(previously: All the posts I will never write Refine Blogpost Day #3: The shortforms I did write  )

 Things that helped my writing

Key levers were (1) writing shortforms instead of posts (2) writing drafts on LW and letting them marinate for a while instead of writing everything in one go (3) writing with other people - especially Adam Shai  (4) lowering my expectations (5) Writing Quickly while maintaining Epistemic Rigor.  (6) starting with a draft of something I wrote on Twitter or ranted to somebody about in person (7) having more personal connections with people that seemed genuinely excited to read what I wanted to write (big difference to being at the Conjecture office in London, surrounded by all kinds of people interested in AI alignment and back at home where I’m the only one).  (8) paying somebody to streamline some of the tasks I have less comparative advantage in. (9) 'somebody is wrong on the internet' writing

What didn’t work

(1) An idea that sounded good in theory but turned out to not work so well was using OtterAI (speech-to-text software) to transcript rants. The transcript are too messy to serve as a useful draft.
(2) Writing with other people often failed when our writing goals & tastes weren't sufficiently aligned. 

Shortforms I wrote this month

Eight reasons to care about Kelly betting: (1) Kelly betting is asymptotically dominant, (2) Evolution selects for Kelly Betting, (3) Selection Theorems and Formal Darwinism, (4) Ergodicity Economics, (5) Kelly betting and Entropy, (6) Relevance of Information, (7) Bayesian Updating, (8) Blackjack

Multi-Step Fidelity causes Rapid Capability Gain Many examples of Rapid Capability Gain can be explained by a sudden jump in fidelity of a multi-step error-prone process. As the single step error rate is  lowered there is a sudden transition from a low fidelity to a high fidelity regime. 

Math research as Game Design In highschool math you follow a recipe to win the game. In proof-based math you have to come up with your own winning strategy. Math research is about creating new games by coming up with simple yet elegant rules that make the game fun to play.

Trapdoor Functions and Prime Insights A common way to think about hard puzzles is that they require an insight or several insights. An insight is analogous to a private key to a trapdoor function. There are cases where there are multiple ways to solve a puzzle but for some puzzles we feel there are certain 'prime insight' that are necessary to solve the puzzle. You cannot break the code without recovering this secret key. 

Artificial/Natural Why do we feel some things are Artificial and some things are Natural? Why do we often feel uneasy about Artificial things, while regarding Natural things as good?

Why do we need mental breaks?  Why do we get mentally tired? Why do we task switch? Three reasons: Global workspace theory, Hopfield networks as a model of human memory, reinforcement learning exploit/explore tradeoffs. 

 


 

12

New Comment

New to LessWrong?