I'm looking for other junior researchers to form a distributed writing group for mutual support. Please get in touch if you're interested.


One to three other people and I.

I do AI alignment research independently. I'm working on the Farlamp project (see the GitHub repo or the project announcement on LessWrong). And I live in Japan.


Read and discuss one another's work.

Duration: 15 min x number of participants


Each at their desk, all connected by video call.




Because in a group we can:

  • Bring up questions, responses, suggestions for each other's research.
  • Review each other's outlines and drafts.
  • Keep each other disciplined.
  • Support each other in other ways.


  • If you want to join, then PM or email me (<given name of President Nixon>.moehn@posteo.de).
  • We start and end our meetings on time.
  • We talk about the state of our work, free-form or following prompts. This format will change based on what we find works well, how far a person is in their research etc.


Idea, content and rationale derived from Booth et al.: The Craft of Research, p. 32. Here is the whole section about it:

Creating a Writing Group

A downside of academic research is its isolation. Except for group projects, you'll read and write mostly alone. But it doesn't have to be that way. Look for someone other than your instructor or adviser who will talk with you about your progress, review your drafts, even pester you about how much you've written. That might be a generous friend, but even better is another writer so that you can comment on each other's ideas and drafts.

Best of all is a group of four or five people working on their own projects who meet regularly to read and discuss one another's work. Early on, each meeting should start with a summary of each person's project in this three-part sentence: I'm working on X because I want to find out Y, so that I (and you) can better understand Z (more about this in 3.4). As your projects advance, develop an opening "elevator story," a short summary of your project that you could give someone on the way to a meeting. It should include your research question, your best guess at an answer, and the kind of evidence you expect to use to support it. The group can then follow up with questions, responses, and suggestions.

Don't limit your talk to just your story, however. Talk about your-readers: Why should they be interested in your question? How might they respond to your argument? Will they trust your evidence? Will they have other evidence in mind? Such questions help you plan an argument that anticipates what your readers ex- pect. Your group can even help you brainstorm when you bog down. Later the group can read one another's outlines and drafts to imagine how their final readers will respond. If your group has a problem with your draft, so will those readers. But for most writers, a writing group is most valuable for the discipline it imposes. It is easier to meet a schedule when you know you must report to others.

Writing groups are common for those writing theses or dissertations. But the rules differ for a class paper. Some teachers think that a group or writing partner provides more help than is appropriate, so be clear what your instructor allows.


4 comments, sorted by Click to highlight new comments since: Today at 10:30 AM
New Comment

What kind of level of knowledge are you looking for? I would be interested in this, but I'm mainly investigating agent foundations, rather than the ML approach.

I would get together with anyone and see how it works out. If the differences in knowledge or topics cause discontent or other problems, we call it quits. Stricter criteria I can't set, because I don't have any experience with writing groups. And when it comes to knowledge of ML/math/research, I'm at a rather low level. Which isn't to say that I'm a complete nitwit. My main achievements are just in other areas.

Would be happy to try this

Cool. Then we're three so far. I'll wait until tomorrow for more responses. Then I'll get back to the responders to schedule a time.

New to LessWrong?