Untitled attachment

Notices where this attachment appears

1. Embed this notice
   mist (ai@cawfee.club)'s status on Tuesday, 17-Oct-2023 15:19:50 JST mist

   in reply to
   @hidden @MercurialBlack @ceo_of_monoeye_dating @jeffcliff @roboneko @scenesbycolleen
   I just played through it. One of the best interactive math explanations I've seen, *despite* its quadruple-vaxxed tone. I actually learned something new because I *incorrectly* guessed that "always cheat" would win because they can exploit "always cooperate."
   
   It slightly annoys me that they changed all the standard terminology. For example, "copycat" is usually called "tit for tat," and "copykitten" is "tit for two tats." I find the phrase "tit for tat" to be funnier too.
   
   Here's a funny bit of lore about the prisoner's dilemma tournament which the book is based on. It was a real public tournament which anybody could enter their own bot into. Lots of professors submitted incredibly sophisticated strategies, and they were all BTFO by the humble "copycat" which was the winner.
   
   There's a little-known way to gank prisoner's dilemma tournaments. What you do is submit one "main" bot and hundreds of "feeder" bots. Your bots start each round by executing a secret "handshake" move sequence; if they recognize each other as both being your bot, then the "feeder" will purposely let the "main" bot exploit it. If the opponent is not recognized as one of your own bots, then your bot will just play "copycat." As long as you can enter enough "feeder" bots into the tournament, your "main" bot will be guaranteed to win. The implication for cult behavior is clear.
   
   Lastly, there's an error in one of the explanations (see pic). None of the games considered in this slideshow are "zero-sum." Zero sum would mean that the payoffs in every case are exactly opposite: if you get +X, then I get -X. Since the "cooperate" case gives positive payout to both players, this game is not zero sum.