夜雨聆风
The Invisible Progress Bar of AGI
Recently, our screens have been flooded with all sorts of AI "magic," but a recent interview with Greg Brockman, co-founder and president of OpenAI, truly set the tech world on fire. He dropped a bombshell: the journey to Artificial General Intelligence (AGI) is already 70% to 80% complete. For those of us chasing new tools, this isn't just a number; it’s a signal that the future we once thought was decades away is practically knocking on our door.
最近，我们的屏幕被各种AI“神迹”刷屏，但OpenAI联合创始人兼总裁Greg Brockman的一场专访，才真正让科技圈炸开了锅。他抛出了一个重磅言论：人类通往AGI的路程，其实已经完成了70%到80%。对于我们这些新工具爱好者来说，这不只是一个数字，而是一个信号——那个我们曾以为还要几十年才能实现的未来，几乎已经在敲门了。
What exactly is AGI in Brockman's eyes? It isn't some sci-fi god that surpasses humans in every single physical and mental aspect. Instead, he defines it as a universal ability to efficiently control computers to handle all kinds of intellectual work, just like a human would. Think about it: AI already outshines us in coding and complex scientific calculations. Even if it still struggles to pick up a coffee cup or navigate a messy room, that doesn't stop it from hitting the core threshold of general intelligence.
在Brockman眼里，AGI到底是什么？它并不是某种在所有物理和智力领域都全面超越人类的“科幻神明”。相反，他将其定义为一种能够像人类一样，高效操控电脑处理各类智力工作的“通用能力”。想想看，现在的AI在写代码、科学计算这些核心任务上已经超越了人类，虽然让它去端杯咖啡或者收拾房间还存在短板，但这并不影响它逼近智力进化的核心门槛。
He shared a fascinating story about a physicist who had been haunted by a specific problem for ages. After submitting the problem to an OpenAI model, a perfect solution was returned in just 12 hours. The physicist was stunned, remarking that for the first time, he felt the model was truly "thinking." This ability to crack top-tier intellectual puzzles gives OpenAI the ultimate confidence: their path—directing pure text models toward AGI—is the right one.
他分享了一个特别有意思的故事：有一位物理学家被某个难题困扰了许久，结果把问题丢给OpenAI的模型后，仅仅12小时就得到了完美的解法。这位物理学家惊叹道，这是他第一次真切地感受到模型是在“思考”。这种攻克人类顶尖智力难题的能力，让OpenAI坚信：他们坚持的纯文本模型直通AGI的路线，是完全正确的。
The Strategy of "Focus" and the Unified Foundation
You might wonder why OpenAI isn't rushing to release the next version of Sora, their video generator that wowed the world. Brockman explained it’s about a hard choice in strategy. In a world of limited computing power, you can’t chase two rabbits at once. Sora is a different branch of the "tech tree" compared to the GPT models. To reach AGI faster, OpenAI decided to go "all in" on the GPT reasoning line, moving Sora into a long-term research plan for robotics.
你可能会好奇，既然Sora视频生成效果那么惊艳，为什么OpenAI最近反而不急着大规模推广它了？Brockman解释说，这涉及到战略上的取舍。在算力极度紧张的今天，你不能同时追两只兔子。Sora和GPT模型位于科技树的完全不同分支，为了更快实现AGI，OpenAI选择“梭哈”GPT推理路线，而将Sora纳入了机器人技术的长期研究计划。
The core of their technical strategy is a "Unified Architecture." Instead of using traditional diffusion models for images or separate systems for voice, OpenAI has brought everything—vision, sound, and text—under the GPT umbrella. This means all their research energy is laser-focused on one single, powerful engine. The real power of AGI lies in its ability to take seemingly unrelated tasks, like scientific research, coding, and casual chat, and run them all on one minimalist technical base.
他们技术战略的核心在于“大一统架构”。OpenAI并没有为图像生成采用传统的扩散模型，也没有为语音搞一套独立的系统，而是将视觉、听觉和文本全部统一到了GPT架构之下。这意味着，他们所有的研发精力都能像激光一样聚焦在一个核心引擎上。AGI真正的强大之处，就在于能将科学研究、编程、聊天这些看似无关的应用，全部装进同一个极简的技术底座中。
This isn't just about technical elegance; it’s a pragmatic business move. Brockman noted that the demand for AI in knowledge work is exploding right now, while robotics isn't quite ready for mass deployment. So, they are prioritizing the tools that can help you finish your report or debug your code today, rather than the AI that can fold your laundry tomorrow.
这不仅仅是为了技术上的优雅，更是一个务实的商业决策。Brockman提到，现在知识工作领域的AI需求正在爆炸式增长，而机器人技术还没到大规模部署的时候。所以，他们优先把算力留给了那些今天就能帮你写报告、改代码的工具，而不是明天才能帮你叠衣服的机器人。
From Chatbox to "Super App"
For those of us used to the ChatGPT chatbox, get ready for a change. Brockman described the future "Super App" as the ultimate evolution of ChatGPT. The chatbox as we know it might eventually vanish. What replaces it? A personal "Gold Medal Assistant"—a digital twin that truly understands your preferences, aligns with your goals, and is incredibly reliable.
对于习惯了ChatGPT对话框的我们来说，准备好迎接变革吧。Brockman描述的未来“超级应用”，是ChatGPT的终极进化形态。我们熟悉的那个聊天框可能会消失，取而代之的是你的“私人金牌助理”——一个深懂你的喜恶、与你目标对齐且极度可靠的数字分身。
This Super App will break down the barriers between different tasks. Imagine telling your AI, "I forgot how to set up hot corners on my Mac," and it just does it for you. Or asking it to analyze a complex spreadsheet and generate charts directly. It’s moving toward a "Codex for everyone." One OpenAI team member already uses it to automatically sort through piles of emails and summarize feedback, while others use it to automatically chop videos into chapters.
这个超级应用将打破不同任务之间的壁垒。想象一下，你只需吩咐一句“我不记得怎么在Mac上设置触发角了”，AI就会自动帮你调好；或者让它分析复杂的表格并直接生成图表。它正在变成“属于每个人的Codex”。OpenAI内部甚至有同事利用它自动梳理海量邮件并提炼核心信息，还有人用它给视频自动划分章节。
Crucially, this app will have a "long-term memory." Early ChatGPT felt like talking to a stranger who forgot you the moment the window closed. The future version will remember every interaction, mastering your specific context to provide hyper-personalized service. They are rolling this out in small steps, starting by making complex tools like Codex much easier for non-technical people to use. No more cryptic error messages; just results.
关键在于，这款应用将具备“长期记忆”。早期的ChatGPT就像一个每次见面都会失忆的陌生人，而未来的版本能记住你们的每一次互动，掌握你的上下文语境，从而提供极度个性化的服务。OpenAI正采取“小步快跑”的方式，第一步就是大幅提升Codex等工具的易用性，让不懂代码的普通人也能跨过报错提示的门槛，直接拿到结果。
Facing the Competition and "Shit-Pile Code"
OpenAI isn't working in a vacuum. Brockman openly admitted the pressure from competitors like Anthropic, whose Claude suite already shows the early shape of a Super App. He confessed that OpenAI previously focused too much on winning coding competitions in sterile laboratory environments, while neglecting the "last mile" of real-world usability.
直面竞争与“屎山代码”
OpenAI并不是在真空中搞研发。Brockman大方承认了来自Anthropic等对手的压力，后者的Claude应用矩阵已经初具超级应用的雏形。他坦言，OpenAI过去太执着于在实验室的纯粹环境下拿编程竞赛冠军，却忽略了现实世界中的易用性，也就是那“最后一公里”。,
Real-world code isn't clean; it’s often what developers call a "shit-pile" (messy and disorganized). Last summer, OpenAI formed a strike team to tackle this head-on, collecting high-fidelity data to train AI to handle the chaos of actual production environments. Now, Brockman claims their products have caught up or even surpassed competitors in blind tests. They are shifting their entire culture from "build a model first, wrap it later" to thinking about the end-user product from the very first day of research.
现实世界的代码可不干净，往往充斥着开发者口中的“屎山”。去年夏天，OpenAI专门组建了突击队攻克这个问题，通过高仿真数据让AI学会处理现实中的突发状况。现在，Brockman称他们的产品在盲测中已经追平甚至超越了竞争对手。他们正在经历一场深刻的转型：从“先做模型再套壳”，转变为从研发的第一天起就思考用户到底会怎么用。
he Mysterious "Spud" and the Smell of Intelligence
There’s a new name floating around the lab: Spud. It’s a brand-new pre-trained foundation model, the fruit of two years of research. Brockman views Spud not as a final destination, but as a critical engine for progress. He described an "AI smell"—that feeling you get when a model is so smart it actively tries to understand and adapt to you, rather than making you struggle to explain things to it.,
实验室里流传着一个新名字：Spud。这是一个全新的预训练基础模型，是过去两年研究心血的结晶。Brockman并不把Spud看作终点，而是一个关键的进步引擎。他描述了一种“AI的气味”——那种当你面对一个足够聪明的模型时，能感觉到它在主动理解你、顺应你，而不是让你费力地去解释自己的意图。
Spud will raise the ceiling of what AI can do—tackling open-ended scientific problems—and lift the floor, making everyday tasks like writing emails or taking notes significantly more useful. It’s about reducing frustration. Brockman noted that while humans are slow to change their "mental models" of what AI can do, once the technology crosses a certain threshold—like helping a terminal cancer patient find a successful alternative treatment—the impact becomes undeniable.
Spud将同时拔高AI的能力上限（解决开放性科学难题）并托起下限（让写邮件、记笔记这些琐事变得更顶用）。它的核心是减少挫败感。Brockman观察到，虽然人类转变对AI的认知很慢，但一旦技术突破了某个临界点——比如帮癌症晚期患者找到有效的替代疗法——AI的价值就会变得无法忽视。
AI Training AI - The Great Takeoff
Perhaps the most mind-bending part of the strategy is the "Automated AI Researcher," expected as early as this autumn. This is the moment AI enters its "takeoff" phase. We’re talking about AI systems that can do the end-to-end work of a research scientist—analyzing data, designing experiments, and verifying results—under the guidance of human experts.
也许整个战略中最让人脑洞大开的，是预计今年秋天就要面世的“自动化AI研究员”。这是AI进入“起飞阶段”的标志。我们说的是那些能自主完成数据分析、实验设计和结果验证的AI系统，它们将在人类专家的指导下，像初级研究员一样工作。
This creates a "positive feedback loop" where AI helps build the next generation of AI. It accelerates the development cycle exponentially. Brockman isn't worried about this being a cost center; he sees it as an income center. He views compute power like hiring a sales team: as long as there is more demand than supply, the more "salespeople" (GPUs) you have, the more value you create.
这创造了一个“正反馈循环”：AI帮着研发下一代AI，让开发周期呈指数级加速。Brockman一点都不担心这会成为成本包袱，反而认为它是“收入中心”。他把算力看作招募销售员：只要市场需求还没被填满，你手里的“销售员”（显卡）越多，创造的价值就越大。
Being the "Shepherd" of the AI Age
So, where does this leave us? Brockman acknowledges the fear surrounding jobs and societal impact. But he offers a different perspective: AI will free us from mechanical brain labor and allow us to return to our human essence—creativity, empathy, and deep connection. The era of "entrepreneurial democracy" is coming, where one person plus AI can do the work of an entire team.
那么，这一切把我们置于何地？Brockman承认人们对失业和社会影响感到恐惧。但他提供了一个不同的视角：AI将把我们从机械性的脑力劳动中解放出来，让我们回归人类的本质——创造力、共情能力和深层的人际连接。“创业民主化”的时代正在到来，一个人加上AI，就能完成过去一整个团队的工作。
His advice is simple but profound: don't just watch from the sidelines. You need to develop a "Sense of Agency." Think of yourself as a project manager. You give the orders, you assign the work, and you hold the accountability. If the AI messes up a website it built for you, it’s still your responsibility.
他的建议简单却深刻：别只在旁边看。你需要培养一种“主体操控感”。把你想象成一个项目经理，你发号施令、分配工作并把控质量。如果AI搞砸了它帮你建的网站，责任依然在你，而不是工具。
The most important first step is to get your hands dirty. Cross the threshold of that blank chatbox. Use the tools until you hit that "aha" moment where you understand their strengths and limits. AI can help you achieve your goals, but it cannot set them for you. You have to decide what you want to create and what value you want to bring to the world.
最重要的第一步是亲自动手。跨过那个空白对话框的门槛，去使用这些工具，直到你产生那种“豁然开朗”的感觉，看清它们的优势和局限。AI能帮你达成目标，但它不能替你设定目标。你得自己决定你想创造什么，想为这个世界带来什么价值。
其他思考：
The "Sense of Agency": Becoming the Shepherd of AI
In the eyes of OpenAI President Greg Brockman, the most critical skill for the AI era is a "Sense of Agency". This means shifting your mindset to that of a project manager who provides direction, gives orders, and assigns tasks. Crucially, while AI acts as a powerful agent, humans must remain the core of the accountability system; if an AI-built website fails and causes loss, the responsibility rests with the human user, not the tool. Ultimately, AI is a tool for fulfillment, but it cannot set goals for you; you must clearly define what you want to create and what value you wish to bring to the world.
“主体操控感”：做AI时代的牧羊人
在OpenAI总裁Greg Brockman看来，AI时代最关键的技能是“主体操控感”。这意味着将你的思维转变为项目经理，由你指引方向、发号施令并分配任务。至关重要的是，虽然AI是一个强大的代理人，但人类必须始终处于问责体系的核心；如果AI建立的网站搞砸了并造成损失，责任在于人类用户，而非工具。归根结底，AI是赋能工具，但它不能替你设定目标；你必须清晰地勾勒出自己想创造什么，以及想为世界带来什么价值。
Strategic Choices: Why OpenAI Paused the Sora Rollout
The decision to slow down Sora's large-scale promotion was a rational choice driven by technical reality and compute constraints. Sora resides on a completely different branch of the "tech tree" than GPT models, requiring a distinct construction method. In an environment of extreme compute shortages, OpenAI opted to "all in" on a unified GPT architecture—an engine that integrates vision, sound, and text under one roof to maximize research efficiency. Furthermore, OpenAI believes that AI applications in knowledge work are on the verge of a true explosion, whereas robotics and video-centric physics understanding are not yet ready for mass deployment. Consequently, Sora has been moved into a long-term research plan for robotics to wait for a more appropriate time.
战略取舍：OpenAI为何暂缓Sora推广
放缓Sora大规模推广的决定是基于技术现实和算力约束的理性选择。Sora位于与GPT模型完全不同的“科技树”分支，其构建方式截然不同。在算力极度紧张的环境下，OpenAI选择“全仓”投入大一统的GPT架构——这是一个将视觉、听觉和文本整合在一起的引擎，旨在实现研发效能的最大化。此外，OpenAI认为知识工作领域的AI应用正处于爆发前夜，而机器人技术和以视频为核心的物理理解尚未成熟到大规模部署的阶段。因此，Sora已被纳入机器人技术的长期研究计划，以等待更合适的时机。
Compute as an Income Center: The Logic of a $110 Billion Investment
Brockman challenges the traditional view of infrastructure costs, arguing that compute power is a "revenue center" rather than a cost burden. He compares GPUs to hiring salespeople: as long as your product is in high demand and you have scalable channels, the more "salespeople" (compute) you have, the more profit you generate. Currently, the growth of compute infrastructure cannot keep pace with the exploding demand for personal AI agents. By investing $110 billion, OpenAI aims to build a solid "profit flywheel" through consumer subscriptions and enterprise services, betting that the immense value released by the technology will far outweigh the initial capital expenditure.
算力即收入：1100亿美元投资背后的逻辑
Brockman挑战了传统的基础设施成本观，认为算力是“收入中心”而非成本负担。他将GPU比作招募销售员：只要你的产品供不应求且拥有规模化的渠道，你手下的“销售员”（即算力）越多，赚取的利润就越多。目前，算力基础设施的增长速度完全赶不上个人AI智能体爆发式的需求。通过投入1100亿美元，OpenAI旨在通过C端订阅和B端企业服务构建稳固的“盈利飞轮”，他们押注这项技术释放的巨大价值将远超最初的资本支出。
#AGI #OpenAI #GregBrockman #AI未来 #超级应用 #科技趋势 #源力研究所
「源力研究所 | Primo Dynamics
——创客码头旗下科技智库，赋能底层创新到商业裂变
聚焦硬核科技创业的「原理级赋能者」，依托创客码头十多年产业资源沉淀。
