Attack on Titan got popular for two big reasons:
- Strikingly good art, in both the manga and the anime versions. Picture a zoomed-out rendering of the literally gigantic Titans towering over a (detailed!) cityscape, followed by kinetic action scenes where humans fight the Titans by using grappling hooks to zip from rooftop to rooftop, dodging attacks like a commando team of Spider-Man clones fighting a giant robot. If Marvel actually made that as a movie scene, would it be dumb? Yes. Would it possibly be awesome? Also yes, especially if it came out when you weren't already jaded by similar fight scenes from other movies.
- A brutal, dystopian world where named characters frequently die. Unlike the Marvel version where heroes get punched into a wall and just pick themselves back up, in Attack on Titan, people die, often gruesomely; the Titans swat humans like flies and eat them like grapes. In a famous scene, human forces launch a large and utterly doomed cavalry charge across an open field, because mass suicide is the only diversion tactic available to let someone else fight effectively. The series includes a number of mass graves and Pyrrhic victories.
Once it was popular, it hit the problems intrinsic in a long-running series that has both a high death count and a horrifying, faceless threat: If you keep killing off named characters, you run out of characters important to the plot. If the threat is a mindless horror that refuses to negotiate, you can't have a good villain monologue. On top of that, you have to escalate somehow, i.e. make a Titan that's even worse than the ones that came before. Say, maybe it's a super-special Titan that could ruin everything, but this one is also a guy who used to be Eren's friend - treachery! Unless Eren can talk him out of it...? Now you have a villain monologue, you're getting as much mileage as possible out of a newly-important character, and the situation for Our Hero is both worse than, and interestingly different from, the last 13 times he fought a big monster in this week's episode of Big Monster Fighting Team.
So it eventually got bogged down in this kind of nonsense, obviously because the author was a victim of his own success and didn't expect to still be writing the thing years later (IIRC, he's admitted as much). The Wikipedia summary cuts out the parts that made it successful in the first place, so if you just look at the summary of each character's biggest plot beat, it's a mystery why anyone would have ever liked it.
You're not supposed to actually seal your mouth - you put just enough tape on that it's annoying or difficult to open your mouth while awake, so that while asleep you won't mouth-breathe by reflex. If you (say) roll over and your mouth would otherwise fall open, the tape keeps your mouth closed or compels you to roll back over and keep breathing through the nose - but if an assassination squad of tiny spiders tried to kamikaze up your nose to suffocate you, you could still snap awake, force open your mouth, and take the tape off.
- Prev
- Next

I've been experimenting with LLMs (as one does, these days) and trying to solve a particular problem: Prepping to run a game of Starfinder 2E. It's working quite well!
The problem with asking an AI about any RPG is that it will confuse the details with every other RPG, and easily hallucinate answers that don't belong to any game at all. The solution: Have the AI extract the content of the Starfinder books (as their digital PDF versions) into Markdown files, i.e. text with annotation. Then give the AI specific rules that it must refer to the extracts for any rules question.
This required some testing to refine - a set of evals that determine whether the AI is getting confused, and why. It's one of the places where an AI can effectively bootstrap itself; I could bring up the common problems with getting an AI to understand a specific RPG, have it write the skill files and evaluations, get advice on which kind of evaluations are useful, have it run the tests and determine if the answers were satisfactory, etc.
But the end result is a project where I can pose a question like, "My level X party is going through an adventure where they fight an enemy necromancer; how much treasure should they get during this part, and can you suggest some treasure that's specifically necromancer-themed but still useful to the party's soldier?" Some follow-up work got me a skill where I can give a brief on a kind of creature I want to homebrew, and it will draft out a couple of options. It saves a huge amount of time and bookkeeping to have approximately-correct stats and abilities, and I can then edit the results up to my standards before inflicting it on my players. Although the average enemy it brainstorms isn't very exciting, sometimes it will borrow an ability I hadn't considered or pick a theme from the book that I can work with. The AI doesn't have taste, but I do, so I can read several proposed monsters and say, "Actually, creature 2 has a cool ability - let's make it acid instead of lightning, and splice it onto creature 4, because that one fits in the adventure better..."
For next week: Formatted handouts for my players in their first starship combat; the book that includes rules for starship combat just came out, and they'll be blasting off. I plan to have handouts for the ship itself and each battle station, so a player whose character sits at the pilot's console can get the handout just for piloting actions, and the players can trade sheets when their characters switch stations.
More options
Context Copy link