Computing Life

10小时攻关，半小时Dev Time：在指甲盖大小的单片机上跑神经网络识别车库门

grapeot — Sat, 01 Aug 2026 23:00:00 -0700

为了解决车库门忘关的问题，我用拇指大小的ESP32-CAM单片机与AI协作做了一个端侧神经网络传感器。本文记录了如何通过打通全自动串口研发闭环、解决数据极度不对称与QAT量化衰减，在10小时内跑通嵌入式AI的完整过程。

10 Hours of AI Collaboration, 30 Mins of Dev Time: Running a Neural Network on a Thumb-Sized Microcontroller for Garage Door Recognition

grapeot — Sat, 01 Aug 2026 22:00:00 -0700

To solve the problem of accidentally leaving my garage door open overnight, I collaborated with AI to build an edge neural network sensor on a thumb-sized ESP32-CAM. This article details how we completed the end-to-end telemetry loop, tackled extreme data imbalance and QAT quantization loss, and got edge AI running on a microcontroller in just 10 hours.

使用AI暴力模拟月全食的绿松石带

grapeot — Mon, 22 Jun 2026 22:00:00 -0700

月食时月面边缘有一条青绿色的窄带，科普说那是臭氧吸收。但为什么是窄带不是整圈？为什么全食最深时反而看不见它？我们从最土的白圆盘开始，一层层加物理，硬算出这条带，在一路翻车里发现了一个更大的问题：AI 太懂物理，反而会把你带进前人留下的近似里。

Simulating the Lunar Eclipse Turquoise Band with Brute-Force AI

grapeot — Mon, 22 Jun 2026 22:00:00 -0700

During a lunar eclipse, a narrow green-blue band appears at the moon's edge; popular science says it is ozone absorption. But why a narrow band instead of a full ring? Why does it vanish at deepest totality? We start from the simplest white disk, layer in physics, and compute the band — and through repeated failures discover a bigger problem: AI knows physics too well, and will happily lead you into approximations left behind by earlier researchers.

使用AI十倍提效，成了模范老黄牛，就能加薪升职了？

grapeot — Sat, 20 Jun 2026 11:00:00 -0700

我用AI提效很成功，产出和rating都是org最高之一，但升职两次都失败了。后来发现一个讽刺的陷阱：正因为手快好用，老板把你当手而非脑，项目零散多变，反而讲不清一年的成果。最擅长用AI的人，反而最容易被AI替代。破解之道是主动设计奖赏系统，把省下的时间用来做判断而非交付更多。

AI 10x'd My Productivity. I Became a Model Workhorse. I Didn't Get Promoted.

grapeot — Sat, 20 Jun 2026 10:00:00 -0700

AI made me a top performer. I was denied promotion twice. Speed makes bosses treat you as a hand, not a brain. The best AI users paradoxically become the most replaceable. The fix: design the incentive structure.

把18亿颗星星画在一张图上，能还原我们拍到的银河吗？

grapeot — Sun, 14 Jun 2026 22:00:00 -0700

从最直白的"一星一像素"出发，八次翻车、六亿颗星，一步一步把银河从真实星表里逼出来。在这个过程中才发现，以前从来没认真想过头顶的星空为什么长这个样子。

Can 1.8 Billion Stars Recreate the Milky Way We Photograph?

grapeot — Sun, 14 Jun 2026 21:00:00 -0700

Starting from the simplest "one star, one pixel" rendering, eight failures and 600 million stars later, the Milky Way slowly emerged from a real star catalog.

用好AI的第二步：先写Skill再执行

grapeot — Tue, 26 May 2026 14:10:00 -0700

用好AI的第二步不是更会写 prompt，而是先外化、再复用。本文讲清 Skill 如何承载工作知识、好 Skill 的三要素，以及如何组织 Skill 文件夹让 Agent 自动找到。

Step Two to Using AI Well: Write the Skill Before You Execute

grapeot — Tue, 26 May 2026 13:10:00 -0700

Step two isn't better prompting. It's externalize first, reuse second. This post explains how Skills carry work knowledge, the three parts of a good Skill, and how to organize them so agents find the right one.