Scattered thoughts on Inkhaven

Here are some scattered thoughts about Inkhaven, the blogging retreat thing I’m doing this month where you have to write a post every day. (This is a meta-post, so I’ll write a second non-meta post today since writing meta-posts is too easy.)

I like being forced to write blog posts

Before Inkhaven, I often spent multiple days not getting anything done. Writing a 500 word blog post doesn’t actually take that much time if you don’t care too much about quality. But I would prefer to write something that’s actually good, and I find having a hard requirement of writing something decent by the end of the day very useful for motivating myself.

I don’t think I’m doing Inkhaven optimally

I think I could be doing Inkhaven a lot better than I currently am. I should spend more time socializing with others, and more time getting feedback on my blog posts. I’ve been pretty reluctant to ask people for feedback on my blog posts because I’ve never written anything good and don’t want to waste people’s time by making them look at my bad writing, but I think getting more feedback might be helpful for improving my writing skills.

Lighthaven is great

Everybody I know who goes to Lighthaven likes it. It has great spaces to write blog posts from. They have a bunch of Apple Studio Displays which are really good 27-inch monitors; the visual quality is great and they charge your laptop through the same USB-C cable used for video output. I wish they had more Apple Studio Displays, it’s hard to articulate why but I just really like these monitors. I think I’ll get a Studio Display for myself after Inkhaven.

The aesthetics of all of the rooms here are also very good. I regret not really spending much time improving my room or office setup while I was working from home at my last job.

Lighthaven is a little too good for me actually. I’ve barely left Lighthaven while I’ve been doing Inkhaven, and I think I’m missing out on a lot of interesting activities and people in the Bay Area. I’m going to spend more time exploring the Bay Area the rest of this month.

Why does nobody fail Inkhaven?

It’s kinda weird how nobody has failed Inkhaven. There have been 1780 person-days of Inkhaven¹ so far and all of them have resulted in a blog post. It’s pretty interesting that it’s so effective! Writing a 500 word blog post is not actually that hard - if you can type at 50 WPM then you can write your daily post in 10 minutes. The hard part is deciding what to write, but if it’s 11:30pm and you don’t have anything it’s not that hard to just do a few paragraphs of stream-of-consciousness or meta-commentary or the bathrooms of Lighthaven, so it’s not that hard to nominally meet the writing requirement.

Can I do independent research?

So I spent a month in Montreal and figured I’d live there and do some independent AI safety research while I was there. I did end up reproducing some interesting results about secret goals in LLMs but I didn’t really do too much beyond that. Inkhaven doesn’t make you do AI safety research, but it provides really good places to do that from and I’ve been much more productive while I’ve been here. Plus I can write blog posts about my research to help meet my daily blog requirement.

But I’m don’t think I’ll be able to keep doing this to the same extent after Inkhaven if I live in Toronto by myself and don’t have any real scaffolding to get myself to research. Getting a job doing this stuff is probably hard but I think it would help with motivation and focus a lot. Maybe I could devise some other scaffolding myself to help me focus?

Plus one of the things I really liked when I did ARENA in January was pair programming with others, and collaborating on solutions. I wish I could do more of that, especially on AI safety research. It’s kinda isolating doing this by myself.