削减 Token 预算而非裁员：科技公司的 AI 降本之道

今日内容一览：[00:26] SpaceX 计划于7月中旬发射星舰第13次试飞[01:53] 美国各地社区掀起反对人工智能数据中心浪潮[03:20] 通过技术优化削减 AI Token 预算以保住员工岗位---[00:04] Welcome to today's Tech Briefing, your daily source for the latest in technology and innovation. Today, we have three exciting stories. We will cover SpaceX's upcoming Starship Flight 13, the growing community resistance against AI data centers, and how companies can optimize their AI token budgets without laying off engineering talent. Let's get started.欢迎收听今天的科技简报，为您带来科技与创新领域的最新动态。今天我们有三个精彩的故事。我们将报道 SpaceX 即将进行的星舰第 13 次飞行、各地社区对 AI 数据中心日益增长的抵制运动，以及公司如何在不解雇工程人才的情况下优化其 AI token 预算。让我们开始吧。---[00:26] SpaceX Targets Mid-July for Starship Flight 13 Launch. SpaceX is preparing for the thirteenth flight test of its Starship system. The team is targeting a launch as early as Thursday, July 16. This launch comes only seven weeks after the previous flight in May. It shows that SpaceX is accelerating its launch cadence (发射频次；节奏；抑扬顿挫). The mission will use the new Starship and Super Heavy V3 vehicles. These are equipped with the advanced Raptor 3 engines. During the flight, Booster 20 will attempt a controlled splashdown in the Gulf of Mexico. Meanwhile, Ship 40 will follow a suborbital path and splash down in the Indian Ocean. A major milestone (里程碑) for this flight is the first deployment of twenty next-generation Starlink V3 satellites. These satellites feature advanced laser links for better communication. They also carry onboard cameras. Six of these cameras will capture live imagery of Starship's heat shield during the stressful reentry phase. To help the cameras, engineers painted several heat shield tiles white to serve as clear imaging targets. They are also testing upgraded tiles on the aft flaps. This flight builds directly on lessons from Flight 12. Hardware and software modifications on Booster 20 and Ship 40 aim to improve engine reliability and vehicle robustness (鲁棒性；强健性). Elon Musk hopes to scale Starship launches to high rates in the coming years. This rapid launch cadence is essential to establish orbital refueling for future Moon and Mars missions.SpaceX 计划于7月中旬发射星舰第13次试飞。SpaceX 正在准备其星舰系统的第十三次飞行测试。团队的目标是最早于 7 月 16 日星期四进行发射。这距离 5 月份的上一次飞行仅隔了七周，显示出 SpaceX 正在加快其发射节奏。本次任务将使用全新的星舰和超重型 V3 载具，它们都配备了先进的猛禽 3 号发动机。在飞行过程中，超重型助推器 20 将尝试在墨西哥湾进行受控溅落。与此同时，飞船 40 将沿着亚轨道轨道飞行，并在印度洋溅落。本次飞行的一个重要里程碑是首次部署 20 颗下一代星链 V3 卫星。这些卫星配备了先进的激光链路以实现更好的通信，它们还携带了星载摄像头。其中六个摄像头将在紧张的再入大气层阶段捕获星舰防热盾的实时图像。为了帮助摄像头拍摄，工程师将几个隔热瓦涂成白色以作为清晰的成像目标。他们还在后翼上测试升级版的隔热瓦。本次飞行直接吸取了第 12 次飞行的教训。助推器 20 和飞船 40 上的硬件和软件改进旨在提高发动机的可靠性和载具的鲁棒性。埃隆·马斯克希望在未来几年将星舰发射频率提高到很高的水平。这种快速的发射节奏对于为未来的月球和火星任务建立轨道加油至关重要。（来源：Teslarati）---[01:53] Communities Across United States Fight Sprawling AI Data Centers. Community opposition against AI data centers is growing fast across the United States. Sprawling AI data centers consume as much energy as entire states. Local residents are worried about rising energy bills, noise, light pollution, and water quality. Commercial energy demand will surpass residential demand for the first time this year because of these facilities. This rapid growth has put data centers under intense scrutiny (监视，仔细检查，审查). Between January and March, protesters blocked or delayed at least 75 projects valued at 130 billion dollars. In Wisconsin, a developer dropped plans for a 12-billion-dollar campus after fierce protests. Local governments are also taking action. Many states have passed new laws to restrict water use and remove tax breaks for these facilities. However, the current patchwork (拼凑物，缝补物) of local regulations is not enough to stop the buildout. At the federal level, lawmakers are debating new bills. Some bills want to pause data center construction, while others want tech giants to pay for their own energy infrastructure. Many communities feel they must fend for themselves against massive projects like Meta's 27-billion-dollar data center in Louisiana and Google's 10-billion-dollar project in Missouri. The battle between tech expansion and community protection is just beginning.美国各地社区掀起反对人工智能数据中心浪潮。在美国各地，针对人工智能数据中心的社区反对之声正在迅速蔓延。庞大的 AI 数据中心消耗的能量与整个州相当。当地居民非常担心电费上涨、噪音、光污染和水质问题。由于这些设施的存在，今年商业能源需求将首次超过居民能源需求。这种快速的增长使数据中心处于严密的社会监视之下。在今年 1 月至 3 月期间，抗议者阻碍或推迟了至少 75 个项目，价值高达 1300 亿美元。在威斯康星州，一家开发商在遭到激烈抗议后放弃了建设 120 亿美元数据中心园区的计划。地方政府也在采取行动。许多州已经通过了新法律，限制这些设施的用水并取消税收优惠。然而，目前拼凑而成的地方法规还不足以阻止这一建设浪潮。在联邦层面，议员们正在对新法案进行辩论。一些法案希望暂停数据中心的建设，而另一些法案则希望科技巨头自己支付能源基础设施的费用。面对路易斯安那州 Meta 270 亿美元的数据中心 and 密苏里州谷歌 100 亿美元的项目等庞大工程，许多社区觉得他们必须自谋出路。科技扩张与社区保护之间的斗争才刚刚开始。（来源：The Verge Tech & Culture）---[03:20] Engineering Solutions to Cut AI Token Budgets and Save Jobs. Many companies are facing a major trade-off. They are spending less money on people and more money on AI tokens. Large tech companies are laying off workers to finance their massive AI investments. However, research shows that cutting staff does not improve business returns. In fact, eighty percent of companies that laid off workers for AI saw no increase in profit. Uber learned this lesson the hard way. It gave AI coding tools to five thousand engineers and exhausted its entire annual token budget in just four months. The actual problem is that companies treat the token bill as fixed and the workforce as flexible. But a token budget can bend if engineers optimize it. The simplest fix is prompt caching. Major API providers now offer prompt caching, which cuts costs by up to ninety percent by not processing the same text repeatedly. For example, security firm ProjectDiscovery restructured its prompts and cut its LLM spending by seventy percent. Other smart strategies include routing tasks to smaller models, using batch processing, and applying prompt compression. Companies should use these measures to save money instead of firing employees. When a company engineers sixty percent off its token bill, it recovers the budget to hire and train young developers. Leaders must decide to invest in their people because human judgment is still essential for high-quality work.通过技术优化削减 AI Token 预算以保住员工岗位。许多公司正面临着一个重大折中。他们花在人身上的钱越来越少，而花在 AI token 上的钱越来越多。大型科技公司正在裁员，以为其庞大的 AI 投资提供资金。然而，研究表明裁员并不能提高业务回报。事实上，80% 为 AI 裁员的公司并没有看到利润增加。优步吸取了这个教训。它向 5,000 名工程师提供了 AI 编程工具，并在短短四个月内就耗尽了全年的 token 预算。实际的问题是，公司将 token 账单视为固定支出，而将劳动力视为可变成本。但如果工程师对其进行优化，token 预算是可以降低的。最简单的解决方法是提示词缓存。主要的 API 提供商现在都提供提示词缓存，它通过避免重复处理相同文本来降低高达 90% 的成本。例如，安全公司 ProjectDiscovery 重组了其提示词，并削减了 70% 的大模型开销。其他聪明的策略包括将任务分流到更小的模型、使用批处理以及应用提示词压缩。公司应该使用这些措施来省钱，而不是解雇员工。当一家公司通过工程优化削减了 60% 的 token 账单时，它就收回了预算来雇佣和培训年轻的开发人员。领导者必须决定投资于他们的人才，因为对于高质量的工作来说，人类的判断仍然是不可或缺的。（来源：Artificial Intelligence News）---[04:49] 📚 今日单词：cadence- /ˈkeɪdns/- n. 发射频次；节奏；抑扬顿挫[04:53] 词组/例句- launch cadence 发射频次；发射节奏- This rapid launch cadence is essential to establish orbital refueling for future Moon and Mars missions.这种快速的发射节奏对于为未来的月球和火星任务建立轨道加油至关重要。---[05:11] That wraps up today's Tech Briefing. Today, we looked at SpaceX's ambitious launch cadence for Starship Flight 13, the growing community resistance against massive AI data centers across the US, and how companies can optimize their token budgets to protect human talent. This brings us to today's discussion question: As AI expansion demands massive energy and competes with local communities, while companies cut payroll to finance it, how do you see the balance between rapid technological progress and community or human interests? Share your thoughts in the comments below. Thank you for listening to today's episode, and we will see you next time!今天的科技简报到这里就结束了。今天，我们关注了 SpaceX 对星舰第 13 次试飞的雄心勃勃的发射节奏，美国各地对大型 AI 数据中心日益增长的社区抵制，以及公司如何通过优化 token 预算来保护人类人才。这把我们带到了今天的讨论话题：随着 AI 的扩张对能源产生巨大需求并与地方社区发生冲突，同时公司通过裁剪员工来为其融资，您如何看待快速的技术进步与社区或人类利益之间的平衡？欢迎在下方评论区分享您的看法。感谢收听今天的节目，我们下期再见！前往小宇宙评论区与主播互动