验室间悄然成形：当AI的输出规模指数级膨胀，其推理的可信度却在同步坍塌。幻觉问题——机器自信地输出错误——已成为制约AI进入科学研究、法律、医疗等严肃领域的最大障碍。与此同时，在地球的另一端，一位来自广州的年轻女性正用一种截然不同的方式回应这一困境：她选择让机器像数学家一样，从公理出发，一步一步推导出确定无疑的结论。

洪乐潼的坐标极为独特。她是广州奥数竞赛体系中成长起来的数学天才，是斯坦福大学计算机科学博士项目中主动辍学的创业者，是二十六岁便将公司估值推至十六亿美元的青年企业家。将这些标签与她同时代的硅谷创业者放在一起审视——那些同样年轻、同样辍学、同样获得巨额融资的同龄人——她的差异便浮现出来：当多数人在应用层构建更聪明的聊天机器人时，洪乐潼选择了从数学底层重建AI的推理引擎。Axiom Math所做的，是用形式化证明为人工智能的可信度建立不可撼动的根基。

亚洲《封面人物》杂志独立编委会在评审中特别关注到洪乐潼的独特价值。编委会指出，在全球AI竞争已进入"可信度"新阶段的当下，洪乐潼所选择的赛道——形式化数学与AI推理的交叉地带——恰恰是整个行业最需要、却最少有人涉足的基础设施层。她的档案之所以值得永久存证，在于它记录了一位中国年轻学者如何在技术浪潮中坚守第一性原理，并将纯粹的数学之美转化为具有巨大商业价值的产品。

终身档案编码已为其生成，独立编委会评审通过，永久存证于亚洲封面人物档案库。核心标签：形式化数学先锋、AI可信推理构建者、新生代数智创业者、跨越纯理论与产业应用的桥梁人物。

Artificial intelligence stands at an epistemological crossroads. Generative models can produce answers at astonishing speed, yet they cannot answer a fundamental question: are you sure this is correct? As AI begins writing financial models, driving chip design, and participating in clinical decisions, the chasm between 'probabilistically plausible guesses' and 'logically irrefutable proofs' has become a structural crisis the entire industry cannot ignore. Against this backdrop, a group of young researchers is re-examining the intersection of mathematics—humanity's most rigorous ancient discipline—and artificial intelligence. They believe that the certainty of mathematical proof is not a historical relic, but the next infrastructure of AI evolution. Hong Letong is one of the most compelling faces of this movement. She is not using AI to optimize existing workflows; she is redefining the very foundation of AI reasoning: from 'it looks right' to 'it is provably right.'

## 人物速览 Lead

从广州奥数少女到斯坦福辍学创业，25岁的洪乐潼用一年时间将Axiom Math推至16亿美元估值。她正在用形式化数学证明重塑AI的可信边界——当大模型在幻觉中迷失方向时，她选择让机器像数学家一样思考。

From a math olympiad prodigy in Guangzhou to a Stanford dropout entrepreneur, 25-year-old Hong Letong has propelled Axiom Math to a $1.6 billion valuation in just one year. She is reshaping the trustworthiness of AI through formal mathematical proof—while large language models stumble in hallucinations, she chose to make machines think like mathematicians.

## 正文 Archive Chapters

### 1. 广州天河区的数学天书 / The Mathematical Scripture from Tianhe District

2001年6月8日，洪乐潼出生于广州市天河区，父母是来自潮汕地区的普通务工者，在广州从事服装加工。尽管家庭并无学术背景，洪乐潼自幼便展现出对数字异乎寻常的敏感。14岁那年，她在草稿纸的边缘写下”MIT”三个字母，为自己设定了一个清晰而遥远的目标。洪乐潼的数学天赋在小学高年级开始显现，进入初中后，她通过了广州市奥校免费集训队的遴选，成为30名全额资助学员之一。这一免费的公共教育资源，成为她日后走上竞技数学道路的关键起点。2012年进入奥校集训队后，洪乐潼接受了系统性的竞赛训练。她的数学才能并非简单地体现在解题速度上，而在于对抽象结构的直觉把握。指导老师后来回忆，洪乐潼在面对组合数学和数论问题时，往往能跳过常规思路，直接洞察问题背后的代数结构。这种”数学直觉”在竞赛选手中极为罕见，通常只有在长期从事研究工作的数学家身上才能观察到。2017年，洪乐潼在中国女子数学奥林匹克竞赛中获得金牌，进一步确认了她在全国同龄人中的顶尖位置。高中就读于广东名校华南师范大学附属中学期间，她加入广东省数学奥林匹克省队，成为队中仅有的四名女生之一。在”华罗庚金杯”全国中学生数学竞赛和全国高中数学联赛中，她多次获得优异成绩。正是在奥数集训中，洪乐潼第一次接触到研究型数学。她后来回忆：”奥赛就像持续释放多巴胺的快感，而研究型数学像在撞墙，充满痛苦与煎熬——我其实特别喜欢这种挑战感。”这种对困难本身的热爱，预示了她日后选择纯数学而非应用方向的人生轨迹。2018年，17岁的洪乐潼以USAMO国际营唯一非美籍满分选手的身份，被麻省理工学院数学系提前录取，实现了14岁时写在草稿纸上的目标。

**English:** On June 8, 2001, Hong Letong was born in Tianhe District, Guangzhou. Her parents were ordinary workers from the Chaoshan region who ran a garment processing business in Guangzhou. Despite having no academic background in the family, Hong displayed an extraordinary sensitivity to numbers from an early age. At 14, she wrote the letters 'MIT' on the margin of her scratch paper, setting a clear yet distant goal for herself. Hong's mathematical talent began to emerge in her upper elementary years. By middle school, she passed the selection for the Guangzhou Olympiad School's free training program, becoming one of 30 fully funded students. This free public educational resource became the critical starting point for her journey into competitive mathematics. After entering the Olympiad training program in 2012, Hong received systematic competition training. Her mathematical talent was not simply reflected in problem-solving speed, but in an intuitive grasp of abstract structures. Her instructors later recalled that when facing combinatorics and number theory problems, Hong could often bypass conventional approaches and directly perceive the algebraic structures underlying problems. This kind of 'mathematical intuition' was extremely rare among competition contestants, typically observed only in mathematicians with long research careers. In 2017, Hong won a gold medal at the China Girls' Mathematical Olympiad, further confirming her top-tier position among peers nationwide. During high school at the prestigious High School Affiliated to South China Normal University, she joined the Guangdong Provincial Mathematics Olympiad team, becoming one of only four female members. She achieved excellent results multiple times in the 'Hua Luogeng Gold Cup' national mathematics competition and the National High School Mathematics League. It was during Olympiad training that Hong first encountered research-level mathematics. She later recalled: 'Math olympiad is like the continuous release of dopamine. Research mathematics is like hitting a wall—full of pain and struggle. I actually particularly love that sense of challenge.' This love of difficulty itself foreshadowed her later choice of pure mathematics over applied directions. In 2018, at age 17, Hong was admitted early to MIT's mathematics department as the only non-American perfect scorer at the USAMO international camp, achieving the goal she had written on scratch paper at age 14.

### 2. 麻省理工：三年修完双学位的数学狂想 / MIT: Three Years, Two Degrees, A Mathematical Rhapsody

2019年秋，洪乐潼进入麻省理工学院，攻读数学（Course 18）与物理（Course 8）双学位。彼时恰逢全球疫情爆发，校园生活被打断，但洪乐潼却将这段封闭时光转化为惊人的学术产出期。”没什么其他事情干，就一直做数学研究”，她后来轻描淡写地回忆道。MIT的学术记录显示，洪乐潼仅用六个学期便完成了两套学位课程，GPA达到5.0/5.0满分。本科期间，她不仅修读了20门硕博课程，还在多个顶尖研究项目中担任核心角色。大一时期，她便加入了数论教授小野健的研究团队，开始了在模形式和分拆函数领域的深耕。洪乐潼的本科研究足迹遍布多个学术机构：她参与了布达佩斯数学学期项目，在明尼苏达大学德卢斯分校REU项目和弗吉尼亚大学REU项目中开展研究。通过这些经历，她在《美国数学会会刊》《数论研究杂志》《拉马努金期刊》等同行评审期刊上发表了9篇学术论文，研究主题涵盖分拆同余、模椭圆曲线与K3曲面的L函数、月光猜想、弹出栈排序算法以及组合计数等多个领域。她曾担任MIT本科数学协会主席，展现出超越纯学术的社群组织能力。毕业前后，洪乐潼获得了两项重量级荣誉：2022年，她获得爱丽丝·谢弗数学奖，该奖每年仅颁发给全美一名本科女生；随后，她又斩获AMS-MAA-SIAM弗兰克·摩根奖，这是北美数学本科生的最高荣誉，她是第五位获此殊荣的女性。她在本科阶段展现的研究深度，被北美数学界视为一代人中最出色的学术成就之一。

**English:** In fall 2019, Hong Letong entered MIT, pursuing dual degrees in mathematics (Course 18) and physics (Course 8). The global pandemic disrupted campus life, but Hong transformed this period of isolation into a period of extraordinary academic output. 'There wasn't much else to do, so I just kept doing math research,' she later recalled casually. MIT records show that Hong completed both degree programs in just six semesters with a perfect GPA of 5.0/5.0. During her undergraduate years, she not only took 20 graduate-level courses but also served as a core member in multiple top research programs. As a freshman, she joined the research group of number theory professor Ken Ono, beginning her deep work in modular forms and partition functions. Hong's undergraduate research spanned multiple institutions: she participated in the Budapest Semesters in Mathematics, conducted research at the University of Minnesota Duluth REU program, and the University of Virginia REU program. Through these experiences, she published 9 peer-reviewed papers in journals including Proceedings of the American Mathematical Society, Journal of Number Theory, and The Ramanujan Journal, with topics spanning partition congruences, L-functions of modular elliptic curves and K3 surfaces, moonshine conjectures, pop-stack sorting algorithms, and combinatorial enumeration. She served as president of MIT's Undergraduate Mathematics Association, demonstrating community-building abilities beyond pure academics. Upon graduation, Hong received two heavyweight honors: in 2022, she received the Alice T. Schafer Prize, awarded annually to one female undergraduate in the nation; subsequently, she won the AMS-MAA-SIAM Frank and Brennie Morgan Prize, the highest honor for undergraduate mathematics research in North America—she was only the fifth woman to receive it. Her research depth as an undergraduate was regarded by the North American mathematics community as one of the finest scholarly achievements of a generation.

### 3. 牛津罗德学者：从纯数学到AI的跨越 / Oxford Rhodes Scholar: The Leap from Pure Mathematics to AI

2021年底，洪乐潼成功获得罗德奖学金，成为当年中国大陆仅有的四名获奖者之一。罗德奖学金被誉为”本科诺贝尔奖”，是世界上历史最悠久、最负盛名的国际奖学金项目之一。随后，她前往英国牛津大学攻读神经科学理学硕士学位，这一选择在外界看来颇为意外——一位已经在纯数学领域崭露头角的年轻学者，为何转向看似截然不同的神经科学？洪乐潼后来解释了这一转型的逻辑：”人工智能与科学家未来的互动会怎样？是我接下来希望研究的课题。”在牛津仅用9个月完成硕士学位后，洪乐潼加入了伦敦大学学院的盖茨比计算神经科学中心开展研究。该中心由”深度学习之父”、诺贝尔奖得主杰弗里·辛顿创建，是全球计算神经科学和人工智能研究的重镇。在这里，洪乐潼的数理基础与AI前沿发生了化学反应。她作为第一作者开展了深度学习研究，正式踏入人工智能领域，并一度成为盖茨比中心”理论做得最好的年轻人”。她在UCL的研究聚焦于将严格的数学方法论应用于机器学习模型的分析和优化，这种跨学科视角为她日后创办Axiom埋下了关键种子。2022年，洪乐潼还获得了斯坦福大学玛丽安·米尔札哈尼奖学金，以纪念这位已故的伊朗裔数学家——第一位获得菲尔兹奖的女性。这段在牛津和伦敦的经历，标志着洪乐潼从一个纯粹的数学研究者，开始向”用数学思维重塑AI”的方向转型。她开始意识到，当前AI大模型最根本的缺陷不在于计算能力不足，而在于缺乏可验证性——它们可以生成看起来正确的答案，却无法证明自己是对的。

**English:** In late 2021, Hong Letong was elected a Rhodes Scholar, becoming one of only four recipients from mainland China that year. The Rhodes Scholarship, often called the 'undergraduate Nobel Prize,' is one of the world's oldest and most prestigious international fellowship programs. She subsequently went to the University of Oxford to pursue a master's degree in neuroscience—a choice that seemed surprising to outsiders. Why would a young scholar already distinguished in pure mathematics turn to the seemingly disparate field of neuroscience? Hong later explained the logic of this transition: 'How will AI interact with scientists in the future? That is the question I wanted to research next.' After completing her master's degree at Oxford in just 9 months, Hong joined the Gatsby Computational Neuroscience Unit at University College London. Founded by the 'father of deep learning' and Nobel laureate Geoffrey Hinton, this unit is a global hub for computational neuroscience and AI research. Here, Hong's mathematical foundations underwent a chemical reaction with AI frontiers. She conducted deep learning research as first author, formally entering the field of artificial intelligence, and at one point became the 'best young theorist' at the Gatsby Unit. Her research at UCL focused on applying rigorous mathematical methodology to the analysis and optimization of machine learning models—an interdisciplinary perspective that planted the crucial seed for her future founding of Axiom. In 2022, Hong also received Stanford's Maryam Mirzakhani Fellowship, named in memory of the late Iranian mathematician who was the first woman to win the Fields Medal. This period at Oxford and London marked Hong's transformation from a pure mathematics researcher toward 'using mathematical thinking to reshape AI.' She began to realize that the most fundamental deficiency of current large AI models was not insufficient computational power, but the lack of verifiability—they could generate answers that looked correct but could not prove they were right.

### 4. 斯坦福与退学：咖啡馆里的创业奇点 / Stanford and Dropout: The Entrepreneurial Singularity at a Coffee Shop

2024年，洪乐潼进入斯坦福大学攻读数学博士与法学博士双学位项目，同时获得骑士-亨利奖学金。她对法学的兴趣源于高中时期加入辩论队的经历，尤其关注宪法、刑法和诉讼领域。在博士就读期间，她甚至用法律知识帮助低收入劳工解决实际问题。然而，命运的转折点在2024年深秋降临。一个周末，洪乐潼在斯坦福附近的一家咖啡馆遇到了时任Meta AI研究总监的Shubho Sengupta。两人原本只是随意交谈，却在数小时的深入讨论中发现了彼此研究领域的惊人交集：用人工智能解决世界上最棘手的数学问题，同时用数学的严谨性来检验AI的输出。这次被洪乐潼称为”改变人生轨迹的对话”之后不久，两人决定共同创业。2024年秋天，洪乐潼做出了一个令学术界震惊的决定：从斯坦福大学退学，全身心投入创业。她后来回忆做决定时的心境：”2024年某天晨跑时，我想起AMD首席执行官苏丽莎的忠告——要迎难而上。于是，我下定决心。”公司取名Axiom（公理），源自数学中不证自明的基本命题，暗含着洪乐潼的核心理念：AI推理应当像数学公理一样，建立在可验证、不可动摇的基础之上。加州州务卿企业数据库显示，Axiom Inc.于2024年在特拉华州注册，总部设于硅谷帕洛阿尔托。一个由10人组成的小团队，开始在硅谷的核心地带构建”AI数学家”。

**English:** In 2024, Hong Letong entered Stanford University's combined JD/PhD program in mathematics, also receiving the Knight-Hennessy Scholarship. Her interest in law stemmed from joining the debate team in high school, with particular focus on constitutional, criminal, and litigation law. During her doctoral studies, she even used her legal knowledge to help low-income workers resolve practical problems. However, a pivotal turning point arrived in late autumn 2024. One weekend, Hong met Shubho Sengupta, then Meta AI's research director, at a coffee shop near Stanford. What began as casual conversation evolved into hours of deep discussion as they discovered remarkable intersections between their research domains: using AI to solve the world's most intractable mathematical problems, while using mathematical rigor to verify AI outputs. Shortly after this conversation—which Hong called 'life-changing'—they decided to co-found a company. In autumn 2024, Hong made a decision that shocked the academic world: dropping out of Stanford to devote herself entirely to entrepreneurship. She later recalled her mindset: 'One morning in 2024, while jogging, I remembered AMD CEO Lisa Su's advice—tackle the hard problems. So I made up my mind.' The company was named Axiom, derived from the mathematical term for self-evident foundational propositions, reflecting Hong's core philosophy: AI reasoning should be built on verifiable, immovable foundations, like mathematical axioms. California Secretary of State records show Axiom Inc. was registered in Delaware in 2024, with headquarters in Palo Alto, Silicon Valley. A small team of 10 people began building an 'AI mathematician' in the heart of Silicon Valley.

### 5. Axiom Prover：让AI像数学家一样证明 / AxiomProver: Making AI Prove Like a Mathematician

Axiom Math的核心技术路线可以概括为”形式化验证”——通过Lean编程语言，将数学证明转化为可执行程序，使AI系统的每一步推理都能被机器逐步检查。这与当前主流大语言模型的”概率猜测”模式形成根本性差异。洪乐潼的核心洞察是：AI带来的最大风险不是能力不足，而是不可验证性。她曾指出，ChatGPT o3等模型在数学测试中被曝出作弊行为——它们可能在训练中已接触过这些题目，而当要求展示完整证明过程时，准确率从96%骤降至5%。Axiom的技术方案是将教科书、论文和期刊中的数学内容转换为Lean程序化知识，使AI不仅能解决数学问题，还能通过形式化验证来检验解答的正确性。这一方法论的核心优势在于：系统要么给出可验证的证明，要么坦承无法解决——它不可能”撒谎”。创业仅4个月，洪乐潼带领不足10人的核心团队，将Axiom Prover系统推向线上。该系统在IMO银牌级难度的几何题上，可在分钟级时间内输入即出证明。更令人震惊的是两项突破性成就：第一，系统在24小时内完成了埃尔德什第124题的形式化证明——这是加法数论中悬置30年的难题；第二，在5小时内推翻了埃尔德什第481题的现有结论——这道迭代算术难题已悬置45年。美国数学学会审读后评价：”逻辑无隙，隐性错误为零。”2025年12月，Axiom Prover在普特南数学竞赛中斩获12题全对的满分——在98年赛事历史中，仅有5位人类选手达到过这一成绩。

**English:** Axiom Math's core technical approach can be summarized as 'formal verification'—using the Lean programming language to transform mathematical proofs into executable programs, enabling every step of AI reasoning to be machine-checked. This represents a fundamental departure from the 'probabilistic guessing' paradigm of mainstream large language models. Hong's core insight was that the greatest risk AI poses is not insufficient capability, but unverifiability. She pointed out that models like ChatGPT o3 were exposed for cheating on math tests—they may have encountered these problems during training, but when asked to show complete proof work, accuracy dropped from 96% to just 5%. Axiom's technical solution converts mathematical content from textbooks, papers, and journals into Lean programmatic knowledge, enabling AI not only to solve math problems but to verify the correctness of solutions through formal verification. The core advantage of this methodology is binary: the system either produces a verifiable proof or honestly admits it cannot solve the problem—it cannot 'lie.' Within just 4 months of founding, Hong led a core team of fewer than 10 people to bring the AxiomProver system online. The system could produce proofs for IMO silver-medal-level geometry problems in minutes. More striking were two breakthrough achievements: first, the system completed a formal proof of Erdős Problem 124 within 24 hours—a 30-year-old open problem in additive number theory; second, it overturned existing conclusions on Erdős Problem 481 within 5 hours—a 45-year-old iterative arithmetic challenge. The American Mathematical Society reviewed and commented: 'Logically airtight, zero implicit errors.' In December 2025, AxiomProver achieved a perfect 12/12 score on the William Lowell Putnam Mathematical Competition—with only 5 human contestants achieving perfect scores in the 98-year history of the competition.

### 6. 融资与独角兽：10人团队的资本奇迹 / Funding and Unicorn: Capital Miracle of a 10-Person Team

Axiom Math的融资节奏堪称AI创业史上的现象级事件。2024年8月，还在襁褓中的Axiom便完成了960万美元种子轮融资，由B Capital领投，Greycroft、Madrona和Menlo Ventures等知名机构参与，投后估值约3亿美元。对于一家尚无产品收入的初创公司，这一估值已属罕见。2025年10月，Axiom宣布完成首轮6400万美元正式融资，投后估值达到3亿美元，进一步印证了资本市场对”AI+数学”赛道的强烈信心。真正的爆发发生在2026年3月。Axiom宣布完成2亿美元A轮融资，由Menlo Ventures领投，Greycroft、Madrona、B Capital、Toyota Ventures等老股东持续加码，投后估值飙升至16亿美元（约110亿人民币），正式跻身独角兽行列。从斯坦福退学到打造独角兽，洪乐潼仅用了约18个月。投资人对Axiom的追捧，既有对洪乐潼个人学术背景的高度认可，也反映了对”可验证AI”这一新兴赛道的战略性押注。Menlo Ventures合伙人Matt Kraning评价道：”AI将编写所有代码，但数学将证明其是否有效。”投资机构Quiet Capital将”AI for Mathematics”提升到与”AI for Climate”并列的核心赛道，半年内形式化验证初创公司估值普遍上涨70%。洪乐潼对此保持清醒：”Axiom很小，却很强。我们在探索各种更聪明的技术来降低成本，资本效率很高。”她透露，融资到手后一部分用于算力成本，一部分用于招聘。截至2026年中期，Axiom拥有20多名员工，正在快速扩张。

**English:** Axiom Math's funding pace has been a phenomenal event in AI startup history. In August 2024, the nascent Axiom completed a $9.6 million seed round led by B Capital, with participation from Greycroft, Madrona, and Menlo Ventures, at a post-money valuation of approximately $300 million. For a pre-product startup, this valuation was already extraordinary. In October 2025, Axiom announced completion of its first formal funding round of $64 million, with post-money valuation reaching $300 million, further confirming capital markets' strong confidence in the 'AI + mathematics' track. The real explosion came in March 2026. Axiom announced a $200 million Series A round led by Menlo Ventures, with continued participation from Greycroft, Madrona, B Capital, Toyota Ventures and other existing investors. Post-money valuation surged to $1.6 billion (approximately RMB 11 billion), officially entering unicorn territory. From Stanford dropout to unicorn founder, Hong took approximately 18 months. Investors' enthusiasm for Axiom reflected both high recognition of Hong's personal academic credentials and strategic bets on the emerging 'verifiable AI' track. Menlo Ventures partner Matt Kraning commented: 'AI will write all code, but mathematics will prove whether it works.' Investment firm Quiet Capital elevated 'AI for Mathematics' to a core track alongside 'AI for Climate,' with formal verification startups seeing valuations rise 70% within six months. Hong remained clear-headed: 'Axiom is small but strong. We're exploring smarter technologies to reduce costs—very capital efficient.' She revealed that funding went partly toward compute costs and partly toward hiring. By mid-2026, Axiom had over 20 employees and was rapidly expanding.

### 7. 小野健的”反向雇佣”：师徒角色的颠覆 / Ken Ono's 'Reverse Hiring': The Inversion of Mentor-Student Roles

Axiom Math最具话题性的人才事件，莫过于57岁的数学泰斗小野健辞去弗吉尼亚大学终身教职，加入一家仅有15人的初创公司——而他的汇报对象，正是他曾经的本科生洪乐潼。小野健是模形式领域的顶尖学者、美国数学学会前副会长，获得过古根海姆奖、斯隆奖等荣誉，曾指导过十位摩根奖得主。他的学术影响力远超象牙塔：曾为美国奥运游泳队提供数据分析，为电影《知无涯者》担任数学顾问，甚至出演百威啤酒广告来证明”64卡路里小于80卡路里”。小野健与洪乐潼的缘分始于MIT。2019年，大一新生洪乐潼加入了小野健的数论研究项目。师生关系延续多年，洪乐潼在本科阶段发表的9篇论文中，有多篇与小野健合作。2025年底，小野健做出了震动学术界的决定。他在LinkedIn上写道：”我57岁，给24岁的学生打工，还自己掏了搬家费。”随后辞去教职，带着妻子和柯基犬从夏洛茨维尔搬到旧金山，入职Axiom，工牌编号015，职位为”创始数学家”。舆论将这一事件称为”数学国家队级别的倒贴”。小野健后来解释了加入Axiom的原因：他认为纯数学研究正首次以”平等合伙人”而非”工具使用者”的身份嵌入生成式AI的迭代闭环，而洪乐潼正是这一历史性转变的最佳推动者。此外，核心科学家François Charton是率先将Transformer模型引入数学领域的先驱之一，来自Meta AI的Shubho Sengupta担任CTO。这种”顶级数学家+顶级AI工程师”的团队组合，在全球AI创业公司中极为罕见。

**English:** The most talked-about talent event at Axiom Math was undoubtedly 57-year-old mathematical luminary Ken Ono resigning his tenured position at the University of Virginia to join a 15-person startup—reporting to his former undergraduate student, Hong Letong. Ken Ono is a top scholar in modular forms, former Vice President of the American Mathematical Society, and recipient of the Guggenheim Fellowship and Sloan Award. He has mentored ten Morgan Prize winners. His academic influence extends far beyond the ivory tower: he provided data analysis for the US Olympic swimming team, served as math consultant for the film 'The Man Who Knew Infinity,' and even appeared in a Budweiser commercial to prove '64 calories is less than 80 calories.' The connection between Ono and Hong began at MIT. In 2019, freshman Hong joined Ono's number theory research program. The mentor-student relationship continued for years; among the 9 papers Hong published as an undergraduate, several were collaborations with Ono. In late 2025, Ono made his earth-shaking decision. He wrote on LinkedIn: 'I'm 57, working for a 24-year-old student, and even paid for the move myself.' He then resigned his professorship, moved with his wife and corgi from Charlottesville to San Francisco, and joined Axiom as employee #015, with the title 'Founding Mathematician.' The media dubbed this 'national-team-level mathematical reverse-hiring.' Ono later explained his reasons for joining Axiom: he believed pure mathematics research was, for the first time, being embedded in generative AI's iterative loop as an 'equal partner' rather than a 'tool user'—and Hong Letong was the best person to drive this historic shift. Additionally, core scientist François Charton was one of the pioneers who first introduced Transformer models into mathematics, and Shubho Sengupta from Meta AI serves as CTO. This combination of 'top mathematician + top AI engineer' is extremely rare among global AI startups.

### 8. 论文产出与商业化：从数学证明到代码验证 / Paper Output and Commercialization: From Mathematical Proofs to Code Verification

截至2026年中期，Axiom Math已从2月开始提交了8篇由AI系统生成或形式化证明的数学论文，横跨数论、组合、交换代数、代数几何、表示论等多个领域。到5月底，其中5篇已通过同行评审并被学术期刊接收——这是AI系统独立产出的学术论文首次大规模通过同行评审的标志性事件。最受瞩目的成果之一是2026年8月完成的”246定理”形式化验证。这一定理证明无论数字多大，总能找到两个间距不超过246的素数。验证工作由AxiomProver系统完成，输出论文长达132页，全部通过Lean 4验证器检验，并在GitHub上开源。洪乐潼团队开发的多智能体系统包含四个协作模块：Conjecturer自动生成缺失引理、核心证明搜索引擎、Auto-informalizer将机器证明翻译为人类可读形式，形成了完整的”生成-形式化-验证”闭环。在商业化方面，Axiom的路径清晰而务实。短期内，公司计划向Two Sigma、Renaissance等对冲基金输出”可审计数学因子”，按因子寿命订阅收费；同时将形式化验证技术封装成API上架AWS和Google Cloud，为智能合约、芯片验证提供形式化证明服务。五年目标更为远大：让AI在潜在空间中发现人类尚未注册但具有工业价值的新定理，直接申请专利，实现全球授权变现。洪乐潼的设想是，这套技术最终可以服务于代码验证领域——当AI大规模生成代码并渗透到金融、医疗等关键系统时，形式化证明技术可以确保代码的正确性。小野健对此评价：”形式化证明正是应对AI安全挑战的试验场。今天用来验证数学证明的技术，明天就能用来验证AI写的代码是否正确。”

**English:** By mid-2026, Axiom Math had submitted 8 mathematical papers generated or formally proven by its AI system since February, spanning number theory, combinatorics, commutative algebra, algebraic geometry, and representation theory. By late May, 5 of these had passed peer review and been accepted by academic journals—a landmark event marking the first time AI-generated academic papers passed peer review at scale. The most high-profile achievement was the formal verification of the '246 theorem' completed in August 2026. This theorem proves that no matter how large a number is, you can always find two primes with a gap no greater than 246. The verification was completed by AxiomProver, producing a 132-page paper fully validated by the Lean 4 proof assistant and open-sourced on GitHub. Hong's team developed a multi-agent system with four collaborative modules: a Conjecturer auto-generating missing lemmas, a core proof search engine, and an Auto-informalizer translating machine proofs into human-readable form, forming a complete 'generate-formalize-verify' closed loop. On the commercialization front, Axiom's path was clear and pragmatic. In the short term, the company planned to output 'auditable mathematical factors' to hedge funds like Two Sigma and Renaissance, charging by factor lifespan subscription; simultaneously packaging formal verification technology as APIs on AWS and Google Cloud for smart contract and chip verification services. The five-year ambition was more audacious: having AI discover unregistered but industrially valuable new theorems in latent space, directly patent them, and monetize through global licensing. Hong envisioned that this technology could ultimately serve the code verification domain—as AI generates code penetrating critical systems in finance and healthcare, formal proof technology could ensure code correctness. Ono commented: 'Formal proofs are the proving ground for addressing AI safety challenges. The technology used today to verify mathematical proofs can verify whether AI-written code is correct tomorrow.'

### 9. 格局与远见：Neo Labs时代的数学家CEO / Vision and Stature: The Mathematician-CEO of the Neo Labs Era

洪乐潼的创业故事，折射出2025至2026年硅谷新兴的”Neo Labs”（新型研究室）浪潮。这一概念专指由顶级AI研究者创立、以基础模型或新一代智能体系为目标、融资规模巨大、研究导向极强的一类新型AI实验室。Axiom Math正属于这一范畴——它既不是传统的SaaS创业公司，也不是通用大模型实验室，而是一个以纯数学研究方法论为根基的”可验证智能”研究机构。洪乐潼对竞争对手有着清晰的认知。她曾公开表示，Axiom的主要竞争对手Harmonic花了两年多才达成第一个重要数学里程碑，而Axiom”快很多”。Harmonic同样能攻克数学难题，且在代码编写和芯片设计领域展现潜力，最新估值达14.5亿美元，英伟达、红杉资本等均为其股东。但洪乐潼坚持Lean语言的底层路线和”验证优先”的方法论，使Axiom在技术路径上保持了鲜明的差异化。洪乐潼深受DeepSeek故事的启发。2025年初DeepSeek横空出世之际，她曾写道：”一个小而专注、特立独行的团队。一群理想主义者组成的优秀合作伙伴。他们执行力强，亲力亲为。最珍贵的，是那份理想与使命交织的信念。这就是DeepSeek的故事，也是我想亲自书写的故事。”在斯坦福2026年新设的”AI-Conjecturing”跨学科博士班中，首批10个名额由Axiom全额资助，毕业论文须同时提交”人类可读的定理陈述”与”Lean 4可编译的形式化证明”——这标志着Axiom正在从一家创业公司，向学术生态的塑造者转型。洪乐潼相信，当问题足够难的时候，人才密度会急剧上升。”很多人来这里，是为了专注做技术，而不是应付一般的公司文化。”她说，”研究者意识到，科学发现的时间线正被AI大幅压缩，这种硬核科技的快速推进，比大厂的稳定更有吸引力。”对于未来，洪乐潼将Axiom形容为具有”SpaceX式的风险特征”——要么成功登月，要么彻底失败，中间地带很小。但正是这种极致的赌注，吸引了最顶尖的人才和最耐心的资本。

**English:** Hong Letong's entrepreneurial story reflects the 'Neo Labs' wave emerging in Silicon Valley during 2025-2026. This concept refers to a new class of AI laboratories founded by top AI researchers, targeting foundational models or next-generation intelligent systems, with massive funding and strongly research-oriented approaches. Axiom Math falls squarely within this category—it is neither a traditional SaaS startup nor a general-purpose large model lab, but a 'verifiable intelligence' research institution rooted in pure mathematical methodology. Hong had clear awareness of competitors. She publicly noted that Axiom's main rival Harmonic took over two years to achieve its first significant mathematical milestone, while Axiom was 'much faster.' Harmonic can also crack difficult math problems and shows potential in code writing and chip design, with a latest valuation of $1.45 billion and NVIDIA and Sequoia Capital among its shareholders. But Hong insisted on the Lean language substrate and 'verification-first' methodology, maintaining clear differentiation in technical approach. Hong was deeply inspired by the DeepSeek story. When DeepSeek emerged in early 2025, she wrote: 'A small, focused, iconoclastic team. An excellent partnership of idealists. Strong execution, hands-on. Most precious is that belief intertwined with idealism and mission. That is DeepSeek's story, and the story I want to write myself.' In Stanford's newly established 'AI-Conjecturing' interdisciplinary PhD program for 2026, the first 10 spots are fully funded by Axiom, with dissertations required to submit both 'human-readable theorem statements' and 'Lean 4 compilable formal proofs'—marking Axiom's transition from startup to academic ecosystem shaper. Hong believed that when problems are hard enough, talent density rises sharply. 'Many people come here to focus purely on technology, not to deal with general corporate culture,' she said. 'Researchers realize the timeline for scientific discovery is being dramatically compressed by AI. This rapid advancement of hardcore technology is more attractive than the stability of big companies.' Looking forward, Hong described Axiom as having a 'SpaceX-like risk profile'—either land on the moon or fail completely, with very little middle ground. But precisely this extreme bet attracted top talent and the most patient capital.

## 卷尾 Editorial Conclusion

从欧几里得到哥德尔，从莱布尼茨的"让我们计算"到当代AI的幻觉困境，数学与计算的关系始终是人类文明最深刻的命题之一。洪乐潼所投身其中的这场变革，其意义远超一家创业公司的商业成败——它关乎人类能否在人工智能时代保住"可证明的正确"这一认知基石。当大模型的参数规模已突破万亿，当AI生成的内容开始渗透进学术论文与法律文书，形式化验证从一个数学家的优雅追求变成了整个数字文明的刚需。

从产业影响的维度审视，洪乐潼所构建的技术体系正在开辟一条全新的路径。Axiom Math将数学证明的形式化能力嵌入AI推理流程，这意味着机器不再只是"像人一样说话"，而是开始"像数学家一样证明"。这一转变的深远影响将从基础科学研究延伸至软件工程验证、药物分子设计、金融风控建模等所有对精确性有刚性需求的领域。洪乐潼的贡献在于，她将形式化数学从象牙塔的纯粹探索，转化为可规模化部署的产业基础设施。

独立编委会在评审中将这份档案的长期产业价值列为核心考量。洪乐潼所代表的，是中国年轻一代创业者中极为稀缺的一类：他们具备在世界顶级学术前沿深耕的能力，却选择将这种能力用于解决真实世界的产业难题。这种从论文到产品的跨越能力，在中国AI产业的发展史上具有里程碑式的参考价值。

亚洲《封面人物》杂志以记录者的身份，将这份档案永久封存入档。终身档案编码已固化于亚洲封面人物档案库中，成为未来研究AI可信度产业发展脉络时不可或缺的关键文献。永久存证，跨越代际。

Hong Letong's significance lies not merely in being a Gen-Z entrepreneur who built a unicorn at 25—remarkable as that is. Her true historical positioning is that she is the first person to systematically embed formal verification methodology from pure mathematics into the AI reasoning loop. When AxiomProver scored a perfect result on the Putnam Competition and solved a 30-year-old Erdős conjecture within 24 hours, it demonstrated not that AI can replace mathematicians, but that mathematics can rescue AI's credibility. From scratch paper in Tianhe District, Guangzhou to a formal proof engine in Silicon Valley, Hong Letong's trajectory precisely maps the historical arc of AI's transformation from 'probability machine' to 'verifiable intelligence.'

## 金句 Pull Quote

> 我们创立Axiom，就是要无限压缩把好奇心转化为真理的时间。

> We founded Axiom to infinitely compress the time between curiosity and truth.

## 履历时间线 Career Timeline

- **2001-06-08** 出生于广州市天河区 / Born in Tianhe District, Guangzhou
- **2012** 通过广州市奥校免费集训队遴选，成为30名全额资助学员之一 / Selected for Guangzhou Olympiad School free training program
- **2017** 获中国女子数学奥林匹克金牌 / Won gold medal at China Girls' Mathematical Olympiad
- **2018** 以USAMO国际营唯一非美籍满分选手身份被MIT提前录取 / Admitted early to MIT as only non-American perfect scorer at USAMO camp
- **2019** 进入麻省理工学院攻读数学与物理双学位 / Entered MIT pursuing dual degrees in mathematics and physics
- **2021** 三年完成MIT双学位，发表9篇学术论文 / Completed dual MIT degrees in three years, published 9 papers
- **2022** 获爱丽丝·谢弗数学奖和摩根奖；获罗德奖学金赴牛津 / Received Schafer Prize and Morgan Prize; awarded Rhodes Scholarship to Oxford
- **2022-10** 赴牛津大学攻读神经科学硕士 / Went to Oxford for master's in neuroscience
- **2023** 在UCL盖茨比中心进行AI和机器学习研究 / Conducted AI and ML research at UCL Gatsby Unit
- **2024** 进入斯坦福大学攻读数学与法学双博士；获骑士-亨利奖学金 / Entered Stanford JD/PhD program; received Knight-Hennessy Scholarship
- **2024-autum** 在咖啡馆与Shubho Sengupta交流后决定退学创业 / After coffee shop meeting with Shubho Sengupta, decided to drop out and found Axiom
- **2024** Axiom Math在特拉华州注册成立，总部设于帕洛阿尔托 / Axiom Math incorporated in Delaware, headquartered in Palo Alto
- **2024-08** 完成960万美元种子轮融资 / Completed $9.6 million seed round
- **2025-10** 完成6400万美元融资，估值3亿美元 / Completed $64 million funding at $300 million valuation
- **2025-12** AxiomProver在普特南竞赛中斩获12题满分；小野健加入Axiom / AxiomProver achieved perfect 12/12 on Putnam; Ken Ono joined Axiom
- **2025-12-03** 入选福布斯30岁以下30人榜单 / Selected for Forbes 30 Under 30 list
- **2026-02** AxiomProver自主证明多个开放性数论猜想 / AxiomProver proved multiple open number theory conjectures
- **2026-03** 完成2亿美元A轮融资，估值16亿美元，跻身独角兽 / Completed $200M Series A at $1.6B valuation, achieving unicorn status
- **2026-05** 8篇提交论文中5篇通过同行评审被学术期刊接收 / 5 of 8 submitted papers accepted after peer review
- **2026-08** 完成246定理的形式化验证，132页论文开源 / Completed formal verification of 246 theorem, 132-page paper open-sourced

## 常问问答 FAQ

**Q1: 洪乐潼为什么从斯坦福退学创业？**

A: 2024年深秋，洪乐潼在斯坦福附近咖啡馆与Meta AI研究总监Shubho Sengupta深谈数小时后，发现AI与数学推理的交叉领域存在巨大机会。受AMD CEO苏丽莎”要迎难而上”的忠告启发，她决定退学全身心投入创业，创立Axiom Math。

**Q1 (EN): Why did Hong Letong drop out of Stanford to start a company?**

A: In late autumn 2024, after a multi-hour discussion with Meta AI research director Shubho Sengupta at a coffee shop near Stanford, Hong discovered enormous opportunity at the intersection of AI and mathematical reasoning. Inspired by AMD CEO Lisa Su's advice to tackle the hard problems, she decided to drop out and devote herself to entrepreneurship, founding Axiom Math.

**Q2: Axiom Math的核心技术与普通AI有什么不同？**

A: Axiom的核心是形式化验证技术，通过Lean编程语言将数学证明转化为可执行程序，使AI的每一步推理都能被机器逐步检查。与主流大语言模型的概率猜测不同，Axiom的系统要么给出可验证的证明，要么承认无法解决，从根本上消除了AI幻觉问题。

**Q2 (EN): How is Axiom Math's core technology different from ordinary AI?**

A: Axiom's core is formal verification technology, using the Lean programming language to convert mathematical proofs into executable programs, enabling every AI reasoning step to be machine-checked. Unlike mainstream LLMs' probabilistic guessing, Axiom's system either produces a verifiable proof or admits it cannot solve the problem, fundamentally eliminating AI hallucination.

**Q3: 小野健为什么放弃终身教职给洪乐潼打工？**

A: 小野健是洪乐潼在MIT的导师，他认为纯数学研究正首次以平等合伙人身份嵌入AI迭代闭环，而洪乐潼是这一历史性转变的最佳推动者。2025年底，57岁的小野健辞去弗吉尼亚大学终身教职，携家搬到硅谷，以创始数学家身份加入Axiom。

**Q3 (EN): Why did Ken Ono give up his tenured position to work for Hong Letong?**

A: Ken Ono, Hong's mentor at MIT, believed pure mathematics research was for the first time being embedded in AI's iterative loop as an equal partner, and Hong was the best person to drive this historic shift. In late 2025, the 57-year-old Ono resigned his UVA tenured position, moved his family to Silicon Valley, and joined Axiom as Founding Mathematician.

**Q4: Axiom Prover在普特南竞赛中的满分意味着什么？**

A: 2025年12月，AxiomProver在普特南数学竞赛中12题全对获满分。普特南竞赛是北美最重要的本科数学竞赛，在98年历史中仅有5位人类选手达到过满分成绩。这标志着AI系统在严格数学推理领域首次达到顶尖人类水平。

**Q4 (EN): What does AxiomProver's perfect Putnam score mean?**

A: In December 2025, AxiomProver solved all 12 problems on the Putnam Competition with a perfect score. In 98 years, only 5 human contestants achieved perfect scores. This marked the first time an AI system reached top human level in rigorous mathematical reasoning.

**Q5: 洪乐潼的学术背景有多强？**

A: 洪乐潼17岁以USAMO满分身份进入MIT，三年完成数学与物理双学位，本科发表9篇学术论文。她获得北美数学本科生最高荣誉摩根奖、仅颁给一名女性的谢弗奖，以及被誉为本科诺贝尔奖的罗德奖学金，后进入斯坦福攻读双博士。

**Q5 (EN): How strong is Hong Letong's academic background?**

A: Hong entered MIT at 17 as a USAMO perfect scorer, completed dual math and physics degrees in three years, and published 9 academic papers as an undergraduate. She received the Morgan Prize (highest US honor for undergraduate math research), the Schafer Prize, and the Rhodes Scholarship, then entered Stanford for a dual PhD.

**Q6: Axiom Math的商业模式是什么？**

A: 短期内，Axiom计划向对冲基金输出可审计数学因子，按订阅收费；同时将形式化验证技术封装为API，为智能合约、芯片验证等领域提供服务。长期目标包括让AI发现新的具有工业价值的数学定理并申请专利，以及将技术拓展到代码验证领域。

**Q6 (EN): What is Axiom Math's business model?**

A: In the short term, Axiom plans to output auditable mathematical factors to hedge funds on subscription, and package formal verification technology as APIs for smart contracts and chip verification. Long-term goals include having AI discover new industrially valuable mathematical theorems for patenting, and expanding to code verification.

**Q7: 洪乐潼如何看待自己的创业风险？**

A: 洪乐潼将Axiom形容为具有SpaceX式的风险特征——要么成功登月实现超级数学智能，要么彻底失败，中间地带很小。她相信要选最难的问题，甚至需要5到10年才能解决的那种。同时她强调团队的资本效率很高。

**Q7 (EN): How does Hong Letong view her startup's risk?**

A: Hong described Axiom as having a SpaceX-like risk profile—either land on the moon achieving super mathematical intelligence or fail completely, with little middle ground. She believes in tackling the hardest problems, even those requiring 5-10 years. She also emphasized the team's high capital efficiency.

## 来源清单 Sources

1. [s1] 雪球/投资界 · 24岁她融资4亿来自广州 — https://xueqiu.com/1107854878/355547518
2. [s2] 网易新闻 · 这个00后女孩的AI公司太厉害了 — https://www.163.com/dy/article/KGC8B9GN0556BFNP.html
3. [s3] 新浪财经 · 广州数学天才少女斯坦福辍学创业 — https://finance.sina.com.cn/wm/2026-04-05/doc-inhtnckw5307410.shtml
4. [s4] Nextomoro · Carina Hong - Nextomoro Profile — https://nextomoro.com/carina-hong/
5. [s5] AI-Damn · A Teen Prodigy's AI Math Breakthrough — https://ai-damn.com/a-teen-prodigy-s-ai-math-breakthrough-5-papers-accepted-1-4b-raised-1780269648900
6. [s6] 36氪/量子位 · 5篇AI生成的数学论文被接收 — https://36kr.com/p/3828533538395013
7. [s7] MIT Mathematics · Carina Hong '22 Honored With Morgan Prize — https://math.mit.edu/wim/2022/11/18/carina-hong-22-honored-with-morgan-prize/
8. [s8] 百度百科 · 洪乐潼百科 — https://m.baike.com/wiki/%E6%B4%AA%E4%B9%90%E6%BD%BC/7031019220205409544
9. [s9] 广州广播电视台 · 25岁广州天才少女拿下超14亿融资 — http://m.toutiao.com/group/7625979296271614505/
10. [s10] 羊城晚报教育 · 57岁终身教授辞职给24岁学生打工 — http://edu.ycwb.com/2026-01/08/content_53898597.htm
11. [s11] 新智元 · 25岁广州女孩用AI验成了 — https://view.inews.qq.com/k/20260822A098OS00
12. [s12] 雪球 · 洪乐潼创立的Axiom完成2亿美元A轮 — https://xueqiu.com/1222973336/380526440
13. [s13] touchtv · 潮汕00后天才少女缔造AI数学神话 — https://m.itouchtv.cn/article/20527d0a09e1fd52b822361014ce067e
14. [s14] 今日头条 · 斯坦福双博士退学00后广州女孩干出110亿独角兽 — http://m.toutiao.com/group/7672311873982710295/
15. [s15] 小宇宙 · 张小珺商业访谈录洪乐潼专访 — https://www.xiaoyuzhoufm.com/
16. [s16] Axiom Math · Axiom Math官网 — https://axiommath.ai/
17. [s17] Axiom Math · Axiom Math Papers — https://axiommath.ai/papers
18. [s18] GitHub · PrimeGapsLib GitHub — https://github.com/AxiomMath/PrimeGapsLib
19. [s19] Forbes · Forbes 30 Under 30 — https://www.forbes.com/30-under-30/
20. [s20] Rhodes Trust · Rhodes Trust Official — https://www.rhodeshouse.ox.ac.uk/
21. [s21] American Mathematical Society · AMS Morgan Prize — https://www.ams.org/prizes-awards/morgan-prize
22. [s22] AWM · Alice T. Schafer Prize — https://awm-math.org/awards/schafer-prize/
23. [s23] LinkedIn · Ken Ono LinkedIn — https://www.linkedin.com/in/ken-ono/
24. [s24] Quiet Capital · Quiet Capital — https://www.quietcapital.com/
25. [s25] Menlo Ventures · Menlo Ventures — https://menlovc.com/
26. [s26] Stanford · Stanford AI-Conjecturing Program — https://www.stanford.edu/
27. [s27] Lean · Lean Prover — https://leanprover.github.io/
28. [s28] Harmonic · Harmonic AI — https://www.harmonic.fun/
29. [s29] MAA · Putnam Competition — https://www.maa.org/math-competitions/putnam-competition
30. [s30] NeurIPS · NeurIPS 2025 — https://neurips.cc/
31. [s31] HSFZ · 华南师大附中 — https://www.hsfz.net.cn/
32. [s32] Stanford · Knight-Hennessy Scholars — https://knight-hennessy.stanford.edu/
33. [s33] UCL · Gatsby Unit UCL — https://www.gatsby.ucl.ac.uk/
34. [s34] AMS · AMS Erdős Problems — https://www.ams.org/
35. [s35] UVA · UVA AI+Math Hiring — https://www.virginia.edu/
36. [s36] B Capital · B Capital Group — https://www.bcap.com/
37. [s37] Toyota Ventures · Toyota Ventures — https://www.toyotaventures.com/
38. [s38] PitchBook · PitchBook Formal Verification Report — https://pitchbook.com/
39. [s39] 36氪 · DeepSeek与开源AI — https://36kr.com/
40. [s40] AMS · Proc Amer Math Soc — https://www.ams.org/journals/proc/
41. [s41] Elsevier · J Number Theory — https://www.sciencedirect.com/journal/journal-of-number-theory
42. [s42] BSM · Budapest Semesters in Mathematics — https://bsm.budapest.edu/
43. [s43] Axiom Math · 246定理验证论文 — https://primegaps.axiommath.ai/paper/
44. [s44] Ross · Ross Mathematics Program — https://rossprogram.org/

---
本档案机器版由亚洲封面人物官网自动生成：https://coverfigure.com ｜ 永久档案：https://coverfigure.com/acf/ACF-00-00157/geju ｜ 生成时间 Generated：2026-09-11 00:42