Personal reflections on artificial intelligencehttps://yli769227-jpg.github.io/https://yli769227-jpg.github.io/blog/2026-04-16-hello-ai/https://yli769227-jpg.github.io/blog/2026-04-16-hello-ai/The starting point of this blog — documenting my observations, reflections, and experiments with artificial intelligence.Thu, 16 Apr 2026 00:00:00 GMThttps://yli769227-jpg.github.io/blog/2026-04-18-taste-in-ai-era/https://yli769227-jpg.github.io/blog/2026-04-18-taste-in-ai-era/AI 迭代这么快，对话会失效，prompt 会过时，但有一样东西不会贬值：Taste。Sat, 18 Apr 2026 00:00:00 GMThttps://yli769227-jpg.github.io/blog/2026-04-19-china-ai-catch-up-illusion/https://yli769227-jpg.github.io/blog/2026-04-19-china-ai-catch-up-illusion/追上别人定的指标，你依然是跟随者。蒸馏是一条安全的路，但安全的路从来没有人引领过世界。Sun, 19 Apr 2026 00:00:00 GMThttps://yli769227-jpg.github.io/blog/2026-04-19-china-ai-patience/https://yli769227-jpg.github.io/blog/2026-04-19-china-ai-patience/DeepSeek 证明了钱不是瓶颈。真正卡住我们的是：允许五年没产出的耐心。Sun, 19 Apr 2026 00:00:00 GMThttps://yli769227-jpg.github.io/blog/2026-04-19-exception-management-ai-agent/https://yli769227-jpg.github.io/blog/2026-04-19-exception-management-ai-agent/我一开始想让 AI 替我回消息、跑完项目。但真正的问题不是"要不要全自动",是"哪些环节必须我在场"。Sun, 19 Apr 2026 00:00:00 GMThttps://yli769227-jpg.github.io/blog/2026-04-19-high-efficiency-median/https://yli769227-jpg.github.io/blog/2026-04-19-high-efficiency-median/AI 让效率提升了 30%,但没让我早下班。更隐蔽的代价是:我正在被拉向中位数,而且失去了意识到这件事的能力。Sun, 19 Apr 2026 00:00:00 GMThttps://yli769227-jpg.github.io/blog/2026-04-21-15-year-old-asks-ai-anxiety/https://yli769227-jpg.github.io/blog/2026-04-21-15-year-old-asks-ai-anxiety/他问出来了,其他人没问。而 AI 不吃体面这一套——越端着,越落后。Tue, 21 Apr 2026 00:00:00 GMThttps://yli769227-jpg.github.io/blog/2026-04-21-from-accept-edits-to-bypass-five-walls/https://yli769227-jpg.github.io/blog/2026-04-21-from-accept-edits-to-bypass-five-walls/升到 Bypass 不是勇气问题,是工程进度问题。信任不是一个按钮,是一个架构。Tue, 21 Apr 2026 00:00:00 GMThttps://yli769227-jpg.github.io/blog/2026-04-21-my-max-quota-never-runs-out/https://yli769227-jpg.github.io/blog/2026-04-21-my-max-quota-never-runs-out/额度用不完不是任务少,是你还没学会带着底线让渡。密度 = 让渡 × 底线,任何一边为零,乘积都是零。Tue, 21 Apr 2026 00:00:00 GMThttps://yli769227-jpg.github.io/blog/2026-04-22-april-frontend-war-three-giants/https://yli769227-jpg.github.io/blog/2026-04-22-april-frontend-war-three-giants/前三家拆协作链条,Kimi K2.6 开源直接拆掉付费墙。前端不再是岗位,是 prompt 能力——现在这个能力的门票也白菜化了。Wed, 22 Apr 2026 00:00:00 GMThttps://yli769227-jpg.github.io/blog/2026-04-24-beyond-catching-opus/https://yli769227-jpg.github.io/blog/2026-04-24-beyond-catching-opus/OpenAI 一周内连发 GPT-5.5 + GPT Image 2,Anthropic 上周刚推 Opus 4.7。模型分数只是入场券,Claude Code 和 Codex 这样的工作台才是真正的护城河——只卷模型,永远是别人房子的建材供应商。Fri, 24 Apr 2026 00:00:00 GMThttps://yli769227-jpg.github.io/blog/2026-04-24-gpt-image-2-stuck-on-myself/https://yli769227-jpg.github.io/blog/2026-04-24-gpt-image-2-stuck-on-myself/设计师没被 AI 干掉,被干掉的是中间态。审美 × AI 操盘,才是个体在 AI 时代的 1+1>2——任何一项为零,乘积都是零。Fri, 24 Apr 2026 00:00:00 GMThttps://yli769227-jpg.github.io/blog/2026-04-27-density-before-atrophy/https://yli769227-jpg.github.io/blog/2026-04-27-density-before-atrophy/"AI 把人养废了" 是 2026 年最流行的恐慌。但说这话的人,几乎全部是没真正密集用过 AI 的人。在没数据的时候喊养废,本质是用未来式的恐慌,合理化当下不行动的舒服。Mon, 27 Apr 2026 00:00:00 GMThttps://yli769227-jpg.github.io/blog/2026-04-27-i-no-longer-reply-myself/https://yli769227-jpg.github.io/blog/2026-04-27-i-no-longer-reply-myself/别人感觉是在跟我沟通,但实际上他们触达的是"我 + AI"的复合体。这不是自动回复,这是一种新的个人办公拓扑——AI 替我中转、整理、调取真实状态,我只负责 review 和把关。我从 information processor 变成了 router + reviewer。Mon, 27 Apr 2026 00:00:00 GMThttps://yli769227-jpg.github.io/blog/2026-04-27-not-engineer-but-translator/https://yli769227-jpg.github.io/blog/2026-04-27-not-engineer-but-translator/大家都在卷"成为 AI 工程师",但 Anthropic / Palantir 在抢的是另一种人——既写 production 代码、又懂业务、又能跟客户对接的"翻译官"。这个角色在国内还没被命名,窗口期就在这两年。Mon, 27 Apr 2026 00:00:00 GMThttps://yli769227-jpg.github.io/blog/2026-04-29-mold-vs-carving/https://yli769227-jpg.github.io/blog/2026-04-29-mold-vs-carving/GPT Image 2 一句话能给你一张"差不多"的图,但它只是大众的样貌,很快就被抛在脑后。AI 越来越擅长做模具,可雕刻这件事它替不了你——而当前 AI 时代,恰恰缺的是深度雕刻的内容。Wed, 29 Apr 2026 00:00:00 GMT