Wednesday, June 8, 2011

Diet, Beliefs Pay Rent, and Fermi

This morning I was able to sit for 20 minutes of concentration easily and spent only about 2+2+10/4ish = 4.5 definitely distracted from the breath. I switched up my diet to an omelet for breakfast, a light salad for lunch, a light salad snack at 3 'cause it was there, and a light corn/chickpeas dinner. I had lots more energy throughout the day - tomorrow I'll go back to my regular schedule to see if I can switch back and forth. During an afternoon meeting, I applied the "do a simple calculation instead of trying to intuit directly to the answer" heuristic which directly led to our team making the right decision on a question with a ~$45k swing in cost-benefit, which we only had a ~75% chance of making right otherwise (partially due to sunk costs), so that's +$10,000. Although actually there's a subsequent spot (after spending some of the effort) we'd have a good chance of killing it, so call it +$5,000 for the team.

I brainstormed a bunch of beliefs I hold <count> 30 actually, and then asked foreach what things the belief prohibited. I didn't take long, but instead brainstormed what they prohibited until getting 1-3 things for each belief - I think that's an important thing, since I want to practice seeing what my beliefs prohibit quickly and automatically, not after a period of focused thought. Then I went back and checked and found 7 I could actually just go test, 4 of which were Fermi calculations and 2 others of which I could and did test without much effort (that leaves 1: Tomorrow after work I will see if I can run 600 feet in 1 minute :D). Finally I did the 4 Fermi calculations which yielded:

  • ~0 evidence in the model for or against buying a car being a good idea (which is some evidence for "buying a car is neutral EV")
  • Quite a lot of evidence in the model for writing an app which assists modeling your revealed preferences as a decision utility being a good idea
  • Quite a lot of evidence in the model that I should cut way back on playing Dominion online, by a factor of about 5 before it's EV neutral again.
  • Quite a lot of evidence in the model that I should tentatively start up working through some of the exercises in Bayesian Data Analysis.


  1. Are you looking for a beginner's Bayesian Stats textbook? I recommend Data Analysis: A Bayesian Tutorial by Sivia.
    I love talking about Bayesian Statistics so let me know if you have questions, especially about what is useful, not useful etc.

  2. Good point, I had skimmed the first few chapters of that a while ago but going back to it is an excellent idea. Much better, I expect, than working through Bayesian Data Analysis at the moment.