2026年7月16日 · Research · Trend · Social Listening

用 last30days 看真实世界最近在吵什么

很多研究材料太干净了。报告已经被整理过，文章已经被编辑过，观点已经被包装过。它们适合建立背景，但不一定能告诉你最近大家真的在兴奋什么、抱怨什么、争论什么。

做博客选题、产品观察或者竞品判断，只看这类材料很容易慢半拍。

last30days 是什么

last30days 的价值，是把时间窗口压到最近 30 天，去看 Reddit、X、YouTube、TikTok、Hacker News、GitHub 和普通网页里的新鲜讨论。它不是替你做最终判断，而是帮你快速扫到最近的语言、情绪和争议点。

来源	它告诉你什么
Reddit	未经编辑的真实意见，带 upvote 数
X / Twitter	最早的反应和争论
YouTube	45 分钟深度的可引用片段
Hacker News	开发者共识
GitHub	PR 速度、issue、release notes
Polymarket	用真金白银投出来的概率，不是嘴炮

来源来自 mvanhorn/last30days-skill（GitHub 55.2K ⭐），它的定位是"用 upvotes、likes 和真金白银打分的 AI agent 搜索引擎，不是编辑筛的"。

我最常用的方式

先问三个问题：

1. 最近谁在讨论这个话题？ 2. 他们用什么词描述问题？ 3. 哪些观点被反复赞同或反复反驳？

这三个问题比"这个方向火不火"更有用，因为它们能把抽象趋势拆成具体人群、具体表达和具体冲突。

一个例子

如果我要写一篇关于 AI Agent 评测的文章，传统做法可能是先找框架、论文和大厂博客。last30days 会补另一层信息：最近开发者吐槽的是 trace 太难读，还是 eval case 难维护？产品人关心的是演示稳定性，还是上线后的安全边界？这些原话会直接影响文章标题和例子。

适用边界

场景	是否该用 last30days
做博客选题、竞品观察、趋势判断	✅ 先扫信号再回一手材料
需要权威长青资料、学术综述	❌ 它是雷达不是结论
只看一个平台就够了	⚠️ 单平台信号有偏差，跨源才有价值

社交平台的声音有偏差，热帖不等于市场真相，高互动也不等于高价值。last30days 更像雷达：先发现哪里有信号，再回到一手材料、产品页面、评论区、GitHub issue 和真实数据里验证。

对我来说，它最适合三类任务：做博客选题时找读者真正关心的问题；做竞品观察时看用户怎么抱怨和替代；做趋势判断时区分"媒体在写"还是"用户真的在用"。

Using last30days to See What People Are Actually Debating Now

Most research material is too clean. Reports are curated, articles are edited, opinions are packaged. They're great for building context, but they won't tell you what people are actually excited about, complaining about, or arguing over right now.

For blog topic selection, product observation, or competitive analysis, relying only on these materials means you're always half a beat behind.

What is last30days

last30days compresses the window to the most recent 30 days and surfaces fresh discussions across Reddit, X, YouTube, TikTok, Hacker News, GitHub, and general web pages. It doesn't make the final judgment for you—it helps you quickly pick up on recent language, sentiment, and points of contention.

Source	What It Tells You
Reddit	Unfiltered real opinions, with upvote counts
X / Twitter	Earliest reactions and debates
YouTube	Quotable clips with 45 min depth
Hacker News	Developer consensus
GitHub	PR velocity, issues, release notes
Polymarket	Market-backed probabilities, not hot air

Sources powered by mvanhorn/last30days-skill (GitHub 55.2K ☆), positioned as "an AI agent search engine scored by upvotes, likes, and real money—not editorial curation."

How I Use It

I start with three questions:

1. Who's been talking about this topic recently? 2. What words are they using to describe the problem? 3. Which views keep getting upvoted—or pushed back against?

These three questions are more useful than "is this direction hot?" because they break abstract trends into specific groups, specific language, and specific conflicts.

Example

If I'm writing about AI Agent evaluation, the traditional approach is to find frameworks, papers, and big-tech blogs. last30days adds another layer: are developers complaining about hard-to-read traces or hard-to-maintain eval cases? Are product people worried about demo stability or production safety boundaries? These exact quotes directly influence the article's title and examples.

Scope of Use

Scenario	Should you use last30days?
Blog topic selection, competitive observation, trend analysis	✅ Scan signals first, then go to primary sources
Need authoritative evergreen content, academic surveys	❌ It's radar, not conclusions
One platform is enough	⚠️ Single-platform signals are biased—cross-source is key

Social platform voices have bias. Trending posts aren't market truth, and high engagement isn't high value. last30days is like radar: detect where signals are, then verify against primary materials, product pages, comment sections, GitHub issues, and real data.

For me, it fits three types of tasks: finding what readers actually care about for blog topics; seeing how users complain and switch for competitive intel; and distinguishing "what media is writing about" from "what users are actually using" for trend analysis.