136. 全球大模型季报第9集:和广密聊,Coding是AGI第二幕、硅谷御三家真相、模型正成为新一代OS
136. 全球大模型季报第9集:和广密聊,Coding是AGI第二幕、硅谷御三家真相、模型正成为新一代OS
Podcast1 hr 22 min
Listen to Episode
Note: AI-generated summary based on third-party content. Not financial advice. Read more.
Quick Insights

Investors should prioritize Anthropic (private) as the current leader in high-value coding agents, with the company projected to reach over $2 billion in ARR next year. While OpenAI (private) temporarily lost focus, watch for the release of their "Sparrow" model as the critical catalyst for them to reclaim the lead in agentic workflows. For a stable, long-term play, Google (GOOGL) remains a high-conviction "safe bet" due to its vertical integration with proprietary TPU hardware and massive distribution via Workspace. High-growth specialized startups like Manus (agents), Eleven Labs (voice), and Perplexity (search) offer significant upside as they capture specific niches of the AI economy. A balanced AGI Portfolio should allocate 20% to top-tier models like the "Big Three," 10% to agent infrastructure, and 10% to AI-driven science.

Detailed Analysis

Based on the detailed discussion in the podcast regarding the current state of the "Silicon Valley Big Three" (OpenAI, Anthropic, and Google) and the shift toward AGI's "Second Act," here are the investment insights and thematic breakdowns.


The "Coding as the New OS" Theme

The transcript highlights a fundamental shift: Coding is no longer just for programmers; it is the primary language of AGI and the "Second Act" of AI development.

  • The Feedback Loop: Coding has the shortest feedback loop in AI training, making it the "Amazon (Books)" of AGI—the first category to be fully conquered before scaling to other sectors.
  • Productivity Explosion: Top 1% of developers are seeing 10x-100x productivity gains. Systems that were 80% human-coded last year are now moving toward <1% human-coded.
  • Economic Impact: Coding automation is viewed as the precursor to automating global GDP.

Takeaways

  • Shift from Chat to Agents: Investors should look past simple "Chatbots" (2C) and focus on "Agentic Power"—models that don't just talk but execute high-quality jobs.
  • Token Usage as a Metric: DAU (Daily Active Users) is becoming less important than Token Usage. High token consumption by "Super-developers" and automated agents is the new indicator of platform value.

Anthropic (Private)

Anthropic is currently described as being in a "spring breeze" phase, having successfully pivoted to a coding-centric strategy.

  • Claude/Solnet 3.5: The success of these models proved that focusing on coding was the correct path when the consumer (2C) chat market became saturated by OpenAI.
  • Culture of Stability: Unlike OpenAI, Anthropic is noted for a more stable research team and a "mission-driven" AGI culture rooted in physics and quantum views of scaling.
  • Revenue Potential: Mentioned as a potential "New Mega 7" company, with the possibility of reaching $2 billion+ in ARR (Annual Recurring Revenue) next year.

Takeaways

  • Bullish Sentiment: Currently viewed as the leader in "Coding Agents." Their focus on data efficiency over raw transformer structural changes is yielding high-performance results.
  • Risk: They face intense competition as OpenAI re-prioritizes coding to catch up.

OpenAI (Private)

OpenAI is described as a "VC-led" organization that may have temporarily lost focus by chasing consumer growth and Sora (video) while neglecting the massive coding market.

  • The "GPT-5" Moment: The upcoming models (internally referred to as "Sparrow" or "Misos") are expected to be the true GPT-5 leap, moving from a "Chat" interface to a "Real Agent" mode.
  • Strategic Pivot: OpenAI reportedly realized only 2-3 months ago that the coding market is 10x to 100x larger than the chatbot market and is now aggressively "all-in" on coding.
  • Valuation: Despite internal friction, they remain a primary contender for the "OS of the future."

Takeaways

  • Watch for "Sparrow": This model release will be the key indicator of whether OpenAI can reclaim the lead in agentic workflows.
  • Platform Risk: The "VC-man" culture (Sam Altman) is noted as a risk for long-term research focus compared to the "Scientist" culture of Anthropic.

Google (GOOGL)

Google is characterized as the "most stable" long-term player due to its vertical integration.

  • The TPU Advantage: Google’s ownership of the hardware layer (TPU) provides a massive "worst-case scenario" safety net that other labs lack.
  • Gemini 3.0: While benchmarks are high, the user experience is noted as lagging. However, their integration into Workspace and the Android ecosystem makes them the most likely candidate for a "Global GDP OS."

Takeaways

  • Long-term Stability: Google is viewed as a "machine" that can withstand talent turnover.
  • Investment Insight: They are a "safe bet" in the AI race due to infrastructure (TPUs) and existing distribution channels, even if they aren't the "fastest" innovators.

Emerging Opportunities & Startups

The transcript identifies several high-growth companies in the AI ecosystem:

  • Cursor: A coding tool currently leading the user experience. However, the transcript suggests a "window of risk"—if top-tier models (OpenAI/Anthropic) integrate these features natively, Cursor may need to sell to a larger player like Tesla/Musk.
  • Manus: Mentioned as a high-potential "AI Agent" startup that could reach $100M to $1B in ARR very quickly.
  • Specialized AI: Eleven Labs (Voice), Suno (Music), and Perplexity (Search) are highlighted as fast-growing companies with strong revenue traction.

Takeaways

  • Hardware/Software Synergy: There is a growing trend of Silicon Valley AI firms seeking hardware talent in Shenzhen, suggesting a future boom in AI Robotics.
  • Investment Strategy: A suggested "AGI Portfolio" includes:
    • 20% in SOTA (State of the Art) Models (The Big Three).
    • 10% in AI for Science (Material science, Brain science).
    • 10% in Agent Infrastructure.

Key Risk Factors

  • The "Hole" in the Middle: While AI will bring long-term prosperity, there is a "difficult hole" in the transition where traditional jobs (especially junior developers and IT services in regions like India) face extreme displacement.
  • Speed of Obsolescence: The technology is moving so fast that companies have "no safety." A lead in one quarter can be wiped out in the next.
  • Human Value: As "intelligence" becomes a cheap machine commodity, the only remaining human value will be "Taste" and "Creativity."
Ask about this postAnswers are grounded in this post's content.
Episode Description
今天给大家带来的是全球大模型季报第9集,这一集的情绪十分复杂。 一方面,你会看到正在急速进化的AI革命。Coding把AI从聊天机器人Chatbot第一幕,推向了能够干活的Agent第二幕。研究员们已经不再亲自写代码。广密带来的核心判断是,Coding是新的”AI加速器",正在加速AGI实现,领先的Coding模型就像领先的GPU。 另一方面,社会层面随之进入白领通缩与失业的窗口。整个社会准备好了应对这么剧烈变化的AI革命吗? OUTLINE: 00:02:00 第9集季报的概览 00:03:28 硅谷体感与洞察 过去一个Q,智能水平进步幅度赶上2025全年,推背感非常强,AI奇点时刻应该很快就到 最关键转折点就是Opus 4.5 → Opus 4.6模型,算是GPT-3 → GPT-4跨代际提升的模型 今年6月之前可能还会再有一个从GPT-3→4水平幅度的跨越,OpenAI/Anthropic新模型都很强,Mythos/Spud下一代也开始训练了,继续加速 Coding会是头部的放大器,最顶尖1的人才能放大10-50倍的生产力 你是否相信Code可以表达数字世界的绝大多数任务?“语言即世界,代码即方案” 如果领先的模型公司不重视Coding,大概率会掉出第一梯队 Coding就像Amazon最早卖书一样,借助卖完书把所有仓储物流用户全都拉通了 Coding在整个AGI的历史进程中处在什么位置? 做好Coding的难点到底在哪里?不只是技术knowhow,更是战略组织和文化问题 00:22:10 硅谷御三家内部真实情况 00:22:10 Anthropic All in Coding不是day1这个团队就想清楚的 创始人特别hands-on亲自看训练数据,重视数据和重视技术细节刻在基因里 偏好“underdog”而非大人物,文化面试非常严格,尤其是看重一个人的文化特质 00:33:35 OpenAI 在这个时代的过去胜利秘诀,可能是下个时代的毒药 过去OpenAI在ChatGPT的成功,让他们专注ToC忽视了Coding 重点说下OpenAI做的不好的地方,可以在构建组织的时候借鉴一下 OpenAI即将要发布的新模型,是真正意义上的GPT-5吧 00:47:13 Gemini Gemini 3被高估 Coding严重落后,Google最大的战略失误 Google是最领先的追随者,资源和布局上没啥绝对短板,但战略跟随OpenAI/Anthropic,掉队可能性很低 00:54:16 Meta TBD 最有机会的挑战者,已经取代xAI,成为硅谷4号种子 00:58:07 xAI xAI短期基本上掉队了,最大问题感觉也是战略摇摆 崩溃始于Elon对整个founding team不满意,完全失去信心 01:02:00 Harness Engineering 01:03:57 中国御三家 01:05:42 模型是新一代操作系统 01:07:01 潜在的社会影响,失业、通缩 01:14:36 硅谷新趋势和投资新思考 《全球大模型季报》系列追踪: 2023: 54. 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 2024Q1: 64. 和广密聊AGI大基建时代:电+芯片=产出智能 2024Q2: 69. 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 2024Q3: 73. 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 2024Q4: 86. 大模型季报年终特辑:和广密预言LLM产品超越Google之路 2025Q1: 97. 25年Q1大模型季报:和广密聊当下最大非共识、AGI的主线与主峰 2025Q2-Q3: 112. 和广密聊大模型季报:分化与收敛、全家桶与垂直整合、L4体验与挖矿窗口 2025Q4: 127. 大模型季报跨年对谈:和广密预言一场AI War、两大联盟和第三个范式Online Learning LINKS: 我们的播客在小宇宙、Apple Podcast、Spotify等全音频平台播出; 我们的视频播客在Bilibili、小红书、视频号、抖音等全视频平台播出; 如果你想服用文字版,请搜索我们工作室的公众号:语言即世界language is world。 DISCLAIMER: 本内容不作为投资建议。 CONTACT: xiaojunzhang@lisw.ai Jump into the new world-and explore with us!😉
About 张小珺Jùn|商业访谈录
张小珺Jùn|商业访谈录

张小珺Jùn|商业访谈录

By 张小珺

努力做中国最优质的科技、商业访谈。 张小珺:财经作者,写作中国商业深度报道,范围包括AI、科技巨头、风险投资和知名人物,也是播客《张小珺Jùn | 商业访谈录》制作人。 如果我的访谈能陪你走一段孤独的未知的路,也许有一天可以离目的地更近一点,我就很温暖:)