Philipp Hauer's Blog
Package by Feature
The Wall of Coding Wisdoms in Our Office
Improving Feedback Flows in Organizations with 'Complete Peer Feedback'
Effective Staff Appraisals with Employee Journey Maps
Don't Put Fat Jars in Docker Images
Slides and Recording of my Talks 'How to Mess up Code Reviews' at the JUG Saxony Day 2019 and the JCON 2019
Modern Best Practices for Testing in Java
MongoDB: Useful Development Tools and Snippets
Vaadin 10+: SASS Integration and CSS Refresh during Development
Focus on Integration Tests Instead of Mock-Based Tests
Package by Feature
The Wall of Coding Wisdoms in Our Office
Improving Feedback Flows in Organizations with 'Complete Peer Feedback'
Effective Staff Appraisals with Employee Journey Maps
Don't Put Fat Jars in Docker Images
Slides and Recording of my Talks 'How to Mess up Code Reviews' at the JUG Saxony Day 2019 and the JCON 2019
Modern Best Practices for Testing in Java
MongoDB: Useful Development Tools and Snippets
Vaadin 10+: SASS Integration and CSS Refresh during Development
Focus on Integration Tests Instead of Mock-Based Tests
Stephen Holiday
Spaced Repetition: My Learning Secret
That First Co-op Job
My Backup Strategy
How Project Rhino leverages Hadoop to build a graph of the music world
SMTP Email Relay for GMail (TLS) with Oozie Using Postfix
Automate Your iPod/iPhone/iPad's Media
Monitor Your Python App With FnordMetric and pyfnordmetric
Stack Overflow Word Trends by Day
Gender Prediction with Python
FidoFetch Architecture
Spaced Repetition: My Learning Secret
That First Co-op Job
My Backup Strategy
How Project Rhino leverages Hadoop to build a graph of the music world
SMTP Email Relay for GMail (TLS) with Oozie Using Postfix
Automate Your iPod/iPhone/iPad's Media
Monitor Your Python App With FnordMetric and pyfnordmetric
Stack Overflow Word Trends by Day
Gender Prediction with Python
FidoFetch Architecture
Devonthink擅长整理本地资料库,Evernote擅长搜集现有信息,GTD擅长项目管理,Roam Research 则擅长笔记写作,而他们都希望能够提高使用者的「生产效率」。
ribbonfarm
A Spectre Is Haunting The West
MJD 59,004
Through a Glass Lightly
New E-Book, and a Portfolio Update
Whistler’s Giantess
Leaking into the Future
Alamut, Bosch, Gaddis: Introduction to Epochal Art
Quarantine Art
Predictable Identities 27: Craving and the Pill
Liminality?…Well, there’s a free sample!
A Spectre Is Haunting The West
MJD 59,004
Through a Glass Lightly
New E-Book, and a Portfolio Update
Whistler’s Giantess
Leaking into the Future
Alamut, Bosch, Gaddis: Introduction to Epochal Art
Quarantine Art
Predictable Identities 27: Craving and the Pill
Liminality?…Well, there’s a free sample!
Ness Labs
From mental map to mental atlas
As we may die
Roam themes: how to style Roam Research with custom CSS
Mental wealth: managing your mental health budget
Productive cognitive load: make the most of your working memory
Thinking in maps: from the Lascaux caves to modern knowledge graphs
How to build a support group
Collectors to Creators Meetup Notes
The anatomy of a perfect educational article
The Goldilocks Principle of Stress and Anxiety
From mental map to mental atlas
As we may die
Roam themes: how to style Roam Research with custom CSS
Mental wealth: managing your mental health budget
Productive cognitive load: make the most of your working memory
Thinking in maps: from the Lascaux caves to modern knowledge graphs
How to build a support group
Collectors to Creators Meetup Notes
The anatomy of a perfect educational article
The Goldilocks Principle of Stress and Anxiety
这段时间看了很多推 RR 的文章,我来唱个反调。我认为这个工具对于绝大多数人来说都是不适用的,理由是(1)我们现存的绝大多数知识并不是编译成网状结构的,(2)只有足够大量的信息才能让网状结构产生明显的效率提升。
对于第(1)点,目前绝大多数的知识都是以书本这种树状结构,或者课程这种线性结构记录下来的,如果你要用 RR 做网状的知识管理,意味着你必须要先理解这些知识,然后自行编译成网状结构,相当于人工进行了一次数据结构的转换。且不说这种转换的成本,传统的树状或线性结构的知识是学术界数十甚至上百年打磨而成的,是基于经验做过高度优化的,你自己编译出来的很可能并不能够比这些传统的更加优异,除非你能够召集一个学术界专门做这种结构转换。事实上,string theory 学术界曾经做过类似的尝试,几年了没啥起色。
对于第(2)点,神经网络的基本原理之一就是必须有足够的数据集才能让算法起作用,否则大概率过拟合。这一点对于RR 的信息管理功能非常重要。对于绝大多数普通人而言,我们日常处理的信息量应该是远远达不到让网状结构产生效率提升的地步的。
所以我认为 RR 这种模式最终会是小众的,因为绝大多数人手里的信息并不适合被以这种网状结构储存或分析,在做了一段时间的尝试以后最终会回归传统的结构。并不是说这种网状结构有什么问题,只是它不适合绝大多数人的需求罢了。我们在判断一个工具是否好用的时候,一定要注意不能仅仅看它提出的概念有多吸引人。每个工具都有自己适用的范围。超出这个范围,再好的工具也会很鸡肋的。
对于第(1)点,目前绝大多数的知识都是以书本这种树状结构,或者课程这种线性结构记录下来的,如果你要用 RR 做网状的知识管理,意味着你必须要先理解这些知识,然后自行编译成网状结构,相当于人工进行了一次数据结构的转换。且不说这种转换的成本,传统的树状或线性结构的知识是学术界数十甚至上百年打磨而成的,是基于经验做过高度优化的,你自己编译出来的很可能并不能够比这些传统的更加优异,除非你能够召集一个学术界专门做这种结构转换。事实上,string theory 学术界曾经做过类似的尝试,几年了没啥起色。
对于第(2)点,神经网络的基本原理之一就是必须有足够的数据集才能让算法起作用,否则大概率过拟合。这一点对于RR 的信息管理功能非常重要。对于绝大多数普通人而言,我们日常处理的信息量应该是远远达不到让网状结构产生效率提升的地步的。
所以我认为 RR 这种模式最终会是小众的,因为绝大多数人手里的信息并不适合被以这种网状结构储存或分析,在做了一段时间的尝试以后最终会回归传统的结构。并不是说这种网状结构有什么问题,只是它不适合绝大多数人的需求罢了。我们在判断一个工具是否好用的时候,一定要注意不能仅仅看它提出的概念有多吸引人。每个工具都有自己适用的范围。超出这个范围,再好的工具也会很鸡肋的。