I was asked recently why it’s worth doing studies that prove discrimination is real. The people facing discrimination - the very people these studies are claiming to help - already know it exists. If you tell any woman you’ve done a study proving that some people talk over women, she’ll be like, yeah, no shit. If you tell a Black driver that some police discriminate during traffic stops, they’ll say the same.
Raindrops on roses and whiskers on kittens
Matrix transposes and not overfittin’
Well-maintained packages, clear docustrings
Networks with very low latency pings…
A frequently cited COVID metric is the positivity rate: that is, the fraction of COVID tests that come back positive. A low positivity rate suggests that an area has enough testing to properly monitor its outbreak; a high positivity rate suggests under-testing. The WHO suggests a 5% positivity rate as a threshold for reopening: ie, only 1 test out of 20 should come back positive.
I spent the last week or so doing this. I learned a lot – my views changed significantly several times, and I realized I disagreed with a lot of what my friends and family were doing. I’m writing about my experience because
My piece on PhD student mental health was published in Times Higher Education. (They graciously allowed me to provide a version that you can read and share without registering.) Thanks very much to the hundreds of PhD students who shared their stories and the many academics who provided feedback on the piece.
Content warning: sexual assault, abortion.
It was nearing midnight and it had been a long day, but I was excited. In the morning I had been reading about conv nets, a powerful deep learning technique that lets computers understand images; in the afternoon I had been chatting with one deep learning researcher about how to use them, in the evening I grabbed dinner with a second deep learning researcher, and now approaching bedtime I was tapping away on my laptop marveling at how, in a few lines of code, I could download a vast mathematical structure that could discriminate between a thousand image classes with near-human accuracy. I was messaging an economist friend about how my field was cooler than his because I was so damn excited.
I wrote this piece three years ago and never published it. But I returned to it recently and it still resonates with me (though a lot has changed since then), so I’m publishing it now.