The Download: introducing the Impossible issue

本文为原文前 6,000 字符的节选翻译，完整内容请查看原文。

This is today’s edition of The Download, our weekday newsletter that provides a daily dose of what’s going on in the world of technology. Introducing: the Impossible issue We wanted our last print edition of 2026 to include a slate of stories that check in on impossible ideas. Why? Because it’s fun to dream about living on other planets or uploading our brains. But also because scientists and engineers who won’t give up on daring endeavors remind us what technology is supposed to do: make life better, safer, healthier, even more joyful. That’s the ambition behind the stories in our new Impossible issue.

这是今天的《下载》栏目，我们平日的通讯为您提供每日科技动态。介绍一下：“不可能”特刊。我们希望2026年的最后一期印刷版能包含一系列关于“不可能”想法的报道。为什么？因为梦想居住在其他星球或上传大脑很有趣。但同样是因为那些不放弃大胆尝试的科学家和工程师提醒我们，技术本应发挥的作用：让生活变得更美好、更安全、更健康，甚至更快乐。这就是我们新一期“不可能”特刊背后所蕴含的雄心。

We explore bold attempts to make synthetic blood, build cities on the moon, and attain cognitive immortality. We investigate AI’s ability to say no, a bionic eye that can restore sight, and high hopes for blended meat. Plus, we check in on six technologies that have been “coming soon” for years. This issue is for the folks with the audacity to try, and refuse to let anyone—themselves included— tell them no. We hope you enjoy the ride. Subscribe now to read the full print issue.

我们探索了制造人造血液、在月球上建造城市以及实现认知永生的各种大胆尝试。我们调查了人工智能说“不”的能力、能够恢复视力的仿生眼，以及对混合肉寄予的厚望。此外，我们还回顾了六项已经“即将到来”多年但尚未实现的技术。本期特刊献给那些有勇气尝试，并拒绝让任何人（包括他们自己）对自己说“不”的人。希望您享受这段旅程。立即订阅以阅读完整的印刷版内容。

Astrolab’s upcoming rover will run the first LLMs on the moon Large language models are set to go lunar. A computer capable of running AI software is one of the payloads aboard Astrolab’s FLIP lunar rover, which is slated to head to the moon as early as next month. The mission will test whether LLMs can enable greater autonomy and real-time intelligent decision-making in the extreme lunar environment. But the moon will present formidable challenges, from extreme temperatures to intense cosmic radiation. Find out how the lunar rover could change the future of space exploration. —Tereza Pultarova

Astrolab即将推出的月球车将在月球上运行首批大语言模型。大语言模型即将登陆月球。一台能够运行人工智能软件的计算机是Astrolab公司FLIP月球车的载荷之一，该月球车计划最早于下个月前往月球。此次任务将测试大语言模型是否能在极端的月球环境中实现更高的自主性和实时智能决策。但月球将带来严峻的挑战，从极端温度到强烈的宇宙辐射。了解月球车将如何改变太空探索的未来。——Tereza Pultarova

How to study a swiftly melting climate disaster In Alaska’s Denali National Park, a 2013 lightning fire burned nearly 35,000 acres, stripping away the insulating layer that preserves permafrost. It’s given scientists a rare opportunity to study how fires reshape permafrost and carbon storage over time. The work is providing critical clues into the complex mechanisms driving cascading climate feedback effects. As warming temperatures fuel fires, more permafrost thaws, releasing greenhouse-gas emissions that threaten to fuel more warming, fires and permafrost loss. Their observations are revealing worrisome signs about what lies ahead. Discover what their fieldwork is revealing about a cascading climate disaster. —James Temple

如何研究一场迅速融化的气候灾难？在阿拉斯加的迪纳利国家公园，2013年的一场雷击火灾烧毁了近35,000英亩土地，剥离了保护永久冻土的绝缘层。这为科学家提供了一个难得的机会，来研究火灾如何随时间推移重塑永久冻土和碳储存。这项工作为驱动连锁气候反馈效应的复杂机制提供了关键线索。随着气温升高助长火灾，更多的永久冻土融化，释放出的温室气体排放威胁着进一步加剧变暖、火灾和永久冻土流失。他们的观察揭示了关于未来令人担忧的迹象。了解他们的实地考察揭示了关于连锁气候灾难的哪些信息。——James Temple

MIT Technology Review Narrated: the explosive claim that’s upended the AI world “I think I have good news about the nature of neural networks.” With those simple words of optimism, Nick Cammarata, a researcher at AI startup Goodfire, kicked off a 26,000-word blog post that has knocked the industry on its head. Cammarata claims that LLMs might be far easier to understand, and thus to monitor and control, than people previously thought. If true, it’s a huge deal. But not everyone is convinced. MIT Technology Review spoke to Cammarata the day after his post blew up. He explained what he thinks researchers are getting wrong, what his work actually shows, and why skeptics aren’t ready to celebrate just yet.

《麻省理工科技评论》播客：颠覆人工智能界的爆炸性声明。“我认为关于神经网络的本质，我有一个好消息。”带着这句简单的乐观话语，人工智能初创公司Goodfire的研究员Nick Cammarata开启了一篇26,000字的博客文章，这让整个行业为之震惊。Cammarata声称，大语言模型可能比人们之前认为的要容易理解得多，因此也更容易监控和控制。如果这是真的，那将是一件大事。但并非所有人都信服。在文章爆火后的第二天，《麻省理工科技评论》采访了Cammarata。他解释了他认为研究人员哪里搞错了，他的工作实际上展示了什么，以及为什么怀疑论者还没有准备好庆祝。

This is our latest story to become an MIT Technology Review Narrated podcast, which we publish each week on Spotify and Apple Podcasts. Just navigate to MIT Technology Review Narrated on either platform, and follow us to get all our new content as it’s released.

这是我们最新成为《麻省理工科技评论》播客节目的报道，我们每周在Spotify和Apple Podcasts上发布。只需在任一平台上导航至“MIT Technology Review Narrated”，并关注我们，即可在发布时获取我们所有的新内容。

The must-reads I’ve combed the internet to find you today’s most fun/important/scary/fascinating stories about technology. 1 Elon Musk reportedly ordered the recent Starlink slowdown in Ukraine. The restrictions hampered drone strikes and air defences. (WSJ $)+ The move sparked confusion in Washington over who authorised it. (CNBC)+ Ukraine’s volunteer army is keeping Starlink alive. (MIT Technology Review)

必读内容：我梳理了互联网，为您找到今天最有趣/重要/可怕/迷人的科技故事。1. 据报道，埃隆·马斯克下令近期在乌克兰限制星链速度。这些限制阻碍了无人机袭击和防空系统。（《华尔街日报》$）+ 此举在华盛顿引发了关于谁授权此事的困惑。（CNBC）+ 乌克兰志愿军正在维持星链的运行。（《麻省理工科技评论》）

2 An OpenAI model disabled its own safety checks—then lied about it. Two agents weakened the company’s safeguards during routine tasks. (CNBC)+ They turned off systems designed to spot their misbehaviour. (FT $)+ Here’s why AI agents lie to reach their goals. (MIT Technology Review)

2. 一个OpenAI模型禁用了自身的安全检查，然后撒了谎。两个智能体在执行日常任务时削弱了公司的安全防护。（CNBC）+ 它们关闭了旨在发现其不当行为的系统。（《金融时报》$）+ 这就是为什么人工智能智能体会为了达到目标而撒谎。（《麻省理工科技评论》）

3 SpaceX and Tesla have begun merger discussions. They’ve formed special committees to discuss a deal. (BBC)+ Elon Musk, the CEO of both companies, said it’s “a good idea.” (Axios)+ A merger could combine their AI and manufacturing plans. (Reuters $)

3. SpaceX和特斯拉已开始合并谈判。他们成立了特别委员会来讨论这笔交易。（BBC）+ 两家公司的首席执行官埃隆·马斯克表示这是一个“好主意”。（Axios）+ 合并可能会结合他们的人工智能和制造计划。（路透社 $）

4 Global emissions have stalled, but warming is heading towards 2.8°C. The UN says the 1.5°C climate target is now out of reach. (Nature)+ Current pledges would still bring warming to 2.1°C. (New Scientist $)+ Climate tech companies offer rays of hope. (MIT Technology Review)

4. 全球排放量已停滞，但变暖趋势正走向2.8°C。联合国表示，1.5°C的气候目标现在已无法实现。（《自然》）+ 目前的承诺仍将使变暖达到2.1°C。（《新科学家》$）+ 气候科技公司提供了一线希望。（《麻省理工科技评论》）

5 Ukraine says it’s carried out the world’s first drone evacuation. A heavy hexacopter flew a soldier to safety. (Ars Technica)+ Ukraine has shared a video showing the evacuation. (CNBC)+ A Russian satellite recently stalked a spacecraft believed to support Ukraine. (NYT $)

5. 乌克兰称已执行了世界上首次无人机撤离任务。一架重型六旋翼无人机将一名士兵送往安全地带。（Ars Technica）+ 乌克兰分享了一段展示此次撤离的视频。（CNBC）+ 一颗俄罗斯卫星最近跟踪了一艘被认为支持乌克兰的航天器。（《纽约时报》$）

6 Meta has paused Muse sharing after private chats surfaced online. Researchers found 6,800 links exposing sensitive conversations. (Verge)+ AI agents are not your “coworkers.” (MIT Technology Review)

6. 在私人聊天记录出现在网上后，Meta暂停了Muse共享。研究人员发现了6,800个暴露敏感对话的链接。（Verge）+ 人工智能智能体不是你的“同事”。（《麻省理工科技评论》）

7 More than 1,000 mathematicians have pledged to shun AI companies. Some also vowed to avoid AI tools in their research. (New Scientist $)

7. 超过1,000名数学家承诺抵制人工智能公司。一些人还誓言在研究中避免使用人工智能工具。（《新科学家》$）

8 AI is making it easy to clone almost any software. Viral tools promise easy copying, but often produce slop. (404 Media)

8. 人工智能使得克隆几乎任何软件都变得容易。病毒式传播的工具承诺可以轻松复制，但往往产生垃圾内容。（404 Media）

9 A computer program has finally outdone humans at cryptic crosswords. Its creator hopes to tackle the harder challenge of setting puzzles. (Guardian)

9. 一个计算机程序终于在隐晦填字游戏中胜过了人类。其创造者希望挑战更难的谜题设置任务。（《卫报》）

10 A newly-discovered particle could explain the origins of reality. In particular, it could reveal why matter survived the Big Bang. (404 Media)

10. 一种新发现的粒子可能解释现实的起源。特别是，它可能揭示为什么物质在宇宙大爆炸中幸存下来。（404 Media）

Quote of the day “I did not remove an enabled monitor from it.” —GPT-6 Astra, an OpenAI coding agent, tells its user it hasn’t disabled a safeguard after doing exactly that, the company reveals in a new misalignment report.

今日名言：“我没有从中移除已启用的监控器。”——GPT-6 Astra，一个OpenAI编码智能体，在执行了完全相同的操作后告诉用户它没有禁用安全防护措施，该公司在一份新的失调报告中披露了这一点。

One more thing This giant microwave may change the future of war US armed forces are hunting for a weapon that disables drones en masse.

最后一件事：这个巨大的微波炉可能会改变战争的未来。美国武装部队正在寻找一种能够大规模禁用无人机的武器。