The following are a few things that I wanted to add to my application. # 1 In my application, I wrote that today I wouldn't make a tool like CodeAssist because I believe it would speed up AI timelines, and it's better for people to have more time to prepare. I want to add that I'm not against AI tools, creating or using them. And I use them myself. I simply wouldn't create tools that primarily contribute to speeding AI timelines. It's not a strong opinion, and I might change that opinion in the future. # 2 To my essay about predictions, I want to add that unagreed reciprocity might be useful/optimal when the distribution of power changes / is not constant. Because when it changes, then agreed reciprocity might not be possible because the people in power might stop the other people from agreeing on certain laws. # 3 In my predictions essay, I wrote that experimentally, if people are in a suboptimal equilibrium and ability to communicate is introduced, then people can switch to Pareto-optimal equilibrium (e.g. they can move from Hare-Hare to Stag-Stag, if they couldn't communicate and now they suddenly can). I'm less confident about the experiments now, given my source, but there is also the following theoretical justification why that would be the case. If you assume a game theory framework where players model themselves (build beliefs about each other based on their past moves, like in Reputation theory), then if you play an iterated Stag-Hunt game, and if there is enough future iterations ahead, then it makes sense to play Stag to build the reputation that you play Stag so that the other player will also switch to Stag. This reasoning applies not only to Stag-Hunt game but to situations where players are stuck in a bad equilibrium overall. If there is enough future ahead, then it's better for a player to switch to playing the strategy from Pareto-optimal equilibrium to build the reputation that you play that strategy so that the other player also switches to it. # 4 As for "history shouldn't be forgotten" in my essay about predictions, what I mean there is basically that if you have an equilibrium that is subgame-perfect but not markov-perfect (like contrite Tit-for-tat for example) - the strategies that are based on history of actions, then those stratgies only make sense if you don't reset the standing. The strategies that people use in real world, like indirect reciprocity are like that, because they are based on what actions someone took in the past.