The Most Human Documentation
summary
An Open Source Vale.sh plugin distills Wikipedia’s "Signs of AI Writing" into 18 rules and runs deterministic linters rather than fuzzy LLM checks; edits are visualized with a HistoryFlow replay of the page and the plugin flags hedging, excessive corrective contrast, and other telltale patterns as warnings. The GitHub repo has 10 stars versus a competing Humanize Library at 17,000 stars; a deliberate typo is left in docs as an experiment and the plugin successfully surfaces it.
AI-written summary · about
chapters (14)
- 0:00Introduction
- 1:02Lindy effect and frustrations
- 3:46Turing test and Most Human Human
- 7:15AI prose rise and flaws
- 10:02Cognitive cost of writing
- 11:46AI patterns on Wikipedia
- 13:52Vale and composable tools
- 16:58Humanizer project success
- 19:25Value of human imperfection
- 23:37AI vs human technical content
- 25:20Manual, mechanized, and automated workflows
- 26:44Documentation as credibility
- 27:46History and future of docs
- 28:27Q&A
related talks
24mDocumentation and AI: How to write right now
43mHow to Use AI to Build Accessible Products
49mNext-Gen Documentation with AI
41mWikiLoop: Big Tech's Open Knowledge contributions
38mThe Cognitive Cost of AI-Assisted Open Source
43mMaking AI transparent with RAG on your own data
35mBuilding AI-powered user interfaces
26mBuild AI Systems for Discernment, Not Approval