What is this post about?
About two years ago I started this blog and then branched out to the EA forum and LessWrong. My experience was overwhelmingly positive (much better than anticipated). This post is intended to encourage people, share some lessons and make it easier to start.
I’d like to thank Max Räuker, Moritz Hanke and Simon Grimm for their feedback.
Clarification: these tips are intended for less experienced writers. I don’t claim to know everything and am interested in improving further. Feedback is welcome.
Clarification 2: Ronja has posted a response to this post in which she argues that my post sets the bar too high. I think my post can be read as such but my intention is to lower the bar of writing posts and get more people to do it.
- Time: I’d argue that following my suggestions doesn’t require much more time than not following them. A good structure makes the writing process faster and iterating doesn’t take that long but both lead to much better posts.
- Quality matters: It’s perfectly fine to write your thoughts in a post and put them online. However, your posts reflect on you. If people don’t like what they see they will not think it’s worth their time and not come back in the future.
TL;DR: writing blog posts is good for lots of reasons, e.g. gaining clarity, education, interaction with others and career building. There are different tips and tricks to write better ranging from cutting content aggressively to quantifying uncertainty. The most important one is to just start writing and then iterate. If you dislike having your own blog, write posts on the forum and hop on other people’s projects.
This was previously posted on the EA forum.
Why write a blog post?
- You learn a lot: Aiming at writing a post is a way of gaining clarity on what you know and what you don’t know. It will structure your thinking and lead you to research more thoroughly. Expecting others to read your post will likely also give you extra motivation to improve the article. Furthermore, I think people underestimate how much you can learn with 5 hours of focused googling, reading abstracts, skimming papers or videos, etc. By now, when I want to get a better grasp of a topic, I start writing a blog post.
- It’s a positive experience: Most of my initial intuitive concerns were very wrong. The world didn’t fall apart after sharing the first post and the feedback I got was positive. Most of the time, not that many people read your first posts and they are usually part of some ingroup, thus more forgiving.
On the other hand, once my blog posts got better, the experience just became more and more satisfying. People responded to my posts, answered important questions, reached out via DM or mail, etc. It makes me feel like I contribute something to solving important problems, which is a great experience.
- You help others: We don’t have the time and energy to research every topic or think about every question in detail. However, we can read a 10-minute summary to get a sufficiently good grasp of the issue. If you did a decent job with your post, most people will be very thankful for that.
- It might get you a job: Work trials at some EA organizations are comparable to writing a blog post, i.e. you are supposed to research a question and write down your results in a given amount of time. Having experience with that type of work is beneficial. Furthermore, your blog can already function as a signal for motivation and quality to potential future employers. I personally know of cases where people were hired because of blogposts they wrote.
What should I write about?
Here are some suggestions. I mostly chose example posts that are of very high quality in my opinion. I would frame them as something to aspire to, not the threshold for starting.
- Write about obvious things. We have a hard time understanding how large the understanding gap between experts and non-experts really is. Thus, extracting and summarizing your latent knowledge is super useful for others. A good example is Neel Nanda’s post “AI alignment a bird’s eye view”.
- Your latest research: Research speed increases by sharing your latest findings. A positive example of such a post is Ajeya Cotra’s “Why AI alignment could be hard with modern deep learning”.
- Clarifying beliefs: Writing down beliefs that are floating around in the community in a clear and concise manner is really beneficial for newcomers and to focus the discussion. Positive examples of such posts include Mathieu Putz’s “An update in favor of trying to make tens of billions of dollars” and Richard Ngo’s “AI safety from first principles”.
- Things you want to understand better: If you want to understand a topic better, just write a blog post about it, e.g. my own “How to sleep better”. Another example would be John G. Halstead’s “Are we going to run out of phosphorous?”.
- Beliefs you want to convince others of: Good examples include Luke Muehlhauser’s “EA needs consultancies” or Holden Karnofsky’s “The most important century” series.
- Deconfusion: Sharing small insights or specifying questions can already help your readership a lot. Someone who does this often and well is Ozzie Gooen.
- A beautiful story: This year’s creative writing contest featured many submissions that were just nice to read. The unweaving of a beautiful thing won but many other submissions are similarly inspiring.
- Question posts: Just writing down and clarifying an important question can spark a valuable discussion within the community. Such a discussion can also be a less effortful starting point for further investigation and future posts.
Tips and tricks:
- Use figures: Figures are often able to communicate more information than text. When people skim your post they will primarily look at figures. Usually, you can create a sufficiently good figure in 10 minutes with PowerPoint and they require less effort than expected. Also, often, Our World in Data has already done the job for you.
- True, understandable, concise–in that order: Write the core steps that are necessary to understand the argument–not more, not less. Make sure that all steps are true, communicate and quantify your uncertainty (see below) if you are unsure.
- Shorter is better. Cut everything that isn’t relevant to your core point. Use footnotes and appendices to provide further nuance where necessary. Often hinting at nuance is sufficient, for example, you can write “
(with some exceptions)” rather than listing all exceptions. This doesn’t imply that you can’t write long posts, just that they should be more information-dense than one usually assumes.
- Don’t provide a false balance: If you want to weigh two arguments and you think one is overwhelmingly stronger, say so and state the reasons. Statements like “On the one hand , on the other hand
” superficially remove you from your responsibility but ultimately misguide your readers. Your informed opinion is part of the service your post provides.
- People skim anyways–make it easy: Google analytics tells me that people spend about 2-5 minutes on a post that takes 10 minutes to read and I also skim sections when reading other posts. Therefore, headlines should be as meaningful as possible, important claims should be bold or italic. Allow your reader to decide what to skim and what to read in detail, don’t try to force them to read everything. If your post is a wall of text or confusing, people won’t bother to read it at all.
- Provide a summary at the top: This not only helps people to save time but also improves clarity–it is easier to follow an argument if you already know its conclusion. Exceptions include short posts (less than 1 page), posts where the title is the summary, and posts that are lists of things, e.g. this post wouldn’t necessarily need a TL;DR.
I think one of the reasons why I find some of Elizier’s posts hard to read is because they start with a wall of text rather than a summary (see e.g. “Causal Diagrams and Causal Models“). It’s a great post but you don’t know what it’s about until you’re halfway through.
- Make the post about one thing only: Focus your post on one question or statement. If you can summarise the post as “this post is about X” you’re fine. If you realize the post is actually about multiple things, split it up and write separate posts whenever possible.
- Your title sets expectations: Depending on your content the title can be a summary, a question or a teaser. I would recommend simple and descriptive titles, e.g. How to be happy by Luke Muehlhauser. The title is a simplification, after all, it doesn’t provide a perfect recipe to happiness but it still contains the gist of the post. In rare circumstances, your title can also depict the vibe of the post, e.g. in Richard Sutton’s The Bitter Lesson.
- Use confidence estimates: The sentence “I believe X (ca. 80%)” is better than just “I believe X”. Words that quantify something such as “overwhelmingly”, “strongly”, “uncertain”, etc. are useful but they carry different meanings for different readers. Thus, using numbers is more precise.
If you are unsure about your number, you can still communicate that, e.g. “ca. 80%” or “80% +- 10” to indicate confidence intervals. You are not legally liable for these numbers and will get them wrong sometimes—readers will be thankful for them anyway.
- Target audience: Different forums require different levels of knowledge, e.g. the EA forum vs. a personal blog. When presenting background knowledge I would recommend providing a very brief summary and references to more detailed introductions.
- Iterate yourself: During the writing process, the first iteration should set up the general structure of the post and dump the text in a stream-of-consciousness-like fashion. Further iterations can be used to focus content, cut text, refactor structure, and improve the wording. I would take some time (at least one day) between iterations to get your head clear and remove your identity from a particular wording. The more time has passed the easier it is to cut bad content. Overall, I think iterating multiple (e.g. 3-5) times is worth the time. If it takes 1 hour to reduce the average reading time by 1 minute without loss of information, 60 readers already justify your effort.
- Write with partners: Writing with others can be more motivating, reduce your workload, improve iteration speed and produce better blog posts overall. I would generally recommend it unless the experience gap between authors is very large.
- Send your post to people: When somebody sends me a post, I’m happy about it. Yet, when I think about sending someone my post or sharing it on Twitter, I feel like I’m intruding or being annoying. All people I asked shared this impression, so you should probably just send your post to people that might enjoy it. In the worst case they decide not to read it. In the best case they learned something or had a good time.
How to get started:
- Create your own blog: With GitHub pages or WordPress you can create a blog in less than an hour. I don’t think a more complex setting is necessary unless you have very specific preferences. You can easily buy a domain name within minutes, e.g. with Namecheap. See e.g. “Why have a blog” for more.
- Write a forum post: Writing a post on the EA forum is super easy and the editor is very intuitive. It’s much less commitment than having your own website.
- Hop on projects: If you just want to try out writing, you can usually join other people’s projects. Writing with multiple people is more fun anyway and you will have someone to hold you accountable.
I’m happy to answer further online or in person at any EA meetup.
One last note
If you want to get informed about new posts you can follow me on Twitter.
If you have any feedback regarding anything (i.e. layout or opinions) please tell me in a constructive manner via your preferred means of communication.