将巨大的厂房变成压迫的象征并非易事，但数据中心行业正在努力尝试。这些装满半智能硅片的巨大方盒子，正日益被视为“撒旦的沙堡”——它们是引发干旱、煮沸海洋的环境灾难。它们作为同样邪恶的人工智能的黑暗殿堂，其角色定位也未能改善公众的态度。数据中心的建设者与厌恶者之间，亟需建立一种基于坦诚的新契约。那种对担忧不屑一顾，或提供华而不实、毫无诚意的误导性说辞的做法，已不再奏效。以能源为例，如果没有可信的数据，就无法评估使用数据中心对环境的影响。让 Claude 写一首打油诗或处理一个数据库需要消耗多少能源？没人知道。目前存在着差异巨大的估算值，但没有哪一个足够可靠，能为个人选择或公共政策的伦理使用提供参考。我们不仅需要宏观层面的了解，还需要足够精确且具有时效性的数据，以区分不同模型和硬件代际之间的差异。唯一能提供此类高质量数据的来源是数据中心运营商。若没有这种合作，没人会相信数据中心的扩张决策背后，不是在不计代价地疯狂争夺吉瓦级的电力。考虑到马斯克的 xAI Colossus 集群等项目在美南地区的落地，这些项目在几乎无视监管许可的情况下安装了甲烷燃烧涡轮机，该行业很难指望获得公众的信任。相反，随之而来的是大量的诉讼。此外，该行业还倾向于拿未来的核能——甚至是核聚变——作为挡箭牌，仿佛这能为今天燃烧碳氢化合物的行为开脱。核能或许最终能分担部分负荷，但目前宣传的许多项目在未来几年内都无法提供电力。至于核聚变，其实现速度恐怕还赶不上计算尺的时代。这绝非解决当前扩张所带来后果的良策。
It's hard to make big sheds symbols of oppression, but the datacenter industry's having a jolly good crack at it. Its big boxes of semi-sentient sand are increasingly seen as satanic sandcastles, environmental disasters that inflict drought while boiling oceans. Their role as dark cathedrals of the equally demonic AIs has done nothing to soften attitudes. A new contract of candour is needed between those who build datacenters and those who despise them. Shrugging off concerns, or offering shiny misdirections with no intent of following through, no longer works. Take energy. Without numbers that can be trusted, it's impossible to judge the environmental effect of using datacenters. How much energy does it take to ask Claude to write a limerick or a database? Who knows. TThere are wildly differing estimates, but nothing reliable enough to inform personal choices or public policy on ethical use. We need to know this not just in broad terms, but with enough accuracy and currency to differentiate between models and hardware generations. The only source of data good enough for this purpose is the datacenter operator. Without that cooperation, nobody's going to believe that anything other than a mad dash for gigawatts at any cost is powering build-out decisions. Given projects such as Musk's xAI Colossus clusters in the Deep South, energized by methane-burning turbines installed with scant regard for regulatory permission, the industry can hardly expect the benefit of the doubt. Lots of lawsuits, though. Then there's the tendency to wave at future nuclear power – and even fusion – as though it excuses burning hydrocarbons today. Nuclear power may eventually shoulder some of the load, but many of the projects now being promoted will not deliver electricity for years. Fusion will be outperformed by slide rules for longer still. This is not an answer to the immediate consequences of the build-out. Meanwhile, we've had pious platitudes that AI is going to get so good it will design itself out of environmental damage. Seen much sign of that lately? Then there's the sector's on-off involvement with genuine innovation. Take the possibilities of marine datacenters powered by tidal and wave generators. The oceans offer abundant cooling and energy, with the tides driven by the gravitational interaction of the Earth, Moon, and Sun. Tidal power in particular is cosmically reliable; we know what the tides will be doing hundreds of years in advance. Seawater is not an ideal coolant, being corrosive and jellyfish-infested, but marine engineers have spent more than a century learning how to work around such problems. As for wave power, there's a lot going on below the radar, although don't wait to hear about it from the big guys. Panthalassa is building wave generators that look disturbingly and fittingly like the spaceship Discovery One from 2001: A Space Odyssey. You know, the home of the most famously insane AI in fiction. The 85-metre structure floats vertically, with a spherical section bobbing at the surface and a long tubular tail extending down through the water column. Wave motion forces internally recirculated water through a turbine, generating electricity for AI hardware inside a sealed, seawater-cooled container. It is designed to operate without anchors, subsea cables or a physical connection to land, avoiding some of the complexity of conventional offshore energy projects. As part of a floating datacenter, it's envisioned that connectivity would be via satellites. Given that orbital networks spend so much of their time over the open oceans populated by sea life notoriously reluctant to spend money on connectivity, this is no bad idea. Floating data farms could also be towed for maintenance or replacement rather than launched into orbit and abandoned. In every way, this is a saner idea than orbiting datacenters. An industry that saw itself as compatible with humans would be pushing harder at floating on water than floating in space. It isn't, because it isn't. Instead, it's operating on the edge of sanity, planning orgasmic overspend on throwing up as much infrastructure as it can as fast as it can, while uttering piteous "stop us before we kill again" pleadings as it struggles to deal with what it has become. This toxic mix of uncontrolled spending, buyer's remorse, misdirection, and HAL 9000 levels of non-alignment with keeping people alive is sharpening concern about environmental damage into activism. By the time the fossil fuel industry was firmly established in the public consciousness as a principal cause of climate change, it had a century of inertia behind it and was woven deeply into the global economy. AI is still groggily emerging from the egg, and it is vulnerable. While it's hard to turn misgivings into action against the global services and strategies of big AI, datacenters are local and vulnerable to local people and local politics. For anyone organizing opposition to AI amid dysfunctional national leadership, datacenters offer an obvious local target. For all the cack-handedness of the industry in perceiving this as a threat, let alone the actual responsibilities in force-feeding a trillion-dollar build-out into the ecosystem, some of us still want long-term, sustainable progress in more and better datacenters supporting more and better computational amplification of human potential. What's happening now is not the way to get there. Inasmuch as there's a datacenter culture – and there is; you could get everyone who makes the decisions that matter into a medium-sized Comic-Con – it needs to adopt radical openness. An open source datacenter is hard to imagine in an industry devoted to hiding its paranoia behind deception, but the same could be said of the mid-90s software sector. Among its very many other benefits, openness builds honesty, and honesty builds trust, and that more than any rare earth or process tweak is the element the AI industry needs most right now. By laying open the calculations and structure behind datacenter engineering, and being open about the decisions shaping the future, more and better options will appear. Participation makes allies; exclusion, enemies. On current evidence, the enemies are growing fast. Get open, get honest, or go home. ®
与此同时，我们听到了许多冠冕堂皇的陈词滥调，声称人工智能将发展到如此先进的程度，以至于能够通过自我设计来消除对环境的破坏。最近你看到这方面的迹象了吗？此外，该行业对于真正的创新一直处于一种断断续续的参与状态。以利用潮汐和波浪发电机供电的海上数据中心为例。海洋提供了充足的冷却资源和能源，而潮汐则是由地球、月球和太阳之间的引力相互作用驱动的。潮汐能尤其具有宇宙级的可靠性；我们甚至在几百年前就能准确预知潮汐的走向。海水虽然不是理想的冷却剂，因为它具有腐蚀性且常有水母出没，但海洋工程师已经花了一个多世纪的时间来学习如何解决这些问题。至于波浪能，虽然有很多进展在暗中进行，但不要指望从那些大公司口中听到。Panthalassa 公司正在建造的波浪发电机，其外观看起来既令人不安又恰如其分地酷似《2001太空漫游》中的“发现号”飞船。要知道，那可是虚构作品中最著名的疯狂人工智能的家园。这个 85 米长的结构垂直漂浮，球形部分在水面起伏，长长的管状尾部延伸至水柱下方。波浪运动迫使内部循环水通过涡轮机，为密封且由海水冷却的容器内的人工智能硬件发电。它的设计无需锚定、海底电缆或与陆地的物理连接，从而避免了传统海上能源项目的一些复杂性。作为浮动数据中心的一部分，其连接方式被设想为通过卫星实现。考虑到轨道网络大部分时间都运行在广阔的海洋上空，而海洋生物显然不愿为连接性付费，这倒不失为一个好主意。浮动数据农场还可以被拖走进行维护或更换，而不是像发射到轨道上那样被遗弃。从各方面来看，这都比轨道数据中心要理智得多。
It's hard to make big sheds symbols of oppression, but the datacenter industry's having a jolly good crack at it. Its big boxes of semi-sentient sand are increasingly seen as satanic sandcastles, environmental disasters that inflict drought while boiling oceans. Their role as dark cathedrals of the equally demonic AIs has done nothing to soften attitudes. A new contract of candour is needed between those who build datacenters and those who despise them. Shrugging off concerns, or offering shiny misdirections with no intent of following through, no longer works. Take energy. Without numbers that can be trusted, it's impossible to judge the environmental effect of using datacenters. How much energy does it take to ask Claude to write a limerick or a database? Who knows. TThere are wildly differing estimates, but nothing reliable enough to inform personal choices or public policy on ethical use. We need to know this not just in broad terms, but with enough accuracy and currency to differentiate between models and hardware generations. The only source of data good enough for this purpose is the datacenter operator. Without that cooperation, nobody's going to believe that anything other than a mad dash for gigawatts at any cost is powering build-out decisions. Given projects such as Musk's xAI Colossus clusters in the Deep South, energized by methane-burning turbines installed with scant regard for regulatory permission, the industry can hardly expect the benefit of the doubt. Lots of lawsuits, though. Then there's the tendency to wave at future nuclear power – and even fusion – as though it excuses burning hydrocarbons today. Nuclear power may eventually shoulder some of the load, but many of the projects now being promoted will not deliver electricity for years. Fusion will be outperformed by slide rules for longer still. This is not an answer to the immediate consequences of the build-out. Meanwhile, we've had pious platitudes that AI is going to get so good it will design itself out of environmental damage. Seen much sign of that lately? Then there's the sector's on-off involvement with genuine innovation. Take the possibilities of marine datacenters powered by tidal and wave generators. The oceans offer abundant cooling and energy, with the tides driven by the gravitational interaction of the Earth, Moon, and Sun. Tidal power in particular is cosmically reliable; we know what the tides will be doing hundreds of years in advance. Seawater is not an ideal coolant, being corrosive and jellyfish-infested, but marine engineers have spent more than a century learning how to work around such problems. As for wave power, there's a lot going on below the radar, although don't wait to hear about it from the big guys. Panthalassa is building wave generators that look disturbingly and fittingly like the spaceship Discovery One from 2001: A Space Odyssey. You know, the home of the most famously insane AI in fiction. The 85-metre structure floats vertically, with a spherical section bobbing at the surface and a long tubular tail extending down through the water column. Wave motion forces internally recirculated water through a turbine, generating electricity for AI hardware inside a sealed, seawater-cooled container. It is designed to operate without anchors, subsea cables or a physical connection to land, avoiding some of the complexity of conventional offshore energy projects. As part of a floating datacenter, it's envisioned that connectivity would be via satellites. Given that orbital networks spend so much of their time over the open oceans populated by sea life notoriously reluctant to spend money on connectivity, this is no bad idea. Floating data farms could also be towed for maintenance or replacement rather than launched into orbit and abandoned. In every way, this is a saner idea than orbiting datacenters. An industry that saw itself as compatible with humans would be pushing harder at floating on water than floating in space. It isn't, because it isn't. Instead, it's operating on the edge of sanity, planning orgasmic overspend on throwing up as much infrastructure as it can as fast as it can, while uttering piteous "stop us before we kill again" pleadings as it struggles to deal with what it has become. This toxic mix of uncontrolled spending, buyer's remorse, misdirection, and HAL 9000 levels of non-alignment with keeping people alive is sharpening concern about environmental damage into activism. By the time the fossil fuel industry was firmly established in the public consciousness as a principal cause of climate change, it had a century of inertia behind it and was woven deeply into the global economy. AI is still groggily emerging from the egg, and it is vulnerable. While it's hard to turn misgivings into action against the global services and strategies of big AI, datacenters are local and vulnerable to local people and local politics. For anyone organizing opposition to AI amid dysfunctional national leadership, datacenters offer an obvious local target. For all the cack-handedness of the industry in perceiving this as a threat, let alone the actual responsibilities in force-feeding a trillion-dollar build-out into the ecosystem, some of us still want long-term, sustainable progress in more and better datacenters supporting more and better computational amplification of human potential. What's happening now is not the way to get there. Inasmuch as there's a datacenter culture – and there is; you could get everyone who makes the decisions that matter into a medium-sized Comic-Con – it needs to adopt radical openness. An open source datacenter is hard to imagine in an industry devoted to hiding its paranoia behind deception, but the same could be said of the mid-90s software sector. Among its very many other benefits, openness builds honesty, and honesty builds trust, and that more than any rare earth or process tweak is the element the AI industry needs most right now. By laying open the calculations and structure behind datacenter engineering, and being open about the decisions shaping the future, more and better options will appear. Participation makes allies; exclusion, enemies. On current evidence, the enemies are growing fast. Get open, get honest, or go home. ®
一个自认为与人类兼容的行业，本应在水上漂浮技术上投入的精力远超在太空漂浮技术上的投入。但事实并非如此，因为它根本不兼容。相反，它正游走在理智的边缘，计划着疯狂的超额支出，以最快的速度尽可能多地堆砌基础设施，同时又发出可怜的“在我们再次犯错前阻止我们”的哀求，挣扎着应对自己已经变成的模样。这种由失控的支出、买家悔恨、误导以及与人类生存目标严重脱节（达到HAL 9000级别）所构成的有毒混合体，正将人们对环境破坏的担忧转化为积极的行动。当化石燃料行业在公众意识中被确立为气候变化的主要原因时，它已经拥有了一个世纪的惯性，并深深植根于全球经济之中。而人工智能才刚刚从蛋壳中蹒跚而出，它还很脆弱。虽然很难将疑虑转化为针对大型人工智能全球服务和战略的行动，但数据中心是本地化的，容易受到当地民众和地方政治的影响。对于任何在国家领导层失能的情况下组织反对人工智能的人来说，数据中心提供了一个显而易见的本地目标。尽管该行业在将此视为威胁方面表现得笨手笨脚，更不用说在向生态系统强行注入万亿美元建设规模方面所承担的实际责任，但我们中的一些人仍然希望看到长期、可持续的进步，即拥有更多更好的数据中心，以支持更多更好的计算能力来放大人类潜能。现在发生的一切并不是实现这一目标的途径。既然存在一种数据中心文化——确实存在，你甚至可以把所有做出关键决策的人都塞进一个中等规模的漫展里——它就需要采取彻底的开放态度。在一个致力于用欺骗来掩盖其偏执的行业中，开源数据中心很难想象，但90年代中期的软件行业也曾面临同样的质疑。
It's hard to make big sheds symbols of oppression, but the datacenter industry's having a jolly good crack at it. Its big boxes of semi-sentient sand are increasingly seen as satanic sandcastles, environmental disasters that inflict drought while boiling oceans. Their role as dark cathedrals of the equally demonic AIs has done nothing to soften attitudes. A new contract of candour is needed between those who build datacenters and those who despise them. Shrugging off concerns, or offering shiny misdirections with no intent of following through, no longer works. Take energy. Without numbers that can be trusted, it's impossible to judge the environmental effect of using datacenters. How much energy does it take to ask Claude to write a limerick or a database? Who knows. TThere are wildly differing estimates, but nothing reliable enough to inform personal choices or public policy on ethical use. We need to know this not just in broad terms, but with enough accuracy and currency to differentiate between models and hardware generations. The only source of data good enough for this purpose is the datacenter operator. Without that cooperation, nobody's going to believe that anything other than a mad dash for gigawatts at any cost is powering build-out decisions. Given projects such as Musk's xAI Colossus clusters in the Deep South, energized by methane-burning turbines installed with scant regard for regulatory permission, the industry can hardly expect the benefit of the doubt. Lots of lawsuits, though. Then there's the tendency to wave at future nuclear power – and even fusion – as though it excuses burning hydrocarbons today. Nuclear power may eventually shoulder some of the load, but many of the projects now being promoted will not deliver electricity for years. Fusion will be outperformed by slide rules for longer still. This is not an answer to the immediate consequences of the build-out. Meanwhile, we've had pious platitudes that AI is going to get so good it will design itself out of environmental damage. Seen much sign of that lately? Then there's the sector's on-off involvement with genuine innovation. Take the possibilities of marine datacenters powered by tidal and wave generators. The oceans offer abundant cooling and energy, with the tides driven by the gravitational interaction of the Earth, Moon, and Sun. Tidal power in particular is cosmically reliable; we know what the tides will be doing hundreds of years in advance. Seawater is not an ideal coolant, being corrosive and jellyfish-infested, but marine engineers have spent more than a century learning how to work around such problems. As for wave power, there's a lot going on below the radar, although don't wait to hear about it from the big guys. Panthalassa is building wave generators that look disturbingly and fittingly like the spaceship Discovery One from 2001: A Space Odyssey. You know, the home of the most famously insane AI in fiction. The 85-metre structure floats vertically, with a spherical section bobbing at the surface and a long tubular tail extending down through the water column. Wave motion forces internally recirculated water through a turbine, generating electricity for AI hardware inside a sealed, seawater-cooled container. It is designed to operate without anchors, subsea cables or a physical connection to land, avoiding some of the complexity of conventional offshore energy projects. As part of a floating datacenter, it's envisioned that connectivity would be via satellites. Given that orbital networks spend so much of their time over the open oceans populated by sea life notoriously reluctant to spend money on connectivity, this is no bad idea. Floating data farms could also be towed for maintenance or replacement rather than launched into orbit and abandoned. In every way, this is a saner idea than orbiting datacenters. An industry that saw itself as compatible with humans would be pushing harder at floating on water than floating in space. It isn't, because it isn't. Instead, it's operating on the edge of sanity, planning orgasmic overspend on throwing up as much infrastructure as it can as fast as it can, while uttering piteous "stop us before we kill again" pleadings as it struggles to deal with what it has become. This toxic mix of uncontrolled spending, buyer's remorse, misdirection, and HAL 9000 levels of non-alignment with keeping people alive is sharpening concern about environmental damage into activism. By the time the fossil fuel industry was firmly established in the public consciousness as a principal cause of climate change, it had a century of inertia behind it and was woven deeply into the global economy. AI is still groggily emerging from the egg, and it is vulnerable. While it's hard to turn misgivings into action against the global services and strategies of big AI, datacenters are local and vulnerable to local people and local politics. For anyone organizing opposition to AI amid dysfunctional national leadership, datacenters offer an obvious local target. For all the cack-handedness of the industry in perceiving this as a threat, let alone the actual responsibilities in force-feeding a trillion-dollar build-out into the ecosystem, some of us still want long-term, sustainable progress in more and better datacenters supporting more and better computational amplification of human potential. What's happening now is not the way to get there. Inasmuch as there's a datacenter culture – and there is; you could get everyone who makes the decisions that matter into a medium-sized Comic-Con – it needs to adopt radical openness. An open source datacenter is hard to imagine in an industry devoted to hiding its paranoia behind deception, but the same could be said of the mid-90s software sector. Among its very many other benefits, openness builds honesty, and honesty builds trust, and that more than any rare earth or process tweak is the element the AI industry needs most right now. By laying open the calculations and structure behind datacenter engineering, and being open about the decisions shaping the future, more and better options will appear. Participation makes allies; exclusion, enemies. On current evidence, the enemies are growing fast. Get open, get honest, or go home. ®
除了诸多其他益处外，开放能建立诚实，而诚实能建立信任。比起任何稀土材料或工艺调整，这才是目前人工智能行业最需要的要素。通过公开数据中心工程背后的计算与结构，并坦诚面对塑造未来的决策，更多更好的选择将会涌现。参与能结交盟友，而排斥只会制造敌人。从目前的迹象来看，敌人正在迅速增加。要么开放，要么诚实，否则就趁早出局。
It's hard to make big sheds symbols of oppression, but the datacenter industry's having a jolly good crack at it. Its big boxes of semi-sentient sand are increasingly seen as satanic sandcastles, environmental disasters that inflict drought while boiling oceans. Their role as dark cathedrals of the equally demonic AIs has done nothing to soften attitudes. A new contract of candour is needed between those who build datacenters and those who despise them. Shrugging off concerns, or offering shiny misdirections with no intent of following through, no longer works. Take energy. Without numbers that can be trusted, it's impossible to judge the environmental effect of using datacenters. How much energy does it take to ask Claude to write a limerick or a database? Who knows. TThere are wildly differing estimates, but nothing reliable enough to inform personal choices or public policy on ethical use. We need to know this not just in broad terms, but with enough accuracy and currency to differentiate between models and hardware generations. The only source of data good enough for this purpose is the datacenter operator. Without that cooperation, nobody's going to believe that anything other than a mad dash for gigawatts at any cost is powering build-out decisions. Given projects such as Musk's xAI Colossus clusters in the Deep South, energized by methane-burning turbines installed with scant regard for regulatory permission, the industry can hardly expect the benefit of the doubt. Lots of lawsuits, though. Then there's the tendency to wave at future nuclear power – and even fusion – as though it excuses burning hydrocarbons today. Nuclear power may eventually shoulder some of the load, but many of the projects now being promoted will not deliver electricity for years. Fusion will be outperformed by slide rules for longer still. This is not an answer to the immediate consequences of the build-out. Meanwhile, we've had pious platitudes that AI is going to get so good it will design itself out of environmental damage. Seen much sign of that lately? Then there's the sector's on-off involvement with genuine innovation. Take the possibilities of marine datacenters powered by tidal and wave generators. The oceans offer abundant cooling and energy, with the tides driven by the gravitational interaction of the Earth, Moon, and Sun. Tidal power in particular is cosmically reliable; we know what the tides will be doing hundreds of years in advance. Seawater is not an ideal coolant, being corrosive and jellyfish-infested, but marine engineers have spent more than a century learning how to work around such problems. As for wave power, there's a lot going on below the radar, although don't wait to hear about it from the big guys. Panthalassa is building wave generators that look disturbingly and fittingly like the spaceship Discovery One from 2001: A Space Odyssey. You know, the home of the most famously insane AI in fiction. The 85-metre structure floats vertically, with a spherical section bobbing at the surface and a long tubular tail extending down through the water column. Wave motion forces internally recirculated water through a turbine, generating electricity for AI hardware inside a sealed, seawater-cooled container. It is designed to operate without anchors, subsea cables or a physical connection to land, avoiding some of the complexity of conventional offshore energy projects. As part of a floating datacenter, it's envisioned that connectivity would be via satellites. Given that orbital networks spend so much of their time over the open oceans populated by sea life notoriously reluctant to spend money on connectivity, this is no bad idea. Floating data farms could also be towed for maintenance or replacement rather than launched into orbit and abandoned. In every way, this is a saner idea than orbiting datacenters. An industry that saw itself as compatible with humans would be pushing harder at floating on water than floating in space. It isn't, because it isn't. Instead, it's operating on the edge of sanity, planning orgasmic overspend on throwing up as much infrastructure as it can as fast as it can, while uttering piteous "stop us before we kill again" pleadings as it struggles to deal with what it has become. This toxic mix of uncontrolled spending, buyer's remorse, misdirection, and HAL 9000 levels of non-alignment with keeping people alive is sharpening concern about environmental damage into activism. By the time the fossil fuel industry was firmly established in the public consciousness as a principal cause of climate change, it had a century of inertia behind it and was woven deeply into the global economy. AI is still groggily emerging from the egg, and it is vulnerable. While it's hard to turn misgivings into action against the global services and strategies of big AI, datacenters are local and vulnerable to local people and local politics. For anyone organizing opposition to AI amid dysfunctional national leadership, datacenters offer an obvious local target. For all the cack-handedness of the industry in perceiving this as a threat, let alone the actual responsibilities in force-feeding a trillion-dollar build-out into the ecosystem, some of us still want long-term, sustainable progress in more and better datacenters supporting more and better computational amplification of human potential. What's happening now is not the way to get there. Inasmuch as there's a datacenter culture – and there is; you could get everyone who makes the decisions that matter into a medium-sized Comic-Con – it needs to adopt radical openness. An open source datacenter is hard to imagine in an industry devoted to hiding its paranoia behind deception, but the same could be said of the mid-90s software sector. Among its very many other benefits, openness builds honesty, and honesty builds trust, and that more than any rare earth or process tweak is the element the AI industry needs most right now. By laying open the calculations and structure behind datacenter engineering, and being open about the decisions shaping the future, more and better options will appear. Participation makes allies; exclusion, enemies. On current evidence, the enemies are growing fast. Get open, get honest, or go home. ®