Sorta Insightful turns nine years old today!

Highlights

I took a break from writing puzzles this year. That’s led to a lot more free time. Puzzle writing has been one of my main recent hobbies, but I’ve found the problem is that I can’t do low key puzzlehunt writing. I either go all-in or I don’t, and when I go all-in, it takes up enough headspace that I have trouble making any other major life decisions. This year was a year where I needed to do other things. I wanted to change things up.

And I did change things up! Some of it was silent backend migrations (both Universal Analytics and my time tracker got deprecated this year), but the most notable change is that I switched career trajectories to AI safety.

I try to keep my Twitter / X usage low, but I retweet all posts whenever I write them, and I’ve noticed my switch into AI safety post has had significantly more Twitter engagement than my other posts. I chalk this up to “the alignment mafia” - a distributed set of people who reshare and promote anything supporting their views on AGI. Listen, I appreciate the support, but I haven’t done anything yet! Chill.

The time it took me to navigate my career change was much less than the time I’ve spent on puzzle writing in the past. I expected to fill that void with more blog writing, but that’s not what happened. So, what filled the void in its place?

Video games. It was video games. A quick review of some of them:

Hi-Fi Rush

Hi-Fi Rush starts simple but gets great once you unlock all the battle mechanics and get into the rhythm based flow. It’s very colorful, never takes itself that seriously, and is filled with fantastic set pieces.

Islands of Insight

Islands of Insight is…fine? It pitched itself as a puzzle-based MMO, where the highlight is a huge set of handmade Nikoli-style logic puzzles in an explorable world. The puzzles are good, but the game is poorly optimized, the social MMO aspects are pretty minimal, and the huge skill tree and unlockables feel like they’re just there to encourage higher engagement, rather than being more fun. The puzzles are great though, and if that’s good enough for you, I had fun with that.

Undertale Yellow

Undertale Yellow is a fantastic fan game, that’s been in development for 7 years and comes out feeling like a canon entry made by Toby Fox. I have small nitpicks about the plot and lack of variety in pacifist boss strategies, but the overall package works. I would have gladly paid money for this, but it’s free. If you liked Undertale check it out.

Hades 1

I bought the first Hades three years ago and never installed it. When Hades 2 went into early access, it was a great excuse to play the first one. I pushed up to Epilogue, all weapon aspects, and 21 heat before setting it down. At some point, it does get easy to fall into the builds you know are overpowered, but that’s the fate of every roguelike. What Hades 1 does well is make you feel overpowered when you get a good run going - and then killing you regardless if you don’t respect the bosses enough. It just does a lot of things right. I don’t like how grindy the story progression gets by the end, where you’re just waiting for a character to provide a dialogue option, but most players will not reach that point.

Incremental Games

Incremental games were a mistake, do not recommend. If you’re unfamiliar with the genre, Cookie Clicker is the prototypical example. You start with a small number of resources, then achieve exponentially larger resources through increasingly complicated progression and automation systems. I tried one of the highly recommended ones, and no joke, it took up 4 months of my life. If you are the kind of person who likes theorycrafting stat builds to maximize damage throughput, incremental games are a game where you only do that. Except, instead of maximizing damage, you’re minimizing time-to-unlock the next progression layer. I was spending 3 hours a day to push my game to a state where I could minimize how much time I’d need to wait to unlock the next layer, and if I didn’t spend those 3 hours I would have had to wait days for progress instead. It felt a lot like machine learning work, where you try to get everything done so that you can let the model train overnight and inspect it next morning. The experience was interesting, but I wouldn’t go through it again.

Statistics

Posts

I wrote 9 posts this year, up from 6 last year.

Time Spent

I spent 139 hours, 56 minutes writing for my blog this year, around 0.75x as much as last year.

View Counts

These are view counts from August 18, 2023 to today.

300   2023-08-18-eight-years.markdown  
247   2023-09-05-efnw-2023.markdown  
628   2023-11-25-neopoints-making-guide.markdown  
15837 2024-01-11-ai-timelines-2024.markdown  
1939  2024-01-21-mh-2024.markdown  
5076  2024-03-23-crew-battle.markdown  
826   2024-04-30-puzzlehunting-201.markdown  
8641  2024-07-08-tragedies-of-reality.markdown  
3923  2024-08-06-switching-to-ai-safety.markdown  

This continues to be a reminder that my view counts are heavily driven by who reshares things. I expected the AI related posts to be popular, but the post on the math of Smash Bros. crew battles is a big outlier. I shared it to the Smash subreddit, someone from there shared it to Hacker News, and that’s why it has so many views. (By the way, I’ve made some minor edits since that post went live, including a proof sketch for the final conjecture. Check it out if you missed it.)

Based on Twitter views, I can also see there’s a 6% clickthrough rate from my tweet saying I was joining an AI safety team to people actually reading the blog post.

Posts in Limbo

Is this a good time to confess that I never look back at the list of in-progress posts I write each year? I just write up the list, never read it, then go “oh yeah, that” when I reread my old posts to prepare the next anniversary post.

I’m no longer sure I get any value from sharing half-baked ideas, and may cut this in the future.

Post about Dominion:

Odds of writing this year: 5%
Odds of writing eventually: 10%

I haven’t touched my draft of this in a long, long time. I’m realizing it’s the kind of thing that could form a good Youtube essay (big fan of the History of the 2002-2005 Yu-Gi-Oh! meta video), but longform video content is not my skillset and not something I’m interested in getting better at.

Post about Dustforce:

Odds of writing this year: 20%
Odds of writing eventually: 60%

One hypothesis of the world is that positive and negative reinforcement plays a much larger role in behavior than people think it does. I’m partial to this, because I can tell I’ve played less Dustforce recently, almost entirely because of my personal laptop developing sticky key issues that make it just a bit more annoying to play. The game is still great, but it’s not a game you want to play with a keyboard that eats 5% of your jumps at random. This also affected my blogging motivation. Imagine trying to write longform text when your E key randomly sends 0 E presses or 2 E presses each time you touch it. So far, blogging has been saved by either writing on my work laptop or plugging in an external keyboard. Neither of those solutions work for Dustforce, since I won’t install it on my work laptop, and my external keyboards have key ghosting issues.

My personal laptop is getting pretty worn down, and I’m going to need to replace it before I travel for Mystery Hunt this year. (Pro tip: gaming laptops tend to prioritize good initial specs over long-term reliability.) One more thing to change - those video games aren’t gonna play themselves.