Local Minimum

Finding a university for PhD and finding an apartment are both about finding the local minimum, which is that one university/apartment that could work. On the contrary, OpenAI is having an obvious bottleneck right now with the recently released not-so-good product and employees’ leaving. But I assume their struggle is much more fun, because they are finding the global minimum.

When stuck in a local minimum in Machine Learning, the thing to do is to increase the learning rate and get pass it. Well, obviously this does not work in real life, so let’s talk about the practical strategy here:

• To find a university for PhD, what you do is to read a lot of papers, find your topic and send applications. This is career-deterministic, so actually this is about finding the global minimum.
• To find an apartment, what you do is to browse a lot of (terribly designed, outdated, probably written in React) websites, send applications and have a lot of virtual showings with shaky cameras. Training neural network (my brain) on this type of data is a more or less waste of time.