Aadit Sheth(@aaditsh)

What AI is actually good at: - Code work you can verify. Tests pass, output matches, you read the d...

8.5内容质量

TL;DR · AI 摘要

AI擅长验证代码、撰写初步文本、处理重复性工作和探索未知领域，但不适用于需要品味、情感决策、实时事实和高风险决策。

核心要点

  • AI擅长验证代码和处理重复性工作
  • AI适用于初步文本撰写和探索未知领域
  • AI不适用于需要品味和情感决策的任务

结构提纲

按章节快速跳转。

  1. §AI的长处

    AI擅长处理需要验证的工作，如代码编写和测试通过。

  2. §AI的局限

    AI不擅长处理需要品味和情感决策的任务。

  3. §应用场景

    AI适用于处理重复性工作和探索未知领域。

思维导图

用一张图看清主题之间的关系。

查看大纲文本（无障碍 / 无 JS 友好）
  • AI的应用与局限

金句 / Highlights

值得收藏与分享的关键句。

#AI#编程#文本处理
打开原文
  • Code work you can verify. Tests pass, output matches, you read the diff.
  • First drafts of bounded text: emails, posts, summaries, briefs. Things you were going to edit anyway.
  • Grunt work: reformatting data, pulling from PDFs, renames," / X

What AI is actually good at: - Code work you can verify. Tests pass, output matches, you read the diff. - First drafts of bounded text: emails, posts, summaries, briefs. Things you were going to edit anyway. - Grunt work: reformatting data, pulling from PDFs, renames, migrations. Basically anything repetitive. - Exploring unfamiliar codebases or domains. Asking dumb questions without burning a colleague's time. - Synthesizing research when you give it the sources or when consensus is stable. - Thinking partner for problems with checkable answers: math, logic, code, structured analysis. What it's not good at: - Anything that requires taste. It can make something look polished. It can't tell you if it's actually good. - Emotional and life decisions. It sounds confident but it doesn't actually know you. That confidence is the dangerous part. - Real-time facts without web access. It will hallucinate prices, dates, and current events with a straight face. - High-stakes stuff you can't verify: legal, medical, financial. If you can't check the answer, don't trust it. - Negotiation or persuasion. It doesn't model the other person. It just generates text that sounds reasonable. - Topics where the internet is wrong. It repeats bad consensus with the same confidence as 2+2. I think the whole point is knowing when to reach for it and when not to. It's a tool. Use it like one.