log
1 watched veritasium on markov chains
- bernoulli, the law of large numbers. initially required independence.
- nebraskov, or the tsar of probability. around 1905, related to russian revolution. h: since law of large numbers was assumed to require independence, then if law of large numbers applies, events are independent. applied it to social statistics which seem to conform to the law of large numbers, therefore decisions were independent, which scientifically proves free will.
- markov, or andrey the furious. just prove dependent events can follow the law of large numbers.
- pushkin, eugene onkein. this was the text markov used. he separated into vowels and consonants. each pair has a probability, each type of letter has a probability. used transition probabilities. chain converges to law of large numbers.
- ulam and neumann used this idea as a base to approximate how much uranium was needed; model multiplication factor on a histogram, showing statistically k. if k = 1, stable; k >1, bomb.
- they used the eniac.
- monte carlo method: ulam's uncle was a gambler. this discovery reminded ulam of the monte carlo casino
- 1993: internet open to the public.
- 1994: yang and filo, founders of search engine yahoo, met masayoshi and got invested in. used keyword trick, but easy to trick (white text on whitespace, which is like what teachers use to trick students who use llms) <- read more on keyword method
- norvig uses the analogy of library card stamps in books
- page and brin used markov chains. consider value of endorsement plus add damping rule.
- 1940s: shannon. instead of v and c, why not take every letter? then not only look at last letter to predict. use previous two. use entire words. you can make better and better predictions by using more and more of the previous words.
- attention is also used by modern llms. previous context informs next prediction, e.g., cell is it biology or prison?
- llms are vulnerable. systems w feedback loops become hard to model w markov chains.
- fun fact: riffling takes 7 to be basically random. overhand shuffle takes over 2k. so don't ask me to shuffle.
2 read ai sticker shock hits corporate america
- exciting to be cassandra. but we'll see. disconnect between what ai is and what it seems to be to the higher-ups and serfs. it's magic! it's not. it's called a lack of cognitive integrity.
- shows bluntly the disconnect in general too.
- relatedly, don't get high on your own supply. seen an awful lot of that.
3 read i wish people were more public
- personally ping-pong between the two. sometimes think i talk too much and share too much. then that the world is made for talking and sharing. sentimentally i prefer the latter.
- big problem is malicious busybodies, which i think is a pretty common character class nowadays. unfortunate.
4 recently, watched terence tao's collab w 3blue1brown
- watched both vids.
- really delightful hearing him speak. i mean, 3b1b is awesome, but tt, jeez. have to listen to him more. he also has a blog, which i've been meaning to read. so naturally, researched rss readers instead. see below.
- humankind has always been fascinated with what lies beyond. really puts into perspective how ingenious people can be.
- didn't take notes cuz i was too busy being gobsmacked. the determination of the orbits, sheer wonder and theivery.
5 some stuff i was looking at.
- 人中有松柏，天下無歲寒。that's a line from 何香凝. among people there are pines & cypresses, below heaven there is no eternal cold. 歲寒知松柏: (only) the year's frost knows pines & cypresses, meaning adversity reveals virtue.
- ellen bass' for janet at the new year. brilliance strewn / across the black felt sky like a / child gone wild with glitter and glue
- probably will try netnewswire
- always thought anki should be good and i would like to use it. these look good though: mochi