我的征尘是星辰大海。。。
The dirt and dust from my pilgrimage forms oceans of stars...
-------当记忆的篇章变得零碎，当追忆的图片变得模糊，我们只能求助于数字存储的永恒的回忆
作者:黄教授
手机视频列表
从2G的MID到AI大模型
视频
音频
原始脚本
从2G MID 到 AI 大模型，技术瓶颈里，需求永远是破局的光。 翻涌的科技浪潮里，总有一些看似精巧的临时解法，在时代的节点上昙花一现，却藏着最朴素的发展规律。 十几年前那段亲历的觅的创业往事，如今再看竟与当下大模型落地的困境惊人相似。 那些被遗忘的时代印记，也道出了比硬件突破、消解巧思更深刻的启示。 那是2G功能机称霸的年代，iPhone 尚未问世，3G还未掀起波澜。 无线网络带宽仅有几十 K 到100K，手机基带只为通话设计，孱弱的算力连解析 HTML 网页都做不到。 可即便技术条件如此窘迫，人们对信息的渴望却从未被束缚。 彼时传统媒体仍占据主流，报纸、杂志、电视都要包年订阅，固定的套餐，缓慢的节奏，挡不住普通人想要随时随地获取新鲜资讯的训的需求。 手机上只有运营商推送的短信新闻，寥寥文字，无链接、无详情。 和如今的社交平台信息传递相比，简陋的令人唏嘘。 可这已是当时移动信息获取的全部。 正是在这样的困境里，我所在的加拿大小公司做出了 MID 的移动互联网设备的尝试。 不纠结端侧算力与带宽的不足，把网页渲染解析全放在服务器，生成压缩后的 JPG 快照传回设备，手机只做简单的显示与交互，用最取巧的方式满足了人们用移动设备看门户网站新闻的需求。 在那个时代，这是夹缝里的最优解，精准踩中了信息需求的痛点，一度在细分市场里充满希望。 可谁也没料到2008年底的回国之行让一切戛然而止，iPhone 问世，3G网络爆发。 带宽从几十 K 跃升至兆级，手机全新的 CPU 架构彻底解放了算力。 曾经所有的技术束缚一夜之间烟消云散，那些费尽心思的巧思瞬间被降维打击，彻底淹没在时代浪潮里。 这段经历最戳人的，从不是临时方案的淘汰，而是看清了科技发展的核心逻辑。 所有技术瓶颈，在真实且强烈的需求面前，终究只是暂时的壁垒。 而需求永远是推动技术飞跃的最核心动力。 当年2G时代，带宽与算力的双重不足，没能阻挡人们对移动互联网的渴求。 这份渴望最终催生了3G 智能手机的革命彻底改写了整个行业。 如今的大模型时代，困境何其相似。 台式机、笔记本跑大模型尚且勉强，手机端更是几乎无法实现。 显存带宽、硬件算力的双重短板，像极了当年的移动网络与功能机。 可人们对人工智能的渴望丝毫不亚于当年对信息的追逐。 我们如今做的显存 优化、量化、分布式推理，种种奇技淫巧，不过是当下瓶颈里的临时解法，和当年的 mid 方案本质无异。 我们总感慨硬件突破能让一切巧思归于无形，却更该明白，从来不是技术瓶颈定义时代，而是未被满足的需求推着技术打破所有壁垒。 当年的信息革命是需求战胜了技术不足。 如今的 AI 浪潮，这份滚烫的需求也必然会催生出下一次硬件与技术的飞跃。 从2G MID 到 AI 大模型，变的是技术形态，不变的是人类对更便捷、更智能生活的向往，这份向往才是科技前行永远的底气。
修正脚本
从2G MID 到 AI 大模型，技术瓶颈里，需求永远是破局的光。 翻涌的科技浪潮里，总有一些看似精巧的临时解法，在时代的节点上昙花一现，却藏着最朴素的发展规律。 十几年前那段我亲历的创业往事，如今再看竟与当下大模型落地的困境惊人相似。 那些被遗忘的时代印记，也道出了比硬件突破、消解巧思更深刻的启示。 那是2G功能机称霸的年代，iPhone 尚未问世，3G还未掀起波澜。 无线网络带宽仅有几十 K 到100K，手机基带只为通话设计，孱弱的算力连解析 HTML 网页都做不到。 可即便技术条件如此窘迫，人们对信息的渴望却从未被束缚。 彼时传统媒体仍占据主流，报纸、杂志、电视都要包年订阅，固定的套餐，缓慢的节奏，挡不住普通人想要随时随地获取新鲜资讯迅猛的需求。 手机上只有运营商推送的短信新闻，寥寥文字，无链接、无详情。 和如今的社交平台信息传递相比，简陋得令人唏嘘。 可这已是当时移动信息获取的全部。 正是在这样的困境里，我所在的加拿大小公司做出了 MID 的移动互联网设备的尝试。 不纠结端侧算力与带宽的不足，把网页渲染解析全放在服务器，生成压缩后的 JPG 快照传回设备，手机只做简单的显示与交互，用最取巧的方式满足了人们用移动设备看门户网站新闻的需求。 在那个时代，这是夹缝里的最优解，精准踩中了信息需求的痛点，一度在细分市场里充满希望。 可谁也没料到2008年底的回国之行让一切戛然而止，iPhone 问世，3G网络爆发。 带宽从几十 K 跃升至兆级，手机全新的 CPU 架构彻底解放了算力。 曾经所有的技术束缚一夜之间烟消云散，那些费尽心思的巧思瞬间被降维打击，彻底淹没在时代浪潮里。 这段经历最戳人的，从不是临时方案的淘汰，而是看清了科技发展的核心逻辑。 所有技术瓶颈，在真实且强烈的需求面前，终究只是暂时的壁垒。 而需求永远是推动技术飞跃的最核心动力。 当年2G时代，带宽与算力的双重不足，没能阻挡人们对移动互联网的渴求。 这份渴望最终催生了3G 智能手机的革命，彻底改写了整个行业。 如今的大模型时代，困境何其相似。 台式机、笔记本跑大模型尚且勉强，手机端更是几乎无法实现。 显存带宽、硬件算力的双重短板，像极了当年的移动网络与功能机。 可人们对人工智能的渴望丝不亚於当年对信息的追逐。 我们如今做的显存 优化、量化、分布式推理，种种奇技淫巧，不过是当下瓶颈里的临时解法，和当年的 mid 方案本质无异。 我们总感慨硬件突破能让一切巧思归于无形，却更该明白，从来不是技术瓶颈定义时代，而是未被满足的需求推着技术打破所有壁垒。 当年的信息革命是需求战胜了技术不足。 如今的 AI 浪潮，这份滚烫的需求也必然会催生出下一次硬件与技术的飞跃。 从2G MID 到 AI 大模型，变的是技术形态，不变的是人类对更便捷、更智能生活的向往，这份向往才是科技前行永远的底气。
英文翻译
From the 2G MID to the AI large model, within the bottlenecks of technology, demand is always the light that breaks the deadlock. In the surging waves of technology, there are always seemingly ingenious temporary solutions that flash in the pan at the nodes of an era, yet they conceal the most fundamental laws of development. The entrepreneurial story I experienced firsthand more than a decade ago now shows a startling resemblance to the current predicament of deploying large models. Those forgotten imprints of the times also reveal insights far deeper than hardware breakthroughs or clever schemes. That was the era dominated by 2G feature phones, before the iPhone was launched, before 3G stirred the waves. Wireless network bandwidth was only tens of K to 100K, phone basebands were designed solely for calls, and the feeble computing power couldn't even parse HTML web pages. Yet even with such constrained technical conditions, people's thirst for information was never restrained. At that time, traditional media still dominated. Newspapers, magazines, and television required annual subscriptions with fixed packages and slow pace, but they couldn't stop ordinary people's urgent need to access fresh information anytime, anywhere. On phones, there were only SMS news pushed by operators—a few lines of text, no links, no details. Compared to the information dissemination on today's social platforms, it was pitifully primitive. But that was all there was for mobile information access at the time. It was in such a predicament that the small Canadian company I was with attempted the Mobile Internet Device (MID). Instead of struggling with the insufficiency of edge computing power and bandwidth, we placed all web page rendering and parsing on the server, generated compressed JPG snapshots, and transmitted them back to the device. The phone only handled simple display and interaction, using the most expedient method to meet people's need to browse news from portal websites on mobile devices. In that era, it was the optimal solution in a tight spot, precisely hitting the pain point of information demand, and it once brimmed with hope in a niche market. Yet no one expected that a return trip to China at the end of 2008 would bring everything to a halt. The iPhone was released, and the 3G network exploded. Bandwidth jumped from tens of K to the megabit level, and the phone's new CPU architecture completely liberated computing power. All the technical constraints that had existed vanished overnight. Those painstakingly crafted clever ideas were instantly crushed by technological leapfrogging and completely drowned in the tide of the era. The most striking takeaway from this experience is not the obsolescence of temporary solutions, but the clear vision of the core logic of technological development. All technical bottlenecks, in the face of real and intense demand, are ultimately only temporary barriers. And demand will always be the core driving force behind technological leaps. Back in the 2G era, the dual shortcomings of bandwidth and computing power couldn't stop people's craving for the mobile internet. That craving eventually gave birth to the revolution of 3G smartphones, completely rewriting the entire industry. Today, in the era of large models, the predicament is strikingly similar. Running large models on desktops and laptops is already challenging, and almost impossible on mobile devices. The dual limitations of memory bandwidth and hardware computing power resemble the mobile networks and feature phones of yesteryear. Yet people's desire for artificial intelligence is no less than their pursuit of information back then. The various tricks we are using now—memory optimization, quantization, distributed inference—are nothing more than temporary solutions in the current bottleneck, no different in essence from the MID solution of those days. We often lament that hardware breakthroughs can render all clever ideas obsolete, but we should understand that it is never technical bottlenecks that define an era; it is unmet demand that pushes technology to break all barriers. Back then, the information revolution was a triumph of demand over technical deficiencies. Now, in the AI wave, this burning demand will inevitably give rise to the next leap in hardware and technology. From the 2G MID to the AI large model, what changes are the forms of technology; what remains unchanged is humanity's aspiration for a more convenient and intelligent life. This aspiration is the eternal foundation that drives technology forward.
back to top