AI News AI资讯 17h ago Updated 1h ago 更新于 1小时前 48

Post-00s Entrepreneurs Take Center Stage: 'Go Big or Go Home' | 36Kr Offline Gathering Season 2 00后创业者站上C位,「go big or go home」|36氪离线聚会第二期

What has always hindered the discovery of truth is not bare-faced lies, but those "profound yet incorrect insights." This quip by the German Lichtenberg over two centuries ago is being played out vividly in today's AI field. Over the past two years, the industry has been enveloped by a set of neat, "profound insights": larger parameters, more GPUs, and massive token consumption have become the yardstick for measuring everything and the totem of belief. A Silicon Valley-style "Token arms race" ha 妨碍真理发现的,从来不是赤裸的谎言,而是那些“极其精辟的错误见解”——德国人利希滕贝格两百多年前的讽刺,在今天的AI领域正被演绎得淋漓尽致。过去两年,行业被一套简洁的“精辟见解”所笼罩:更大的参数、更多的GPU、更海量的Token消耗,几乎成了衡量一切的标尺和信仰的图腾。一场硅谷式的“Token军备竞赛”悄然上演,仿佛谁能烧掉最多的算力,谁就自动赢得了通往AGI的门票。这套逻辑如此直观,以至于没人愿意多问一句:这消耗的方向,是对的吗?当你把Token当作燃料疯狂倾倒时,那台引擎的图纸,你真的看懂了吗?

68
Hot 热度
72
Quality 质量
65
Impact 影响力

Analysis 深度分析

This is precisely the most paradoxical landscape of the current AI ecosystem. We are obsessed with measuring "input" yet reluctant to examine the substance of the "output." Consequently, the industry is filled with two opposing voices: on one side, the clamor of grand narratives, constantly predicting comprehensive replacement of humanity; on the other, the silence of pragmatists, who in their corners strive to ground AI in reality, solving those specific and gritty real-world problems. A recent "offline gathering" hosted by 36Kr acted like a touchstone. It did not indulge in the former, but instead invited several founders born after 2000 to the stage, pressing them on more fundamental, even somewhat "clumsy" questions: What real productivity has AI actually created? How does it change a company's organization? How should we, amidst the bubbles and hype, anchor our own course?

The three young founders on stage constituted, in unison, a corrective to the "Token frenzy." They all chose "non-consensus" paths, venturing beyond the hottest wave of large models to tackle AI's toughest challenges—the physical world and scientific laws. Zheng Jiaxi, who works on underwater robots, has no interest in generic applications; he focuses on the meticulous inspection needs of offshore energy platforms amidst typhoon-force waves. Jin Ruofan, developing multi-agent systems for biopharmaceuticals, was already thinking back in 2022 when ChatGPT dazzled the world—not about how smooth the dialogue was, but why it couldn't invoke tools and enter laboratories to solve those complex, long-cycle scientific discovery loops. And the youngest humanoid robotics founder, Huang Yi, uses a controversial "full-stack open-source" approach, attempting to break the curse of slow hardware iteration. He wants the pace of robotic evolution to keep up with this heart-pounding era.

Their choices are essentially a collective rebellion against the industry's "profound incorrect insights." While mainstream consensus chases "smarter brains," they are working on "more dexterous bodies" and "more specialized domain knowledge." While others compare the parameter scale of models, they compete on the speed of iteration in interacting with the real world. The example from Jin Ruofan's team on virus research is stunning: a vague input leads to a genuine scientific hypothesis. This is not the merit of piling up model parameters, but of the Agent architecture behind it understanding and restructuring the research workflow. This signals a key turning point: AI's productivity is moving from the single dimension of "generating text" to the composite dimension of "driving processes." It is no longer just a mouth that can chat, but is becoming a pair of hands that can operate lab equipment, a pen that can draw design blueprints, and eyes and limbs that can plunge into the deep sea.

This also raises a deeper anxiety: Why hasn't AI, which theoretically promises huge efficiency revolutions, triggered a leap in organizational efficiency within many enterprises? The problem may not lie in the technology itself, but in our use of an old map to find a new continent. The bureaucracy of the industrial age and the flat structures of the internet age have not directly adapted to AI-native demands. AI, as "productivity itself," requires a thorough restructuring of an organization's decision-making processes, knowledge management, and collaboration models. What it needs is not just a few tools labeled with AI, but a data-driven, intelligent organizational form capable of real-time perception, rapid decision-making, and agile response. This is far more difficult and profound than simply introducing a large model API.

Youth is not a privilege in this transformation. On the contrary, it means a more rigorous scrutiny. Capital and the market are no longer paying for grand narratives in a PPT; they need to see solid progress, verifiable closed loops, and precise targeting of industry pain points. Huang Yi securing over 120 orders isn't because the "humanoid robot" story is sexy, but because his open-source strategy genuinely lowers the barrier to ecosystem participation, allowing universities and developers to enter the field at an affordable cost. Zheng Jiaxi and Jin Ruofan's projects are also rooted in critical areas like energy safety and biomedicine. No matter how advanced the technology, it cannot bear fruit if it is detached from the soil of industrial reality.

Perhaps this is the picture the second half of AI entrepreneurship should look like: bid farewell to the blind worship of "token consumption" and return to the devout pursuit of "problem-solving effectiveness." Young entrepreneurs are forced to prove in a shorter time that they not only understand AI, but also understand industry, understand organizations, and understand those non-linear pain points in a complex world. They need to be simultaneously technical experts, product managers, and industry insight leaders. This challenge is immensely difficult, but that is precisely where the opportunity lies. When everyone crowds into the same lane to compare whose horse-drawn carriage is faster, those quietly starting to develop internal combustion engines, though initially clumsy and misunderstood, may be truly touching the pulse of the era.

The future of AI will not belong to those who are best at explaining it, but to those who are best at applying it to solve concrete problems. The process of discovering truth ultimately cannot be completed in studies and conference halls; it requires repeated trial and error and iteration in muddy factories, turbulent waves, and complex laboratories. Those "profound incorrect insights" will still proliferate because they are easy and effortless to spread, but true progress always belongs to the "literal-minded" who dare to question consensus and blaze trails in the real world through action. This group of young entrepreneurs is becoming such "literal-minded" people. They may be unpolished, but they are closer than ever to the essence of AI as a productivity—not a triumph of rhetoric, but a triumph of action.

妨碍真理发现的,从来不是赤裸的谎言,而是那些“极其精辟的错误见解”——德国人利希滕贝格两百多年前的讽刺,在今天的AI领域正被演绎得淋漓尽致。过去两年,行业被一套简洁的“精辟见解”所笼罩:更大的参数、更多的GPU、更海量的Token消耗,几乎成了衡量一切的标尺和信仰的图腾。一场硅谷式的“Token军备竞赛”悄然上演,仿佛谁能烧掉最多的算力,谁就自动赢得了通往AGI的门票。这套逻辑如此直观,以至于没人愿意多问一句:这消耗的方向,是对的吗?当你把Token当作燃料疯狂倾倒时,那台引擎的图纸,你真的看懂了吗?

这正是当下AI生态最吊诡的景观。我们痴迷于度量“投入”,却懒于审视“产出”的实质。于是,行业充斥着两种声音:一边是宏大叙事的喧嚣,不断预言着对人类的全面替代;另一边则是务实者的沉默,他们在角落里,试图让AI接上地气,解决那些具体而粗糙的真实问题。最近36氪的一场“离线聚会”像一块试金石,它没有沉溺于前者,而是把几个00后创始人请上台,追问一些更本质的、甚至有些“笨拙”的问题:AI到底创造了什么真实生产力?它如何改变一家公司的组织?我们该如何在泡沫与热潮中,锚定自己的航向?

台上三位年轻的创始人,恰好构成了对“Token狂热”的一次集体纠偏。他们不约而同地选择了“非共识”路径,在最热门的大模型浪潮之外,去触碰AI最硬的骨头——物理世界和科学规律。做水下机器人的郑嘉熙,对通用的东西提不起兴趣,他盯着的是台风巨浪下,离岸能源平台那毫厘不能差的巡检需求;搞生物制药多智能体的金若凡,早在2022年ChatGPT惊艳世界时,思考的就不是对话有多流畅,而是它为何没能调用工具、走进实验室,去解决那些结构复杂、流程漫长的科学发现闭环。而最年轻的人形机器人创始人黄一,更是用“全栈开源”这种争议性的姿态,试图打破硬件迭代缓慢的魔咒,他想让机器人的进化速度,跟得上这个时代令人心悸的节奏。

他们的选择,本质上是对行业“精辟错误见解”的一次集体叛逃。当主流共识在追逐“更聪明的大脑”时,他们在解决“更灵巧的身体”和“更专业的领域知识”;当别人在比拼模型的参数规模时,他们在比拼与真实世界交互的迭代速度。金若凡团队那个关于病毒研究的例子堪称惊艳:模糊的输入,却能导向一个真实的科学假设。这并非模型参数堆砌的功劳,而是其背后Agent架构对科研流程的理解和重构。这暗示着一个关键转折:AI的生产力,正从“生成文本”的单一维度,迈向“驱动进程”的复合维度。它不再是一个只会聊天的嘴,而正在成为一双能操作实验仪器、一支能绘制设计图纸、一双能潜入深海的眼睛和手。

而这也引出了更深层的焦虑:为什么在理论上能带来巨大效率革命的AI,在很多企业内部却未能引发组织效率的飞升?问题或许不在于技术本身,而在于我们用旧地图在寻找新大陆。工业时代的科层制、互联网时代的扁平化,都未能直接适配AI原生的需求。AI作为“生产力本身”,它要求组织的决策流程、知识管理和协作模式进行一场彻底的重构。它需要的不是几个挂上AI标签的工具,而是一种能够实时感知、快速决策、灵活响应的数据化、智能化组织形态。这远比引入一个大模型API要困难和深刻得多。

年轻,在这场变革里绝非特权。相反,它意味着更严苛的审视。资本和市场不再为一个PPT里的宏大叙事买单,他们需要看到扎实的进展、可验证的闭环和对产业痛点的精准刺穿。黄一能拿到超120台订单,不是因为“人形机器人”的故事性感,而是因为他的开源策略实实在在降低了生态参与门槛,让高校和开发者能以可负担的成本进入赛道。郑嘉熙和金若凡的项目,同样扎根于能源安全和生物医药这类国计民生领域,技术再前沿,脱离产业实际的土壤也开不出果实。

这或许才是AI创业的下半场应有的图景:告别对“Token消耗量”的盲目崇拜,回归对“问题解决度”的虔诚追寻。年轻的创业者们正被迫在更短的时间内,证明自己不仅理解AI,更理解产业、理解组织、理解复杂世界里那些非线性的痛点。他们需要同时是技术专家、产品经理和行业洞察者。这种挑战无比艰巨,但也恰恰是机会所在。当所有人都挤在同一条赛道上比较谁的马车更快时,那些悄悄开始研发内燃机的人,尽管初期笨拙、不被理解,却可能真正触碰到了时代的脉搏。

AI的未来,不会属于那些最会解释它的人,而会属于那些最会运用它去解决具体问题的人。真理的发现过程,终究无法在书斋和会议厅里完成,它需要在泥泞的工厂、汹涌的海浪和复杂的实验室中,一次次试错、一次次迭代。那些“精辟的错误见解”依然会流行,因为它省力、易传播,但真正的进步,永远属于那些敢于质疑共识、用行动在真实世界里开辟路径的“笨人”。这群年轻的创业者,正在成为这样的“笨人”,他们或许稚嫩,却比任何时候都更接近AI作为生产力的本质——不是言辞的胜利,而是行动的凯旋。

Disclaimer: The above content is generated by AI and is for reference only. 免责声明:以上内容由 AI 生成,仅供参考。

大模型 大模型 机器人 机器人 开源 开源
Share: 分享到: