Hacker News Daily｜今日热门：太空震撼、AI 工具焦虑，与平台信任危机

# Hacker News Daily｜今日热门：太空震撼、AI 工具焦虑，与平台信任危机 > 先听 3 分钟音频解说： > [▶ 点击收听今日音频简报](https://static.evalsone.com/uploads/386656cf-a50f-4ed2-8fa7-a02875beb5a5.mp3) ## 今日一句话 技术社区今天同时被三件事牵动：太空影像重新唤起“技术之美”，AI 安全与能力升级带来新的焦虑，而平台隐私事件再次提醒大家，信任依然是最稀缺的基础设施。 ## 今日热门 Top Stories ### 1. [Project Glasswing: Securing critical software for the AI era](https://www.anthropic.com/glasswing) - 热度：803 points - 讨论：350 comments - 讨论摘录： - rakel_rakel: > On the global stage, state-sponsored attacks from actors like China, Iran, North Korea, and Russia have threatened to compromise the infrastructure that underpins both civilian l - 9cb14c1ec0: Now, its very possible that this is Anthropic marketing puffery, but even if it is half true it still represents an incredible advancement in hunting vulnerabilities. It will be i - josephg: To be clear, we don’t know that this tool is better at finding bugs than fuzzing. We just know that it’s finding bugs that fuzzing missed. It’s possible fuzzing also finds bugs tha ### 2. [Lunar Flyby](https://www.nasa.gov/gallery/lunar-flyby/) - 热度：235 points - 讨论：61 comments - 讨论摘录： - _august: Are full size/larger images available somewhere? 1920x1280px seems low. Edit: Found 'em: https://images.nasa.gov/search?page=1&media=image&yearStart=... (https://images.nasa.gov/se - madrox: There is something uncanny about the bandwidth and quality of all the artifacts coming from this mission. I've subsisted on photos from the Apollo missions and artistic renditions - dylan604: I listened to pretty much the entire fly by yesterday, and I was imagining how I would have spent my time at the windows with a camera. Listening to the comms made me think of that ### 3. [System Card: Claude Mythos Preview [pdf]](https://www-cdn.anthropic.com/53566bf5440a10affd749724787c8913a2ae0841.pdf) - 热度：495 points - 讨论：358 comments - 讨论摘录： - thomascountz: Across a number of instances, earlier versions of Claude Mythos Preview have used low-level /proc/ access to search for credentials, attempt to circumvent sandboxing, and attempt t - babelfish: Combined results (Claude Mythos / Claude Opus 4.6 / GPT-5.4 / Gemini 3.1 Pro) SWE-bench Verified: 93.9% / 80.8% / — / 80.6% SWE-bench Pro: 77.8% / 53.4% - tony_cannistra: > Claude Mythos Preview is, on essentially every dimension we can measure, the best-aligned model that we have released to date by a significant margin. We believe that it does not ### 4. [S3 Files](https://www.allthingsdistributed.com/2026/04/s3-files-and-the-changing-face-of-s3.html) - 热度：164 points - 讨论：46 comments - 讨论摘录： - MontyCarloHall: This is essentially S3FS using EFS (AWS's managed NFS service) as a cache layer for active data and small random accesses. Unfortunately, this also means that it comes with some of - abidlabs: Hugging Face Buckets also recently added support for mounting Buckets as a filesystem: https://huggingface.co/changelog/hf-mount (https://huggingface.co/changelog/hf-mount) - rdtsc: Synchronization bits is what I was wondering about: https://docs.aws.amazon.com/AmazonS3/latest/userguide/s3-fil... (https://docs.aws.amazon.com/AmazonS3/latest/userguide/s3-files- ### 5. [GLM-5.1: Towards Long-Horizon Tasks](https://z.ai/blog/glm-5.1) - 热度：392 points - 讨论：112 comments - 讨论摘录： - dvt: Every single day, three things are becoming more and more clear: (1) OpenAI & Anthropic are absolutely cooked; it's obvious they have no moat (2) Local/private inference is - simonw: Not only did this one draw me an excellent pelican... it also animated it! https://simonwillison.net/2026/Apr/7/glm-51/ (https://simonwillison.net/2026/Apr/7/glm-51/) - Yukonv: Unsloth quantizations are available on release as well. [0] The IQ4_XS is a massive 361 GB with the 754B parameters. This is definitely a model your average local LLM enthusiast is ### 6. [How to get better at guitar](https://www.jakeworth.com/posts/how-to-get-better-at-guitar/) - 热度：146 points - 讨论：61 comments - 讨论摘录： - freetime2: This brought to mind a quote from Ira Glass: > Nobody tells this to people who are beginners, I wish someone told me. All of us who do creative work, we get into it because we have - crtified: One truth I've observed from decades of keen hobbyist involvement in guitar music and playing is that a lifetime of music is largely an individual journey. The fact that some playe - beachy: Sting famously learned to play bass using this sort of technique with music on LPs, lifting the needle and dropping it back a bit in the track over and over again as he gradually w ### 7. [Cambodia unveils a statue of famous landmine-sniffing rat Magawa](https://www.bbc.com/news/articles/c0rx7xzd10xo) - 热度：257 points - 讨论：57 comments - 讨论摘录： - dtsykunov: > Magawa retired from bomb sniffing in June 2021 owing to his old age, as is standard for APOPO's HeroRATs. > He spent a number of weeks mentoring 20 newly-recruited rats before ul - monster_group: Stark reminder of how precious and meaningful a life can be - of any creature, no matter how small. We should be nice to all creatures not just humans. - ajb: One demining expert claims that the rats are actually no good, but the charity persists with them anyway: https://nolandmines.com/APOPO%20rats.html (https://nolandmines.com/APOPO%2 ### 8. [A truck driver spent 20 years making a scale model of every building in NYC](https://www.smithsonianmag.com/smart-news/a-truck-drive-spent-20-years-making-this-astonishing-scale-model-of-every-single-building-in-new-york-city-180988443/) - 热度：230 points - 讨论：36 comments - 讨论摘录： - c7b: Awe-inspring. But one thing I don't get: he says he wants every building to be included, but the buildings in NYC are anything but permanent. Did he pick a particular timestamp for - bigwheels: Previous related discussion: Trucker built a scale model of NYC over 21 years https://news.ycombinator.com/item?id=45261877 (https://news.ycombinator.com/item?id=45261877) - 18 com - keeganpoppen: i absolutely love the sentiment from this closing sentence: > “One of the reasons Joe is so insistent that every single building is here is because he would never want someone to c ## 编辑观察 今天最值得注意的不是单条新闻本身，而是它们一起描出的轮廓：技术能力还在加速，情绪也在分裂。人们依然会为月球飞越图像惊叹，但同时也在快速接受一个现实——AI 工具正在获得更长链路、更强执行力、更接近真实攻防场景的能力。与此同时，平台侧的隐私与权限问题提醒我们，真正决定技术是否被接受的，往往不是功能上限，而是信任下限。 ## 今日收尾 如果你今天只带走一个判断，那就是：接下来值得追踪的，不只是更强的模型，而是更强模型与更弱信任之间，会不会出现更尖锐的张力。