基于音频「天文学的AI革命：在海量数据中寻找新天体，让微弱的宇宙信号被看见｜硅谷101年度线下大会」完整转写整理（音频1357秒，约22.6分钟）。本场为英文演讲（讲者为中文母语者），开场一段是节目中文旁白。讲者自述、并经西湖大学官网核实为孙凤梧（Fengwu Sun）：西湖大学天文系助理教授（2026年9月入职），此前在哈佛大学天文台做博士后（再之前在亚利桑那大学史都华天文台），JWST NIRCam（近红外相机）仪器科学团队成员。除标注「经核实」的三处（讲者身份、Lazuli项目、天文学论文LLM痕迹统计）外，文中事实与数字均为讲者或节目旁白口述口径，未做外部核验；转写疑似错误文末统一标注。同源系列前作：AI助理争夺战篇。

先说结论

天文学正在成为第一个「数据产能彻底压垮人类注意力」的基础科学。孙凤梧在硅谷101 Alignment 2026年度大会的这场演讲里给出了一组硬数字：Nancy Grace Roman太空望远镜每天产出约1.4TB数据，约为韦布望远镜（JWST）的20倍；薇拉·鲁宾天文台（Vera Rubin Observatory）每晚产出10—20TB、约100万条「警报」，满负荷运行时将达到每晚700万条；到2030年前后，平方公里阵列（SKA）射电望远镜每天将产生约2PB数据——而全球天文学家总共只有约3万人，讲者算了一笔账：即便全员上阵，每人每晚也得回复约200封警报邮件，「这绝对不可扩展」。结论因此很直接：天文研究流程正在被AI倒逼重构——机器学习已经把引力透镜的发现从「一个博士生一篇论文」变成「一次批量识别约400个」；AI智能体在帮讲者设计韦布望远镜的新观测模式；望远镜四百年来由人做主的运行决策链正在交给AI；自监督算法能把数据深度提升约2.5倍，让6.5米口径「等效于」一台20米望远镜。与此同时，私人资本正按「AI创业公司」的方式造望远镜——Eric Schmidt资助的Lazuli私人空间天文台、可以折叠进星舰一次发射十台的Nautilus方案。对AI从业者，讲者的邀请同样明确：天文数据无标签、全公开、每天新增，是自监督算法最理想的试验场；而「天文学家每天收集的都是来自地球之外的新鲜数据」，恰好是对模型崩溃（model collapse）担忧的解药。

一、100亿美元买「大」与「冷」：韦布望远镜与宇宙的第一缕光

演讲的前半段是一场高质量的「看图说话」，但它解释了一个常被忽略的机制：为什么一台望远镜值100亿美元，答案是「大」与「冷」两个字的物理组合。

讲者从韦布望远镜发布的第一张深场照片讲起——那张由美国总统拜登在白宫揭晓的星系团照片，发布于新冠期间，「是那一年里为数不多的好消息」（讲者语）。照片里几乎每个光点都是一个星系，每个星系里有上千亿颗恒星（讲者口述量级，见文末标注）。这台望远镜1996年完成设计，造了约25年，2021年12月25日由阿丽亚娜5号火箭从库鲁发射场送入距地球150万公里的第二拉格朗日点（L2）。为什么要跑那么远？讲者给了一个车间式的类比：炼钢工人手里举着一小块熔化的金属，背景却是炉膛的高温——地球本身就是一个约300开尔文的黑体，如果你想看红外波段的微弱信号，整个天空都是「炉膛」。只有在L2那个永远晒不到太阳也照不到地球光的地方把仪器冷下来，才能把背景噪声压下去，再用6.5米的大镜子多收光。大镜子负责「看见」，冷负责「不被晃瞎」，这就是韦布的全部设计哲学。

同一套逻辑也解释了那些著名图像：船底座星云的「宇宙悬崖」是恒星风推挤尘埃的前线；哈勃拍的「创造之柱」里尘埃遮住了星光，而韦布的红外波段能穿透尘埃、把尘埃本身从「吸收」变成「发光」——这就是「走向红外的力量」。

讲者自己的研究是「深场」：寻找宇宙的第一缕光。方法有二。一是把望远镜对准一块没有亮星的区域长时间凝视；二是更聪明的办法——利用爱因斯坦广义相对论，星系团的引力势相当于一片放大镜，等于「把韦布这台放大镜对准另一台放大镜（星系团）」，去找宇宙极早期的暗弱天体。这套方法确实奏效：讲者展示的图像中，「在一片什么都没有的地方有一个小红点」，把它的光按波长拆开后确认，这个星系存在于大爆炸后约3亿年——只相当于宇宙138亿年历史的约2%—3%——而且光谱里检测到了氧。讲者特意停顿强调这个细节：这个星系里的恒星爆炸已经造出了氧，「我们此刻呼吸的氧气，就是从大爆炸后3亿年前后开始被造出来的」。发现它时，这是已知最遥远的星系；讲者补充说现在被确认的最远天体比它「还要再远一点」。韦布还带来了一类全新天体：「小红点」（little red dot）——它们看起来几乎在所有特征上都像恒星，但因为太远又太亮，天文学界认为其能量来自黑洞吞噬气体、把引力势能转化为光。

这一段看似与AI无关，其实是整场演讲的地基：正是「大+冷+红外」把望远镜的灵敏度和数据量同时推上数量级，才有了后面的洪水。

二、数据洪水：从每天1.4TB到每天2PB，天文学家只有三万个

后半段演讲的主线是数据速率。讲者先用一个动机铺垫：宇宙是四维的（三维空间加时间演化），天文学家真正想做的是给整个宇宙「拍电影」。韦布已经在做小型版本——连续两年盯着同一片天区，找出「某年亮起、次年熄灭」的点，那就是超新星。而真正为「拍电影」设计的，是2026年8月30日（讲者口径）发射的Nancy Grace Roman太空望远镜：口径2.4米，不如韦布大，但强在探测器——18片4K×4K的探测器拼成3亿像素的焦平面（讲者形容「和手机里类似的器件」），每天产出约1.4TB数据，约为韦布的20倍；而且这些数据发生在150万公里外的L2，必须先从深空传回地面。讲者称Roman「此刻已经在取数据」——屏幕上展示的还是望远镜调焦前的恒星图像，团队正在聚焦和调试。

地面上的薇拉·鲁宾天文台则把「广角」推到极致：它在智利，拥有人类造过的最大相机——32亿像素，约10倍于Roman；望远镜口径8.5米（讲者口述，该镜实际约8.4米，见文末标注），每晚产出10—20TB数据，此刻已在巡天。Rubin的核心玩法是时域巡天：对同一天区反复拍照再做差分，找出「动了的东西」——小行星、暂现源。2025年它发布的第一张图像里，每一个光点都是一个星系。问题在于产出端：Rubin现在每晚产生约100万条警报事件，满负荷时将是每晚700万条；而全球天文学家只有约3万人——「如果我们所有人每晚都处理这些数据，每人要回复约200封警报邮件，这绝对不可扩展。」讲者随后展示了各大望远镜数据速率随时间的曲线：2025年出现跳升拐点（Roman+Rubin），2030年前后Square Kilometer Array（平方公里阵列，地面射电望远镜阵列）加入后，单是它每天就产生约2PB数据。一句话总结：「我们的领域正在经历大洪水，必须依赖AI，因为我们的注意力已经不够用了。」

三、AI的四个切口：找透镜、设计观测、开望远镜、压噪声

讲者给了四个正在发生的AI用例，每个都对应天文流程里一个原本由人占据的环节。

**第一，批量发现引力透镜。**星系团弯曲背景星系的光形成弧状「透镜」，是测量星系距离、研究宇宙膨胀历史的关键工具。讲者回忆：大约十年前，一个博士生在数据里找到一两个这样的系统就够发论文毕业了；如今借助机器学习跑遍所有存档数据，人们一次性批量识别出约400个引力透镜——发现效率从「人年级」压缩到「跑一遍模型」。

**第二，用AI智能体设计新观测模式。**讲者坦言自己「最近是Codex云端智能体的用户」（原话转写存疑，见文末标注）：他把韦布非额定观测模式下的全部数据喂给AI，让AI告诉他自己，色散后的光谱落在探测器的哪些位置——由此「榨」出一种使用韦布望远镜的新观测模式。这个用例有旁证：经西湖大学官网核实，讲者本人正是韦布NIRCam宽场无缝光谱（WFSS）观测模式的在轨调试与定标者，也就是说，他在用AI加速的正是自己的老本行。

**第三，AI接管望远镜的运行决策链。**选目标、指向、云来了要不要关圆顶、发现超新星要不要立刻转过去——讲者强调「这条决策链四百年来都由人类完成」，如今可以交给AI闭环执行。

**第四，自监督压噪声、白捡口径。**讲者的清华同行开发了一个自监督模型：望远镜凝视同一天区时会把数据叠加成深场二维图像，但若去建模噪声随时域的涨落规律，就能从同样的数据里提取比传统方法深约2.5倍的信号。这是什么概念？「相当于你用的不再是一台6.5米的韦布，而是一台20米的望远镜——而20米的望远镜是根本没办法发射上天的。」GPU不够，算法来凑，这是数据侧「等效口径」的扩张。

演讲里还插了一个略带自嘲的行业观察：有研究统计，近年来超过一半的天文学论文带有大语言模型痕迹，但只有0.8%的论文声明了这一点（讲者引用原话：「超过一半的近期论文显示出大语言模型痕迹，但只有0.8%声明了」）。经核实，该统计来自arXiv预印本（编号2609.10664）：2025年天文学论文中约54%带有语言模型痕迹，声明率0.81%，约每66篇带痕迹的论文才有1篇声明。讲者的态度不是指控同行作弊，而是「领域变化太快，我们还不知道最好的做法是什么」。

四、像AI创业公司一样造望远镜：Lazuli、NDA与星舰运力

如果说前三个切口是「用AI做天文」，第四个变化更结构性：天文学家开始像AI创业公司一样造望远镜。

对照系是哈勃：上世纪70年代设计、90年代发射、现在还在写提案申请观测时间，「至少要到2030年代才退役」，NASA甚至还在论证延寿任务。国家旗舰望远镜的迭代周期以三十年计。但私人资本正在改写节奏：讲者介绍了由前Google CEO Eric Schmidt资助的Lazuli——「第一台私人空间天文台」。经核实（Schmidt Sciences在AAS冬季年会的发布）：Lazuli是Eric与Wendy Schmidt夫妇通过Schmidt Sciences资助的3米口径空间望远镜，镜子比哈勃还大，配备日冕仪等三台仪器，目标2029年前发射，由2011年诺贝尔物理学奖得主Saul Perlmutter领衔的团队提出、亚利桑那大学承造其中两台仪器——讲者转写稿中的「202011年诺奖」即为2011年之误。讲者强调的三个细节都指向「创业公司化」：望远镜上几乎所有部件都能买到商业现货；项目团队在发布前签了保密协议（NDA）；「我们对它所知甚少——它可能是某种封闭式望远镜，像OpenAI或Claude那样」（句中Cloudy疑为Claude之转写误）。以NDA方式闭门造镜、以私人资本承担风险、以商业器件压缩成本——这套叙事正是硅谷听众最熟悉的剧本。

另一个方案把「快迭代」推到极致：亚利桑那大学团队提出的Nautilus空间天文台。把十面6.5米级镜子折叠进星舰（Starship）的整流罩，一次发射、在轨展开，相当于「一口气发射十台韦布」。讲者的算术很直白：造十台的平均单台成本远低于只造一台，而且可以更快迭代——「这些年人们已经能造50厘米级的镜子做这类实验」。当发射运力不再是瓶颈，望远镜就从「三十年磨一剑」变成了「批量生产、小步快跑」。

这条链条值得展开：星舰运力+商业现货器件+私人资本 → 望远镜迭代周期从三十年压到两三年 → 观测资源的权力结构开始从国家机构向私人基金会转移 → 二阶影响是开放科学的张力（Lazuli建设期全程NDA，但其官方口径宣称未来数据与申请时间完全开放）→ 对读者的启发是：长周期硬件行业的重构点往往不在算法，而在「运力与器件的商品化」这个不起眼的前置条件——识别你所在行业的「星舰时刻」，比押注哪个算法更重要。

五、为什么天文是AI的天然试验场

收尾时讲者把话锋转向台下的AI研究者，给出了天文对AI界的三重价值，也完成了整场演讲的闭环。

其一，天文数据「完全没有标签」——望远镜拍下什么就是什么，是自监督算法天然的试验场；其二，天文界「把收集到的每一个光子毫无保留地向公众开放」，没有门槛；其三，也是讲者认为最重要的一点：如果说整个学科里只有一个「开放科学」，那就是天文学——「因为天文学望向的是外部，而不是这颗行星上的任何东西。」他顺手回击了当下AI圈最流行的焦虑：如果你担心模型在递归生成的数据上训练会崩溃，「那我可以告诉你，天文学家每天收集的都是来自地球之外的新鲜数据」。天文学有四百年的收集与识别历史，而现在——这句话是全场演讲的落点——「AI正在帮助我们看见超出我们想象的东西」（AI is helping us to see more than what we can imagine）。

把整场演讲压成几条可带走的链条：

  • 注意力瓶颈链条：传感器像素与巡天节奏（Roman 3亿像素、Rubin 32亿像素每晚差分）→ 2025年数据速率跳升一个量级 → 3万天文学家对每晚700万条警报，人工筛选模型崩溃 → AI接管过滤、识别与调度 → 天文学家的角色从「看数据」迁移到「设计系统与验收结果」。任何「采集能力远超分析能力」的行业都会被同样倒逼；判断你自己领域的「数据洪水时点」，比判断算法好坏更值钱。
  • 无标签试验场链条：数据天然无标签+每晚新增+全公开 → 成为自监督算法的理想基准与训练场 → 算法进步反哺天文（清华团队2.5倍深度即是实例）→ 二阶影响是天文成为AI4Science的公共测试床，且物理世界的新鲜数据是模型崩溃的解药。对AI从业者：无标签、持续更新、可公开的数据资产，是这个时代最被低估的战略资源。
  • 设施创业化链条：星舰运力+商业器件+私人资本 → 望远镜从三十年一代变成两三年迭代（Lazuli、Nautilus）→ 观测资源的权力从国家向私人基金会转移，「NDA闭门造镜、数据开放运行」的新平衡正在形成 → 关注点：私人天文台的数据开放承诺能否兑现。
  • 规范真空链条：LLM进入论文写作（54%带痕迹）→ 声明率仅0.8%，约66:1 → 工具渗透速度远超规范建立速度 → 讲者的态度是「不是作弊，是不知道最好的做法」。启发：在规范真空期，先建可验证的流程（数据、代码、提示公开），比道德声明更有效。

接下来值得盯的观察指标：Roman完成调焦后的第一批科学数据与开放节奏；Rubin警报量逼近每晚700万条的进程与社区过滤器生态；SKA建造进度（2030年前后、每天2PB）；Lazuli从NDA走向发射（目标2029年前）的每一步；以及天文学期刊对LLM使用声明的政策落地。天文学的AI革命不是「AI能不能帮科学家」的锦上添花，而是一个四百年的学科在数据洪水下别无选择的基础设施重建——这也是它值得所有行业旁观的原因。

ASR转写校正与存疑说明

本场演讲为英文，Qwen3-ASR整体识别质量良好。已确认校正项已同步应用到文末附录；存疑项保留原文。逐项如下：

  1. 「G.W.C.」「GWC」「GwC」「GWT」→ JWST（James Webb Space Telescope，韦布太空望远镜）。讲者自我介绍段明确说「short name for the James Webb Space Telescope」，全篇统一。
  2. 「Feng Wusun」→ Fengwu Sun（孙凤梧）；「the New West Lake University」→ Westlake University（西湖大学）。经西湖大学官网教师页核实：2026年9月入职任天文系助理教授，此前哈佛大学天文台博士后（2024—2026），JWST NIRCam仪器科学团队成员。
  3. 「Near Cam」→ NIRCam（韦布近红外相机）。
  4. 「硅谷幺零幺Alignment二零二六大会」→ 硅谷101 Alignment 2026大会（数字中文读法转写）。
  5. 「Nobel Prize laureate back in 202011」→ 2011（Saul Perlmutter获诺贝尔物理学奖年份，经外部核实）。
  6. 「Legacy Survey of Space of Time」→ Legacy Survey of Space and Time（LSST）（鲁宾天文台「时空遗产巡天」）。
  7. 「Square Kilometer Array … launch on the ground」→ built on the ground（SKA为建造于地面的射电阵列，非发射入轨；按上下文确认）。
  8. 「a 100 million 100 billion years away」（韦布首图中的星系距离）：疑似口误或转写误差，量级应为数十亿至百亿光年，保留原文。
  9. 「each of the galaxy will have another billion of stars」「one billion stars」：疑为「hundred billion（千亿）」量级之口误，按讲者口述保留。
  10. 「Codex cloud」：疑似指OpenAI Codex云端智能体（或同类云端Agent产品），未确认，保留原文。
  11. 「UME search」（韦布两年时域超新星搜索项目）：项目名转写存疑，未确认，保留原文。
  12. 「like OpenAI or Cloudy」：「Cloudy」疑似「Claude」（Anthropic），未确认，保留原文。
  13. 「8.5 meter」（Rubin望远镜口径）：讲者口述；该望远镜（Simonyi巡天望远镜）主镜口径约8.4米，按口述保留。
  14. 「NASA launched in August thirtieth」（Roman发射时间）：按讲者口径保留；讲者称其已发射、正在调焦取数，未做外部核验。

转写原文（修正版）

以下为整份转写稿全文，已按上方校正表中已确认项（第1—7条）修正（JWST、NIRCam、讲者姓名与校名、大会名、2011年、LSST、SKA地面建造），存疑项保留原文；[起-止] 时间戳保留原样。原转写由 Qwen3-ASR（Intel Arc GPU，120秒分段）生成。

[0s - 120s] 人类正在迎来一个全新的宇宙探索时代。新一代观测设施正在以前所未有的速度收集来自宇宙深处的数据，但一个新的问题也随之出现：这些望远镜每天可能产生数TB甚至数十TB的数据，它们远远超出了科学家人工分析的能力。而AI正在成为天文学家探索宇宙的新工具。当AI开始帮助人类探索宇宙，我们又将发现哪些过去无法看到的秘密呢？接下来就让我们回到硅谷101 Alignment 2026大会现场，来看看这场精彩的演讲。Hello, everyone. My name is Fengwu Sun, and I’m a assistant professor at Westlake University in Hangzhou, China, which is also a very famous city for for AI right now. And before that, I was a postdoc at the Harvard University, and I was a a member of the JWST NIRCam Science Team. So JWST is is short name for the James Webb Space Telescope. That’s the biggest telescope that we launched to into the space in the human history. NIRCam is the primary instrument on on JWST That’s taking all of the beautiful image that you just show. And today, I’m going to show you some progress in our field, including how beautiful our other images are, and and also how AI is changing our field in teaching us how to make movies of the the universe and teaching us how to how to look at our data. Ah, okay. So I I believe most people have been looking at this image. This was one of the first image that JWST released. Ah, is actually released by Joe Biden at at the White House. And at that time, during the COVID, this is one of the few good news over the whole year. And as you can see, nearly everything that in the in the in the picture is a galaxy that are something like a 100 million 100 billion years away from us, and each of the galaxy will have another billion of stars. Ah, and the telescope was designed.

[120s - 240s] Back in 1996, and it took almost 30 years to 25 years to build the telescope, test everything, and launch to the space with the Ariane 5 in Kourou. And these are our last view of the of the of the telescope when before it sent out to the second Lagrangian point on December 25th. And you may wonder why NASA spent 10 billion dollars to build this this telescope, 6.5 meter telescope, even bigger than a school bus, to to the space. I’m going to introduce a little bit on how us extragalactic astronomers are doing research every day. Basically, we are like this this worker in the iron factory. You can see that this man may hold a very small melting metal in his hand, but the background is so high. Consider the Earth. The Earth itself is 300 Kelvin blackbody, and if you want to observe something really in the infrared, the background is so so so high. And only when you cool down the the background, launching the telescope far away from the from the sun, from the from the Earth, in a place that you will never receive any sunlight or Earth light, you will identify a scientific target from a lower background. And that’s the power of being cold. And the whole design of the James Webb Space Telescope is you have a large mirror to collect more light and make it cold to suppress the background noise, and that makes the James Webb Space Telescope. And arguably, this is one of the most beautiful images that people have been seeing from the James Webb Space Telescope. It shows we call it cosmic cliff because you can clearly see the the the the the front of of the star wind. So galaxies are made of a sorry stars are made of a hydrogen gas, and to form hydrogen gas, you have to use the dust grains in in the in the space to to accumulate the hydrogen atoms to let them react. And on one side of the the nebulae that you you you you you are still having a lot of dust that the undergoing star formation.

[240s - 360s] And on the other side, the young stars have already formed, and the the blow away gas. There’s such kind of front line. And similarly, if you can compare the the image taken by Hubble and the James Webb for the star forming regions in in the Milky Way, you can see clear difference. On the left hand side, that is a very iconic view of the we call the Pillar of Creation. The dust here, the main purpose is to to to the the dust here obscure the light coming from the the the the star forming region. But on the right hand side, if you go to the Webb viewings, you definitely see more stars. That because you go to infrared, the light can penetrate the the the dust more easily, and you actually see the dust in emission, but not in absorption anymore. So that’s the power of going to the infrared. Talking about a little bit about my own research, my own research focus on deep field. The main science goal of of our group, our collaboration, is trying to find the first light in the universe. What is the first astronomical object that form in the universe, and how where you can find them, and what how do they look like? And usually we do two ways. One, we we just point telescope to to some place that do not have very bright star, and just integrate as long as possible. Just take data, stare at it, and that will be some image on the on the right. And there are a little bit smarter way that you actually point into a structure we call the galaxy cluster. And because of the Albert Einstein’s general relativity, the gravitational potential of this galaxy cluster will form something like a magnifying glass. So you are actually pointing a magnifying glass, which is JWST, to another magnifying glass, which is galaxy cluster, to find very faint, very far away astronomical object at the beginning of the universe. And actually we are quite successful with JWST. That’s the whole purpose why JWST was was built. So this is a the the one of the deepest image that that we our team receive, and just.

[360s - 480s] In the middle of nowhere, we identify a small red dot there, and we split light at different wavelengths and confirmed this galaxy was existed 300 million years after the Big Bang. Right now, the universe has a history of roughly 13.8 billion years, and this is roughly two or three percent of the at two or three percent of the age when the when the universe just formed. And most amazingly, if you observe at other different different wavelengths, you actually get oxygen detection in in the spectrum. And think about this: the galaxy is already producing oxygen by exploding stars at that time and produce all of the oxygen that we are breathing today in this room. Okay, starting from the 300 million years after the Big Bang. And that was not that when we discuss when discovered this this object, this was the the most distant galaxy we know. And right now, this is the most distant object that we have confirmed so far, a little bit more further. And still in the vast universe with you know black background, and you you you you look through each of the object, measure each of their distance, and you finally find one target through such kind of large data set. Another interesting that astronomers are paying attention these days is if you go to look back at the first image that the White House released, actually in this region there’s a small red compact object in this in this region. We were very curious about what that is when the data was released, and later people astronomers come up with a name for that. And you probably know that astronomers are very bad in giving names, and we just call them little red dot. And because it’s little and also red and also looks like dot, this is a type of new type of star that JWST identified. They look like stars, exactly look like stars in nearly all of the features, but actually we believe because they are so far away and so bright.

[480s - 600s] They are actually powered by black hole, so it’s considered that the black hole is creating gas, just eating gas, and then when the gas fall into the black hole, is emitting light, turn the gravitational potential into into the the light. If you go back to the the place that near to us, you will see beautiful galaxy like this, and you see so many bubbles. Each of the bubbles was blown away by a supernovae that the star died and explode, maybe a few hundred million years ago, and they also produce oxygen, carbon that make all of the life on the on the on the on this planet. And speaking about my my favorite image, this is actually Jupiter. So since 1609, when the first astronomer Galileo pointed his telescope into the sky, this is one of the first object that he observed, and since then every telescope actually observed the Jupiter, and this is the best image of Jupiter I have ever seen. This is the the Great Red Spot, and on the top and bottom that’s actually the aurora light, so the northern light. If you go if you stand on the on the Jupiter at the the time, you will see aurora above your head, and this is the the Saturn that you can see the the the ring and the gap. Okay, so I’m showing a lots of image. They are they are they are 2D image, right? But we we know that the universe is four dimensional, right? You have space three axes, and also is evolve as function of time, so it’s it’s 4D, and astronomers are making working very hard to try to make a movie of the the whole universe, and perhaps that’s very informative for the world model that we just discuss, right? And and think about this, we are actually we are actually working on this UME search the James Webb Space Telescope, so you can look at the same the region same region of the sky over the two years, and you can identify some dot that goes off in one year and and fade away in the other year, and we call that supernovae, basically a star end of their life.

[600s - 720s] And one of the new telescope, space telescope that NASA is launching, NASA launched in August thirtieth, is a so-called Nancy Grace Roman Space Telescope. And one of the main purposes of a Roman Space Telescope is try to make a movie of the universe by finding a supernovae. Which by finding them, we sort of know they are they are like candles. You can put candles at different places and measure the the distance to it, so it can actually help us to map the universe. So this is the artist’s impression of the Roman Space Telescope. It’s not as large as the JWST. It’s only two point four meter across, but the the the power of the Roman Space Telescope is has very beautiful detectors. So each of them, this is the we call it focal plane. They have eighteen detector, basically the same detector, similar detector as you use on your phone. Each of them is a four K times four K pixel. So combine all of these eighteen chips together, you are actually get a three hundred million pixel at once. And consider if it keep observing the sky, the data that it will produce is roughly one point four terabytes per day, and that’s roughly twenty times of a JWST how JWST is producing data. And remember, these data do not happen on the ground. They happen at L two, which is one point five million kilometer away, and you have to transfer the data from the deep space to the ground. It’s actually already taking data right now. So these are data that before the telescope is focused. So the star, every image that you see here is a is a star, but before the focus. So the the team is working very hard in focusing the telescope and taking the real data, make sure telescope works. As for for us who are searching for the first light of universe, we care about how Roman is observing the the universe at the very beginning. So here I’m so showing you a numerical simulation obtained by astronomers, and all of the background image are the gas, the hydrogen at at at that time. And you know that the galaxy, they are like cities on the on on on the

[720s - 840s] On the map, they are not randomly they are randomly distributed, but they form structures, right? And all of these these these sign circles represent one galaxy that should be detected by Roman. And think about it, this is the first image of JWST that you just saw, and JWST can only observe such a very small patch of sky, but Roman will get get a very large view of it. And basically, that’s that that shows how the how the data rate expansion in the in the astronomy is going on or the upcoming years. Another very important infrastructure in astronomy is so called Vera Rubin Observatory, which is conducting a survey called the Legacy Survey of Space and Time. So this is a telescope that in in the mountain of Chile, and it’s ground based telescope, not launching into space. It’s doing survey every night right now. The special thing with this this telescope is the camera is the biggest camera that human being ever built. So it’s a 3.2 gigapixels, which is 10 times more than than what they have Roman. So it was producing is is actually producing data right now at the rate of 10 or 20 terabytes per day. And the telescope is 8.5 meter, so it’s it’s one of the largest telescope that built on the on the ground. And you can see how the mirror of the telescope, how how large is this camera? This is camera compared with the engineering team on the on on the ground. And the the the observatory has been observing. This was a data taken in last year, and this is the first image of the of the telescope release. You can see these beautiful galaxies, clusters, and each of the each of the light you see is a galaxy. Each of galaxy have one billion stars there at different distance. One way that Rubin is observing is observing we call it in time domain to compare the the difference of image. As you can see here, there are moving star go across it. We call it asteroid. So one key thing in astronomy is to identify things move in the sky because most of the thing they do not.

[840s - 960s] In the sky, and that’s how Roman is identifying asteroids within the the first data set. And Rubin is actually producing roughly one million alerts, these kind of events, every night. And when Rubin is run at full speed, it will go to produce seven million alerts every night. But consider on the planet, there’s only thirty thousand astronomers. So if all of us should work on the data every night, we have to reply two hundred emails about the the alerts. So it’s it’s definitely not scalable. And I’m showing here how the data rate of different telescopes, from major and ground and space telescope, as a function of time. Over 2025, there’s a major data rate flood because of Roman, because of Rubin. And if the Square Kilometer Array, which is a radio telescope, built on the ground around 2030, this data this telescope will produce roughly two petabytes of data, PB of data, every day. And basically, no astronomer will go to analyze all of data. So what I want to say is our field is undergoing a major data flood, and we we have to rely on AI because we we are running out of our attention. And I want to give you a few cases why we are what we are doing these days by using AI in our field. So the first thing is a gravitational lensing that you see the first image of the JWST, and this galaxy cluster is bending the light from the background galaxy into this beautiful arc like like shown here. Maybe 10 years ago, one PhD student can identify one or two system like this in the data, and then publish a paper. You publish two paper, you get the PhD degree. But but but these days, you because such kind of big data rate, you you you have to rely on machine learning method to go through all of data that people have been taking and publish 400 ish gravitational lens together to identify these.

[960s - 1080s] Kind of things you can measure the the the distance of galaxy and that is very powerful for understanding how the universe is expanding as a function of time. On my side, I’m also have a user of Codex cloud these days. So actually, I’m this is one of my research. Actually, I can use the the data from JWST in a under use observing mode and feed all my data into all the data to to AI and let AI tell me where are the spectrum the the dispersed light land on the detector and help me to get a new observing mode a new way to use the James Webb Space Telescope to observe the universe. And people have been using the AI to run run the telescope. Basically, these these jobs was done by by human beings over past 400 years. You you have to pick a target and point your telescope to to to the target and then if there is a cloud going on, okay, you have to point your telescope other places or or or close your your dome or if you find something that is going off like a supernova, you point the telescope to to the exact place. All of this decision chain was was by human being, but these days you can you can do it in with AI. And this is from my friends at Tsinghua. They develop a a self-supervised model that actually when we look at stare when we point our telescope in a certain piece of sky, we are we we stack the data together so you have a very deep 2D image of it. But if you look at the temporal the the the the the the time domain of of the noise, actually you can understand how noise is fluctuating. So you are getting deeper data than what you are getting from the traditional method. And these people using this method, they can get the the data deeper by roughly a 2.5 factor of 2.5. So that means you are using JWST.

[1080s - 1200s] We see not only like a 6.5 meter telescope, but something like 20 meter telescope that there’s no way to launch that into space right now. And our field is being heavily impacted by by large language model over the last two years or so. And there’s a recent paper saying that more than half the recent astronomy paper are written by large language models. And one interesting quote from the from the from the abstract is more than half of the recent paper therefore show large language model trace, but only 0.8 percent of them declare that in the paper. So I’m not saying that astronomers are cheating with to each other at these days, but actually the field is changing so fast that we don’t know what is the best way to proceed. And another interesting thing is I’m talking about the data, but talking about the the place that generate data, which is telescope. These days astronomers are building start to build telescope like AI startup, which is very interesting. I want to share to you. So when astronomers are computing, you know that the computer evolve very fast over the last 60 years. But when you talk about observing, let’s say Hubble Space Telescope, this is telescope design in 1970s, launched in the space in 1990s. We are still writing the telescope proposal to apply for the time and use it right now, and it won’t retire until 2030 at least. And the NASA is still thinking about the mission to save Hubble to make it run longer. And right now the the private fund is kicking in for for building telescope. So this is the this is project funded by Eric Schmidt, the the former Google CEO, is called Lazuli. Is this is the first private space observatory that will be launched to the space? They collaborate with people like Saul Perlmutter, who is a Nobel Prize laureate back in 2011, and also people at University of Arizona to build all of the the the telescope and going to launch it in two or three years. Nearly all of the things on the telescope they can buy it as a commercial product, and the all of the the project team signed NDA before releasing the telescope idea to the people.

[1200s - 1320s] We know very little bit about this telescope. It’s probably some close, close telescope like OpenAI or Cloudy. But we believe this telescope will will bring huge impact to to the to the to our research in astrophysics. And another way to look at this is is this mission concept called Nautilus Space Observatory, which is by a professor at the University of Arizona. Think about the Starship. If you fold ten mirrors, each of the size is six point five meter, to the Starship, and when it get launched, each of the mirrors get get expanded when they when they go to the space. And basically, you will launch ten telescope like like James Webb Space Telescope at the same time. And and when you build with ten telescope, you you can do faster iteration. Okay, these years we can people people start to build telescope like fifty centimeters, and they can do this kind of experiment. And when you build ten telescope like in this way and launch in one shot, the the the price will be much cheaper than building the average price will be much cheaper than building just one, as you can can imagine. So this is my final slide. I’m I’m very happy to report some of the recent progress that we made in the in the field of astronomy. And I would like to align with with with the AI researchers leaders in in this room. First of all, as an astronomer, I believe that we definitely need AI to prepare for the ongoing data flood. And this is already at the scale that no no of the astronomical team can handle. And we definitely need that for the for the this this discovery and also the breakthrough make have to make decision. And other interesting thing with our data is our data comes with no label at all, so it’s a very good test ground for all of the self-supervising algorithm. And astronomer release every photon that we collect to the public without any without without any barriers. And the universe is giving us giving us this this gift. And another thing very importantly is if there’s only one open science in

[1320s - 1357s] Among the whole discipline, then that will be astronomy, because astronomy is looking outward, not on anything on the ground in this planet. If you believe that model collapse when trained on recursively generated data, then I will tell you that astronomers are collecting fresh data every day and from nowhere from the planet. And astronomy is modern astronomy is a field that has already get 400 years of history. We collect a lot of data and start to identify things. Right now, AI is helping us to see more than what we can imagine, and that’s all I want to share. Thank you very much.