Today is another robot special. The guest is Xie Chen, founder and CEO of Kuanglun Intelligence, who previously served as the head of autonomous driving simulation at Nvidia, Cruise, and NIO. Our topic is very specific: simulation and synthetic data. Today's embodied intelligence has not yet found an effective formula for scaling law, and data is a key bottleneck. Wang He, the founder of Galaxy General and guest of our 106th episode, mentioned that real data accounts for only 1% of their training data, and synthetic data is the mainstay. In today's episode, I talked with Xie Chen about the practical details of simulation and synthetic data. 02:00 Quick Q&A; starts 02:48 High-frequency vocabulary analysis: Sim2Real (from simulation to reality), Sim2Real gap, synthetic data 04:31 From Cruise to Nvidia to NIO, how to do synthetic data and simulation? 14:11 What is the specific process of making synthetic data? What is the ratio of synthetic data to real data? 16:17 The difference between intelligent driving and embodied intelligence in synthetic data (intelligent driving is a visual game, and physical interaction is the most critical for embodied intelligence) 32:41 What is the physical Real2Sim (reality to simulation) workflow? How to evaluate a successful simulation? Key technical nodes? 46:18 Physical Intelligence (π) has a dilemma attitude towards simulation and synthetic data 48:55 Spicy comments on Meta's $30 billion acquisition of Scale AI and Alexandr Wang's extremely aggressive behavior 53:57 Current bottlenecks in synthetic data 55:25 Global embodied intelligence industry chain Mapping: Hardware company (Yushu) Base model company (π, Skild, Nvidia and DeepMind) Software and hardware integration companies landing in vertical fields (Figure, Tesla Optimas, The Bot Company) Companies that focus on simulation for end-to-end landing (Kuanglun) ("Tesla Optimas' management culture is completely different from π") 01:09:22 There are entrepreneurial opportunities in the embodied model layer in the United States. In my opinion, ByteDance, Xiaomi, and Ideal are more suitable for being the "brain" in China. 01:15:33 Lao Huang said internally: NV is a simulation company 01:21:25 The final model should be cross-universe, cross-world, and cross-ontology (improving the ability to cross the universe is essentially improving generalization) 01:23:28 The embodied intelligence industry is still in the GPT-1 stage and has not yet found a formula for scaling law. 01:28:21 I just started my business and started learning from embodied undergraduate studies. 01:37:37 Final quick Q&A; [Robot Special] Detailed explanation of robot base models and VLA classic paper - "Humans are the most intelligent VLA" Talking with Wang He about the academic marginal history of embodied intelligence and the artificial chaos after the capital bombing

Original title: 109. 机器人遭遇数据荒？与谢晨聊：仿真与合成数据、Meta天价收购和Alexandr Wang

Original description: <figure><img src="https://image.xyzcdn.net/Flo18nNUSP7OUNlTf8UgCdHxio6O.jpg" /></figure><p>今天又是一集机器人专场。嘉宾是光轮智能创始人兼CEO谢晨，他曾在英伟达、Cruise及蔚来汽车担任自动驾驶仿真负责人。我们的话题非常具体，即:仿真与合成数据。</p><p>今天的具身智能尚且没有找到scaling law的有效配方，其中，数据是一个关键卡点。我们106集的嘉宾银河通用创始人王鹤就提到，真实数据在他们的训练数据比重仅仅1%，合成数据挑起大梁。</p><p>今天这集节目，我与谢晨聊了聊仿真与合成数据的实操细节。</p><figure><img src="https://image.xyzcdn.net/Fm2F9n8vMm_n-xafqMi98xs3T3K8.png" /></figure><figure><img src="https://image.xyzcdn.net/FvO83aFZZwjNNrtt9QcAh3xGuJzF.png" /></figure><figure><img src="https://image.xyzcdn.net/FvV-R5FBydYHGZAMyXAV1K1A9iJT.png" /></figure><p>02:00 开始的快问快答</p><p>02:48 高频词汇解析：Sim2Real（从仿真到现实）、Sim2Real的gap、合成数据</p><p>04:31 从Cruise到英伟达到蔚来，怎么做合成数据和仿真？</p><p>14:11 制作合成数据的具体流程？合成数据与真实数据的配比？</p><p>16:17 在合成数据上，智能驾驶和具身智能的区别（智能驾驶是视觉的游戏，具身智能的物理交互最关键）</p><p>32:41 物理的Real2Sim（真实到仿真）工作流是怎样的？怎么评估成功的仿真？关键技术节点？</p><p>46:18 Physical Intelligence（π）对仿真与合成数据的两难态度</p><p>48:55 辣评Meta 300亿美金收购Scale AI和极其aggressive的Alexandr Wang</p><p>53:57 合成数据目前面临的瓶颈</p><p>55:25 全球具身智能产业链Mapping：</p><p>硬件公司（宇树）</p><p>基座模型公司（π、Skild、英伟达和DeepMind）</p><p>在垂域落地的软硬结合公司（Figure，特斯拉Optimas、The Bot Company）</p><p>以仿真为中心做端到端落地的公司（光轮）</p><p>（“特斯拉Optimas的管理文化和π完全不一样”）</p><p>01:09:22 美国存在具身模型层的创业机会，中国在我看来字节、小米、理想更适合做“大脑”</p><p>01:15:33 老黄在内部说：NV is a simulation company</p><p>01:21:25 终局的模型应该是是跨宇宙、跨世界、跨本体（提升跨宇宙的能力，本质是提升泛化性）</p><p>01:23:28 具身智能的产业还在GPT-1阶段，还没找到scaling law的配方</p><p>01:28:21 我创业刚开始，从具身的本科开始学起</p><p>01:37:37 最后的快问快答</p><figure><img src="https://image.xyzcdn.net/FvVbUNblF7FHIjfdp3MmmbAdLZ8G.png" /></figure><p>【机器人专场】</p><p><a href="https://www.xiaoyuzhoufm.com/episodes/67f28c6e0decaeb0943fb14a" rel="noopener noreferrer nofollow" target="_blank">逐篇讲解机器人基座模型和VLA经典论文——“人就是最智能的VLA”</a></p><p><a href="https://www.xiaoyuzhoufm.com/episodes/6857f2174abe6e29cb65d76e" rel="noopener noreferrer nofollow" target="_blank">和王鹤聊，具身智能的学术边缘史和资本轰炸后的人为乱象</a></p><figure><img src="https://image.xyzcdn.net/Fn7o36NtUYpCM_rQiFj1LW-TIwk8.JPG" /></figure>

Today's guest is Dr. Yu Kai, founder and CEO of Horizon Robotics. In the past 49 years of his life, **he has navigated the academic circles of Germany and the United States, the Chinese internet circle, the venture capital circle, the capital circle, and the automotive circle.** In each circle and "jianghu" (江湖, a term for the martial arts world or a specific community), he started as an unknown nobody and leveled up. In the end, he did well in every circle. A former senior executive who has dealt with him commented, **Yu Kai is one of the most socially intelligent scientists.** Yu Kai graduated from Nanjing University and the University of Munich. After graduation, he worked for Siemens and NEC Research Institute. In 2012, he returned to China to join Baidu, and in 2015, he left to found Horizon Robotics. Coincidentally, 2025 marks the 10th anniversary of Horizon Robotics' founding. **This year, I spoke with Dr. Yu Kai twice, and this episode is his oral history.** With the outbreak of the large language model wave, more artificial intelligence scientists are pouring into the entrepreneurial track from universities. Yu Kai's entrepreneurial philosophy may give everyone some inspiration - **entrepreneurship is not only about technology and business, nor is it just about conflict and competition, but also a "jianghu" story of people coming and going.** Just like Zhang Zuolin's line in the TV series "Young Marshal": "Jianghu is not about fighting, it's about social skills." In 2025, we will progress together with AI! 03:06 **Entering the Academic Jianghu** Initially unknown in academic circles, a fortune teller said I would be "unknown and work in vain" before the age of 24. I have published over 100 papers. I am very intoxicated, and I would appreciate my previous papers late at night. Stories of meeting Geoffrey Hinton, Yann LeCun, and Andrew Ng. There was a very silent person sitting opposite me, and no one paid attention to him. He was eating alone - this person is Richard Sutton, who recently won the Turing Award. 31:18 **Entering the Internet Jianghu Again** I should be the first Chinese AI scholar who studied in the United States to return to China. I immediately wrote to Geoffrey Hinton, and he replied: Kai, that's great, but would you mind if I also asked other companies? The authorization I received at the time was to bid up to $24 million. After $24 million, I had to discuss each bid with the domestic side. In order to win with a small probability, I took the lead and offered $12 million. "Hey, you see Geoffrey Hinton doesn't seem to show up at meetings often, what's he doing...?" I asked him: Hey, Andrew (Andrew Ng), what are you doing? How is everything going? I started to test him. Andrew Ng was shocked! He said: You tricked me into Baidu, and you ran away yourself, that's not cool! 51:19 **Entering the Entrepreneurial Jianghu Again** I made three investments: I bought Nvidia, I bought Tesla, and I wholeheartedly invested in Horizon Robotics. This guy told me: Brother, you know what? My status at home now depends on that sentence of yours! When Horizon Robotics was just founded, I looked at it, and Nvidia was only a $10.7 billion company, now it's 3 trillion! What is one frustration that Andrew Ng had while leading Google Brain at Google? Not being able to buy GPUs! Consensus is either wrong or worthless. What is your business secret? What is something you saw that others didn't see? Is there a bug in this world? Is there a narrow door to the future that most people haven't paid attention to? 01:11:21 **Entering the Capital Jianghu Too** We didn't write a single page of BP (business plan) and raised the first round. I thought: Wow, life is so easy! As a result, in the second round, I discovered that I met with 50-60 institutions, and none of them placed an order. It was particularly tough... no one understood... What I said was almost dry... for a long time... in the dark... and no one was moved. I set an iron rule: the first time I meet with an investor, it must not be in his office, but in my office. I kept pretending! I said: I really don't have time, I am just a focused, low-EQ scientist, tinkering with my own things, and I'm too lazy to deal with you. We created the legendary 12 mini-rounds of Series C, taking 1.6 billion US dollars in one go - this is also a counter-consensus - without adding 1 cent of valuation in the middle. Wow, Horizon Robotics actually has 102 shareholder investment institutions, I don't even know how I managed to get them. 01:21:39 **Switching to the Automotive Jianghu** Scientists often have this problem when starting a business: 360-degree scanning. After Zeng Ming's class, many of our classmates went back and cut directions and laid off teams. One night while sleeping, I suddenly woke up in shock: Damn, this is not right! With Changan: deliberately losing the game, you have to lose elegantly, discreetly, and deliberately. With Li Xiang: Li Xiang told me when we climbed the mountain in early 2019: You should focus on the automotive direction. With He Xiaopeng: I haven't conquered Xiaopeng yet - sometimes you have to attack head-on, sometimes you have to go around. With Wang Chuanfu: We seized the opportunity window, which is equivalent to opening a small crack in the door, and we rushed in with a "whoosh." 02:09:48 **I am not a Jianghu Person** My role model for leaders is Liu Bang. Do you know who my favorite character is in these movies? Rhett Butler in "Gone with the Wind." My surname is Yu, and the company's name is Horizon - leeway, leeway, always leave leeway for people and things. Intelligent driving: OEMs (original equipment manufacturers) will not self-develop in the future, it is a standardized function. 3 years to complete 100% hands-off, 5 years to complete 100% eyes-off, 10 years to complete 100% minds-off. What is the death door? The CUDA of robots. Next-generation chip innovation. 02:35:23 **Final Quick Q&A** I think the world is a pre-written program, and everyone is acting according to the script. 02:39:26 **Additional Extras** Teaching skills: If you are determined to resign, don't say anything bad about the company. Yan Junjie's hairstyle is like mine (joke). Andrew Ng and I seriously discussed entrepreneurship in the United States. I drank Maotai with a college graduate to adjust him, unlike Li Xiang who is decisive. Why is the WeChat profile picture Guan Yu? [From Steam Engine to Self-Driving] Series 《3-Hour Interview with Li Xiang (Podcast Version): Otaku, AI, Family, Games and Ladder》 《Chatting with He Xiaopeng about FSD, "Swimming in a Sea of Blood," Heroes and Dogs in Troubled Times》 《Dialogue with Ola Källenius, Global CEO of Mercedes-Benz: CEO in Transition and 139-Year-Old Mercedes-Benz in Transition》 《Chatting with Lou Tiancheng about Robotaxi and ACRush: "The better L2 is done, the further away L4 is"》 Text version of this episode: 《Dialogue with Yu Kai: The World is More Than Swords and Shadows, It's a Jianghu Story of People Coming and Going》

Original title: 108. 余凯口述30年史：世界不止刀光剑影，是一部人来人往的江湖故事

Original description: <figure><img src="https://image.xyzcdn.net/Flo18nNUSP7OUNlTf8UgCdHxio6O.jpg" /></figure><p>今天的嘉宾是地平线创始人兼CEO余凯博士。</p><p>在过去49年人生中，<strong>他一路闯关过德美学术圈、中国互联网圈、创投圈、资本圈、汽车圈。</strong>在每个圈子和江湖，都从籍籍无名的无名小卒开始升级打怪。到最后，在每个圈子，他混得都不错。</p><p>一位与他打过交道的前企业高层评价，<strong>余凯是科学家里非常具有社会智慧的一位。</strong></p><p>余凯毕业于南京大学和慕尼黑大学，毕业后，先后就职西门子、NEC研究院，于2012年回国加入百度，又于2015年离职创立地平线。</p><p>很巧的是，2025年正好是地平线创立10年。<strong>今年上半年，我与余凯博士聊了两次，这集节目是他的一部口述史</strong>。</p><p>随着大语言模型浪潮爆发，更多人工智能科学家从高校系统涌入创业轨道。余凯的创业观，也许能给大家一些启示——<strong>创业不仅是技术和商业，也不仅仅有刀光剑影，更是一部人来人往的江湖故事。</strong></p><p>就像电视剧《少帅》张作霖的台词：“江湖不是打打杀杀，江湖是人情世故。”</p><p>2025年，我们和AI共同进步！</p><figure><img src="https://image.xyzcdn.net/Fm2F9n8vMm_n-xafqMi98xs3T3K8.png" /></figure><figure><img src="https://image.xyzcdn.net/FtYwXS7GiDXB0ddOvtgKa35Ak6qg.png" /></figure><figure><img src="https://image.xyzcdn.net/FvV-R5FBydYHGZAMyXAV1K1A9iJT.png" /></figure><blockquote>03:06 <strong>初入学术江湖</strong></blockquote><p>一开始在学术圈籍籍无名，算命先生说我24岁前“籍籍无名，劳而无功”</p><p>发表过100篇论文，我很陶醉，夜深人静都会翻我以前的paper自我欣赏</p><p>结识Geoffrey Hinton、Yann LeCun、吴恩达的故事</p><p>我这边的对面坐了一个人特别的沉默，没人搭理他，一个人在那吃闷饭——这个人叫Richard Sutton，前段时间拿了图灵奖</p><blockquote>31:18 <strong>再入互联网江湖</strong></blockquote><p>我应该是旅美人工智能华人学者第一个回国的</p><p>我立刻就跟Geoffrey Hinton写信，他回信：Kai，挺好的，但你介不介意我也问一下其他公司？</p><p>我当时拿到的授权是，最高出到2400万美金，2400美金以后，每一次出价就要跟国内商量</p><p>我为了小概率能赢，抢先第一个出价，1200万美金</p><p>“哎呀，你看Geoffrey Hinton开会好像不太出现啊，他在干嘛…？”</p><p>我就问他：唉，Andrew（吴恩达）你在干嘛？各方面怎么样？开始试探他</p><p>吴恩达一下子震惊到了！说：你小子把我忽悠到百度，你自己跑掉，太不够意思了吧？</p><blockquote>51:19 <strong>又入创业江湖</strong></blockquote><p>我做了3个投资：买了英伟达，买了特斯拉，全身心把我投到地平线</p><p>这个哥们跟我讲：兄弟，你知道吗？我现在在我家的地位，就靠你那句话！</p><p>地平线刚创立那一天我看了一下，英伟达才是一个107亿美金公司，现在是3万亿！</p><p>吴恩达在Google lead谷歌大脑，有一个frustration（沮丧）是什么？不能买GPU！</p><p>共识要么是错的，要么是没价值的</p><p>你的商业的secret是什么？有什么东西你看见了别人没有看见？这个世界是不是有Bug？这个世界是不是有通向未来的窄门，而大部分人没有关注到？</p><blockquote>01:11:21 <strong>也入资本江湖</strong></blockquote><p>我们一页BP没写，就融了第一轮，我觉得：哎呀，Life is so easy！</p><p>结果第二轮就发现，见了50-60家机构，没一个下单。特别tough……没人理解……</p><p>我说的简直是口干舌燥……地老天荒……昏天黑地……也没人动心</p><p>我定了一个铁律：我跟投资人第一次见面，绝不能在他办公室，一定要在我办公室</p><p>我继续装！我说：我真的没时间，我就是一个专注的、情商低的科学家，正在倒腾我自己的事情，懒得理你</p><p>我们创造了C轮业界传奇的12小轮，一把拿了16亿美金——这也是一个反共识——中间没有加1分钱估值</p><p>哇，地平线竟然有102家股东投资机构，我都不知道我怎么磕出来的</p><blockquote>01:21:39<strong> 转战汽车江湖</strong></blockquote><p>科学家创业通常有这个问题：360度扫射</p><p>曾鸣那堂课上完以后，我们班好多同学回去都去砍方向、裁团队</p><p>有天晚上睡觉，我梦中突然一惊：我靠，这样不对啊！</p><p>和长安：故意输球，你们要优雅地、不露声色地、故意地输啊</p><p>和李想：李想在2019年初，我们俩爬山他讲：你应该聚焦汽车方向</p><p>和何小鹏：我现在还没有磕下小鹏————有的时候你要强攻，有的时候你要迂回</p><p>和王传福：我们逮着机会窗口，相当于这个门开一个小缝，咱们就呲溜一声冲进去</p><blockquote>02:09:48<strong> 我不是江湖人</strong></blockquote><p>领导者我的role model是刘邦</p><p>电影这些角色，你知道我最喜欢谁吗？《飘》里的白瑞德</p><p>我的名字姓余，公司的名字地平线——余地，余地，做人做事永远要留有余地</p><p>智能驾驶：主机厂未来不会自研，它是一个标准化的功能</p><p>3年完成100%hands-off，5年完成100%eyes-off，10年完成100%minds-off</p><p>死门是什么？</p><p>机器人的CUDA</p><p>下一代芯片创新</p><blockquote>02:35:23<strong> 最后的快问快答</strong></blockquote><p>这个世界我认为是写好了程序，每个人都是按照剧本来演</p><blockquote>02:39:26 <strong>补充花絮</strong></blockquote><p>传授技巧：如果你决心离职，不要说公司任何不好</p><p>闫俊杰的发型像我（玩笑）</p><p>我和吴恩达在美国serious讨论过创业</p><p>我为了调一个校招生喝茅台，不像李想手起刀落</p><p>微信头像为什么关公？</p><figure><img src="https://image.xyzcdn.net/FvVbUNblF7FHIjfdp3MmmbAdLZ8G.png" /></figure><p>【从蒸汽机到无人驾驶】系列</p><p><a href="https://www.xiaoyuzhoufm.com/episodes/67769bd815a5fd520e8fa318">《对李想的3小时访谈（播客版）：宅男、AI、家庭、游戏和天梯》</a></p><p><a href="https://www.xiaoyuzhoufm.com/episodes/6695032837236c546e4c2e0f">《和何小鹏聊，FSD、“在血海游泳”、乱世中的英雄与狗熊》</a></p><p><a href="https://www.xiaoyuzhoufm.com/episodes/68300e93fcbc2e206b58eb2b">《对话奔驰全球CEO康林松：转型期CEO和转型之中的139岁奔驰》</a></p><p><a href="https://www.xiaoyuzhoufm.com/episodes/66bdb98233591c27be49e931">《和楼天城聊聊Robotaxi和ACRush：“L2做得越厉害，离L4越远”》</a></p><p>本集文字版：<a href="https://mp.weixin.qq.com/s/TQ-pOkEi412kYvrJE2F1TQ">《对话余凯：世界不止刀光剑影，是一部人来人往的江湖故事》</a></p><figure><img src="https://image.xyzcdn.net/Fn7o36NtUYpCM_rQiFj1LW-TIwk8.JPG" /></figure>

This episode doesn't have grand narratives; it's quite casual. The guest is Meng Qiu, founding partner of Qingliu Capital and former Technology VP of Baidu. Longtime listeners of "Business Interview" may know that Meng Qiu basically returns every year to catch up on the current venture capital climate and her own life. In the wolf-culture-dominated Chinese investor circle, Meng Qiu has always been very Buddhist and Taoist. This episode is even more relaxed. She frankly says that work is quite boring, so after discussing serious topics, we also talked about reading, travel, movies, and girls' chattering. (This episode was recorded at the end of April) Our podcast program is first launched on Tencent News. You can go and follow us to get the program information and more news as soon as possible :) 02:00 Has the capital winter of 2025 passed? No… 04:00 Has the emergence of DeepSeek made AI application entrepreneurship more active? No… 10:45 Current experience of various Bots: fortune teller? Simp? Especially commenting on WeChat, Yuanbao, and Xiaohongshu 25:28 Discuss how to make Agent in WeChat? Is a universal Agent established? 31:25 Entrepreneurial opportunities and entrepreneurs for vertical Agents 35:52 Current organizations tend to be small, which may benefit young entrepreneurs 37:42 Why are organizations smaller, but the amount of financing is higher? 38:18 In addition to Agent, what I am also looking at is embodied intelligence (simulators are very important) 43:57 Wearable devices 54:54 Large model companies 58:31 Work has been very boring in the past two years, my travel journey 01:03:55 My reading journey 01:12:34 Talking about the film and television industry (Meng Qiu is an independent director of China Film), "Good Things" and feminism Meng Qiu's previous programs: "1. Chatting with investor Meng Qiu about California, the investment cold wave, and Lin Daiyu" "21. Large models and the real market temperature from the perspective of investors｜Chatting with Meng Qiu about ChatGPT" "65. Has the key to venture capital failed? Chatting with Meng Qiu: Hibernation, a game with fewer people, and rodents" 【More Information】 Contact us: Weibo @张小珺-Benita For more information, please follow the official account: 张小珺

Original title: 107. 和梦秋的catch-up：创投挺无聊，也聊聊旅行读书和女性主义

Original description: <figure><img src="https://image.xyzcdn.net/Flo18nNUSP7OUNlTf8UgCdHxio6O.jpg" /></figure><p>这集没有宏大叙事，相当随性。嘉宾是清流资本创始合伙人、百度前技术VP梦秋。</p><p>关注《商业访谈录》比较久的朋友可能知道，梦秋基本每年都会来返场一次，和我们一起catch-up当下的创投水温以及她自己的生活。在狼性文化蓬勃的中国投资人圈里，梦秋一直是很佛系也很道家的存在。</p><p>这一集更是松弛，她直言工作挺无聊，所以在聊了正经话题以后，我们也聊了聊读书、旅行、观影和女生的碎碎念。</p><p>（本次节目录制在4月底）</p><figure><img src="https://image.xyzcdn.net/Fm2F9n8vMm_n-xafqMi98xs3T3K8.png" /></figure><figure><img src="https://image.xyzcdn.net/FgO-fNdZ9-g2JgGptgIspHPz5Me5.png" /></figure><blockquote><p>我们的播客节目在<a href="https://view.inews.qq.com/u/8QIf3n5c64Ucuzne7gI%3D?devid=FF4E49E6-9C89-4986-A413-04E856F31262&amp;qimei=766696f2cd8f313d744bc2c9000012918102&amp;uid=100161026780" rel="noopener noreferrer nofollow" target="_blank">腾讯新闻首发</a>，大家可以前往关注哦，这样可以第一时间获取节目信息和更多新闻资讯：）</p></blockquote><figure><img src="https://image.xyzcdn.net/FvV-R5FBydYHGZAMyXAV1K1A9iJT.png" /></figure><p>02:00 2025年资本寒冬过去了吗？没…</p><p>04:00 DeepSeek的出现，让AI应用创业变得活跃了吗？没…</p><p>10:45 现阶段各种Bot的体验：神婆？舔狗？尤其点评微信、元宝和小红书</p><p>25:28 探讨一下，微信里怎么做Agent？通用Agent成立吗？</p><p>31:25 垂直Agent的创业机会和创业者</p><p>35:52 现在的组织倾向于小组织，这可能利好年轻创业者</p><p>37:42 为啥组织更小，融资额却更高了？</p><p>38:18 除了Agent，还在看的是具身智能（仿真器很重要）</p><p>43:57 可穿戴设备</p><p>54:54 大模型公司</p><p>58:31 这两年工作很boring，我的旅行之路</p><p>01:03:55 我的读书之路</p><p>01:12:34 聊影视行业（梦秋是中影独董）、《好东西》和女性主义</p><figure><img src="https://image.xyzcdn.net/FvVbUNblF7FHIjfdp3MmmbAdLZ8G.png" /></figure><p>梦秋此前的节目：</p><p><a href="https://www.xiaoyuzhoufm.com/episodes/62cc4eb0fa15142e17251617" rel="noopener noreferrer nofollow" target="_blank">《1. 和投资人梦秋聊聊加州、投资寒潮和林黛玉》</a></p><p><a href="https://www.xiaoyuzhoufm.com/episodes/6430e7c89361a4e7c3f586f0" rel="noopener noreferrer nofollow" target="_blank">《21. 投资人视角下的大模型和市场真实水温｜和梦秋聊ChatGPT》</a></p><p><a href="https://www.xiaoyuzhoufm.com/episodes/66261c27200abebe6e7635af" rel="noopener noreferrer nofollow" target="_blank">《65. 风险投资的钥匙失灵了吗？和梦秋聊：蛰伏、更少人的游戏和啮齿动物》</a></p><p>【更多信息】</p><p>联络我们：微博<a href="https://weibo.com/u/6486678714" rel="noopener noreferrer nofollow" target="_blank">@张小珺-Benita</a></p><p>更多信息欢迎关注公众号：张小珺</p><figure><img src="https://image.xyzcdn.net/Fn7o36NtUYpCM_rQiFj1LW-TIwk8.JPG" /></figure>

今天继续《商业访谈录》的机器人专场，嘉宾是北京大学助理教授、银河通用创始人兼CTO王鹤。 王鹤毕业于清华和斯坦福大学。**他给我们从“具身智能”的学术缘起开始聊起，这是一个学术流派从一个学科中萌芽到边缘再到主流渗透的全过程。** 而随着ChatGPT诞生，“具身智能”这个小众概念，在过去2年成了新的资本宠儿——**但一时间，也带来了新的乱象。** 我们探讨了一些具身智能产业界关键问题： 1/具身智能起源于计算机视觉的学术流派，视觉、语言、智能的关系是什么？为什么VLM（视觉语言模型）的表现显著弱于LLM（大语言模型）？ 2/具身智能的最大困境之一是数据采集，合成数据是正解吗？具体应该怎么做？ 3/如果大模型提倡的是“智能即产品”，那么具身智能呢？王鹤的回答是“生产力即产品”。 去年底，英伟达创始人黄仁勋来华访问。答谢宴上，王鹤不仅和黄仁勋同桌，而且就在做黄仁勋旁边（挨着坐）。在节目最后，我们也聊了聊这个有趣的插曲——他提到，**那晚黄仁勋吃了不少水煮肉片。** 2025，我们和AI共同进步！ 我们的播客节目在腾讯新闻首发，大家可以前往关注哦，这样可以第一时间获取节目信息和更多新闻资讯：） 03:00 开始的自问自答 **05:58 语言不是智能的本质，而是“一次跃变”** “具身智能”和“机器人”是不同学术流派 **“具身智能”起源于“计算机视觉”的研究流派** 视觉有智能吗？纯视觉智能的可解释性差，是端到端的 语言不是智能的本质，不能说没有语言就没有智能 **智能的本质是什么？“一种视情况对环境做出反应的能力”** 语言是人类能产生这么高智能的“一次跃变” 视觉的本质是一种非常强的sensor（传感器） **25:08 具身智能的学术边缘史** 具身智能最早兴起的task（任务）是，导航 **加入视觉模态，强调Perception–Action Loop（感知-动作循环），成为具身智能研究流派能立起来的核心叙事** 标志性事件：“具身智能是计算机视觉未来的三颗北极星之一”（李飞飞） 我和Skild创始人Deepak Pathak在Facebook人工智能实验室FAIR打过交道 **41:15 我的学术之路** 2016年，博士第一个项目：从人类视频里学多步的人与物体交互过程的生成（动画领域） 在Stanford博士第一年，在不喜欢的方向非常挣扎，后来换组、换方向 **Stanford是高度自由的市场：你可以随时踢你老板，你老板可以随时踢你** 第一篇论文憋了很久，很绝望 完全从视频中学习，学习世界模型，还没成为当下能推进具身智能的技术 我的第二个项目：位姿估计和合成数据相关 2020年李开复曾在湾区丽思卡尔顿组织brunch，观点分歧 回国坚定以家庭机器人为目标推进research，根本没有allies（盟军） **01:25:08 具身智能的软件和硬件是螺旋上升的问题** ChatGPT火了以后，很多人开始找我创业，我说创不了 所有工业机械臂在去年的全球总产值才1000亿RMB，和理想一家车企产值相当 如果采取不成熟的激进的硬件方案，对智能会是一种拖累 在这个硬件基础上，我们的方案是，做相对专用的智能和越来越通用的智能 **VLM为什么显著弱于LLM？**互联网视觉数据/所有人眼观测的覆盖〈〈〈互联网文字数据/人类所有说的话的覆盖（VLM数据不够，VLA的Action数据是最近两年才开始收集的） **01:44:34 我们要避免陷入以下泥潭** 这一代具身智能公司相比此前机器人公司，差异在哪？ 在我看来，具身智能公司如果陷入以下两个泥潭，天花板会很有限： 1、“长期漂浮”的公司；2、“算不过来账”的公司，边际成本不降 我们要做一个应用场景内的泛化（现在选择的是货架场景） 在我看来，机器人领域的头部效应很重 **01:55:17 具身智能是，“生产力即产品”** 雇人摇操采真实数据的成本到底有多高？一笔经济账 真实数据在我们训练数据的比重是1%，合成数据管线挑起大梁 行业内的tricky现象：把没有功能的机器人卖给别人（这是一种商业模式） 关于合成数据和Sim-to-Real（仿真到现实迁移）的常见误区 有出货量后的数据回流和数据飞轮 **如果大模型是“智能即产品”，那么具身智能就是“生产力即产品”** **02:13:51 资本轰炸后的人为乱象** 谁在创造生产力，谁在讲故事，这是最乱的——这个源自美国 对Figure的估值400亿美元的两种逻辑 有的人胆子很大，不告诉别人我是摇操，但实际摇操 **呼吁：真实展示！不要摇操！** 5年内我们一定要有万台以上的应用，如果做不到这个，我们这个领域就被证伪了！ 不要去搞一些砸我们行业招牌的事情！这些模式是很可怕的，是在砸这个行业的饭碗 通用机器人的到来不要想得那么快 **02:25:25 一个插曲** **去年黄仁勋访华为什么和黄仁勋同桌且在旁边？聊了什么？** 黄仁勋能吃辣，吃了很多水煮肉片 02:28:26 最后的快问快答 【机器人专场】 逐篇讲解机器人基座模型和VLA经典论文——“人就是最智能的VLA” 【更多信息】 联络我们：微博@张小珺-Benita 更多信息欢迎关注公众号：张小珺

Original title: 106. 和王鹤聊，具身智能的学术边缘史和资本轰炸后的人为乱象

Original description: <figure><img src="https://image.xyzcdn.net/Flo18nNUSP7OUNlTf8UgCdHxio6O.jpg" /></figure><p>今天继续《商业访谈录》的机器人专场，嘉宾是北京大学助理教授、银河通用创始人兼CTO王鹤。</p><p>王鹤毕业于清华和斯坦福大学。<strong>他给我们从“具身智能”的学术缘起开始聊起，这是一个学术流派从一个学科中萌芽到边缘再到主流渗透的全过程。</strong></p><p>而随着ChatGPT诞生，“具身智能”这个小众概念，在过去2年成了新的资本宠儿——<strong>但一时间，也带来了新的乱象。</strong></p><p>我们探讨了一些具身智能产业界关键问题：</p><p>1/具身智能起源于计算机视觉的学术流派，视觉、语言、智能的关系是什么？为什么VLM（视觉语言模型）的表现显著弱于LLM（大语言模型）？</p><p>2/具身智能的最大困境之一是数据采集，合成数据是正解吗？具体应该怎么做？</p><p>3/如果大模型提倡的是“智能即产品”，那么具身智能呢？王鹤的回答是“生产力即产品”。</p><p>去年底，英伟达创始人黄仁勋来华访问。答谢宴上，王鹤不仅和黄仁勋同桌，而且就在做黄仁勋旁边（挨着坐）。在节目最后，我们也聊了聊这个有趣的插曲——他提到，<strong>那晚黄仁勋吃了不少水煮肉片。</strong></p><p>2025，我们和AI共同进步！</p><figure><img src="https://image.xyzcdn.net/Fm2F9n8vMm_n-xafqMi98xs3T3K8.png" /></figure><figure><img src="https://image.xyzcdn.net/Fuo-Y2NiU-ETwS5ypbWWUtvAVBaQ.png" /></figure><blockquote>我们的播客节目在<a href="https://view.inews.qq.com/u/8QIf3n5c64Ucuzne7gI%3D?devid=FF4E49E6-9C89-4986-A413-04E856F31262&amp;qimei=766696f2cd8f313d744bc2c9000012918102&amp;uid=100161026780">腾讯新闻首发</a>，大家可以前往关注哦，这样可以第一时间获取节目信息和更多新闻资讯：）</blockquote><figure><img src="https://image.xyzcdn.net/FvV-R5FBydYHGZAMyXAV1K1A9iJT.png" /></figure><p>03:00 开始的自问自答</p><blockquote><strong>05:58 语言不是智能的本质，而是“一次跃变”</strong></blockquote><p>“具身智能”和“机器人”是不同学术流派</p><p><strong>“具身智能”起源于“计算机视觉”的研究流派</strong></p><p>视觉有智能吗？纯视觉智能的可解释性差，是端到端的</p><p>语言不是智能的本质，不能说没有语言就没有智能</p><p><strong>智能的本质是什么？“一种视情况对环境做出反应的能力”</strong></p><p>语言是人类能产生这么高智能的“一次跃变”</p><p>视觉的本质是一种非常强的sensor（传感器）</p><blockquote><strong>25:08 具身智能的学术边缘史</strong></blockquote><p>具身智能最早兴起的task（任务）是，导航</p><p><strong>加入视觉模态，强调Perception–Action Loop（感知-动作循环），成为具身智能研究流派能立起来的核心叙事</strong></p><p>标志性事件：“具身智能是计算机视觉未来的三颗北极星之一”（李飞飞）</p><p>我和Skild创始人Deepak Pathak在Facebook人工智能实验室FAIR打过交道</p><blockquote><strong>41:15 我的学术之路</strong></blockquote><p>2016年，博士第一个项目：从人类视频里学多步的人与物体交互过程的生成（动画领域）</p><p>在Stanford博士第一年，在不喜欢的方向非常挣扎，后来换组、换方向</p><p><strong>Stanford是高度自由的市场：你可以随时踢你老板，你老板可以随时踢你</strong></p><p>第一篇论文憋了很久，很绝望</p><p>完全从视频中学习，学习世界模型，还没成为当下能推进具身智能的技术</p><p>我的第二个项目：位姿估计和合成数据相关</p><p>2020年李开复曾在湾区丽思卡尔顿组织brunch，观点分歧</p><p>回国坚定以家庭机器人为目标推进research，根本没有allies（盟军）</p><blockquote><strong>01:25:08 具身智能的软件和硬件是螺旋上升的问题</strong></blockquote><p>ChatGPT火了以后，很多人开始找我创业，我说创不了</p><p>所有工业机械臂在去年的全球总产值才1000亿RMB，和理想一家车企产值相当</p><p>如果采取不成熟的激进的硬件方案，对智能会是一种拖累</p><p>在这个硬件基础上，我们的方案是，做相对专用的智能和越来越通用的智能</p><p><strong>VLM为什么显著弱于LLM？</strong>互联网视觉数据/所有人眼观测的覆盖〈〈〈互联网文字数据/人类所有说的话的覆盖（VLM数据不够，VLA的Action数据是最近两年才开始收集的）</p><blockquote><strong>01:44:34 我们要避免陷入以下泥潭</strong></blockquote><p>这一代具身智能公司相比此前机器人公司，差异在哪？</p><p>在我看来，具身智能公司如果陷入以下两个泥潭，天花板会很有限：</p><p>1、“长期漂浮”的公司；2、“算不过来账”的公司，边际成本不降</p><p>我们要做一个应用场景内的泛化（现在选择的是货架场景）</p><p>在我看来，机器人领域的头部效应很重</p><blockquote><strong>01:55:17 具身智能是，“生产力即产品”</strong></blockquote><p>雇人摇操采真实数据的成本到底有多高？一笔经济账</p><p>真实数据在我们训练数据的比重是1%，合成数据管线挑起大梁</p><p>行业内的tricky现象：把没有功能的机器人卖给别人（这是一种商业模式）</p><p>关于合成数据和Sim-to-Real（仿真到现实迁移）的常见误区</p><p>有出货量后的数据回流和数据飞轮</p><p><strong>如果大模型是“智能即产品”，那么具身智能就是“生产力即产品”</strong></p><blockquote><strong>02:13:51 资本轰炸后的人为乱象</strong></blockquote><p>谁在创造生产力，谁在讲故事，这是最乱的——这个源自美国</p><p>对Figure的估值400亿美元的两种逻辑</p><p>有的人胆子很大，不告诉别人我是摇操，但实际摇操</p><p><strong>呼吁：真实展示！不要摇操！</strong></p><p>5年内我们一定要有万台以上的应用，如果做不到这个，我们这个领域就被证伪了！</p><p>不要去搞一些砸我们行业招牌的事情！这些模式是很可怕的，是在砸这个行业的饭碗</p><p>通用机器人的到来不要想得那么快</p><blockquote><strong>02:25:25 一个插曲</strong></blockquote><p><strong>去年黄仁勋访华为什么和黄仁勋同桌且在旁边？聊了什么？</strong></p><p>黄仁勋能吃辣，吃了很多水煮肉片</p><p>02:28:26 最后的快问快答</p><figure><img src="https://image.xyzcdn.net/FvVbUNblF7FHIjfdp3MmmbAdLZ8G.png" /></figure><p>【机器人专场】</p><p><a href="https://www.xiaoyuzhoufm.com/episodes/67f28c6e0decaeb0943fb14a">逐篇讲解机器人基座模型和VLA经典论文——“人就是最智能的VLA”</a></p><p>【更多信息】</p><p>联络我们：微博<a href="https://weibo.com/u/6486678714">@张小珺-Benita</a></p><p>更多信息欢迎关注公众号：张小珺</p><figure><img src="https://image.xyzcdn.net/Fn7o36NtUYpCM_rQiFj1LW-TIwk8.JPG" /></figure>

《商业访谈录》 has interviewed many CEOs and senior executives of Chinese new energy vehicle companies. Today's guest is Wang Xin, head of autonomous driving and connected car R&D at Mercedes-Benz China, a century-old German car company. We talked about the 20-year transformation of China's auto industry, and the transformation and secret stories of a German car company. Our podcast is first released on Tencent News, so you can follow it to get the latest program information and more news :) **Major Industrial Transformation** 01:25 20 years ago, even Bird mobile phones made cars 07:54 I used to work at automotive Tier 1 Delphi for 18 years, and joined Mercedes-Benz 3 years ago, which is behind a major industrial transformation 09:30 Several technology cycles in the global auto industry in the past 20 years (before 2004, 2004-2014, 2014-2020, 2020-present) 11:31 Now it has transformed into a data-driven era, and the era of Tier 1 black box delivery is over **Right to Speak** 27:40 Is the right to speak of the Chinese team and the German headquarters fought for? 28:27 Mercedes-Benz China R&D team organizational structure, communication mechanism and battle 34:08 The battle culture of German companies is different from that of American companies 41:23 What processes are required when features designed and produced for China are exported globally? **New Technology** 43:21 Intelligentization is an irreversible trend, but it should not be radical 46:50 Vehicle-to-vehicle communication needs to be redefined after L3 is achieved 51:54 The relationship between technology and luxury: if intelligence is equalized, has the standard of luxury changed? 01:01:49 The process of switching from rule-based algorithms to end-to-end last year was quite painful 01:04:40 LiDAR is a good redundancy 01:05:35 CLA and Baobao cooperate with large language models **139-year-old Car Company** 01:09:36 People-oriented 01:11:08 Safety steps 01:13:08 The world's first car driver was the wife of the founder of Mercedes-Benz 01:15:00 What is it like to work in a century-old company - what is glory? What is the burden? 01:17:48 Once-in-a-century transformation and major changes 01:33:22 Does Mercedes-Benz CEO Ola Källenius get angry? Related episodes: Dialogue with Mercedes-Benz Global CEO Ola Källenius: CEO in Transition and 139-year-old Mercedes-Benz in Transition 【More Information】 Contact us: Weibo @张小珺-Benita For more information, please follow the official account: 张小珺

Original title: 105. 和奔驰王忻聊，产业大转折下的德国汽车、话语权和技术battle

Original description: <figure><img src="https://image.xyzcdn.net/Flo18nNUSP7OUNlTf8UgCdHxio6O.jpg" /></figure><p>《商业访谈录》访谈过很多中国新能源车企的CEO和高层，今天的嘉宾来自一家德国百年车企，他是奔驰中国自动驾驶与车联网研发负责人王忻。</p><p>我们聊了聊中国汽车产业20年变革的历程，以及一家德国车企的转型与秘密故事。</p><figure><img src="https://image.xyzcdn.net/Fm2F9n8vMm_n-xafqMi98xs3T3K8.png" /></figure><figure><img src="https://image.xyzcdn.net/FjMoypYgmSWwR6Hv1JpFkZRvS8Yu.png" /></figure><blockquote><p>我们的播客节目在<a href="https://view.inews.qq.com/u/8QIf3n5c64Ucuzne7gI%3D?devid=FF4E49E6-9C89-4986-A413-04E856F31262&amp;qimei=766696f2cd8f313d744bc2c9000012918102&amp;uid=100161026780" rel="noopener noreferrer nofollow" target="_blank">腾讯新闻首发</a>，大家可以前往关注哦，这样可以第一时间获取节目信息和更多新闻资讯：）</p></blockquote><figure><img src="https://image.xyzcdn.net/FvV-R5FBydYHGZAMyXAV1K1A9iJT.png" /></figure><blockquote><p><strong>产业大转型</strong></p></blockquote><p>01:25 20年前，就连波导手机也做过汽车</p><p>07:54 我曾在汽车Tier 1德尔福18年，3年前加入奔驰，背后是产业大转折</p><p>09:30 过去20年全球汽车产业的几个技术周期（2004年以前，2004-2014年，2014-2020年，2020年至今）</p><p>11:31 现在转变成数据驱动的时代，Tier 1黑盒交付的时代不再</p><blockquote><p><strong>话语权</strong></p></blockquote><p>27:40 中国团队和德国总部的话语权是争夺过来的吗？</p><p>28:27 奔驰中国研发团队组织架构、沟通机制和battle</p><p>34:08 德国企业的battle文化和美国企业是不同的</p><p>41:23 为中国设计生产的功能要反向输出全球的时候，需要哪些流程？</p><blockquote><p><strong>新技术</strong></p></blockquote><p>43:21 智能化是不可逆的趋势，但不能激进</p><p>46:50 车车通讯在L3实现以后需要重新定义</p><p>51:54 科技和豪华的关系：如果智能平权，豪华的标准变了吗</p><p>01:01:49 去年从规则算法切换到端到端的过程挺煎熬的</p><p>01:04:40 激光雷达是一个很好的冗余</p><p>01:05:35 CLA和豆包合作大语言模型</p><blockquote><p><strong>139岁车企</strong></p></blockquote><p>01:09:36 以人为本</p><p>01:11:08 安全的步骤</p><p>01:13:08 世界上第一位汽车驾驶员是奔驰创始人的太太</p><p>01:15:00 在百年企业工作是什么体验——荣耀是什么？负担是什么？</p><p>01:17:48 百年一遇的大转型、大变革</p><p>01:33:22 奔驰CEO康林松会发脾气吗？</p><figure><img src="https://image.xyzcdn.net/FvVbUNblF7FHIjfdp3MmmbAdLZ8G.png" /></figure><p>相关单集：</p><p><a href="https://www.xiaoyuzhoufm.com/episodes/68300e93fcbc2e206b58eb2b" rel="noopener noreferrer nofollow" target="_blank">对话奔驰全球CEO康林松：转型期CEO和转型之中的139岁奔驰</a></p><p>【更多信息】</p><p>联络我们：微博<a href="https://weibo.com/u/6486678714" rel="noopener noreferrer nofollow" target="_blank">@张小珺-Benita</a></p><p>更多信息欢迎关注公众号：张小珺</p><figure><img src="https://image.xyzcdn.net/Fn7o36NtUYpCM_rQiFj1LW-TIwk8.JPG" /></figure>

As AI's software capabilities spill over into hardware, in addition to embodied intelligence, smart glasses may be another industry that will benefit. Today's guest is Misa Zhu, the founder of smart glasses company Rokid. In the first half of 2025, Misa appeared in a speech wearing smart glasses developed by his company, which once attracted attention. **This year also marks his 11th year of entrepreneurship in the hardware dark forest.** We started by talking about the acquisition of his first company by Alibaba for 10 million US dollars—we talked about **Jack Ma and Wu Ma**, and also about his second venture, **the China-US comparison, stages, and trends of the smart glasses market.** Our podcast program premiered on Tencent News. You can go and follow it to get program information and more news as soon as possible :) 02:00 Quick Q&A starts 02:36 Alibaba acquired my first startup for 10 million US dollars, all converted into stocks 05:14 In the worst times, Jack Ma talked to me and introduced Joe Cai (Joseph Tsai), and also introduced Dr. Wang Jian 08:05 I have two weeks to pay salaries, and there are only 4,000 yuan in the account 15:55 When I was a senior executive at Alibaba, Wu Ma (Wu Yongming) proposed to do AI and established M lab 22:43 Rokid's financing, Jack Ma's advice 27:40 Wu Ma was my direct boss back then, comments on Wu Ma 31:41 Important decision in 2019: switch from AI to AR track within a week 48:00 Will organ-like hardware switch from mobile phones to smart glasses? 59:17 After the important decision, half of the employees were laid off and a building was emptied 01:05:45 First PMF after the transformation 01:09:55 The current smart glasses are in the intermediate stage from BlackBerry to iPhone 1 01:11:52 AI's expansion on hardware: embodied intelligence, wearable intelligence 01:13:05 In smart glasses, the first half of next year will be the time to compete with giants 01:19:29 Jack Ma summarized 4 opportunities for startups to compete with giants: 4 "no's" 01:23:38 Different definitions of smart glasses products in China and the United States 01:41:35 The first company value is playfulness, and the boss is always the trouble maker 01:48:32 Talking about entrepreneurs in Hangzhou 01:59:05 The dark forest of hardware entrepreneurship 02:27:00 Final quick Q&A 【More Information】 Contact us: Weibo @张小珺-Benita For more information, please follow the official account: 张小珺

Original title: 104. 和Rokid祝铭明聊，吴妈、阿里、硬件创业黑森林的第11年

Original description: <figure><img src="https://image.xyzcdn.net/Flo18nNUSP7OUNlTf8UgCdHxio6O.jpg" /></figure><p>随着AI的软件能力向硬件溢出，除了具身智能，智能眼镜或许是另一个会受益的产业。</p><p>今天的嘉宾是智能眼镜公司Rokid创始人祝铭明（Misa），2025上半年Misa佩戴其公司开发的智能眼镜出现在一次演讲中，一度引发关注，<strong>今年也是他在硬件黑森林里创业的第11个年头。</strong></p><p>我们从他的第一家公司1000万美金被阿里并购开始聊起——聊了聊<strong>马云和吴妈</strong>，也聊了聊他的第二段创业、<strong>智能眼镜市场的中美对比、阶段与趋势</strong>。</p><figure><img src="https://image.xyzcdn.net/Fm2F9n8vMm_n-xafqMi98xs3T3K8.png" /></figure><figure><img src="https://image.xyzcdn.net/FmOqd2jZnPpPWSi6lw-jBpb8-qYE.png" /></figure><blockquote>我们的播客节目在<a href="https://view.inews.qq.com/u/8QIf3n5c64Ucuzne7gI%3D?devid=FF4E49E6-9C89-4986-A413-04E856F31262&amp;qimei=766696f2cd8f313d744bc2c9000012918102&amp;uid=100161026780">腾讯新闻首发</a>，大家可以前往关注哦，这样可以第一时间获取节目信息和更多新闻资讯：）</blockquote><figure><img src="https://image.xyzcdn.net/FvV-R5FBydYHGZAMyXAV1K1A9iJT.png" /></figure><p>02:00 开始的快问快答</p><p>02:36 阿里1000万美金收购我的第一个创业公司，全部换成了股票</p><p>05:14 最糟糕的时候，马云找我聊，引荐了Joe Cai（蔡崇信），又引荐了王坚博士</p><p>08:05 我还有两个星期发薪水，账上只有4000块</p><p>15:55 在阿里当高管，吴妈（吴泳铭）提出想做AI，成立M lab</p><p>22:43 Rokid的融资、马云的建议</p><p>27:40 吴妈当年是我的顶头上司，对吴妈的comments</p><p>31:41 2019年重要决策：一星期内从AI切换AR赛道</p><p>48:00 像器官一样的硬件会从手机切换到智能眼镜？</p><p>59:17 重要决策之后裁员了一大半，清空了一幢楼</p><p>01:05:45 转型后第一次PMF</p><p>01:09:55 现在的智能眼镜在黑莓到iPhone 1的中间阶段</p><p>01:11:52 AI在硬件上的展开：具身智能、随身智能</p><p>01:13:05 在智能眼镜，明年上半年会是与巨头竞争的时间点</p><p>01:19:29 马云总结创业公司和巨头竞争的4个机会：4个不</p><p>01:23:38 中美定义智能眼镜产品的不同</p><p>01:41:35 公司价值观第一条是玩心，老板总是那个trouble maker</p><p>01:48:32 聊聊杭州创业者们</p><p>01:59:05 硬件创业的黑森林</p><p>02:27:00 最后的快问快答</p><p>【更多信息】</p><p>联络我们：微博<a href="https://weibo.com/u/6486678714">@张小珺-Benita</a></p><p>更多信息欢迎关注公众号：张小珺</p><figure><img src="https://image.xyzcdn.net/Fn7o36NtUYpCM_rQiFj1LW-TIwk8.JPG" /></figure>

Today's guest is another AI application entrepreneur, Chen Mian, the founder of Lovart. His product has become another Agent with a certain global reputation in 2025 after Manus. The difference is that Manus is a general Agent, and Lovart is a vertical Agent, for designers. Rather than saying he is a CEO who makes products, **his mental state is closer to that of a "fighting CEO".** This interview took place after Lovart became popular. You can feel the overflowing joy of Chen Mian at this moment, after experiencing a series of resentments in the past 2 years, such as subsidy wars, product delisting, only 4,000 yuan left in the account, and failing to raise funds. This is a snapshot of the mental state of an Agent entrepreneur in 2025. The curtain of the wave has just been raised. 2025, look forward to our common progress with AI! Our podcast program premiered on Tencent News. You can go and follow it to get program information and more news as soon as possible :) 03:00 Quick questions and answers **Wandering** 05:00 A 90s' 10 years of mobile Internet experience with constant job hopping (Tencent, 360, Baidu, Didi, Mobike, Meituan, Missfresh, ByteDance Education, and Jianying) 07:02 Experienced two battles, the peak was when the battle was in full swing, and then it was a mess 13:58 Starting from 0 to 1 to make Guagualong, just promoted to Byte 4-1, and encountered the "Double Reduction" policy 15:18 Would it be better to change the choice? **AI is here, I feel rescued** 25:25 AI is at least the invention of the computer, a change comparable to the information revolution (intelligence vs informatization) 28:58 The moment of being redeemed: "Hope is the antidote to all pain, and the meaning of all pain" 29:51 Avoid the main track of large models and the main axis of language, and choose multi-modality and creation **2023: One second I won first place in China, the next second it was delisted, layoffs, and no money** 36:00 The first investor I met was Zhang Yutong 37:43 June-September 2023, I fought with all my heart! ——Burned 2 million US dollars in 3 months 39:03 One second I just won first place in China, the next second it was delisted, layoffs, and no money 40:45 How to view advertising? How to view Kimi advertising? 42:35 What is the feeling of being delisted? Crashing 44:09 The company's account only has 4,000 yuan left 45:17 What is the current customer acquisition cost? How to effectively acquire users? 49:38 The opportunity is fleeting, and the results must be expanded when the rhythm is good **2024: Crazy investment** 50:05 2024 crazy investment, one round of financing per month, closed 3 rounds 52:21 We are very clear about the limitations of the first generation product liblib, and began to consider the second generation product 55:58 How was the second-generation product Lovart pre-researched? **2025: Lovart is popular** 59:48 If this designer is called Lovart, he/she also loves art, which is quite cool 01:01:47 What does competing for the "world's first XX Agent" bring? 01:03:00 Why is getting an invitation code a standard feature? 01:03:56 After Lovart became popular 01:07:30 Know-how of AI application entrepreneurship **This is the most!! awesome!! thing in my entrepreneurship!!** 01:19:57 I am a Gemini, sometimes crazy, sometimes very soft 01:24:04 Coexist with anxiety, just do it! 01:25:26 This is the most!! awesome!! thing in my entrepreneurship!! 01:28:00 But at this moment it's so cool!!! I was happy for a while - it's my simple happiness hahahahaha 01:28:32 No amount of money or job title can buy it 01:32:35 Innovation in unfamiliar fields is like repeatedly sliding a match on wet wood, igniting and extinguishing; until one day, you seize a gap, ignite the wood, and the fire spreads throughout the cave 01:33:58 At the end of 2023, I went to the Hillhouse office, and in the sunlight, I was in a daze **Make a big scene, leave quietly** 01:35:00 Childhood: wandering, martial arts novels and computer games 01:26:01 I don't know where my hometown is, I can only keep moving forward 01:38:46 Advice for other AI application entrepreneurs 01:42:29 Final quick questions and answers [Agent Entrepreneurship Trilogy in the First Half of 2025] 3-hour interview with Manus founder Xiao Hong: The world is not a linear extrapolation, but an important variable in the game 3-hour interview with YouWare founder Ming Chaoping: Today's Agent is like a gorilla just picking up a burning stick Lovart founder Chen Mian reviews the past two years of application entrepreneurship: This moment is so cool!! hahahahaha [More Information] Contact us: Weibo @张小珺-Benita For more information, please follow the official account: 张小珺

Original title: 103. Lovart创始人陈冕复盘应用创业这两年：这一刻就是好爽啊！！哈哈哈哈哈

Original description: <figure><img src="https://image.xyzcdn.net/Flo18nNUSP7OUNlTf8UgCdHxio6O.jpg" /></figure><p>今天的嘉宾又是一位AI应用创业者，Lovart创始人陈冕。</p><p>他的产品成为2025年既Manus之后，另一个在全球斩获一定知名度的Agent。不同的是，Manus是通用Agent，Lovart是垂直Agent，面向设计师使用。</p><p>与其说他是做产品的CEO，<strong>他的精神状态更贴近一名“战斗型CEO”。</strong></p><p>这次访谈发生Lovart火了之后，你能感受到陈冕在过去2年遭遇了补贴战争、产品下架、账上只剩4000块现金的绝境、怎么都融不到资等一系列愤懑之后——此时此刻，充斥着的要溢出的快乐。</p><p>这是2025年对一位Agent创业者精神状态的截取。浪潮的大幕才刚刚拉开。</p><p>2025，期待我们和AI共同进步！</p><figure><img src="https://image.xyzcdn.net/Fm2F9n8vMm_n-xafqMi98xs3T3K8.png" /></figure><figure><img src="https://image.xyzcdn.net/Fhg0fHfEjPJE44bF_5suSeFO9TmJ.png" /></figure><blockquote>我们的播客节目在<a href="https://view.inews.qq.com/u/8QIf3n5c64Ucuzne7gI%3D?devid=FF4E49E6-9C89-4986-A413-04E856F31262&amp;qimei=766696f2cd8f313d744bc2c9000012918102&amp;uid=100161026780" rel="noopener noreferrer nofollow" target="_blank">腾讯新闻首发</a>，大家可以前往关注哦，这样可以第一时间获取节目信息和更多新闻资讯：）</blockquote><figure><img src="https://image.xyzcdn.net/FvV-R5FBydYHGZAMyXAV1K1A9iJT.png" /></figure><p>03:00 开始的快问快答</p><blockquote><strong>漂泊</strong></blockquote><p>05:00 一个90后的不断跳槽的10年移动互联网经历</p><p>（腾讯、360、百度、滴滴、摩拜、美团、每日优鲜、字节教育和剪映）</p><p>07:02 经历了两次战斗，战斗正酣的时候是顶点，后面一地鸡毛</p><p>13:58 从0到1做瓜瓜龙，刚升字节4-1，就撞上双减了</p><p>15:18 换一种选择，会更好吗？</p><blockquote><strong>AI来了，觉得自己被解救了</strong></blockquote><p>25:25 AI至少是电脑的发明，比肩信息革命的变革（智能化vs信息化）</p><p>28:58 被救赎的一刻：“希望是一切痛苦的解药，是一切痛苦的意义”</p><p>29:51 避开大模型主航道和语言主轴，选择多模态、创作</p><blockquote><strong>2023年：前一秒赢了中国第一，下一秒被下架了、裁员了、没钱了</strong></blockquote><p>36:00 见的第一个投资人是张予彤</p><p>37:43 2023年6月-9月，我全情的战斗！——3个月烧了200万美金</p><p>39:03 前一秒刚赢了中国第一，下一秒被下架了、裁员了、没钱了</p><p>40:45 怎么看投流？怎么看Kimi投流？</p><p>42:35 被下架什么心情？奔溃啊</p><p>44:09 公司账上只剩4000块</p><p>45:17 现在获客成本是多少？怎么有效获取用户？</p><p>49:38 时机稍纵即逝，好的节奏时一定要扩大战果</p><blockquote><strong>2024年：哐哐哐狂投</strong></blockquote><p>50:05 2024年哐哐狂投，一个月一轮融资，close了3轮</p><p>52:21 我们非常清楚第一代产品liblib的局限性，开始考虑第二代产品</p><p>55:58 第二代产品Lovart是怎么预研的？</p><blockquote><strong>2025年：Lovart火了</strong></blockquote><p>59:48 如果这个设计师叫Lovart，他/她又Love art，还蛮酷的</p><p>01:01:47 争抢“全球第一个XX Agent”究竟带来什么？</p><p>01:03:00 为啥搞邀请码成了标配？</p><p>01:03:56 Lovart火了之后</p><p>01:07:30 AI应用创业的know-how</p><blockquote><strong>这是我创业最！！爽的！！东西！！</strong></blockquote><p>01:19:57 我是双子座，时而发狂，时而很软</p><p>01:24:04 与焦虑共生，就是干！</p><p>01:25:26 这是我创业最！！爽的！！东西！！</p><p>01:28:00 但在这一刻就是好爽啊！！！我爽了好一会儿——就是我朴实的快乐哈哈哈哈哈</p><p>01:28:32 给我多少钱、给我多少职级，都买不到</p><p>01:32:35 在陌生领域的创新，就像用火柴在潮湿的木头上反复地滑动，点燃又熄灭；直到有一天，你抓住了某一个缝隙，把木柴点燃，火势弥漫整个山洞</p><p>01:33:58 2023年底去高瓴办公室，阳光中，我恍惚了</p><blockquote><strong>大闹一场，悄然离去</strong></blockquote><p>01:35:00 童年：漂泊、武侠小说和电脑游戏</p><p>01:26:01 我不知道故乡是哪，只能一直往前走</p><p>01:38:46 给其他AI应用创业者的建议</p><p>01:42:29 最后的快问快答</p><figure><img src="https://image.xyzcdn.net/FvVbUNblF7FHIjfdp3MmmbAdLZ8G.png" /></figure><p>【2025上半年Agent创业三部曲】</p><p><a href="https://www.xiaoyuzhoufm.com/episodes/67c3d80fb0167b8db9e3ec0f">对Manus创始人肖弘的3小时访谈：世界不是线性外推，做博弈中的重要变量</a></p><p><a href="https://www.xiaoyuzhoufm.com/episodes/68372c9631215eb5063bcdb1">对YouWare创始人明超平3小时访谈：今天Agent像大猩猩刚拿起一根烧火棍</a></p><p><a href="https://www.xiaoyuzhoufm.com/episodes/68455e0a6dbe9284e75c6fbf">Lovart创始人陈冕复盘应用创业这两年：这一刻就是好爽啊！！哈哈哈哈哈</a></p><p>【更多信息】</p><p>联络我们：微博<a href="https://weibo.com/u/6486678714" rel="noopener noreferrer nofollow" target="_blank">@张小珺-Benita</a></p><p>更多信息欢迎关注公众号：张小珺</p><figure><img src="https://image.xyzcdn.net/Fn7o36NtUYpCM_rQiFj1LW-TIwk8.JPG" /></figure>

Today's episode of "Business Interviews" welcomes its first co-host, the familiar Li Guangmi. Guangmi invited Zhang Xiangyu, Chief Scientist of the large model company StepUp Starry, to talk about the past, present, and future technological frontiers of multi-modality. In this episode, Zhang Xiangyu elaborates on: **his 10-year history of participating in multi-modality, his new thinking on multi-modality, and his predictions for the next "GPT-4 moment."** He mentioned a detail: during the training process, he discovered an inexplicable phenomenon - the model's general conversational ability, emotional intelligence, and knowledge all became stronger as the model grew larger, but the model's reasoning ability (especially in mathematics) first increased and then flattened out, and then decreased as it expanded further - this point has not yet triggered widespread discussion in the industry. He also gave his own explanation for this **strange phenomenon**. Below is the conversation between Guangmi and Xiangyu. 2025, let's progress together with AI! Our podcast is first released on Tencent News, you can go and follow it to get program information and more news as soon as possible :) **10-Year History of Multi-Modal Research: Confusion and Turning Points** 02:00 Zhang Xiangyu's academic experience and personal research focus 12:25 Learning history from CV (Computer Vision) to NLP (Natural Language Processing) 17:14 In 2022, I became pessimistic about relying solely on vision to learn the "GPT moment in the CV field" 18:22 What's wrong with the pure vision domain? **A generative model like GPT can simultaneously possess generation, understanding, and human alignment, while these three are separated in static images.** 24:23 I stopped researching static image representation and conceived a new research topic: using the alignment relationship between vision and language in the short term 29:10 After trying, I still couldn't achieve the integration of image understanding, generation, and alignment. I got an increasingly strong generation model and an increasingly strong understanding model, which did not have a superimposed effect - why is it so difficult to integrate? 38:45 **I was very confused for more than half a year, but a turning point appeared at this moment.** **Strange things, clues, and solutions discovered in training large models** 41:11 A perplexing strange thing was discovered during the training process: **the model's general conversational ability, emotional intelligence, and knowledge are indeed stronger the larger the model, but the model's reasoning ability (especially in mathematics) performance first increases and then flattens, and then decreases as it expands further.** 43:10 Some clues: larger models tend to skip steps and are not honest when doing math problems 44:33 After analysis, **this is the essential defect of next token prediction** 45:42 A larger compression rate does not necessarily correspond to higher calculation accuracy, let's do a thought experiment 47:27 "Feature collapse phenomenon" of generative models 50:48 The solution is to introduce RL (Reinforcement Learning) 53:28 **The core of o1 is the pattern of the chain of thought** - "do thinking models, pattern is all you need" 01:01:52 When the model reaches a certain step, there are two branches in front of it - go left? Or go right? - Can it be solved within one token? (critical decision) - No, so introduce a reflection pattern 01:10:16 The essence of the o1 paradigm is a Meta-CoT, which is CoT's CoT **New Thinking and New Progress on Multi-Modal Research** 01:10:57 After studying o1, I returned to study why visual generation has such poor controllability, and I had some clues 01:15:13 Simply putting generation and understanding together is very difficult, missing an important link CoT 01:15:54 **Started a new project in the middle of last year: visual understanding (Long CoT in visual space)** 01:19:06 Let me reveal the results to you after trying for half a year! 01:21:30 The o series not only generalizes the domain, but also generalizes the pattern, which is more attractive 01:22:16 Game-like problems are difficult to generalize, and there are many invalid thoughts and low-level errors 01:24:07 The reflection pattern inspired by o1 is distributed in the pre-training corpus 01:31:31 There are two theories about pre-training plus multi-modal data: does it affect text IQ? Or does it enhance the scaling law? 01:36:43 Walk on two legs in the future: expand the pre-training corpus and expand the action space 01:45:42 How long until the "GPT-4 moment" of multi-modality? **Predicting the next "GPT-4 Moment"** 01:46:56 long context and multi-model collaboration 02:07:09 Architecture is not important, architecture serves algorithms and systems (why I say Linear Transformer is not essential) 02:08:30 **The next "GPT-4 moment"? Model's online learning/autonomous learning** 02:21:22 Clarify some views on Agent 02:25:00 Although humans do not have generative organs, they have world models 02:26:34 Our intelligence level is still struggling for vision, and the robotics field is running ahead 【More Information】 Contact us: Weibo @张小珺-Benita For more information, please follow the official account: 张小珺

Original title: 102. 和张祥雨聊，多模态研究的挣扎史和未来两年的2个“GPT-4时刻”

Original description: <figure><img src="https://image.xyzcdn.net/Flo18nNUSP7OUNlTf8UgCdHxio6O.jpg" /></figure><p>今天这集，《商业访谈录》第一次迎来一位co-host，是大家熟悉的李广密。</p><p>广密邀请了大模型公司阶跃星辰的首席科学家张祥雨，来聊聊，多模态的前世今生和未来技术的前沿走向。</p><p>张祥雨在这集节目详细阐述了：<strong>他参与的多模态的10年历史，对多模态的全新思考，以及所预见的下一个“GPT-4时刻”。</strong></p><p>他提到一个细节：在训练过程中他曾经发现一件百思不得其解的现象——模型的通用对话能力、情商和知识量都是随着模型变大变得更强，但模型的推理能力（尤其是数学）表现却是先上升后平缓，再扩大反而是下降——这点在业界还未引发广泛讨论。关于这个<strong>怪现象</strong>，他也给出了自己的解答。</p><p>下面是广密和祥雨的聊天。</p><p>2025，我们和AI共同进步！</p><figure><img src="https://image.xyzcdn.net/Fm2F9n8vMm_n-xafqMi98xs3T3K8.png" /></figure><figure><img src="https://image.xyzcdn.net/FiSVQGPuUlWbTkbF5UYXQXUufs8Q.png" /></figure><blockquote>我们的播客节目在<a href="https://view.inews.qq.com/u/8QIf3n5c64Ucuzne7gI%3D?devid=FF4E49E6-9C89-4986-A413-04E856F31262&amp;qimei=766696f2cd8f313d744bc2c9000012918102&amp;uid=100161026780" rel="noopener noreferrer nofollow" target="_blank">腾讯新闻首发</a>，大家可以前往关注哦，这样可以第一时间获取节目信息和更多新闻资讯：）</blockquote><figure><img src="https://image.xyzcdn.net/FvV-R5FBydYHGZAMyXAV1K1A9iJT.png" /></figure><blockquote><strong>多模态研究的10年史：迷茫和转机</strong></blockquote><p>02:00 张祥雨的学术经历和个人研究主线</p><p>12:25 CV（计算机视觉）向NLP（自然语言处理）的学习历史</p><p>17:14 2022年我开始对单纯靠视觉学出“CV领域的GPT时刻”比较悲观</p><p>18:22 纯视觉这个domain有什么问题？<strong>GPT这样的生成模型你可以同时拥有生成、理解和人类对齐，而静态图像这三者是割裂的</strong></p><p>24:23 我停止了对静态图像表征的研究，构思新的研究主题：短期内利用视觉和语言的对齐关系</p><p>29:10 经过尝试还是没做到图像的理解、生成和对齐一体化，我得到一个越来越强的生成模型，和一个越来越强的理解模型，没有起到叠加效果——为什么如此难以融合？</p><p>38:45 <strong>做了大半年十分迷茫，但在此刻出现了转机</strong></p><blockquote><strong>训练大模型发现的怪事、蛛丝马迹与办法</strong></blockquote><p>41:11 训练过程中发现了一件百思不得其解的怪事：<strong>模型的通用对话能力、情商、知识量确实模型越大越强，但模型的推理能力（尤其是数学）表现是先上升后平缓，再扩大反而是下降</strong></p><p>43:10 一些蛛丝马迹：更大的模型做数学题倾向于跳步，不老实</p><p>44:33 经过分析，<strong>这是next token prediction的本质缺陷</strong></p><p>45:42 更大的压缩率未必对应更高的计算精度，我们来做一个思想实验</p><p>47:27 生成模型的“特征坍缩现象”</p><p>50:48 解决方案就是引入RL（强化学习）</p><p>53:28 <strong>o1的核心是思维链的pattern</strong>——“做思考模型，pattern is all you need”</p><p>01:01:52 当模型走到某一步，摆在面前有两个分支——走左边？还是走右边？——一个token之内到底能不能解决？（critical decision）——不能，所以引入反思pattern</p><p>01:10:16 o1范式的本质是一种Meta-CoT ，是CoT的CoT</p><blockquote><strong>对多模态研究的新思考和新进展</strong></blockquote><p>01:10:57 研究完o1，返回研究为什么视觉生成可控性这么差，就有了眉目</p><p>01:15:13 简单把生成和理解做到一起，难度非常大，缺失了重要一环CoT</p><p>01:15:54 <strong>去年中开启新的project：视觉理解（视觉空间的Long CoT）</strong></p><p>01:19:06 尝试了半年，结果给大家透露一下吧！</p><p>01:21:30 o系列不仅泛化了domain，更吸引人的是泛化了pattern</p><p>01:22:16 博弈类问题是难以泛化的领域，有很多无效思考和低级错误</p><p>01:24:07 o1激发的反思pattern，在预训练语料中都有分布了</p><p>01:31:31 关于预训练加多模态数据有两种说法：影响了text智商？还是增强了scaling law？</p><p>01:36:43 往后两条腿走：扩充预训练语料和扩展动作空间</p><p>01:45:42 多模态的“GPT-4时刻”还有多久</p><blockquote><strong>预见下一个“GPT-4时刻”</strong></blockquote><p>01:46:56 long context和多模型协作</p><p>02:07:09 架构不重要，架构是服务算法和系统的（为什么我说Linear Transformer不本质）</p><p>02:08:30<strong> 下一个“GPT-4时刻”？模型的在线学习/自主学习</strong></p><p>02:21:22 澄清一些有关Agent的观点</p><p>02:25:00 人虽然没有生成器官，但人有世界模型</p><p>02:26:34 我们的智能水平还在为视觉挣扎，机器人领域在抢跑</p><p>【更多信息】</p><p>联络我们：微博<a href="https://weibo.com/u/6486678714" rel="noopener noreferrer nofollow" target="_blank">@张小珺-Benita</a></p><p>更多信息欢迎关注公众号：张小珺</p><figure><img src="https://image.xyzcdn.net/Fn7o36NtUYpCM_rQiFj1LW-TIwk8.JPG" /></figure>

The guest today is Ming Chaoping (Xiao Ming/Leon), an AI application entrepreneur. In 2024, the AI narrative was still about large models. *Commercial Interview* interviewed Yang Zhilin, Wang Xiaochuan, Li Kaifu and other founders of large model companies. In the blink of an eye, the AI narrative in 2025 has become about application companies and Agents—**a new protagonist has appeared.** This episode, like the previous "3-Hour Interview with Xiao Hong, Founder of Manus: The World is Not a Linear Extrapolation, Be an Important Variable in the Game", **also comes from the front lines of "AI application explosion" and "Agent explosion."** In China's venture capital circle, Ming Chaoping is an entrepreneur who has received relatively consensus capital support from the very beginning. Born in 1995, he has worked on products at OnePlus, ByteDance, and Moonshot. This is his first time as CEO, and the first product he launched is called YouWare. He has some connections with our previous two guests: one is Yang Zhilin. In 2023, he had an in-depth conversation with Yang Zhilin for 10 hours, from day to night, and decided to join Moonshot after the conversation. The other is Xiao Hong. Sometimes I hear people in the venture capital industry compare Xiao Ming and Xiao Hong, saying that they both belong to **"Hands-on, product-oriented entrepreneurs."** Is this the case? Today's 3-hour interview hopes to present Xiao Ming's true state. Whether it is true is up to everyone to decide. However, although they are often associated, Xiao Hong and Xiao Ming have never met. Looking forward to 2025, we will progress together with AI : ) Our podcast is premiered on Tencent News, you can go and follow us, so you can get program information and more news as soon as possible : ) 03:16 Quick Q&A; begins **Those scattered, rebellious, and frustrating teenage years** 04:36 Childhood and teenage fragments 06:52 I debated at Wuhan University, especially good at the fourth speaker 13:00 The most important thing I learned from debating: **"Always have a third-party perspective, debating is not about convincing your opponent," "Turn yourself into an idiot in 1 second" (by Zhang Xiaolong)** 14:40 I basically slept in the lab for the last two years of college, immersed in the "Intelligent Vehicle Competition" and won the national prize 19:43 **Oh, so miserable, a painful experience - I was the team member who lost the most competitions and also won the most best debater awards** 23:06 That was the senior's retirement match, and I still feel very guilty today **The first three stops on the road to product manager: OnePlus, ByteDance, Moonshot** 26:37 First stop on the road to product manager: OnePlus mentor took us to take the subway and go shopping 30:18 "Experience is not data": Battery life data and battery life experience are not equal. 95-100% and 0-5% battery level are the times when users are most sensitive and anxious about the experience. 33:58 Second stop on the road to product manager: I was very uncomfortable when I first went to ByteDance and felt like I was a rookie 37:42 **What are the disadvantages of ByteDance's product methodology? "It will wipe out many flashes of inspiration"** 39:35 "Data is the 'rearview mirror of driving', but it cannot guide you forward" 41:03 **ByteDance doesn't have Steve Jobs, ByteDance doesn't have Zhang Xiaolong, but ByteDance has Yiming** - Yiming has talked to some very small startup teams. 42:08 What do you think of the fact that many entrepreneurs have emerged from ByteDance in recent years, but none have achieved great success? 43:28 **I talked with Yang Zhilin for 10 hours in 2023**, talking about music, art, hobbies, products, and past experiences. It was really 8 pm when we were going to eat pizza, and I said, "Why don't you tell me about the technology?" 50:14 The popularity and sudden stop of the overseas product Noisee 01:03:03 ByteDance was able to become ByteDance because it coincided with several important variables of the times (the popularity of mobile devices, bandwidth speed, recommendation engine) **Jump off the big ship and start a business!** 01:05:05 Immediately encountered the bitter lesson 01:11:59 The instinct to carve patterns, uncontrollably giving it more scaffolding, you will run counter to the biggest variable of this era 01:12:45 Suddenly realized that something was wrong - the product was stopped before it was launched 01:13:18 Insomnia epiphany: **One of the key indicators in the AI era is "token consumption speed", and we must pursue "per token valuation"** 01:16:33 **The "shell" is underestimated, it should be called "container" and "environment"** (the environment is the reactor of people) 01:17:52 It is quite irresponsible to only give users a Chatbot input box 01:21:18 Today's Coding development is similar to Camera in those days. In the early days, talking about cameras was about "people holding SLRs." The huge change was the emergence of a new group of people - "mobile photographers" 01:23:50 The early trend variable is the emergence of a new group of people, with rapid growth. **Today's new group of people is "Vibe Coder"** 01:25:23 Is Anthropic playing the role of Sony today? Other startups spend time on Camera/base models, or on - with the iteration of Camera, Snapchat, Instagram, TikTok, TikTok Live appeared **Today's Agent is like a gorilla just picking up a burning stick** 01:37:12 Two possible future ecosystems for Agents: analogous to Singapore vs. the United States 01:40:44 Page rank becomes Agent rank 01:42:07 If you turn all to C companies into to B companies, you will be resisted by everyone 01:44:02 The network effect of Agents 01:46:02 We also want to become OS Agent! - The path doesn't tell you : ) 01:46:30 I said to the team: **"99.9% of us are going to die"** 01:46:57 Today's Agent is like a gorilla picking up a stone and starting to smash things 01:47:58 "Always believe that the Model will get better, always believe that the Model has nothing to do with you" 01:49:50 The underlying model is creating smarter people, and application companies are adapting to our production needs through environment/experience 01:53:54 The OS Agent I envision: it is alive 01:58:26 Agents will next appear like tribes in human society, encountering trust issues, and needing ID cards and password locks 02:03:33 Observations on AI technology and products in the past 2 years (consuming tokens and squeezing intelligence in a more efficient way) **First time as CEO** 02:14:50 Provide emotional value to your employees 02:19:53 Post-90s founders are more confident, more free and easy, and more rebellious 02:21:17 Fundraising is booming, but I feel like I'm walking on thin ice 02:23:35 Consciously fighting against Ego 02:33:45 Chess player and the person playing chess 02:36:57 Final quick Q&A; Related episodes: Talking with Yang Zhilin about the year of large model entrepreneurship: the increment of human ideals, the probabilistic non-consensus, and Sora Talking with Wang Xiaochuan about the year of re-entrepreneurship: Responding to Zhu Xiaohu and the third possibility of China's AGI Talking with Li Kaifu: What should we do if the United States forms an AGI hegemony? 3-Hour Interview with Xiao Hong, Founder of Manus: The World is Not a Linear Extrapolation, Be an Important Variable in the Game 【More Information】 Contact us: Weibo @张小珺-Benita, Xiaohongshu @张小珺 For more information, please follow the official account: 张小珺

Original title: 101. 对YouWare创始人明超平3小时访谈：今天Agent像大猩猩刚拿起一根烧火棍

Original description: <figure><img src="https://image.xyzcdn.net/Flo18nNUSP7OUNlTf8UgCdHxio6O.jpg" /></figure><p>今天的嘉宾是明超平（小明/Leon），一名AI应用创业者。</p><p>2024年的AI叙事还是大模型，《商业访谈录》访谈了杨植麟、王小川、李开复等大模型公司创始人；稍一转眼，2025年的AI叙事已然变成应用公司和Agent——<strong>新的主角登场了。</strong></p><p>这集节目和往期<a href="https://www.xiaoyuzhoufm.com/episodes/67c3d80fb0167b8db9e3ec0f" rel="noopener noreferrer nofollow" target="_blank">《对Manus创始人肖弘的3小时访谈：世界不是线性外推，做博弈中的重要变量》</a>一样，<strong>也是来自一线“AI应用爆发”、“Agent爆发”的前沿声音。</strong></p><p>在中国创投圈，明超平是一位创业伊始就受到资本相对共识的创业者。他出生于95年，曾先后在OnePlus、ByteDance、Moonshot做产品。这是他第一次做CEO，发的第一个产品叫YouWare。</p><p>他和我们此前的两位嘉宾有一些渊源：一个是杨植麟，2023年他和杨植麟深谈了10个小时，从白天到黑夜，聊完决定加入Moonshot；另一个是肖宏，有时候我会听到创投业人士将小明与小红对比来聊，说他们都属于<strong>“Hands-on型、产品型创业者”</strong>。</p><p>是不是这样呢？今天的3小时访谈希望能呈现小明的真实状态，是不是大家说了算。</p><p>不过，虽然老被关联，小红与小明至今没见过。</p><p>期待2025，我们和AI共同进步：）</p><figure><img src="https://image.xyzcdn.net/Fm2F9n8vMm_n-xafqMi98xs3T3K8.png" /></figure><figure><img src="https://image.xyzcdn.net/FnQOGRhNOAkScnGSulOXQlc0KLhZ.png" /></figure><blockquote><p>我们的播客节目在<a href="https://view.inews.qq.com/u/8QIf3n5c64Ucuzne7gI%3D?devid=FF4E49E6-9C89-4986-A413-04E856F31262&amp;qimei=766696f2cd8f313d744bc2c9000012918102&amp;uid=100161026780" rel="noopener noreferrer nofollow" target="_blank">腾讯新闻首发</a>，大家可以前往关注哦，这样可以第一时间获取节目信息和更多新闻资讯：）</p></blockquote><figure><img src="https://image.xyzcdn.net/FvV-R5FBydYHGZAMyXAV1K1A9iJT.png" /></figure><p>03:16 开始的快问快答</p><blockquote><p><strong>那些散漫的叛逆的挫败的少年成长</strong></p></blockquote><p>04:36 童年和少年片段</p><p>06:52 我在武大打辩论，尤其擅长四辩</p><p>13:00 辩论让我学会的最重要的：<strong>“永远具有第三方视角，辩论不是说服你的对手”，“1秒钟把自己变成傻子”（by张小龙）</strong></p><p>14:40 大学后两年基本睡在实验室里，沉浸式打“智能汽车竞赛”，拿了国奖</p><p>19:43 <strong>哎，好惨，惨痛的经历——我是输掉比赛最多的队员，也是拿最佳辩手最多的队员</strong></p><p>23:06 那是学长的退役比赛，我到今天还很愧疚</p><blockquote><p><strong>产品经理之路的前三站：OnePlus、ByteDance、Moonshot</strong></p></blockquote><p>26:37 产品经理之路第一站：OnePlus导师带我们去坐地铁、逛商场</p><p>30:18 “体验不是数据”：续航数据和续航体验不划等号，95-100%和0-5%电量是用户对体验最敏感焦虑的时候</p><p>33:58 产品经理之路第二站：刚去字节极不适应，觉得自己很菜</p><p>37:42 <strong>字节产品方法论劣势是什么？“它会磨灭掉很多灵光一现的创意”</strong></p><p>39:35 “数据是‘开车的后视镜’，但它不能指引你前进”</p><p>41:03 <strong>字节没有乔布斯，字节没有张小龙，但字节有一鸣啊</strong>——有一些很小很小的创业团队，一鸣都聊过了</p><p>42:08 怎么看字节过去这些年出来了许多创业者，但无人大成？</p><p>43:28 <strong>23年和杨植麟聊了10个小时</strong>，聊音乐、艺术、爱好、产品、过去的经历，实在是到晚上8点我们要去吃pizza，我说“要不给我讲讲技术吧”</p><p>50:14 海外产品Noisee的走红与骤停</p><p>01:03:03 字节能成为字节，契合了时代几个重要变量（移动设备普及、带宽速度、推荐引擎）</p><blockquote><p><strong>跳下大船创业啦！</strong></p></blockquote><p>01:05:05 立马就遇到the bitter lesson（苦涩的教训）</p><p>01:11:59 想雕花的本能，情不自禁给它更多脚手架，你会和这个时代的最大变量背道而驰</p><p>01:12:45 突然意识到，这个东西不对劲——产品没上线就停掉了</p><p>01:13:18 失眠的顿悟：<strong>AI时代关键指标之一是“token消耗速度”，要追求“per token valuation”</strong></p><p>01:16:33 <strong>“壳”被低估了，应该叫“容器”和“环境”</strong>（环境是人的反应器）</p><p>01:17:52 只给用户一个Chatbot输入框，是蛮不负责任的</p><p>01:21:18 今天Coding发展和当年Camera类似，早期聊相机说的是“拿着单反的人”，巨大变化是出现了新的人群——“手机摄影师”</p><p>01:23:50 早期趋势变量是出现新的人群，增速快，<strong>今天的新人群是“Vibe Coder（氛围编程师）”</strong></p><p>01:25:23 今天Anthropic是不是承担索尼的角色？其他创业公司把时间花在Camera/基座模型上，还是花在——随着Camera迭代出现了Snapchat、Instagram、TikTok、TikTok Live</p><blockquote><p><strong>今天的Agent就像大猩猩刚拿起一根烧火棍</strong></p></blockquote><p>01:37:12 Agent未来可能的两种生态：类比新加坡vs美国</p><p>01:40:44 Page rank变成Agent rank</p><p>01:42:07 如果你把所有to C公司都变成to B公司，会受到大家的反抗</p><p>01:44:02 Agent的网络效应</p><p>01:46:02 我们也想成为OS Agent呀！——路径不告诉你：）</p><p>01:46:30 我对团队说：<strong>“咱们99.9%是要死掉的”</strong></p><p>01:46:57 今天的Agent像一个大猩猩拿起石头开始砸东西</p><p>01:47:58 “永远相信Model会变好，永远相信Model和你无关”</p><p>01:49:50 基础模型在造更聪明的人，应用公司在通过环境/经验适用我们的生产需求</p><p>01:53:54 我设想的OS Agent：它是活的</p><p>01:58:26 Agent接下来会像人类社会出现部落，遇到信任问题，需要身份证、密码锁</p><p>02:03:33 过去2年对AI技术和产品的观察（以更高效的方式消耗token、压榨智能）</p><blockquote><p><strong>第一次做CEO</strong></p></blockquote><p>02:14:50 给你的员工提供情绪价值</p><p>02:19:53 90后founders更自信、更洒脱、更叛逆</p><p>02:21:17 融资风生水起，我却感觉如履薄冰</p><p>02:23:35 有意识地对抗Ego</p><p>02:33:45 棋手和对弈的人</p><p>02:36:57 最后的快问快答</p><figure><img src="https://image.xyzcdn.net/FvVbUNblF7FHIjfdp3MmmbAdLZ8G.png" /></figure><p>相关单集：</p><p><a href="https://www.xiaoyuzhoufm.com/episodes/65e16b5b6144a933b1d968b5" rel="noopener noreferrer nofollow" target="_blank">和杨植麟聊大模型创业这一年：人类理想的增量、有概率的非共识和Sora</a></p><p><a href="https://www.xiaoyuzhoufm.com/episodes/65f77b6e6764957079e5d8eb" rel="noopener noreferrer nofollow" target="_blank">和王小川聊再创业这一年：回应朱啸虎与中国AGI第三种可能</a></p><p><a href="https://www.xiaoyuzhoufm.com/episodes/6730aa0bf373fe5d4d215d0c" rel="noopener noreferrer nofollow" target="_blank">和李开复聊聊：如果美国形成AGI霸权，我们应该怎么办？</a></p><p><a href="https://www.xiaoyuzhoufm.com/episodes/67c3d80fb0167b8db9e3ec0f" rel="noopener noreferrer nofollow" target="_blank">对Manus创始人肖弘的3小时访谈：世界不是线性外推，做博弈中的重要变量</a></p><p>【更多信息】</p><p>联络我们：微博<a href="https://weibo.com/u/6486678714" rel="noopener noreferrer nofollow" target="_blank">@张小珺-Benita</a>，小红书<a href="https://www.xiaohongshu.com/user/profile/5fede947000000000100603e?xhsshare=CopyLink&amp;appuid=5fede947000000000100603e&amp;apptime=1710049463" rel="noopener noreferrer nofollow" target="_blank">@张小珺</a></p><p>更多信息欢迎关注公众号：张小珺</p><figure><img src="https://image.xyzcdn.net/Fn7o36NtUYpCM_rQiFj1LW-TIwk8.JPG" /></figure>

In late March 2025, Mercedes-Benz global CEO Ola Källenius visited China for 9 days, during which I interviewed him. Mercedes-Benz is the world's first company to invent gasoline vehicles and is now in the most significant period of transformation in its 139-year history. It can be said that **Ola Källenius is the "CEO of a transitional period" who is leading this critical transformation.** Born in Sweden in 1969, he joined Mercedes-Benz in 1993 and has spent most of his career there. Six years ago, in 2019, he became the global CEO of Mercedes-Benz. It is worth mentioning that he is the first non-German CEO in Mercedes-Benz history upon taking office. I talked with Mr. Källenius about **his key strategic decisions in the past 6 years (including luxury car strategy, electrification strategy), the successes and failures of the Chinese market, whether technological equality and luxury cars are paradoxical, and the most significant critical transformation in Mercedes-Benz history under his leadership.** What I want to present to you is a 139-year-old giant in transition. Our podcast program premiered on Tencent News. You can go and follow it to get program information and more news as soon as possible :) * **04:12 Part 1: Talking about the Chinese market** * What are your most important views on the Chinese market in the past 6 years? * Your market share in China has been declining for the past 3 years, what are the reasons? * Does this mean you have lost the electric vehicle battle in China? * Do you emphasize luxury car strategy rather than electrification transformation? * Do you like the color TVs, refrigerators, and large sofas in Chinese cars? * Will your success or failure in the Chinese market determine the success or failure of the global transformation? * What innovative strategies are planned to reverse the situation in the Chinese market? * **18:24 Part 2: Talking about AI and new technologies** * The outside world says Tesla is 10 years ahead of you in electric vehicles, how do you respond? * But if you can't control all the technology, can you control your luxury cars? * As the automotive industry shifts to electrification and intelligent driving, Mercedes-Benz has not fully led these new technologies. Can Mercedes-Benz still dominate luxury? * You are testing solid-state batteries, can you share more progress? * How will artificial intelligence change the rules of the game in the global automotive industry? * China's DeepSeek is rising globally, would you consider cooperation? * With "technology democratization," technology is no longer super exclusive. Do consumers still need luxury cars? * If you had to choose between luxury and technology, which would you choose? * If Karl Benz were still alive, which one do you think he would choose? * **40:17 Part 3: Talking about the CEO in transition and Mercedes-Benz in transition** * Is the 139-year-old Mercedes-Benz giant now at an unprecedented turning point in its history? * Back to May 22, 2019, the day you took over as CEO, what happened that day? * As the helmsman of the era of change, please tell us the biggest difficulties you have faced so far. * Tesla and Chinese car companies are still founder-driven, while German automakers have gone through generations of professional managers. Will this make German automakers more conservative? * Do you ever feel that the 139-year-old giant is transforming slowly? * When you make every major decision, do you feel that Mr. Benz is watching you? Does this put a lot of pressure on you? * If you could ask Karl Benz a question, what would you ask?

Original title: 100. 对话奔驰全球CEO康林松：转型期CEO和转型之中的139岁奔驰

Original description: <figure><img src="https://image.xyzcdn.net/Flo18nNUSP7OUNlTf8UgCdHxio6O.jpg" /></figure><p>2025年3月底，奔驰汽车全球CEO康林松先生（Ola Källenius）来华9天，期间我对他做了一次访谈。</p><p>奔驰是世界上第一个发明了燃油车的公司，现在正处于奔驰139年历史上最重大变革时期，可以说<strong>康林松是主导这场关键变革的“一名转型期CEO”。</strong></p><p>他1969年出生于瑞典，1993年加入了奔驰，绝大多数职业生涯都在奔驰；6年前，他在2019年担任了奔驰全球CEO。值得一提的是，他是奔驰历史上第一位在上任时非德裔的CEO。</p><p>我和康林松先生聊了聊<strong>他上任6年的重要战略决策（包括豪华车战略、电动化战略）、中国市场的成与败、科技平权与豪华车是否有悖论，以及在他领导之下的这场奔驰有史以来最重大的关键变革。</strong></p><p>我想给大家呈现的是一个，转折之中的139岁巨人。</p><figure><img src="https://image.xyzcdn.net/Fm2F9n8vMm_n-xafqMi98xs3T3K8.png" /></figure><figure><img src="https://image.xyzcdn.net/FreJb-NMmeSH5rseRc5RzdK5mm-5.png" /></figure><blockquote>我们的播客节目在<a href="https://view.inews.qq.com/u/8QIf3n5c64Ucuzne7gI%3D?devid=FF4E49E6-9C89-4986-A413-04E856F31262&amp;qimei=766696f2cd8f313d744bc2c9000012918102&amp;uid=100161026780">腾讯新闻首发</a>，大家可以前往关注哦，这样可以第一时间获取节目信息和更多新闻资讯：）</blockquote><figure><img src="https://image.xyzcdn.net/FvV-R5FBydYHGZAMyXAV1K1A9iJT.png" /></figure><ul> <li><strong>04:12 第一部分：谈中国市场</strong></li> <li>过去6年，你对中国市场最重要的观点是什么？</li> <li>你们在中国的市场份额过去3年一直下降，什么原因导致的?</li> <li>这是否意味你们在中国的电动车之战中，输了？</li> <li>你们更强调豪华车战略，而不是电动化转型，是这样吗？</li> <li>你喜欢中国汽车里的彩电、冰箱、大沙发吗？</li> <li>你们在中国市场的成败会决定全球转型的成败吗?</li> <li>计划采用哪些创新策略来扭转在中国市场的局面?</li> <li><strong>18:24 第二部分：谈AI和新技术</strong></li> <li>外界说特斯拉在电动车上比你们领先10年，你如何回应?</li> <li>但如果你们无法控制所有技术，你们能控制自己的豪华车吗？</li> <li>随着汽车行业转向电动化和智能驾驶，奔驰没有完全引领这些新技术，奔驰还能主导豪华吗？</li> <li>你们正在测试固态电池，能否分享更多进展？</li> <li>人工智能将如何改变全球汽车行业的游戏规则？</li> <li>中国DeepSeek正在全球范围内崛起，你会考虑合作吗？</li> <li>随着“技术民主化”，技术不再是超级排他性，消费者还需要豪华车吗？</li> <li>如果必须在豪华和科技之间选择，你会选择哪一个？</li> <li>如果卡尔本茨先生还在世，你觉得他会选哪一个？</li> <li><strong>40:17 第三部分：谈转型期CEO和转型之中的奔驰</strong></li> <li>拥有139年历史的奔驰巨头，如今正处于其历史上前所未有的转折点上？</li> <li>回到2019年5月22日，你接任首席执行官的那一天，那天都发生了什么？</li> <li>作为变革时期掌舵者，请说出你至今面临过的最大困境</li> <li>特斯拉和中国车企仍然是创始人驱动，而德国汽车制造商经历了几代职业经理人，这是否会让德国汽车制造商更加保守？</li> <li>你有没有感觉过，这位139岁的巨人转型缓慢？</li> <li>当你做出每一个重大决定时，有没有觉得本茨先生在关注着你？这会让你有很大压力吗？</li> <li>如果能问卡尔本茨先生一个问题，你想问什么？</li></ul><figure><img src="https://image.xyzcdn.net/Fn7o36NtUYpCM_rQiFj1LW-TIwk8.JPG" /></figure>

In 2021, Sam Altman personally invested $375 million in Helion Energy, an American nuclear fusion startup, his largest personal bet to date. Helion boasts that it will build the world's first 50-megawatt fusion power plant by 2028. Musk has a different view. He once said, "We have an inexhaustible nuclear fusion reactor above our heads – the sun." He believes solar energy is the fundamental path to solving humanity's energy problems. However, in the eyes of many, **controlled nuclear fusion is still the "holy grail of energy."** As we move towards AGI today, energy will be the biggest bottleneck for civilization's evolution – after all, AGI may not fear humans, but it certainly fears power outages. In this episode, I invited Yang Zhao, the founder of Energy Singularity, a Chinese controlled nuclear fusion startup, to talk about it. Compared to AI, controlled nuclear fusion is a longer and less traveled entrepreneurial path. **It is almost facing one of the most complex physical problems in human history, standing on the edge of technology and human civilization, exploring technology.** In the program, Yang Zhao gave us a popular science introduction to the cutting-edge technology of controlled nuclear fusion; as a participant in China's controlled nuclear fusion cause, he also relatively clearly calculated **how much funding is still needed for humans to tame controlled nuclear fusion? How much further do we have to go?** We also talked about what our world and our civilization will be like in the more distant future when energy becomes infinite. Our podcast program premiered on Tencent News, you can go and follow it, so you can get program information and more news as soon as possible :) Quick Q&A from 03:00 **High-frequency professional term explanation** 04:10 Nuclear fusion, nuclear fission, controlled nuclear fusion, tokamak, high-temperature superconducting tokamak, there are only 3 full low-temperature superconducting devices in the world 13:22 The "International Thermonuclear Experimental Reactor (ITER)" jointly launched by many countries, a super-large tokamak device, has invested 25 billion euros and has a construction period of 30 years 14:47 Both high-temperature superconducting materials and low-temperature superconducting materials are low-temperature (high-temperature superconductivity can reduce the device volume by two orders of magnitude under the condition of energy gain, which also means that the construction cost is reduced by about two orders of magnitude) 19:17 A key indicator: Q value/energy gain is determined by the triple product (plasma density × temperature × confinement time), the world's highest Q value has just passed 1, and currently pursuing Q>10 **History of Controlled Nuclear Fusion** 21:55 Starting from Einstein's mass-energy equation E=MC², very small mass loss will produce huge energy 27:09 From hydrogen bombs to inertial confinement to magnetic confinement, different magnetic field shapes correspond to different magnetic confinement bifurcation technology routes 27:50 In the 1960s, the Soviet Union thought of the tokamak route with a donut-like magnetic field configuration 28:50 There are probably more than 100 tokamak devices in the world 29:11 Transition from the era of using copper to make tokamaks to using superconductors to make tokamaks 30:17 In 2024, we built the world's first full high-temperature superconducting tokamak ("Honghuang 70" device) **4 Years of Nuclear Fusion Entrepreneurship** 34:01 2021 idea: Maybe high-temperature superconductivity significantly reduces the device volume and reduces the cost by two orders of magnitude 39:43 After thinking clearly, build a team, starting with 4 people 41:05 Yang Zhao's personal background: Stanford Ph.D. direction is relatively basic physics, quantum gravity, string theory, the intersection of quantum gravity and quantum information, basic physics that is far from this world 46:36 Human science's "pearl" and "lifetime" series 51:05 Between graduating with a Ph.D. and starting a nuclear fusion business? Doing EIR (Entrepreneur in Residence) at GSR Ventures in the United States, the first venture combining artificial intelligence and music education 55:36 Early 2021, market research for nuclear fusion research institutes and suppliers **"Honghuang 70", "Honghuang 170" and Jingtian Magnet** 59:37 How is "Honghuang 70" built? What are the steps to make a full high-temperature superconducting tokamak? 01:07:33 Problems are constantly emerging in every link. The closer you get to the real object, the bigger and more problems you have, and the higher the cost of modification and repair 01:15:43 The significance of "Honghuang 70" 01:18:49 The "three-step" to the first controlled nuclear fusion product 01:21:10 The significance of Jingtian Magnet (large-aperture magnet) and "Honghuang 170" 01:35:18 Comments on the world's 3 full low-temperature superconducting devices (EAST in Hefei, China, KSTAR in South Korea, and JT-60SA in Japan) 01:38:40 "Honghuang 380", made according to the requirements of a demonstration power station that can run for a long time with complete energy **How much further do humans have to go to tame controlled nuclear fusion?** 01:41:40 Sam Altman's largest personal investment to date is Helion Energy: "The magnetic field configuration is linear, unlike ours which is a donut" 01:45:06 What is the relationship between nuclear fusion and AI? 01:52:13 The division and differences of the Chinese and American nuclear fusion market pattern 01:54:10 Our technology route is similar to that of CFS (Commonwealth Fusion Systems, a federal nuclear fusion system company spun off from MIT) 01:56:40 The raw material for true fusion commercialization needs to use deuterium-deuterium to generate electricity, not deuterium-tritium 02:03:16 What will the world be like when energy is infinite? 02:04:57 Talking about self and organization, climbing and falling 02:30:57 Final quick Q&A Energy Singularity site and device diagram located in Lingang, Shanghai: (Above: Outside the company) (Above: Inside the factory) (Above: "Honghuang 70" under construction) (Above: "Honghuang 70" plasma) (Above: The moment "Honghuang 70" was completed) (Above: "Jingtian Magnet") (Above: The testing system of "Jingtian Magnet", "Jingtian Magnet" lies in a large tank) 【More information】 Contact us: Weibo @张小珺-Benita For more information, please follow the official account: 张小珺

Original title: 99. 对能量奇点创始人杨钊3小时访谈：人类驯服可控核聚变还有多少路程？

Original description: <figure><img src="https://image.xyzcdn.net/Flo18nNUSP7OUNlTf8UgCdHxio6O.jpg" /></figure><p>2021年，Sam Altman以个人名义向美国核聚变初创公司Helion Energy注资3.75亿美金，这是他迄今最大的一笔个人下注。Helion豪言称，将在2028年前建成全球首座50兆瓦聚变电厂。</p><p>马斯克持不同看法。他曾说：“我们的头顶一直就有一个取之不尽、用之不竭的核聚变反应堆——太阳”。他相信太阳能才是人类能源问题的根本路径。</p><p>不过，在许多人眼中，<strong>可控核聚变仍然是“能源界的圣杯”。</strong></p><p>随着今天我们向AGI迈进，能源将是文明演进的最大瓶颈——毕竟，AGI或许不惧怕人类，但一定害怕断电。</p><p>这集节目，我邀请了中国可控核聚变创业公司、能量奇点创始人杨钊来聊聊。相比AI，可控核聚变是一条更漫长、更人迹罕至的创业之路。</p><p><strong>它几乎是面对人类有史以来最复杂的物理难题之一，站在科技与人类文明的边界上，做技术摸索。</strong></p><p>节目中，杨钊帮我们做了一次关于可控核聚变的前沿技术科普；作为中国可控核聚变事业的参与者，他也相对清晰地计算出了，<strong>人类驯服可控核聚变还需要多少资金要消耗？还有多少路程要走？</strong>我们也聊了聊，在更远处的未来，当能源成为无限，我们的世界、我们的文明又将怎样？</p><figure><img src="https://image.xyzcdn.net/Fm2F9n8vMm_n-xafqMi98xs3T3K8.png" /></figure><figure><img src="https://image.xyzcdn.net/FpvoKOCqm9m1BUQBqMrHw96qKSQf.png" /></figure><blockquote>我们的播客节目在<a href="https://view.inews.qq.com/u/8QIf3n5c64Ucuzne7gI%3D?devid=FF4E49E6-9C89-4986-A413-04E856F31262&amp;qimei=766696f2cd8f313d744bc2c9000012918102&amp;uid=100161026780" rel="noopener noreferrer nofollow" target="_blank">腾讯新闻首发</a>，大家可以前往关注哦，这样可以第一时间获取节目信息和更多新闻资讯：）</blockquote><figure><img src="https://image.xyzcdn.net/FvV-R5FBydYHGZAMyXAV1K1A9iJT.png" /></figure><p>03:00 开始的快问快答</p><blockquote><strong>高频专业名词解释</strong></blockquote><p>04:10 核聚变、核裂变、可控核聚变、托卡马克、高温超导托卡马克，全世界只有3台全低温超导装置</p><p>13:22 多国联合推出的“国际热核聚变实验堆计划”（ITER）， 一个超大型托卡马克装置，已投入资金250亿欧元，建设周期30年</p><p>14:47 高温超导材料和低温超导材料都是低温（高温超导在的能量增益的条件下，可以将装置体积缩小两个数量级，也意味建造成本大约缩小两个数量级）</p><p>19:17 一个关键指标：Q值/能量增益由三乘积（等离子体的密度×温度×约束时间）决定，Q值全球最高刚过1，目前追求Q&gt;10</p><blockquote><strong>可控核聚变的历史</strong></blockquote><p>21:55 从爱因斯坦质能方程E=MC²开始说起，非常小质量损失会产生巨大能量</p><p>27:09 从氢弹到惯性约束到磁约束，不同磁场形状对应不同磁约束分叉技术路线</p><p>27:50 上个世纪60年代，苏联想到用甜甜圈一样的磁场位形托卡马克路线</p><p>28:50 全世界大概有100台以上托卡马克装置</p><p>29:11 从用铜做托卡马克的时代过渡到用超导做托卡马克</p><p>30:17 2024年，我们建成全世界第一台全高温超导托卡马克（“洪荒70”装置）</p><blockquote><strong>核聚变创业这4年</strong></blockquote><p>34:01 2021年想法：也许高温超导显著缩小装置体积，将成本两个数量级降低</p><p>39:43 想清楚以后搭团队，最开始4个人</p><p>41:05 杨钊的个人背景：斯坦福博士方向是比较底层的物理，量子引力、弦论、量子引力和量子信息的交叉，离这个世界比较远的基础物理</p><p>46:36 人类科学的“明珠”和“有生之年”系列</p><p>51:05 从博士毕业到核聚变创业之间？在金沙江创投美国做EIR（驻场准创业者）、第一段人工智能和音乐教育结合的创业</p><p>55:36 2021年初，针对核聚变科研院所和供应商的市场调研</p><blockquote><strong>“洪荒70”、“洪荒170”和经天磁体</strong></blockquote><p>59:37 “洪荒70”是怎么建造的？做一台全高温超导托卡马克需要几步？</p><p>01:07:33 每个环节都在不断出问题，你越接近实物状态，你的问题越大、问题越多，改动修补成本越高</p><p>01:15:43 “洪荒70”的意义</p><p>01:18:49 通往第一个可控核聚变商品的“三步走”</p><p>01:21:10 经天磁体（大孔径磁体）和“洪荒170”的意义</p><p>01:35:18 点评世界上3台全低温超导装置（中国合肥的EAST、韩国的KSTAR、日本的JT-60SA）</p><p>01:38:40 “洪荒380”，按照完整能长时间运行的示范电站的要求做</p><blockquote><strong>人类驯服可控核聚变还有多少路程？</strong></blockquote><p>01:41:40 Sam Altman迄今为止最大的一笔个人投资是Helion Energy：“磁场位形是直线性的，不像我们是甜甜圈”</p><p>01:45:06 核聚变和AI的关系是什么？</p><p>01:52:13 中美核聚变市场格局的分割与差异</p><p>01:54:10 我们和CFS（Commonwealth Fusion Systems，美国麻省理工学院分拆出来的联邦核聚变系统公司）技术路线是相似的</p><p>01:56:40 真正聚变商业化的原料需要用氘氘去发电，而不是氘氚</p><p>02:03:16 当能源无限，世界会怎么样？</p><p>02:04:57 聊聊自我与组织、登山与跌落</p><p>02:30:57 最后的快问快答</p><figure><img src="https://image.xyzcdn.net/FvVbUNblF7FHIjfdp3MmmbAdLZ8G.png" /></figure><p>位于上海临港的能量奇点现场及装置图：</p><figure><img src="https://image.xyzcdn.net/FtoDf-jIxBPk5UcLWXsH2bx6cpnq.heic" /></figure><p>（上图：公司外）</p><figure><img src="https://image.xyzcdn.net/FlXzl7UYF1m6EqwuwPNgPVt2nOIg.heic" /></figure><p>（上图：厂房内）</p><figure><img src="https://image.xyzcdn.net/Fmm2ggP6exvPmlcbSpSV3ta_rbvk.jpg" /></figure><figure><img src="https://image.xyzcdn.net/Fny0cq8_EMT6r75CF2acahr18qBQ.jpg" /></figure><p>（上图：“洪荒70”建设中）</p><figure><img src="https://image.xyzcdn.net/FmZ1w85ssuv4Oc1hYJHPk-mt9KlZ.jpg" /></figure><p>（上图：“洪荒70”等离子体）</p><figure><img src="https://image.xyzcdn.net/FmyHUa_akdXsPDNFCo9NWhlX_GI7.jpg" /></figure><p>（上图：“洪荒70”建成时刻）</p><figure><img src="https://image.xyzcdn.net/FvpfDdcsAYING4Ev9n-fa70EG9lK.jpg" /></figure><figure><img src="https://image.xyzcdn.net/FkU1jHH4FxDVg2GZjeRcWyJ9ScsB.jpg" /></figure><figure><img src="https://image.xyzcdn.net/Fu-NkpbGsZ8xE-9SvZprld6PefFT.jpg" /></figure><p>（上图：“经天磁体”）</p><figure><img src="https://image.xyzcdn.net/FgvxPre7Y6V5DFnNKqYD4KJVmoZB.jpg" /></figure><p>（上图：“经天磁体”的测试系统，“经天磁体”躺在大罐子里）</p><p>【更多信息】</p><p>联络我们：微博<a href="https://weibo.com/u/6486678714" rel="noopener noreferrer nofollow" target="_blank">@张小珺-Benita</a></p><p>更多信息欢迎关注公众号：张小珺</p><figure><img src="https://image.xyzcdn.net/Fn7o36NtUYpCM_rQiFj1LW-TIwk8.JPG" /></figure>

今天的嘉宾是清华大学交叉信息研究院助理教授、星动纪元创始人陈建宇。他的研究和创业方向都是人形机器人。大语言模型浪潮爆发后，学界和工业界看见了机器人从专用走向通用的可能迹象，机器人革命随之而来。其中，本轮革命最重要的是，对机器人底层架构，也就是机器人“大脑”的探索。但通用机器人还在科学研究阶段，处于产业发展早期。这集节目，陈老师将带领大家，概览式阅读机器人基座模型和当下最前沿的架构VLA架构（Vision-Language-Action Model，视觉语言动作模型）的经典论文。希望我们的节目能直观地帮助更多人靠近科学前线，感受技术之美，并且能直观感知当前技术拐点。还是那句话：期待2025，我们和AI共同进步！（因为因为，陈老师真的分享了很多很多的动图和视频，本集结合视频服用效果更佳噢！可以前往：含投屏的视频版本。嘿嘿！预祝你学得开心！学得顺利啦！）我们的播客节目在腾讯新闻首发，大家可以前往关注哦，这样可以第一时间获取节目信息和更多新闻资讯：）02:30 陈建宇的研究和创业方向04:11 讲解开始前，先提问几个小问题17:36 当下最大变量：从专用模型到通用模型（robot foundation model）的可能性21:12 大模型浪潮爆发后，机器人领域经历了两个阶段：从利用基础模型进行机器人研究（leveraging foundation models in robotics）到为机器人预训练基础模型（pretraining foundation models for robotics）第一阶段：利用基础模型进行机器人研究（leveraging foundation models in robotics）21:59 机器人传统三板块：Planning+Perception+Actuation（规划+感知+执行）——第一步，用LLM（Large Language Model，大语言模型）替代Planning23:54 由Google Robotics团队提出的具身智能开创性论文Say Can《Do As I Can, Not As I Say: Grounding Language in Robotic Affordances》（中文名：我能做到，而不是我说到：将语言与机器人的可供性相结合）27:03 第二步，用VLM（Vision-Language Models，视觉语言模型）替代Perception27:52 来自Google的论文《Inner Monologue: Embodied Reasoning through Planning with Language Models》（中文名：内心独白：通过语言模型规划进行具身推理）29:51 由清华和上海姚期智研究院提出的《DoReMi: Grounding Language Model by Detecting and Recovering from Plan-Execution Misalignment》（中文名：DoReMi：通过检测和恢复规划-执行不一致来落地语言模型）32:47 第三步，想把Actuation进一步自动化，用Code LM（专门用于代码相关任务的大型语言模型）来替代Actuation32:24 由李飞飞团队提出的《VoxPoser: Composable 3D Value Maps for Robotic Manipulation with Language Models》（中文名：VoxPoser：使用语言模型进行机器人操作的可组合3D价值地图）第二阶段：为机器人预训练基础模型（pretraining foundation models for robotics）38:36 VLA端到端模型（Vision-Language-Action Model，视觉语言动作模型）——“人是很智能的VLA Agent”39:53 关于VLA的经典论文及分类：40:17 Aloha论文《Learning Fine-Grained Bimanual Manipulation with Low-Cost Hardware》（中文名：学习用低成本硬件进行精细双手操作）47:36 Mobile Aloha论文《Mobile ALOHA: Learning Bimanual Mobile Manipulation with Low-Cost Whole-Body Teleoperation》（中文名：移动ALOHA：使用低成本全身远程操作学习双手移动操作）50:15 论文《A Generalist Agent》介绍了一个名为Gato的通用型人工智能代理（中文名：通用型代理）52:45 RT-1论文《RT-1: Robotics Transformer for Real-World Control at Scale》（中文名：RT-1：机器人Transformer用于大规模现实世界控制）59:02 Octo论文《Octo: An Open-Source Generalist Robot Policy》（中文名：Octo：一个开源的通用机器人策略）01:02:20 CrossFormer论文《Scaling Cross-Embodied Learning: One Policy for Manipulation, Navigation, Locomotion and Aviation》（中文名：扩展跨具身学习：操控、导航、运动和飞行的统一策略）01:06:58 字节跳动AI Lab的两个工作GR-1和GR-2：《Unleashing Large-Scale Video Generative Pre-Training For Visual Robot Manipulation》（为视觉机器人操控释放大规模视频生成预训练模型）《A Generative Video-Language-Action Model with Web-Scale Knowledge for Robot Manipulation》（用于机器人操作的网络规模知识生成视频-语言-动作模型》）01:15:02 Palm-E论文《PaLM-E: An Embodied Multimodal Language Model》（中文名：PaLM-E：具身多模态语言模型）01:20:02 当前VLA最有名的开山工作：Google推出的RT-2论文《RT-2：Vision-Language-Action Models Transfer Web Knowledge to Robotic Control》（中文名：RT-2：视觉-语言-动作模型将网络知识迁移到机器人控制中）01:26:05 RT-X论文《Open X-Embodiment: Robotic Learning Datasets and RT-X Models》（中文名：开放X具身：机器人学习数据集与RT-X模型）01:31:16 《OpenVLA: An Open-Source Vision-Language-Action Model》（约等于开源版RT-2）（中文名：OpenVLA：一个开源的视觉-语言-动作模型）01:32:56 陈建宇课题组《HiRT: Enhancing Robotic Control with Hierarchical Robot Transformers》（中文名：HiRT：利用分层机器人Transformer增强机器人控制）01:38:40 Figure AI Helix，没发论文，但是今年Figure最新架构01:39:28 Pi0论文《π₀: A Vision-Language-Action Flow Model for General Robot Control》（中文名：π₀：一个视觉-语言-动作的流模型用于通用机器人控制）01:41:36 英伟达最近发布的GROOT N1模型《GR00T N1: An Open Foundation Model for Generalist Humanoid Robots》（中文名：GR00T N1：通用人形机器人的开放基础模型）01:42:32 《Diffusion Policy: Visuomotor Policy Learning via Action Diffusion》（中文名：扩散策略：通过动作扩散进行视觉运动策略学习）01:47:39 清华发布的《RDT-1B: A Diffusion Foundation Model for Bimanual Manipulation》（中文名：RDT-1B：双手操作机器人的扩散基础模型）01:51:04 《Prediction with Action: Visual Policy Learning via Joint Denoising Process》（动作预测：通过联合去噪过程进行视觉策略学习）和续作《Video Prediction Policy: A Generalist Robot Policy with Predictive Visual Representations》（视频预测策略：一个预测视觉表征的通才机器人策略）02:03:06 两个未来方向：《UP-VLA: A Unified Understanding and Prediction Model for Embodied Agent》（UP-VLA：具身智能体的统一理解与预测模型）《Improving Vision-Language-Action Model with Online Reinforcement Learning》（通过在线强化学习改进视觉-语言-动作模型）02:09:22 最后的提问【技术之美】系列：逐句讲解DeepSeek-R1、Kimi K1.5、OpenAI o1技术报告——“最优美的算法最干净”逐篇讲解DeepSeek关键9篇论文及创新点——“勇敢者的游戏”逐篇讲解DeepSeek、Kimi、MiniMax注意力机制新论文——“硬件上的暴力美学”【更多信息】联络我们：微博@张小珺-Benita，小红书@张小珺更多信息欢迎关注公众号：张小珺

很多人在催更《全球大模型季报》的2025年第一集，在Q1的最后一天，终于和大家见面了！这一集广密依然带来了信息满满的有关于全球大模型最新的的前沿认知。经历了最近几个月的全球AI格局巨变，他最大的变化是，重新坚信了Pre-training（预训练）——认为只有Pre-training才能决定模型内在的上限，涌现新能力，而Post-training+RL（后训练+强化学习）是加强。在这一集季报中，我们对于Q1的全球明星DeepSeek、作为模型“盗火者”的Manus、OpenAI的烟雾弹、硅谷的认知分歧与价值观、未来的范式级新路线，都进行了一一讨论。更重要的是，他更新了在一位AGI原教旨主义者的眼中，AGI的主线、珠峰与路途。希望《全球大模型季报》能持续陪伴你，2025，我们和AI共同进步！我们的播客节目在腾讯新闻首发，大家可以前往关注哦，这样可以第一时间获取节目信息和更多新闻资讯：）04:22 Pre-training叙事卷土重来今天最大非共识是Pre-training空间还非常大，Pre-training决定了模型内在的上限还是要再喊一下，大家还是要重视Pre-training了，这个最本质OpenAI现在不那么重视Pre-training，为什么？OpenAI的Pre-train核心团队一直挺动荡11:56 Coding是最通用赛博世界的环境，是模型的手我对两年内实现AGI有前所未有的信心Coding意义不在于编程本身，而在于Coding是数字经济GDP活动最重要的环境，是最通用的赛博世界的环境Coding是比搜索引擎和推荐引擎重要的东西19:55 OpenAI vs Anthropic：战略是不同组织能力的表达OpenAI和Anthropic同宗同源，最开始路线一样，但走着走着，核心战略bet或路线已经发生了分化OpenAI是frontier team做出了O系列，frontier team的老大Mark Chen现在成为了仅次于Sam/Greg的三号人物硅谷的认知分歧？这个问题本质是：智能重要，还是流量重要？我有点担心OpenAI过早的走向一家消费互联网公司30:18 一位AGI原教旨主义眼中的AGI roadmap（路线图）智能提升是唯一主线，智能本身就是最大应用今天回头看，ChatGPT只是这座高山山脚的第一站，后面还有很多个山头：Coding、Coding Agent、General Agent、AI for Science、RoboticsChatGPT只是前菜，接下来Agent才是正餐今天还是围绕智能主线，最重要的是push智能能力往上走，做应用的要构建一个环境或容器，承接研究溢出的智能红利文生图有可能是OpenAI烟雾弹今天做Robotics Foundation model/Research的做法不够本质26/27年可能是AI for Science爆发的时间点43:00 智能的本质是什么？这是个极好的问题——大家有想法可以打在评论区：）人类进化就3个关键词：1. 生存，2. 探索，3. 自动化智能进步的衡量标志是什么？一个Chatbot对话可能消耗几千个Token，一个Perplexity搜索大概几十K Token，但一个Manus平均可能要70-80万个Token48:03 Agent是新物种“智能水平离AGI越近，可能就越像宇宙大爆炸”Agent落地最关键的3个能力：1. Long Context reasoning, 2. Tool use, 3. Instruction following指令遵循AGI接下来的milestone是long-term memory，这个会取代long context55:49 未来范式级的路线，可能Online Learning是一个如果说未来还有范式级的路线，可能Online Learning是一个，让模型可以在线自主探索并学习对GPU或者英伟达叙事影响有多大？怎么看待贾扬清的公司（Lepton AI）被卖掉？英伟达在下一盘什么大棋？01:02:45 模型与产品的关系、壁垒和商业模式今天定价为什么20美元，是copy SaaS的定价吗？但SaaS不会消耗大量token裸模型发布的时代即将结束？形成壁垒主要是两个：一是成为Cloud，OpenAI自己变成微软的Azure Cloud；二是成为OS，要有生态，后面打造新的Operating System投资人怎么投AI应用？模型长期会把产品吃掉吗？本质是，feature system vs Learning system哪个更快Perplexity/Cursor/Manus都是“模型的盗火者”01:15:11 全球大模型公司竞争格局和全球AI产品公司GPT-4.5算不算领先？GPT-5为什么一直在跳票？OpenAI有没有失败的风险？怎么看待OpenAI支持了Anthropic的MCP协议？OpenAI和微软为什么会有裂痕？分家对微软影响多大？Manus vs Perplexity，都是执行力很强的团队，被称作“套壳之王”理想的投资组合：25% Anthropic, 25% Bytedance, 10% OpenAI, 10% Mira Thinking Machine Lab, 5% SSI, 5% Cursor, 5% Manus, 另外15%还没想好如果DeepSeek融资，我会放基金的25%01:54:32中美格局：如何跨越地缘封锁科技投资不是靠“混”能混出结果的，很多VC investor到处混圈子，其实没意义，还是得靠“创造”【全球大模型季报】系列2023年：口述全球大模型这一年：人类千亿科学豪赌与参差的中美景观2024年Q1：和广密聊AGI大基建时代：电+芯片=产出智能2024年Q2：口述全球大模型这半年：Perplexity突然火爆和尚未爆发的AI应用生态2024年Q3：AGI范式大转移：和广密预言草莓、OpenAI o1和self-play RL2024年Q4：大模型季报年终特辑：和广密预言LLM产品超越Google之路【免责声明】单纯内容分享，不作为投资建议。【更多信息】联络我们：微博@张小珺-Benita，小红书@张小珺更多信息欢迎关注公众号：张小珺

今天发布的是和理想汽车自动驾驶研发副总裁郎咸朋的聊天。本次访谈发生在2024年12月，和我们之前发布的《对李想的3小时访谈》在同一时期进行。郎咸朋13-18年在百度做自动驾驶，18年加入理想，过去10年都在中国的自动驾驶领域。他以亲历者的视角聊了自动驾驶10年演进史，详解了其中的关键节点和技术细节。这次谈话更像是对自动驾驶的一次技术科普。我觉得我们聊的还不错，所以决定分享给大家。（因为访谈发生在去年，如果大家听到今年，指的是24年；如果听到去年，指的是23年。）期待2025，我们和AI共同进步！我们的播客节目在腾讯新闻首发，大家可以前往关注哦，这样可以第一时间获取节目信息和更多新闻资讯：）01:32 10年前的自动驾驶当成有轨电车研发，现在看起来很可笑04:30 2018年，从高精地图+激光雷达到BEV+Transformer，Tesla是标杆12:07 当年激光雷达50-60万/台，早期百度/Cruise一辆车7-8个激光雷达，传感器成本远高于这辆车（当时我们在百度，一辆车500万人民币）13:09 为什么特斯拉要用视觉解决问题？为什么自己造芯片？15:16 特斯拉一辆车的传感器+芯片成本？一辆车有几个芯片？20:06 特斯拉总在用“升维”方式解决问题25:06 激光雷达和camera解决方案区别28:46 端到端、“我们以前做自动驾驶都做错了”41:14 我的工作经历：13-18年在百度，18年开始在理想50:50 “L3不是L2的延长，而是L4的先导”01:15:15 端到端是最典型的强化学习，端到端+VLM+世界模型是RL架构01:26:40 2024年3月李想对智驾团队发火01:32:23 “卫城”项目：“他就觉得你一定要跪下来求他”“老子就算死也要站着死”01:35:51 想过职业生涯栽在这儿吗？李想脾气不太好？【从蒸汽机到无人驾驶】系列对李想的3小时访谈（播客版）：宅男、AI、家庭、游戏和天梯和何小鹏聊，FSD、“在血海游泳”、乱世中的英雄与狗熊和楼天城聊聊Robotaxi和ACRush：“L2做得越厉害，离L4越远”从蒸汽机到无人驾驶3｜和孟醒聊特斯拉FSD进化史从蒸汽机到无人驾驶4｜Waymo和它的对手们：我暗中考察了四个月【更多信息】联络我们：微博@张小珺-Benita，小红书@张小珺更多信息欢迎关注公众号：张小珺

今天的嘉宾是肖弘（小红/小宏），一名AI应用创业者。我们从去年在不同时期进行谈话，可以算是展开了一场“接力式访谈”。大模型的模型能力还在迅速变化，身处其中的创业者要不断根据外部环境身段柔软地时时调整姿势。《商业访谈录》希望记录一名AI应用创业者，在技术变革之中、当一切都处于不稳定状态下的持续思考历程。而这个历程富有魅力之处就在于，它是变化的，而且还会继续变化。我们正在开启的2025可能会是AI应用爆发的元年、Agent爆发的元年，这集节目正是来自一线“AI应用爆发”、“Agent爆发”的前沿声音。肖弘提供了一种身处浪花中创业者的心态：“世界不是线性外推的，要让自己成为博弈中的重要变量。”期待2025，我们和AI共同进步！我们的播客节目在腾讯新闻首发，大家可以前往关注哦，这样可以第一时间获取节目信息和更多新闻资讯：）第一次访谈发生在2024年秋天，彼时小红刚完成一轮融资。（提醒大家注意的是，第一次访谈中，当我们说到今年、去年这些词语的时候，可能脑海里需要转化一下——“去年”指的是2023年，“今年”指的是2024年）03:03 开始的快问快答05:15 连续创业者、第一段创业、沮丧时刻、不同年龄段毕业生的最优选17:04 已经把大学赚的钱花完了，山穷水尽33:16 “VC是很贵的集资手段”，它的贵不体现在你不好的时候，而是体现在好的时候42:10 最重要的转折点：预判大厂的预判、《大空头》、等待、诱惑52:27 每次觉得自己技能点还可以的时候，就会冒出一个新维度——这次是资本56:19 嗅到泡沫的味道58:15 卖过公司founder的生活和心态、“让你的生活状态变得昂贵是更昂贵的”01:05:55 第二段创业的第一款产品：浏览器插件、Monica.im、ChatGPT for Google01:28:04 从2022年底创业开始每年重要的决策01:48:38 有模型 vs 没模型，“贸工技”vs“技工贸”，“模型是技术平权”02:02:14 我脑海中大模型应用的分类和方法论：主场景补充、模型能力带来的变化、模型能力在特定领域的外溢第二次访谈发生在2025年春节后，此时DeepSeek一时间改写了中国AI应用底层生态，我们坐下来又聊了一次。这次话题主要围绕他即将要发布的Agent产品（Manus，但当时还没上线），他讲述了对新产品的完整思考过程。02:09:32 2025年春节真是梦回2023年呐！02:10:09 我尝试把火过的AI应用作为少量数据点，总结规律做预测02:20:00 开源一个AI创业idea：预判大模型的下一个能力是什么，先做好应用，在那里等着02:21:01 大模型原厂做什么 vs 应用公司做什么02:23:51 DeepSeek最佛却取得最好结果，精神上给了大家鼓励：BE YOURSELF！！02:26:27 从产品角度解析为什么DeepSeek全球爆火，而OpenAI o1遗憾错过（02:29:00-02:29:44 注：这里有几处口误，不是Perplexity而是DeepSeek）02:31:12 对即将要发布的新产品的完整思考：国内Agent的第一枪！02:47:49 这几天正在经历“A-ha moment”，真的觉得在制造生命一样的东西02:49:27 在一切都不稳固的状态下，AI应用创业者应该保持什么心态？02:51:36 大厂能理解你的创新的时候就是很危险的时候02:58:24 用时代的年龄思考，而不是用生理的年龄思考03:01:02 我让DeepSeek解释“贪、嗔、痴”03:01:53 为什么不做底层模型？03:04:39 Peak Ji（Manus首席科学家）问黄仁勋：接下来几年什么事情发生会让你觉得很惊讶？黄仁勋的回答：Basically nothing。03:09:08 Founder是没得选的03:12:20 Founder应该用“博弈的方式”思考，而不是用“逻辑推理的方式”思考03:15:00 一位founder的生活和对世界的认知03:17:50 最后的快问快答【更多信息】联络我们：微博@张小珺-Benita，小红书@张小珺更多信息欢迎关注公众号：张小珺

今天这集节目延续我们的论文系列。我邀请MIT计算机科学与人工智能实验室的在读博士松琳，来给大家解读上个星期DeepSeek和Kimi发布的全新技术报告。DeepSeek和Kimi又一次技术对垒。在同一天发布论文，两篇集中在改进注意力机制以处理长文本任务上。而春节前，MiniMax也发布了一篇注意力机制相关的论文。松琳将带领大家阅读这3篇注意力机制有关的文章，解析不同模型公司的技术哲学和路线选择。我们希望能让更多人领略AI科技平权，体验技术之美。2025，我们和AI共同进步！（如果如果，你觉得光听还不够刺激，觉得一定要坐在电脑前看着投屏、拿起纸笔学习更有沉浸感…如果你实在是真心想要找虐的话…请前往：含投屏的视频版本。预祝你学习顺利啦！）我们的播客节目在腾讯新闻首发，大家可以前往关注哦，这样可以第一时间获取节目信息和更多新闻资讯：）02:30 讲解开始前，先提问几个小问题15:36 DeepSeek最新论文《Native Sparse Attention: Hardware-Aligned and Natively Trainable Sparse Attention》讲解论文中文名：《原生稀疏注意力：硬件对齐且原生可训练的稀疏注意力》路线：稀疏注意力机制本篇工作最大亮点：Native Sparse Attention 全线压制 Full Attention01:19:14 Kimi最新论文《MoBA: Mixture of Block Attention for Long-Context LLMs》讲解论文中文名：《MoBA：面向长文本上下文的块注意力混合架构》路线：稀疏注意力机制01:44:42 MiniMax春节前的论文《MiniMax-01: Scaling Foundation Models with Lightning Attention》讲解论文中文名：《MiniMax-01：利用闪电注意力扩展基础模型》路线：线性注意力机制02:30:07 最后强化学习一下【技术之美】系列：逐句讲解DeepSeek-R1、Kimi K1.5、OpenAI o1技术报告——“最优美的算法最干净”逐篇讲解DeepSeek关键9篇论文及创新点——“勇敢者的游戏”【更多信息】联络我们：微博@张小珺-Benita，小红书@张小珺更多信息欢迎关注公众号：张小珺

今天这集节目是春节前，我和前今日头条用户产品负责人、前MiniMax产品负责人张前川的一次聊天。张前川从2005年开始入行成为产品经理，经历过百度、360、字节、知乎等公司，最后一次出现在新闻是他从大模型公司MiniMax离职。这次聊天分成两个部分：第一部分是他作为一位20年的产品经理的思考，特别是他观察到的中国内容产品演变史。第二部分是他最近半年的所思所想。和上一集节目张亚勤对AGI勾勒的乐观前景不同，关于AGI，张前川发出了对人类威胁的预警，并分析人类灭绝的n种可能性。他告诉我，他接下来会考虑成立一家pro-human的非营利组织。（本集节目录制于春节前）我们的播客节目在腾讯新闻首发，大家可以前往关注哦，这样可以第一时间获取节目信息和更多新闻资讯：）内容产品的20年演变史 02:00 张前川自我介绍：一位产品经理的20年 （曾供职过的企业：百度、360、字节、新浪微博、知乎、MiniMax） 03:39 搜索引擎、拉里·佩奇和俞军 17:35 2005-2025，过去20年，内容分发产品的演化历史和内生逻辑 24:28 推荐引擎、今日头条和张一鸣 28:28 小红书在推荐引擎范式的延长线上吗？为什么小红书不是字节做的？ 39:29 微信送礼物的功能maybe对于小红书更有意义 42:50 泛化如水流有方向：女生向男生，年轻向年长用户的泛化更容易 52:00 面试时张一鸣问我的问题离开大模型公司的张前川，发出AGI预警这里对AGI推演和92集张亚勤的观点截然不同 54:10 AI是人类的最后一个发明，加速度无限大 57:00 把AI当目的 vs 把人当目的（更多公司是在for AI的目的？） 01:09:08 Agent就是“baby阶段的AGI” 01:19:56 人类灭绝的n种可能性（人利用AI灭绝人类/AI有意识的灭绝人类/AI无意识的灭绝人类） 01:24:30 人类如何幸福生存下去？ 01:26:03 如果AI产品收费模式是付费制，大部分人类会逐渐失去和AI的连接 01:40:21 AGI公司的承诺和制度安排 01:58:48 人类不要逃避危险本身而逃避思考【更多信息】联络我们：微博@张小珺-Benita，小红书@张小珺更多信息欢迎关注公众号：张小珺

人人都谈论AGI，但在可见的未来，AGI到底是一幅什么样的画面？针对这个话题，我邀请了清华大学智能产业研究院（AIR）院长、前百度公司总裁张亚勤院士，来聊聊。在他的脑海中，人工智能会按照信息智能〉物理智能〉生物智能的图谱逐步实现AGI。这种构想和信仰之下，AGI呈现的是一个乐观的前景画面。但这不是人们预感AI进化的全部图景，在之后的节目中，我也会推出对AGI的另一种思考与隐忧。我们的播客节目在腾讯新闻首发，大家可以前往关注哦，这样可以第一时间获取节目信息和更多新闻资讯：） 01:45 在AI冬眠期，微软亚洲研究院成立了 03:57 微软怎么管理亚洲研究院？目标如何设置？ 05:53 清华智能产业研究院（AIR）的目标和路径，不做基层通用大模型 10:45 我脑海中的AGI地图：信息智能〉物理智能〉生物智能 12:40 信息智能是大脑的智商，5年内达到AGI水平 14:25 物理智能首先落地是自动驾驶，人形机器人需要10年 15:51 机器人分三类：家庭机器人、工厂机器人、社会机器人 21:05 人形机器人：未来可能每个人有一个你的copy、你的“分身” 22:25 为什么大模型大幅加速了自动驾驶和具身智能的开发速度？ 39:38 生物智能：20年实现，人的寿命更长，诞生新物种 45:56 构想10年后的世界 51:27 最后推荐两本书【更多信息】联络我们：微博@张小珺-Benita，小红书@张小珺更多信息欢迎关注公众号：张小珺

2025年这个春节，DeepSeek一举改写了全球AGI大叙事。在万般热闹之际，我们特别想沉下来做一些基础科普工作。在《商业访谈录》89集节目中，我邀请了加州大学伯克利分校人工智能实验室在读博士生潘家怡，为大家对照解读了春节前的DeepSeek-R1-Zero、R1、Kimi发布的K1.5，以及OpenAI更早发布的o1技术报告。这些模型聚焦的都是大模型最新技术范式，RL强化学习，简单来说就是o1路线。今天这集，我邀请的是香港科技大学计算机系助理教授何俊贤。他的研究方向是大模型推理，从很早就开始关注DeepSeek的系列研究。我们会focus在最近引发全球AI届关注的DeepSeek上。何老师将带领大家从DeepSeek的第1篇论文开始，阅读经过挑选的这家公司历史上发布的9篇论文。我们希望帮助大家从一个更延续、更长期、也更技术底层的视角来理解DeepSeek，以及它所做的复现与创新工作；与此同时也希望能让更多人感受到技术之美。（如果如果，你觉得光听还不够刺激，觉得一定要坐在电脑前看着投屏、拿起纸笔学习更有沉浸感…如果你实在是真心想要找虐的话…请前往：含投屏的视频版本。嘿嘿！预祝你学习顺利啦！2025我们和AI共同进步！）我们的播客节目在腾讯新闻首发，大家可以前往关注哦，这样可以第一时间获取节目信息和更多新闻资讯：）03:01 讲解开始前，先提问几个小问题整体风格：Open、Honest、低调、严谨的科学态度DeepSeek基座模型21:00 《DeepSeek LLMScaling Open-Source Language Models with Longtermism》技术讲解45:48 《DeepSeekMoE: Towards Ultimate Expert Specialization in Mixture-of-Experts Language Models》技术讲解01:06:40 《DeepSeek-V2: A Strong, Economical, and Efficient Mixture-of-Experts Language Model》技术讲解01:40:17 《DeepSeek-V3 Technical Report》技术讲解DeepSeek推理模型02:05:03 《DeepSeek-Coder: When the Large Language Model Meets Programming - The Rise of Code Intelligence》技术讲解02:12:16 《DeepSeek-Coder-V2: Breaking the Barrier of Closed-Source Models in Code Intelligence》技术讲解02:47:18 《DeepSeek-Prover: Advancing Theorem Proving in LLMs through Large-Scale Synthetic Data》和《DeepSeek-Prover-V1.5: Harnessing Proof Assistant Feedback for Reinforcement Learning and Monte-Carlo Tree Search》技术讲解02:52:40 《DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning》技术讲解03:01:41 9篇论文到这里都讲完啦！最后我们一起强化学习一下！关于强化学习往期节目：AGI范式大转移：和广密预言草莓、OpenAI o1和self-play RL｜全球大模型季报4和OpenAI前研究员吴翼解读o1：吹响了开挖第二座金矿的号角王小川返场谈o1与强化学习：摸到了一条从快思考走向慢思考的路逐句讲解DeepSeek-R1、Kimi K1.5、OpenAI o1技术报告——“最优美的算法最干净”开源一场关于DeepSeek的高质量闭门会：一场关于DeepSeek的高质量闭门会：“比技术更重要的是愿景”【更多信息】联络我们：微博@张小珺-Benita，小红书@张小珺更多信息欢迎关注公众号：张小珺

1年前，在我的报道《朱啸虎讲了一个中国现实主义AIGC故事》中，他的观点淋漓尽致地展现了一个现实版中国AI故事。他用“我们一看就知道，这个肯定没戏”，“我们一开始就说了，我就不看好大模型”，“ 我都不愿意去聊，你知道吗？这没有意义”，表态绝不会投资6家中国大模型创业公司中的任何一家。然而，时隔1年，Allen Zhu对待通用人工智能的态度出现了惊奇的逆转。这是朱啸虎现实主义故事的1周年连载。希望《商业访谈录》能持续为大家呈现中国AI市场的风云变幻与市场中人的心态起伏。2025，期待我们和AI共同进步！我们的播客节目在腾讯新闻首发，大家可以前往关注哦，这样可以第一时间获取节目信息和更多新闻资讯：） 01:30 DeepSeek快让我相信AGI了 03:08 这是对领先者的curse、诅咒 05:21 闭源模型还有没有价值？是很严峻的灵魂拷问 06:00 20天做到2000万DAU：DeepSeek是全球App增速历史第一，不需要任何限定语 07:00 梁文锋觉得，意识不一定是一个非常高技能、高门槛的事情 11:19 不用担心在别人的地基上盖房子，对AI应用公司是极大的解放 14:40 初始语料需要博士级别、各个领域专家级别的人来打标签 15:53 DeepSeek这次唯一没有公开的可能就是预训练语料 17:40 OpenAI成本很高，如果不能持续保持领先，挑战会挺大 18:09 中国大模型公司都需要重新思考 20:05 今天的DeepSeek是追赶者还是创新者角色？ 21:02 怎么看梁文锋？梁文锋是你的反面吗？ 21:40 我肯定会投啊！我肯定会投！——这个价格已经不太重要了，关键是参与在这里面 22:59 我今天上午和梁文锋说，R1可能会被认为是机器AI意识的元年 24:00 至少搜索肯定是被彻底取代了——这是毫无疑问的！ 26:14 AI产品的数据飞轮价值不大，这是我这两年最大一个教训 27:51 如果你是DeepSeek CEO，你当下最关切、最紧要需要解决的问题是什么？ 31:32 怎么看特朗普上任第二天宣布的Stargate（星际之门）项目？怎么看英伟达？ 33:29 “朱啸虎们”怎么看待“梁文锋们”？ 34:07 复盘1年前《朱啸虎讲了一个中国现实主义AIGC故事》里的观点，哪些被打脸？哪些更坚定？ 36:16 尤其在中国，你可以假设底层模型是免费的！ 37:07 回顾2024年的大模型行业，你会把哪些时刻当作关键节点？ 40:45 字节今天如果马上改开源追赶，那也不太容易 41:30 今年会不会看到通义千问把自己的生态向DeepSeek兼容，这样一个标志性事件？ 42:08 点评新出现的AI产品（包括但不限于Perplexity、Cursor、Devin） 47:39 聊小红书、具身智能、杭州和理想汽车 52:26 现在全世界就只有中美有AI能力，出海都是Low-Hanging Fruits（低垂的果实） 52:57 我今天看到在新加坡的泡泡玛特，买盲盒还要配货，你想到吗？！买盲盒都要配货！！ 55:30 答DeepSeek问 59:50 最后的快问快答朱啸虎往期节目：你们要的朱啸虎，来了1年前的报道：《朱啸虎讲了一个中国现实主义AIGC故事》【更多信息】联络我们：微博@张小珺-Benita，小红书@张小珺更多信息欢迎关注公众号：张小珺