伤痕我心

How Taliban Stormed Across Afghanistan, Smashing NATO-Trained Army in Ten Days_我的网站

A |     The August 15, 2021 fall of Kabul shocked the world, signaling the collapse of the West’s 20-year-long, multi-trillion-dollar nation-building experiment in Afghanistan in a single day.,Less than two weeks before the cataclysm, US and Afghan officials issued statement after statement expressing confidence in the Afghan military’s ability to fight on without American support on the ground.,“[The Afghan Army has] got greater numbers, they’ve got an air force, and, oh by the way, their air force is also flying…This is not an incompetent air force – they know how to fight from the air and they are. They have modern weaponry. They have military organizational ability as well as 20 years of training in the field by – not just the United States but our NATO and Allied partners. So they have a lot of advantages. It’s really just now about using those advantages,” Pentagon spokesman John Kirby told reporters in Washington on August 11, less than 100 hours before Kabul’s collapse.,In the chaotic few days that followed, Afghan Air Force personnel wound up flying nearly 50 helicopters and fixed-wing aircraft out of the country to neighboring Uzbekistan.,What happened? How did the US and NATO’s orderly withdrawal – which began in May 2021 amid fears that the Taliban would attack Western forces if they didn’t stick to the terms of the 2020 Trump-Taliban Doha peace deal, turn into such a humiliating and disorganized rout?,Paper Tiger vs Shoeless Army,On paper, the Afghan government and army seemed prepared for a years-long slog, similar to the one put up by the Soviet-backed Democratic Republic of Afghanistan, which managed to hold out against the Mujahedeen independently between 1989 and early 1992, even as the jihadists received billions in arms and cash from the CIA, Saudi and Egyptian intelligence, Pakistan and China. By comparison, the NATO-backed Afghan Army had serious problems with foreign support for the Taliban, with the insurgent group’s only major foreign backer consisting of Pakistan’s Inter-Services Intelligence.,The Afghan military boasted a total size of some 300,000 personnel – the 16th largest army in the world, along with an impressive array of $85 billion in US-gifted weaponry, including tens of millions of guns, grenades and rocket launchers, 8,500 Humvees, hundreds of armored personnel carriers, infantry fighting vehicles, tanks, mine-resistant ambush protected vehicles, and thousands of heavy and light trucks.,The Air Force’s inventory included more than two hundred Russian and US-made attack and utility helicopters, dozens of transport planes, and close to 30 Cessna 208 and A-29 Super Tucano ground attack and COIN jets.,US 'Lost Track' of Afghan War Aims, Says Retired Marine General12 August, 17:00 GMT,Facing off against them were the Taliban: a rag-tag group of militants whose core fighting force was estimated to consist of about 60,000 fighters, some of them shoeless, no air force, and limited mobility due to their adversary’s total air superiority and control of fuel.,Behind the scenes, a sprinkling of media reports and testimonies from Afghan soldiers and officials told a sobering story previewing the coming disaster, discussing the widespread disillusionment, demoralization and corruption centered around military and police structures based more on patronage, personal power and enrichment than competence and aptitude.,“There is no motivation for the army to fight for the corrupt government and corrupt politicians here,” Ahmad Wali Massoud, Afghanistan’s former envoy to the UK, told the Wall Street Journal on August 10, a few days before Kabul’s collapse.,“They are not fighting for [then-Afghan President Ashraf] Ghani. They haven’t even been fed properly. Why should they fight? For what? They are better off with the Taliban, which is why they are switching sides like that,” Massoud explained.,US Military Vehicles Falling Into Taliban Hands at Alarming Rate, Report Finds3 July 2021, 15:14 GMT,Ghani himself would end up fleeing the country as Kabul fell, with his entourage reportedly carrying so much money in suitcases from the state coffers that he had to leave some of it behind. Ghani has categorically dismissed these reports and denied allegations besmirching his good name.,John Sopko, the US inspector general for Afghanistan, warned Congress in March 2021 about the state of the Afghan’s military’s demoralization, admitting that “endemic corruption” was a threat, and that it was “fueling” the insurgency and giving the Taliban the means to gain public support.,Another problem, according to a May 2021 report by the Associated Press, was lack of professionalism. Attiqullah Amarkhiel, a former Afghan Army major general, told the news agency that while the Moscow-allied Afghan Army of 1989 consisted of about 150,000 high school educated troops. By contrast, he said, the NATO-trained crack troops of the Western-backed Afghan government had just six-to-eight weeks of basic training, with most recruits uneducated and/or allied to warlords.,“Then we had quality. Today we have quantity,” Amarkhiel said.,Blitzkrieg,This demoralization and corruption, combined with what The Intelligencer characterized as an institutionally built-in deep overdependence on US coalition support, culminated in the events of early-mid-August 2021.,By mid-July of that year, the Taliban was able to take advantage of coalition forces’ withdrawal by establishing control over large swathes of the Afghan countryside. However, no major cities and none of the country’s 34 provincial capitals were under the group’s control at the time. Throughout its two-decade insurgency campaign, the Taliban only launched one major strategic offensive against a government-controlled city – striking Kandahar in May 2011 before retreating back into the countryside.,Fast forward a decade, and the militant group would finally cast off their inhibitions about offensives cities.,On August 6, the militants captured Zaranj, capital of the southern province of Nimruz near the border with Iran, in a lightning attack, with government forces putting up light resistance and local bureaucrats fleeing the city.,A day later, the militants captured Sheberghan, capital of Jawzjan province, after heavy fighting. On August 8, Sar-e-Pul, Kunduz and Taluqan, three provincial capitals in the north of the country, fell, followed by Aybak in the northern Samangan province on August 9.,August 10 witnessed the collapse of Farah in the west and Pul-e-Khumri in Afghanistan’s center. A day later, Faizabad, capital of the northeastern province of Badakhshan also fell.,Animation showing Taliban's inexorable march towards Kabul. “The problem with the army was not lack of training or weapons. The most important thing in war is politics”Read @AmyKazmin @KatrinaManson and @b_parkyn's story on why the Afghan army foldedhttps://t.co/bjjJdPp15J pic.twitter.com/Ll6JrHQWLH— Steven Bernard (@sdbernard) August 15, 2021,August 12-13 would become the darkest days for the Afghan army and government up to that point, with the cities of Ghazni, Herat, Kandahar, Lashkar Gah, Qala-e-Naw, Feruz Koh, Pul-e-Alam, Terenkot and Qalat falling to the Taliban in the space of 48 hours. On August 14, the strategic northern city of Mazar-i-Sharif fell.,Finally, on August 15, 2021, Kabul and the eastern city of Jalalabad collapsed – falling, awkwardly, while tens of thousands of coalition troops, contractors, diplomats, Afghan puppet administration assistance and Western intelligence agents were still in the country.,© AP Photo / Rahmat GulA U.S. Chinook helicopter flies over the city of Kabul, Afghanistan, Sunday, Aug. 15, 2021. Taliban fighters entered the outskirts of the Afghan capital on Sunday, further tightening their grip on the country as panicked workers fled government offices and helicopters landed at the U.S. Embassy.A U.S. Chinook helicopter flies over the city of Kabul, Afghanistan, Sunday, Aug. 15, 2021. Taliban fighters entered the outskirts of the Afghan capital on Sunday, further tightening their grip on the country as panicked workers fled government offices and helicopters landed at the U.S. Embassy.© AP Photo / Rahmat Gul,In a post-op analysis of the events, General Yasin Zia, the former chief of staff of the Afghan army, estimated that some 4,000 security forces troops were killed, and 1,000 more gone missing, in the deadly weeks between July 1 and August 15, 2021 – with that figure representing about one half of the average of 8,000 Afghan security personnel killed each year over the previous five years.,Zia estimated that in total, some 92,000 members of the Afghan security forces died in the line of duty since 2001. Up to 100,000 Afghan civilians, tens of thousands of Taliban fighters, over 3,500 US and coalition troops, and 4,000+ Western mercenaries also perished.,* An organization under UN sanctions for terrorist activities.。    Token市场迈入“爆发元年”。

B | 截至2026年3月，我国Token日均调用量已突破140万亿，较2024年初实现超千倍增长；与此同时，科技巨头AI大模型API价格陆续下调，最高降幅达99%。         向上的需求与下行的单价，共同将Token经济推向全新的发展阶段。带着对Token行业未来范式的探寻，财闻走访了位于北京中关村软件园的软通动力集团全球总部。         在软通动力大厦的一层，科技感十足的展厅生动记录了这家软件企业向 AI转型的历程。         这家从IT服务起家的公司，在2024年完成对同方计算机业务板块的战略并购后，整合“软实力”与“硬科技”，正全力转向“软硬件一体、全栈智能”的新叙事。         软通动力展厅一角 曹井雪/摄          真正站在AI落地最前线的，是软通动力Token智能事业本部负责人、高级副总裁霍宇。

C | 这位长期泡在To B业务一线的老兵，一边把握着最前沿的AI技术，一边链接着企业最真实的需求场景，这也让他对Token经济有更为独到的理解。         霍宇照片 图源：采访者供图          公共资源，是很多人对Token未来属性的判断。然而，霍宇却觉得这个类比“并不严谨”。

D | 据他分析，目前Token“标准化”程度不足，且不同场景对模型和Token的需求差异显著，因此难以达到标准化程度高的公共资源水平。         他以大模型举例，高价模型与低价模型之间的质量可能会有极大的差距。同样地，Token市场方兴未艾，龙蛇混杂。如花大价钱买来的高性能Token，但背后实际调用模型版本可能与宣传不一致，这会导致运行的速度、稳定性等受到负面影响。C端用户对此感知并不明显，而B端的任何一次中断都可能转化为直接的经济损失。         “企业愿意为高质量、不间断、高性能的Token服务付费。”Token的热潮虽被C端驱动，但真正的“战场”却在B端的生产级应用。

E | AI正改变着企业的发展方向，同时也重塑着人才格局。         以下是访谈全文：          企业尚处于探索阶段，“Token不经济”仍在持续          财闻：近期，“Token不经济”的说法也引发热议——企业消耗Token产生的成本可能超过实际收益。你判断这是短期失衡，还是会长期存在？          霍宇：这种现象还会持续一段时间。

F | AI的演进速度极快。2025年我们有一个很深的感触——无论是内部研发的产品，还是协助客户做的项目，往往刚完成开发，立项流程都还没走完，技术已经过时了。         这本质上是企业必须交的一笔“学费”。就像持续学习一样，刚掌握的知识可能已不是最前沿的，我们必须不断向上探索。但如果不迈出这一步，与先进技术之间的鸿沟就永远无法跨越。         AI还有一个特殊的门槛，就是“体感”。

G | 没有真金白银的投入和实际应用，企业就无法真正摸清它的能力边界。         企业的天性追求“成本—收益”的确定性，而AI是基于概率的生成式系统，与传统的确定性程序逻辑截然不同。

H | 要把这两种逻辑融合到业务流程里，势必涉及组织架构乃至思维方式的重构，这需要相当长的时间。         因此，当下阶段，“投入大于产出”的局面难以避免。         财闻：这个“学费”要交到什么时候？是Token单价降到足够低，还是企业的应用能力追上来之后？          霍宇：两方面因素都在起作用。Token降价是大势所趋。目前国内厂商的主流路径就是降本增效，今年以来业界发布的几项新技术，几乎都指向降低成本和架构优化。         但更关键的，是企业必须想清楚一个根本问题——AI到底能做什么。         AI的本质是概率性智能，而非确定性工具。其优势在于寻找最优路径，它拥有无限可能，同时也伴随着幻觉的副作用。         这意味着，大量确定性业务场景其实并不需要调用最昂贵的模型。但行业目前的普遍习惯，仍然是直接追逐市面上最顶尖、最贵的模型，这造成了许多不必要的成本消耗。         真正追求投资回报的企业，必然会转向更具性价比的模型和私有化部署。开源模型生态如今十分活跃，一旦部署在自有服务器上，竞争的维度就不再是Token的按量消耗，而是算力、芯片和电力成本——这本质上是资本性支出的逻辑：一次性投入算力完成部署，长期享受收益。         不过，这一转变不会像外界想象的那么迅速，目前整个产业还处在早期探索阶段。这一轮AI热潮最初是由消费端驱动的，个人开发者、自媒体和小型团队在探索中受益显著，而真正的企业级大规模应用还很有限，产业红利尚未真正释放。         标准化程度不足，Token产品尚不具备公共资源属性          财闻：此前Token降价趋势明显，有观点预测Token将像“电力”一样逐渐成为公共资源。你如何看待这一类比？在你看来，Token的定价模式和资源属性最终会走向何方？          霍宇：这个说法并不严谨。云计算刚兴起时，同样被预判会像水电一样成为公共资源，但发展了近二十年，这一预期仍未实现。         一个重要误区在于，水电是高度标准化的产品，分类相对简单。

I | 而一家大厂关于云服务的产品就多达数百款，Token服务的种类同样极为繁杂。产品标准化程度不足，想要成为基础设施，进程自然会慢得多。         未来，Token很可能会走向两极分化。一类类似“民用电”——企业部署本地模型，首要考量是数据安全与稳定执行，而非追求参数最优；另一类则像“超高压电”——最顶尖的模型，但可能仅面向少数企业开放，某些能力过强、存在风险的模型并不会向所有人全面放开。（例如最新版本的Claude Mythos，由于其能力太强、存在安全防控风险，因此不向所有人开放。）          价格低廉的模型一定会出现，而且会越来越多。         财闻：你提到Token市场“不够标准化”，具体体现在哪些方面？          霍宇：目前市场较为混乱。

J | 同样是调用某个模型，官方原版的服务和第三方渠道提供的版本，质量可能天差地别。背后的缓存命中率、推理优化技术、服务保障水平都不一样，有些甚至卡到一字一顿，根本无法正常使用。         更严重的是，部分渠道声称提供的是某款主流模型，实际调用的却是参数量更小、成本更低的版本；还有通过非官方渠道带来服务质量差异的情况。这就导致用户产生误判，觉得“这个模型怎么变笨了，连简单任务都处理不好”。         打个比方：就像电动车，不同厂商的差距不只在于电机本身，座椅、中控、内饰等周边配置的组合协同，决定了车主的真实体验。Token服务不能只是一台裸露的电机，而是一整个配套的设备。如果只给企业一个基础引擎，让企业自行组装，最终结果可能根本无法运行。这就是标准缺失带来的问题。         财闻：这种混乱的市场格局，根源在哪里？          霍宇：根源在于，Token经济此前是由C端市场驱动的。

K | 消费级娱乐场景对服务中断有极高的容忍度，但进入生产环节后，任何一次中断都可能转化为直接的经济损失。那些估值位居前列的企业级Token平台，核心竞争力正是提供附带SLA承诺的API服务，保障不中断、有赔付。而C端服务普遍缺乏这类承诺，服务器一旦繁忙，对话可能直接中断。         当前，Token的使用正在从C端转向B端，真正的“战场”在企业级市场。市场之所以看好AI，是因为发现企业愿意为高质量、不间断、高性能的Token服务付费。企业购买的并非智能本身，而是可靠、稳定、有兜底保障的一整套能力。         软硬件一体服务在AI市场仍较稀缺          财闻：科技大厂与运营商都在布局Token业务，软通动力与它们相比，核心差异在哪里？          霍宇：科技大厂是强产品导向、技术导向。

L | 它们的AI研究员成本高昂，但让这些从事前沿研究的人才去做企业AI落地，往往并不匹配。大厂推出的AI产品整体上依然是像“运营专家”等偏重C端的工具。         而打开我们的专家广场后，看到的是“智能化养猪”“智能仓库分拣”。这些场景或许不那么新潮，但解决的却是一个一个的B端真实问题。

M | 所以，我们最核心优势在于大企业客户和软硬一体。目前市场上能够提供软硬一体AI服务的公司并不多，许多AI企业硬件能力很强，但与客户之间还隔着较远的距离。         软通动力睿动智能体云平台专家广场          财闻：软通动力自研Token业务的同时，也战略投资了硅基流动。

N | 内外两个生态如何协同？          霍宇：硅基流动侧重于生产侧和推理优化，拥有独特的技术能力，并且对所有模型保持中立。

o | 软通的优势则在于更贴近客户——无论是IT服务还是硬件交付，都是直接面对客户的某个具体场景或功能模块。         我们的客户需求十分多元，不会只锁定某一个模型，往往需要同时接入多种不同的模型，与硅基流动的互补性很强。         财闻：你负责的Token智能事业群，在整个战略中扮演什么角色？          霍宇：我们事业群的前身就专注于智能体业务，为企业提供定制化的智能体服务。在积累了众多案例之后会发现，仅凭Token本身，很难满足许多复杂场景下的客户要求，必须在源头上进行把控。         因此，业务链条自然延伸到了Token的全生命周期，包括聚合、分发和治理。         “Token治理”是当前很多客户关心的议题：投入了大量成本购买Token，效果究竟如何？回报体现在哪里？是否存在资源浪费？我们要做的，就是把每一环都管理起来——Token来自哪里，分发到了哪个环节，最终用在怎样的场景，以及场景匹配什么模型就足够。         AI时代改变企业人才需求          财闻：软通动力的AI业务当前处于什么发展阶段？何时能带动公司整体业绩反转？          霍宇：我认为目前还处于“战略跟进阶段”。当下AI领域的人才和算力仍然是稀缺资源，成本高企是处于AI转型中的企业普遍经历的境况。只要决心投入，现阶段的投入产出比普遍不够亮眼。但关键在于，要不要跟进？大部分企业的共识是：必须跟进。         AI不是一次简单的技术迭代，它可能带来整个产业的重新洗牌，真正进入“卖铲子”的阶段，才是真正开始赚钱的时候。         财闻：对比转型前后，软通动力为客户解决问题的场景和能力发生了什么变化？          霍宇：我们服务的场景并没有本质变化，但智能化程度截然不同。

p |          过去我们解决的是信息化的问题，随后进入数字化阶段，如今则迈向智能化阶段——核心目标要么是持续提升效率，要么是帮助客户获取更深层次的业务洞察。从宏观层面看，企业的核心刚需始终未变：提效、降本、打造生态链、拓展业务边界。         真正发生巨变的，是对人才的要求。以往到客户现场，面对的是确定性问题，解决方式在某种程度上是依靠增加人力数量来解决。         但现在随着AI的应用，客户会对单个业务员提出更多需要解决的需求。要回答这个问题，需要敏锐的洞察力、创造力、深厚的行业理解，还需要出色的沟通表达技巧和技术动手能力。国内企业想要FDE（Forward Deployment Engineer，前线部署工程师）人才——每个人都必须同时是行业顾问和技术专家。这是一个对综合素质要求极高的角色。         财闻：那这些复合型人才是外部招聘还是内部培养？          霍宇：两种方式都有，但目前人才市场热度极高，供给看似充足，实则良莠不齐，真正合适的人选并不好找。

q | 有过真实项目落地经验、具备实际体感的人，在各家企业都十分稀缺。         因此，我们主要还是从现有员工中选拔优秀骨干。那些真正带过项目的高级项目经理和高级架构师，本质上每个人都是FDE，只是目前还没有足够多的AI落地项目让他们反复锤炼。一旦有机会做成三四个项目，这种体感就会逐渐建立起来。         财闻：有一种说法是，AI时代软件行业会被颠覆性地替代，你怎么看？          霍宇：对此不必过度担忧。软件行业并不会消亡，只是进入门槛将显著降低。

r | AI目前最具价值的应用场景之一就是编程，未来的工作方式可能是：你只需将需求描述清楚，软件即可快速生成。         正如蒸汽机出现后，制造业非但没有消失，反而吸引了更多人涌入。我们很多员工现在也在使用AI辅助编程、撰写方案、进行分析，AI让软件开发变得前所未有的高效。我们当前所做的许多工作，本质上是在推动软件的标准化程度不断提升。         过去软件行业迟迟未能形成“工厂”概念，正是因为标准化水平不够。而这一轮AI浪潮，本质上正在让软件实现批量生成。

Current article:http://cmjy.shaihanhuangwoshang.cyou/991/fst3.html

Published on:06:42:55