Podcast Lab
I listen to a lot of tech podcasts. Some episodes go over my head because they assume background I don't have. The Dwarkesh ones especially - an episode can spend an hour deep in debt markets or chip supply chains. When I want to actually understand one of those episodes, I make a page for it here. The page explains the basic concepts first, then goes through what was discussed in the episode. I retain a lot more this way.
What happens once AI can automate AI research? →
Ryan Greenblatt expects AI to fully take over AI research around 2031, and after that a year of automated work could produce four or five years of progress. The debate leans on machine-learning vocabulary the hosts never define, so this page explains reinforcement learning, verifiability, algorithmic progress, reward hacking, and alignment specs, and then goes through the argument and its takeover scenario.
Who pays for the AI buildout? →
Dylan Patel thinks AI infrastructure will need trillions of dollars a year, and that a lot of it will be borrowed. Borrowing at that scale pushes up interest rates for everyone. The finance concepts the episode assumes, from dollars per megawatt through debt, crowding out, and discount rates, each get a section with a model to play with, and the argument itself comes after.