On Fermi Problems
-
Often you have a factor or two you have NO domain knowledge about at all, then just give your estimate to your best effort. It works because it becomes a random walk on log scale when you multiply each steps. Therefore, the mean is getting closer to the expectation
-
The more steps you have, the higher the standard deviation becomes. However, you often have better estimate on the smaller steps. This is essentially a trade off. But more steps also suggest your mental model is better
-
If the guess is too far off from the actual value, the reason of error is more valuable than the result itself
-
Only keep the part that is the deciding factor in the estimation
How many people in the world are talking on their cell phones in any given minute?
Each month i use 3 hours of cell phone, so 1/ (8 * 30) chance i will be using cellphone * 8 billion people
How many iPhone screen repairmen are there in the United States?
400 mil people, 60% with iphone, will repair the phone once in 1.5 years. So 160 mil repairs each year. Each repair takes 2 hours so each person can repair 4 * 365 = 1500 per year, i.e., about 100k iphone repair person!
How many people in US die each year?
400 mil people, birth rate at 1.5 per woman, so in 80 years, about 300 mil people are born, each year 3.5 mil are born ~ roughly equal the # of deaths