Curated. This brought together a few different interests of mine (pedagogy, wargames, and chaos theory as a field) and presented some newer ideas.
I’ve heard of a few variants of Simulation Games like the ones described here. I’ve participated in and heard a bunch about AI Futures wargames, and have a friend who was considering running a “Local Politics simulation” to get rationalists more familiar with how local politics goes.
I’ve felt a bit sus about them, since, they have a lot of degrees of freedom about how to arrange the scenario. (AI Futures wargames do have the benefit of us getting to see the resolution in a few years and see how it compares, but, like, not till it’s too late)
Each individual component of “have students participate in a historical multiple times” is relatively obvious, at least in retrospect, but, individually it would be much less interesting to do nonhistorical ones, do them only once, or do them without students.
Historical means there’s some ground truth you can compare them against and check that the simulation is behaving remotely reasonably. (A quick Claude search says there are lots of wargame scenarios that simulate future battles or events, but, very few that are historical, and run multiple times).
The students are interesting because they turn what would be a very expensive process into one that actually has a market-fit that makes it natural to run repeats of.
The notion of using history-limited LLMs to run these much faster/cheaply has some obvious concerns/limitations (i.e. re AI judgment), but, it seems interesting as a source of rough intuitions. (I’ve heard they are notably less smart, I’m interested in whether there have been attempts to solve that with Deepseek like “have them do a bunch of math or something with fewer historical implications”)
Curated. This brought together a few different interests of mine (pedagogy, wargames, and chaos theory as a field) and presented some newer ideas.
I’ve heard of a few variants of Simulation Games like the ones described here. I’ve participated in and heard a bunch about AI Futures wargames, and have a friend who was considering running a “Local Politics simulation” to get rationalists more familiar with how local politics goes.
I’ve felt a bit sus about them, since, they have a lot of degrees of freedom about how to arrange the scenario. (AI Futures wargames do have the benefit of us getting to see the resolution in a few years and see how it compares, but, like, not till it’s too late)
Each individual component of “have students participate in a historical multiple times” is relatively obvious, at least in retrospect, but, individually it would be much less interesting to do nonhistorical ones, do them only once, or do them without students.
Historical means there’s some ground truth you can compare them against and check that the simulation is behaving remotely reasonably. (A quick Claude search says there are lots of wargame scenarios that simulate future battles or events, but, very few that are historical, and run multiple times).
The students are interesting because they turn what would be a very expensive process into one that actually has a market-fit that makes it natural to run repeats of.
The notion of using history-limited LLMs to run these much faster/cheaply has some obvious concerns/limitations (i.e. re AI judgment), but, it seems interesting as a source of rough intuitions. (I’ve heard they are notably less smart, I’m interested in whether there have been attempts to solve that with Deepseek like “have them do a bunch of math or something with fewer historical implications”)