plzdontkillus diary (#2)
week two is done. i no longer feel like front flipping into a wood chipper. we're all going to make it laddies.
rereading my week one diary, i realize i made it sound like i wasn't having fun and that i hated the other fellows. to the contrary, i'm really yucking it up with everyone over here. it could be said that many people here have copious "lore" and "aura." let it be known.
also, i feel good about what ronny, aella, and the team are trying to do. they've been navigating through a set of non-obvious dilemmas (e.g. wanting to get people to make more ai alignment content, while also not wanting to guilt people into content-compliance via statusmogging).
first, a general observation. lighthaven is an extraordinary institution. it's great enough that i think it's fine when people apply to programs here specifically because they want to be here. i agree that this seems like a poor motivation, but it's a natural consequence of things being good. maybe you don't like lighthaven, or don't like rationalists in general, but i think you should appreciate that lighthaven events regularly have that "utopian feel." i'm thinking of robin hanson going around lessonline in shock going, "w-why is everyone here so interesting? it doesn't make sense." i'm thinking of scott sumner saying "at nearly seventy years of age, i've found my paradise at lighthaven." i'm thinking of inkhaven alums getting months-long comedowns once they return to their home cities.
ronny cleaves to a nice, pro-social value of wanting people to think of themselves as "providing information" rather than "convincing people." i didn't hear him unpack this model, but i like this distinction when i ruminate on it. the "convincing people" frame scopes in quite a lot of tricks which treat your audience like substrate. all of the following are helpful for "convincing people": being attractive, shaming low status outgroupers, toxoplasmic debate tricks, pandering to biases, insecurities, and resentment. in kantian terms, you, the convincer, are training yourself to treat people as "mere means," and not "ends in themselves." in contrast, the "providing information" frame respects people's epistemic autonomy.
after reading a lot of machiavelli, i'm no longer sure that the "providing information" frame is adequate for politics. it is, in any case, good for the soul, and as such should be the method of first resort when it comes to political persuasion.
posting on X has been going well. the follower count is growing smoothly. i am mainly cracking jokes there that either make me laugh or pander to my mutuals. the videos i post on my account have slightly farther reach than my text tweets (i've been calling this "mutual spillover" in my head). my current model for X growth is:
- get to know my mutuals better
- make jokes for them
- make videos to acquire mutual spillover
- repeat
my day 6 video is baudrillard at disneyland. this builds on the format i cloned the day before. i am greatly pleased by this video. it does pretty much everything i want it do.
this reinforces to me the utility and the fun of mimicking content that annoys me.
i enjoy finding japanese and korean cooking footage and editing it down. "food" is a very satisfying content vertical. it is also appealingly innocuous.
my day 7 video is novel superalignment technique. this is my first face video, and also my most successful one.
the inspiration for this video was like this: i saw a stack of unclaimed copies of "if anyone builds it everyone dies," and thought it would be funny if we stuffed thirty people in a tiny room and had them read copies of IABIED in silence together in a cultic manner. once me and a couple lads gathered the people together, we started throwing together different ideas. one of the fellows, matthew, started this moronic/genius in-joke of a "viral ai dance", which consists of doing the 6-7 hand-move horizontally. anyways, so as b footage for my cult-video idea, we took footage of the cultists doing the ai dance together, and i quickly threw together this shitpost video.
the ai yassified vishal at the end is a winner.
my day 8 video is conor mcgregor striking mastercat. this was a surprisingly tricky video to make, as many ai video platforms do not want to allow their users to make videos that do motion capture transfer from footage of celebrities. i've been using a few sneaky techniques to rout around this problem.
i also made sure the captions in this video are nice, and formatted in a way i think is funny / engaging. i think though that the captions are too low on the video, and should have been put in the apple negative space above the cat.
my day 9 video is the beach runner.
the original idea here was to make another fun baudrillard video. jeredino lavagnino gave me the following idea: take the part of amerique where baudrillard is talking about how runners are masochists who implement penal servitude into their own bodies -- then overlay a bunch of spacey runner footage over this. i loved this idea, but once i started working on the imagery for this video, i realized it would have to be serious and uninterrupted by the human voice.
this is, of course, another agi parable.
it was surprisingly difficult to generate the imagery for this video in my head. i spent a couple hours walking around lighthaven, trying make myself dissociate in order to see the video clearly: the runner tying his shoes multiple times; running on the beach; running backwards away from the great eye; running backwards away from the eye, but being pulled over to it across a bridge; the eye blooming in flower; the runner dissolving into the eye.
i think the gamelan fusion music is very good.
my day 10 video is a shiny homer simpson slopvid. i've been working on shiny homers, because i've been trying to clone pleometric's "homer let the barts out" tiktoks. the lipsyncing for this has proved challenging, as, again, many ai video platforms do not want to let their users make copyrighted characters lip sync to anything. i haven't found a successful workaround for this yet.
my day 11 video is some cult video b footage. what happened here is i finished my day 12 video, which was 21 minutes of edited footage, but it took me 7 hours to export the fucking thing (which involved reconstructing the editing on the video multiple times). serves me right for using capcut after it betrayed me the first time. i'll be using da vinci resolve exclusively from now on.
my day 12 video is low cortisol ai doomer yapfest, which is my first foray into yap-memer content. i enjoyed doing this quite a lot (even though the editing took an excruciatingly painful 15 hours, due to a comically extensive barrage of technical problems).
i enjoy this format, and will create more longform doom-yap.
libidos at slophaven are high, yet no sex is happening. how interbesting...