Hi! I finished another week at Inkhaven so I’m writing another update email. I’m still in Inkhaven, the writing retreat where you have to write a blog post every day. I’ve had a lot of output over the 13 days I’ve been here!

I’ve been thinking about meta-models for LLM interpretability recently and I think they’re a really cool way of interpreting how LLMs think! I’ve done a bunch of experiments and written up some research notes and general thoughts about meta-models (links below). Also I wrote about a bunch of other things, because I have to be writing something every day. If you have any thoughts about meta-models please let me know! (ideally through Discord (@moreloops) or by commenting on my LessWrong post about them)

Posts since last update

• Don't default to doing nothing on April 13, 2026
• Secret fields on Canadian income tax returns on April 12, 2026
• Scattered thoughts on Inkhaven on April 12, 2026
• Why I'm excited about meta-models for interpretability on April 11, 2026
• Blog posts I won’t write on April 10, 2026
• Latent reasoning oracles research notes on April 9, 2026
• Getting chat-tuned models to act kinda like base models on April 8, 2026
• Why was cybersecurity automated before AI R&D? on April 7, 2026
• Judge prediction markets by depth, not volume on April 6, 2026

Inkhaven

I've been really liking Inkhaven so far! It's been a lot of fun writing posts every day. You should check out the Inkhaven website which has posts from all the residents; there are a lot of great posts from others. (I really liked ForeverHaven which invented the genre of Inkhaven meta-fiction.)

This email technically counts as a daily post for Inkhaven (since you can read it on the web archive page), but I'm going to try very hard to write a second post today because it kinda feels like cheating to count this as my daily post.

Future

I'm pretty unsure what I'm going to write about for the rest of Inkhaven. I've already posted all of the obviously good things on my list of possible posts, so my blog posts will probably be on a much more varied set of topics than they've been on so far. I might write about some cybersecurity research I've done in the past. Also might write about my life history. I might even write some fiction! (If I completely run out of ideas I might even have to resort to meta-posting about Inkhaven and my lack of ideas.)

I haven't been exploring the SF Bay Area as much as I planned to while I've been here for Inkhaven. (Lighthaven is such a nice place that you never want to leave.) I'm going to visit SF this week so I can explore it a bit.

I guess I should start planning what I'm going to do after Inkhaven soon; right now I don't have any plans for what I'm going to spend my time on after April. I'm probably going to hang around the Bay Area for another ~week after Inkhaven so I can spend some more time exploring and meeting people here. Once I have to leave SF (sad!) I'm going to head back to Toronto. I might spend some more time in Montreal too.

I'm not sure what I'm going to do once I'm back in Canada; I might try doing some more independent research on meta-models for interpretability, which are so cool. Maybe I'll get a full-time job. idk.

Please reach out if you have any thoughts or insights for me! Replying to this email might work but I haven't tested it, so preferably you should DM me on Discord as @moreloops.