Context for this post: the Global Challenges Project (GCP) is a workshop for people exploring AI safety (and biosecurity) careers. I’m not talking about Google Cloud Platform.

This was probably my worst summer ever. After ~10 weeks of the most socialization I’d ever had, I went back to my largely (but not totally) dead social life at home. In hindsight, even though I desperately wanted a break from Irvine, this was a terrible mistake.

Going from being around your friends twice a week or more to being around a friend once every few weeks is unbelievably painful. I’d spent the majority of my life being a largely non-social person, but this was the first time it genuinely hurt. Combining this with my newfound OCD sapped my soul. It’s insane how much mental bandwidth these two things can eat up. I found it harder to read and write than ever before.

So, obviously, I was very excited to leave my house to travel to Berkeley for EAGx and GCP. This would be the first time I’d taken a trip somewhere without my parents in nearly a decade. I’d also be in the Bay for the first time since I was, like, nine.

I actually don’t want to talk about my time at EAGx that much. I had some good one-on-ones, and obviously I was incredibly grateful to be around my friends, but I left with the same level of unease I came with. Was I truly, actually… NGMI? Say it ain’t so!

Anyway, I do want to talk about GCP. I was surprised by a few things.

First, I was surprised by how easy talking to a bunch of new people would be. I guess I’d been operating under my pre-2026 logic and hadn’t accounted for the fact that all of us were here for the same thing and were (generally) familiar with the same ideas. This isn’t necessarily true in other social settings, and it changed the way I’ll be thinking about events like these moving forward.

Second, I was surprised by how valuable “random” one-on-ones can be. A big portion of GCP is having one-on-ones, and you get to request certain people you want to talk to. Obviously, because multiple participants can request the same person you want to talk to, you can’t speak with every single person you request. I wouldn’t say that any of my one-on-ones were actually bad, but some were significantly more valuable than others. The ones that were the most valuable, in my opinion, were with people I hadn’t explicitly requested. I’m not sure if this is a taste thing that I should look into; it could be something to think about.

Third, I was surprised that Hamming circles sound extremely corny but can also be extremely high value. If you don’t know what that is (I had no idea), a Hamming circle is when you get a small group of people (at GCP, 3-4 people per circle), and you all take turns focusing on one person’s pressing issues. Any issue someone has can be brought up in the Hamming circle.

This sounds like an extremely vulnerable experience. That’s because it is. I almost didn’t participate, but I’m really glad I did. I think it helped me realize how much of my perceived doom was built up by my mind. Like, am I going to get sent to Chemistry Hell for having a bad summer? Or am I incompetent because I hadn’t written in months? Or am I ZERO-CONTEXT because I can’t pull out the entire list of every blog post I’ve ever read? Not necessarily, actually! I should probably stop beating myself up mentally over these things and actually start acting on them!

To sum it up, GCP was really good. I’d highly recommend that anyone seriously considering AI safety participate in one. It sucks that there’s a three-week gap between it and the quarter starting.

Berkeley is really cool. They should bury their power lines.