compression, oil and progress
summary of all the things i read and watched this week
started the day with aaron swartz’s productivity blogpost suggested by atharva. it’s very insightful and full of great advices. like the one below. this can be applied for anything big your life especially software projects.
still in awe of this polaris dawn video
read this lesswrong post about compression. the author presents some interesting points. it got me thinking about languages and how some languages are really different from each other as in how much dense they are - like how much ideas they compress in less words as possible like latin. it makes you wonder if intelligence is really the ability to compress knowledge
one more great post by scott alexander about “simulations”, prompting in chatgpt etc
started another day with discussing internal representation in ml with chatGPT. got to know some cool stuff
info dump on twitter - https://x.com/arpitingle/status/1835950597658542484
troubleshooting and debugging hardware projects is really hard. still working on that rpi project
read this interesting post on the case for filling the rivers
studied bayes theorem and watched the rational animations video about it
great post analysing openai’s o1 model
writer is really skeptical about o1 being a single model but some oai researcher already told that o1 is a single model in a tweet. really tells you how hard openai is cooking
watched the dwarkesh podacst with daniel yergin
learned many things
john rockefeller got three times more rich when standard oil broke and also it led to may innovations in the industry
teddy roosevelt ran the election on the basis of breaking the company (Standard Oil)
gasoline was a literally a waste product in the initial days
how middle east and oil is so tied
china’s vulnerability with energy
renewables getting entangled with geopolitics
i highly recommend reading the works in progress magazine. it’s available for free on their site
i love the stripe ecosystem - first stripe press and now this mag. it’s really incredible and every big org should strive to build such beautiful things
patrick is doing a lot for progress studies. the field seems very interesting. i should read more about it
found this very funny yt channel https://www.youtube.com/@introvertedmadness/videos
also i had this idea about a study. not sure if it will work
Owl Posting strikes again! this was a really great read
also yud cooked me on twitter lol
watched this incredible mit seminar by Hyung Won Chung (OpenAI) - Don’t teach. Incentivize: Scale-first view of Large Language Models
i agree that oai doesn’t publish much about the “research internals” but they have some of the best talks available online
also this video about o1 preview was amazing lol.
rewatched this episode and made some notes
clayton’s this tweet reminded me of sama’s Moore's Law for Everything essay. i really liked his American Equity Fund idea.
went down inside mech interp rabbit hole (again)
found a cool tool made by openai and this cool video video by welch labs
working on frontend is really annoying sometimes. also time to add some changes to my personal site. bookshelf page, some more lists. i have a lot of pending blogs. i write half and then start writing on a new one and the cycle continues lol. new blog next i promise.
SEE YA NEXT WEEK!