Comments by "Christian Baune" (@programaths) on "$400,000 a Year and 10 Hours A Week At GOOGLE" video.

  1. When I worked for the state, I made a deduplication software. It used Bayesian inference, and I found unintuitive things like there are more family names than first names. So, if you grouped people by family names, you ended up with smaller groups than if you grouped by first name. It's counterintuitive because you usually call people by their first name and rarely notice collisions. But you would be better off calling them by family names to avoid collisions. Which means that family name is a better discriminator. Better than date of birth. So, I had to write a paper on how the data is vectorized, fed to the deduplicator, and then restored. That paper was just unreadable because of the mathematics involved in the biases (weights). Summations over set theory and Bays. I didn't know how to make it simple without making it a thousand pages, so I wrote a few formulas, explained some observations, and made a table showing step-by-step what happens to a pair being compared (even if the software doesn't compare teams, but that's a very high-level view and in the end, that's kind of what you would do). The manager looked at it and told me it was convincing enough. I asked if he understood the formulas, and he told me: "No, but that looks very impressive, that will do." All he needed was that it looked correct and good, even if nobody understood it. 😂 He was a good manager because he prevented me from doing stupid things and focused on the technical part of my job. One day, I was writing an email to show how provably stupid a higher-up was, and that manager ran to me to say not to send "that" email. His first word was, "Don't send it" and then he explained the politics around it and that I should ignore it and do something else instead of sending that email and possibly getting fired. We could have done a mini-series about what happened there; it is just unbelievable ^^ The most comical one was when I was tasked to prepare some load tests, then I ran one iteration and got a direct phone call asking me to cease testing because servers went down. The real test was 50k times this with a slight ramp-up and cool-down. However, the test was too fast to send the data required to register a pupil. ^^ The load test was canceled, and it was fingers crossed in the hope school would not rush to do it on the last day. Organic smoothing. (That didn't happen ^^) In big companies, there are always some idiots who are too important and don't want to listen because they know better, and that lands pretty well for interesting situations.
    2