1/ podcast

80000 hours: build career with a positive social impact

Making sense with Sam Harris

2/ books

originals by adam grant


3/ data projects

learn from datakind

4/ advice on developing on leadership


start a proposal & communication for stakeholders, and then take it to your manager

start a group organizing people to discuss and work on something




it is interesting to continue watching Dropbox and its financial statementsfor the past year and here is what I learned:

1/ revenue increasing 13.5%, cost around the same. revenue in 2021H1 is 135 million. (market cap is around 12 billion)
2/ The revenue growth can be decomposed into: paid user growth around 7.8%, ARPU around 7%.
This is different from what we learned a year ago. Half of the revenue comes from ARPU growth. (back then we learned from Ravi that revenue is driven by # of paid users)
3/ I have cleared out my position, since I think the price has arrived at the level appropriate for next 2 years.



Argument: Time series can be very useful in Internet Industry?

As usual, we will list all the opposing arguments:

1/ Time series is a very narrow area(only for time series)

2/ Real time series forecast/detection problems, can only be solved based on a case-by-case manner

3/ Time series analysis was popular in Academia in 90’s. It becomes much less popular since then, especially comparing to CV and NLP. Also time series anlaysis frequentists lost the debate the Bayesianists.

4/ (Traditional) time series analysis is based on strong assumptions like: stationarity.

5/ Time series analysis methods like causality is not so useful.