躲在世界的外面，愉快地张望，
Joined February 2009
- Tweets47.9K
- Following240
- Followers14K
- Likes7.2K
受不了ChatGPT了，重写了system prompt
gist.github.com/goldengrape/…
goldengrape retweeted
SolidCraft 0.19.0 is out
solidcraft.io
- New Compact layout and a redesigned Assistant
- 14 new themes, including Tokyo Night (pictured), Nord, Catppuccin Mocha, Solarized Light and One Dark
- Free orbit: spin right over the top, great with a 3D mouse
- View cube glides are back, orbiting stays instant
- Chamfer, sketch, parameter and import fixes
Note: designs saved in 0.19.0 won't open in older versions
goldengrape retweeted
Social media addiction is real. Here’s how to get sober.
sciam.com/article/were-addic…
用 Kaldi 音素识别验证 TTS 发音
quaily.com/goldengrape/p/usi…
goldengrape retweeted
注意一些罕见字罕用法，即使不是多音字也是极有可能读错的，特别是对Gemini TTS/flash，比如穴位名称，可以用“攒竹穴”来测试，
对于多音字的正确率，gemini tts flash比gemini tts差一些。
edge-tts给定单读音文字，几乎不会读错（我没遇到过），可以作为回退
goldengrape retweeted
这道多音字tts题我做过，是这样解：
1. 让AI标记出文稿易读错的字词，甚至是所有多音字。
2. 让AI替换成常见同音字
3. 用TTS生成语音，并切割出标记位置前后2-3秒
4. 对标记段落人工听音审核
5. 如果都读对了，通过。如果有没读对的，回退至edge-tts，生成语音并拼接。
以上过程丢给个agent做就行
做空：从金融交易到社会监督——利益、真相与道德边界
quaily.com/goldengrape/p/sho…
notebook与音频概览：notebook.google.com/notebook…
goldengrape retweeted
実はペットボトルのネジって種類がいくつもあるんですよ。
そんな沢山ある種類の図面を閲覧できるサービスがあったりします。
手元にあるセブンのマスカット＆ピーチのボトルは、φ28のPCO 1881あたりのねじを使ってるっぽい気がしたり、しなかったり……😎
isbt.com/resources/isbt-thre…
goldengrape retweeted
Got the best results with Opus 5.5 so far.
ScanRuler.com includes a full parametric CAD workspace for that!
goldengrape retweeted
My latest free tool, ScanRuler.com, can automatically reverse-engineer your 3D Scans! Just tell your AI agent to connect to the ScanRuler MCP give it an stl and tell it to reverse engineer it. A detailed video and a custom skill will be released in the next few days!
goldengrape retweeted
硬件入门一直被许多人视为门槛极高的一道坎：元器件繁杂、接线容易短路、软硬件协同调试困难……但如果把复杂的电路彻底拆解成可交互的“积木”，会是怎样的体验？
开发者 Dilum Sanjaya 基于 Claude Opus 构建了一款极其惊艳的交互式学习应用：Learn Arduino，主打“一次搞懂一个电路”，把原本枯燥抽象的硬件开发变成了所见即所得的极简探索。
这款工具的亮点非常直接：
零负担渐进式路径：摒弃传统教程漫长的理论轰炸，从最基础的认识板卡（Meet the Board）、点亮一颗 LED（Blink an LED），到按键检测、电位器调压与多色混合，循序渐进拆解每个动作。
高质感 3D 可视化电路：UI 极其克制优雅，没有任何冗余元素。开发板、面包板与元器件全部以拟真且清爽的视角呈现，接线逻辑一目了然，彻底消除新手“插错线怕烧板子”的心理恐惧。
即时代码与硬件联动：右侧内置代码编辑窗口（Try It Yourself），修改参数即可实时在模拟电路中查看对应状态，让 pinMode 和 digitalWrite 的底层行为变成肉眼可见的物理反馈。
无论是想要拓展嵌入式领域的纯软件工程师、正在寻找生动教学素材的 STEAM 教师，还是打算带孩子做硬件启蒙的家长，这种把复杂工业概念重构为直观体验的产品都极具启发性。
好的工具从来不是替你偷懒，而是用直观的设计消解前期的认知负荷，把枯燥的摸索转化为纯粹的创造乐趣。
体验地址：learn-arduino-red.vercel.app…
goldengrape retweeted
To train user models / personal agents that best understand us, it helps to have access to all of our scattered digital footprint. For my first release in my PhD, I’m open-sourcing **Homecoming**, a desktop app that automatically aggregates all your personal context across all applications that you’ve accessed over your lifetime. It helps auto-request your data packages from over 2000 supported websites, and manages them locally in one place you can query, explore, and use for any downstream application. All your messages, emails, tiktok scrolling activity, goodreads, netflix, strava… even your 6th grade Wattpad account. You can even explore the features companies use to target ads at you.
This project was super fun to work on as I realized there is so much data about me that I didn’t even know existed. It’s a nice feeling to gain sovereignty over my data.
All data stays local, unless you choose otherwise. Lmk how it is!
Blog: benshi34.github.io/#/blog/ho…
Link to download: github.com/benshi34/homecomi…
goldengrape retweeted
Now you can co-create Slide Decks & Video Overviews to make sure you're getting *exactly* what you want. Here's how it works:
—Select a suggested focus topic from your sources (or type your own)
—Select the format, length, language, and visual style
—Review and edit the slide outline or video producer notes
—Generate, voila!
Available to all Pro and Ultra subscribers, try it out and let us know what you think!
goldengrape retweeted
The algorithms that run your feeds on social media are like “heat-seeking missiles that identify psychological vulnerability,” one psychologist told SciAm associate editor Allison Parshall.
sciam.com/article/our-love-h…
goldengrape retweeted
SolidCraft 0.17 is out ⚡
solidcraft.io
Much faster for big designs:
- Reopening a saved design takes seconds, not minutes
- Large STEP assemblies import in about a minute
- No more "Not responding" while opening or saving
- Moving parts in big assemblies is near instant
Also new:
- Loft guide rails and two-sided extrudes through all
- Mouse gestures: a ring of tools on the right button
- Danish, Vietnamese & Malay translations
- 78 bug fixes across modelling, STEP & sheet metal
232 changes from 61 contributors. Thank you!
goldengrape retweeted
概率论和统计学是生活中非常有用的学科。
举个例子，假设尊界一共卖出 100 万台车，我们随机不放回抽样三台，刹车在类似的制动情况下都被踩断。那尊界车（该测试条件下）总体的故障率点估计值是 100%，双侧 95% 置信区间为 29.24%–100%。
严格一点的统计学解释是：我们使用了一种在重复抽样中至少有 95% 覆盖率的方法，得到总体故障率的估计区间为 29.24%–100%。
其实这个问题的核心，远不是尊界卖出多少车。用 ChatGPT 做了个工具（见附图），比较尊界销量 1 千到 1 百万之间的故障率双侧 95% 置信区间变化情况。我们可以发现尊界即使仅售出 1 千台车，随机不放回抽样三台车刹车都断裂，统计学推断的总体故障率双侧 95% 置信区间下限也有 29.40%，接近 1 百万台车销量时的结果。
尊界的问题关键在于，三台车刹车都在类似的制动情况下断裂。即使真实总体故障率仅 1%（很高了），随机抽样三台车都故障的概率才大概 0.0001%。
当然，这里简化了一些问题，比如抽样方法是否真随机，比如后续生产的车工艺是否变化，比如测试方法是否可靠……现在有了 AI，我们自己就可以测试不同的条件。
所以，上述置信区间不能直接当作尊界实际总体故障率的估计。但至少我们清楚：如果一种故障非常罕见，那么随机抽取三辆车全部出现同类故障，是概率极低的事件。
如果尊界车的故障率真的在可接受水平，我们现在看到的现象，用上海话来说就是“眼眼叫碰着眼眼叫”。
