JELADA BLOG

Runtime Notes

A writing archive about code, reading, links, images, and everyday judgment.

Language

Home

Feeds

Author

Capture

Language

Auto 中文 EN

Stream

Thoughts+Links+Media

看得有点毛骨悚然，「周秦之变」也许在文化上不成立。周人就已经把「差序格局」-> 双标相对主义、「齐户编民」-> 计划经济玩模拟经营游戏、「社会秩序应符合天道」深入植入了这片土地。

Thought2026-08-25

2026-04-02

现在电脑上是不是可以同时存在系统1和系统2了

现在这被称为 harness 了，像驯化不稳定的马一样让 LLM 稳定的输出智力。

Thought2026-08-11

历代社会风俗事物考

正在读《历代社会风俗事物考》

BookReading2026-08-10尚秉和5/5

怎么让 LLM 像人非常有难度，而且似乎也很重要。想像这样一种场景：即使每个人都有一个通晓世间所有知识的 AI 陪伴在身边，社会的整体潜力会增加，但也不多，因为这还是需要人主动触发，而「主动性」很难仅靠知识提升。所以除了给人知识（方便）之外，还需要另一种辅助 AI，以培育人的元能力。

Thought2026-07-23

LLM 写代码可以，是不是写合同也行？

Thought2026-07-14

现在的 AI 服务器还会重新走一遍大型机、小型机、台式机、便携笔记本、手机这样的小型化之路吗？如果可以的话，人们当然愿意一个 AI 以助手的身份和实体贴近自己身边的，就像科幻作品中那样。

Thought2026-05-20

Codex cli 新出的 goal 命令去做因子挖掘的工作，效率很高，一晚上能跑出来比较可用的。

Thought2026-05-11

2026-03-13

For code generation, LLMs are already more than capable. Maybe the next domain is ops and security. Imagine this: a truly sharp security engineer watching every server log around t…

量化交易似乎也天然合适，对 alpha 的追寻拥有无休止的精力、100% 的纪律与「理性」、24 小时的交易市场，这都是人永远不可能拥有的。

Thought2026-04-23

AI 现在可以很廉价地吐出一套非常复杂的方案，看起来面面俱到，似模似样，实际越复杂它的幻觉越严重。可行的实践法则，还是让它从极简方案开始，然后配上可观测性，形成闭环后逐渐优化、生长。

Thought2026-04-23

观点的角度很高

一篇很好的控制理论与ai的综述

Link2026-04-15

为什么不懂物理的机器人反而赢了yage.ai

控制理论 ai

2026-03-15

Even if GPT-5.4 Pro drafts a detailed RFC, and Codex then breaks it down into milestones for implementation, once you are dealing with tens of thousands of lines of code, it is sti…

Thought2026-04-12

现在电脑上是不是可以同时存在系统1和系统2了

Thought2026-04-02

2026-03-27

Codex needs more than remote control. It also needs to be able to log into remote machines over ssh on its own.

For gathering information, one side needs web fetch , and the other needs access to a remote machine.For workflow orchestration, there needs to be a feedback loop: the spec shoul…

Thought2026-03-31

We learn all kinds of grand theoretical models about society, the state, and history. They seem to offer the individual a kind of psychological comfort, a sense of "thinking one un…

Thought2026-03-29

Codex needs more than remote control. It also needs to be able to log into remote machines over ssh on its own.

Thought2026-03-27

2026-03-15

Even if GPT-5.4 Pro drafts a detailed RFC, and Codex then breaks it down into milestones for implementation, once you are dealing with tens of thousands of lines of code, it is sti…

This is a natural place to apply feedback control.

Thought2026-03-15

Even if GPT-5.4 Pro drafts a detailed RFC, and Codex then breaks it down into milestones for implementation, once you are dealing with tens of thousands of lines of code, it is sti…

Thought2026-03-15

For code generation, LLMs are already more than capable. Maybe the next domain is ops and security. Imagine this: a truly sharp security engineer watching every server log around t…

Thought2026-03-13

If the default assumption behind vibe coding is that no human will read the code, then maybe the constraint that each file stay under a few hundred lines no longer means much. Mayb…

Thought2026-03-13

Walking the dog this morning, I saw green buds pushing out on the willow branches by the river. A light breeze passed through, and the branches tilted upward. It suddenly struck me…

Thought2026-03-06

Writing

Posts

Hello, blog.jelada.dev

This is the first post, used to verify the full pipeline for the posts list, detail page, RSS, and sitemap.

2026-03-05release blog

Tag aggregation sample: cross-type reuse

An extra post to verify how tags aggregate across thoughts, posts, and links.

2026-03-05product experience release note content strategy

Browse all posts